Você está na página 1de 240

1

Version released: December 20, 2012

Introduction to Mathematical Logic


Hyper-textbook for students by Vilnis Detlovs, Dr. math., and Karlis Podnieks, Dr. math. University of Latvia
This ork is licensed under a reative ommons License and is copyri!hted " 2000-2012 by us, Vilnis Detlo#s and $arlis %odnieks& 'ections 1, 2, ( of this book represent an extended translation of the correspondin! chapters of the book: V. Detlovs, !lements of Mathematical Logic, )i!a, *ni#ersity of +at#ia, 1,-., 2/2 pp& 0in +at#ian1& 2ith kind permission of Dr& Detlo#s&

Vilnis Detlovs. Memorial Page


In "re"aration 3 fore#er 0ho e#er, since 2000, used successfully in a real lo!ic course for computer science students1& This hyper-textbook contains links to: 2ikipedia, the free encyclopedia4 5acTutor History of 5athematics archi#e of the *ni#ersity of 't 6ndre s4 5ath2orld of 2olfram )esearch&

Table of Contents
)eferences&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&( 1& 7ntroduction& 2hat 7s +o!ic, )eally8&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&. 1&1& Total 9ormali:ation is %ossible;&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&/ 1&2& %redicate +an!ua!es&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&10 1&(& 6xioms of +o!ic: 5inimal 'ystem, <onstructi#e 'ystem and <lassical 'ystem&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&21&.& The 9la#or of %ro#in! Directly&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&(= 1&/& Deduction Theorems&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&.1 2& %ropositional +o!ic&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&/0 2&1& %ro#in! 9ormulas <ontainin! 7mplication only&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&/0 2&2& %ro#in! 9ormulas <ontainin! <on>unction&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&/1 2&(& %ro#in! 9ormulas <ontainin! Dis>unction&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&/. 2&.& 9ormulas <ontainin! ?e!ation 3 5inimal +o!ic&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&/= 2&/& 9ormulas <ontainin! ?e!ation 3 <onstructi#e +o!ic&&&&&&&&&&&&&&&&&&&&&&&&&&-. 2&-& 9ormulas <ontainin! ?e!ation 3 <lassical +o!ic&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&-/ 2&=& <onstructi#e @mbeddin!& Ali#enkoBs Theorem&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&-, 2&C& 6xiom 7ndependence& *sin! <omputers in 5athematical %roofs&&&&&&&&=1 (& %redicate +o!ic&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&C(&1& %ro#in! 9ormulas <ontainin! Duantifiers and 7mplication only&&&&&&&&&&C(&2& 9ormulas <ontainin! ?e!ations and a 'in!le Duantifier&&&&&&&&&&&&&&&&&&&&&CC (&(& %ro#in! 9ormulas <ontainin! <on>unction and Dis>unction&&&&&&&&&&&&&&&&,C (&.& )eplacement Theorems&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&100 (&/& <onstructi#e @mbeddin!&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&10C .& <ompleteness Theorems 05odel Theory1&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&11C .&1& 7nterpretations and 5odels&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&11C .&2& <lassical %ropositional +o!ic E Truth Tables&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&1(0 .&(& <lassical %redicate +o!ic E AFdelBs <ompleteness Theorem&&&&&&&&&&&&&1(, .&.& <onstructi#e %ropositional +o!ic 3 $ripke 'emantics&&&&&&&&&&&&&&&&&&&&&&&1/C /& ?ormal 9orms& )esolution 5ethod&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&1=C /&1& %renex ?ormal 9orm&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&1C0 /&2& 'kolem ?ormal 9orm&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&1,0 /&(& <on>uncti#e and Dis>uncti#e ?ormal 9orms&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&1,/ /&.& <lause 9orm&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&200 /&/& )esolution 5ethod for %ropositional 9ormulas&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&20/&-& HerbrandBs Theorem&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&21. /&=& )esolution 5ethod for %redicate 9ormulas&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&220 -& 5iscellaneous&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&2(( -&1& ?e!ation as <ontradiction or 6bsurdity&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&2((

#eferences
$ilbert D., %ernays P. &'()*+ Arundla!en der 5athematik& Vol& 7, Gerlin, 1,(., .=1 pp& 0)ussian translation a#ailable1 Kleene ,. . &'(-.+ 7ntroduction to 5etamathematics& Van ?ostrand, 1,/2 0)ussian translation a#ailable1 Kleene ,. . &'(/0+ 5athematical +o!ic& Hohn 2iley I 'ons, 1,-= 0)ussian translation a#ailable1 Mendelson !. &'((0+ 7ntroduction to 5athematical +o!ic& 9ourth @dition& 7nternational Thomson %ublishin!, 1,,=, ..0 pp& 0)ussian translation a#ailable1 Podnieks K. &'((0+ 2hat is 5athematics: AFdelBs Theorem and 6round& 1,,=-2012 0a#ailable online, )ussian #ersion a#ailable1&

'. Introduction. 1hat Is Logic, #eally2


13#4I456 7n this book, "redicate language is used as a synonym of first order language, formal theory 3 as a synonym of formal system, deductive system, constructive logic 3 as a synonym of intuitionistic logic, algorithmically solvable 3 as a synonym of recursively solvable, algorithmically enumerable 3 as a synonym of recursively enumerable. 2hat is lo!ic8
'ee also 9actasia +o!ic by )o!er Gishop Hones&

7n a sense, lo!ic represents the most !eneral means of reasoning used by people and computers& 2hy are means of reasonin! important8 Gecause any body of data may contain not only facts #isible directly& 9or example, assume the follo in! data: the date of birth of some person J is Hanuary 1, 2000, and yesterday, 'eptember 1., 2010 some person K killed some person L& Then, most likely, J did not kill L& This conclusion is not represented in our data directly, but can be deri#ed from it by usin! some means of reasonin! 3 a7ioms 0Mback!round kno led!eN1 and rules of inference& 9or example, one may use the follo in! statement as an axiom: M5ost likely, a person of a!e 10 canOt kill anybodyN& There may be means of reasonin! of different le#els of !enerality, and of different ran!es of applicability& The abo#e Mkiller axiomN represents the lo est le#el 3 it is a #ery specific statement& Gut one can use la s of physics to deri#e conclusions from hisPher data& Theories of physics, chemistry, biolo!y etc& represent a more !eneral le#el of means of reasonin!& Gut can there be means of reasonin! applicable in almost e#ery situation8 This 3 the most !eneral 3 le#el of means of reasonin! is usually re!arded as logic& Gut is lo!ic absolute 0i&e& uniQue, predestined1 or relati#e 0i&e& there is more than one kind of lo!ic18 7n modern times, an absolutist position is some hat incon#enient 3 you must defend your MabsoluteN concept of lo!ic a!ainst heretics and dissidents, but #ery little can be done to exterminate these people& They may freely publish their concepts on the 7nternet& 'o let us better adopt the relati#ist position, and define lo!ic0s1 as any common frame8ork for building theories& 9or example, the so-called

/ absolute !eometry can be #ie ed as a common lo!ic for both the @uclidean and non-@uclidean !eometry& Aroup axioms ser#es as a common lo!ic for theories in#esti!atin! mathematical structures that are subtypes of !roups& 6nd, if you decide to rebuild all mathematical theories on your fa#orite set theory, then you can #ie set theory as your lo!ic&
<an there be a common lo!ic for the entire mathematics8 To a#oid the absolutist approach let us appreciate all the existin! concepts of mathematics 3 classical 0traditional1, constructi#ist 0intuitionist1, ?e 9oundations etc& Rf course, enthusiasts of each of these concepts must propose some specific common frame ork for buildin! mathematical theories, i&e& some specific kind of lo!ic& 6nd they do& <an set theory 0for example, currently, the most popular #ersion of it 3 Lermelo-9raenkelBs set theory1 be #ie ed as a common lo!ic for the classical 0traditional1 mathematics8 Kou may think so, if you do not ish to distin!uish bet een the first order notion of natural numbers 0i&e& discrete mathematics1 and the second order notion 0i&e& ScontinuousS mathematics based on set theory or a subset of it1& Rr, if you do not ish to in#esti!ate in parallel the classical and the constructi#ist 0intuitionist1 #ersions of some theories&

'.'. 9otal :ormali;ation is Possible6


Aottlob 9re!e 01C.C-1,2/1 <harles '& %eirce 01C(,-1,1.1 Gertrand )ussell 01C=2-1,=01 Da#id Hilbert 01C-2-1,.(1

Ho far can e proceed ith the mathematical ri!or 3 ith the axiomati:ation of some theory8 <omplete elimination of intuition, i&e& full reduction of all proofs to a list of axioms and rules of inference, is this really possible8 The ork by Aottlob 9re!e, <harles '& %eirce, Gertrand )ussell, Da#id Hilbert and their collea!ues sho ed ho this can be achie#ed e#en ith the most complicated mathematical theories& 6ll mathematical theories ere indeed reduced to systems of axioms and rules of inference ithout any admixture of sophisticated human skills, intuitions etc& Today, the lo!ical techniQues de#eloped by these brilliant people allo ultimate axiomati:ation of any theory that is based on a stable, self-consistent system of principles 0i&e& of any mathematical theory1& 2hat do they look like 3 such S100T ri!orousS theories8 They are called formal theories 0the terms Mformal systemsN and Mdeducti#e systemsN also are used1 emphasi:in! that no step of reasonin! can be done ithout a reference to an exactly formulated list of axioms and rules of inference& @#en the most Sself-e#identS lo!ical principles 0like, Sif A implies B, and B implies C, then A implies CS1 must be either formulated in the list of axioms and rules explicitly,

or deri#ed from it& 2hat is, in fact, a mathematical theory8 7t is an <engine< generating theorems& Then, a formal theory must be an Sen!ineS !eneratin! theorems ithout in#ol#in! of human skills, intuitions etc&, i&e& by means of a precisely defined al!orithm, or a computer pro!ram& The first distincti#e feature of a formal theory is a "recisely defined 0SformalS1 language used to express its "ro"ositions& S%recisely definedS means here that there is an al!orithm allo in! to determine, is a !i#en character strin! a correct proposition, or not& The second distincti#e feature of a formal theory is a precisely defined 0SformalS1 notion of "roof& @ach proof pro#es some proposition, that is called 0after bein! pro#ed1 a theorem& Thus, theorems are a subset of propositions& 7t may seem surprisin! to a mathematician, but the most !eneral exact definition of the Sformal proofS in#ol#es neither axioms, nor inference rules& ?either Sself-e#identS axioms, nor SplausibleS rules of inference are distincti#e features of the SformalityS& 'peakin! strictly, Sself-e#identS is synonymous to Saccepted ithout ar!umentationS& Hence, axioms andPor rules of inference may be S!ood, or badS, Strue, or falseS, and so may be the theorems obtained by means of them& The only definitely #erifiable thin! is here the fact that some theorem has been, indeed, "roved from by usin! some set of axioms, and by means of some set of inference rules& Thus, the second distincti#e feature of SformalityS is the possibility to verify the correctness of "roofs mechanically, i&e& ithout in#ol#in! of human skills, intuitions etc& This can be best formulated by usin! the 0since 1,(- 3 precisely defined1 notion of al!orithm 0a Smechanically applicable computation procedureS1: 3 theory T is called a formal theory, if and only if there is an algorithm allo8ing to verify, is a given te7t a correct "roof via "rinci"les of T, or not. 7f somebody is !oin! to publish a Smathematical textS callin! it S proof of a theorem in theory TS, then e must be able to #erify it mechanically hether the text in Question is really a correct proof accordin! to the standards of pro#in! accepted in theory T& Thus, in a formal theory, the standards of reasonin! should be defined precisely enou!h to enable #erification of proofs by means of a precisely defined al!orithm, or a computer pro!ram& 0?ote that e are discussin! here verification of ready "roofs, and not the much more difficult problem 3 is some proposition pro#able in T or not, see @xercise 1&1&/ belo and the text after it1& 6xioms and rules of inference represent only one 0but the most popular;1 of the possible techniQues of formali:ation&

= 6s an unpractical example of a formal theory let us consider the game of chess, let us call this StheoryS CHESS& +et Bs define as "ro"ositions of CHESS all the possible positions 3 i&e& allocations of some of the pieces 0kin!s included1 on a chessboard 3 plus the fla!: S hite to mo#eS or Sblack to mo#eS& Thus, the set of all possible positions is the language of CHESS& The only a7iom of CHESS is the initial position, and the rules of inference 3 the rules of the !ame& The rules allo passin! from some propositions of CHESS to some other ones& 'tartin! ith the axiom and iteratin! mo#es allo ed by the rules e obtain theorems of CHESS& Thus, theorems of CHESS are defined as all the possible positions 0i&e& propositions of CHESS1 that can be obtained from the initial position 0the axiom of CHESS1 by mo#in! pieces accordin! to the rules of the !ame 0i&e& by Spro#in!S #ia the inference rules of CHESS1& !7ercise '.'.'. <ould you pro#ide an unpro#able proposition of CHESS8 2hy is CHESS called a formal theory8 2hen somebody offers a Smathematical textS P as a proof of a theorem A in CHESS, this means that P is a record of some chess-!ame stopped in the position A& <heckin! the correctness of such SproofsS is a borin!, but an easy task& The rules of the !ame are formulated precisely enou!h 3 e could rite a computer pro!ram that ill execute the task& !7ercise '.'... Try estimatin! the si:e of this pro!ram in some pro!rammin! lan!ua!e& Rur second example of a formal theory is only a bit more serious& 7t as proposed by %aul +oren:en, so let us call this theory L& Pro"ositions of L are all the possible S ordsS made of letters a, b, for example: a, b, aa, aba, baab& Thus, the set of all these S ordsS is the language of L& The only a7iom of L is the ord a, and L has t o rules of inference: X U- Xb, and X U- aXa& This means that 0in L1 from a proposition X e can infer immediately the propositions Xb and aXa& 9or example, the proposition aababb is a theorem of L: a U- ab U- aaba U- aabab U- aababb rule1 rule2 rule1 rule1 This fact is expressed usually as L U- aababb 0 S+ pro#es aababbS, U- bein! a Sfallen TS1& !7ercise '.'.). a1 Verify that L is a formal theory& 0Hint: describe an al!orithm allo in! to determine, is a seQuence of propositions of L a correct proof, or not&1 b1 0%& +oren:en1 Verify the follo in! property of theorems of +: for any X, if L U- X, then L U- aaX& Rne of the most important properties of formal theories is !i#en in the

C follo in! !7ercise '.'.*. 'ho that the set of all theorems of a formal theory is algorithmically enumerable, i&e& sho that, for any formal theory T, a al!orithm 6T can be defined that prints out on an 0endless1 paper tape all theorems of this theory 0and nothin! else1& 0Hint: e ill call T a formal theory, if and only if e can present an al!orithm for checkin! texts as correct proofs #ia principles of reasonin! of T& Thus, assume, you ha#e . functions: GenerateFirstText ! 3 returns Text, Generate"extText ! 3 returns Text, #sCorrectProof Text! 3 returns tr$e or fa%se, ExtractTheorem Text! 3 returns Text, and you must implement the functions GenerateFirstTheorem ! 3 returns Text, Generate"extTheorem ! 3 returns Text1& *nfortunately, such algorithms and "rograms cannot solve the "roblem that the mathematicians are mainly interested in= is a given "ro"osition A "rovable in T or not2 2hen, executin! the al!orithm 6T, e see our proposition A printed out, this means that A is pro#able in T& 'till, in !eneral, until that moment e cannot kno in ad#ance hether A ill be printed out some time later or it ill not be printed at all& 4ote. 6ccordin! to the official terminolo!y, al!orithmically enumerable sets are called Srecursi#ely enumerable setsS, in some texts 3 also Slistable setsS& !7ercise '.'.-. a1 Describe an al!orithm determinin! hether a proposition of L is a theorem or not& b1 <ould you ima!ine such an al!orithm for CHESS8 Rf course, you can, yet&&& Thus you see that e#en, ha#in! a relati#ely simple al!orithm for chec&ing the correctness of proofs, the problem of pro'abi%ity can be a #ery complicated one& T is called a solvable theory 0more precisely 3 algorithmically solvable theory1, if and only if there is an al!orithm allo in! to check hether some proposition is pro'ab%e by usin! the principles of T or not& 7n the @xercise 1&1&/a you pro#ed that L is a sol#able theory& 'till, in the @xercise 1&1&/b you established that it is hard to state hether CHESS is a Sfeasibly sol#ableS theory or not& Determining the "rovability of "ro"ositions is a much more com"licated task than checking the correctness of ready "roofs. 7t can be pro#ed that most mathematical theories are unsolvable, the elementary 0first order1 arithmetic of natural numbers and set theory included 0see, for example, 5endelson V1,,=W, or %odnieks V1,,=W, 'ection -&(1& 7&e& there is no al!orithm allo in! to determine, is some arithmetical proposition pro#able from the axioms of arithmetic, or not& 4ote. 6ccordin! to the official terminolo!y, al!orithmically sol#able sets are called Srecursi#e setsS&

, ?ormally, mathematical theories contain the ne!ation symbol not& 7n such theories sol#in! the problem stated in a proposition A means pro#in! either A, or pro#in! notA 0Sdispro#in! AS, Srefutin! AS1& 2e can try to sol#e the problem by usin! the enumeration al!orithm of the @xercise 1&1&.: let us ait until A or notA is printed& 7f A and notA ill be printed both, this ill mean that T is an inconsistent theory 0i&e& usin! principles of T one can pro#e some proposition and its ne!ation1& 7n !eneral, e ha#e here . possibilities: a1 A ill be printed, but notA solution1, b1 notA ill be printed, but A solution1, ill not 0then the problem A has a positi#e ill not 0then the problem A has a ne!ati#e

c1 A and notA ill be printed both 0then T is an inconsistent theory1, d1 neither A, nor notA ill be printed& 7n the case d1 e may be aitin! fore#er, yet nothin! interestin! ill happen: usin! the principles of T one can neither pro#e nor dispro#e the proposition A, and for this reason such a theory is called an incom"lete theory( The famous incompleteness theorem pro#ed by $urt AFdel in 1,(0 says that most mathematical theories are either inconsistent or incom"lete 0see 5endelson V1,,=W or %odnieks V1,,=W, 'ection -&11& !7ercise '.'./. 'ho that any 0simultaneously1 consistent and complete formal theory is sol#able& 0Hint: use the al!orithm of the @xercise 1&1&., i&e& assume that you ha#e the functions GenerateFirstTheorem ! E returns Text, Generate"extTheorem ! E returns Text, and implement the function #sPro'ab%e Text! 3 returns tr$e or fa%se1& 2here the consistency and completeness come in8 !7ercise '.'.0. a1 Verify that Sfully axiomatic theoriesS are formal theories in the sense of the abo#e !eneral definition& 0Hint: assume, that you ha#e the follo in! functions: GenerateFirstText ! E returns Text, Generate"extText ! E returns Text, #sPropositon Text! E returns tr$e or fa%se, #sAxiom Proposition! E returns tr$e or fa%se, there is a finite list of inference rule names: X) 1, &&&, )nY, function App%y )$%e"ame, List*fPropositions! E returns Proposition or fa%se, and you must implement the functions #sCorrectProof List*fPropositions! E returns tr$e or fa%se, ExtractTheorem Proof! E returns Proposition1& b1 0for smart students1 2hat, if, instead of X) 1, &&&, )nY, e ould ha#e an infinite list of inference rules, i&e& functions Aenerate9irst)ule01, Aenerate?ext)ule01 returnin! )ule?ame8

10

'... Predicate Languages


$istory
9or a short o#er#ie of the history, see Duantification& 'ee also: 6ristotle 0(C.-(22 G<1 3 in a sense, the Sfirst lo!icianS, S&&& as not primarily a mathematician but made important contributions by systemati:in! deducti#e lo!ic&S 0accordin! to 5acTutor History of 5athematics archi#e1& Aottlob 9re!e 01C.C-1,2/1 3 S7n 1C=, 9re!e published his first ma>or ork Begriffsschrift, eine +er arithmetischen nachgebi%+ete Forme%sprache +es reinen ,en&ens 0<onceptual notation, a formal lan!ua!e modelled on that of arithmetic, for pure thou!ht1& 6&Aeor!e and ) Heck rite: ((( #n effect, it constit$tes perhaps the greatest sing%e contrib$tion to %ogic e'er ma+e an+ it -as, in any e'ent, the most important a+'ance since 6ristotle( ((( 7n this ork 9re!e presented for the first time hat e ould reco!nise today as a lo!ical system ith ne!ation, implication, uni#ersal Quantification, essentially the idea of truth tables etc&S 0accordin! to 5acTutor History of 5athematics archi#e1& <harles 'anders %eirce 01C(,-1,1.1: S&&& He as also interested in the 9our <olour %roblem and problems of knots and linka!es&&& He then extended his fatherBs ork on associati#e al!ebras and orked on mathematical lo!ic and set theory& @xcept for courses on lo!ic he !a#e at Hohns Hopkins *ni#ersity, bet een 1C=, and 1CC., he ne#er held an academic post&S 0accordin! to 5acTutor History of 5athematics archi#e1& Hilary %utnam& %eirce the +o!ician& Historia .athematica, Vol& ,, 1,C2, pp& 2,0-(01 0an online excerpt a#ailable, published by Hohn 9& 'o a1& #ichard %eatty. %eirceBs de#elopment of Quantifiers and of predicate lo!ic& "otre ,ame /( Forma% Logic, Vol& 10, ? 1 01,-,1, pp& -.-=-& Aeraldine Grady& 9rom %eirce to 'kolem& 6 ?e!lected <hapter in the History of +o!ic& @lse#ier 'cience: ?orth-Holland, 2000, 2000, -2/ pp& 0online o#er#ie at http:PP &else#ier&comP psPfindPbookdescription&c sZhomeP-21/(/Pdescription[description 1&

2hen tryin! to formali:e some piece of our 0until no 3 informal1 kno led!e, ho should e proceed8 2e ha#e an informal #ision of some domain consistin! of Mob>ectsN& 2hen speakin! about it, e are utterin! #arious propositions, and some of these propositions are re!arded as MtrueN statements about the domain& Thus, our first formali:ation task should be definin! of some forma% %ang$age, allo in! to put all our propositions about the domain in a uniform and precise ay& 6fter this, e can start considerin! propositions that e are re!ardin! as

11 MtrueN statements about the domain& There may be an infinity of such statements, hence, e canBt put do n all of them, so e must or!ani:e them someho & 'ome minimum of the statements e could simply declare as axioms, the other ones e could try to deri#e from the axioms by usin! some r$%es of inference& 6s the result, 'ection1& e could obtain a forma% theory 0in the sense of the pre#ious

7n mathematics and computer science, the most common approach to formali:ation is by usin! of the so-called "redicate languages, first introduced by A& 9re!e and <& '& %eirce&
07n many textbooks, they are called first order languages, see belo second order lan!ua!es&1 the arnin! about

*sually, lin!uists analy:e the sentence SHohn lo#es GritneyS as follo s: /ohn 3 sub>ect, %o'es 3 predicate, Britney 3 ob>ect& The main idea of predicate lan!ua!es is as follo s: instead, let us rite %o'es /ohn, Britney!, here %o'es x, y! is a t o-ar!ument predicate, and /ohn, and Britney both are ob>ects& 9ollo in! this ay, e could rite 0 x, y! instead of x0y& This approach 3 reducin! of the human lan!ua!e sentences to #ariables, constants, functions, predicates and >uantifiers 0see belo 1, appears to be flexible enou!h, and it is much more uniform hen compared to the #ariety of constructs used in the natural human lan!ua!es& 6 unified approach is much easier to use for communication ith computers&
6nother example: SGritney orks for G57 as a pro!rammerS& 7n a predicate lan!ua!e, e must introduce a (-ar!ument predicate Sx orks for y as :S, or orks0x, y, :1& Then, e may put the abo#e fact as: orks0Gritney, G57, %ro!rammer1&

Language Primitives Thus, the informal #ision behind the notion of predicate lan!ua!es is centered on the so-called SdomainS 3 a 0non-empty81 collection of Sob>ectsS, their SpropertiesS and the SrelationsS bet een them, that e ish to SdescribeS 0or SdefineS81 by usin! the lan!ua!e& This #ision ser#es as a !uide in definin! the lan!ua!e precisely, and further 3 hen selectin! axioms and rules of inference& ?b@ect Variables Thus, the first kind of lan!ua!e elements e ill need are ob@ect variables 0sometimes called also indi#idual #ariables, or simply, #ariables1& 2e need an unlimited number of them1: x, y, :, x1, y1, :1, &&& The abo#e-mentioned SdomainS is the intended Sran!eS of all these #ariables&

12
!7am"les. 11 Guildin! a lan!ua!e that should describe the Sdomain of peopleS, e must start by introducin! S#ariables for peopleS: x denotes an arbitrary person& 21 Guildin! the lan!ua!e of the so-called first order arithmetic, e are thinkin! about Sall natural numbersS as the ran!e of #ariables: 0, 1, 2, (, ., &&& 0x denotes an arbitrary natural number1& (1 Guildin! the lan!ua!e of set theory, e think about Sall setsS as the ran!e of #ariables: x denotes an arbitrary set&

M,omain of peop%e1 represente+ as a 2.L c%ass +iagram 4ote. 'ince our screens and printers allo only a limited number of pixels per inch, in principle, e should !enerate #ariable names by usin! a finite set of characters, for example, by usin! a sin!le letter x: x, xx, xxx, xxxx, xxxxx, &&& ?b@ect onstants The next possibility e may ish to ha#e in our lan!ua!e are the so-called ob@ect constants 0sometimes called indi#idual constants, constant letters, or simply, constants1 3 names or symbols denotin! some specific Sob>ectsS of our SdomainS&
!7am"les& 11 7n our Slan!ua!e for peopleS people: Hohn, Gritney etc& e may introduce constants denotin! particular

21 7n the lan!ua!e of first order arithmetic, e may ish to introduce t o constants 3 0 and 1 to denote S:eroS and SoneS 3 t o natural numbers ha#in! specific properties& (1 7n the lan!ua!e of set theory, e could introduce a constant denotin! the empty set, but there is a ay to do ithout it as ell 0for details, %odnieks V1,,=W, 'ection 2&(1&

:unction onstants

1( 7n some lan!ua!es e may need also the so-called function constants 0sometimes called function letters1 3 names or symbols denotin! specific functions, i&e& mappin!s bet een Sob>ectsS of our SdomainS, or operations on these ob>ects&
!7am"les& 11 7n our Slan!ua!e for peopleS e ill not use function constants& 21 7n the lan!ua!e of first order arithmetic, e ill introduce only t o function constants S\S and S]S denotin! the usual addition and multiplication of natural numbers, i&e& the t oar!ument functions x\y and x]y& (1 7n the lan!ua!e of set theory, e could introduce function constants denotin! set intersections x y , unions x y , set differences x3y, po er sets %0x1 etc&, but there is a ay to do ithout these symbols as ell 0for details, %odnieks V1,,=W, 'ection 2&(1&

7n mathematics, normally, e are ritin! f x, y! to denote the #alue of the function f for the ar!ument #alues x, y& This 0the so-called SprefixS notation1 is a uniform ay suitable for functions ha#in! any number of ar!uments: f x!, g x, y!, h x, y, 3! etc& 7n our e#eryday mathematical practice some of the t oar!ument functions 0in fact, operations1 are represented by usin! the more con#enient SinfixS notation 0x\y, x]y instead of the uniform \0x, y1, ]0x, y1, etc&1& 4ote& 7n a sense, ob>ect constants can be #ie ed as a special case of function constants 3 an ob>ect constant is a M:ero-ar!ument functionN& Predicate onstants The last 0but the most important;1 kind of primiti#es e need in our lan!ua!e are the so-called "redicate constants 0sometimes called predicate letters1 3 names or symbols denotin! specific "ro"erties 0of1 or relations bet een Sob>ectsS of our SdomainS& 4ote& *sin! SpredicateS as the unifyin! term for SpropertyS and SrelationS may seem some hat unusual& Gut some kind of such unifyin! term is necessary& %roperties are, in fact, unary 0i&e& one-ar!ument1 SpredicatesS, for example, Sx is redS& )elations are, t o- or more-ar!ument SpredicatesS, for example, Sx is better than yS, or Sx sends y to :S&
!7am"les& 11 7n our Slan!ua!e for peopleS the class dia!ram abo#e1: .a%e x! E means Sx is a maleS4 Fema%e x! e ill use the follo in! predicate constants 0see

E means Sx is a femaleS4 E means Sx is mother of yS4 E means Sx is father of yS4 E means Sx and y are married, y bein! ifeS4

.other x, y! Father x, y!

.arrie+ x, y!

x^y E means Sx an y are the same personS&

1.
The first t o constants represent, in fact, SpropertiesS 0or, SclassesS1 of our ob>ects& The other . constans represents SrelationsS bet een our ob>ects& The term SpredicateS is used to include both #ersions& 2e do not introduce Person x! as a predicate because our domains consists of persons only& 21 7t may seem stran!e to non-mathematicians, yet the most popular relation of ob>ects used in most mathematical theories, is e>uality 0or identity1& 'till, this is not stran!e for mathematicians& 2e can select an ob>ect x in our SdomainS by usin! a #ery specific combination of properties and relations of it, and then 3 select another ob>ect y 3 by usin! a different combination& 6nd after this 0sometimes it may take many years to do1 e pro#e that x^y, i&e& that these t o different combinations of properties and relations are possessed by a sin!le ob>ect& 5any of disco#eries in mathematics could be reduced to this form& 7n the lan!ua!e of first order arithmetic, eQuality S^S is the only necessary predicate constant& RtherSbasicS relations must be reduced to eQuality& 9or example, the relation x_y for natural numbers x, y can be reduced to eQuality by usin! the addition function and the formula :0x\:\1^y1& (1 7n the lan!ua!e of set theory a specific predicate constant SinS denotes the set membership relation: Sx in yS means Sx is a member of yS& The eQuality predicate x^y also ill be used 3 it means Sthe sets x an y possess the same membersS&

The uniform ay of representation suitable for predicates ha#in! any number of ar!uments is a!ain the SprefixS notation: p x!, 4 x, y!, r x, y, 3! etc& 7n the real mathematical practice, some of the t o-ar!ument predicates are represented by usin! the SinfixS notation 0for example, x^y instead of the uniform ^0x, y1, etc&1&
AeroBargument "redicate constants2 7n an interpretation, each such predicate must become either StrueS, or SfalseS& Hence, paradoxically, :ero-ar!ument predicate constants beha#e like Spropositional #ariablesS 3 they represent assertions that do not possess a meanin!, but possess a Struth #alueS&

,ummary of Language Primitives Thus, the specification of a predicate lan!ua!e includes the follo in! "rimitives: 11 6 countable set of ob>ect #ariable names 0you may !enerate these names, for example, by usin! a sin!le letter SxS: x, xx, xxx, xxxx, &&&1& 21 6n empty, finite, or countable set of ob>ect constants& (1 6n empty, finite, or countable set of function constants& To each function constant a fixed ar!ument number must be assi!ned& .1 6 finite, or countable set of predicate constants& To each predicate constant a fixed ar!ument number must be assi!ned& Different sets of primiti#es yield different predicate lan!ua!es&
!7am"les& 11 Rur Slan!ua!e for peopleS is based on: a1 ob>ect #ariables x, y, :, &&&4 b1 ob>ect constants: Hohn, Gritney, &&&4 c1 function constants: none4 d1 predicate constants: .a%e x!, Fema%e x!, .other x, y!, Father x, y!, .arrie+ x, y!, x0y&

1/
21 The lan!ua!e of first order arithmetic is based on: a1 ob>ect #ariables x, y, :, &&&4 b1 ob>ect constants: 0, 14 c1 function constants: x\y, x]y4 d1 predicate constant: x^y& (1 The lan!ua!e of set theory is based on: a1 ob>ect #ariables x, y, :, &&&4 b1 ob>ect constants: none4 c1 function constants: none4 d1 predicate constants: x in y, x^y&

The remainin! part of the lan!ua!e definition is common for all predicate lan!ua!es& 9erms and formulas Gy usin! the lan!ua!e primiti#es, 0compound1 formulas& e can build terms, atomic formulas and

9erms are expressions used to denote ob>ects and functions: a1 Rb>ect #ariables and ob>ect constants 0if any1, are terms& b1 7f f is a k-ar!ument function constant, and t 1, &&&, tk are terms, then the strin! f0t1, &&&, tk1 is a term& c1 There are no other terms&
!7am"les& 11 7n our Slan!ua!e for peopleS only #ariables x, y, :, &&&, and ob>ect constants /ohn, Britney, &&& are terms& 21 7n the lan!ua!e of first order arithmetic, for addition and multiplication the SinfixS notation is used: if t1, t2 are terms, then 0t1\t21 and 0t1]t21 are terms& Rf course, the ob>ect constants 0, 1 and #ariables x, y, :, &&& are terms& @xamples of more complicated terms: 0x\y1, 001\11]01\111, 0001\11]x1\11& (1 7n the lan!ua!e of set theory, #ariables x, y, :, &&& are the only kind of terms&

7f a term does not contain #ariable names, then it denotes an Sob>ectS of our SdomainS 0for example, 001\11\11 denotes a specific natural number 3 the number (1& 7f a term contains #ariables, then it denotes a function& 9or example, 000x]x1\0y]y11\11 denotes the function x 2\y2\1& 01arning6 ?ote that the lan!ua!e of first order arithmetic does not contain a function constant denotin! the e7"onentiation xy, thus, for example, e must rite x]x instead of x2&1 Rf course, the key element of our efforts in describin! Sob>ectsS, their properties and relations, ill be assertions, for example, the commutati#e la in arithmetic: 00x\y1^0y\x11& 7n predicate lan!ua!es, assertions are called formulas 0or, sometimes, ell formed formulas 3 ff-s, or sentences1& 3tomic formulas 0in some other textbooks: elementary formulas, prime formulas1 are defined as follo s: a1 7f p is a k-ar!ument predicate constant, and t1, &&&, tk are terms, then the strin! p0t1, &&&, tk1 is an atomic formula&

1b1 There are no other atomic formulas& 9or the eQuality symbol, the SinfixS notation is used: if t1, t2 are terms, then 0t1^t21 is an atomic formula&
!7am"les& 11 7n our Slan!ua!e for peopleS, the follo in! are examples of atomic formulas: .a%e x!, Fema%e Britney!, .a%e Britney! 0not all formulas that are ell formed, must be true;1, Father x, Britney!, .other Britney, /ohn!, .arrie+ x, y!( 21 'ummary of the atomic formulas of the lan!ua!e of first order arithmetic: a1 constants 0 and 1, and all #ariables are terms4 b1 if t 1 and t2 are terms, then 0t1\t21 and 0t1]t21 also are terms4 c1 atomic formulas are built only as 0t1^t21, here t1 and t2 are terms& (1 7n the lan!ua!e of set theory, there are only t o kinds of atomic formulas: x^y 0 here x and y are arbitrary #ariables1&

x y , and

7n the lan!ua!e of first order arithmetic, e#en by usin! the only a#ailable predicate constant S^S atomic formulas can express a lot of cle#er thin!s: 00x\01^x14 00x\y1^0y\x114 00x\0y\:11^00x\y1\:114 00x]01^014 00x]11^x14 00x]y1^0y]x114 00x]0y]:11^00x]y1]:114 000x\y1]:1^00x]:1\0y]:111& !7ercise '...'. 6s the next step, translate the follo in! assertions into the lan!ua!e of first order arithmetic 0do not use abbre#iations;1: 2]2^., 2]2^/, 0x\y12 ^ x2\2xy\y2& C om"oundD :ormulas 9he follo8ing definition is common for all "redicate languages. !ach language is s"ecific only by its set of language "rimitives. To rite more complicated assertions, e ill need compound formulas, built of atomic formulas by usin! a fixed set of "ro"ositional connectives and >uantifiers 0an in#ention due to A& 9re!e and <& '& %eirce1& 7n this book, e ill use the follo in! set: 7mplication symbol: G`< means Sif G, then <S, or SG implies <S, or S< follo s from GS& <on>unction symbol, B C means SG and <S& Dis>unction symbol, B C means SG, or <, or bothS, i&e& the so-called nonexclusi#e SorS& ?e!ation symbol, B means Snot G& *ni#ersal Quantifier, x B means Sfor all x: GS& @xistential Quantifier, x B means Sthere is x such that GS& The idely used eQui#alence connecti#e a can be deri#ed in the follo in!

1= ay: Ga< stands for (( B 5C )( C 5 B )) & 7f you like usin! the so-called exclusi#e SorS 0MG, or <, but not bothN1, you could define G xor < as ( B 6 C ) & 1arning6 9or pro!rammers, con>unction, dis>unction and ne!ation are familiar Slo!ical operationsS 3 unlike the implication that is not used in SnormalS pro!rammin! lan!ua!es& 7n pro!rammin!, the so-called 79statements, hen compared to lo!ic, mean a different thin!: in the statement 79 x^y TH@? ::^1=, the condition, x^y is, indeed, a formula, but the SconseQuenceS ::^1= is not a formula 3 it is an executable statement& 7n lo!ic, in G`< 0Sif G, then <S1, G and < both are formulas& 2e define the notion of formula of our predicate lan!ua!e as follo s: a1 6tomic formulas are formulas& b1 7f G and < are formulas, then ( B 5 C ) , ( B C ) , ( B C ) , and ( B ) also are formulas 0G and < are called sub-formulas1& c1 7f G is a formula, and x is an ob>ect #ariable, then 0xG1 and 0xG1 also are formulas 0G is called a sub-formula1& d1 07f you like so,1 there are no other formulas&
'ee also: ?otes on +o!ic ?otation on the 2eb by Peter ,uber&

Kno8ledge #e"resentation by Means of Predicate Languages


!7am"les& 11 7n our Slan!ua!e for peopleS, the follo in! are examples of compound formulas:
(( Father ( x , y ))( .other ( x , y ))) ( x ( y (( Father ( x , y )) 5 ( .a%e ( x ))))) ( x ( y (( .other ( x , y )) 5 ( 7 .a%e ( x ))))) ( x ( y ( .other ( y , x )))) ( x ( .a%e ( x ) Fema%e ( x )))

Sx is a parent of yS Sfathers are malesS 3 could ser#e as an a7iom; Smothers are not malesS 3 could be derived from the axioms; Seach x has some y as a motherS 3 could ser#e as an a7iom; 2hat does it mean8 7t could ser#e as an a7iom;

x ( y ( 3 (( .other ( x , 3 ) .other ( y , 3 )) 5 ( x = y ))))


x ( y ( 3 (( Father ( x , 3 ) Father ( y , 3 )) 5 ( x = y ))))

2hat does it mean8 7t could ser#e as an a7iom;

1C

21 'ome simple examples of compound formulas in the lan!ua!e of first order arithmetic: 1arning6 'peakin! strictly, predicate symbols S_S, SbS, S cS, SdS, SeS etc& do not belon! to the lan!ua!e of first order arithmetic& 0u0x^0u\u111 0u000x\u1\11^y11 ( 0< y $ ( x =( y$))) Sx is an e#en numberS Sx is less than yS, or, x_y Sx is di#isible by yS, speakin! strictly, x_y must be replaced by u000x\u1 \11^y11&

(( 1< x )(( y ( 3 ((( y < x )( 3 < x ))( x=( y3 ))))))) formula prime x!, Sx is a prime numberS, speakin! strictly, the ( subformulas of the kind x_y should be replaced by their full #ersion of the kind u000x\u1\11^y11& SThere are infinitely many prime numbersS 0one of the first mathematical theorems, V7 century G<1, speakin! strictly, _x must be replaced by u000 \u1\11^x11, and prime x! must be replaced by the abo#e formula&

( - ( x (( - < x )( prime ( x )))))

x y ( 0 < y 5 3 $ ( $ < y x = y3 +$)) 2hat does it mean8

(1 'ome simple examples of compound formulas in the lan!ua!e of set theory:


( y ( y x )) ( 3 (( 3 x ) 5 ( 3 y ))) (( 3 (( 3 x ) 6 ( 3 y ))) 5 ( x = y ))

Sx is a non-empty setS Sx is a subset of yS, or xcy 2hat does it mean8 2ill ser#e as an a7iom; Sx contains :ero or one memberS Sx is union of y and :S, or x^y*:

( y ( 3 (( y x)( 3 x )) 5 y = 3 )) ( $ (( $ x ) 6 (( $ y )( $ 3 ))))

Rf course, having a "redicate language is not enough for expressin! all of our kno led!e formally, i&e& for communicatin! it to computers& <omputers do not kno in ad#ance, for example, ho to handle sexes& 2e must tell them ho to handle these notions by introducin! a7ioms& Thus, the abo#ementioned formulas like as ( x ( .a%e ( x ) Fema%e ( x ))) , or ( x ( y (( Father ( x , y )) 5 ( .a%e ( x )))))

1, ill be absolutely necessary as a7ioms& 6s e ill see later, lan!ua!e \ axioms \ lo!ic ^ theory, i&e& in fact, to formulate all of our kno8ledge formally, 8e must create theories. !7ercise '..... Translate the follo in! assertions into our Slan!ua!e for peopleS: Sx is child of yS4 Sx is !rand-father of yS4 Sx is brother of xN4 Mx is sister of yS4 !7ercise '...). Translate the follo in! assertions into the lan!ua!e of first order arithmetic: Sx and y do not ha#e common di#isorsS 0note: 1 is not a di#isor;14 S 2 is not a rational numberS& p 02arnin!; p 4( 2 = ) , and x ( x x = 2) are not correct solutions& 2hy81
4

1arning6 ,ome ty"ical errors6 11 Tryin! to say Mfor all xb2, 90x1N, do not rite x ( x > 2 F ( x )) E this ould imply that x x89!; Gut, ho about x^18 The correct #ersion: x x89FF x!!( 21 'ome computer pro!rammers do not like usin! the implication connecti#e `, tryin! to rite formulas as conditions of 79- or 2H7+@-statements, i&e& by usin! con>unction, dis>unction and ne!ation only& This SapproachS makes most lo!ical tasks much harder than they really are; 5ore than that E some people try sayin!, for example, S%ersons are not DepartmentsS, as follo s: x ( Person ( x )7 ,epartment ( x )) E instead of the correct #ersion: x ( Person ( x ) 5 ,epartment ( x )) ( (1 Do not use abbre#iations at this early sta!e of your studies& 9or example, do not rite x89!yF x, y! to say that Sfor all x that are b2, there is y such that 90x, y1S& 7nstead, you should rite x x895yF x, y!!& 'imilarly, instead of 0a89!bG a, b!, you should rite a ( a >2 b G ( a , b )) ( To say Sthere is one and only one x such that 90x1S, you shouldnBt F x! 0at this sta!e of your studies1, you should rite, instead, rite :x

(( xF ( x ))( x; ( x9 (( F ( x; ) F ( x9 )) 5 ( x; = x9 ))))) ( .1 %redicates cannot be substituted for ob>ect #ariables& 9or example, ha#in! ( predicate constants -or&ing x, y!, Person x!, ,epartment y!, do not try ritin! -or&ing Person x!, ,epartment y!! to say that Sonly persons are orkin!, and

20 only in departments<( The correct #erssion: x y ( -or&ing ( x , y ) 5 Person ( x ) ,epartment ( y )) ( /1 Tryin! to say Meach person is orkin! in some departmentN, do not rite x y ( Person ( x ) ,epartment ( y ) 5 -or&ing ( x , y )) & The correct #ersion: x ( Person ( x ) 5 y ( ,epartment ( y )-or&ing ( x , y ))) & 2hat is the difference8 !7ercise '...*& Try in#entin! your o n predicate lan!ua!e& %repare and do your o n @xercise 1&2&2 for it& !7ercise '...-& 7n computer science, currently, one the most popular means of kno led!e representation are the so-called UML class diagrams and ? L 0*5+ E *nified 5odelin! +an!ua!e, R<+ E Rb>ect <onstraint +an!ua!e1& The abo#e dia!ram representin! our Mdomain of peopleN is an example& 7n our Mlan!ua!e of peopleN, put do n as many a7ioms of the domain you can notice in the dia!ram& 9or example, Me#ery person is either male, or femaleN, Mall fathers are malesN, Me#ery person has exactly one motherN, Ma person can marry no more than one personN etc& ManyBsorted Languages
5aybe, you ha#e to describe t o or more kinds of Sob>ectsS that you do not ish to reduce to Ssub-kindsS of one kind of Sob>ectsS 0for example, inte!er numbers and character strin!s1& Then you may need introducin! for each of your SdomainsS a separate kind 0SsortS1 of ob>ect #ariables& 7n this ay you arri#e to the so-called manyBsorted "redicate languages& 7n such lan!ua!es: a1 each ob>ect constant must be assi!ned to some sort4 b1 for each function constant, each ar!ument must be assi!ned to some some sort, and function #alues must be assi!ned to a 0sin!le1 sort4 c1 for each predicate constant, each ar!ument must be assi!ned to some sort& 7n many-sorted predicate lan!ua!es, the term and atomic formula definitons are some hat more complicated: buildin! of the term f0t 1, &&&, tk1 or the formula p0t1, &&&, tk1 is allo ed only, if the sort of the term t i #alues coincides ith the sort of the i-th ar!ument of f or p respecti#ely& 6nd the Smeanin!S of Quantifiers depends on the sort of the #ariable used ith them& 9or example, x means Sfor all #alues of x from the domain of the sort of xS& Theoretically, many-sorted lan!ua!es can be reduced to one-sorted lan!ua!es by introdudin! the correspondin! predicates 'orti0x1 0Sthe #alue of x belon!s to the sort iS1& 'till, in applications of lo!ic 0for example, in computer science1 the many-sorted approach is usually more natural and more con#enient& 0'ee Chapter ;=( .any>Sorte+ First *r+er Logic, by Hean Aallier&1

1arning about second order languages6


7n our definition of predicate lan!ua!es only the follo in! kinds of primiti#es ere used: ob>ect #ariables, ob>ect constants, function constants and predicate constants& Kou may ask: ho about function variables and "redicate variables8 9or, you may ish to denote by r San

21
arbitrary propertyS of your Sob>ectsS& Then, r0x1 ould mean Sx possess the property rS, and you ould be able to say somethin! about Sall propertiesS, for example, r xy0x^y`0r0x1ar0y11& 7n this ay you ould ha#e arri#ed at a second order language; 7n such lan!ua!es, function and predicate #ariables are allo ed& Gut properties lead to sets of ob>ects, for example, Xx U r0x1Y ould mean the set of all ob>ects that possess the property r& Gut, hy should e stop at the properties of ob>ects8 Ho about Sproperties of sets of ob>ectsS etc&8 6s it as detected lon! a!o, all kinds of sets can be fully treated only in set theory; Thus, instead of buildin! your o n second order lan!ua!e, you should better try applyin! your fa#orite 0Sfirst orderS1 set theory& 6n unpleasant conseQuence: the existence of the 0much less si!nificant1 notion of second order lan!ua!es forces many people to call predicate lan!ua!es Sfirst order languagesS E to emphasi:e that, in these lan!ua!es, the only kind of #ariables allo ed are ob>ect #ariables& Rn the other hand, hen tryin! to implement realistic formal reasoning soft8are, then usin! of some second order constructs is, as a rule, more efficient than implementin! of a pure first order reasonin!& 'ee, for example, "otices of the A.S, 'pecial 7ssue on 9ormal %roof, Vol& //, ? 11, 200C 0a#ailable online1& 9or details, see: 'econd-order-lo!ic& 6bout second order arithmetic see )e#erse 5athematics& 6bout an almost 0but not 100T1 successful attempt to create a set theory Sas simple as lo!icS 0by Aeor! <antor and Aottlob 9re!e1 3 see %odnieks V1,,=W, 'ection 2&2&

?mitting Parentheses Rur formal definitions of terms and formulas lead to expressions containin! many parentheses& +et us recall, for example, our formula expressin! that Sx is a prime numberS: (( 1 < x )(( y ( 3 ((( y < x )( 3 < x ))( x =( y 3 ))))))) & 'uch formulas are an easy readin! for computers, yet incon#enient for human readin! 0and e#en more incon#enient 3 for puttin! them correctly1& 7n the usual mathematical practice 0and in pro!rammin! lan!ua!es1 e are allo ed to impro#e the look of our formulas by omittin! some of the parentheses E accordin! to 0some of1 the follo in! rules: a1 Rmit the outermost parentheses, for example, e may rite 6`0G`<1 instead of the formally correct 06`0G`<11& 7n this ay e may impro#e the final look of our formulas& 'till, if e ish to use such formulas as parts of more complicated formulas, e must restore the outermost parentheses, for example: 06`0G`<11`D& b1 2e may rite, for example, simply: x + y + 3 + $ , x y3 $ , A BC , , A BC , , x y 3 $ F instead of the more formal 00x\y1\:1\u, 00x]y1]:1]u, (( A B )C ) , , (( A B )C ) , , x0y0:0u091111&

22 7n this ay follo s: e can simplify the abo#e expression Sx is a prime numberS as ( 1< x )( 7( y 3 (( y < x )( 3 < x )( x =( y 3 ))))) & c1 2e can apply the so-called "riority rules& 9or example, the priority rank of multiplications is supposed to be hi!her than the priority rank of additions& This rule allo s ritin! x\y]: instead of the more formal x\0y]:1 E because of its hi!her priority rank, multiplication must be Sperformed firstS& The most popular priority rules are the follo in!: c11 The priority rank of function constants is hi!her than the priority rank of "redicate constants& This allo s, for example, ritin! x]y ^ y]x instead of 0x]y1^0y]x1, or x y 3 E instead of x ( y 3 ) & c21 The priority rank of "redicate constants is hi!her than the priority rank of "ro"ositional connectives and >uantifiers& This allo s, for example, ritin! y < x 3 < x instead of ( y < x )( 3 < x ) & c(1 The priority rank of >uantifiers is hi!her than the priority rank of "ro"ositional connectives& This allo s, for example, ritin! x F y G instead of ( x ( F ))( y ( G )) , or ritin! x F instead of ( x ( F )) & c.1 The priority rank of negations is hi!her than the priority rank of con@unctions and dis@unctions& This allo s, for example, ritin! A B instead of ( 7 A)( 7 B ) & c/1 The priority rank of con@unctions and dis@unctions is hi!her than the priority rank of im"lications& This allo s, for example, ritin! A 5 A B instead of A 5 ( A B ) & 7n the usual mathematical practice some additional priority rules are used, but some of them are not allo ed in the common pro!rammin! lan!ua!es& To a#oid confusions do not use too many priority rules simultaneously; 6ccordin! to the abo#e priority rules, e can simplify the abo#e expression Sx is a prime numberS obtainin! a form that is much easier for human readin! 0but is some hat complicated for computers to process it1: 1< x 7 y 3 ( y < x 3 < x x = y 3 ) & 6s you see, all the abo#e rules are mere abbre#iations& 7n principle, you could use any other set of abbre#iation rules accepted by your audience& 7f computers ould do lo!ic themsel#es, they ould not need such rules at all 0except, maybe, for displayin! some of their results to humans, but hy81& !7ercise '.../. STranslateS the follo in! assertions into our Slan!ua!e for peopleS:

2( Sx and y are siblin!sS4 Sx and y are brothersS4 Mx and y are sistersN4 Mx is cousin of yN4 Mparents of x and y are marriedN4 construct formulas expressin! as much ell-kno n relationships bet een people as you can& Gut ho about the predicate 3ncestorC7, yD E Sx is an ancestor of yS8 <ould it be expressed as a formula of our Slan!ua!e for peopleS8 The first idea E let us SdefineS this predicate recursi#ely: x y ( Father ( x , y ) .other ( x , y ) 5 Ancestor ( x , y )) 4 x y 3 ( Ancestor ( x , y ) Ancestor ( y , 3 ) 5 Ancestor ( x , 3 )) & The second rule declares the transiti#ity property of the predicate& The abo#e t o formulas are a7ioms, allo in! to deri#e the essential properties of the predicate Ancestor x, y!& Gut ho about a sin!le formula 90x, y1 in the Slan!ua!e for peopleS, expressin! that Sx is an ancestor of yS8 'uch a formula should be a tricky combination of formulas Father x, y!, .other x, y! and x0y& 6nd such a formula is im"ossible; 'ee arlos 3reces& %h&D& Thesis, 2000, 0a non-tri#ial;1 Theorem 1&2& !7ercise '...0. STranslateS the follo in! assertions into the lan!ua!e of first order arithmetic: Sx and y are t in primesS 0examples of t in pairs: (,/4 /,=4 11,1(4 1=,1,4&&&1, SThere are infinitely many pairs of t in primesS 0the famous T in %rime <on>ecture1, Sx is a po er of 2S 02arnin!; n0x^2n1 is not a correct solution& 2hy8 Gecause exponentiation does not belon! to the lan!ua!e of first order arithmetic&1, S@ach positi#e e#en inte!er d. can be expressed as a sum of t o primesS 0the famous Aoldbach <on>ecture1& :ree Variables and %ound Variables The abo#e expression Sx is a prime numberS: 1< x y 3 ( y < x 3 < x x = y 3 ) contains ( #ariables: x E occurs . times in terms, y E 2 times in terms and 1 time in Quantifiers, : E occurs 2 times in terms and 1 time in Quantifiers& Rf course, x is here a SfreeS #ariable 3 in the sense that the Struth #alueS of the formula depends on particular S#aluesS taken by x& Rn the contrary, the Struth #alueS of the formula does not depend on the particular S#aluesS taken by the t o SboundS #ariables y and : E the Quantifiers y, : force these #ariables to

2. Srun across their entire ran!eS& 5ore precisely, first, e ill count only the occurrences of #ariables in terms, not in Quantifiers& 6nd second, e ill define a particular occurrence o x of a #ariable x in 0a term of1 a formula 9 as a free occurrence or a bound occurrence accordin! to the follo in! rules: a1 7f 9 does not contain Quantifiers x, x, then ox is free in 9& b1 7f 9 is xA or xA, then ox is bound in 9& c11 7f 9 is G H , G H , or A`H, and ox is free in A 0or in H1, then ox is free in 9& c21 7f 9 is fA, yA, or yA, free in 9& here y is not x, and ox is free in A, then ox is

d11 7f 9 is G H , G H , or A`H, and ox is bound in A 0or in H1, then ox is bound in 9& d21 7f 9 is fA, yA, or yA 0 here y is any #ariable, x included1, and o x is bound in A, then ox is bound in 9& Thus, the abo#e formula 1< x y 3 ( y < x 3 < x x = y 3 ) contains . free occurrences of x, 2 bound occurrences of y, and 2 bound occurrences of :& !7ercise '...G& Verify that an occurrence of x in 9 cannot be free and bound simultaneously& 0Hint: assume that it is not the case, and consider the seQuence of all sub-formulas of 9 containin! this particular occurrence of x&1 9ormally, e can use formulas containin! free and bound occurrences of a sin!le #ariable simultaneously, for example, x >1 5 x ( x >1) & Rr, many bound occurrences of a sin!le #ariable, for example, ( xF ( x ) xG ( x )) xH ( x ) means the same as ( xF ( x ) yG ( y )) 3H ( 3 ) & 'till, e do not recommend usin! a sin!le #ariable in many different roles in a sin!le formula& 'uch formulas do not cause problems for computers, but they may become incon#enient for human readin!& +et us say, that x is a free variable of the formula 9, if and only if 9 contains at least one free occurrence of x, or 9 does not contain occurrences of x at all& 7f a formula contains free #ariables, i&e& #ariables that are not bound by Quantifiers 0for example: x = 0 x =1 1, then the Struth #alueS of such formulas may depend on particular #alues assi!ned to free #ariables& 9or

2/ example, the latter formula is StrueS for x^1, yet it is SfalseS for x^2& 9ormulas that do not contain free occurrences of #ariables, are called closed formulas, for example: - x ( - < x prime ( x )) & <losed formulas represent Sdefinite assertions about ob>ects of theoryS, they are expected to be 0but not al ays really are1 either StrueS, or SfalseS& 9erm ,ubstitution To say that x is a free #ariable of the formula 9, e may ish to rite 90x1 instead of simply 9& )eplacin! all free occurrences of x by a term t yields an SinstanceS of the formula 9& 7t ould be natural to denote this SinstanceS by 90t1& 9or example, if 90x1 is y0y\y^x1 and t is :]:\:, then 90t1, or 90:]:\:1 denote y0y\y^:]:\:1& Ho e#er, if t ould be y]y\y, then 90t1, or 90y]y\y1 y0y\y^y]y\y1& 7s this really 90y]y\y18 ill

ould be

Thus, sometimes, substitutions can lead to cra:y results& 6nother example: in our expression Sx is a prime numberS, let us replace x by y& 2ill the resultin! formula mean Sy is a prime numberS8 +etBs see: 1< y y 3 ( y < y 3 < y y = y 3 ) & 'ince y_y is al ays false, the second part fy:0&&&1 is true, hence, the latter formula means simply that S1 is less than yS, and not that Sy is a prime numberS& Rf course, e failed because e replaced a free #ariable x by a #ariable y in such a ay that some free occurrence of x became bound by a Quantifiers for y 0y1& 7n this ay e deformed the initial meaning of our formula& The follo in! simple rule allo s to a#oid such situations& 'uppose, x is a free #ariable of the formula 9& 2e ill say that the substitution :C7HtD 0i&e& the substitution of the term t for x in the formula 91 is admissible, if and only if no free occurrences of x in 9 are located under Quantifiers that bind #ariables contained in t& 7f the substitution 90xPt1 is admissible, then, by replacin! all free occurrences of x in 9 by t, of course, e do not chan!e the initial meanin! of the formula 90x1, and hence, e may safely denote the result of this substitution by 90t1& !7ercise '...(. 7s xPy an admissible substitution in the follo in! formulas8 2hy8 x = 0 y ( y > 3 ) 4

2x =0 y ( y > x ) & !7ercise '...'I. a1 5athematicians: think o#er the analo!y bet een bound #ariables in lo!ic and bound #ariables in sum expressions and inte!rals& b1 %ro!rammers: think o#er the analo!y bet een bound #ariables in lo!ic and loop counters in pro!rams&

'.). 37ioms of Logic= Minimal ,ystem, onstructive ,ystem and lassical ,ystem
9he Problem of #easoning ?o e !o on to the second phase of formali:ation: after ha#in! defined a forma% %ang$age 0predicate lan!ua!e1 allo in! to put do n propositions about our domain of interest, and ha#in! formulated some of the propositions as axioms, and must introduce some means of reasoning allo in! to deri#e other statements that are MtrueN of our domain& 7ndeed, ha#in! formulated some fra!ment of our kno led!e as a set of axioms 61, &&&, 6n in some predicate lan!ua!e +, e do not think that 61, &&&, 6n represent all statements that are MtrueN of the ob>ects e are tryin! to in#esti!ate& 5any other statements ill follo8 from 61, &&&, 6n as conseQuences& !7am"le& 6ssume, e ha#e formulated the follo in! axioms in our Slan!ua!e for peopleS: x ( .a%e ( x ) Fema%e ( x )) 4 x ( .a%e ( x ) Fema%e ( x )) , and the follo in! facts: .a%e ( /ohn ) ? Fema%e ( Britney ) & Then e do not need to formulate 7 Fema%e ( /ohn ) ? 7 .a%e ( Britney ) as separate facts& These facts can be deri#ed from the already re!istered facts& The problem of reasonin!: Sformula 9 follo s from 6 1, &&&, 6nS, hat does it mean8 The ans er must be absolutely precise, if e ish to teach reasoning to com"uters& ,olution of the Problem 9irst of all, let us notice that there are axioms and rules of inference that are applicable to any predicate lan!ua!es, independently of the specific features of their domains& 'uch axioms and rules could be called M!enerally #alidN& 9or example, assume, some formula 9 has the follo in! form: ( B 5 , ) 5 (( C 5 , ) 5 ( BC 5 , )) ,

2= here G, <, D are some formulas& Then 9 is MtrueN independently of the specific facts represented in the formulas G, <, D& 'imilarly, the follo in! rule of inference 5% is applicable independently of the facts represented in the formulas G, <: Ha#in! deri#ed the formulas G, G`<, deri#e the formula <& 7f e ha#e G`D and G`D already deri#ed, then 3 by applyin! the rule 5% to the abo#e formula 9 3 e deri#e that B C 5 , & ?o , e ill try formulatin! a complete set of S!enerally #alidS principles 0axioms and rules of inference1 of Slo!ically correct reasonin!S& SAenerally #alidS means that these principles ill be applicable to any predicate lan!ua!e& 7&e& for any fixed predicate lan!ua!e +, e ish to formulate a uniform list of logical a7ioms and inference rules that ould allo formali:ation of principles of reasonin! that are S#alidN for all lan!ua!es& 'uch principles are called sometimes Spure lo!icalS principles& The existence of such !eneral principles 0and e#en, in a sense, a complete system of them1 is the result of a 2/00 year lon! history of !reat disco#eries 0or in#entions8 E see belo 1& 6ristotle 0(C.-(22 G<1, Aottlob 9re!e 01C.C-1,2/1, <harles 'anders %eirce 01C(,-1,1.1& Gertrand )ussell 01C=2-1,=01 E SThe Princi"ia Mathematica is a three-#olume ork
on the foundations of mathematics, ritten by Gertrand )ussell and 6lfred ?orth 2hitehead and published in 1,10-1,1(& 7t is an attempt to deri#e all mathematical truths from a elldefined set of axioms and inference rules in symbolic lo!ic& The main inspiration and moti#ation for the %rincipia as 9re!eBs earlier ork on lo!ic, hich had led to some contradictions disco#ered by )ussell&S 0accordin! to 2ikipedia1&

Da#id Hilbert 01C-2-1,.(1, 2ilhelm 6ckermann 01C,--1,-21&


D.$ilbert, 1.3ckermann& Arund:g!e der theoretischen +o!ik& Gerlin 0'prin!er1, 1,2C 0see also: Hilbert and 6ckermannBs 1,2C +o!ic Gook by 'tanley ?& Gurris1&

The first #ersion of lo!ical axioms as introduced in 1C=, by A& 9re!e in his abo#e-mentioned Begriffsschrift& The next important #ersion as proposed in 1,10-1,1( by G& )ussell and 6& 2hitehead in their famous book Principia .athematica& 6nd finally, in 1,2C D& Hilbert and 2& 6ckermann published in their abo#e-mentioned book, in a sense, the final #ersion of lo!ical axioms& 5odifications of this #ersion are no used in all textbooks of mathematical lo!ic& 7n our #ersion, the axioms ill be represented by means of the so-called a7iom schemas 0pro!rammers mi!ht call them temp%ates1& @ach schema 0template1

2C represents an infinite, yet easily reco!ni:able collection of sin!le axioms& 9or example, schema +(: B C 5 B may represent the follo in! axioms 0Sinstances of the schemaS1 in the lan!ua!e of first order arithmetic: x= y x= x 5 x= y , 11 =11+1=1 +1 5 11=1 , and many other axioms: take any formulas G, <, and you ill obtain an axiom B C 5 B & 2e ill not specify properties of the e>uivalence connecti#e in axioms& 2e ill re!ard this connecti#e as a deri#ed one: Ga< ill be used as an abbre#iation of ( B 5 C )(C 5 B ) & 37ioms of Logic 'uppose, e ha#e specified some predicate lan!ua!e +& 2e adopt the follo in! 1/ axiom schemas as the logical a7ioms for the language L. 7n the axiom schemas +1-+11 belo , G, < and D are any formulas in the lan!ua!e +& The first t o axiom schemas +1, +2 represent the SdefinitionS of the implication connecti#e: L': B 5 ( C 5 B ) 0 hat does it mean81,

L.: ( B 5 ( C 5 , )) 5 (( B 5 C ) 5 ( B 5 ,)) 0 hat does it mean81& The follo in! axiom schemas +(3+/ represent the SdefinitionS of the 6?Dconnecti#e 0con>unction1: L): L*: L-:

B C 5 B 0 hat does it mean81,


B C 5 C 0 hat does it mean81, B 5 ( C 5 B C ) 0 hat does it mean81&

The follo in! axiom schemas +-3+C represent the SdefinitionS of the 0nonexclusi#e1 R)-connecti#e 0dis>unction1: L/:

B 5 BC 0 hat does it mean81,

L0: C 5 BC 0 hat does it mean81, LG: ( B 5 , ) 5 (( C 5 , ) 5 ( BC 5 , )) 0 hat does it mean81&

2, The next xiom schema +, represents the SdefinitionS of the ?R-connecti#e& 7n fact, it is a formal #ersion of a proof method ell-kno n in mathematics E refutation by deri#in! a contradiction 0)e+$ctio a+ abs$r+$m1: L(: ( B 5 C ) 5 (( B 5 C ) 5 B ) 0 hat does it mean81& The next axiom schema +10 represents the famous principle S<ontradiction 7mplies 6nythin!S 0Ex contra+ictione se4$it$r 4$o+%ibet, or Ex fa%so se4$it$r 4$o+%ibet1: L'I: B 5 ( B 5 C ) 0 hat does it mean81& The follo in! axiom schema +11 represents the famous La8 of !7cluded Middle 0Terti$m non +at$r!: L'': B B 0 hat does it mean81&

The abo#e 11 schemas 0plus the .o+$s Ponens rule of inference, see belo 1 represent the classical "ro"ositional logic in the lan!ua!e +& ?o , the SdefinitionsS of the uni#ersal and existential Quantifiers follo & 7n the follo in! axiom schemas +12, +1(, 9 is any formula, and t is a term such that the substitution 90xPt1 is admissible 0in particular, t may be x itself1: L'.: x F ( x ) 5 F ( t ) mean81, L'): F ( t ) 5 x F ( x ) mean81& 0in particular, x F ( x ) 5 F ( x ) , 0in particular, F ( x ) 5 x F ( x ) , hat does it hat does it

7n the follo in! schemas +1., +1/, 9 is any formula, and A is a formula that does not contain x as a free #ariable: L'*: x ( G 5 F ( x )) 5 ( G 5 x F ( x )) 0 hat does it mean81, L'-: x ( F ( x ) 5 G) 5 ( x F ( x ) 5 G ) 0 hat does it mean81& #ules of Inference 7n the follo in! rules of inference, G, < and 9 are any formulas& Modus Ponens: G`<4 G U- < 0 hat does it mean81& Generalization: 90x1 U- x F ( x ) 0 hat does it mean81& This list of lo!ical axioms and rules of inference represents the so-called classical "redicate logic in the predicate lan!ua!e + 0or, simply E the

(0 classical logic in the lan!ua!e +1& ,ome of the logical a7ioms are <8rong, but useful<6 The axioms +1, +2 represent the 0currently1 most popular #ersion of Sdefinin!S the implication connecti#e& 6bout other 0eQui#alent1 #ersions E containin! ( or . axioms E see Hilbert, Gernays V1,(.W 0<hapter 7771 and @xercise 1&/&2& The axiom +, represents the 0currently1 most popular #ersion of Sdefinin!S the ne!ation connecti#e& 6bout other 0eQui#alent1 #ersions E see Hilbert, Gernays V1,(.W 0<hapter 7771, @xercise 2&.&2& Three of the abo#e axiom schemas seem to be 0at least partly1 problematic& 9or example, ho do you find the funny axiom +10: B 5 ( B 5 C ) 8 7f fG and G ere true simultaneously, then anythin! ere true8 Ex contra+ictione se4$it$r 4$o+%ibet@ 7s this a really StrueS axiom8 Rf course, it is not& 'till, this does not matter: e do not need to kno , ere < StrueS or not, if fG and G ere StrueS simultaneously& Gy assumin! that Sif fG and G ere true simultaneously, then anythin! ere trueS 8e greatly sim"lify our logical a""aratus& 9or example, e ill pro#e in 'ection 2&- that, in the classical lo!ic, ffG`G& This simple formula canBt be pro#ed ithout the Scra:yS axiom +10 0see 'ection 2&C1& 7n fact, the first axiom + 1: G`0<`G1 also is funny& 7f G is 0unconditionally1 true, then G follo s from <, e#en if < has nothin! in common ith G8 5oreo#er, in @xercise 1&.&20d1 e ill see that the axioms +1, +, allo pro#in! that fG, G U- f<, i&e& if fG and G ere true simultaneously, then anythin! ere false 0thus, in a sense, + 1 contains already /0T of + 10;1& 6fter this, could e think of +1 as a really StrueS axiom8 Rf course, e canBt& 'till, this does not matter: if G is 0unconditionally1 true, then e do not need to kno , follo s G from < or not& Gy assumin! that Sif G is true, then G follo s from anythin!S 8e greatly sim"lify our logical a""aratus& The abo#e t o phenomena are called "arado7es of the material im"lication, see %aradoxes of 5aterial 7mplication by %eter 'uber, and 9alsity 7mplies 6nythin! by 6lexander Go!omolny& 5ay our decision to S!reatly simplifyS the lo!ical apparatus ha#e also some undesirable conseQuences8 +et us consider the follo in! formula 90x1: y0child0x, y1`9emale0y11& 7t seems, 90x1 is intended to mean: S6ll the children of x are femaleS& Ho e#er, in our system of lo!ic, 90x1 is re!arded as true also, if x does not ha#e children at all; 7f you do not ha#e children at all, then all your children are female; Rr male8 Rr smart8 @tc& 'eems funny, but

(1 is, in fact, harmless&&& onstructive Logic 'till, the most serious problem is caused by the axiom + 11: G#fG E the +a of @xcluded 5iddle& Ho can e think of +11 as a StrueS axiom, if 0accordin! to AFdelBs 7ncompleteness Theorem1 each sufficiently stron! consistent theory contains undecidable propositions8 7&e& e postulate that either G, or fG Smust be trueS, yet for some G e cannot pro#e neither G, nor fG; $no in! that G#fG is StrueS may inspire us to ork on the problem, but it may appear useless, if e do not succeed&&& 'hould e retain +11 as an axiom after this8 'ome other strange conse>uences of +11 also should be mentioned 0see @xercise 2&-&.1: B ( B 5 C ) , ( B 5 C )(C 5 B ) , (( B 5C ) 5 B ) 5 B 0the so-called PeirceJs La81& 9or these 0and some other1 reasons some people re>ect + 11 as a S#alidS lo!ical axiom& The abo#e list of 1/ axiom schemas as it stands is called the classical logic& Gy excludin! +11 from the list the so-called constructive 0historically, and in most textbooks E intuitionistic1 logic is obtained& 6s a concept, it as introduced by +uit:en @!bertus Han Grou er in 1,0C:
L. !. K. %rou8er. De onbetrou baarheid der lo!ische principes 0The unreliability of the lo!ical principles1, TiA+schrift 'oor BiAsbegeerte, 2 01,0C1, pp&1/2-1/C&

Grou erBs main ob>ection as a!ainst non-constructi#e proofs enabled mainly by SimproperS use of the +a of @xcluded 5iddle&

hich are

9or ele!ant examples of non-constructi#e proofs see <onstructi#e 5athematics by Dou!las Grid!es in 'tanford @ncyclopedia of %hilosophy& 4ote. 6 similar kind of non-constructi#e reasonin! is represented by the socalled Double ?e!ation +a : ffG`G, see 'ection 2&-& 6s a formal system, the intuitionistic lo!ic in 1,(0: as formulated by 6rend Heytin!

3. $eyting. Die formalen )e!eln der intuitionistischen 5athematik& Sit3$ngsberichte +er Pre$ssischen A&a+emie +er Bissenschaften, Physi&a%isch>mathematische C%asse , 1,(0, pp&.2/-&

The constructi#e concept of lo!ic differs from the classical one mainly in its

(2 interpretation of dis>unction and existence assertions: E To pro#e G#< constructively, you must pro#e G, or pro#e <& To pro#e B C by usin! the classical lo!ic, you are allo ed to assume ( BC ) as a hypothesis to deri#e a contradiction& Then, by the +a of @xcluded 5iddle ( B'C )( B'C ) you obtain B C & Ha#in! only such a Sne!ati#eS proof, you may be unable to determine, hich part of the dis>unction G#< is true E G, or <, or both& $no in! that B C is StrueS may inspire you to ork on the problem, but it may appear useless, if you do not succeed&&& E To pro#e xG0x1 constructively, you must pro#ide a particular #alue of x such that G0x1 is true& To pro#e xG0x1 by usin! the classical lo!ic, you are allo ed to assume xfG0x1 as a hypothesis to deri#e a contradiction& Then, by the +a of @xcluded 5iddle xB ( x ) xB ( x ) you obtain xG0x1& Ha#in! only such a Sne!ati#eS proof, you may be unable to find a particular x for hich G0x1 is true& $no in! that xG0x1 is StrueS may inspire you to ork on the problem, but it may appear useless, if you do not succeed&&& 4ote& 7nformally, e may re!ard existence assertions as Shu!e dis>unctionsS& 9or example, in the lan!ua!e of first order arithmetic, xG0x1 could be Sthou!htS as B ( 0) B ( 1 ) B ( 2 )&&& , i&e& as an infinite SformulaS& Thus, the abo#e t o theses are, in a sense, SeQui#alentS& The constructi#e 0intuitionist1 lo!ic is one of the !reat disco#eries in mathematical lo!ic E surprisin!ly, a complete system of constructi#e reasonin! 0as e ill see later, in 'ection .&.1 can be obtained simply by droppin! the +a of @xcluded 5iddle from the list of #alid lo!ical principles&
'ee also 7ntuitionistic +o!ic by Hoan 5oscho#akis in 'tanford @ncyclopedia of %hilosophy& +uit:en @!bertus Han Grou er 01CC1-1,--1: SHe re>ected in mathematical proofs the %rinciple of the @xcluded 5iddle, hich states that any mathematical statement is either true or false& 7n 1,1C he published a set theory, in 1,1, a measure theory and in 1,2( a theory of functions all de#eloped ithout usin! the %rinciple of the @xcluded 5iddle&S 0accordin! to 5acTutor History of 5athematics archi#e1& S<omo Heinrich 'chol: solia decir en sus cursos: no son ni Heide!!er ni 'artre los #erdaderos reno#adores de la filosofia, sino Grou er porQue shlo il ha atacado el bastihn dos #eces milenario del platonismo: la concepcihn de los entes matematicos como cosas en si&S 0Quoted after 6ndris )& )a!!io, @scritos <ompletos, Prometeo Libros, 20021&

Minimal Logic Gy excludin! both +10 and +11 the so-called minimal logic is obtained& 7t introduced by 7n!ebri!t Hohansson in 1,(-: as

I.Kohansson. Der 5inimalkalkgl, ein redu:ierter intuitionistischer 9ormalismus& Compositio .athematica, 1,(-, Vol& ., ?1, pp&11,-1(-&

(( 6s a separate concept, the minimal lo!ic is much less si!nificant than the constructi#e lo!ic& 7ndeed, since it allo s pro#in! of fG, G U- f< 0in a sense, /0T of +10;1, droppin! of +10 is not a #ery bi! step& :irst ?rder 9heories Ha#in! defined our predicate lan!ua!e +, and ha#in! formulated for + all the lo!ical axioms and rules of inference, do e need more8 To complete the formali:ation of our informal #ision of our domain of interest, e must formulate at least some specific axioms describin! the specific feat$res of the +omain& +o!ical axioms and rules of inference are #alid for any domains, i&e& they are Mcontent-freeN in the sense that, by usin! them only, one cannot deri#e specific information about the domain& 9or example, one cannot deri#e from 0any1 lo!ic, that x ( .a%e ( x ) Fema%e ( x )) & To communicate this fact to the computer, axiom& e must formulate it as a specific

6nd, as e ill pro#e in 'ection .&(, e ill ne#er need introducin! of specific r$%es of inference& 6ll e need are the t o lo!ical rules of inference 3 .o+$s Ponens and Genera%i3ation& Thus, as the result of the formali:ation process, first order theories& @ach first order theory T includes: a1 a specific predicate language +0T14 b1 logical a7ioms and rules of inference for this lan!ua!e 0classical or constructi#e #ersion may be adopted, see belo 14 c1 a set of s"ecific 0nonBlogicalD a7ioms of T&
6s the first example, letBs use our Slan!ua!e for peopleS to build a Mtheory for peopleN& @xamples of instances of lo!ical axioms for the Mlan!ua!e for peopleN: +1: x ( .a%e ( x ) 5 ( Fema%e ( x ) 5 .a%e ( x ))) 4 +-: x y ( .other ( x , y ) 5 .other ( x , y ) Father ( x , y )) 4 +11: .a%e ( /ohn ).a%e ( /ohn ) 4 +12: x ( Fema%e ( x )) 5 Fema%e ( Britney ) 4 etc& 6nd letBs introduce the follo in! non-lo!ical axioms:
x ( .a%e ( x ) Fema%e ( x )) 4

ill obtain the so-called

(.
x ( .a%e ( x ) Fema%e ( x )) 4 x y ( Father ( x , y ) 5 .a%e ( x )) 4 x ( y ( 3 (( Father ( x , 3 ) Father ( y , 3 )) 5 ( x = y ))))

&&&

!7ercise '.).'. @xtend this list of axioms as far as you can& 7s your list complete8 2hat do you mean by McompleteN8 6nother example of a first order theory %eano arithmetic1: The lan!ua!e of %6: a1 The constants 0 and 1, and all #ariables are terms& b1 7f t1 and t2 are terms, then 0t1\t21 and 0t1]t21 also are terms& c1 6tomic formulas are built as 0t1^t21, here t1 and t2 are terms& 'ince e can use, for example, the expression 2x 2-(y2-1^0 as a shortcut for 01\11]x]x^01\1\11]y]y\1, e can say simply that, in first order arithmetic, atomic formulas of are arbitrary Diophantine eQuations& 7nstances of lo!ical axioms +1: x =0 5 ( y =1 5 x = 0) 4 +-: x = y 5 x = y 3 =1 4 +11: 0= 17 ( 0= 1) 4 +12: x ( x = 1 ) 5 x = 1 4 etc& The specific 0non-lo!ical1 axioms of first order arithmetic: x^x, x^y`y^x, x^y`0y^:`x^:1, x^y`x\1^y\1, f00^x\11, x\1^y\1`x^y, x\0^x, x\0y\11^0x\y1\1, x]0^0, x]0y\11^0x]y1\x, B ( 0 ) x ( B ( x ) 5 B ( x + 1 )) 5 xB ( x ) , here G is any formula& The axioms =-10 represent recursi#e definitions of addition and multiplication& 6s the last the so-called induction schema is listed& 9or the most popular a7iom system of set theory 3 see Lermelo-9raenkelBs set theory& for the lan!ua!e of first order arithmetic:

E the so-called first order arithmetic P3 0also called

Proofs and 9heorems 7n !eneral, any seQuence of formulas 9 1, 92, &&&, 9m could be re!arded as a

(/ 0correct or incorrect1 formal "roof 0or simply, a proof1 of its last formula 9m& 7n a correct "roof, formulas can play only the follo in! roles: a1 Axioms& 'ome formulas may be instances of lo!ical or non-lo!ical axioms& b1 Conse4$ences of earlier formulas, obtained by usin! rules of inference& 9or example, if 92/ is 6, and 9(. is 6`G, and 9/1 is G, then e can say that 9/1 has been obtained from 92/ and 9(. by usin! the .o+$s Ponens rule& Rr, if 9-2 is <0x1, and 9-( is x<0x1, then e can say that 9-( has been obtained from 9-2 by usin! the Genera%i3ation rule& c1 Hypotheses& 'ome formulas may appear in the proof as Spro#ed else hereS, or e#en ithout any >ustification, simply by assumin! that they are StrueS& Thus, the follo in! notation can describe the actual status of a formal proof: VTW: 61, 62, &&&, 6n U- G, here T is a first order theory 0it determines hich formulas are axioms and hich are not1, 61, 62, &&&, 6n are all the hypotheses used in the proof, and G is the formula pro#ed by the proof& @ach formula in such a proof must be either an axiom, or a hypothesis from the set 61, 62, &&&, 6n, or it must be obtained from earlier formulas 0in this proof1 by usin! a rule of inference& Kou may read the abo#e notation as Sin theory T, by usin! formulas 61, 62, &&&, 6n as hypotheses, the formula G can be pro#edS& 9or the first examples of real formal proofs see the next 'ection 1&., Theorem 1&.&1 and Theorem 1&.&2& 7n the real mathematical practice, hen pro#in! VTW: 6 1, 62, &&&, 6n U- G, e may apply some theorem D that already has been pro#ed earlier& 7f e ould simply insert D into our formal proof, then, formally, this ould yield only a proof of VTW: 61, 62, &&&, 6n, D U- G, i&e& D ould be Qualified as a hypothesis& To obtain the desired formal proof of VTW: 6 1, 62, &&&, 6n U- G, e must insert not only D itself, but the entire "roof of D; 7n this ay e obtain the follo in! 9heorem '.).'. 7f there is a proof VTW: 6 1, 62, &&&, 6n, D U- G, and a proof VTW: 61, 62, &&&, 6n U- D, then there is a proof of VTW: 61, 62, &&&, 6n U- G& Proof. The proof of VTW: 61, 62, &&&, 6n, D U- G is a seQuence of formulas 9 1, 92, &&&, D, &&&, 9m, G, and the proof of VTW: 6 1, 62, &&&, 6n U- D is some seQuence of formulas A1, A2, &&&, Ap, D& +et us replace D by A1, A2, &&&, Ap, D:

(91, 92, &&&, A1, A2, &&&, Ap, D, &&&, 9m, G, and eliminate the duplicate formulas& This seQuence is a proof of VTW: 6 1, 62, &&&, 6n U- G& D&@&D& 7f, in some proof, hypotheses are not used at all, then e may rite simply VTW: U- G, or e#en T U- G, and say that G is a theorem of theory T& Rf course, by usin! axioms directly one almost ne#er can pro#e really complicated theorems& 'till, e can retain our simple formal definition of the notion of theorem because of the follo in! orollary '.).'. 7f there is a proof of VTW: 61, 62, &&&, 6n U- G, and proofs of VTW: U- 61, VTW: U- 62, &&&, VTW: U- 6n, then there is a proof of VTW: U- G& Proof. 7mmediately, by Theorem 1&(&1& onsistency 'ometimes, a seemin!ly plausible set of axioms allo s deri#in! of contradictions 0the most strikin! example E )ussellBs paradox in the Snai#eS set theory1& 6 formula 9 is called a contradiction in the theory T, if and only if VTW: U- 9 and VTW: U- f9, i&e& if T pro#es and dispro#es 9 simultaneously& Theories containin! contradictions are called inconsistent theories& Thus, T is called a consistent theory4 if and only if T does not allo deri#in! of contradictions& ?ormally, for a first order theory, the set of all theorems is infinite, and, therefore, consistency cannot be verified em"irically& 2e may only hope to establish this desirable property by means of some theoretical "roof 0see %odnieks V1,,=W, 'ection /&. for a more detailed discussion of this problem1& 9or theories adoptin! the abo#e lo!ical axioms, inconsistency is, in a sense, Sthe orst possible propertyS& 7ndeed, the axiom + 10: B 5 ( B 5 C ) says that if a theory allo s deri#in! a contradiction, then, in this theory, anythin! is pro#able& 7n 'ection 2&. e ill E ithout +10 E pro#e /0T of it: B 5 ( B 5 C ) & Thus, e#en ithout +10: if a theory allo s deri#in! a contradiction, then, in this theory, anythin! is dispro#able& 7s consistency enou!h for a theory to be SperfectS8 7n 'ection .&( e ill pro#e the so-called 5odel @xistence Theorem: if a first order theory is consistent, then there is a SmodelS 0a kind of Smathematical realityS1 here all its axioms and theorems are StrueS&

(= om"leteness 7f a formula contains free #ariables, i&e& #ariables that are not bound by Quantifiers 0for example: x = 0 x =1 1, then the Struth #alueS of such formulas may depend on particular #alues assi!ned to free #ariables& 9or example, the latter formula is StrueS for x^1, yet it is SfalseS for x^2& 9ormulas that do not contain free occurrences of #ariables, are called closed formulas, for example: - x ( - < x prime ( x )) & <losed formulas represent Sdefinite assertions about the ob>ects of our theoryS, and they are expected to be either StrueS, or SfalseS& Rr, in a first order theory, they are expected to be either pro#able, or dispro#able 0refutable1& The abo#e closed formula 0statin! that Sthere are infinitely many prime numbersS1 is pro#able E if our theory is first order arithmetic& T is called a com"lete theory, if and only if for each closed formula 9 in the lan!ua!e of T: VTW: U- 9 or VTW: U- f9, i&e& if and only if T pro#es or dispro#es any closed formula of its lan!ua!e& 7n other ords: a complete theory can sol#e any problem from the domain of its competence& 7n an incom"lete theory, some closed formulas 0Sdefinite assertions about the ob>ects of theoryS1 can be neither pro#ed, not dispro#ed& Thus, an incomplete theory canBt sol#e some of the problems from the domain of its competence& 9ormally, accordin! to this definition, an inconsistent theory is complete& 7ndeed, the axiom +10: fG`0G`<1 says that if a theory allo s deri#in! a contradiction, then, in this theory, anythin! is pro#able, i&e& it is a complete theory& Rf course, if T ould be both consistent and complete, then e could call it Sabsolutely perfectS& *nfortunately, AFdelBs incompleteness theorem says thata all fundamental mathematical theories are either inconsistent or incom"lete, i&e& none of them is Sabsolutely perfectS 0see 5endelson V1,,=W or %odnieks V1,,=W, 'ection -&11& !7ercise '.)... )e-formulate the abo#e axiom system for a manyBsorted "redicate language 0or, see <hapter 10& 5any-'orted 9irst Rrder +o!ic, by Hean Aallier&1

'.*. 9he :lavor of Proving Directly


9heorem '.*.'& V+1, +2, 5%W: U- 6`6 for any formula 6& 2hat does it mean8

(C 7tBs the so-called refle7ivity "ro"erty of im"lication& The follo in! seQuence of formulas represents a proof of the formula 6`6: 7tBs the axiom schema L.: 06`00<`61`611`006`0<`611` 0G`0<`D11`00G`<1`0G 011 06`611 `D11, ith G ^ 6, < ^ <`6, D ^ 6& 021 6`00<`61`61 7tBs the axiom schema L': G`0<`G1, ith G ^ 6, < ^ <`6& 7t follo s from 011 and 021 by .o+$s Ponens& 7tBs the axiom schema L': G`0<`G1, ith G ^ 6, < ^ <& 7t follo s from 0(1 and 0.1 by .o+$s Ponens&

0(1 06`0<`611`06`61

0.1 6`0<`61

0/1 6`6

6s you can see, the proof is easy to #erify, but it could be hard to build it from scratch& S2hyS should e take Sthe axiom + 2 ith G ^ 6, < ^ <`6, D ^ 6S for 0118
$o8 could one invent a "roof like the above one2 Rne of the #ersions could be as follo s& 9irst, letBs try to find an axiom, from hich e could !et 6`6 as a conseQuence& Gy tryin! +1, i&e& G`0<`G1, and settin! G^<^6, e could obtain 6`06`61, a dead end, perhaps& 'o, letBs try +2, i&e& 0G`0<`D11`00G`<1`0G`D11& Gy settin! G^D^6 e obtain 06`0<`611`006`<1`06`611& 7t seems to be a !ood decision E because the first premise 6`0<`61 is, in fact, +1& Hence, by applyin! the 5% rule, e obtain 06`<1`06`61& ?o , ho to make 6`< Spro#ableS8 'ince < is, in fact, an arbitrary formula, e can replace < by <`6, obtainin! 06`0<`611`06`61& The premise is here, a!ain, + 1, hence, by applyin! the 5% rule, e obtain 6`6& D&@&D& Gy performin! all our replacements at the #ery be!innin!, e obtain the abo#e proof of the formula 6`6& VGT2, the abo#e t o smart SoperationsS E obtainin! 6`6 ithin +2, and makin! +1 of 6`<, are applications of the socalled unification, a #ery !eneral and #ery important method used in intellectual computer pro!rams, for details, see 'ection /&=&W

9heorem '.*..& V+1, +2, 5%W: 6`G, G`< U- 6`<, for any formulas 6, G, <& 2hat does it mean8 7tBs the so-called La8 of ,yllogism 0by 6ristotle1, or the transitivity "ro"erty of im"lication& The follo in! seQuence of formulas represents a proof of the formula 6`<

(, from the hypotheses 6`G and G`<: 011 6`G 021 G`< 06`0G`<11`006`G1`06 0(1 `<11 0.1 0G`<1`06`0G`<11 0/1 6`0G`<1 0-1 06`G1`06`<1 0=1 6`< Hypothesis& Hypothesis& 7tBs the axiom schema L.: 0G`0<`D11`00G`<1`0G`D11, ith G ^ 6, < ^ G, D ^ <& 7tBs the axiom schema L': G`0<`G1, ith G ^ G`<, < ^ 6& 7t follo s from 021 and 0.1 by .o+$s Ponens& 7t follo s from 0(1 and 0/1 by .o+$s Ponens& 7t follo s from 011 and 0-1 by .o+$s Ponens&

4ote. Rnly axiom schemas +1 and +2 , and inference rule .o+$s Ponens are used for pro#in! the Theorems 1&.&1 and 1&.&2& Hence, these theorems 8ill remain valid for any logical system containing L', L. and Modus Ponens. !7ercise '.*.'. Guild seQuences of formulas representin! the follo in! proofs 0only the axiom schemas +1 and +2 and .o+$s Ponens are necessary1: a1 V+1, 5%W: 6 U- G`6 0a seQuence of ( formulas1& 2hat does it mean8 b1 V+2, 5%W: 6`G, 6`0G`<1 U- 6`< 0a seQuence of / formulas1& 2hat does it mean8 c1 V+1, +2, 5%W: 6`0G`<1 U- G`06`<1 0a seQuence of , formulas E thanks to %a#el 6ndreye# for the idea1& 2hat does it mean8 7tBs the so-called Premise Permutation La8& d1 V+1, +2, 5%W: 6`06`G1 U- 6`G 0easy solution - a seQuence of , formulas, smart solution by 6rnold Rstro#sky 3 C formulas1& 2hat does it mean8 9heorem '.*.)& V+1., 5%, AenW 7f 9 is any formula, and A is any formula that does not contain x as a free #ariable, then A`90x1 U- A`x90x1&

.0 The follo in! seQuence of formulas represents a proof of the formula A`x90x1 from the hypothesis A`90x1: 011 A`90x1 021 x0A`90x11 0(1 x0A`90x11`0A`x90x11 0.1 A`x90x1 Hypothesis& 7t follo s Aenerali:ation& from 011 by

7tBs the axiom schema L'* 0A does not contain x as a free #ariable1& 7t follo s from 021 and 0(1 by .o+$s Ponens&

!7ercise '.*..& Guild seQuences of formulas representin! the follo in! proofs 09 is any formula, and A is a formula that does not contain x as a free #ariable1: a1 V+1/, 5%, AenW: 90x1`A U- x90x1`A 0a seQuence of . formulas1& 2hat does it mean8 b1 V+(-+/, 5%W: A B U- B A 0a seQuence of C formulas1& 2hat does it mean8 c1 V+--+C, 5%W: U- A B 5 B A 0a seQuence of / formulas1& 2hat does it mean8 d1 V+1, +,, 5%W: G, fG U- f< 0a seQuence of , formulas1& 2hat does it mean8 7tBs /0T of the axiom +10; e1 V+(, +., +,, 5%W: U- ( A A) 0a seQuence of / formulas1& 2hat does it mean8 7tBs the so-called La8 of 4onB ontradiction& f1 V+1, +C, +10, 5%W: U- A B 5 ( A5 B ) 0a seQuence of / formulas1& 2hat does it mean8 !1 V+C, +11, 5%W: 6`G, f6`G U- G 0a seQuence of = formulas1& 2hat does it mean8 h1 V+1-+C, 5%W: A 5 B U- AC 5 B C 0a seQuence of 11 formulas1& 2hat does it mean8 i1 V+1-+11, 5%W: U- A( A5 B ) 0a seQuence of 1/ formulas1& 2hat does it mean8 Does it mean anythin! at all8 !7ercise '.*.) 0for smart students1& <ould one build shorter seQuences pro#in! the formulas of @xercise 1&.&1 c, d1 and @xercise 1&.&2 b, d18 @#!eny Vihro#

.1 #erified in 2011 that any proof of the formula of @xercise 1&.&1 d1 lon!er than / formulas& ill be

'.-. Deduction 9heorems


7f, by assumin! G as a hypothesis, e ha#e pro#ed <, then e ha#e pro#ed that G implies <& This natural ay of reasonin! is formali:ed in the so-called deduction theorems 0introduced by HacQues Herbrand and 6lfred Tarski1:
K. $erbrand. )echerches sur la thiorie de la dimonstration& %hD Thesis, *ni#ersity of %aris, 1,(0 0appro#ed in 6pril 1,2,1& 3. 9arski. *eber eini!e fundamentale Ge!riffe der 5etamathematik& S<omptes )endus de 'iances de la 'ociiti des 'ciences et des +ettres de Varso#ie, <lasse 777S, 1,(0, Vol&2(, pp& 22-2,&

2e ill pro#e t o such theorems 3 Deduction Theorem 1 0for propositional lo!ic1 and Deduction Theorem 2 0for predicate lo!ic1& 9heorem '.-.' CDeduction 9heorem 'D& 7f T is a first order theory, and there is a proof of VT, 5%W: 61, 62, &&&, 6n, G U- <, then there is a proof of V+1, +2, T, 5%W: 61, 62, &&&, 6n U- G`<& 7&e& ha#in! a .o+$s Ponens proof of < from the hypotheses 6 1, 62, &&&, 6n, G, e can build a .o+$s Ponens proof of G`< from the hypotheses 6 1, 62, &&&, 6n& 7t appears that, usually, pro#in! of VT, 5%W: &&& G U- < is easier 0technically simpler1 than pro#in! of VT, 5%W: &&& U- G`<& !7ercise '.-.' 0for smart students1& Do not read the proof belo & Try pro#in! yourself& Proof 0thanks to 'er!ey $o:lo#ich for the idea, see also $leene V1,-=W, @xercise 10<1& 2e must define a procedure allo in! to con#ert any proof of VT, 5%W: 61, 62, &&&, 6n, G U- < into a proof of V+1, +2, T, 5%W: 61, 62, &&&, 6n UG`<& The easy ay to do this ould be usin! an induction by the number of formulas in the proof of VT, 5%W: 61, 62, &&&, 6n, G U- <& Gut e ill use a more

.2 ele!ant idea& 6ny proof of VT, 5%W: 61, 62, &&&, 6n, G U- < is a seQuence of formulas 91, 92, &&&9m& 2e ill replace each formula 9i by ( or / formulas, the last of these bein! the formula G`9i, retainin! our seQuence as a #alid proof& 2e must consider the follo in! cases: 11 9 is an axiom 0i&e& an instance of a lo!ical axiom or a non-lo!ical axiom of T1& )eplace 9 by ( formulas: 9, 9`0G`91, G`9& The second formula is an instance of +1, the third formula is obtained from the first t o ones by usin! .o+$s Ponens& 21 9 is one of the hypotheses 6 i& )eplace 9 by ( formulas: 9, 9`0G`91, G`9& The second formula is an instance of + 1, the third formula is obtained from the first t o ones by usin! .o+$s Ponens& (1 9 is G& )eplace 9 by the / formulas from the proof of Theorem 1&.&1, here D can be any formula: 0G`00D`G1`G11`00G`0D`G11`0G`G11 0an instance of +21, G`00D`G1`G1 0an instance of +11, G`0D`G11`0G`G1 0by .o+$s Ponens1, G`0D`G1 0an instance of +11, G`G 0by .o+$s Ponens1& The last formula is here, of course, G`9& .1 9 is deri#ed from some pre#ious formulas 9 i and 9> by .o+$s Ponens, 9i ha#in! the form 9>`9 0i&e& 9>`9 and 9> yield 9 by .o+$s Ponens1& Then, the formulas G`9>, G`09>`91 are already present in the con#erted proof 0they appeared durin! the replacement operations applied to the formulas 9> and 9>`91& 'o, replace 9 by ( formulas: 0G`09>`911`00G`9>1`0G`911 0an instance of +21, 0G`9>1`0G`91 0by .o+$s Ponens1, G`9 0by .o+$s Ponens1& Thus, hat e ha#e no , is a correct proof in V+ 1, +2, 5%W that is usin! the

.( hypotheses 61, 62, &&&, 6n, but not G; The last formula of this proof is G`< 0because < is the last formula of our initial proof of V+ 1, +2, 5%W: 61, 62, &&&, 6n, G U- <1& Thus, e ha#e a proof of V+1, +2, 5%W: 61, 62, &&&, 6n U- G`<& D&@&D& The abo#e proof of Deduction Theorem 1 includes, in fact, an algorithm allo in! to obtain a proof of V+1, +2, 5%W: 61, 62, &&&, 6n U- G`< from a !i#en proof of V+1, +2, 5%W: 61, 62, &&&, 6n, G U- <& The resultin! proof is lon!er than the !i#en one: if the !i#en proof consists of m formulas, then the resultin! proof consists of Dm or (m\2 formulas1& orollaries '.-.'. 11 7f there is a proof of VT, 5%W: 61, 62, &&&, 6n, G1, G2, &&&, Gk U- <, then there is a proof of V+1, +2, T, 5%W: 61, 62, &&&, 6n U- 0G1`0G2`0&&&`0Gk`<1&&&111& 21 7f T includes 0or pro#es1 schemas + 1, +2, then, if there is a proof of VT, 5%W: 61, 62, &&&, 6n, G U- < then there is a proof of VT, 5%W: 61, 62, &&&, 6n U- G`< & 7n particular, if VT, 5% W: G U- <, then VT, 5%W: U- G`<& 6nd, if VT, 5% W: G, <U- D, then VT, 5%W: U- G`0<`D1& Proof. 11 Gy iteratin! Deduction Theorem 1& 21 7f T is a theory hich includes or pro#es the schemas +1, +2, then V+1, +2, T, 5%W is eQui#alent to VT, 5%W& D&@&D& !7ercise '.-.. 0for smart students1& 7n earlier #ersions of lo!ical axioms, instead of the axiom +2, in some texts, the follo in! ( axioms ere in use: +21: 06`06`G11`06`G1, +22: 06`0G`<11`0G`06`<11 0i&e& the %remise %ermutation +a 1, +2(: 06`G1`00G`<1`06`<11 0the +a property of implication1& of 'yllo!ism, or the transiti#ity

Verify that both #ersions, i&e& V+1, +2, 5%W and V+1, +21, +2(, +2(, 5%W are eQui#alent& 0Hint: a1 'ee 'ection 2&1 to #erify that V+1, +2, 5%W pro#es +21, +2(, +2(& b1 Verify that V+1, +21, +2(, +2(, 5%W pro#es +2 either directly, or by pro#in! the Deduction Theorem 1 for V+1, +21, +2(, +2(, 5%W&1 !7ercise '.-.) 0thanks to 'er!ey $o:lo#ich for the idea1&

.. a1 %ro#e the follo in! S!enerali:ationS of the .o+$s Ponens rule: V+1, +2, 5%W: 0D1`0D2`&&&0Dk`G1&&&1, 0D1`0D2`&&&0Dk`0G`<11&&&1 U0D1`0D2`&&&0Dk`<1&&&1& b1 %ro#e the follo in! S!enerali:ationS of the axiom +1. 0formulas D1, D2, &&&, Dk do not contain x as a free #ariable1: V+1, +2, +1., 5%W: U- x0D1`0D2`&&&0Dk`90x11&&&1 ` 0D1`0D2`&&& 0Dk`x90x11&&&1& !7ercise '.-.* 0for smart students1& 7n#esti!ate the si:e 0the number of formulas1 of the proof of V+1, +2, 5%W: 61, 62, &&&, 6n, U- G`< as a function f0m1 of the si:e m of the proof of V+ 1, +2, 5%W: 61, 62, &&&, 6n, G U- <& Kou may ish to report your result& 2e ill publish your report on the eb as an appendix to this book& The current record holder is 'er!ey $o:lo#ich, 200.: f0m1 c (m\2& 7mpro#e this result, or pro#e that it is the best one possible& !7ercise '.-.- 0for smart students1& 7n#esti!ate the si:e 0the number of instances of atomic formulas1 of the proof of V+ 1, +2, 5%W: 61, 62, &&&, 6n, UG`< as a function !0m1 of the si:e m of the proof of V+ 1, +2, 5%W: 61, 62, &&&, 6n, G U- <& Kou may ish to report your result& 2e ill publish your report on the eb as an appendix to this book& The current record holder is $irils 'olo#>o#s, 200C: !0m, n1 c =m\2.nE2, here n is the number of instances of atomic formulas in the formula G& 7mpro#e this result, or pro#e that it is the best one possible& 1arning6 Generalization involved... ?o , hat, if in the proof of 6 1, 62, &&&, 6n, G U- < not only .o+$s Ponens, yet also Genera%i3ation is used8 2e must be careful, because, tryin! SsimplyS to apply Deduction Theorem 1, e can obtain cra:y results& 7ndeed, ha#in! a formula 90x1, by Genera%i3ation, e obtain the formula x90x1& Thus, 90x1 U- x90x1& 7f Deduction Theorem 1 could be extended to Aen ithout any restrictions, then e could conclude that U- 90x1`x90x1& 7f this is true for any x, it is true also for x^2, hence, U9021`x90x1& Thus, if the number 2 is prime, then all numbers are prime8 'o, let us try deri#in! a restricted formulation of the Deduction Theorem E it seems, e should "rohibit a""lication of 5en to the free variables of % E the hypothesis Sto be mo#edS& 9heorem '.-.. CDeduction 9heorem .D& 7f T is a first order theory, and there is a proof of

./ VT, 5%, AenW: 61, 62, &&&, 6n, G U- <, here Genera%i3ation is not applied to the free #ariables of G, then there is a proof of V+1, +2, +1., T, 5%, AenW: 61, 62, &&&, 6n U- G`<& Proof. 2e must extend the abo#e proof of the Deduction Theorem 1 that consisted of . cases& 9irst, e must extend the first case: 1B1 9 is an axiom 0i&e& an instance of a lo!ical axiom or a non-lo!ical axiom of T1& )eplace 9 by ( formulas: 9, 9`0G`91, G`9& The second formula is an instance of +1, the third formula is obtained from the first t o ones by usin! .o+$s Ponens& 6nd e must add the follo in! case: /1 9 is deri#ed from some pre#ious formula 9 i by Genera%i3ation, thus, 9 ha#in! the form x9i, here x is not free in the formula G& )eplace 9 by the follo in! ( formulas: x0G`9i1`0G`x9i1, x0G`9i1, G`x9i& 'ince x is not free in G, the first formula is an instance of + 1.& The second formula is obtained by Genera%i3ation from the formula G`9i that is already present in the con#erted proof 0it appeared durin! the replacement operation applied to the formula 9i1& The third formula is obtained from the first t o ones by usin! .o+$s Ponens& Thus, hat e ha#e no , is a correct proof in V+1, +2, +1., 5%, AenW that is usin! the hypotheses 61, 62, &&&, 6n, but not G; The last formula of this proof is G`< 0because < is the last formula our initial proof of V+1, +2, +1., 5%, AenW: 61, 62, &&&, 6n, G U- <1& Thus, e ha#e a proof of V+ 1, +2, +1., 5%, AenW: 61, 62, &&&, 6n U- G`<& D&@&D& orollary '.-..& 11 7f there is a proof of VT, 5%, AenW: 61, 62, &&&, 6n, G1, G2, &&&, Gk U- <, here Genera%i3ation is not applied to the the free #ariables of the formulas G1, G2, &&&, Gk, then there is a proof of V+1, +2, +1., T, 5%, AenW: 61, 62, &&&, 6n U- 0G1`0G2`0&&&`0Gk`<1&&&111&

.21 7f G is a closed formula, and there is a proof of VT, 5%, AenW: 61, 62, &&&, 6n, G U- <, then there is a proof of V+1, +2, +1., T, 5%, AenW: 61, 62, &&&, 6n U- G`<& (1 7f T is a theory hose axioms include schemas +1, +2, +1., then, if there is a proof of VT, 5%, AenW: 61, 62, &&&, 6n, G U- <, here Genera%i3ation is not applied to the the free #ariables of G, then there is a proof of VT, 5%, AenW: 61, 62, &&&, 6n U- G`<& 7n particular, if VT, 5%, AenW: G U- <, here Genera%i3ation is not applied to the free #ariables of G, then VT, 5%, AenW: U- G`<& Proof. 'imilar to the proof of the abo#e <orollaries of Deduction Theorem 1& 1arning6 Previously "roved theorems involved... 7n the real mathematical practice, hen pro#in! VT, 5%, AenW: 6 1, 62, &&&, 6n U<, e may ish to apply some theorem D that e ha#e already pro#ed earlier& 7f e ould simply insert D into our formal proof, then, formally, this ould yield only that VT, 5%, AenW: 6 1, 62, &&&, 6n, D U- <& To obtain the desired formal proof of VT, 5%, AenW: 61, 62, &&&, 6n U- <, e must insert not only D itself, but the entire "roof of D; 'till, ith the Deduction Theorem 2 this may be problematic& 7f e are pro#in! VT, 5%, AenW: 61, 62, &&&, 6n, G U- < ith the intention to apply Deduction Theorem 2 0to obtain VT, 5%, AenW: 61, 62, &&&, 6n U- G`<1, then, before insertin! the proof of D, e must ensure that, in this proof, Genera%i3ation is not applied to the free #ariables of G& Gut, of course, the ori!inal proof of D could contain such Genera%i3ations; To sol#e this problem, e could try, in the proof of D, before insertin! it, rename simultaneously all the #ariables to hich Genera%i3ation is applied and hich are free #ariables in G& Gut this simultaneous renamin! may affect the bound #ariables of D, and thus E destroy the intended use of D& The problem is sol#ed completely by the follo in! extension of the Deduction Theorem 2:

.= 9heorem '.-.) CDeduction 9heorem .3D& 7f there is a proof of VT, 5%, AenW: 61, 62, &&&, 6n, G U- <, here, after % a""ears in the "roof, Genera%i3ation is not applied to the free #ariables of G, then there is a proof of V+1, +2, +1., T, 5%, AenW: 61, 62, &&&, 6n U- G`<& 7ndeed, ha#in! such a theorem, e obtain the necessary orollary '.-.). 7f there is a proof of VT, 5%, AenW: 61, 62, &&&, 6n, G, D U- <, here, after G appears in the proof, Genera%i3ation is not applied to the free #ariables of G, and there is a proof of VT, 5%, AenW: 61, 62, &&&, 6n U- D, then there is a proof of VT, 5%, AenW: 61, 62, &&&, 6n U- G`<& Proof of the orollary. 7n the proof of VT, 5%, AenW: 6 1, 62, &&&, 6n, G, D U- <, first, mo#e all the hypotheses 61, 62, &&&, 6n to the be!innin!& Then, immediately after them, insert the proof of VT, 5%, AenW: 6 1, 62, &&&, 6n U- D& ?o e ha#e a proof of VT, 5%, AenW: 6 1, 62, &&&, 6n, G U- <, here, after G appears in the proof, Genera%i3ation is not applied to the free #ariables of G& Gy Deduction Theorem 26, then there is a proof of VT, 5%, AenW: 6 1, 62, &&&, 6n U- G`<& D&@&D& Proof of the Deduction 9heorem .3. +et us modify the abo#e proof of the Deduction Theorem 2& 2e must define a procedure allo in! to con#ert any allo ed proof of VT, 5%, AenW: 61, 62, &&&, 6n, G U- < into a proof of V+ 1, +2, T, 5%, AenW: 61, 62, &&&, 6n U- G`<& *nlike the abo#e proof, let us lea#e unchan!ed all the formulas of the proof of VT, 5%W: 61, 62, &&&, 6n, G U- < before % a""ears in the "roof& 6fter this, starting 8ith %, e ill replace each formula 9 by ( or / formulas, one of them bein! the formula G`9& 2e must consider the follo in! cases: 11, 21, (1 E as in the proof of the Deduction Theorem 1&

.C .1 9 is deri#ed from some pre#ious formulas 9 i and 9> by .o+$s Ponens, 9i ha#in! the form 9>`9 0i&e& 9>`9 and 9> yield 9 by .o+$s Ponens1& Then, . subcases are possible& .a1 9> and 9>`9 both appear before G, i&e& they remain unchan!ed in the con#erted proof& +et us replace 9 by the follo in! ( formulas: 9, 9`0G`91, G`9& The second formula is an instance of + 1, the third formula is obtained by usin! .o+$s Ponens from the first t o ones& .b1 9> appears before G, and 9>`9 is G or appears after G& Then, the formulas 9> and G`09>`91 are already present in the con#erted proof& +et us replace 9 by the follo in! / formulas: 0G`09>`911`00G`9>1`0G`911 0an instance of +21, 0G`9>1`0G`91 0by .o+$s Ponens1, 9>`0G`9>1 0an instance of +11, G`9> 0by .o+$s Ponens1, G`9 0by .o+$s Ponens1& .c1 9> is G or appears after G, and 9>`9 appears before G& Then, the formulas G`9> and 9>`9 are already present in the con#erted proof& +et us replace 9 by the follo in! / formulas from the proof of Theorem 1&.&2: 0G`09>`911`00G`9>1`0G`911 0an instance of +21, 09>`91`0G`09>`911 0an instance of +11, G`09>`91 0by .o+$s Ponens1, 0G`9>1`0G`91 0by .o+$s Ponens1, G`9 0by .o+$s Ponens1& .d1 9> and 9>`9 both are G or appear after G& Then, the formulas G`9 > and G`09>`91 are already present in the con#erted proof 0they appeared durin! the replacement operations applied to the formulas 9 > and 9>`91& +et us replace 9 by the follo in! ( formulas: 0G`09>`911`00G`9>1`0G`911 0an instance of +21, 0G`9>1`0G`91 0by .o+$s Ponens1, G`9 0by .o+$s Ponens1&

., /1 9 is deri#ed from some pre#ious formula 9i by Genera%i3ation, thus, 9 is in the form x9i& Then, 2 subcases are possible& /a1 9i appears before G& Then x is not free in G& +et us replace 9 by the follo in! ( formulas: 9 0by Genera%i3ation, x is not free in G1, 9`0G`91 0an instance of +11, G`9 /b1 9i is G or appears after G& Then x is not free in G, and the formula G`9 i that is already present in the con#erted proof 0it appeared durin! the replacement operation applied to the formula 9 i1& +et us replace 9 by the follo in! ( formulas: x0G`9i1 0by Genera%i3ation, x is not free in G1, x0G`9i1`0G`x9i1 0an instance of +1., since x is not free in G1, G`x9i 0by .o+$s Ponens1& Thus, hat e ha#e no , is a correct proof in V+ 1, +2, +1., T, 5%, AenW that is usin! the hypotheses 61, 62, &&&, 6n, but not G; The last formula of this proof is G`< 0because < is the last formula our initial proof of VT, 5%, AenW: 6 1, 62, &&&, 6n, G U- <1& Thus, e ha#e a proof of V+ 1, +2, +1., T, 5%, AenW: 61, 62, &&&, 6n U- G`<& D&@&D& !7ercise '.-./ 0for smart students1& 7n some other textbooks, a some hat different system of lo!ical axioms is used: instead of the axioms + 1., +1/ and the Genera%i3ation rule the follo in! t o rules of inference are used: A`90x1 U- A`x90x1 0-7ntroduction14 90x1`A U- x90x1`A 0-@limination1& Rf course, here, A is a formula that does not contain x as a free #ariable& Verify that both systems are eQui#alent in all of their #ersions 0minimal, constructi#e, and classical1&

/0

.. Pro"ositional Logic
Aeor!e Goole 01C1/-1C-.1: S7n 1C/. he published An #n'estigation into the La-s of Tho$ght, on Bhich are fo$n+e+ the .athematica% Theories of Logic an+ Probabi%ities( Goole approached lo!ic in a ne ay reducin! it to a simple al!ebra, incorporatin! lo!ic into mathematics& He pointed out the analo!y bet een al!ebraic symbols and those that represent lo!ical forms& 7t be!an the al!ebra of lo!ic called Goolean al!ebra hich no finds application in computer construction, s itchin! circuits etc&S 0accordin! to 5acTutor History of 5athematics archi#e1& 'ee also: 5.%oole& The <alculus of +o!ic& The Cambri+ge an+ ,$b%in .athematica% /o$rna% , #ol& ( 01C.C1 0a#ailable online at http:PP &maths&tcd&iePpubPHist5athP%eoplePGooleP<alc+o!icP, published by Da#id )& 2ilkins1&

..'. Proving :ormulas ontaining Im"lication only


+et us return to the @xercise 1&.&10d1, formulas pro#in! the follo in!: d1 V+1, +2, 5%W: 6`06`G1 U- 6`G& Did you try the next step 3 pro#in! of dB1 V+1, +2, 5%W: U- 06`06`G11`06`G18 9or pro#in! directly 3 almost an impossible task; ?o , ha#in! deduction theorems, e can simplify the task of pro#in! d1, and make the task of pro#in! dB1 feasible& More "recisely E the task of "roving that dD and dJD are "rovable. 7ndeed, 011 6`06`G1 021 6 0(1 6`G 0.1 G Hypothesis& Hypothesis& Gy 5%, from 011, 021& Gy 5%, from 021, 0(1& here you produced a seQuence of ,

Thus, e ha#e established that 6`06`G1, 6 U- G& ?o , by Deduction Theorem 1, V+1, +2, 5%W: 6`06`G1 U- 6`G&

/1 6nd let us apply this theorem once more, V+1, +2, 5%W: U- 06`06`G11`06`G1& 4ote. 7n fact, e pro#ed here only 6`06`G1, 6 U- G, but 8e did not "rove dD and dJD, i.e. 8e did not "roduce the corres"onding se>uences of formulas. 1e @ust "roved that these se>uences do e7ist6 To produce them really, e must apply the al!orithm described in the proof of Deduction Theorem 1& !7ercise ..'.'. 7ma!ine applyin! the al!orithm described in the proof of Deduction Theorem 1: a1 to the abo#e . formula seQuence 3 producin! a seQuence of .. formulas pro#in! V+1, +2, 5%W: U- 06`06`G11`06`G14 b1 to your , formula proof of 0d1 3 producin! a seQuence of 2, formulas pro#in! the same&

1arning6 6l ays be careful

hen selectin! hypotheses& 9or

example, to pro#e the stran!e formula 0the so-called %eirceBs +a 1 U006`G1`61`6 0it is pro#able in the classical lo!ic, not in the constructi#e lo!ic;1, you can try pro#in! that 06`G1`6 U- 6, but not 6`G, 6 U- 6& 2hy8 Gecause, by Deduction Theorem 1, from 6`G, 6 U- 6 it follo s that 6`G U6`6 and U- 06`G1`06`61, or 6 U- 06`G1`6 and U- 6`006`G1`61& 2here do you see U- 006`G1`61`6 here8 !7ercise ..'... %ro#e the follo in! V+1, +2, 5%W: a1 U- 006`G1`06`<11`06`0G`<11& 2hat does this formula mean8 b1 U- 06`G1`00G`<1`06`<11& 2hat does this formula mean8 7tBs another #ersion of the so-called La8 of ,yllogism 0by 6ristotle1, or the transitivity "ro"erty of im"lication& @xplain the difference bet een this formula and Theorem 1&.&2: 6`G, G`< U- 6`<& c1 U- 06`0G`<11`0G`06`<11& 2hat does this formula mean8 7tBs another #ersion of the so-called Premise Permutation La8. @xplain the difference bet een this formula and @xercise 1&.&10c1: 6`0G`<1 U- G`06`<1&

.... Proving :ormulas ontaining on@unction


9heorem ....'& a1 V+/, 5%W 6, G U- A B & b1 V+(, +., 5%W: A B U- 6, A B U- G& +et us pro#e 0a1&

/2 011 6 021 G 0(1 0.1 0/1 A 5 ( B 5 A B )


B 5 A B A B

Hypothesis& Hypothesis& 6xiom L-: B 5 ( C 5 BC ) G ^ 6, < ^ G& Gy 5%, from 011 and 0(1& Gy 5%, from 021 and 0.1& ith

?o , let us pro#e 0b1& 011 021 0(1 6 Thus, 011 021 0(1 G Thus, A B U- G& Theorem 2&2&1 allo s easy pro#in! of e>uivalences& +et us recall that Ga< is defined as a shortcut for ( B 5 C )(C 5 B ) & Rf course, e ill call G and < eQui#alent formulas, if and only if U- Ga<& 9or example, by Theorem 1&.&1, V+1, +2, 5%W U- 6`6, hence, V+1, +2, +/, 5%W U- ( A 5 A )( A5 A) , i&e& V+1, +2, +/, 5%W U- 6a6& Rf course, 0a1 of the @xercise 2&1&2 is the re#erse formula of the axiom +2& Hence, by Theorem 2&2&1: V+1, +2, +/, 5%W U- 06`0G`<11 a 006`G1`06`<11& Gy 0c1 of the @xercise 2&1&2, and Theorem 2&2&1: V+1, +2, +/, 5%W U- 06`0G`<11a0G`06`<11 ?o , let us pro#e another form of the +a of 'yllo!ism, or Theorem 1&.&2 A B U- 6& A B
A B 5 B A B

Hypothesis& 6xiom L): BC 5 B < ^ G& Gy 5%, from 011 and 021& ith G ^ 6,

A B 5 A

Hypothesis& 6xiom L*: BC 5 C < ^ G& Gy 5%, from 011 and 021& ith G ^ 6,

/( V+1, +2, 5%W: 6`G, G`< U- 6`<: V+1-+., 5%W U- ( A 5 B )( B 5C ) 5 ( A5 C ) & 011 021 0(1 ( A 5 B )( B 5C ) ( A 5 B )( B 5C ) 5 ( A5 B ) ( A 5 B )( B 5C ) 5 ( B 5 C ) Hypothesis& 6xiom L): BC 5 B 6`G, < ^ G`<& 6xiom L*: BC 5 C 6`G, < ^ G`<& Gy 5%, from 011 and 021& Gy 5%, from 011 and 0(1& Gy by the transiti#ity property of implication 0Theorem 1&.&21& ith G ^ ith G ^

0.1 6`G 0/1 G`< 0-1 6`<

Thus, e ha#e established that V+1-+., 5%W: ( A 5 B )( B 5C ) U- 6`<& Gy Deduction Theorem 1, V+1-+., 5%W U- (( A5 B )( B 5 C )) 5 ( A5 C ) & !7ercise ....'. %ro#e the follo in! V+1- +/, 5%W: a1 6`G, 6`< U- A 5 BC & 2hat does it mean8 b1 U- ( A 5 B )( A 5C ) 5 ( A 5 BC ) & 2hat does it mean8 c1 A 5 BC U- 6`G& 2hat does it mean8 d1 A 5 BC U- 6`<& 2hat does it mean8 e1 U- ( A 5 BC ) 5 ( A5 B )( A 5C ) & 2hat does it mean8 Hence, V+1- +/, 5%W: U- ( A 5 BC ) 6 ( A5 B )( A 5C ) & !7ercise ...... %ro#e the follo in!, V+1- +/, 5%W: a1 U- A B 6 B A & 2hat does it mean8 That con@unction is commutative& b1 U- A( B C ) 6 ( A B )C & 2hat does it mean8 That con@unction is associative& c1 U- A A6 A & 2hat does it mean8 That con@unction is idem"otent& !7ercise ....). %ro#e the follo in!, V+1- +/, 5%W: a1 U- ( A 5 ( B 5 C )) 6 ( A B 5 C ) & 2hat does it mean8

/. b1 U- ( A 5 B ) 5 ( AC 5 B C ) & 2hat does it mean8 The con#erse formula ( A C 5 B C ) 5 ( A5 B ) cannot be true& @xplain, hy& c1 6 U- B 6 B A & 2hat does it mean8 !7ercise ....*. +et us recall that the eQui#alence connecti#e 6aG is defined as a shortcut for ( A 5 B )( B 5 A) & %ro#e the follo in! properties of this connecti#e V+1- +/, 5%W: 0a1 U- 6a6 0reflexi#ity1, 0b1 U- 06aG1`0Ga61 0symmetricity1, 0c1 U- ( A 6 B )( B 6C ) 5 ( A6 C ) 0transiti#ity1&

..). Proving :ormulas ontaining Dis@unction


!7ercise ..).'. %ro#e the follo in! V+1, +2, +--+C, 5%W: a1 V+C, 5%W: 6`<, G`< U- A B 5C & 2hat does it mean8 b1 V +/, +--+C, 5%W: U- A B 6 B A & 2hat does it mean8 That dis@unction is commutative& c1 V+1, +2, +/, +--+C, 5%W: U- A A 6 A & 2hat does it mean8 That dis@unction is idem"otent& 9heorem ..).I. V+1, +2, +C, 5%W: 7f there is a proof of 61, 62, &&&, 6n, G U- D, and a proof of 61, 62, &&&, 6n, < U- D, then there is a proof of A1 , A2 , &&& , An , BC U- D& !7ercise ..)... %ro#e Theorem 2&(&0& %ro#in! that dis@unction is associative reQuires some sophistication 0smart students: do not read the proof belo , try pro#in! yourself1: V+1, +2, +/, +--+C, 5%W: U- A( B C ) 6 ( A B )C & 9irst, ima!ine, e are tryin! to obtain, for example,

// A( B C ) 5 ( A B )C , from the axiom +C: ( A 5 ( A B )C ) 5 (( B C 5 ( A B )C ) 5 ( A( BC ) 5 ( A B )C )) 7t seems, the second premise from +C: BC 5 ( A B )C also could be obtained

( B 5 ( A B )C ) 5 ((C 5 ( A B )C ) 5 ( BC 5 ( A B )C )) & Thus, the ` part of the task ill be sol#ed, if e could pro#e that U011 A 5 A( BC ) , UA 5 A( BC ) ?o , let us pro#e that U021 0(1 B 5 A( B C ) & B 5 BC BC 5 A( B C ) 6xiom L/& 6xiom L0: C 5 BC < ^ G#<& ith G ^ 6, B 5 A( B C ) , U- C 5 A( BC ) & 6xiom L/: B 5 BC < ^ B C & ith G ^ 6,

0.1

B 5 A( B C ) ?o , let us pro#e that U- C 5 A( BC ) &

9rom 021 and 0(1, by the transiti#ity property of implication 0Theorem 1&.&21&

0/1 0-1

C 5 BC BC 5 A( B C )

6xiom L0& 6xiom L0: C 5 BC < ^ G#<& ith G ^ 6,

0=1

C 5 A( BC )

9rom 0/1 and 0-1, by the transiti#ity property of implication 0Theorem 1&.&21&

!7ercise ..).). a1 %ro#e the con#erse: V+1, +2, +--+C, 5%W: U( A B )C 5 A( BC ) & b1 %ro#e 0use Deduction Theorem 11 that V+ 1, +2, +--+C, 5%W: U-

/( A 5 B ) 5 ( AC 5 B C ) & 2hat does it mean8 The con#erse formula ( A C 5 B C ) 5 ( A 5 B) cannot be true& @xplain, hy& c1 %ro#e that V+1, +2, +--+C, 5%W: U- 6`G, <`D U- AC 5 B , & 2hat does it mean8 The follo in! theorem corresponds to the ell-kno n distributive "ro"erty of 0number1 addition to multiplication: 0a\b1c ^ ac\bc& Rf course, the SdualS distributi#e property 0i&e& 3 of multiplication to addition1 does not hold for numbers: ab\c^0a\c10b\c1 ould imply ab\c^ab\ac\bc\cc, c^ac\bc\cc, and, if c_b0, then 1^a\b\c& 'till, surprisin!ly, in lo!ic, 9heorem ..).'. on@unction is distributive to dis@unction, and dis@unction is distributive to con@unction= V+1-+C, 5%W: U- ( A B )C 6 ( A C )( B C ) & V+1-+C, 5%W: U- ( A B )C 6 ( A C )( B C ) & 9irst, let us pro#e that U- ( A B )C 5 ( A C )( B C ) & 011 %ro#e U- A B 5 ( AC )( BC ) 021 %ro#e U- C 5 ( AC )( B C ) 0(1 U- ( A B)C 5 ( AC )( BC ) !7ercise ..).*. a1 %ro#e 011 and 021& b1 Do not read the follo in! proof& Try pro#in! yourself& ?o , let us pro#e the con#erse: U- ( A C )( B C ) 5 ( A B )C & 4ote. The proof belo starts ith < as a hypothesis& 2hy not ith ( A C )( B C ) 8 Gecause, e ill use Deduction Theorem 1 to pro#e the intermediate formula 0-1 C 5 ( B C 5 ( A B )C ) , not the final result; 011 021 0(1 0.1 0/1 < G`< C 5 ( A B)C B 5 ( A B )C BC 5 ( A B )C Hypothesis& 9rom 011& 6xiom +=& 9rom 021 and 0(1& 9rom 0.1 and 0(1& 9rom 011 and 021, by @xercise 2&(&10a1&

/= 0-1 0=1 0C1 0,1 0101 0111 C 5 ( B C 5 ( A B )C ) ( B 5 A B) 5 ( BC 5 ( A B )C ) A 5 ( B 5 A B ) A 5 ( B C 5 ( A B )C ) AC 5 ( BC 5 ( A B )C ) ( A C )( B C ) 5 ( A B )C 9rom 011-0/1, by Deduction Theorem 1& @xercise 2&(&(0b1& 6xiom +(& 9rom 0C1 and 0=1& 9rom 0,1 and 0-1& 9rom 0101, by @xercise 2&2&(0a1&

?o , e must pro#e that U- ( A B )C 5 ( A C )( B C ) & 011 %ro#e U021 %ro#e U0(1 A 5 ( C 5 ( AC )( B C )) B 5 ( C 5 ( AC )( B C ))

%ro#e U( A B)C 5 ( AC )( BC )

!7ercise ..).-. %ro#e the abo#e 011, 021 and 0(1& 9inally, e must pro#e that U- ( A C )( B C ) 5 ( A B )C & !7ercise ..)./. %ro#e that&

..*. :ormulas ontaining 4egation E Minimal Logic


9heorem ..*.'& a1 7f V+1, +2, +,, 5%W: 61, 62, &&&, 6n, G U- <, and V+1, +2, +,, 5%W: 61, 62, &&&, 6n, G U- f<, then V+1, +2, +,, 5%W: 61, 62, &&&, 6n U- fG& 2hat does this mean8 b1 V+(, +., +,, 5%W: U- ( A A) & 2hat does it mean8 7tBs the so-called

/C La8 of 4onB ontradiction. Proof. a1 Gy Deduction Theorem 1, 61, 62, &&&, 6n U- G`<, and 61, 62, &&&, 6n U- G`f<& +et us continue this proof by addin! the axiom L(: 0G`<1`00G`f<1`fG1 as the next step& 6fter this, by applyin! 5% t ice e obtain fG& D&@&D& b1 'ee @xercise 1&.&2 0e1& !7ercise ..*.'. a1 0for smart students1 7n#esti!ate the si:e 0the number of formulas1 of the proof of V+1, +2, +,, 5%W: 61, 62, &&&, 6n, U- fG as a function f0k, m1 of the si:es k, m of the proofs of V+1, +2, +,, 5%W: 61, 62, &&&, 6n, G U< and 1, 62, &&&, 6n, G U- f<& Kou may ish to report your result& 2e ill publish your report on the eb as an appendix to this book& The current record holder is 6i!a )omane, 200C: f0k, m1 c (0k\m1\=& 7mpro#e this result, or pro#e that it is the best possible one& b1 V+1, +2, +,, 5%W: 6, fG U- f06`G1& Rr, V+ 1-+., +,, 5%W: UA B 5 7( A 5 B) & 2hat does it mean8 c1 U- V+1, +2, +,, 5%W: 06`f61`f6& 2hat does it mean8 3ttention= nonBconstructive reasoning6 7n 'ection 2&-, e ill use the classical lo!ic V+1-+11, 5%W to pro#e the con#erse formula of 0c1: 7 ( A 5 B ) 5 A7 B , i&e& the eQui#alence 7 ( A 5 B ) 6 A7 B & This formula cannot be pro#ed in the constructi#e lo!ic V+ 1-+10, 5%W 0see 'ection 2&C1& 9heorem ..*... V+1, +2, +,, 5%W: U- 06`G1`0fG`f61& 2hat does it mean8 7tBs the so-called ontra"osition La8& 4ote& The follo in! form of Theorem 2&.&2 is called Modus 9ollens: V+1, +2, +,, 5%W: U- 6`G, fG U- f6& 3ttention= nonBconstructive reasoning6 7n 'ection 2&-, e ill use the classical lo!ic V+1-+11, 5%W to pro#e the con#erse formula 0fG`f61`06`G1, i&e& the eQui#alence 06`G1a0fG`f61& 2e ill see also that these formulas cannot be pro#ed in the constructi#e lo!ic V+ 1-+10, 5%W 0see 'ection 2&C1& !7ercise ..*... a1 %ro#e Theorem 2&.&2& b1Verify that, in our axiom system, the +a of ?on-<ontradiction and the <ontraposition +a could be used instead of the axiom + ,& 5ore precisely:

/, pro#e +, in the lo!ic V+1-+/, +a of ?on-<ontradiction, <ontraposition +a , 5%W& Ge careful: do not use theorems dependin! on the axiom +,& 9heorem ..*.). V+1-+,, 5%W: U- 06`fG1a0G`f61& 2hat does it mean8 9irst e pro#e that U- 06`fG1`0G`f61& 011 6`fG 021 G 0(1 06`G1`006`fG1`f61 0.1 6`G 0/1 06`fG1`f6 0-1 f6 Hypothesis& Hypothesis& 6xiom L(: 0G`<1`00G`f<1`fG1 ith G ^ 6, < ^ G& 9rom 021 by 6xiom +1 and 5%& 9rom 0(1 and 0.1& 9rom 011 and 0/1&

Thus, by Deduction Theorem 1, U- 06`fG1`0G`f61& Gy s appin! 6 and G e obtain the con#erse formula: U- 0G`f61`06`fG1& D&@&D& 3ttention= nonBconstructive reasoning6 1arning6 The 0#ery similar to Theorem 2&.&(1 formula 0f6`G1a0fG`61 can be pro#ed only in the classical lo!ic; 9heorem ..*.*. V+1, +2, +,, 5%W: U- 6`ff6& 2hat does it mean8 011 6 021 0f6`61`00f6`f61`ff61 0(1 6`0f6`61 0.1 f6`6 0/1 0f6`f61`ff6 0-1 ff6 Hypothesis& 6xiom L(& 6xiom L'& 9rom 011 and 0(1 by 5%& 9rom 021 and 0.1 by 5%& 9rom 0/1 and Theorem 1&.&1 by 5%&

3ttention= nonBconstructive reasoning6 7n 'ection 2&-, e ill use the classical lo!ic V+1-+11, 5%W to pro#e the con#erse formula U- ff6`6, i&e& the eQui#alence U- ff6a6 0the so-called Double 4egation La81& 2e ill see also 0'ection 2&C1 that these formulas cannot be pro#ed in the constructi#e

-0 lo!ic V+1-+10, 5%W& 'till, in the minimal lo!ic e can pro#e 0Grou er, 1,2(81: 9heorem ..*.-. V+1, +2, +,, 5%W: U- fff6af6& 2hat does it mean8 7ndeed, by Theorem 2&.&., U- f6`fff6& Gy the <ontraposition +a 0Theorem 2&.&21, U- 06`ff61`0fff6`f61& Hence, by Theorem 2&.&., Ufff6`f6& D&@&D& Theorem 2&.&/ 0and some of the follo in! formulas in this and in the next section containin! double ne!ations1 may seem uninterestin! to people belie#in! unconditionally in the eQui#alence ff6a6& 'till, it seems interestin! 0at least 3 for a mathematician1 to obtain a !eneral characteri:ation of lo!ical formulas that do not depend on the +a of @xcluded 5iddle& 7n 'ection 2&= e ill use these formulas to pro#e the ele!ant and non-tri#ial Ali#enkoBs theorem: a1 6 is pro#able in the classical propositional lo!ic 0i&e& in V+1-+11, 5%W1, if and only if ff6 is pro#able in the constructi#e propositional lo!ic 0i&e& in V+1-+10, 5%W1, b1 f6 is pro#able in the classical propositional lo!ic, if and only if f6 is pro#able in the constructi#e propositional lo!ic& 9heorem ..*./. a1 V+1, +2, +,, 5%W: U- 0f6`61`ff6& 2hat does it mean8 b1 V+1, +2, +-, +=, +,, 5%W: U- ff06#f61& 2hat does it mean8 7n this eak form, the +a of @xcluded 5iddle can be Spro#ed constructi#elyS& The formula 77 ( A7 A) can be pro#ed in the constructi#e lo!ic, but A7 A canBt 3 as e ill see in 'ection 2&C& !7ercise ..*.). %ro#e 0a1 and 0b1 of Theorem 2&.&-& The axiom +11 canBt be used in these proofs; 0Hint for 0b1: deri#e a contradiction from f06#f61&1 9heorem ..*.0. V+1-+,, 5%W: a1 U- 06`G1`0ff6`ffG1& 2hat does it mean8 b1 U- ff06`G1`0ff6`ffG1& 2hat does it mean8 c1 U- 06`0G`<11`0ff6`0ffG`ff<11& 2hat does it mean8 d1 ff06`G1, ff0G`<1 U- ff06`<1& 2hat does it mean8 e1 ff6, ff06`G1 U- ffG& 2hat does it mean8 The con#erse of 0a1: 0ff6`ffG1`06`G1 cannot be pro#ed in the constructi#e lo!ic 0see 'ection 2&C1& To pro#e 0a1, e must simply apply t ice the <ontraposition +a : 06`G1`0fG`f61`0ff6`ffG1& 6nd, of course, 0e1 is an easy conseQuence of 0b1&

-1 ?o , let us pro#e 0b1& 011 021 ff06`G1 ff6 Hypothesis& Hypothesis& 9rom 0a1, by transposin! 6`G and ff6, by the %remise %ermutation +a & 9rom 021 and 0(1& Gy the <ontraposition +a & 9rom 0.1 and 0/1& Gy the <ontraposition +a & 9rom 0-1 and 0=1& 9rom 011 and 0C1& Gy Theorem 2&.&/& 9rom 0,1 and 0101&

0(1

ff6`006`G1`ffG1

0.1 0/1 0-1 0=1 0C1 0,1

06`G1`ffG 006`G1`ffG1`0fffG`f06`G11 fffG`f06`G1 0fffG`f06`G11`0ff06`G1`ffffG1 ff06`G1`ffffG ffffG

0101 ffffG`ffG 0111 ffG +et us pro#e 0c1& 011 6`0G`<1 021 ff6 0(1 ffG 0.1 ff6`ff0G`<1 0/1 ff0G`<1 0-1 ffG`ff< 0=1 ff< Hypothesis& Hypothesis& Hypothesis&

Thus, by Deduction Theorem 1, U- ff06`G1`0ff6`ffG1&

9rom 011, by 0a1& 9rom 021 and 0.1& 9rom 0/1, by 0b1& 9rom 0(1 and 0-1&

Thus, by Deduction Theorem 1, U- 06`0G`<11`0ff6`0ffG`ff<11&

-2 ?o e can pro#e 0d1& 9irst, let us take 0c1 6`<: ith 6 ^ 6`G, G ^ G`<, < ^

011U- 006`G1`00G`<1`06`<111`0ff06`G1`0ff0G`<1`ff06`<111& 021 U- 06`G1`00G`<1`06`<1 0(1 ff06`G1 0.1 ff0G`<1 0/1 ff06`<1 Gy transiti#ity of implication and Deduction Theorem 1& Hypothesis& Hypothesis& 9rom 011, 0(1 and 0.1&

9heorem ..*.G. V+1-+,, 5%W: a1 U- 77 ( A B ) 6 ( 77 A77 B ) & 2hat does it mean8 b1 U- 77 A77 B 5 77 ( A B ) & 2hat does it mean8 3ttention= nonBconstructive reasoning6 The con#erse of 0b1: 77 ( A B ) 5 77 A7 7 B cannot be pro#ed in the constructi#e lo!ic 0see 'ection 2&C1& 2hat does it mean8 7f e simply succeed in deri#in! a contradiction from 7 ( A B ) , then, perhaps, e do not ha#e a method allo in! to decide, hich part of 77 A77 B is true 3 ff6, or ffG8 !7ercise ..*.*& %ro#e Theorem 2&.&C& 0Hint: use the result of @xercise 2&2&(0a1, if needed&1 9heorem ..*.(. V+1, +2, +,, 5%W U- f6`06`fG1 0compare 1&.&20d11& 2hat does it mean8 ith @xercise

7tBs a eak form of the Scra:yS axiom L'I: f6`06`G1& This axiom says: S<ontradiction implies anythin!S& 7n the minimal lo!ic e can pro#e /0T of +10: S<ontradiction implies that all is ron!S& Rf course, this /0T-pro#ability of +10 decreases the si!nificance of the minimal lo!ic accordin!ly& Proof& 'ee @xercise 2&.&/& 9heorem ..*.'I. V+1-+,, 5%W: a1 U- 7 A7 B 5 7 ( A B ) & 7tBs a half of the so-called :irst de Morgan La8& 2hat does it mean8 b1 U- 7 ( A B ) 6 7 A7 B & 7tBs the so-called ,econd de Morgan La8& 2hat does it mean8 3ttention= nonBconstructive reasoning6 The second half of 0a1 3 the con#erse implication, i&e& the eQui#alence 7 ( A B ) 6 7 A7 B can be pro#ed in the classical lo!ic, yet not in the constructi#e lo!ic 0see 'ection 2&C1& @xplain,

-( hy&
6u!ustus de 5or!an 01C0--1C=11: SHe reco!nised the purely symbolic nature of al!ebra and he as a are of the existence of al!ebras other than ordinary al!ebra& He introduced de 5or!anBs la s and his !reatest contribution is as a reformer of mathematical lo!ic&S 0accordin! to 5acTutor History of 5athematics archi#e1&

*se <ontraposition +a to pro#e 0a1 and 0b`1 in @xercise 2&.&/& +et us pro#e 0bj1& 001 7 A 7 B Hypothesis& 9rom 001, by 6xiom +(& 9rom 001, by 6xiom +.& 9rom 011, by Theorem 2&.&,: f6`06`f<1& < is any formula& 9rom 021, by Theorem 2&.&,: fG`0G`f<1& < is any formula& 9rom 0(1 and 0.1, by 6xiom +C: ( A 5 7C ) 5 (( B 5 7C ) 5 ( A B 5 7C )) & )epeat 0(1-0/1 ith ff< instead of f<& 9rom 0/1 and 0-1, by 6xiom +,: ( A B 57 C )5 (( A B 5 77C ) 57 ( A B ))

011 f6 021 fG 0(1 6`f< 0.1 G`f< 0/1 0-1 0=1 A B 57 C A B 57 7C

7 ( A B )

Thus, by V+1, +2W Deduction Theorem 1, V+1-+,, 5%W U- 7 A7 B 5 7 ( A B ) & !7ercise ..*.-. %ro#e: a1 Theorem 2&.&,& b1 0a1 and 0b`1 of Theorem 2&.&10& 0Hint: use <ontraposition +a 1& c1 V+1-+,, 5%W: U- ( A 5 B ) 5 7 ( A7 B ) & 2hat does it mean8 <ompare ith @xercise 2&.&1& d1 V+1-+C, 5%W: U- A B 5 (( A 5 B ) 5 B ) & 2hat does it mean8 3ttention= nonBconstructive reasoning6 The con#erse implication of 0a1, 7 ( A7 B ) 5 ( A 5 B) cannot be pro#ed in the constructi#e lo!ic 0see 'ection 2&C1& @xplain, hy& 'till, e ill pro#e this formula in the classical

-. lo!ic& The con#erse of 0b1: (( A5 B ) 5 B ) 5 A B cannot be pro#ed in the constructi#e lo!ic 0see 'ection 2&C1& @xplain, hy& 'till, e ill pro#e this formula in the classical lo!ic&

..-. :ormulas ontaining 4egation E onstructive Logic


7n this book, constructive logic is used as a synonym of intuitionistic logic; <onstructi#e lo!ic includes the Scra:yS axiom L'I: fG`0G`<1, but re>ects the +a of @xcluded 5iddle L'': G#fG as a !eneral lo!ical principle& 9heorem ..-.'. a1 V+10, 5%W: 6, f6 U- G& 2hat does it mean8 b1 V+1, +2, +C, +10, 5%W: UA B 5 ( 7 A5 B ) & 2hat does it mean8

c1 V+1, +C, +10, 5%W: U- 7 A B 5 ( A5 B ) & 2hat does it mean8 Rf course, 0a1 follo s directly from +10, by 5%& !7ercise ..-.'. %ro#e 0b1 and 0c1 of Theorem 2&/&1& ?ote: hen pro#in! 0c1, you cannot use Deduction Theorem 1 0because of the missin! axiom + 21& 'o, simply build a seQuence of / formulas representin! the proof of 0c1& 3ttention= nonBconstructive reasoning6 The con#erse of 0b1, i&e& ( 7 A5 B ) 5 A B cannot be pro#ed in the constructi#e lo!ic 0see 'ection 2&C1& @xplain, hy& The con#erse of 0c1, i&e& ( A 5 B ) 5 7 A B cannot be pro#ed in constructi#e lo!ic 0see 'ection 2&C1& @xplain, hy& 'urprisin!ly, 0b1, i&e& the rule A B , 7 A U- G seems to be a Quite a SnaturalS lo!ical principle, yet it cannot be pro#ed ithout the Scra:yS axiom + 10; 2hy not8 Gecause it implies +10; 7ndeed, 011

A B 5 ( 7 A5 B )

Hypothesis& Hypothesis& Hypothesis&

021 f6 0(1 6 0.1 0/1 0-1 G A 5 A B A B

6xiom +-& Gy 5%, from 0(1 and 0.1& Gy 5%, from 011, 0/1 and 021&

-/ Hence, by Deduction Theorem 1, V+1, +2, +-, 5%W: A B 5 ( 7 A5 B ) Uf6`06`G1& 7n 'ection 2&C e ill pro#e that +10 cannot be deri#ed from +1-+,, hence, 0b1 also cannot be deri#ed from +1-+, 0i&e& ithout +101& 9heorem ..-... V+1-+10, 5%W: a1 U- 0ff6`ffG1`ff06`G1& 7tBs the con#erse of Theorem 2&.&=0b1& Hence, V+1-+10, 5%W: U- ff06`G1a0ff6`ffG1& b1 U- ff6`0f6`61& 7tBs the con#erse of Theorem 2&.&-0a1& Hence, V+1-+10, 5%W: U- ff6a0f6`61& c1 UA7 A 5 ( 7 7 A 5 A ) & 2hat does it mean8 d1 U- ff0ff6`61& 2hat does it mean8 Rf course, 0b1 is an instance of the a7iom L'I. To pro#e 0a1, let us pro#e that V+ 1-+10, 5%W: ff6`ffG, f06`G1 U- fG, ffG& Then, by Theorem 2&.&1, 0a1 U- 0ff6`ffG1`ff06`G1& !7ercise ..-..& a1 %ro#e that V+1-+10, 5%W: ff6`ffG, f06`G1 U- fG, ffG& b1 %ro#e 0c1 and 0d1 of Theorem 2&/&2& !7ercise ..-.). %ro#e that in V+1-+10, 5%W: a1 6 U- B 6 B7 A & 2hat does it mean8 b1 U- B( A7 A) 6 B & 2hat does it mean8 c1 U- (( A7 A) B)C 6C & 2hat does it mean8

../. :ormulas ontaining 4egation E lassical Logic


7f you a!ree to adopt the formula B7 B , i&e& the +a of @xcluded 5iddle 06xiom +11 in the list of 'ection 1&(1, you can pro#e, first of all, the so called Double 4egation La8: 9heorem ../.'. V+1-+11, 5%W: U- ff6 ` 6& Hence, V+1-+11, 5%W: U- ff6 a 6& 7ndeed, by Theorem 2&/&2, V+1-+10, 5%W: U- A7 A 5 ( 7 7 A 5 A ) , hence, V+1-+11, 5%W: U- ff6`6& D&@&D&

-7n the minimal lo!ic e pro#ed Theorem 2&.&.: V+ 1, +2, +,, 5%W: U- 6`ff6& Hence, V+1-+11, 5%W: U- ff6 a 6& 3ttention= nonBconstructive reasoning6 The formula ff6`6 cannot be pro#ed in the constructi#e lo!ic, see 'ection 2&C& 2hy8 Gecause it represents a kind of non-constructi#e reasonin!& 7ndeed, ima!ine, you ish to pro#e that xG0x1& 6ssume the contrary, fxG0x1, and deri#e a contradiction& Thus you ha#e pro#ed&&& the ne!ation of fxG0x1, i&e& f fxG0x1& To conclude xG0x1 from f fxG0x1, you need the Double ?e!ation +a & Hence, by adoptin! this la as a lo!ical principle, you ould allo non-constructi#e existence proofs 3 if you pro#e xG0x1 by assumin! fxG0x1, and deri#in! a contradiction, then you may not obtain a method allo in! to find a particular x satisfyin! G0x1& !7ercise ../.'. %ro#e that V+C, +11, 5%W: 6`G, f6`G U- G& Rr, by Deduction Theorem 1, V+1, +2, +C, +11, 5%W: 06`G1`00f6`G1`G1& 2hat does it mean8 This formula cannot be pro#ed in the constructi#e lo!ic 0see 'ection 2&C1& @xplain, hy& 7n the classical lo!ic, you can pro#e also the full form of the La8= 9heorem ../... V+1-+11, 5%W: U- 06`G1 a 0fG`f61& 2e pro#ed a half of this +a in the minimal lo!ic as Theorem 2&.&2: V+ 1, +2, +,, 5%W: U- 06`G1`0fG`f61& +et us pro#e the remainin! half: V+ 1-+11, 5%W U- 0fG`f61 ` 06`G1& 011 021 0(1 0.1 0/1 0-1 fG`f6 6 ff6`ffG 6`ff6 ffG`G G Hypothesis& Hypothesis& 9rom 011, by the first half& Double ?e!ation +a & Double ?e!ation +a & 9rom 0.1, 0(1 ans 0/1& ontra"osition

Gy Deduction Theorem 1, V+1-+11, 5%W U- 0fG`f61 ` 06`G1& 3ttention= nonBconstructive reasoning6 The formula 0fG`f61 ` 06`G1 cannot be pro#ed in the constructi#e lo!ic, see 'ection 2&C& @xplain, hy& !7ercise ../.'3. %ro#e that in V+1-+11, 5%W: a1 U- 0f6`G1a0fG`61 0compare ith Theorem 2&.&(1&

-= b1 U- 06`G1`00f6`fG1`0Ga611& 3ttention= nonBconstructive reasoning6 These t o formulas cannot be pro#ed in the constructi#e lo!ic, see 'ection 2&C& 9heorem ../.). V+1-+11, 5%W: U- ( A B ) 6 A B & 7tBs the so-called :irst de Morgan La8. 6 half of this +a e pro#ed in the minimal lo!ic as Theorem 2&.&100a1: V+ 1+,, 5%W U- 7 A7 B 5 7 ( A B ) & +et us pro#e the remainin! half: V+1-+11, 5%W U- 7 ( A B ) 5 7 A7 B & 3ttention= nonBconstructive reasoning6 This formula cannot be pro#ed in the constructi#e lo!ic, see 'ection 2&C& @xplain, hy& +et us start by pro#in! 7 ( 7 A7 B ) 5 A B & 011 021 0(1 0.1

7 ( 7 A7 B ) 7 ( 7 A' 7 B ) 5 77 A77 B
77 A77 B 5 77 ( A B ) 77 ( A B )

Hypothesis& Gy the 'econd de 5or!an +a -Theorem 2&.&100b1& Theorem 2&.&C0a1& V+1-+,, 5%W; 9rom 011, 021 and 0(1&

Thus, by Deduction Theorem 1, V+1-+,, 5%W U- 7 ( 7 A7 B ) 57 7 ( A B) & Gy applyin! the first half of the <ontraposition +a 0pro#able in the minimal lo!ic1: V+1-+,, 5%W U- 777 ( A B ) 57 7 (7 A7 B ) & Gy Theorem 2&.&/: V+1-+,, 5%W U- 7 ( A B ) 5 777 ( A B ) , hence, V+1-+,, 5%W U- 7 ( A B ) 5 77 ( 7 A7 B) & ?o , by the Double ?e!ation +a , V+1-+11, 5%W U- 77 ( 7 A7 B) 5 7 A7 B , hence, V+1-+11, 5%W U- 7 ( A B ) 5 7 A7 B & D&@&D& 7n the classical lo!ic, e can express implication by ne!ation and dis>unction& 7ndeed, e already kno that V+1-+10, 5%W: U- 7 A B 5 ( A5 B ) 0Theorem 2&/&10c11& 9heorem ../.*. a1 V+1-+C, 5%W: AC U- ( A 5 B ) 5 B C & Hence, V+1+C, 5%W: A7 A U- ( A 5 B ) 5 7 A B & b1 V+1-+11, 5%W: U- ( A 5 B ) 6 7 A B & Rf course, 0b1 follo s from 0a1 and Theorem 2&/&10c1& +et us pro#e 0a1&

-C 011 6, 6`G U- G 021 6, 6`G UBC Gy 6xiom +-& Gy Deduction Theorem 1&

0(1 6 U- ( A 5 B ) 5 B C 0.1 <, 6`G U- < 0/1 <, 6`G UB C

Gy 6xiom +=& Gy Deduction Theorem 1& Gy 6xiom +C&

0-1 < U- ( A 5 B ) 5 B C 0=1 6#< U- ( A 5 B ) 5 B C

!7ercise ../..& %ro#e that in V+1-+11, 5%W: a1 U- B( A7 A) 6 B & 2hat does it mean8 b1 U- (( A7 A) B )C 6C & 2hat does it mean8 c1 U- (( A5 B ) 5 B ) 5 A B & 2hat does it mean8 Hence, by @xercise 2&.&/0d1, V+1-+11, 5%W: U- (( A5 B ) 5 B ) 6 A B & !7ercise ../.)& %ro#e that in V+1-+11, 5%W: a1 U- ( A 5 B ) 6 7 ( A7 B ) & 2hat does it mean8 b1 U- 7 ( A 5 B ) 6 A7 B & 2hat does it mean8 c1 U- A B 6 ( 7 A5 B ) & 2hat does it mean8 d1 U- A B 67 ( A5 7 B) & 2hat does it mean8 e1 0for smart students1 Try detectin!, hich parts of these eQui#alences are pro#able: 11 in the minimal lo!ic, 21 in the constructi#e lo!ic& ,trange formulas !7ercise ../.*& %ro#e in V+1-+11, 5%W the follo in! stran!e formulas: a1 U- A( A5 B ) & 2hat does it mean8 Does it mean anythin! at all8 b1 U- ( A 5 B )( B 5 A) & 2hat does it mean8 Does it mean anythin! at all8 The most cra:y theorem of the classical propositional lo!ic8 c1 U- (( A5 B ) 5 A )5 A & 2hat does it mean8 Does it mean anythin! at all8 7t is the so-called PeirceJs La8 from:
. ,. Peirce. Rn the al!ebra of lo!ic: 6 contribution to the philosophy of notation& American

-,
/o$rna% of .athematics, 1CC/, #ol&=, pp&1C0-202&

..0. onstructive !mbedding. 5livenkoJs 9heorem


+et us recall some of the results of pre#ious sections concernin! double ne!ations: 9heorem ..*.*. V+1, +2, +,, 5%W: U- 6`ff6& 9heorem ..*.-. V+1-+,, 5%W: U- fff6af6& 9heorem ..*./CbD. V+1-+,, 5%W: U- 77 ( A7 A) & 7n this +a of @xcluded 5iddle can be Spro#ed constructi#elyS& 9heorem ..*.0. V+1-+,, 5%W: a1 U- 06`G1`0ff6`ffG1& b1 U- ff06`G1`0ff6`ffG1& c1 U- 06`0G`<11`0ff6`0ffG`ff<11& d1 ff06`G1, ff0G`<1 U- ff06`<1& e1 ff6, ff06`G1 U- ffG& 9heorem ..*.G. V+1-+,, 5%W: a1 U- 77 ( A B ) 6 (77 A77 B ) & b1 U- 77 A77 B 5 77 ( A B ) & 9heorem ..-... V+1-+10, 5%W: a1 U- 0ff6`ffG1`ff06`G1& 7tBs the con#erse of Theorem 2&.&=0b1& d1 U- ff0ff6`61& 9heorem ../.'. V+1-+11, 5%W: U- ff6 a 6& Does it mean that for any formula 6: if V+1-+11, 5%W: U- 6, then V+1-+10, 5%W: U- ff68 0The con#erse is ob#ious: if V+1-+10, 5%W: U- ff6, then V+1-+11, 5%W: U- 6 by Theorem 2&-&1&1 7ma!ine, e ha#e a proof of V+1-+11, 5%W: U- 6& 7t is a seQuence of formulas )1, )2, &&&, )n, here )n ^ 6& 7f this seQuence does not contain instances of the axiom +11, then it is a proof of V+1-+10, 5%W: U- 6 as ell& Hence, accordin! to Theorem 2&.&., V+1-+10, 5%W: U- ff6 7f the seQuence )1, )2, &&&, )n contains some instances of +11, i&e& formulas eak form, the

=0 ha#in! the form B7 B , then, accordin! to Theorem 2&.&-0b1, e could try replacin! each such formula by a seQuence pro#in! that V+ 1-+,, 5%W: U77 ( B 7 B ) & 7t appears that each of the formulas ff)1, ff)2, &&&, ff)n is pro#able in V+1-+10, 5%W& a1 7f )k is an instance of the axioms + 1-+10, then V+1-+10, 5%W: U- ff)k 0Theorem 2&.&.1& b1 7f )k is an instance of the axiom +11, then V+1-+10, 5%W: U- ff)k 0Theorem 2&.&-0b11& c1 ?o , let us assume that i, > _ k, and ) i, )> U- )k directly by 5%, i&e& )> is )i`)k& 2e kno already that V+1-+10, 5%W: U- ff)i and V+1-+10, 5%W: Uff0)i`)k1& Gy Theorem 2&.&=0b1, V+1-+,, 5%W: Uff0)i`)k1`0ff)i`ff)k1& Hence, V+1-+10, 5%W: U- ff)k& Gecause 6 ^ )n, 1,2,: e ha#e pro#ed the remarkable Ali#enkoBs theorem from

V.5livenko& 'ur QuelQues points de la lo!iQue de 5& Grou er& Aca+emie )oya%e +e Be%gi4$e, B$%%etins +e %a c%asse +es sciences, 1,2,, ser&/, #ol&1/, pp&1C(-1CC& Valery 7#ano#ich Ali#enko 01C,--1,.0, see http:PP &math&ruPhistoryPpeopleP!li#enko, in )ussian1 is best kno n by the so-called Ali#enko-<antelli theorem in probability theory&

5livenkoJs 9heorem& V+1-+11, 5%W: U- 6, if and only if V+ 1-+10, 5%W: U- ff6& Rr: a formula 6 is pro#able in the classical propositional lo!ic, if and only if its double ne!ation ff6 is pro#able in the constructi#e propositional lo!ic& This theorem pro#ides a kind of a Sconstructi#e embeddin!S for the classical propositional lo!ic: any classically pro#able formula can be Spro#edS in the constructi#e lo!ic, if you simply put t o ne!ations before it& orollary& V+1-+11, 5%W: U- f6, if and only if V+1-+10, 5%W: U- f6& Rr: a Sne!ati#eS formula f6 is pro#able in the classical propositional lo!ic, if and only if it is pro#able in the constructi#e propositional lo!ic& 7ndeed, if V+1-+11, 5%W: U- f6, then by Ali#enkoBs theorem, V+ 1-+10, 5%W: Ufff6, and by Theorem 2&.&/, V+1-+10, 5%W: U- f6& D&@&D& !7ercise ..0.'. %ro#e the follo in! #ersion of Ali#enkoBs theorem 0see $leene V1,/2W1: a1 7f V+1-+11, 5%W: 61, 62, &&&, 6n U- <, then V+1-+10, 5%W: ff61, ff62, &&&, ff6n U- ff<&

=1 b1 7f V+1-+11, 5%W: f61, f62, &&&, f6n, G1, G2, &&&, Gp U- f<, then V+1-+10, 5%W: f61, f62, &&&, f6n , ffG1, ffG2, &&&, ffGp U- f<&

..G. 37iom Inde"endence. Using om"uters in Mathematical Proofs


7f one of our axioms +i could be pro#ed by usin! the remainin! n>; axioms, then e could simplify our lo!ical system by droppin! + i as an axiom& 6 strikin! example: 9heorem ..G.'& The axiom L(: 06`G1`006`fG1`f61 can be pro#ed in V+1, +2, +C, +10, +11, 5%W&
This fact as established by 6u!usts $urmitis 0on the eb, also: 6& 6& $urmit1: 3. 3. Kurmitis& Rn independence of a certain axiom system of the propositional calculus& Proc( Lat'ian State 2ni'ersity, 1,/C, Vol& 20, ?(, pp& 21-2/ 0in )ussian1&

The follo in! proof of +, in V+1, +2, +C, +10, +11, 5%W is due to Hanis 'edols 01,(,-20111& 9irst, let us establish that the formula 06`f61`f6 can be pro#ed in V+ 1, +2, +C, +10, +11, 5%W 0in @xercise 2&.&1 e established that V+ 1, +2, +,, 5%W: U06`f61`f61: 011

( A 5 7 A) 5 (( 7 A 5 7 A) 5 ( A7 A) 57 A ) 6xiom +C&
Hypothesis& This is pro#able in V+1, +2, 5%W 0Theorem 1&.&11& 6xiom +11& 9rom 011, 021, 0(1 and 0.1, by 5%&

021 6`f6 0(1 f6`f6 0.1 A' 7 A

0.1 f6

=2 Gy Deduction Theorem 1 0 hich is #alid for any lo!ical system containin! V+1, +2, 5%W1&

0-1 06`f61`f6

?o let us establish that in V+1, +2, +10, 5%W: 6`G, 6`fG U- 6`f6& 0=1 0C1 0,1 0,1 6`G 6`fG 6 G Hypothesis& Hypothesis& Hypothesis& 9rom 0=1, 0,1, by 5%& 9rom 0C1, 0,1, by 5%& 6xiom +10& 9rom 0,1, 0101 and 0111 by 5%& Gy Deduction Theorem 1 0 hich is #alid for any propositional system containin! V+1, +2, 5%W1&

0101 fG 0111 fG`0G`f61 0121 f6 01(1 6`G, 6`fG U- 6`f6

9inally, let us mer!e the proofs of 0-1 and 01(1, then by 5% e obtain f6, i&e& V+1, +2, +C, +10, +11, 5%W: 6`G, 6`fG U- f6& ?o , by Deduction Theorem 1 0 hich is #alid for any propositional system containin! V+1, +2, 5%W1 e obtain the axiom +,: V+1, +2, +C, +10, +11, 5%W: 06`G1`006`fG1`f61& D&@&D& 2hat should e do after establishin! that one of our axioms is SdependentS8 Do you think, e should drop +, as an axiom of our lo!ical system8 9irst, letBs note that 8e have "roved L( by using three "roblematic a7ioms: +1, +10, +11& Gut +, itself is not problematic; 'econdly, +, cannot be pro#ed in V+1-+C, +10, 5%W 0see Theorem 2&C&2 belo 1& Hence, if e ould drop +,, then, instead of a simple definition classical lo!ic ^ constructi#e lo!ic \ +11,

=( e ould ha#e a more complicated one: constructi#e lo!ic ^ classical lo!ic 3 +11 \ +,& ?o , the Question of Questions: Is La8 of !7cluded Middle an inde"endent logical "rinci"le2 7&e&, could e pro#e the +a of @xcluded 5iddle 0the axiom + 11: B7 B 1 by usin! the other axioms 0i&e& V+1-+10, 5%W1 as e pro#ed +, in V+1, +2, +C, +10, +11, 5%W8 7f not, ho could e demonstrate that this is impossible at all8 Ho could e demonstrate that some lo!ical principle is inde"endent, i&e& that it cannot be deri#ed from other principles8 +et us assume, e ha#e an al!orithm 4 computin! for each formula 6 some its SpropertyS Q061 such that: a1 Q0+11 is true, Q0+21 is true, &&&, Q0+101 is true 0i&e& the axioms +1-+10 possess property 41& b1 7f Q061 is true and Q06`G1 is true, then Q0G1 is true 0i&e& 5% Spreser#esS property Q1& Hence, Q091 is true for all the formulas 9 that are pro#able in V+ 1+10, 5%W& c1 Q0+111 is false 0+11 does not possess property 41& 7f e could obtain such a property 4, then, of course, this ould demonstrate that +11 cannot be pro#ed in V+1-+10, 5%W, i&e& that the La8 of !7cluded Middle is an inde"endent logical "rinci"le& The most popular ay of introducin! such properties of formulas are the socalled Smulti-#alued lo!icsS or Smany-#alued lo!icsS, introduced by Han +ukasie ic: and @mil %ost:
K.Lukasie8ic;. R lo!ice tro> artoscio e>& )$ch Fi%o3ofic3ny L-o-!, 1,20, #ol& /, pp& 1-,1=1 !.Post& 7ntroduction to a !eneral theory of elementary propositions ( Amer( Ao$rn( math(, 1,21, #ol& 21, pp&1-(-1,/ )ead more: 5any-Valued +o!ic by 'ie!fried Aott ald in 'tanford @ncyclopedia of %hilosophy&

9or example, let us consider a kind of Sthree-#alued lo!icS, here 0 means SfalseS, 1 3 Sunkno nS 0or ?*++ 3 in terms of 'D+1, 2 3 StrueS& Then it ould be natural to define con>unction and dis>unction as A B ^ min06,G1 A B ^ max06,G1&

=. Gut ho should e define implication and ne!ation8 6 G 0 0 0 1 0 2 1 0 1 1 1 2 2 0 2 1 2 2 A B 0 0 0 0 1 1 0 1 2 A B 6`G 0 1 2 1 1 2 2 2 2 i1 i2 i( i. i/ ii= iC i,

6 f6 0 i10 1 i11 2 i12 Thus, theoretically, e ha#e here to explore: (, ^ 1,-C( #ariants of implication definitions and (( ^ 2= ne!ation definitions& Do you think, it ould be natural to set the #alues of f6 as follo s8 6 f6 0 2

1 1 2 0 e ould try buildin! a SnaturalS three-#alued lo!ic, in hich S1S

Kes, if

=/ ould mean, indeed, Sunkno nS& To fill in the MnaturalN table of three-#alued implication, e could use, for example, the classical eQui#alence 06`G1af6#G& 7n this ay e could obtain the MnaturalN three-#alued lo!ic used, for example, for handlin! of ?*++-#alues in 'D+& Ho e#er, our aim is here, in a sense, >ust the opposite of MnaturalN& 2e consider Sunder the abo#e truth tables, formula 6 al ays takes StrueS #aluesS as a kind of the abo#e-mentioned SpropertyS Q061& Hence, e ill try to define the tables for implication and ne!ation in such a ay that: a1 the axioms +1, +2, &&&, +10 al ays take StrueS #alues 0i&e& 21, b1 5odus %onens Spreser#esS takin! al ays StrueS #alues 0i&e& if the formulas 6 and 6`G are al ays 2, then G also is al ays 21, c1 the axiom +11 sometimes takes the #alues 0 or 1& Gecause of S#iolatin!S +11, the definitions of implication and ne!ation, ha#in! these properties, cannot be 100T natural& 'o, e must explore 0at least some of1 the SunnaturalS #ersions as ell& !7ercise ..G.'& De#elop a simple 0recursi#e1 computer pro!ram recei#in! as input: a1 6ny such Struth tablesS& b1 6ny formula 9 consistin! of letters 6, G, <, and propositional connecti#es& and printin! out Struth #aluesS of the formula 9, for example, if 9 ^ G`06`G1: 6 G G`06`G1 0 0 0 1 0 2 1 0 1 1 1 2 2 0 2 2 2 2 2 2 2 ill

=2 1 2 2 2 2

7n this example the axiom +1 al ays takes StrueS #alues& %erhaps, e should be interested only in those #ariants of our Struth tablesS that SsatisfyS at least the axioms +1, +2, &&&, +C forcin! them al ays to take StrueS #alues&
4ote& 'ee my #ersion of the pro!ram in <\\: header file, implementation&

Thus, e consider Sunder the abo#e truth tables, formula 6 al ays takes StrueS #aluesS as a kind of the SpropertyS Q061& 2ill 5odus %onens preser#e this property8 7f 6 is StrueS, and 6`G is StrueS, ho could G be not8 +et us consider the rele#ant part of the abo#e truth tables 0i&e& the part here 6 is StrueS1: 6 G 6`G 2 0 2 1 2 2 i= iC i,

7f e ould consider only those #ariants of our Struth tablesS here i = ^ 0 or 1, iC ^ 0 or 1, and i, ^ 2, then, if G ould not be 2 for some #alues of its ar!uments, then 6`G also ould not be 2 for the same #alues of ar!uments& Hence, if e restrict oursel#es to Struth tablesS ith i= ^ 0 or 1, iC ^ 0 or 1, and i, ^ 2, then 5% preser#es the property of Sbein! trueS& I.e., from <true< formulas MP can derive only <true< formulas. The next idea: if e ish the axiom + -: 6`6#G al ays takin! the #alue StrueS 0i&e& the #alue 21, then, if 3L%, then 3F% must be .. Thus, of all the (, ^ 1,-C( possible implication definition #ariants only the follo in! (]2]2 ^ 12 #ariants are orth of explorin!: 6 G 6`G 0 0 0 1 2 2

== 0 2 2

1 0 i.^0,1,2 1 1 1 2 2 2

2 0 i=^0,1 2 1 iC^0,1 2 2 2

!7ercise ..G..& a1 Verify that under any of these 12 implication definitions the axioms +(, +., +-, += al ays take the #alue 2, i&e& you do not need testin! these axioms any more& b1 9or each of the axioms, +1, +2, +/ and +C, determine all the possible combinations of the #alues of i., i=, iC forcin! it to take al ays the #alue 2& 4ote. The SintersectionS of b1 consists of / combinations 0see the results file [001& !7ercise ..G.)& @xtend your pre#ious computer pro!ram by addin! - nested loops: for i.^0 to 2, for i=^0 to 1, for iC^0 to1, for iaa^0 to 2, for ib^0 to 2, for ic^0 to 2& +et the pro!ram print out only those #ariants of Struth tablesS that make StrueS all the axioms +1-+C& 05y pro!ram yields 1(/ such #ariants, see the results file [001& Thus, no e ha#e 1(/ #ariants of Struth tablesS that make StrueS all the axioms +1-+C& +et us search amon! them for the #ariants that allo pro#in! of axiom independency results e are interested in& 37iom L( 7n Theorem 2&C&1 e established that the axiom L(: 06`G1`006`fG1`f61 can be pro#ed in V+1-+C, +10, +11, 5%W& 'till, 9heorem ..G..& The axiom +, cannot be pro#ed in V+1-+C, +10, 5%W& Proof& +et your pro!ram print out only those #ariants of Struth tablesS that make StrueS all the axioms +1-+C, and make: +, 3 not StrueS, and +10 3 StrueS& 5y pro!ram yields -- such #ariants, see the results file [01& 7 like especially the 0most natural81 #ariant [((:

=C 7mplication #ariant [(: 2 2 2 2 2 2 0 1 2 +1-+C true& Variant [((& ?e!ation: 2 1 0 +, not true& +10 true& +11 not true& 6 G 6`G 0 0 0 1 0 2 1 0 1 1 1 2 2 0 2 1 2 2 2 2 2 2 2 2 0 1 2

6 f6 0 1 2 2 1 0

'ee the extended results file [1 for this #ariant& *nder this #ariant the axioms + 1-+C and +10 are StrueS& 6s e kno , under this #ariant, by 5%, from StrueS formulas only StrueS formulas can be deri#ed& The axiom +, is not StrueS under this #ariant: 6 G 06`G1`006`fG1`f61 0 0 0 1 0 2 1 0 2 2 2 1

=, 1 1 1 2 2 0 2 1 2 2 1 1 2 2 2

Hence, +, cannot be pro#ed in V+1-+C, +10, 5%W& D&@&D& 7n a similar ay, e can obtain other independence results& 37iom L'I 9heorem ..G.)& The Scra:yS axiom L'I: fG`0G`<1 cannot be pro#ed in the minimal lo!ic V+1-+,, 5%W, and e#en not in V+1-+,, +11, 5%W& Proof& +et your pro!ram print out only those #ariants of Struth tablesS that make StrueS all the axioms +1-+C, and make: +, 3 StrueS, +10 3 not StrueS, and +11 3 StrueS& 5y pro!ram yields - such #ariants, see the results file [02& 7 like especially the 0some hat natural81 #ariant [1: 7mplication #ariant [1: 2 2 2 0 2 2 0 1 2 +1-+C true& Variant [1& ?e!ation: 2 2 1 +, true& +10 not true& +11 true& 'ee the extended results file [2 for this #ariant& *nder this #ariant the axioms +1-+, and +11 are StrueS& 6s e kno , under this #ariant, by 5%, from StrueS formulas only StrueS formulas can be deri#ed& The axiom +10 is not StrueS under this #ariant: 6 G f6`06`G1 0 0 0 1 0 2 1 0 1 1 1 2 2 2 2 2 2 2

C0 2 0 2 1 2 2 0 1 2

Hence, +10 cannot be pro#ed in V+1-+,, +11, 5%W& D&@&D& 37iom L'' ?o , let us pro#e the main result of this section: 9heorem ..G.*. The +a of @xcluded 5iddle L'': B7 B cannot be pro#ed in the constructi#e propositional lo!ic V+1-+10, 5%W& 7&e& the La8 of !7cluded Middle is an inde"endent logical "rinci"le& Proof& +et your pro!ram print out only those #ariants of Struth tablesS that make StrueS all the axioms +1-+C, and make: +, 3 StrueS, +10 3 StrueS, +11 3 not StrueS& 5y pro!ram yields only one such #ariant, see the results file [0(: 7mplication #ariant [1: 2 2 2 0 2 2 0 1 2 +1-+C true& Variant [1& ?e!ation: 2 0 0 +, true& +10 true& +11 not true& 'ee the extended results file [( for this #ariant& *nder this #ariant the axioms +1-+10 are StrueS& 6s e kno , under this #ariant, by 5%, from StrueS formulas only StrueS formulas can be deri#ed& The axiom +11 is not StrueS under this #ariant: G fG 0 1 2 2 0 0
B 7 B

2 1 2

Hence, +11 cannot be pro#ed in V+1-+10, 5%W& D&@&D& The results file [0( pro#es also the follo in! 9heorem ..G.- Cthanks to %a#els 5ihailo#s for a correction D. The follo in! 0classically pro#able1 formulas cannot be pro#ed in the constructi#e propositional lo!ic V+1-+10, 5%W: ff6 ` 6 0fG ` f61 ` 06`G1

C1 0f6`G1`0fG`61 0ff6 ` ffG1 ` 06`G1 ( A 5 B ) 5 7 A B (( A5 B ) 5 B ) 5 A B 006`G1`61`6 7 ( A 7 B ) 5 ( A 5 B ) 7 ( A 5 B ) 5 A 7 B A( A 5 B ) 06`G1`00f6`fG1`0G`611 7ndeed, all these formulas take non-StrueS #alues under the Struth tablesS from the proof of Theorem 2&C&.& The follo in! three formulas also cannot be pro#ed in the constructi#e propositional lo!ic, yet, unfortunately, the Struth tablesS from our proof of Theorem 2&C&. do not allo pro#in! this: 7 ( A B ) 5 7 A 7 B 77 ( A B ) 5 77 A7 7 B ( A 5 B )( B 5 A) 7ndeed, under the abo#e Struth tablesS, these formulas al ays take StrueS #alues 0see results file [0(1& 6nother interestin! conclusion: add these three formulas as additional a7ioms to &L'BL'I, MP+ E and L'' 8ill remain still un"rovable6 Thus, e did not succeed in buildin! a three-#alued lo!ic that ould allo sho in! that the latter three formulas cannot be pro#ed in the constructi#e propositional lo!ic& 7s it possible at all to build a multi-#alued lo!ic that ould separate constructi#ely pro#able propositional formulas from unpro#able ones8 $urt AFdel sho ed in 1,(2 that this is impossible: none of the finitelyB valued logics <matches< e7actly the constructive "ro"ositional logic=
K. 5Mdel& Lum intuitionistischen 6ussa!enkalkgl, An3eiger A&a+emie +er Bissenschaften Bien, 5ath&-natur iss& $lasse, 1,(2, Vol& -,, pp&-/---&

!7ercise ..G.*& a1 0for smart students1 Verify that the latter three formulas cannot be pro#ed in the constructi#e propositional lo!ic V+ 1-+10, 5%W& Rr, see 'ection .&.& b1 Verify that any of the follo in! formulas could be used 3 instead of B7 B 3 as the axiom +11 of the classical propositional lo!ic: i1 ( A 5 B ) 5 7 A B , ii1 ffG`G, iii1 f06`G1`6 0Hint: since all these formulas are pro#able in V+1-+11, 5%W, it remains to pro#e + 11 in V+1-+10, 5%W \ 0i1, in V+1-+10, 5%W \ 0ii1, and in V+1-+10, 5%W \ 0iii11& c1 Verify that ith ffG`G instead of +11 the Scra:yS axiom +10 becomes

C2 100T deri#able from the other axioms& %erhaps, this is hy many textbooks prefer the combination +1-+, \ ffG`G as the axiom list for the classical propositional lo!ic& Gut, then, e are forced to define the constructi#e propositional lo!ic not as a subset of the classical one, but as the classical lo!ic ith the axiom ffG`G replaced by the Scra:yS axiom + 10: fG`0G`<1; 37iom L'I again... 9inally, let us check hich of the main results of 'ections 2&/ 0constructi#e lo!ic1 and 2&- 0classical lo!ic1 depend on the Scra:yS axiom + 10: f6`06`G1& +et your pro!ram print out only those #ariants of Struth tablesS that make StrueS all the axioms +1-+C, and make: +, 3 StrueS, +10 3 not StrueS& 5y pro!ram yields - such #ariants, see the results file [0.& 'urprisin!ly, in all these #ariants +11 also is StrueS 0thus, the results file [0. eQuals the results file [021& 6s the most producti#e appears 7mplication #ariant [1: 2 2 2 0 2 2 0 1 2 +1-+C true& Variant [1& ?e!ation: 2 2 1 +, true& +10 not true& +11 true& <onstructi#ely pro#able formulas ?ot true: ( A B) 5 (( 7 A) 5 B ) ?ot true: (( 7 A ) B ) 5 ( A 5 B ) ?ot true: 00ff61`0ffG11`0ff06`G11 ?ot true: 0ff61`00f61`61 ?ot true: ( A ( 7 A)) 5 (( 7 7 A) 5 A) ?ot true: ff00ff61`61 <lassically pro#able formulas True: ( 77 ( A B )) 5 (( 77 A)( 7 7 B )) True: ( 7 ( A B)) 5 (( 7 A )( 7 B )) ?ot true: 0ff61`6 ?ot true: 00fG1`0f611`06`G1 ?ot true: 00f61`G1`00fG1`61 ?ot true: 00ff61`0ffG11`06`G1 True: ( A 5 B ) 5 (( 7 A) B ) ?ot true: (( A5 B ) 5 B )5 ( A B ) ?ot true: 006`G1`61`6 ?ot true: ( 7 ( A( 7 B ))) 5 ( A5 B ) True: 06`G1`000f61`G1`G1 ?ot true: ( 7 ( A 5 B )) 5 ( A( 7 B )) ?ot true: A( A 5 B ) True: ( A 5 B )( B 5 A)

C( ?ot true: 06`G1`000f61`0fG11`0G`611 Thus, the follo in! constructi#ely pro#able formulas cannot be pro#ed in the minimal lo!ic V+1-+,, 5%W 0and e#en in V+1-+,, +11, 5%W1, i&e& they cannot be pro#ed ithout the Scra:yS axiom +10: ( A B) 5 ( 7 A 5 B ) 7 A B 5 ( A5 B ) 0ff6`ffG1 ` ff06`G1 ff6 ` 0f6`61 A7 A 5 ( 77 A 5 A ) ff0ff6`61 6nd the follo in! classically pro#able formulas cannot be pro#ed Scra:yS axiom +10 Cthanks to %a#els 5ihailo#s for a correctionD: ff6`6 0fG`f61`06`G1 0f6`G1`0fG`61 0ff6`ffG1`06`G1 (( A5 B ) 5 B ) 5 A B 006`G1`61`6 7 ( A 7 B ) 5 ( A 5 B ) 7 ( A 5 B ) 5 A 7 B A( A 5 B ) 06`G1`00f6`fG1`0G`611 !7ercise ..G.- 0thanks to 'tanisla# Aolubco# for the idea1& Gut ho remainin! four 0classically pro#able1 formulas: a1 ( A 5 B ) 5 7 A B , b1 7 ( A B ) 5 7 A7 B , c1 77 ( A B ) 5 77 A7 7 B , c11 06`G1`00f6`G1`G1, d1 ( A 5 B )( B 5 A) 8 'ho that the formulas 0a, b, c1 can be pro#ed ithout the Scra:yS axiom +10, i&e pro#e them in V+1-+,, +11, 5%W& 0Hint: use Theorem 2&-&. 0a1 V+ 1-+C, 5%W: A7 A U- ( A 5 B ) 5 7 A B( 1& 9or smart students: ho about the remainin! formulas 0c1, d18
Using com"uters in mathematical "roofs Do you trust the above "roofs2 %ersonally, 7 trust much more my ability to rite 0relati#ely1 error-free computer pro!rams than my ability to carry out error-free mathematical proofs& Gut ho about you8 Rf course, you do not need trustin! my 0or your o n1 pro!ram !eneratin! the results files [00, [01, [02, [0( and [0.& 2e used these files only to select the Struth tableS

ithout the

about the

C.
#ariants allo in! to pro#e our independence results& The critical points to be trusted are 0see my implementation file1 : a1 the recursi#e pro!ram int 5y9ormula::Value6t0int 6, int G, int <1 and b1 the character strin! analy:er int 5y9ormula::6naly:e0int ]pRperation, 6nsi'trin! ]p'ub9ormula1, 6nsi'trin! ]p'ub9ormula21 Kou may ish to remo#e your orries by #erifyin! directly that under all the ( truth table #ariants used abo#e: a1 the axioms + 1-+C are true, and b1 the axioms + ,, +10, +11 and other formulas are true or not true accordin! to the !oal of each particular proof& Gefore you ha#e performed this 100T, you can feel the flavor of using com"uters in mathematical "roofs 07 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it, 7 trust this proof, 7 do not trust it,&&&1 *nfortunately, in more complicated cases the situation does not allo 0i&e& manual #erification of the solution found by a computer1: the abo#e simple exit

SThe 9our <olour Theorem as the first ma>or theorem to be pro#ed usin! a computer, ha#in! a proof that could not be #erified directly by other mathematicians& Despite some orries about this initially, independent #erification soon con#inced e#eryone that the 9our <olour Theorem had finally been pro#ed& Details of the proof appeared in t o articles in 1,==& )ecent ork has led to impro#ements in the al!orithm&S 06ccordin! to the article: The 9our <olour Theorem in 5acTutor History of 5athematics archi#e1& The proof of the 9our <olour Theorem as completed in 1,=- by $enneth 6ppel and 2olf!an! Haken, see their photo!raphs published in @uropean 5athematical 'ociety, ?e sletter ?o& .-, December 2002, pp& 1/-1,& SThe best-kno n, and most debated, instance is the use of computer analysis by $enneth 6ppel and 2olf!an! Haken of the *ni#ersity of 7llinois in their 1,=- proof of the four-colour con>ecture 0that four colours suffice to colour in any map dra n upon a plane in such a ay that countries hich share a border are !i#en different colours1& 9irst put for ard in 1C/2, the con>ecture had become perhaps the most famous unsol#ed problem in mathematics, resistin! a multitude of efforts at proof for o#er a century& 6ppel and HakenBs demonstration rested upon computeri:ed analysis, occupyin! 1,200 hours of computer time, of o#er 1,.00 !raphs& The analysis of e#en one of those !raphs typically ent beyond hat an unaided human bein! could plausibly do: the ensemble of their demonstration certainly could not be checked in detail by human bein!s& 7n conseQuence, hether that demonstration constituted SproofS as deeply contro#ersial&&&S 0accordin! to Donald 5ac$en:ie& <omputers and the 'ociolo!y of 5athematical %roof& 7n: Tren+s in the History an+ Phi%osophy of .athematics , Rdense: *ni#ersity of 'outhern Denmark, 200., pp&-=-C-1& 'ee also $en 6ppel on the .<T proof, December 1,,C )obin 2ilson& 9our <olours 'uffice& @uropean 5athematical 'ociety, ?e sletter ?o& .-, December 2002, pp& 1/-1, 0online copy1&

C/
The 9our <olor Theorem, ?o#ember 1(, 1,,/, by )obin Thomas& 6 computer-checked proof of the 9our <olour Theorem, 200., by Aeor!es Aonthier& Doron Leilber!er& Rpinion /.: 7t is 7mportant to $eep +ookin! for ?on-<omputer %roofs of the 9our-<olor Theorem, Gut ?ot 9or the S*sualS )easons 0a#ailable online at http:PP &math&rut!ers&eduPf:eilber!PRpinion/.&html1& T o other famous computer assisted mathematical proofs: - 7n 1,C,, by usin! <ray super-computers, <lement 2& H& +am finished his proof that finite pro>ecti#e plane of order 10 is impossible 0for details see %ro>ecti#e plane in 2ikipedia1& - 7n 1,,C, Thomas <& Hales finished his proof of $epler con>ecture about the densest arran!ement of spheres in space 0Hohannes $epler con>ectured it in 1-11, for details see $epler con>ecture in 2ikipedia1& 'ee lo!ical soft are links selected by %eter 'uber& Visit the 5i:ar %ro>ect&

C-

). Predicate Logic
).'. Proving :ormulas ontaining Nuantifiers and Im"lication only
9heorem ).'.I. V+1, +2, +12, +1(, 5%W U- xG0x1`xG0x1& 2hat does it mean8 7t prohibits Sempty domainsS& 7ndeed, 011 xG0x1 021 xG0x1`G0x1 0(1 G0x1 0.1 G0x1`xG0x1 0/1 xG0x1 Hypothesis& 6xiom +12& Gy 5%& 6xiom +1(& Gy 5%&

Thus, by V+1, +2, 5%W Deduction Theorem 1, there is a proof of V+ 1, +2, +12, +1(, 5%W U- xG0x1`xG0x1& 9heorem ).'.'. a1 V+1, +2, +12, +1., 5%, AenW U- x0G`<1`0xG`x<1& 2hat does it mean8 b1 V+1, +2, +12-+1/, 5%, AenW U- x0G`<1`0xG`x<1& 2hat does it mean8 +et us pro#e 0a1& 011 x0G`<1 021 xG 0(1 x0G`<1`0G`<1 0.1 G`< 0/1 xG`G 0-1 G Hypothesis& Hypothesis& 6xiom +12: x90x1`90x1& 9rom 011 and 0(1, by 5%& 6xiom +12: x90x1`90x1& 9rom 021 and 0/1, by 5%&

C= 0=1 < 0C1 x< 9rom 0.1 and 0-1, by 5%& 9rom 0=1, by Aen&

7n this proof, Aen is applied only to x, hich is not a free #ariable in x0G`<1 and xG& Thus, by V+1, +2, +1., 5%, AenW Deduction Theorem 2, there is a proof of V+1, +2, +12, +1., 5%, AenW U- x0G`<1 ` 0xG`x<1& +et us pro#e 0b1& 011 x0G`<1 021 x0G`<1`0G`<1 0(1 G`< 0.1 <`x< 0/1 G`x< 0-1 x0G`x<1 0=1 x0G`x<1`0xG`x<1 0C1 xG`x< Hypothesis& 6xiom +12: x90x1`90x1& 9rom 011 and 021, by 5%& 6xiom +1(: 90x1`x90x1& 9rom 0(1 and 0.1, by transiti#ity of implication V+1, +2, 5%W& 9rom 0/1, by Aen& 6xiom +1/: x090x1`A1`0x90x1`A1 0x< does not contain x as a free #ariable1& 9rom 0-1 and 0=1, by 5%&

7n this proof, Aen is applied only to x, hich is not a free #ariable in x0G`<1& Thus, by V+1, +2, +1., 5%, AenW Deduction Theorem 2, there is a proof of V+1, +2, +12-+1/, 5%, AenW U- x0G`<1 ` 0xG`x<1& D&&D& 9heorem ).'... a1 V+1, +2, +/, +12, +1., 5%, AenW U- xyG0x, y1ayxG0x, y1& 2hat does it mean8 b1 V+1, +2, +/, +1(, +1/, 5%, AenW U- xyG0x, y1ayxG0x, y1& 2hat does it mean8 c1 V+1, +2, +12-+1/, 5%, AenW U- xyG0x, y1`yxG0x, y1& 2hat does it mean8 The con#erse implication xyG0x, y1`yxG0x, y1 cannot be true& @xplain, hy& +et us pro#e 0b1&

CC 011 021 0(1 0.1 0/1 0-1 G0x, y1`xG0x, y1 xG0x, y1`yxG0x, y1 G0x, y1`yxG0x, y1 90x1`A U- x90x1`A yG0x, y1`yxG0x, y1 xyG0x, y1`yxG0x, y1 6xiom +1( ith 90x1 ^ G0x, y1& 6xiom +1( ith 90y1 ^ xG0x, y1& 9rom 011 and 021, by transiti#ity of implication V+1, +2, 5%W& @xercise 1&.&(0a1: V+1/, 5%, AenW, x not free in A& 9rom 0(1, by 0.1, ith 90y1 ^ G0x, y1, A ^ xyG0x, y1& 9rom 0/1, by 0.1, ith 90x1 ^ yG0x, y1, A ^ xyG0x, y1&

The proof of the con#erse implication V+ 1, +2, +1(, +1/, 5%, AenW U- yxG0x, y1`xyG0x, y1 is identical& ?o , by 6xiom +/ e obtain the eQui#alence 0b1& D&@&D& !7ercise ).'.'. %ro#e 0a1 and 0c1 of Theorem (&1&2& !7ercise ).'... %ro#e in the constructi#e lo!ic, V+1-+10, +12-+1/, 5%, AenW U- x0G0x1`<0x11 ` 0xG0x1`x<0x11&

)... :ormulas ontaining 4egations and a ,ingle Nuantifier


3ttention= nonBconstructive reasoning6 fxG ` xfG& This formula is accepted in the classical lo!ic: if no x can possess the property G, then there is an x that does not possess G& 7t represents non-constructi#e reasonin! in its ultimate form: let us assume, all x-s possess the property G, if e succeed in deri#in! a contradiction from this assumption, then 3 hat8 7s this a proof that there is a particular x that does not possess the property G8 Does our proof contain a method allo in! to build at least one such x8 7f not, do e ha#e a SrealS proof of xfG8 Ho many formulas can be built of the formula G by usin! ne!ations and a sin!le Quantifier8 ffffffffffxffffffffffG

C, ffffffffffxffffffffffG +et us recall Theorem 2&.&/ V+1-+,, 5%W: U- fff6af6& 7&e&, any number of ne!ations can be reduced to :ero, one, or t o, and thus e obtain (]2]( ^ 1C formulas to be in#esti!ated& The follo in! table represents the results of this in#esti!ation from 3.$eyting. Rn eakened Quantification& /o$rna% of Symbo%ic Logic, 1,(-, #ol&11, pp&11,-121 0see also $leene V1,/2W, 'ection (&/1& 9able ).. I III

xG xfG ---------------------------------------------------------------------------------------ffxG ffxfG ^^^^^^^^^^^^^^^^^^^^^^^^^^ fxffG xffG ^^^^^^^^^^^^^^^^^^^^^^^^^^ ffxffG fxG fxfG II xG -------------------------------------------xffG -------------------------------------------ffxG ffxffG fxfG IV xfG ffxfG fxffG fxG

Legend& a1 7n the classical lo!ic, ithin each of the . !roups all formulas are eQui#alent, for example, in !roup 777: fxGaxfG& Rf course, formulas of different !roups cannot be eQui#alent 0explain, hy1& b1 T o formulas ithin a !roup are constructi#ely eQui#alent, if and only if they ha#e no separatin! lines bet een them& 9or example, in !roup 77: constructi#ely, fxfGaffxG, but not fxfGaxG 0explain, hy1& 6ll the formulas of the !roup 7V are constructi#ely eQui#alent& c1 7f t o formulas 91, 92 ithin a !roup 091 3 abo#e, 92 3 belo 1 are separated by a single line, then: constructi#ely, 91`92, and ff092`911, but not 92`91& 9or example, in !roup 77: constructi#ely, xG`fxfG, and ff0fxfG`xG1, but not fxfG`xG 0explain, hy1&

,0 d1 7f t o formulas 91, 92 ithin a !roup 091 3 abo#e, 92 3 belo 1 are separated by a double line, then: constructi#ely, 91`92, but not 92`91, and e#en not ff092`911& 9or example, in !roup 777: constructi#ely, xfG`fxG, but not fxG`xfG, and e#en not ff0fxG`xfG1 0try to explain, hy1& Thus, the implication fxG`xfG could be Qualified as su"erBnonB constructive& ?o , let us pro#e the implications necessary for the positi#e part of the abo#e le!end to be true& 5rou" I 7-1& <onstructi#ely, V+1, +2, +,, 5%W: U- xG`ffxG 7mmediately, by Theorem 2&.&. V+1, +2, +,, 5%W: U- 6`ff6& 7-2& <onstructi#ely, V+1-+,, +12, +1., 5%, AenW: U- ffxG`xffG 011 xG`G 021 ffxG`ffG 0(1 x0ffxG`ffG1 0.1 ffxG`xffG 6xiom +12: x90x1`90x1& 9rom 011, by Theorem 2&.&=0a1 V+1-+,, 5%W: U- 06`G1`0ff6`ffG1& 9rom 021, by Aen& 9rom 0(1, by 6xiom +1.: x0A`90x11`0A`x90x11&

7-(& <onstructi#ely, V+1, +2, +,, 5%W: U- xffG`ffxffG 7mmediately, by Theorem 2&.&. V+1, +2, +,, 5%W: U- 6`ff6& 7-.& <onstructi#ely, V+1, +2, +,, +12, +1/, 5%, AenW U- ffxffG`fxfG 011 021 xffG`ffG fffG`fxffG 6xiom +12: x90x1`90x1& 9rom 011, by the <ontraposition +a 3 Theorem 2&.&2& V+1, +2, +,, 5%W: U06`G1`0fG`f61& Gy Theorem 2&.&. V+1, +2, +,, 5%W: U- 6`ff6&

0(1

fG`fffG

,1 9rom 021 and 0(1, by transiti#ity of implication 3 Theorem 1&.&2 V+1, +2, 5%W& 9rom 0.1, by Aen& 9rom 0/1, by 6xiom +1/: x090x1`A1`0x90x1`A1& 9rom 0-1, by the <ontraposition +a V+1, +2, +,, 5%W&

0.1 0/1 0-1 0=1

fG`f xffG x0fG`fxffG1 xfG`fxffG ffxffG`fxfG

7-/& 7n the classical lo!ic, V+1-+11, +1(, +1., 5%, AenW: U- fxfG`xG 011 fG`xfG 021 fxfG`ffG 0(1 ffG`G 0.1 fxfG`G 0/1 x0fxfG`G1 0-1 fxfG`xG 6xiom +1(: 90x1`x90x1& 9rom 011, by the <ontraposition +a V+1, +2, +,, 5%W& <lassical lo!ic, Theorem 2&-&1 V+1-+11, 5%W: U- ff6 ` 6 9rom 021 and 0(1, by transiti#ity of implication V+1, +2, 5%W& 9rom 0.1, Gy Aen& 9rom 0/1, by 6xiom +1.: x0A`90x11`0A`x90x11&

Thus, e ha#e pro#ed that in Aroup 7, constructi#ely, 9 1`92`9(`9.`9/, and, in the classical lo!ic, 9/`91& 7&e& e ha#e pro#ed that in Aroup 7: a1 in the classical lo!ic, all the formulas are eQui#alent, and b1 constructi#ely, upper formulas imply lo er formulas& 7--& <onstructi#ely, V+1, +2, +,, +1(, +1., 5%, AenW: U- fxfG`xffG 011 fG`xfG 021 fxfG`ffG 6xiom +1(: 90x1`x90x1& 9rom 011, by the <ontraposition +a V+1, +2, +,, 5%W&

,2 0(1 x0fxfG`ffG1 0.1 fxfG`xffG 9rom 021, by Aen& 9rom 0(1, by 6xiom +1.: x0A`90x11`0A`x90x11&

Thus, e ha#e pro#ed that in Aroup 7, constructi#ely, V+ 1, +2, +,, +12 -+1/, 5%, AenW: 9(`9.`9/`9(, i&e& that formulas 9(, 9., 9/ are constructi#ely eQui#alent& 9or Aroup 7, it remains to pro#e 7-=& <onstructi#ely, V+1-+10, 5%W U- ff0ffxG`xG1 7mmediately, by Theorem 2&/&20d1 V+1-+10, 5%W U- ff0ff6`61& 5rou" II 77-1& <onstructi#ely, V+1, +2, +,, +12-+1/, 5%, AenW U- xG`xffG 011 G`ffG 021 x0G`ffG1 0(1 xG`xffG Gy Theorem 2&.&. V+1, +2, +,, 5%W: U6`ff6& 9rom 011, by Aen& 9rom 021, by Theorem (&1&10b1 V+1, +2, +12-+1/, 5%, AenW

77-2& <onstructi#ely, V+1-+,, +12-+1/, 5%, AenW U-xffG`ffxG 011 G`xG 021 ffG`ffxG 0(1 x0ffG`ffxG1 0.1 xffG`ffxG 6xiom +1(: 90x1`x90x1& 9rom 011, by Theorem 2&.&=0a1 V+1-+,, 5%W: U- 06`G1`0ff6`ffG1& 9rom 021, by Aen& 9rom 0(1, by Theorem (&1&10b1 V+1, +2, +12-+1/, 5%, AenW

77-(& <onstructi#ely, V+1-+,, +12-+1/, 5%, AenW U- ffxG`ffxffG 7mmediately from 77-1, by 9rom 011, by Theorem 2&.&=0a1 V+ 1-+,, 5%W: U06`G1`0ff6`ffG1& 77-.& <onstructi#ely, V+1-+,, +12, +1/, 5%, AenW U- ffxffG`fxfG

,( 011 xfG`fG 021 ffG`fxfG 0(1 x0ffG`fxfG1 0.1 xffG`fxfG 0/1 ffxffG`fffxfG 0-1 fffxfG`fxfG 0=1 ffxffG`fxfG 6xiom +12: x90x1`90x1& 9rom 011, by the <ontraposition +a 3 Theorem 2&.&2& V+1, +2, +,, 5%W: U06`G1`0fG`f61& 9rom 021, by Aen& 9rom 0(1, by 6xiom +1/: x090x1`A1`0x90x1`A1& 9rom 0.1, by Theorem 2&.&=0a1 V+1-+,, 5%W: U- 06`G1`0ff6`ffG1& Theorem 2&.&/ V+1-+,, 5%W: Ufff6af6 9rom 0/1 and 0-1, by transiti#ity of implication V+1, +2, 5%W&

77-/& 7n the classical lo!ic, V+1-+11, +1(, +1., 5%, AenW: U- fxfG`xG 011 fxfG`ffxG 021 ffxG`xG 0(1 fxfG`xG 77-- V+1, +2, +,, +1(, +1., 5%, AenW, see belo & <lassical lo!ic, Theorem 2&-&1 V+1-+11, 5%W: U- ff6 ` 6 9rom 011 and 021, by transiti#ity of implication V+1, +2, 5%W&

Thus, e ha#e pro#ed that in Aroup 77, constructi#ely, 9 1`92`9(`9.`9/, and, in the classical lo!ic, 9/`91& 7&e& e ha#e pro#ed that in Aroup 77: a1 in the classical lo!ic, all the formulas are eQui#alent, and b1 constructi#ely, upper formulas imply lo er formulas& 77--& <onstructi#ely, V+1, +2, +,, +1(, +1., 5%, AenW U- fxfG`ffxG 011 G`xG 6xiom +1(: 90x1`x90x1&

,. 9rom 011, by the <ontraposition +a 3 Theorem 2&.&2& V+1, +2, +,, 5%W: U06`G1`0fG`f61& 9rom 021, by Aen& 9rom 0(1, by 6xiom +1.: x0A`90x11`0A`x90x11& 9rom 0.1, by the <ontraposition +a 3 Theorem 2&.&2& V+1, +2, +,, 5%W: U06`G1`0fG`f61&

021 fxG`fG 0(1 x0fxG`fG1 0.1 fxG`xfG

0/1 fxfG`ffxG

Thus, e ha#e pro#ed that in Aroup 77, constructi#ely, V+ 1-+,, +12-+1/, 5%, AenW: 9(`9.`9/`9(, i&e& that formulas 9(, 9., 9/ are constructi#ely eQui#alent& 77-=& <onstructi#ely, V+1-+10, 5%W: U- ff0ffxG`xG1 7mmediately, by Theorem 2&/&2 V+1-+10, 5%W: U- ff0ff6`61& Thus, constructi#ely, ff09(`911, and 91`92`9(`9.`9/`9(& Gy Theorem 2&.&=0d1, V+1-+,, 5%W ff06`G1, ff0G`<1 U- ff06`<1& Thus, in fact, e ha#e pro#ed that in Aroup 77, for all i, >, constructi#ely, ff09 i`9>1 0a kind of S eak eQui#alenceS1& 5rou" III 777-1& <onstructi#ely, V+1, +2, +,, 5%W: U- xfG`ffxfG 7mmediately, by Theorem 2&.&. V+1, +2, +,, 5%W: U- 6`ff6& 777-2& <onstructi#ely, V+1, +2, +,, +12, +1/, 5%, AenW: U- ffxfG`fxffG 011 xffG`ffxffG 021 ffxffG`fxfG 0(1 xffG`fxfG 7-( V+1, +2, +,, 5%W, see abo#e& 7-. V+1, +2, +,, +12, +1/, 5%, AenW, see abo#e& 9rom 011 and 021, by transiti#ity of implication V+1, +2, 5%W&

,/ 0.1 ffxfG`fxffG 9rom 0(1, by the <ontraposition +a V+1, +2, +,, 5%W&

777-(& <onstructi#ely, V+1-+,, +12, +1., 5%, AenW: U- fxffG`fxG 011 xG`ffxG 021 ffxG`xffG 0(1 xG`xffG 0.1 fxffG`fxG 7-1 V+1, +2, +,, 5%W, see abo#e& 7-2 V+1-+,, +12, +1., 5%, AenW 9rom 011 and 021, by transiti#ity of implication V+1, +2, 5%W& 9rom 0(1, by the <ontraposition +a V+1, +2, +,, 5%W&

777-.& 7n the classical lo!ic, V+1-+11, +1(, +1., 5%, AenW: U- fxG`xfG 011 fxfG`xG 021 fxG`ffxfG 0(1 ffxfG`xfG 0.1 fxG`xfG 7-/: in the classical lo!ic, V+1-+11, +1(, +1., 5%, AenW 9rom 011, by the <ontraposition +a V+1, +2, +,, 5%W& <lassical lo!ic, Theorem 2&-&1 V+1-+11, 5%W: U- ff6 ` 6 9rom 021 and 0(1, by transiti#ity of implication V+1, +2, 5%W&

Thus, e ha#e pro#ed that in Aroup 777, constructi#ely, 9 1`92`9(`9., and, in the classical lo!ic, 9.`91& 7&e& e ha#e pro#ed that in Aroup 777: a1 in the classical lo!ic, all the formulas are eQui#alent, and b1 constructi#ely, upper formulas imply lo er formulas& 777-.& <onstructi#ely, V+1, +2, +,, +1(, +1., 5%, AenW: U- fxffG`ffxfG 011 fxfG`xffG 021 fxffG`ffxfG Thus, 7-- V+1, +2, +,, +1(, +1., 5%, AenW 9rom 011, by the <ontraposition +a V+1, +2, +,, 5%W&

e ha#e pro#ed that in Aroup 777, constructi#ely, 92`9(`92, i&e& that

,formulas 92, 9( are constructi#ely eQui#alent& 777-/& <onstructi#ely, V+1-+10, 5%W: U- ff0ffxfG`xfG1 7mmediately, by Theorem 2&/&2 V+1-+10, 5%W: U- ff0ff6`61& 5rou" IV 7V-1& <onstructi#ely, V+1, +2, +,, 5%W: U- xfG`ffxfG 7mmediately, by Theorem 2&.&. V+1, +2, +,, 5%W: U- 6`ff6& 7V-2& <onstructi#ely, V+1-+,, +12-+1/, 5%, AenW: U- ffxfG`fxffG 011 xffG`fxfG 9rom 77-2, 77-(, 77-. V+1-+,, +12-+1/, 5%, AenW, by transiti#ity of implication V+1, +2, 5%W& 9rom 011, by the <ontraposition +a V+1, +2, +,, 5%W&

021 ffxfG`fxffG

7V-(& <onstructi#ely, V+1, +2, +,, +12-+1/, 5%, AenW: U- fxffG`fxG 011 xG`xffG 021 fxffG`fxG 77-1 V+1, +2, +,, +12-+1/, 5%, AenW 9rom 011, by the <ontraposition +a V+1, +2, +,, 5%W&

7V-.& <onstructi#ely, V+1, +2, +,, +1(, +1., 5%, AenW: U- fxG`xfG 011 G`xG 021 fxG`fG 0(1 x0fxG`fG1 0.1 fxG`xfG 6xiom +1(: 90x1`x90x1& 9rom 011, by the <ontraposition +a V+1, +2, +,, 5%W& 9rom 021, by Aen& 9rom 0(1, by 6xiom +1.: x0A`90x11`0A`x90x11&

Thus, e ha#e pro#ed that in Aroup 7V all the formulas are constructi#ely eQui#alent& 6nd thus, e ha#e pro#ed the positi#e part of the le!end of Table (&2& The ne!ati#e part of the le!end asserts that the follo in! 0classically pro#able1

,= formulas cannot be pro#ed constructi#ely: 011 ffxG`xG 021 xffG`ffxG 0(1 ff0xffG`ffxG1 0.1 xffG`xG 0/1 ffxG`xffG 0-1 ffxfG`xfG 0=1 fxG`fxffG 0C1 ff0fxG`fxffG1 'ee Aroup 7& 'imply, an instance of 0the nonconstructi#e1 ff6`6& 'ee Aroup 7& ,u"erBnonBconstructive: e#en ff021 is non-constructi#e; ff021& 'ee Aroup 7& 'ee Aroup 77& ?early, an instance of 0the nonconstructi#e1 ff6`6& 'ee Aroup 77& 'tron!er than simply nonconstructi#ity of ff6`68 'ee Aroup 777& 'imply, an instance of 0the non-constructi#e1 ff6`6& 'ee Aroup 777& ,u"erBnonBconstructive: e#en ff0=1 is non-constructi#e; ff0=1& 'ee Aroup 777&

2e ill pro#e these facts in 'ection .&/ 0see @xercise .&/&11& 'till, the most strikin! 0classically pro#able1 non-constructi#e Quantifier implications correspond to existence proofs #ia re+$ctio a+ abs$r+$m: ff0C1 is constructi#ely pro#able, but 0C1 is not, see Aroup 77& 7f e kno ho to deri#e a contradiction from xfG, then may be, e do not kno ho to find a particular x such that G& 0,1 is eaker than 0C1, but still nonconstructi#e, see Aroup 77& 7f e kno ho to deri#e a contradiction from xfG, then may be, e do not kno ho to deri#e a contradiction from fxffG& #en ff0101 is non-constructi#e, see Aroup 777& 7f e kno ho to deri#e a contradiction from xG, then may be, e do not kno ho to find a particular x such that fG&

0C1

fxfG`xG

0,1

fxfG`ffxffG

0101 fxG`xfG

,C 0111 is eaker than 0101, but still super-nonconstructi#e 0i&e& e#en ff0111 is nonconstructi#e1, see Aroup 777& 7f e kno ho to deri#e a contradiction from xG, then may be, e do not kno ho to deri#e a contradiction from fxfG&

0111 fxG`ffxfG

).). Proving :ormulas ontaining on@unction and Dis@unction


9heorem ).).'. a1 V+1-+/, +12, +1., 5%, AenW: U- x ( B C ) 6 xB xC & b1 V+1, +2, +--+C, +1., 5%, AenW: U- xB xC 5 x ( BC ) & The con#erse formula x ( B C ) 5 xB xC cannot be true& @xplain, hy& !7ercise ).).'. %ro#e V+(-+/, +12, 5%, AenW: x ( B C ) U- xB xC V+(-+/, +12, 5%, AenW: xB xC U- x ( B C ) & 'ince, in your first proof, Aen has been applied only to x, hich is not a free #ariable in x0GI<1, then, by Deduction theorem 2 V+1, +2, +1., 5%, AenW e obtain that V+1- +/, +12, +1., 5%, AenW: U- x ( B C ) 5 xB xC & 'imilarly, in your second proof, Aen has been applied only to x, hich is not a free #ariable in xB xC , then, by Deduction theorem 2 V+ 1, +2, +1., 5%, AenW e obtain that V+1- +/, +12, +1., 5%, AenW: U- xB xC 5 x ( BC ) & ?o , by Theorem 2&2&10a1 V+/W: 6, G U- A B , e obtain the eQui#alence 0a1 of Theorem (&(&1& +et us pro#e 0b1: U- xB xC 5 x ( BC ) & 011 021 B 5 B'C x ( B 5 B C ) 6xiom +-& 9rom 011, by Aen& and

,, 0(1 0.1 0/1 0-1 0=1 xB 5 x ( BC ) C 5 BC x ( C 5 B C ) xC 5 x ( BC ) 9rom 021, by Theorem (&1&10a1 V+1, +2, +12, +1., 5%, AenW& 6xiom +=& 9rom 0.1, by Aen& 9rom 0/1, by Theorem (&1&10a1 V+1, +2, +12, +1., 5%, AenW&

xB xC 5 x ( BC ) 9rom 0(1 and 0-1, by 6xiom +C&

The summary is V+1, +2, +--+C, +12, +1., 5%, AenW& D&@&D& 9heorem ).)... a1 V+1-+C, +12-+1/, 5%, AenW: U- x ( BC ) 6 xB xC & b1 V+1-+/, +1(-+1/, 5%, AenW: U- x ( BC ) 5 xB xC & The con#erse implication xB xC 5 x ( BC ) cannot be true& @xplain, hy& 9or the proof of x ( BC ) 5 xB xC , see @xercise (&(&20a1 belo & +et us pro#e xB xC 5 x ( BC ) & 011 021 0(1 0.1 0/1 0-1 0=1 B 5 BC x ( B 5 B C ) 6xiom +-& Gy Aen& Gy Theorem (&1&10b1: V+1, +2, +12-+1/, 5%, AenW U- x0G`<1`0xG`x<1 6xiom +=& Gy Aen& Gy Theorem (&1&10b1&

xB 5 x ( BC )
C 5 B C x ( C 5 B C ) xC 5 x ( BC )

xB xC 5 x ( B C ) 9rom 0(1 and 0-1, by 6xiom +C&

The summary is V+1, +2, +--+C, +12-+1/, 5%, AenW& ?o , by Theorem 2&2&10a1 V+/W: 6, G U- A B , 0a1& D&@&D& e obtain the eQui#alence ith + 1( and

!7ercise ).)... a1 %ro#e 0a`1 of Theorem (&(&2 0Hint: start finish by applyin! +1/&1

100 b1 %ro#e 0b1 of Theorem (&(&2& 0Hint: first, assume BC , deri#e xB xC , and apply Deduction Theorem 11& The con#erse implication xB xC 5 x ( BC ) cannot be true& @xplain, hy&

).*. #e"lacement 9heorems


6n example: Hence, e kno that lo! xy =lo! x + lo! y and lo! x y = ylo! x &

a b a b lo! 2 ( =lo! 2 +lo! ( = alo! 2 +blo! ( &

6nother example: e kno that 0in the classical lo!ic1: U- ( A 5 B ) 6 7 A B & Hence, the formula 0J`K1`L Sshould beS eQui#alent to 7 ( X 5 E ) F , and to 7 ( 7 X E ) F & 2e kno also that 7 ( A B ) 6 7 A7 B , hence, e can continue: 0J`K1`L Sshould beS eQui#alent to ( 77 X 7 E ) F , and to ( X 7E ) F 0since U-ff6a61& *ntil no , in our lo!ic, mathematical ar!ument& e could not use this #ery natural kind of

7n this section e ill pro#e meta-theorems that ill allo replacin! subformulas by eQui#alent formulas& 9or example, if ha#e pro#ed the formula xG`D, and e kno that U- Ga<, then e can replace G by <, obtainin! the formula x<`D& These theorems ill make the abo#e treatment of the formula 0J`K1`L completely le!al& 2e ill pro#e also that the meanin! of a formula does not depend on the names of bound #ariables used in it& 9or example, U- 0xG0x1`<1a0yG0y1`<1& 4ote& To pro#e all these replacement theorems logic V+1-+,, +12-+1/, 5%, AenW& ,ubBformulas and ?ccurrences 7ntuiti#ely, G is a sub-formula of <, if G is a formula, and G is a part 0substrin!1 of <& Gut note that a sub-formula may appear in the same formula more than once, as, for example, in the follo in! instance of the axiom + 1: 7%C7D`0x<0x1`7%C7D1& Thus, it ould be more correctly to speak about occurrences of sub-formulas& 7n the abo#e example, there are t o occurrences of the formula xG0x1& The formal definition is as follo s: e ill need only the minimal

101 a1 o0G1 is an occurrence in G in G& b1 7f o0G1 is an occurrence of G in <, then o0G1 is an occurrence of G in f<, C , , ,C , C , , ,C , <`D, and D`<& b1 7f o0G1 is an occurrence of G in <, then o0G1 is an occurrence of G in x<, and x<& 2e can define also the notion of "ro"ositional occurrences: a1 o0G1 is a propositional occurrence in G in G& b1 7f o0G1 is a propositional occurrence of G in <, then o0G1 is a propositional occurrence of G in f<, C , , ,C , C , , ,C , <`D, and D`<& 7ntuiti#ely, o0G1 is a propositional occurrence of G in <, if, in <, no Quantifiers stand o#er o0G1& #e"lacement Lemma '. 7n the minimal lo!ic, V+1-+,, 5%W: 0a1 6aG U- 06`<1a0G`<1 0b1 6aG U- 0<`61a0<`G1 0c1 6aG U- AC 6 B C 0d1 6aG U- C A6 C B 0e1 6aG UAC 6 B C V+1-+/, 5%W V+1-+/, 5%W V+1-+/, 5%W V+1-+/, 5%W V+1-+C, 5%W V+1-+C, 5%W V+1-+,, 5%W

0f1 6aG U- C A6 C B 0!1 6aG U- f6afG

ase CaD. 2e ill first pro#e that V+1, +2, +., 5%W: 6aG U- 06`<1`0G`<1& 011 ( A 5 B )( B 5 A) 6aG 3 hypothesis& Hypothesis& 9rom 011, by 6xiom +.& 9rom 0(1 and 021, by transiti#ity of implication V+1, +2, 5%W&

021 6`< 0(1 G`6 0.1 G`<

Thus, by V+1, +2, 5%W Deduction Theorem 1, V+1, +2, +., 5%W: 6aG U06`<1`0G`<1&

102 7n a similar ay, e can pro#e that V+1, +2, +(, 5%W: 6aG U- 0G`<1`06`<1& ?o , by Theorem 2&2&10a1, e obtain 0a1& D&@&D& !7ercise ).*.'. %ro#e 0b, c, d 1 of )eplacement +emma 1& ase CeD= 6aG U- AC 6 B C & 2e ill first pro#e that V+1, +2, +(-+C, 5%W: 6aG U- AC 5 B C & 011 021 ( A 5 B )( B 5 A) A C 6aG 3 hypothesis& Hypothesis& 9rom 011, by 6xiom +(& 6xiom +-& 9rom 0(1 and 0.1, by transiti#ity of implication V+1, +2, 5%W& 6xiom +=&

0(1 6`G 0.1 0/1 0-1 0=1 B 5 BC


A 5 BC

C 5 B C

( A 5 BC ) 5 (( C 5 B C ) 5 ( AC 5 BC )) 6xiom +C& 9rom 0=1, 0/1, 0-1 and 021, by 5%&

0C1 G#<

Thus, by V+1, +2, 5%W Deduction Theorem 1, V+1-+C, 5%W: 6aG U- AC 5 B C & 7n a similar ay, e can pro#e that V+1-+C, 5%W: 6aG U- BC 5 AC & ?o , by Theorem 2&2&10a1, e obtain 0e1& !7ercise ).*... %ro#e 0f 1 of )eplacement +emma 1& ase CgD. 2e ill first pro#e that V+1-+,, 5%W: 6aG U- f6`fG& 011 021 ( A 5 B )( B 5 A) f6 Hypothesis& Hypothesis&

10( 0(1 0.1 0/1 0-1 0=1 G`6 0G`61`00G`f61`fG1 0G`f61`fG G`f6 fG 9rom 011, by 6xiom +.& 6xiom +,& 9rom 0(1 and 0.1, by 5%& 9rom 021, by 6xiom +1& 9rom 0-1 and 0/1, by 5%&

Thus, by V+1, +2, 5%W Deduction Theorem 1, V+1-+,, 5%W: 6aG U- f6`fG& 7n a similar ay, e can pro#e that V+1-+,, 5%W: 6aG U- fG`f6& ?o , by Theorem 2&2&10a1, e obtain 0!1& This completes our proof of the )eplacement +emma 1& #e"lacement 9heorem '. +et us consider three formulas: G, GB, <, here G is a sub-formula of <, and o0G1 is a "ro"ositional occurrence of G in < 0i&e& no Quantifiers stand o#er o0G11& +et us denote by <B the formula obtained from < by replacin! o0G1 by GB & Then, in the minimal lo!ic, V+1-+,, 5%W: GaGB U- <a<B& Proof& 7nduction by the SdepthS of the propositional occurrence o0G1& Induction base= depth ^ 0& Then < is G, and <B is GB& The conclusion is ob#ious& Induction ste". 7f < is not G, then one of the follo in! holds: a1 < is 9`A, and o0G1 is in 9& b1 < is 9`A, and o0G1 is in A& c1 < is F G , and o0G1 is in 9& d1 < is F G , and o0G1 is in A& e1 < is F G , and o0G1 is in 9& f1 < is F G , and o0G1 is in A& !1 < is f9, and o0G1 is in 9& <ase 0a1& Gy induction assumption, V+1-+,, 5%W: GaGB U- 9a9B& Gy )eplacement +emma 10a1, V+1-+,, 5%W: 9a9B U- 09`A1a09B`A1& Thus, V+1-+,, 5%W: GaGB U- <a<B& !7ercise ).*.). )epeat the abo#e ar!ument for the remainin! cases 0b, c, d, e,

10. f, !1& D&@&D& ?o , e can use the replacement ar!ument mentioned at the be!innin! of this section 3 at least, for propositional occurrences of eQui#alent sub-formulas& #e"lacement Lemma .. 7n the minimal lo!ic, V+1-+,, +12-+1/, 5%, AenW: 0a1 Ga< U- xGax< 0b1 Ga< U- xGax< ase CaD& 2e ill first pro#e that V+1, +2, +(, +12, +1., 5%, AenW: Ga< U- xG`x<& 011 ( B 5 C )(C 5 B ) Ga< 3 hypothesis& Hypothesis& 9rom 011, by 6xiom +(& 9rom 021, by 6xiom +12: xG0x1`G0x1& Gy 5%& Gy Aen& V+1-+/, +12, +1., 5%, AenW V+1-+/, +12-+1/, 5%, AenW

021 xG 0(1 G`< 0.1 G 0/1 < 0-1 x<

Thus, e ha#e pro#ed that V+(, +12, 5%, AenW: Ga<, xG U- x<& 'ince Aen has been applied only to x, hich is not a free #ariable in xG, then, by Deduction Theorem 2 V+1, +2, +1., 5%, AenW e obtain that V+1, +2, +(, +12, +1., 5%, AenW: Ga< U- xG`x<& 7n a similar ay, e acan pro#e that V+1, +2, +., +12, +1., 5%, AenW: Ga< U- x<`xG& ?o , by Theorem 2&2&10a1, e obtain 0a1& ase CbD& 2e ill first pro#e that V+1, +2, +(, +12, +1., 5%, AenW: Ga< U- xG`x<& 011 ( B 5 C )( C 5 B ) Ga< 3 hypothesis&

10/ 021 xG 0(1 G`< 0.1 U- <`x< 0/1 G`x< 0-1 x0G`x<1 0=1 U- x0G`x<1`0xG`x<1 0C1 x< Hypothesis& 9rom 011, by 6xiom +(& 6xiom +1(& 9rom 0(1 and 0.1, by transiti#ity of implication V+1, +2, 5%W& Gy Aen& 6xiom +1/: x090x1`A1`0x90x1`A1& 9rom 0=1, 0-1 and 021, by 5%&

Thus, e ha#e pro#ed that V+ 1, +2, +(, +1(, +1/, 5%, AenW: Ga<, xG U- x<& 'ince Aen has been applied only to x, hich is not a free #ariable in xG, then, by Deduction Theorem 2 V+1, +2, +1., 5%, AenW e obtain that V+1, +2, +(, +12-+1/, 5%, AenW: Ga< U- xG`x<& 7n a similar ay, e can pro#e that V+1, +2, +., +12-+1/, 5%, AenW: Ga< U- x<`xG& ?o , by Theorem 2&2&10a1, e obtain 0b1& D&@&D& #e"lacement 9heorem .. +et us consider three formulas: G, GB, <, here G is a sub-formula of <, and o0G1 is any occurrence of G in <& +et us denote by <B the formula obtained from < by replacin! o0G1 by GB & Then, in the minimal lo!ic, V+1-+,, +12-+1/, 5%, AenW: GaGB U- <a<B& Proof& 7nduction by the SdepthS of the occurrence o0G1& Induction base= depth ^ 0& Then < is G, and <B is GB& The conclusion is ob#ious& Induction ste". 7f < is not G, then one of the follo in! holds: a1-!1 3 as in the proof of )eplacement Theorem 1& h1 < is x9, and o0G1 is in 9& i1 < is x9, and o0G1 is in 9&

10ase ChD. Gy induction assumption, V+1-+,, +12-+1/, 5%, AenW: GaGB U9a9B& Gy )eplacement +emma 20a1, V+ 1-+,, +12-+1/, 5%, AenW: 9a9B Ux9ax9B& Thus, V+1-+,, +12-+1/, 5%, AenW: GaGB U- <a<B& ase CiD. Gy induction assumption, V+1-+,, +12-+1/, 5%, AenW: GaGB U- 9a9B& Gy )eplacement +emma 20b1, V+1-+,, +12-+1/, 5%, AenW: 9a9B U- x9ax9B& Thus, V+1-+,, +12-+1/, 5%, AenW: GaGB U- <a<B& D&@&D& 4o8 Conly no86D , 8e may use in our "roofs the re"lacement argument mentioned at the beginning of this section. 3nd no8, for any e>uivalent subBformulas6 9inally, let us pro#e that the meanin! of a formula does not depend on the names of bound #ariables used in it& 7ntuiti#ely, it Smust be soS, but no e can pro#e this intuition as a meta-theorem& #e"lacement Lemma ). 7f the formula G does not contain the #ariable y, then 0in the minimal lo!ic1: a1 V+/, +12, +1., 5%, AenW: U- xG0x1ayG0y1 b1 V+/, +1(, +1/, 5%, AenW: U- xG0x1ayG0y1& 9irst, let us pro#e V+12, +1., 5%, AenW: U- xG0x1`yG0y1& 011 U- xG0x1`G0y1 021 U-y0xG0x1`G0y11 6xiom +12: x90x1`90t1& G0x1 does not contain y, hence, G0xPy1 is an admissible substitution& Gy Aen&

6xiom +1.: U0(1 y0xG0x1`G0y11`0xG0x1` x0A`90x11`0A`x90x11& yG0y11 xG0x1 does not contain y& 0.1 U- xG0x1`yG0y1 Gy 5%& ?o , let us pro#e V+12, +1., 5%, Aen0x1W: U- yG0y1`xG0x1& 011 U- yG0y1`G0x1 6xiom +12: x90x1`90t1& G0x1 does not contain y, hence, G0y1 contains only free occurrences of y, i&e& G0yPx1 is an admissible substitution&

10= 021 U- x0yG0y1`G0x11 Gy Aen&

6xiom +1.: U0(1 x0yG0y1`G0x11`0yG0y1` x0A`90x11`0A`x90x11& yG0y1 does not contain x as a free xG0x11 #ariable& 0.1 U- yG0y1`xG0x1 Gy 5%&

?o , by Theorem 2&2&10a1, e obtain 0a1& To pro#e 0b1, first, let us pro#e V+1(, +1/, 5%, Aen0y1W: U- yG0y1`xG0x1& 011 U- G0y1`xG0x1 021 U-y0G0y1`xG0x11 6xiom +1(: 90t1`x90t1& G0x1 does not contain y, hence, G0xPy1 is an admissible substitution& Gy Aen&

6xiom +1/: U0(1 y0G0y1`xG0x11`0yG0y1` x090x1`A1`0x90x1`A1& xG0x1 xG0x11 does not contain y& 0.1 U- yG0y1`xG0x1 Gy 5%& ?o , let us pro#e V+1(, +1/, 5%, Aen0x1W: U- xG0x1`yG0y1& 011 U- G0x1`yG0y1 021 U- x0G0x1`yG0y11 6xiom +1(: 90t1`x90x1& G0x1 does not contain y, hence, G0y1 contains only free occurrences of y, i&e& G0yPx1 is an admissible substitution& Gy Aen&

6xiom +1/: U0(1 x0G0x1`yG0y11`0xG0x1` x090x1`A1`0x90x1`A1& yG0y1 yG0y11 does not contain x as a free #ariable& 0.1 U- xG0x1`yG0y1 D&@&D& #e"lacement 9heorem ). +et y be a #ariable that does not occur in a formula 9, containin! an occurrence of a Quantifier x 0or x1& +et us replace by y all Gy 5%&

?o , by Theorem 2&2&10a1, e obtain 0b1&

10C occurrences of the #ariable x bound by this particular Quantifier occurrence& +et us denote the resultin! formula by 9B& Then, in the minimal lo!ic, V+1-+,, +12-+1/, 5%, AenW: U- 9a9B& Proof. Thus, the formula 9 contains a sub-formula xG0x1 0or xG0x11, and e ish to replace it by y0G0y1 0or yG0y11, here y does not occur in 9& Gy )eplacement +emma (, in the minimal lo!ic, U- xG0x1ayG0y1, and UxG0x1ayG0y1& Hence, by )eplacement +emma 2, in the minimal lo!ic, U9a9B& D&@&D& ?o let us repeat our example& 2e kno that 0in the classical lo!ic1: U- ( A 5 B ) 6 7 A B & Hence, the formula 0J`K1`L is eQui#alent to 7 ( X 5 E ) F , and to 7 ( 7 X E ) F & 2e kno also that 7 ( A B ) 6 7 A7 B , hence, e can continue: 0J`K1`L is eQui#alent to ( 77 X 7 E ) F , and to ( X 7E ) F 0since U-ff6a61& ?o , in our lo!ic, e can use this #ery natural kind of mathematical ar!ument&

).-. onstructive !mbedding


Ali#enkoBs Theorem 0see 'ection 2&=1 pro#ides a simple Sconstructi#e embeddin!S for the classical propositional lo!ic: any classically pro#able formula can be Spro#edS in the constructi#e lo!ic, if you put t o ne!ations before it& This theorem does not hold for the predicate lo!ic& 9or example 0see 'ection (&21, 77-/& 7n the classical lo!ic, V+1-+11, +1(, +1., 5%, AenW: U- fxfG`xG& The double ne!ation of this formula, i&e& the formula ff0fxfG`xG1 cannot be pro#ed in the constructi#e predicate lo!ic 0see 'ection .&/1& Thus, instead of the simple operation ff9, e must search for a more complicated embeddin! operation& Ho e#er, !7ercise ).-.I 0for smart students1& Verify that a formula 9 is pro#able in the classical predicate lo!ic, if and only if ff9 is pro#able in the constructi#e predicate lo!ic plus the follo in! axiom schema: xffG`ffxG 0the socalled ,o$b%e "egation Shift schema, see 7ntuitionistic +o!ic by Hoan 5oscho#akis in 'tanford @ncyclopedia of %hilosophy& The first embeddin! operation $olmo!oro# 01,0(-1,C=1 in as introduced by 6ndrey ?ikolae#ich

10, 3.4.Kolmogorov& Rn the principle tertium non datur& .atem( sborni&, 1,2/, #ol&(2, pp&-.----= 0in )ussian1& 6 Quote from 6 'hort Gio!raphy of 6&?& $olmo!oro# by %aul 5&G& Vitanyi follo s: S$& !ot interested in mathematical lo!ic, and in 1,2/ published a paper in 5athematicheskii 'bornik on the la of the excluded middle, hich has been a continuous source for later ork in mathematical lo!ic& This as the first 'o#iet publication on mathematical lo!ic containin! 0#ery substantial1 ne results, and the first systematic research in the orld on intuitionistic lo!ic& $& anticipated to a lar!e extent 6& Heytin! Bs formali:ation of intuitionistic reasonin!, and made a more definite correlation bet een classical and intuitionistic mathematics& $& defined an operation for kembeddin!B one lo!ical theory in another& *sin! this 3 historically the first such operation, no called the k$olmo!oro# operationB 3 to embed classical lo!ic in intuitionistic lo!ic, he pro#ed that application of the la of the excluded middle in itself cannot lead to a contradiction& 7n 1,(2 $& published a second paper on intuitionistic lo!ic, in hich for the first time a semantics as proposed 0for this lo!ic1, free from the philosophical aims of intuitionism& This paper made it possible to treat intuitionistic lo!ic as constructi#e lo!ic&S 'ee also $olmo!oro# <entennial& 2e ill in#esti!ate the follo in! #ersion of an embeddin! operation: to obtain R091, in a formula 9, put t o ne!ations before: a1 e#ery atomic formula, b1 e#ery dis>unction, c1 e#ery existential Quantifier& 5ore precisely, let us define the follo in! embeddin! operation R 0you may ish to compare it ith some other #ersions possessin! similar properties1: ?"eration ? Detlo#s V1,-.W
7f 9 is an atomic formula, then R091 is ff9& R09`A1 is R091`R0A1&
* ( F G ) is * ( F ) * (G ) & * ( F G ) is 77 (* ( F )* ( G))

?"eration K $olmo!oro# V1,2/W


$091 is ff9&

?"eration ?J ?"eration ?o AFdel V1,((W, Aent:en V1,(-W, see $leene V1,/2W see $leene V1,/2W
RB091 is 9& Ro091 is 9&

ff0$091`$0A11
77 ( C ( F ) C ( G ))

7 (* G ( F ) 7* G ( G)) * G ( F ) * G ( G )

Ro091`Ro0A1
* o ( F )* o ( G)

77 ( C ( F ) C ( G ))

7 (7* G ( F ) 7* G ( G ))

7 (7*o ( F ) 7*o ( G ))

110
R0f91 is fR091& R0x91 is xR091& fff$091, or f$091] fRB091 ffx$091 xRB091 fxfRB091 fRo091 xRo091 fxfRo091

R0x91 is ffxR091& ffx$091

0]1 Gy Theorem 2&.&/, V+1-+,, 5%W: U- fff$091af$091& 9or example, let us take the abo#e formula fxfG`xG& 7f G is an atomic formula, then R0fxfG`xG1 is fxfffG`ffxffG, i&e& fxfG`ffxffG The latter formula is constructi#ely pro#able 0see 'ection (&2, Aroup 771& Lemma ).-.'& 9or any formula 9, in the classical lo!ic, U- 9aR091& Proof. Gy induction& +et us recall Theorem 2&-&1: V+1-+11, 5%W U- ff6 a 6& '. Induction base: 9 is an atomic formula& Then R091 is ff9& Gy Theorem 2&-&1, V+1-+11, 5%W U- ff9a9, hence, in the classical lo!ic, U- R091a9& .. Induction ste". ase .a= 9 is G#<& Then R091 is ff0R0G1#R0<11& 011 R0G1aG 021 R0<1a< 0(1 0.1 0/1 0-1 BC 6 * ( B )C * ( B )C 6 * ( B)* ( C ) * ( B )* ( C ) 6 77 ( * ( B )* ( C )) BC 6 77 (* ( B )* ( C )) , i&e& 9aR091 7nduction assumption& 7nduction assumption& 9rom 011, by )eplacement Theorem 1& 9rom 021, by )eplacement Theorem 1& Theorem 2&-&1: V+1-+11, 5%W Uff6 a 6& Gy transiti#ity of implication&

ase .b= 9 is xG& Then R091 is ffxR0G1& 011 R0G1aG 021 xGaxR0G1 7nduction assumption& 9rom 011, by )eplacement Theorem 2&

111 0(1 xR0G1affxR0G1 Theorem 2&-&1: V+1-+11, 5%W U- ff6 a 6&

0.1 xGaffxR0G1, i&e& 9aR091 Gy transiti#ity of implication& ase .c= 9 is G`<& ase .d= 9 is GI<& ase .e= 9 is fG& ase .f= 9 is xG& !7ercise ).-.'&%ro#e 0c, d, e, f1& D&@&D& 'till, the key feature of the formulas ha#in! the form R091 is !i#en in Lemma ).-..& 9or any formula 9, there is a proof of V+1-+,, +12, +1., 5%,AenW: U- ffR091aR091& 7&e&, in the minimal lo!ic, e may drop the double ne!ation before R091 0before an arbitrary formula, e can do this only in the classical lo!ic1& 4ote. 7n some other textbooks, if ffAaA can be pro#ed in the constructi#e lo!ic, then A is called a stable formula& Thus, the embeddin! R091 is a stable formula for any 9& Proof. Gy Theorem 2&.&., V+1, +2, +,, 5%W: U- 6`ff6& Thus, it remains to pro#e U- ffR091`R091& +et us proceed by induction& '. Induction base: 9 is an atomic formula& Then R091 is ff9, and ffR091`R091 is ffff9`ff9& +et us recall Theorem 2&.&/: V+ 1-+,, 5%W Ufff6af6& Hence, by takin! 6 ^ f9: V+1-+,, 5%W U- ffff9`ff9, i&e& V+1-+,, 5%W U- ffR091`R091& .. Induction ste". ase .a: 9 is BC , or xG, or fG& Then R091 is 77 ( * ( B)* ( C )) , or ffxR0G1, or fR0G1& Hence, ffR091`R091 is fffA`fA, here A is 7 ( * ( B )* ( C )) , or fxR0G1, or R0G1& +et us recall Theorem 2&.&/: V+1+,, 5%W U- fff6af6& Hence, V+1-+,, 5%W U- fffA`fA, i&e& V+1-+,, 5%W U- ffR091`R091& ase .b= 9 is G`<& Then R091 is R0G1`R0<1& Gy induction assumption, V+1, +2, +12, +1., 5%, AenW: U- ffR0G1`R0G1, and U- ffR0<1`R0<1&

112 011 ffR0<1`R0<1 021 ff0R0G1`R0<11 0(1 ffR0G1`ffR0<1 0.1 R0G1`ffR0G1 0/1 R0G1`R0<1, i&e& R091 7nduction assumption& ffR091 3 hypothesis& Gy Theorem 2&.&=0b1: V+1-+,, 5%W Uff06`G1`0ff6`ffG1& Gy Theorem 2&.&., V+1, +2, +,, 5%W: U- 6`ff6& 9rom 0.1, 0(1 and 011, by transiti#ity of implication V+1, +2, 5%W&

Hence, since Aen is not applied here at all, by Deduction Theorem 1 V+1, +2, 5%W e obtain that V+1-+,, +12, +1., 5%, AenW U- ffR091`R091& ase .c= 9 is BC & Then R091 is * ( B )* ( C ) & Gy induction assumption, V+1, +2, +12, +1., 5%, AenW: U- ffR0G1`R0G1, and U- ffR0<1`R0<1& 011 021 77 ( * ( B)* ( C )) 77* ( B )77 * ( C ) ffR091 3 hypothesis& 9rom 011, by Theorem 2&.&C0a1, V+1-+,, 5%W U- 77 ( A B ) 6 ( 77 A77 B ) & 9rom 021, by 6xiom +(& 9rom 021, by 6xiom +.& 9rom 0(1, by induction assumption& 9rom 0.1, by induction assumption& 9rom 0/1 and 0-1, by 6xiom +/&

0(1 ffR0G1 0.1 ffR0<1 0/1 R0G1 0-1 R0<1 0=1 * ( B )* (C ) , i&e& R091

Hence, since Aen is not applied here at all, by Deduction Theorem 1 V+1, +2, 5%W e obtain that V+1-+,, +12, +1., 5%, AenW U- ffR091`R091& ase .d= 9 is xG& Then R091 is xR0G1& Gy induction assumption, V+1-+,, +12, +1., 5%, AenW: U- ffR0G1`R0G1& 2e must pro#e that UffxR0G1`xR0G1&

11( 011 ffxR0G1`xffR0G1 021 U- ffR0G1`R0G1 0(1 U- x0ffR0G1`R0G11 0.1 U- xffR0G1`xR0G1 0/1 U- ffxR0G1`xR0G1 D&@&D& Lemma ).-.)& 7f 9 is one of the 0classical1 axioms +1-+11, +12-+1/, then, in the constructi#e lo!ic, V+1-+10, +12-+1/, 5%, AenW: U- R091& 4ote. The axiom +10 ill be used in the proof of +emma (&/&( only once 3 to pro#e that R0+101 is pro#able in the constructi#e lo!ic& Gut, of course, R0+101 cannot be pro#ed in the minimal lo!ic, hence, in the +emma (&/&(, the constructi#e lo!ic cannot be replaced by the minimal one& Proof. ase '. 9 0as an axiom schema1 does not contain dis>unctions and existential Quantifiers, i&e& if 9 is +1, +2, +(, +., +/, +,, +10, +12, or +1.&, then R091 is an instance of the same axiom as 9, i&e& V9W: U- R091& 9or example, if 9 is + 1, i&e& G`0<`G1, then R091 is R0G1`0R0<1`R0G11, i&e& R091 is an instance of the same axiom +1& ase .a& 9 is +-: B 5 BC & Then R091 is * ( B ) 5 77 ( * ( B )* ( C )) , and VV+1, +2, +-, +,, 5%W U- R091& 7ndeed: 011 021 0(1 * ( B ) 5 * ( B )* ( C ) 6xiom +-& 'ection (&2, 7-2: V+1-+,, +12, +1., 5%, AenW U- ffxG`xffG 7nduction assumption Gy Aen& 9rom 0(1, by Theorem (&1&10a1, V+1, +2, +12, +1., 5%, AenW U- x0G`<1`0xG`x<1& 9rom 011 and 0.1, by transiti#ity of implication V+1, +2, 5%W&

Gy Theorem 2&.&., V+1, +2, +,, * ( B )* ( C ) 5 77 ( * ( B )* ( C )) 5%W: U- 6`ff6& * ( B ) 5 77 ( * ( B )* ( C )) Gy transiti#ity of implication V+1, +2, 5%W&

ase .b& 9 is +=: C 5 BC & Then R091 is * ( C ) 57 7( * ( B )* ( C )) ,

11. and VV+1, +2, +=, +,, 5%W U- R091& %roof is similar to <ase 2a& ase .c. 9 is +C: ( B 5 , ) 5 (( C 5 , ) 5 ( B C 5 ,)) & Then R091 is ( * ( B ) 5 * ( , )) 5 (( * ( C ) 5 * ( , )) 5 ( 77 ( * ( B )* (C )) 5* ( , ))) & 011 ffR0D1`R0D1 021 R0G1`R0D1 0(1 0R0<1`R0D1 0.1 77 ( * ( B)* ( C )) Gy +emma (&/&2, V+1-+,, +12, +1., 5%,AenW: U- ffR091`R091& Hypothesis& Hypothesis& Hypothesis&

( * ( B ) 5 * ( , )) 5 (( * ( C ) 5* ( , )) 5 ( * ( B )* ( C ) 5* ( , ))) & 0/1 6xiom + & C 0-1 0=1

* ( B ) '* ( C ) 5 * ( , )

Gy 5%&

9rom 0-1, by Theorem 2&.&=0a1, 77 ( * ( B)* ( C )) 5 77* ( , ) V+1-+,, 5%W U06`G1`0ff6`ffG1 Gy 5%& 9rom 011, by 5%&

0C1 ffR0D1 0,1 R0D1

Hence, since Aen is not applied after hypotheses appear in the proof, by Deduction Theorem 26 V+1, +2, +1., 5%, AenW e obtain that V+1-+,, +12, +1., 5%,AenW U- R091& ase .d. 9 is +11: G#fG& Then R091 is 77 ( * ( B)7* ( B)) & +et us recall Theorem 2&.&-0b1: V+1-+,, 5%W U- 77 ( A7 A) & Hence, V+1-+,, 5%W UR091& ase .e. 9 is +1(: 90t1`x90x1& Then R091 is R090t11`ffxR090x111, and VV+1, +2, +,, +1(, 5%W U- R091& 7ndeed: 011 R090t11`xR090x11 021 xR090x11`ffxR090x11 6xiom +1(& Gy Theorem 2&.&., V+1, +2, +,, 5%W: U6`ff6&

11/ 0(1 U- R090t11`ffxR090x11 Gy transiti#ity of implication V+1, +2, 5%W&

ase .f. 9 is +1/: x090x1`A1`0x90x1`A1& Then R091 is x0R090x11`R0A11`0ffxR090x11`R0A11& 011 ffR0A1`R0A1 021 x0R090x11`R0A11 0(1 ffxR090x11 0.1 Gy +emma (&/&2, V+1-+,, +12, +1., 5%,AenW: U- ffR091`R091& Hypothesis& Hypothesis&

x0R090x11`R0A11` 0xR090x11`R0A11& 6xiom +1/: x090x1`A1`0x90x1`A1& Gy 5%& 9rom 0.1, by Theorem 2&.&=0a1, V+1-+,, 5%W U- 06`G1`0ff6`ffG1 Gy 5%& 9rom 011, by 5%&

0/1 xR090x11`R0A1 0-1 ffxR090x11`ffR0A1 0=1 ffR0A1 0C1 R0A1

Hence, since Aen is not applied after hypotheses appear in the proof, by Deduction Theorem 26 V+1, +2, +1., 5%, AenW e obtain that V+1-+,, +12, +1., +1/, 5%,AenW U- R091& D&@&D& 9heorem ).-.*& 7n the classical lo!ic, V+1-+11, +12-+1/, 5%, AenW: G1, G2, &&&, Gn U- <, if and only if, in the constructi#e lo!ic, V+1-+10, +12-+1/, 5%, AenW: R0G11, R0G21, &&&, R0Gn1 U- R0<1& 7n particular, a formula 9 is pro#able in the classical lo!ic, if and only if the formula R091 is pro#able in the constructi#e lo!ic& Proof. 1& +et V+1-+11, +12-+1/, 5%, AenW: G1, G2, &&&, Gn U- <& 7nduction by the len!th of the shortest proof&

11Induction base. 7f < is an axiom, then, by +emma (&/&(, in the constructi#e lo!ic, U- R0<1& 7f < is Gi, then R0Gi1 U- R0<1 in any lo!ic& Induction ste". 7f < is deri#ed by 5% from G and G`<, then, by induction assumption, in the constructi#e lo!ic: R0G11, R0G21, &&&, R0Gn1 U- R0G1, and R0G11, R0G21, &&&, R0Gn1 U- R0G`<1& +et us mer!e these t o proofs& 'ince R0G`<1 is R0G1`R0<1, then, by 5%, in the constructi#e lo!ic: R0G 11, R0G21, &&&, R0Gn1 UR0<1& 7f < is xG0x1, and is deri#ed by Aen from G0x1, then, by induction assumption, in the constructi#e lo!ic: R0G 11, R0G21, &&&, R0Gn1 U- R0G0x11& Hence, by Aen, in the constructi#e lo!ic: R0G11, R0G21, &&&, R0Gn1 UxR0G0x11, i&e& R0G11, R0G21, &&&, R0Gn1U- R091& D&@&D& 2& +et in the constructi#e lo!ic: U- R0G 11, R0G21, &&&, R0Gn1 U- R0<1& Gy +emma (&/&1, in the classical lo!ic, U- Gi`R0Gi1 for all i, and U- R0<1`<& Hence, in the classical lo!ic, G1, G2, &&&, Gn U- <& D&@&D& orollary ).-.-. 7f, in the classical lo!ic, G1, G2, &&&, Gn U- C 7C , then, in the constructi#e lo!ic, R0G11, R0G21, &&&, R0Gn1 U- * ( C )7* ( C ) & 7&e&, if the postulates G1, G2, &&&, Gn are inconsistent in the classical lo!ic, then the postulates R0G11, R0G21, &&&, R0Gn1 are inconsistent in the constructi#e lo!ic& Rr: if the postulates R0G11, R0G21, &&&, R0Gn1 are consistent in the constructi#e lo!ic, then the postulates G1, G2, &&&, Gn are consistent in the classical lo!ic& orollary ).-./. 7f, for some predicate lan!ua!e, the classical lo!ic is inconsistent, then so is the constructi#e lo!ic& Rr: if, for some "redicate language, the constructive logic is consistent, then so is the classical logic 0AFdel V1,((W, Aent:en V1,(-W1& 1arning6 <orollary (&/&- does not extend immediately to first order theories, ha#in! their o n specific non-lo!ical axioms& 7t must be #erified separately for each theory; 9or example, !7ercise ).-.. 0for smart students1& Verify that, if the constructi#e first order arithmetic is consistent, then so is the classical first order arithmetic 0AFdel V1,((W, Aent:en V1,(-W1& 0Hint: #erify that, a1 atomic formulas of arithmetic are stable 3 this is the hard part of the proof, b1 if 9 is an axiom of arithmetic,

11= then so is R091&1 Thus, the nonBconstructivity does not add contradictions Cat leastD to arithmetic& 7f it ould, then e could deri#e Sconstructi#eS arithmetical contradictions as ell& K. 5Mdel& Lur intuitionistischen 6rithmetik und Lahlentheorie& Ergebnisse eines mathematischen Co%%o4$i$ms, 1,((, Vol& ., pp& (.-(C& 5erhard 5ent;en& Die 2iderspruchsfreiheit der reinen Lahlentheorie& .athematische Anna%en, 1,(-, Vol& 112, pp& .,(-/-/& 6bout constructi#e embeddin! operations as a !eneral notion see 4ikolai 3.,hanin& @mbeddin! the classical lo!ical-arithmetical calculus into the constructi#e lo!ical-arithmetical calculus& ,o&%a+i A" SSS), 1,/., #ol& ,., ?2, pp&1,(-1,- 0in )ussian1&

11C

*. om"leteness 9heorems CModel 9heoryD


*.'. Inter"retations and Models
7n principle, to do the so-called pure mathematics, i&e& simply to pro#e theorems, one needs only <synta7< E axioms and rules of inference& 6nd, for com"uters, this is the only ay of doin! mathematics; Gut ho about <semantics< E about our intended S#isionN from hich e started desi!nin! our predicate lan!ua!e and formulatin! axioms8 9irst of all, e must understand that there is no ay of formulatin! 0in the predicate lan!ua!e and axioms1 of a%% the feat$res of our domain of interest& 9or example, hat information can be deri#ed about the person Britney from our Mtheory for peopleN of 'ection 1&(8 6ll e can deri#e ill be formulated in terms of the follo in! predicates:
5ale0x1 E means Sx is a maleS4 9emale0x1 E means Sx is a femaleS4 5other0x, y1 E means Sx is mother of yS4 9ather0x, y1 E means Sx is father of yS4 5arried0x, y1 E means Sx and y are marriedS4 x^y&

Thus, there is no ay of obtainin! from such a theory of any information that canBt be formulated in these terms, for example, about the a!e, colour of eyes etc& 6nd thus, by communicatin! our predicate lan!ua!e and our axioms to a computer, e ha#e communicated only a small part of all the features of our domain of interest& $ence, neither our "redicate language, nor our a7ioms can s"ecify our initial OvisionP com"letely. 3nd, if so E in "rinci"le, one can imagine many different OvisionsP behind our ne8 language and a7ioms6 7t may seem that if, instead of our MpeopleBs domainN that is #ery rich in details, e ill consider, for example, natural numbers, then the situation ill become better, and e ill be able to describe our informal M#isionN unambi!uously8 9or example, let us considered the lan!ua!e of first order arithmetic 0lan!ua!e primiti#es: x, y, &&&, 0, 1 \, ], ^1, and the follo in! non-usual SGoolean #isionS G behind it&

11, a1 6s the domain of G 0the Star!etS set of ob>ects1, instead of the set of all natural numbers, let us consider the set of SGoolean #aluesS D G ^ Xt, fY& Thus, no , the #ariables x, y, &&& can take only #alues t, f& b1 The interpretation mappin! intG assi!ns: to the ob>ect constant 1 3 the ob>ect t, to the ob>ect constant 0 3 the ob>ect t, thus: intG001^f, intG011^t& c1 To the function constant S\S e assi!n the ell-kno n +isA$nction truth table: intG0\1^ G G , to the function constant S]S 3 the ell-kno n conA$nction truth table: intG0]1^ G G & d1 To the predicate constant S^S 3 the eQuality predicate for the set D G, i&e& intG0^1 ^ X0t, t1 , 0f, f1Y& 7s this #ision S orseS than the usual one in#ol#in! MrealN natural numbers8 7t seems, it is orse, because the follo in! axiom of arithmetic: x\1^y\1 ` x^y is false under this #ision& 7ndeed, set x^0 and y^1: 0\1^1\1 ` 0^1& Here, the premise is true: 0 +1= f t =t =1, 1+1 =t t =t =1 , but the conclusion is not: 0^1 means f^t& Rn the other hand, the follo in! theorem of Goolean al!ebra: x\x^x is true under the abo#e SGoolean #isionS 0 t t =t , f f = f 1, but it is false under the usual #ision in#ol#in! natural numbers& Thus, if t o theories share the same lan!ua!e 0as do Goolean al!ebra and first order arithmetic1, then the S#alidityS of a #ision may depend on the formulas 0axioms and theorems1 that e are expectin! to be true& 7f e consider only a language, then many different and e#en stran!e interpretations-#isions ill be possible& Gut if e consider a theory 0i&e& a lan!ua!e plus some specific axioms1, then only a part of the interpretations-#isions ill be #alid 3 only those ones, under hich the specific axioms of our theory ill be true& 'uch interpretations-#isions are called models of the theory&
3nother e7am"le: in our Slan!ua!e for peopleS e used names of people 0 /Britney, /ohn, Paris, Peter, (((1 as ob>ect constants and the follo in! predicate constants: 5ale0x1 E means Sx is a maleS4 9emale0x1 E means Sx is a femaleS4 5other0x, y1 E means Sx is mother of yS4 9ather0x, y1 E means Sx is father of yS4 5arried0x, y1 E means Sx and y are marriedS4 x^y E means Sx an y are the same personS&

120
?o , let us fix the list of . names: Britney, /ohn, Paris, Peter, and let us consider the follo in! interpretation H of the lan!ua!e: a1 The domain 3 and the ran!e of #ariables 3 is DH ^ Xbr, >o, pa, peY 0. character strin!s1& b1 intH0Gritney1^br, intH0Hohn1^>o, intH0%aris1^pa, intH0%eter1^pe& c1 intH05ale1 ^ X>o, peY4 intH09emale1 ^ Xbr, paY& d1 intH05other1 ^ X0pa, br1, 0pa, >o1Y4 intH09ather1 ^ X0pe, >o1, 0pe, br1Y& e1 intH05arried1 ^ X0pa, pe1, 0pe, pa1Y& f1 d1 intH0^1 ^ X0br, br1, 0>o, >o1, 0pa, pa1, 0pe, pe1Y& *nder this interpretation, Min this small orldN, it is true that, Smothers are femalesS, and that Sall fathers are married peopleS 0under this interpretation, not in the real orld;1& Thus, under this interpretation, the correspondin! formulas x05other0x1`9emale0x11 and x09ather0x1`y 5arried0x, y11 are Qualified as true& Gut, under this interpretation, Min this small orldN, it not true that Seach person possess a motherS& The correspondin! formula xy 5other0y, x1 is Qualified as false& !7ercise *.'.I. Guild another interpretation 0a Mcra:yN one;1 of our M. people lan!ua!eN, under hich the follo in! formulas are true: Msome people are both male and femaleN, Mthere are sex-less peopleN, Ma person may marry herselfN, Ma person may be mother of herselfN& Gy introducin! s"ecific nonBlogical a7ioms, i&e& by introducin! M. people theoryN instead of pure axiom-less M. people lan!ua!eN e can disQualify this Mcra:yN interpretation& 9or example, the follo in! axioms are false under it:
x ( .a%e ( x ) Fema%e ( x )) ? x 7 ( .a%e ( x ) Fema%e ( x )) &

Model 9heory <ould the notion of Marbitrary #isionN be defined precisely8 9or a particular predicate lan!ua!e and particular axioms 3 is there only one M#isionN possible8 Tryin! to ans er these Questions, e arri#e at the so-called model theory& Model theory is a #ery specific approach to in#esti!ation of formal theories& 5odel theory is usin! 0up to1 the full po er of set theory& In model theory, 8e investigate formal theories by using set theory as a metaBtheory.
Paul %ernays, in 1,/C: S6s Gernays remarks, syntax is a branch of number theory and semantics the one of set theory&S 'ee p& .=0 of $ao 1ang& @7AHTK K@6)' R9 9R*?D6T7R?6+ 'T*D7@'& ,ia%ectica, Vol& 12, 7ssue (., pp& .---.,=, December 1,/C 0a#ailable online at Glack ell 'yner!y1&

7n 'ections .&1-.&( e ill de#elop model theory for the classical logic , and in 'ections .&.-.&/ 3 model theory for the constructi#e lo!ic& 7n the classical model theory, e ill replace our #a!ue S#isionsN by relati#ely ell-defined mathematical structures 3 the so-called inter"retations& 6s e ill see, interpretations are allo ed to be non-constructi#e&

121 Technically, an interpretation ill be a relati#ely ell-defined ay of assi!nin! Sprecise meanin!sS to all formulas of a predicate lan!ua!e& 6ny particular predicate lan!ua!e allo s multiple ays of assi!nin! Sprecise meanin!sS to its formulas 3 m$%tip%e interpretations& Inter"retation of a language E the s"ecific "art +et + be a predicate lan!ua!e containin! ob>ect constants c1, &&&, ck, &&& , function constants f1, &&&, fm, &&&, and predicate constants p1, &&&, pn, &&&& 6n interpretation H of the lan!ua!e + consists of the follo in! t o entities: a1 6 nonBem"ty set DH 3 the domain of interpretation 0it ill ser#e first of all as the ran!e of ob>ect #ariables1& 0Kour fa#orite set theory comes in here&1 b1 6 mappin! intH that assi!ns: - to each ob>ect constant ci 3 a member intH0ci1 of the domain DH 0thus, ob>ect constants SdenoteS particular ob>ects in DH1, - to each function constant fi 3 a function intH0fi1 from DH x &&& x DH into DH 0of course, intH0fi1 has the same number of ar!uments as fi1, - to each predicate constant pi 3 a predicate intH0pi1 on DH, i&e& a subset of DH x &&& x DH 0of course, intH0pi1 has the same number of ar!uments as pi1& Thus, in a sense, the mappin! intH assi!ns Smeanin!S to the lan!ua!e primiti#es& The most popular example 3 let us consider the so-called standard inter"retation ' of first order 0%eano1 arithmetic %6: a1 The domain is D' ^ X0, 1, 2, &&&Y E the set of all natural numbers Sas kno itS 0more precisely 3 as you define it in your fa#orite set theory1& e

b1 The mappin! int' assi!ns: to the ob>ect constant 0 3 the number 0, to the ob>ect constant 1 3 the number 1, to the function constant S\S E the function x\y 0addition of natural numbers1, to the function constant S]S E the function x]y 0multiplication of natural numbers1, to the predicate constant S^S E the predicate x^y 0eQuality of natural numbers1&
Ket another interpretation H1 of the same lan!ua!e: a1 The domain is DH1 ^ Xe, a, aa, aaa, &&&Y E the set of all strin!s built of the letter SaS 0e is the empty strin!1& b1 The mappin! intH1 assi!ns: to the ob>ect constant 0 3 the empty strin! e, to the ob>ect

122
constant 1 3 the strin! SaS, to the function constant S\S E the concatenation function of strin!s, to the function constant S]S E y times concatenation of x, to the predicate constant S^S E the strin! eQuality predicate& Ket another interpretation H2 0there is no interpretation of the lan!ua!e1: a1 The domain is DH2 ^ XoY 3 a sin!le ob>ect o& b1 The mappin! intH2 assi!ns: to the ob>ect constant 0 3 the ob>ect o, to the ob>ect constant 1 3 the same ob>ect o, to the function constant S\S E the only possible function f0o,o1^o, to the function constant S]S E the only possible function f0o,o1^o, to the predicate constant S^S E the predicate X0o, o1Y& ay to disQualify it as a formally correct

'ome time later, e ill use specific non-lo!ical a7ioms to disQualify 0at least some of1 such SinadeQuateS interpretations& Ha#in! an interpretation H of the lan!ua!e +, e can define the notion of true formulas 0more precisely E the notion of formulas that are true under the inter"retation H1& 6s the first step, terms of the lan!ua!e + are interpreted as members of D H or functions o#er DH& 7ndeed, terms are defined as ob>ect constants, or ob>ect #ariables, or their combinations by means of function constants& The term ci is interpreted as the member intH0ci1 of DH& The #ariable xi is interpreted as the function Ji0xi1 ^ xi& 6nd, if t ^ fi0t1, &&&, tQ1, then intH0t1 is defined as the function obtained by substitutin! of functions intH0t11, &&&, intH0tQ1 into the function intH0fi1& 9or example 0first order arithmetic1, the standard interpretation of the term 01\11\1 is the number (, the interpretation of 0x\y\11]0x\y\11 is the function 0x\y\112& Im"ortant Q nonBconstructivity6 ?ote that, for an infinite domain D H, the interpretations of function constants may be nonBcom"utable functions& Gut, if they are all computable, then e can compute the S#alueS of any term t for any combination of #alues of #ariables appearin! in t& 6s the next step, the notion of true atomic formulas is defined& Rf course, if a formula contains #ariables 0as, for example, the formula x\y^11, then its Struth-#alueS must be defined for each combination of #alues of these #ariables& Thus, to obtain the truth-#alue of the formula p i0t1, &&&, tQ1 for some fixed #alues of the #ariables contained in t1, &&, tQ, e must first ScomputeS the #alues of these terms, and then substitute these #alues into the predicate intH0pi1&

12( 9or example 0first order arithmetic1, under the standard interpretation ', the formula x\y^1 ill be true, if and only if either x takes the #alue 0, and y takes the #alue 1, or x takes the #alue 1, and y takes the #alue 0& Rther ise, the formula is false& Im"ortant Q nonBconstructivity6 ?ote that, for an infinite domain D H, the interpretations of predicate constants may be nonBcom"utable predicates& Gut, if they ere all computable, then e could compute the Struth #alueS of any atomic formula 9 for any combination of #alues of #ariables appearin! in 9& Inter"retations of languages Q the standard common "art 6nd finally, e define the notion of true com"ound formulas of the lan!ua!e + under the interpretation H 0of course, for a fixed combination of #alues of their free #ariables1: a1 Truth-#alues of the formulas fG, BC , BC and G`< must be computed from the truth-#alues of G and < 0by usin! the ell-kno n classical truth tables 3 see 'ection .&2 belo 1& b1 The formula xG is true under H, if and only if G0c1 is true under H for all members c of the domain DH& c1 The formula xG is true under H, if and only if there is a member c of the domain DH such that G0c1 is true under H& 9or example 0first order arithmetic1, the formula y (( x = y + y )( x = y + y +1)) says that Sx is e#en or oddS& *nder the standard interpretation ', this formula is true for all #alues of its free #ariable x& 'imilarly, xy0x\y^y\x1 is a closed formula that is true under '& Im"ortant Q nonBconstructivity6 7t may seem that, under an interpretation, any closed formula is Seither true or flaseS& Ho e#er, note that, for an infinite domain DH, the notion of Strue formulas under HS is extremely nonB constructive: to establish, for example, the truth-#alue of the formula xG, or the formula xy0x\y^y\x1, e must #erify the truth of G0c1 for infinitely many #alues of c 0or a\b^b\a for infinitely many #alues of a and b1& Rf, course, this #erification cannot be performed on a computer& 7t can only 0sometimes1 be pro#ed&&& in some other theory& The Sde!ree of constructi#ityS of the formulas like as xy<0x,y1, xy:D0x,y,:1 etc& is e#en less than that&&&
!m"ty Domains2 Do you think, e should consider also em"ty domains of interpretation8 6ccordin! to the

12.
axiom +1(: 0G`G1`x0G`G1, hence, x0G`G1& 7n an empty domain, this formula ould be false& Thus, to co#er the empty domain, e ould be forced to re-consider the axioms andPor re-consider the traditional meanin! of x E see 0c1 abo#e& +et us concentrate on non-empty domains only&

+et us say that a formula of the lan!ua!e + is al8ays true under the interpretation H, if and only if this formula is true for all combinations of #alues of its free #ariables& 9hree Kinds of :ormulas 7f one explores some formula 9 of the lan!ua!e + in #arious interpretations, then three situations are possible: a1 9 is true in all interpretations of the lan!ua!e +& 9ormulas of this kind are called logically valid formulas& b1 9 is true in some interpretations of +, and false Q in some other interpretations of +& c1 9 is false in all interpretations of + 0then, of course, f9 is true in all interpretations1& 9ormulas of this kind are called unsatisfiable formulas. 9ormulas that are Snot unsatisfiableS 0i&e& formulas of kinds 0a1 and 0b11 are called, of course, satisfiable formulas& !7ercise *.'.'. Verify that: a1 : is satisfiable, if and only if R: is not logically valid& b1 : is logically valid, if and only if R: is unsatisfiable& Logically Valid :ormulas 'ome formulas are al ays true under all interpretations, for example: ( B 5 C )(C 5 , ) 5 ( B 5 ,) , F ( x ) 5 xF ( x ) , xF ( x ) 5 F ( x ) , x ( F ( x ) 5 G ( x )) 5 ( xF ( x ) 5 xG ( x )) , x ( F ( x ) 5 G ( x )) 5 ( xF ( x ) 5 xG ( x )) , x ( G ( x ) H ( x )) 5 ( xG ( x ) xH ( x )) ,

x ( G ( x ) H ( x ))5 ( xG ( x ) xH ( x )) &
'uch formulas are called logically valid& 5ore precisely, in a predicate lan!ua!e +, a formula is called logically valid, if and only if it is true in all inter"retations of the lan!ua!e + for all #alues of its free #ariables& Thus, a lo!ically #alid formula is true independently of its Smeanin!S E the

12/ interpretations of constants, functions and predicates used in it& Gut note that here, the 0classical;1 inter"retations of "ro"ositional connectives and >uantifiers remain fi7ed. 7n a sense, lo!ically #alid formulas are Mcontent-freeN: they do not !i#e us any specific information about features of ob>ects they are Mspeakin!N about& Im"ortant Q nonBconstructivity6 The notion of lo!ically #alid formulas is doubly nonBconstructive in the sense that the uni#ersal Quantifier Sfor all interpretationsS is added to the 0already1 non-constructi#e definition of a true formula& 6s e ill see in, all the axioms of our classical lo!ical axiom system V+ 1-+1/, 5%, AenW are lo!ically #alid formulas& 6nd that inference rules 5% and Aen !enerate only lo!ically #alid formulas& 7&e& e ill pro#e that all the formulas that can be "roved in the classical logic &L 'BL'-, MP, 5en+, are logically valid. 6s an example, let us #erify that the axiom +12: x90x1`90t1 is lo!ically #alid& +et us assume the contrary, i&e& that, under some interpretation H, for some #alues of its free #ariables, + 12 is false& 6ccordin! to the classical truth tables, this could be only, if and only if x90x1 ere true, and 90t1 ere false 0under the interpretation H, for the same abo#e-mentioned #alues of free #ariables1& +et us ScomputeS the #alue of the term t for these #alues of free #ariables 0since the substitution 90xPt1 is admissible, t may contain only these #ariables1, and denote it by c& Thus, 90c1 is false& Gut x90x1 is true, hence, 90a1 is true for all a in the domain D H, i&e& 90c1 also is true& <ontradiction& Hence, +12 is true under all interpretations for all combinations of its free #ariables 0if any1& !7ercise *.'... Verify that the remainin! - of the abo#e formulas are lo!ically #alid& 0Hint: follo the abo#e example E assume that there is an interpretation H such that the formula under Question is false for some #alues of its free #ariables, and deri#e a contradiction&1 Is our a7iom system of logic "o8erful enough to "rove 3LL the logically valid formulas2 The ans er is positi#e E see AFdelBs <ompleteness Theorem in 'ection .&(: a formula is lo!ically #alid, if and only of it is pro#able in the classical lo!ic V+1-+11, +12-+1/, 5%, AenW& Gut, of course, there are formulas that are not lo!ically #alid& 9or example, ne!ations of lo!ically #alid formulas are false in all interpretations, i&e& they are not lo!ically #alid& 'uch formulas are called unsatisfiable formulas& Gut there are formulas that are true in some interpretations, and false E in some other ones& 6n example of such formulas: the axiom of arithmetic x\1^y\1 `

12x^y considered abo#e& To conclude that some formula is not lo!ically #alid, e must build an interpretation H such that the formula under Question is false for some #alues of its free #ariables& 6s an example, let us #erify that the formula x ( p ( x ) 4 ( x )) 5 x p ( x ) x 4 ( x ) is not lo!ically #alid 0p, Q are predicate constants1& 2hy it is not8 Gecause the truth-#alues of p0x1 and Q0x1 may beha#e in such a ay that p ( x )4 ( x ) is al ays true, but neither x p0x1, nor x Q0x1 is true& 7ndeed, let us take the domain D ^ Xa, bY, and set: 7 a b "C7D true false >C7D false true

7n this interpretation, p ( a )4 ( a ) ^ true, p ( b ) 4 ( b ) ^ true, i&e& the premise x ( p ( x ) 4 ( x )) is true& Gut the formulas x p0x1, x Q0x1 both are false& Hence, in this interpretation, the conseQuent x p ( x ) x 4 ( x ) is false, and thus, x ( p ( x ) 4 ( x )) 5 x p ( x ) x 4 ( x ) is false& 2e ha#e built an interpretation, makin! false the formula under Question& D&@&D& Rn the other hand, this formula is satisfiable 3 there is an interpretation under hich it is true& 7ndeed, let us take D^XaY as the domain of interpretation, and let us set p0a1^Q0a1^true& Then all the formulas x ( p ( x ) 4 ( x )) , x p ( x ) , x 4 ( x ) become true, and so is the entire formula under consideration& D&@&D& !7ercise *.'.). Verify that the follo in! formulas are satisfiable, but not lo!ically #alid 0p, Q, r are predicate constants1: a1 p ( x , y ) p ( y , 3 ) 5 p ( x , 3 ) , b1 Q0x1`x Q0x1, c1 0x Q0x1`x r0x11`x0Q0x1`r0x11, c11 x0p0x1`G1`0x p0x1`G1, here G does not contain x, d1 xy p0x, y1`yx p0x, y1, e1 x 4 ( x ) x r ( x ) 5 x ( 4 ( x )r ( x )) , f1 x 7 p ( x , x ) x y 3 ( p ( x , y ) p ( y , 3 ) 5 p ( x , 3 )) ` x y ( x = y p ( x , y ) p ( y , x )) &

12= Hint& 9or the domain D^Xa, bY, use table form to define your interpretation of a binary predicate letter r0x,y1, for example, 7 a a b b y a b a b rC7, yD false true true false

!7ercise *.'.*. 7s the follo in! formula lo!ically #alid, or not 0p, Q are predicate constants1: 0x p0x1`x Q0x11`x0p0x1`Q0x11& 0Hint: follo the abo#e example E use natural numbers or other ob>ects tryin! to build an interpretation H such that the formula under Question is false for some #alues of its free #ariables&1 ,atisfiability 2e already kno that, in a predicate lan!ua!e +, a formula 9 is called satisfiable, if and only if there is an inter"retation of the lan!ua!e + such that 9 is true for some #alues of its free #ariables 0 e ill say also that 9 is satisfied under this inter"retation1& 6 set of formulas 91, &&&, 9n, &&& is called satisfiable, if and only if there is an interpretation under hich the formulas 91, &&&, 9n, &&& are satisfied sim$%taneo$s%y& !7am"les& a1 9ormula x p0x1 is, of course, not lo!ically #alid, but it is satisfiable, because it is true in the follo in! interpretation H: D H^XbY, p0b1 is true& b1 9ormulas x]0^0, x\y^y\x and x\0y\:1^0x\y1\: are not lo!ically #alid 0see @xercise .&1&= belo 1, but they are satisfiable, because they are true under the standard intepretation of arithmetic& !7ercise *.'.-. a1 Verify that the formula xy0p0x1`p0y11 is al ays true in all one-element interpretations 0i&e& hen the interpretation domain consists of a sin!le element1, but is false in at least one t o-element interpretation 0p is a predicate constant1& b1 Verify that the formula

12C x y 3 [( p ( x ) 6 p ( y ))( 4 ( y ) 6 4 ( 3 ))( r ( 3 ) 6 r ( x ))] is al ays true in all one- and t o-element interpretations, but is false in at least one three-element interpretation 0p, Q, r are predicate constants1& c1 %ro#e that the formula xy 90x,y1 is lo!ically #alid, if and only if so is the formula x 90x, !0x11, here ! is a function constant that does not appear in 9& d1 %ro#e that the formula xy: 90x,y,:1 is satisfiable, if and only if so is the formula xy 90x, y, h0x,y11, here h is a function constant that does not appear in 9& Logical onse>uences S9 implies AS, or Sthe formula A follo s from the formula 9S E hat should this mean in !eneral8 7f 9 is true, then A is true8 6l ays, or under some specific conditions8 +et us specify all these SconditionsS as formulas 61, &&&, 6n 0the formula 9 included1& Then, A follo s from 6 1, &&&, 6n unconditionally 0Slo!icallyS1, i&e& if 61, &&&, 6n are all true, then A must be true ithout any other conditions& 'ince the notion of StrueS e ha#e formali:ed as Strue in interpretationS, e can formali:e the notion of Slo!ical conseQuenceS as follo s: A is a logical conse>uence of 61, &&&, 6n, if and only if 5 is true under any inter"retation, under 8hich 3', ..., 3n are all true& Rr, as follo s: A is a logical conse>uence of 61, &&&, 6n, if and only if 5 is true in any model of 3', ..., 3n& !7ercise *.'./& Verify that: a1 The formula A is a lo!ical conseQuence of formulas 61, &&&, 6n, if and only if the formula A1&&& An 5G is lo!ically #alid& b1 7f the set of formulas 61, &&&, 6n is satisfiable, then the formulas G, fG cannot both be lo!ical conseQuences of 61, &&&, 6n& c1 The formula A is a lo!ical conseQuence of formulas 6 1, &&&, 6n, if and only if the set 61, &&&, 6n, fA is unsatisfiable&

12, 2e ill pro#e in 'ection .&( that A is a lo!ical conseQuence of 61, &&&, 6n, if and only if V+1-+11, +12-+1/, 5%, AenW: A1&&& An U- A, i&e& if the formula A1&&& An 5G is pro#able in the classical lo!ic& 9heories and 9heir Models 7f T is a first order theory, and H is an interpretation of its lan!ua!e, and if H makes true the specific axioms of T, then 0traditionally1 H is called a model of T& 9or non-mathematical people, the term Smodel of a theoryS may seem some hat stran!e: in SnormalS branches of science, theories ser#e as a basis for buildin! models of natural phenomena, technical de#ices etc& Gut only the term is stran!e 0Supside do nS1 here, the process is the same as in SnormalS branches of science: first order theories S!enerateS their models, and these models can be used for modelin! natural phenomena, technical de#ices etc& ,"ecific a7ioms of a first order theory 9 are not logically valid formulas6 They are not true in all interpretations, they are true only in the models of T& 5odels of T E it is a "ro"er subclass of all the possible interpretations& 9or example, the Sob#iousS arithmetical axioms like as x\0^x 0or, theorems like as x\y^y\x1 are not lo!ically #alid& 7f e ould interpret 0 as the number St oS, then x\0 and x ill be eQual; +o!ically #alid formulas must be true under all interpretations; !7ercise *.'.0. a1 Verify that, if a theory has a model, then the set of its specific axioms is satisfiable& b1Verify that x^x, x]0^0, x\y^y\x and x\0y\:1^0x\y1\: are satisfiable, but not lo!ically #alid formulas& 6s e already noted abo#e, in a sense, logically valid formulas <do not contain information< 0are Mcontent-freeN1 E >ust because they are true in all interpretations, i&e& they are true independently of the Smeanin!S of lan!ua!e primiti#es& 7ndeed, let us consider the formulas x\0^x ` x\0^x, and x\0^0 ` x\0^0& Goth are lo!ically #alid, but do e !et more information about :ero and addition after readin! them8 6nother example: 2]2^/ ` 2]2^/, or 2]2^. ` 2]2^., these formulas also are lo!ically #alid, but do they help in computin! the #alue of 2]28 The specific axioms of some theory T, on the contrary, do Scontain informationS E they separate a proper subclass of all interpretations E models of T& Do the axioms of first order arithmetic MspecifyN the standard interpretation ', i&e& are the axioms of first order arithmetic true in this inter"retation only2

1(0 ?o, there are many nonBstandard inter"retations makin! these axioms true; 5ore: ?on-standard arithmetic in 2ikipedia& 9ransitive Predicates and #ecursion +et us return to the problem that e considered already in 'ection 1&2& Ho about the predicate 3ncestorC7, yD E Sx is an ancestor of yS8 <ould it be expressed as a formula of our Slan!ua!e for peopleS8 The first idea E let us SdefineS this predicate recursi#ely: x y ( Father ( x , y ).other ( x , y ) 5 Ancestor ( x , y )) 4 x y 3 ( Ancestor ( x , y ) Ancestor ( y , 3 ) 5 Ancestor ( x , 3 )) & The second rule declares the transiti#ity property of the predicate& The abo#e t o formulas are a7ioms, allo in! to deri#e essential properties of the predicate 6ncestor0x, y1& Gut ho about a sin!le formula 90x, y1 in the Slan!ua!e for peopleS, expressin! that Sx is an ancestor of yS8 'uch a formula should be a tricky combination of formulas 9ather0x, y1, 5other0x, y1 and x^y& 6nd such a formula is im"ossible; 9or the proof 3 see arlos 3reces& %h&D& Thesis, 2000, Theorem 1&2& !7ercise *.'.G 0for smart students1& @xplain the precise meanin! of the statement: in the Slan!ua!e for peopleS, formula 90x, y1 expresses that Sx is an ancestor of yS&

*... lassical Pro"ositional Logic Q 9ruth 9ables


@mil +eon %ost 01C,=-1,/.1& S&&& %ostBs %h&D& thesis, in hich he pro#ed the completeness and consistency of the propositional calculus described in the Principia .athematica by introducin! the truth table method& He then !eneralised his truth table method, hich as based on the t o #alues StrueS and SfalseS, to a method hich had an arbitrary finite number of truth-#alues&&& 7n the 1,20s %ost pro#ed results similar to those hich AFdel, <hurch and Turin! disco#ered later, but he did not publish them& He reason he did not publish as because he felt that a Bcomplete analysisB as necessary to !ain acceptance&S 06ccordin! to 5acTutor History of 5athematics archi#e1& 9irst, let us consider the classical "ro"ositional logic& Here, each formula is built of some is built of some MatomsN G1, G2, &&&, Gn by usin! propositional connecti#es only 0i&e& BC , BC , 7 B , B 5C 1& Rur axioms for this lo!ic e represented as axiom schemas +1-+11, in hich the letters G, <, D could be replaced by any formulas&

1(1 7s our list +1-+11 of classical propositional axiom schemas McompleteN8 6renBt some necessary axiom schemas missin! there8 7f somethin! necessary is missin!, e must add it to the list& This problem as sol#ed by @mil +& %ost in 1,20& He pro#ed that if one 8ould add to L'BL'' as an a7iom schema any formula that canJt yet be "roved from these a7ioms, then one 8ould obtain a system, in 8hich all formulas are "rovable, i.e. an inconsistent system. Thus, nothin! is missin! in our list of classical propositional axioms& %ost pro#ed his theorem by usin! the so-called classical truth tables 0a specific interpetation 3 in terms of the abo#e 'ection .&11& @ach propositional atom may take any of t o truth-#alues 3 tr$e and fa%se& 6nd, if e already kno the truth-#alues of the formulas G, <, then e can use truth tables to compute the truth-#alues of the formulas BC , BC , 7 B , B 5C & 7f G is false, and < is false, then BC is false& 7f G is false, and < is true, then BC is false& 7f G is true, and < is false, then BC is false& 7f G is true, and G is true, then BC is true& % 0 0 0 0 1 0 1 0 0 1 1 1 7f G is false, and < is false, then BC is false& 7f G is false, and < is true, then BC is true& 7f G is true, and < is false, then BC is true& 7f G is true, and < is true, then BC is true& % 0 0 0 0 1 1 1 0 1 1 1 1 7f G is false, then fG is true& 7f G is true, then fG is false& B C B C

1(2 % R% 0 1 1 0 ?o problems so far& 7f G is false, and < is false, then G`< is hat8 True8 9alse8 Gut, hy8 7f G is false, and < is true, then G`< is hat8 True8 9alse8 Gut, hy8 7f G is true, and < is false, then G`< is false, of course& 7f G is true, and < is true, then G`< is hat8 %erhaps, not false8 Hence, true8 Ho to ans er the ( hatBs8 7f G is false, then G`< possesses no real meanin!& 6nd, if e already kno that G is true, and < is true, then G`< is no more interestin!& Gut, if a definite Struth-#alueS for G`< is mandatory in all cases, then e can great%y simp%ify the sit$ation by assumin! that G`< is al ays true, except, if G is true, and < is false& Thus: 7f G is false, and < is false, then G`< is true& 7f G is false, and < is true, then G`< is true& 7f G is true, and < is false, then G`< is false& 7f G is true, and < is true, then G`< is true& % %F

0 0 1 0 1 1 1 0 0 1 1 1 This definition is eQui#alent to sayin! that G`< is true, if and only if 7 ( B7 C ) is true or: G`< is false, if and only if G is true, and < is false& 7n this ay, ha#in! any formula 9 and some assi!nment of truth-#alues to its atoms, e can compute the truth-#alue of 9& Gut hat ould happen to some propositional formula 9, if e ould try a%% the possib%e truth-#alues of all the propositional atoms occurrin! in 98 There are three possibilities: 9 takes only true #alues4 9 takes only false #alues4

1(( 9 takes both of #alues& Lemma *...'. *nder the classical truth tables, all the classical propositional axioms +1-+11 take only true #alues& Proof. 9irst, let us #erify +11 and +10: % R% 0 1 1 0 % 1 1 R% %F 1 1 0 1 R%FC%F D 1 1 1 1
B B

0 0 1 0 1 1 1 0 0 1 1 0

!7ercise *...'. Verify +1-+,& 'ee also: STruth TablesS from The 2olfram Demonstrations %ro>ect& <ontributed by: Hector Lenil& Lemma *..... *nder the classical truth tables, if the formulas G and G`< take only true #alues, then so does <& 7&e& from Sal ays trueS formulas, 5odus %onens allo s deri#in! only of Sal ays trueS formulas& Proof. +et us assume that, in some situation, < takes a false #alue& 7n the same situation, G and G`< take true #alues& 7f G is true, and < is false, then G`< is false& <ontradiction& Hence, < takes only true #alues& D&@&D& 4ote& 7n the proof of +emma .&2&2, only the third ro of implication truth table as si!nificant: if G is true, and < is false, then G`< is false; 9heorem *...) Csoundness of the classical "ro"ositional logicD. 7f V+1-+11, 5%W: U- 9, then, under the classical truth tables, 9 takes only true #alues& 7n particular: the classical propositional lo!ic is consistent 3 in the sense that one cannot pro#e V+1-+11, 5%W: U- G7 G , for any formula A& Proof. Gy induction, from +emmas .&2&1 and .&2&2&

1(. om"leteness of lassical Pro"ositional Logic Ho about the con#erse statement of Theorem .&2&(: if, under the classical truth tables, formula 9 takes only true #alues, then V+ 1-+11, 5%W: U- 98 7&e&, are our a7ioms "o8erful enough to "rove any formula that is taking only true values2 The ans er is SyesS: 9heorem *...* Ccom"leteness of the classical "ro"ositional logicD. 6ssume, the formula 9 has been built of formulas G 1, G2, &&&, Gn by usin! propositional connecti#es only& 7f, under the classical truth tables, for any truth-#alues of G 1, G2, &&&, Gn, formula 9 takes only true #alues, then: a1 in the constructi#e lo!ic, V+1-+10, 5%W: B17 B1 , B 27 B 2 , &&& , B n7 Bn U- 9, b1 in the classical lo!ic, V+1-+11, 5%W: U- 9& orollary *...*& The classical propositional axioms V+1-+11, 5%W are ScompleteS in the sense that if one ould add any formula that canBt yet be pro#ed from these axioms, then one ould obtain a system, in hich all formulas are pro#able, i&e& an inconsistent system& Rf course, 0b1 follo s from 0a1 immediately E all the premises B17 B1 , B 27 B 2 , &&& , B n7 B n are instances of the axiom +11& The corollary also follo s immediately& 7ndeed, if some formula 9 canBt be pro#ed from V+1-+11, 5%W, then it takes false #alue for some combination of truth-#alues of its atoms& )eplace each true atom by the formula 6`6, and each false atom 3 by f06`61& 7n this ay e obtain a formula 9B that takes only false #alues, i&e& f9B takes only true #alues, and hence, can be pro#ed from V+1-+11, 5%W& Thus, if e ould add 9 to V+1-+11, 5%W as an axiom schema, then, in this system, the formulas 9B and f9B ill be pro#able, and by +10 3 any formula ill be pro#able& 4ote. 6ssume, the formula 9 is built of atoms G 1, G2, &&&, Gn by usin! propositional connecti#es only& 7f, under the classical truth tables, for any 0possible and impossible1 truth-#alues of G1, G2, &&&, Gn, formula 9 takes only true #alues, then 9 is called a tautology& Theorem .&2&. says that any tautology can be "roved in the classical "ro"ositional logic& <ompleteness of the classical propositional lo!ic %ost in his 1,20 %h&D& thesis, and published as as first pro#ed by @mil +&

!. Post& 7ntroduction to a !eneral theory of elementary propositions ( American /o$rna% of .athematics, 1,21, #ol& .(, pp&1-(-1C/&

1(/
6bout the history, see also: #ichard Aach& <ompleteness before %ost: Gernays, Hilbert, and the de#elopment of propositional lo!ic& The B$%%etin of Symbo%ic Logic, 1,,,, #ol& /, ?(, pp&((1-(-- 0online copy a#ailable1&

9ollo in! an ele!ant later idea by +as:lo $almar before tryin! to pro#e this theorem&

e need t o simple lemmas Acta

L. Kalmar& *eber 6xiomatisiebarkeit des 6ussa!enkalkuels& scientiari$m mathematicar$m S3ege+!& 1,(.-(/& #ol& =, pp& 222-2.(&

Lemma *...-. 7n the constructi#e lo!ic, one can ScomputeS the classical truth#alues of 7 B , B 5 C , B C , B C in the follo in! sense: 4egation V W: fG U- fG Im"lication V+10, 5%W: fG, f< U- G`< on@unction Dis@unction

V+1, +2, +(, +,, 5%W: V+1-+,, 5%W: fG, f< U7 ( B C ) fG,f< U7 ( B C )

V+1, +2, +,, V+10, 5%W: 5%W: fG, < U- G`< G U- ffG

V+1, +2, +(, +,, 5%W: V+=, 5%W: fG, < U- 7 ( BC ) fG, < U- BC

V+1, +2, +,, 5%W: V+1, +2, +., +,, 5%W: V+-, 5%W: G, f< U- f0G`<1 G, f< U- 7 ( BC ) V+1, 5%W: G, < U- G`< V+/, 5%W: G, < U- BC G, f< U- BC V+-, 5%W: G, < U- BC

4ote& Thus, to ScomputeS the classical truth-#alues, the axiom +11 is not necessary; Proof. R% SB R% 7mmediately, in any lo!ic& % SB R R% Gy Theorem 2&.&.& V+1, +2, +,, 5%W: U- 6`ff6& R%, SB %F R%, R SB %F

1(Gy axiom +10: fG`0G`<1 e obtain fG U- G`<& This co#ers both cases& %, R SB RC%F D This is exactly Theorem 2&.&10c1 V+1, +2, +,, 5%W& %, SB %F Gy axiom +1: <`0G`<1 e obtain < U- G`<& R%, R SB ( B C ) R%, SB ( B C ) Gy axiom +(: BC 5 B and the <ontraposition +a 0Theorem 2&.&21 V+ 1, +2, +,, 5%W: U- 06`G1`0fG`f61 e obtain U- 7 B 5 7 ( B C ) , and fG U7 ( BC ) & This co#ers both cases& %, R SB ( B C ) Gy axiom +.: BC 5 C and the <ontraposition +a 0Theorem 2&.&21 V+1, +2, +,, 5%W: U- 06`G1`0fG`f61 e obtain U- 7C 5 7( BC ) , an f< U7 ( B C ) & %, SB B C e obtain G, < U- BC & Gy axiom +/: B 5 ( C 5 BC ) R%, R SB ( B C ) Gy Theorem 2&.&100b1& R%, SB BC e obtain < U- BC & Gy axiom +=: C 5 B'C %, R SB BC %, SB BC e obtain G U- BC & This co#ers both cases& Gy axiom +-: B 5 BC D&@&D& 6s the next step, e ill !enerali:e +emma .&2&/ by sho in! ho to ScomputeS truth-#alues of arbitrary formula 9, hich is built of formulas G 1, G2, &&&, Gn by usin! more than one propositional connecti#e& 9or example, let us take the formula BC 5 B C : G < BC
B C

B C 5 B C

1(= 0 0 0 0 1 1 1 0 1 1 1 1 0 0 0 1 1 0 0 1

2e ill sho that, in the constructi#e lo!ic V+1-+10, 5%W: fG, f< U- BC 5 B C , fG, < U- 7 ( BC 5 B C ) , G, f< U- 7 ( BC 5 B C ) , fG, f< U- BC 5 B C & Lemma *.../. 6ssume, the formula 9 has been built of formulas G1, G2, &&&, Gn by usin! propositional connecti#es only& 6ssume that, if the formulas G1, G2, &&&, Gn take the truth-#alues #1, #2, &&&, #n respecti#ely, then, for these #alues, formula 9 takes the truth-#alue & Then, in the constructi#e lo!ic, e can ScomputeS the truth-#alue of 9 in the follo in! sense: V+1-+10, 5%W: #1G1, #2G2, &&&, #nGn U- 9, here: 9 denotes 9, if is true, and f9, if #i is true, and fGi, if #i is false& Proof. Gy induction& Induction base. 9 is one of the formulas G i& Then ^#i, and, of course, in any lo!ic, #iGi U- 9& Induction ste". ?ote that +emma .&2&/ represents the assertion of +emma .&2&- for formulas built of G1, G2, &&&, Gn by usin! a sin!le propositional connecti#e& 1& : is R5& Gy the induction assumption, V+1-+10, 5%W: #1G1, #2G2, &&&, #nGn U- BA, here B represents the truth-#alue of A& Gy +emma .&2&/, V+1-+10, 5%W: BA U- 9, hence, V+1-+10, 5%W: #1G1, #2G2, &&&, #nGn U- 9& 2& : is 5 o $, here o is implication, con>unction, or dis>unction& Gy the induction assumption, V+1-+10, 5%W: #1G1, #2G2, &&&, #nGn U- BA, is false, and # iGi denotes Gi, if

1(C here B represents the truth-#alue of A, and V+1-+10, 5%W: #1G1, #2G2, &&&, #nGn U- BBH, here BB represents the truth-#alue of H& Gy +emma .&2&/, V+1-+10, 5%W: BA, BBH U- 9, hence, V+1-+10, 5%W: #1G1, #2G2, &&&, #nGn U- 9& D&@&D& Proof of 9heorem *...*CaD. Gy +emma .&2&-: V+1-+10, 5%W: G1, #2G2, &&&, #nGn U- 9, V+1-+10, 5%W: fG1, #2G2, &&&, #nGn U- 9, because 9 takes only true #alues& Gy V+1, +2, 5%W Deduction Theorem 1, V+1-+10, 5%W: #2G2, &&&, #nGn U- G1`9, V+1-+10, 5%W: #2G2, &&&, #nGn U- fG1`9, +et us mer!e these t o proofs and append an instance of the axiom +C: U- ( B1 5 F ) 5 (( 7 B1 5 F ) 5 ( B17 B1 5 F )) & Hence, by 5%: V+1-+10, 5%W: #2G2, &&&, #nGn U- B17 B;5 F , and V+1-+10, 5%W: G1#fG1, #2G2, &&&, #nGn U- 9& Gy repeatin! this operation e obtain Theorem .&2&.0a1: V+1-+10, 5%W: B17 B1 , B 27 B 2 , &&& , B n7 Bn U- 9& D&@&D& om"utational om"le7ity of the Problem 9rom no on, e could for!et our ability of pro#in! formulas in the classical propositional lo!ic, learned in 'ection 2& 7ndeed, in order to #erify, is a formula pro#able in V+1-+11, 5%W, or not, e can simply check, under the classical truth tables, takes this formula only true #alues, or not& 7s this checkin! really simpler than pro#in! of formulas in V+1-+11, 5%W8 7f the formula contains n different atoms 6, G, <, &&&, then its truth table contains 2n ro s that must be checked one by one& Rf course, if the formula

1(, contains 2 atoms 0like as ( A 5 B ) 5 7 A B , or ( atoms 0like as the 6xiom +21, then its truth table consists of . or C ro s E for most people this is a feasible task& Gut the Struth tableS for a formula containin! (2 atoms contains four billions of ro s to check&&& 'o, let us try in#entin! a more efficient al!orithm8 7t seems, e ill ne#er succeed E the problem of determinin! the classical pro#ability of propositional formulas belon!s to the complexity class Mco-?%completeN, see Goolean satisfiability problem in 2ikipedia& 6nd the problem of determinin! the constructive pro#ability of propositional formulas is e#en harder 3 it belon!s to the complexity class M%'%6<@-completeN, see:
#ichard ,tatman. 7ntuitionistic propositional lo!ic is polynomial-space complete, Theoretical <omputer 'cience , 01,=,1, pp& -=3=2 0online copy a#ailable1&

*.). lassical Predicate Logic Q 5MdelJs om"leteness 9heorem


$urt AFdel 01,0--1,=C1 SHe is best kno n for his proof of AFdelBs 7ncompleteness Theorems& 7n 1,(1 he published these results in Hber forma% $nentschei+bare SIt3e +er Principia .athematica $n+ 'er-an+ter Systeme ( (((AFdelBs results ere a landmark in 20th-century mathematics, sho in! that mathematics is not a finished ob>ect, as had been belie#ed& 7t also implies that a computer can ne#er be pro!rammed to ans er all mathematical Questions&S 06ccordin! to 5acTutor History of 5athematics archi#e1& 6s Da#id Hilbert and 2ilhelm 6ckermann published in D.$ilbert, 1.3ckermann& Arund:ue!e der theoretischen +o!ik& Gerlin 0'prin!er1, 1,2C their, in a sense, SfinalS #ersion of the axioms of classical lo!ic, they obser#ed: S2hether the system of axioms is complete at least in the sense that all the lo!ical formulas hich are correct for each domain of indi#iduals can actually be deri#ed from them, is still an unsol#ed Question&S
0Quoted after ,. . Kleene& The 2ork of $urt AFdel& SThe Hournal of 'ymbolic +o!icS, December 1,=-, Vol&.1, ?., pp&=-1-==C 'ee also: Hilbert and 6ckermannBs 1,2C +o!ic Gook by 'tanley ?& Gurris1&

7ndeed, as e ill #erify belo , a1 all axioms of the classical lo!ic 0+ 1-+11, +12-+1/1 are lo!ically #alid, b1 the inference rules 5%, Aen allo to pro#e 0from lo!ically #alid formulas1 only lo!ically #alid formulas& Hence, in this

1.0 ay only lo!ically #alid formulas can be pro#ed& 'till, is our list of lo!ical axioms complete in the sense that all logically valid formulas can be "roved2 E the Question asked by Hilbert and 6ckermann in 1,2C& The ans er is SyesS E as $urt AFdel established in 1,2,, in his doctoral dissertation S*eber die Vollstlndi!keit des +o!ikkalkuelsS0#isit AFdelBs 6rchi#e in the %rinceton *ni#ersity +ibrary1& The correspondin! paper appeared in 1,(0: K. 5Mdel. Die Vollstlndi!keit der 6xiome des lo!ischen 9unktionenkalkuels& S5onatshefte fuer 5athematik und %hysikS, 1,(0, Vol&(=, pp&(.,-(-0& 5MdelJs om"leteness 9heorem. 7n any predicate lan!ua!e, a formula is lo!ically #alid, if and only if it can be pro#ed by usin! the classical lo!ic V+ 1+11, +12-+1/, 5%, AenW& 7n fact, a more !eneral theorem can be pro#ed: 9heorem *.).I 0Thanks to 'une 9olda!er for the idea&1& 7f T is a first order theory ith classical lo!ic, then some formula 9 is al ays true in all models of T, if and only if T pro#es 9& AFdelBs <ompleteness Theorem follo s from Theorem .&(&0, if the set of specific axioms of T is empty& 9irst, let us pro#e the easy part 0sometimes called the soundness theorem1 E that all the formulas that can be pro#ed by usin! the classical lo!ic V+ 1-+11, +12-+1/, 5%, AenW are lo!ically #alid& Lemma *.).'. 6ll the axioms of the classical lo!ic 0+1-+11, +12-+1/1 are lo!ically #alid& Proof. 11 *nder the classical truth tables, the propositional axioms + 1-+11 take only true #alues 0+emma .&2&11& Hence, these axioms are true under all interpretations& 2a1 L'.: x90x1`90t1, here 9 is any formula, and t is a term such that the substitution 90xPt1 is admissible& +et us assume that, under some interpretation H, for some #alues of its free #ariables, +12 is false& 6ccordin! to the classical truth tables, this could be only, if and only if x90x1 ere true, and 90t1 ere false 0under the interpretation H, for the same abo#e-mentioned #alues of free #ariables1& +et us ScomputeS the #alue of the term t for these #alues of free #ariables 0since the substitution 90xPt1 is admissible, t may contain only these #ariables1, and denote it by c& Thus, 90c1 is false& Gut x90x1 is true, hence, 90a1 is true for all a , / , i&e& 90c1 also is true& <ontradiction& Hence, + 12 is true under all

1.1 interpretations for all combinations of its free #ariables& 2b1 L'): 90t1`x90x1, here 9 is any formula, and t is a term such that the substitution 90xPt1 is admissible& 'imilarly, see @xercise .&(&1& 2c1 L'*: x0A`90x11`0A`x90x11, here 9 is any formula, and A is a formula that does not contain x as a free #ariable& +et us assume that, under some interpretation H, for some #alues of its free #ariables, +1. is false& 6ccordin! to the classical truth tables, this could be only, if and only if x0A`90x11 ere true, and A`x90x1 ere false 0under the interpretation H, for the same abo#e-mentioned #alues of free #ariables1 7f x0A`90x11 is true, then A`90c1 is true for all c , / & 'ince A does not contain x, this means that if A is true, then 90c1 is true for all c ,/ & 9rom the orher side, if A`x90x1 is false, then A is true, and x90x1 is false& 6nd finally, if x90x1 is false, then 90c1 is false for some c ,/ & Gut, as e established abo#e, if A is true, then 90c1 is true for all c ,/ & <ontradiction& Hence, under all interpretations, +1. is true for all combinations of its free #ariables& 2d1 L'-: x090x1`A1`0x90x1`A1, here 9 is any formula, and A is a formula that does not contain x as a free #ariable& 'imilarly, see @xercise .&(&1& D&@&D& !7ercise *.).'. Verify that the axioms +1( and +1/ are lo!ically #alid& Lemma *.)... 9rom lo!ically #alid formulas, inference rules 5% and Aen allo deri#in! only of lo!ically #alid formulas&& Proof. 1& Modus Ponens& 6ssume, G and G`< are lo!ically #alid formulas& Gy 5%, e deri#e <& 6ssume, < is not lo!ically #alid, i&e&, under some interpretation H, for some #alues of its free #ariables, < is false& *nder this interpretation H, for these #alues of free #ariables of <, the formulas G and G`< are true& Then, accordin! to the classical truth tables, < also must be true& <ontradiction& Hence, < is lo!ically #alid& 2& 5enerali;ation& 6ssume, 90x1 is lo!ically #alid, but x90x1 is not, i&e&, under some interpretation H, for some #alues of its free #ariables, x90x1 is false& Hence, under this interpretation H, for these #alues of free #ariables of x90x1, there is c ,/ such that 90c1 is false& Gut 90x1 is lo!ically #alid, i&e&

1.2 90c1 is true& <ontradiction& Hence, x90x1 is lo!ically #alid& D&@&D& orollary *.).) Csoundness of the classical "redicate logicD & 6ll the formulas that can be pro#ed by usin! the classical lo!ic V+ 1-+11, +12-+1/, 5%, AenW, are lo!ically #alid& Proof. 7mmediately, by +emmas .&(&1 and .&(&2& !7ercise *.).'T. Verify that if, under an interpretation H, all specific a7ioms of a theory T are true, then all theorems of T also are true under H& 0Hint: each theorem < is pro#ed by usin! some finite set of specific axioms, let us denote by G the con>unction of these axioms, consider the formula G`<, and use <orollary .&(&(&1 Rf course, the abo#e soundness theorem is the easy half of AFdelBs <ompleteness Theorem& To complete the proof, e must pro#e the con#erse: if some formula is lo!ically #alid, then it can be pro#ed by usin! the classical lo!ic V+1-+11, +12-+1/, 5%, AenW& $enkinJs Model !7istence 9heorem AFdelBs initial #ery complicated proof from 1,2, as !reatly simplified in 1,.=, hen +eon Henkin obser#ed in his %h&D& thesis that the hard part of the proof can be !enerali:ed, simplified and best presented as the so-called 5odel @xistence Theorem& The result as published in 1,.,:
L. $enkin. The completeness of the first-order functional calculus& SH& 'ymbolic +o!icS, 1,.,, #ol&1., pp&1/,-1--& 'ee also HenkinBs later account of his disco#ery: L. $enkin. The disco#ery of my completeness proofs& SThe Gulletin of 'ymbolic +o!icS, 1,,-, #ol&2, ?2, pp&12=-1/C& HenkinBs proof as further simplified by Aisbert Hasen>l!er in 1,/(: 5. $asen@Uger& @ine Gemerkun! :u HenkinBs Ge eis fuer die Vollstlndi!keit des %rldikatenkalkuels der ersten 'tufe& SH& 'ymbolic +o!icS, 1,/(, #ol&1C, pp&.2-.C&

7f T is an inconsistent theory, then there are no models of T& 7ndeed, if T pro#es a contradiction, i&e& a formula of the kind B B , then, in a model of T, the formula G must be true and false simultaneously& This is imposssible& Hence, if there is a model of T, then T is consistent& The con#erse Question: could it be possible that T is a consistent theory, but there are no models of T8 The ans er is !i#en in $enkinJs Model !7istence 9heorem. 7f a first order classical formal theory is

1.( consistent 0in the sense that, by usin! the classical lo!ic, it does not pro#e contradictions1, then there is a finite or countable model of this theory 0i&e& an interpretation ith a finite or countable domain, under hich all axioms and theorems of the theory are al ays true1& 7n the 1,20s, some people insisted that mere consistency of a theory 0in the syntactic sense of the ord E as the lack of contradictions1 is not sufficient to re!ard it as a meanin!ful theory E as a Stheory of somethin!S& 5odel @xistence Theorem says the contrary E 0syntactic;1 consistency of a theory is sufficient: if a theory does not contain contradictions, then it is a <theory of something< E it describes at least some kind of Smathematical realityS& 9or example, you may think that @uclidean !eometry is Smeanin!lessS E because it does not describe 100T correctly the spacial properties of the *ni#erse& Gut itBs your problem, not @uclidBs E use another theory, if necessary& @uclidean !eometry describes its o n kind of Smathematical realityS 3 and 100T correctly; +et us assume the 5odel @xistence Theorem 0 e 'ection1& Proof of 9heorem *.).I. 7f T pro#es 9, then 9 is al ays true in all models of T 0@xercise .&(&1J1& ?o , let us assume that some formula 9 is al ays true in all models of theory T, yet it cannot be pro#ed in T& +et us consider the theory TB in the lan!ua!e of T hich contains 0besides the axioms of T1 an additional non-lo!ical axiom E the ne!ation of 9, i&e& the formula fx1&&&xn9, here x1, &&, xn are exactly all the free #ariables of 9 0if 9 contains free #ariables x1, &&, xn, then, to ne!ate its assertion, e must add to 9 the Quantifiers x1&&&xn1& 'ince 9 cannot be deri#ed from the axioms of T, 9J is a consistent theory. 7ndeed, if TB ould be inconsistent, i&e& e could pro#e in TB some formula < and its ne!ation f<, then e had proofs of VTW: fx1&&&xn9 U- <, and VTW: fx1&&&xn9 U- f<& 'ince fx1&&&xn9 is a closed formula, by Deduction Theorem 2, VTW: U- fx1&&&xn9 `<, and VTW: U- fx1&&&xn9 `f<& ?o , by axiom +,: 0G`<1`0G`f<1`fG, e obtain that VTW: U- ffx1&&&xn9& Gy the 0classical1 Double ?e!ation +a , this implies VTW: U- x1&&&xn9, and by axiom +12: xG0x1`G0x1 E VTW: U- 9& Gut, by our assumption, 9 cannot be pro#ed in T& Hence, TB is a consistent theory& ?o , by the 5odel @xistence Theorem, there is a model of TB, i&e& an interpretation H that makes all its axioms al ays true& *nder this interpretation, ill pro#e it later in this

1.. all axioms of T are al ays true, i&e& H is a model of T& 6nd the formula fx1&&&xn9 0as an axiom of TB1 also is true under H& Rn the other hand, since 9 is al ays true in all models of T, it is al ays true also under the interpretation H& Hence, formulas x1&&&xn9 and fx1&&&xn9 both are al ays true under H& This is impossible, hence, 9 must be pro#able in T& D&@&D& 1& 'uch a simple proof seems almost impossible; 2e are pro#in! that the lo!ical axioms and rules of inference are stron! enou!h, but here come these axioms in8 They come in E in the proof of the 5odel @xistence Theorem& This theorem says that if some formal theory T does not ha#e models, then the lo!ical axioms and rules of inference are stron! enou!h to deri#e a contradiction from the axioms of T& Gut the proof of the 5odel @xistence Theorem that e ill consider belo , is positi#e, not ne!ati#e; 2& The abo#e simple proof seems to be extremely non-constructi#e; S7f 9 is al ays true in all models of T, then it can be pro#ed in TS& Ho could e obtain this proof8 'till, ho do e kno that 9 is true in all models of T8 Rnly, if e had a constructi#e procedure that is #erifyin! this, e could ask for an al!orithm con#ertin! such procedures into proofs in T; Proof of $enkinJs Model !7istence 9heorem !7ercise *.).) 0for smart students1& %ro#e HenkinBs 5odel @xistence Theorem by usin! the follo in! smart ideas due to +& Henkin and A& Hasen>l!er& +et T be a consistent theory& 2e must build a model of T& 2hat kind of SbricksS should e use for this Sbuildin!S8 Idea V'= let us use ob>ect constants of the lan!ua!e; 'o, let us add to the lan!ua!e of T an infinite set of ne ob>ect constants d1, d2, d(, &&& 0and adopt the correspondin! additional instances of lo!ical axioms1& %ro#e that this extended theory T0 is consistent& The model e are buildin! must contain all Sob>ectsS hose existence can be pro#ed in T0& Idea V.= for each formula 9 of T0 ha#in! exactly one free #ariable 0for example, x1 let us add to the theory T 0 the axiom x90x1`90di1, here the constant di is uniQue for each 9& 7f T0 pro#es x90x1, then this constant di ill represent in our model the Sob>ectS x ha#in! the property 9& %ro#e that this extended theory T1 is consistent& 7dea [(: pro#e the 0non-constructi#e1 +indenbaumBs lemma: the axiom set of any consistent theory can be extended in such a ay, that the extended theory is consistent and complete 0the axiom set of the extended theory may be not al!orithmically sol#able1& 6fter this, extend T1 to a consistent complete theory T2& Idea V*= let us take as the domain of the interpretation 5 the set of all those terms of T 0 that do not contain #ariables& 6nd let us interpret each function constant f as the Ssyntactic

1./ constructor functionS fB, i&e& let us define the #alue fB0t 1, &&&, tn1 simply as the character strin! Sf0t1, &&&, tn1S& 9inally, let us interpret each predicate constant p as the relation pB such that pB0t1, &&&, tn1 is true in 5, if and only if T 2 pro#es pB0t1, &&&, tn1& To complete the proof, pro#e that an arbitrary formula A is al ays true in 5, if and only if T2 pro#es A& Hence, all theorems of the initial theory T are al ays true in 5&
6dolf +indenbaum 01,0.-1,.11, his ife Hanina Hosiasson-+indenbaum 01C,,-1,.11&

LindenbaumJs Lemma& 6ny consistent first order theory can be extended to a consistent complete theory& 5ore precisely, if T is a consistent first order theory, then, in the lan!ua!e of T, there is a set 6 of closed formulas such that T\6 is a consistent complete theory& 07n !eneral, T\6 is not a formal theory in the sense of 'ection 1&1, see belo &1 4ote& Gy T\6 e denote the first order theory in the lan!ua!e of T, obtained from T by addin! the formulas of the set 6 as non-lo!ical axioms& !7ercise *.).*. Verify that, in any predicate lan!ua!e +, only countably many formulas can be !enerated& 7&e& produce an al!orithm for printin! out a seQuence 90, 91, 92, &&& containin! all the formulas of +& Proof of the LindenbaumJs Lemma 03ttention= nonBconstructive reasoning6D +et us use the al!orithm of the @xercise .&(&. printin! out the seQuence 9 0, 91, 92, &&& of all formulas in the lan!ua!e of T, and let us run throu!h this seQuence, processin! only those formulas 9i that are closed& 6t the #ery be!innin!, the set of ne axioms 60 is empty& 6t the step i, e already ha#e some set 6i-1 of ne axioms& 7f the formula 9i is not closed, let us i!nore it, and set 6 i^6i-1& ?o , let us suppose that 9i is a closed formula& 7f T\6i-1 pro#es 9i, or T\6i-1 pro#es f9i, then e can i!nore this formula, and set 6i^6i-1& 7f T\6 does not pro#e neither 9 i, nor f9i, then let us simply add 9i 0or f9i, if you like it better1 to our set of ne axioms, i&e& set Ai= Ai1{ Fi } & @tc&, ad infinitum& 6s the result of this process formulas A= A0 A1 A2&&& Ai &&& & e obtain a set of closed

+et us pro#e that T\6 is a consistent complete theory& onsistency. 7f T\6 ould be inconsistent, e ould ha#e a proof of VT\6W UC 7C for some formula <& 7f, in this proof, no axioms from the set 6

1.ould be used, inconsistent& e ould ha#e a proof of VTW U- C 7C , i&e& T ould be

Rther ise, the proof of VT\6W U- C 7C could contain a finite number of axioms G1, &&&, Gk from the set 6& +et us arran!e these axioms in the seQuence, as e added them to the set 6& Thus e ha#e a proof of VTW: G 1, &&&, Gk UC 7C & +et us recall Theorem 2&.&10a1: 7f 61, 62, &&&, 6n, G U- C 7C , then 61, 62, &&&, 6n U- fG& Hence, e ha#e a proof of VTW: G 1, &&&, Gk-1 U- fGk& Gut this is impossible E e added Gk to the set 6 >ust because T\6i-1 could not pro#e neither Gk, nor fGk& D&@&D& om"leteness. 2e must #erify that, for any closed formula 9 in the lan!ua!e of T, either T\6 U- 9, or T\6 U- f9& +et us assume, this is not the case for some closed formula 9& Rf course, 9 appears in the abo#e seQuence 9 0, 91, 92, &&& as some 9i& 7f neither T\6 U- 9, nor T\6 U- f9, then neither T\6 i-1 U- 9i, nor T\6i1 U- f9i& 7n such a situation e ould add 9 to the set 6, hence, e ould ha#e T\6 U- 9& D&@&D& This completes the proof of +indenbaumBs +emma& 3ttention= nonBconstructive reasoning6 T\6 is a some hat stran!e theory, because, in !eneral, e do not ha#e an al!orithmical decision procedure for its axiom set& 7ndeed, to decide, is some closed formula 9 an axiom of T\6, or not, e must identify 9 in the seQuence 9 0, 91, 92, &&& as some 9i, and after this, e must #erify, hether T\6i-1 pro#es 9i, or T\6i-1 pro#es f9i, or none of these& Thus, in !eneral, T\6 is not a formal theory in the sense of ion 1&1& Proof of the Model !7istence 9heorem 03ttention= nonBconstructive reasoning6D 7nspired by the beautiful exposition in 5endelson V1,,=W& ,te" '. 2e must build a model of T& 2hat kind of SbricksS should e use for this Sbuildin!S8 7dea [1: let us use ob>ect constants of the lan!ua!e; 'o, in order to prepare enou!h SbricksS, let us add to the lan!ua!e of T a countable set of ne ob>ect constants d1, d2, d(, &&& 0and extend the definitions of terms, atomic formulas and formulas accordin!ly, and add ne instances of lo!ical axioms accordin!ly1& +et us pro#e that, if T is consistent, then this extended theory T0 also is consistent& 7f T0 ould be inconsistent, then, for some formula <, e could obtain a proof of VT0W: U- C 7C & 7f, in this proof, ob>ect constants from the set Xd1, d2, d(,

1.= &&&Y ould not appear at all, then, in fact, e had a proof of VTW: U- C 7C , i&e& e could conclude that T is inconsistent& Gut, if the ne ob>ect constants do appear in the proof of VT0W: U- C 7C 8 Then, let us replace these constants by any #ariables of T that do not appear in this proof 0this is possible, since each predicate lan!ua!e contains a countable set of ob>ect #ariables1& 6fter these substitutions, the proof becomes a #alid proofs of T, because: a1 The lo!ical axioms remain #alid& b1 The non-lo!ical axioms of T do not contain the ob>ect constants d1, d2, d(, &&&, i&e& they do not chan!e& c1 6pplications of inference rules 5% and Aen remain #alid& Hence, VTW: U- C G 7C G , here the formula <B has been obtained from < by the abo#e substitutions& 7&e&, if T0 ould be inconsistent, then T also ould be inconsistent& ,te" .. The model e are buildin! must contain all Sob>ectsS hose existence can be pro#ed in T0& 7dea [2: for each formula 9 of T 0 ha#in! exactly one free #ariable 0for example, x1 let us add to the theory T 0 the axiom x90x1`90di1, here the constant di is uniQue for each 9& 7f T0 pro#es x90x1, then this di ill represent in our model the Sob>ectS x ha#in! the property 9& +et us pro#e that, if T is consistent, then this extended theory T1 also is consistent& ?ote that in T1 the same lan!ua!e is used as in T0& To implement the 7dea [2 correctly, first let us use the al!orithm of the @xercise .&(&. printin! out the seQuence 9 0, 91, 92, &&& of all formulas in the lan!ua!e of T0, and let us run throu!h this seQuence, processin! only those formulas 9i that ha#e exactly one free #ariable& +et us assi!n to each such formula 9i a uniQue constant dc0i1 in such a ay that dc0i1 does not appear neither in the non-lo!ical axioms of T, nor in 9 i, nor in the axioms y9>0y1`9>0dc0>11 for all formulas 9> precedin! 9i in the seQuence 90, 91, 92, &&&& 6nd, if x is the 0only1 free #ariable of 9 i, let us adopt x9i0x1`9>0dc0i11 as an axiom of T1& ?o , let us assume that the extended theory T1 is inconsistent, i&e& that, for some formula < of T0, e ha#e a proof of VT 1W: U- C 7C & 7n these proofs, only a finite number n of axioms x90x1`90dc0911 could be used& 7f n^0, then e ha#e VT0W: U- C 7C , i&e& then T0 is inconsistent&

1.C 7f nb0, then let us mark the axiom x90x1`90d c0911 ith 9 ha#in! the lar!est index in the seQuence 90, 91, 92, &&&& 6nd, in the proof of VT1W: U- C 7C , let us replace the constant c091 by some #ariable y that does not appear in this proof 0this is possible, since each predicate lan!ua!e contains a countable set of #ariables1& 6fter this substitution, the proof remain a #alid proof of T 1, because: a1 The lo!ical axioms remain #alid& b1 The non-lo!ical axioms of T do not contain the constant c091, i&e& they do not chan!e& c1 The axiom x90x1`90dc0911 becomes x90x1`90y1& 'ince 9 does not contain the constant c091, the premise x90x1 does not chan!e& d1 The remainin! n-1 axioms y9>0y1`9>0dc0>11 of T1 do not contain the constant c091, i&e& they do not chan!e& e1 6pplications of inference rules 5% and Aen remain #alid& Thus e ha#e no another proof of a contradiction E VT 1W: U- C G 7C G , here the formula <B has been obtained from < by substitutin! y for c091& +et us recall Theorem 2&.&10a1: 7f 61, 62, &&&, 6n, G U- C 7C , then 61, 62, &&&, 6n U- fG& +et us take the formula x90x1`90y1 for G, and <B- for <& Thus, there is a proof of f0x90x1`90y11, here only lo!ical axioms, non-lo!ical axioms of T, and the remainin! n-1 axioms y9>0y1`9>0dc0>11 of T1 are used& +et us recall the @xercise 2&-&(0b1 V+1-+11, 5%W: U- 7 ( A 5 B ) 6 A7 B & Thus, f0x90x1`90y11 is eQui#alent to xF ( x )7 F ( y ) , and, in fact, e ha#e a proof of x90x1, and a proof of f90y1& Gy applyin! Aen to the second formula, e obtain a proof of yf90y1, hich is eQui#alent to fy90y1 0indeed, let us recall 'ection (&2, Table (&2, Aroup 7V, constructi#ely, U- xfGafxG1& Gy )eplacement Theorem (, fy90y1 is eQui#alent to fx90x1& Thus, e ha#e a proof of a contradiction xF ( x )7 xF ( x ) , here only lo!ical axioms, non-lo!ical axioms of T, and the remainin! n-1 axioms y9>0y1`9>0dc0>11 of T1 are used& +et us repeat the abo#e chain of reasonin! another n-1 times to eliminate all occurrences of the axioms x90x1`90dc0911 from our proof of a contradiction& 7n this ay e obtain a proof of a contradiction in T0, hich is impossible 0see 'tep 11& Hence, T1 is a consistent theory& ,te" ). 7dea [(: let us use the 0non-constructi#e;1 +indenbaumBs lemma, and

1., extend T1 to a consistent complete theory T2& ?ote that in T2 the same lan!ua!e is used as in T0& ,te" *. +et us define an interpretation 5 of the lan!ua!e of T 0, in hich all theorems of T2 ill be al ays true& 'ince all theorems of the initial theory T are theorems of T2, this ill complete our proof& 7dea [.: let us take as the domain D5 of the interpretation 5 the 0countable; E #erify;1 set of all constant terms of T0, i&e& terms that do not contain #ariables 0this set of terms is not empty, it contains at least the countable set of ob>ect constants added in 'tep 11& 6nd let us define interpretations of ob>ect constants, function constants and predicate constants as follo s& a1 The interpretation of each ob>ect constant c is the constant c itself& b1 The interpretation of a function constant f is the Ssyntactic constructor functionS fB, i&e&, if f is an n-ary function constant, and t 1, &&&, tn are constant terms, then the #alue fB0t1, &&&, tn1 is defined simply as the character strin! Sf0t1, &&&, tn1S 0Quotation marks i!nored1& c1 The interpretation of a predicate constant p is the relation pB such, if p is an n-ary predicate constant, and t1, &&&, tn are constant terms, then pB0t1, &&&, tn1 is defined as true in 5, if and only if T 2 pro#es p0t1, &&&, tn1 0note that T2 is a consistent complete theory, i&e& it pro#es either p0t 1, &&&, tn1, or fp0t1, &&&, tn1, but not both;1& ,te" -& To complete the proof, e must pro#e that, in the lan!ua!e of T 0, an arbitrary formula A is al ays true in 5, if and only if T 2 pro#es A 0let us denote this, as usual, by T2 U- A& This ill be pro#ed, if e ill pro#e that, if x1, &&&, xm is the set of at least all free #ariables contained in the formula A, and t1, &&&, tm are constant terms, then A0t1, &&&, tm1 is true in 5, if and only if T 2 UA0t1, &&&, tm1& The proof ill be by induction& Induction base: A is an atomic formula p0s1, &&&, sn1, here p is a predicate constant, and the terms s1, &&&, sn contain some of the #ariables x 1, &&&, xm& 7n s1, &&&, sn, let us substitute for x1, &&&, xm the terms t1, &&&, tm respecti#ely& 7n this ay e obtain constant terms sB1, &&&, sBn& Thus A0t1, &&&, tm1 is simply p0sB1, &&&, sBn1& Gy definition 0see 'tep .1, p0sB1, &&&, sBn1 is true, if and only if T2 U- p0sB1, &&&, sBn1, i&e&, if and only if T2 U- A0t1, &&&, tm1& D&@&D&

1/0 Induction ste". 4ote. 'ince, T2 is a complete consistent theory, for any closed formula 9, T 2 pro#es either 9, or f9 0but not both1& Hence, if e kno that 9 is true in 5, if and only if T2 U- 9, then e can conclude that 9 is false in 5, if and only if T 2 U- f9& 7ndeed, if 9 is false, then 9 is not true, i&e& T 2 does not pro#e 9, i&e& T2 Uf9& 6nd, if T2 U- f9, then T2 does not pro#e 9, i&e& 9 is not true, i&e& 9 is false& 6nd con#ersely: if e kno that 9 is false in 5, if and only if T 2 U- f9, then e can conclude that 9 is true in 5, if and only if T2 U- 9& 7ndeed, if 9 is true, then f9 is not true, i&e& T2 does not pro#e f9, i&e& T2 U- 9& 6nd, if T2 U- 9, then T2 does not pro#e f9, i&e& 9 is not false, i&e& 9 is true& ase '= A is fH& Then, accordin! to the classical truth tables, A0t1, &&&, tm1 is true in 5, if and only if H0t 1, &&&, tm1 is false in 5& Gy the induction assumption, H0t1, &&&, tm1 is true in 5, if and only if T 2 U- H0t1, &&&, tm1& Then, by the abo#e note, since H0t1, &&&, tm1 is a closed formula, H0t1, &&&, tm1 is false in 5, if and only if T2 U- fH0t1, &&&, tm1, i&e&, if and only if T2 U- A0t1, &&&, tm1& D&@&D& ase .= A is H`$& Then, accordin! to the classical truth tables, A0t 1, &&&, tm1 is false in 5, if and only if H0t 1, &&&, tm1 is true in 5, and $0t1, &&&, tm1 is false in 5& Gy the induction assumption, H0t 1, &&&, tm1 is true in 5, if and only if T 2 UH0t1, &&&, tm1, and $0t1, &&&, tm1 is true in 5, if and only if T 2 U- $0t1, &&&, tm1& Gy the abo#e note, $0t1, &&&, tm1 is false in 5, if and only if T 2 U- f$0t1, &&&, tm1& Hence, A0t1, &&&, tm1 is false in 5, if and only if T2 U- H0t1, &&&, tm1, and T2 U- f$0t1, &&&, tm1, or, A0t1, &&&, tm1 is true in 5, if and only if not 0T2 U- H0t1, &&&, tm1, and T2 U- f$0t1, &&&, tm11& +et us recall Theorem 2&2&1 and @xercise 2&-&(0a1, V+ 1-+11, 5%W: U( A 5 B ) 6 7 ( A7 B ) & 7n T2, all the axioms of the classical lo!ic are adopted, hence 0#erify;1, A0t1, &&&, tm1 is true in 5, if and only if T2 U- H0t1, &&&, tm1`$0t1, &&&, tm1,

1/1 or, A0t1, &&&, tm1 is true in 5, if and only if T2 U- A0t1, &&&, tm1& D&@&D& ase )= A is H C & Then, accordin! to the classical truth tables, A0t 1, &&&, tm1 is true in 5, if and only if H0t1, &&&, tm1 is true in 5, and $0t1, &&&, tm1 is true in 5& Gy the induction assumption, H0t1, &&&, tm1 is true in 5, if and only if T 2 U- H0t1, &&&, tm1, and $0t1, &&&, tm1 is true in 5, if and only if T2 U- $0t1, &&&, tm1& +et us recall Theorem 2&2&1& 7n T 2, all the axioms of the classical lo!ic are adopted, hence 0#erify;1, A0t1, &&&, tm1 is true in 5, if and only if T2 U- H ( t 1 , &&& , t m) C ( t 1 , &&& ,t m ) , or, A0t1, &&&, tm1 is true in 5, if and only if T2 U- A0t1, &&&, tm1& D&@&D& ase *= A is H C & Then, accordin! to the classical truth tables, A0t 1, &&&, tm1 is false in 5, if and only if H0t 1, &&&, tm1 is false in 5, and $0t 1, &&&, tm1 is false in 5& Gy the induction assumption, and by the abo#e note, H0t 1, &&&, tm1 is false in 5, if and only if T 2 U- fH0t1, &&&, tm1, and $0t1, &&&, tm1 is false in 5, if and only if T2 U- f$0t1, &&&, tm1& +et us recall Theorem 2&2&1 and Theorem 2&.&100b1: V+1-+10, 5%W U- 7 ( A B ) 6 7 A7 B 0the so-called 'econd de 5or!an +a 1& 7n T2, all the axioms of the classical lo!ic are adopted, hence 0#erify;1, A0t1, &&&, tm1 is false in 5, if and only if T2 U- 7 ( H ( t 1 , &&& , t m) C ( t 1, &&& , t m )) , or, A0t1, &&&, tm1 is false in 5, if and only if T 2 U- fA0t1, &&&, tm1& Thus, by the abo#e note, A0t1, &&&, tm1 is true in 5, if and only if T2 U- A0t1, &&&, tm1& D&@&D& ase -= A is xH& Then, by definition, A0t 1, &&&, tm1 is true in 5, if and only if H0x, t1, &&&, tm1 is Strue for some xS, i&e&, if and only if H0t, t 1, &&&, tm1 is true in 5 for some constant term t in 5& Gy the induction assumption, H0t, t 1, &&&, tm1 is true in 5, if and only if T2 U- H0t, t1, &&&, tm1& +et us recall our abo#e 'tep 2& 'ince H0x, t1, &&&, tm1 is a formula containin! exactly one free #ariable, in T2 e ha#e an axiom xH0x, t1, &&&, tm1`H0cH, t1, &&&, tm1, here cH is an ob>ect constant&

1/2 9irst, let us assume that A0t1, &&&, tm1 is true in 5& Then H0t, t1, &&&, tm1 is true in 5 for some constant term t in 5, hence, T 2 U- H0t, t1, &&&, tm1 for this particular t& +et us recall the axiom + 1(: 90t1`x90x1& 'ince t is a constant term, this axiom is #alid for t& 2e need the follo in! instance of + 1(: H0t, t1, &&&, tm1`xH0x, t1, &&&, tm1& 7n T2, all the axioms of the classical lo!ic are adopted, hence, T2 U- H0t, t1, &&&, tm1`xH0x, t1, &&&, tm1, and, by 5%, T2 U- xH0x, t1, &&&, tm1, i&e& T2 U- A0t1, &&&, tm1& D&@&D& ?o , let us assume that T2 U- A0t1, &&&, tm1, i&e& T2 U- xH0x, t1, &&&, tm1& Gy the abo#e-mentioned axiom, T2 U- xH0x, t1, &&&, tm1`H0cH, t1, &&&, tm1, here cH is an ob>ect constant& Thus, by 5%, T 2 U- H0cH, t1, &&&, tm1& 'ince cH is a constant term, by the induction assumption, if T 2 U- H0cH, t1, &&&, tm1, then H0cH, t1, &&&, tm1 is true in 5& Hence, H0cH, t1, &&&, tm1 is true in 5, i&e& H0x, t1, &&&, tm1 is true Sfor some xS, i&e& xH0x, t1, &&&, tm1 is true in 5, i&e& A0t1, &&&, tm1 is true in 5& D&@&D& ase /= A is xH& Gy definition, A0t1, &&&, tm1 is true in 5, if and only if H0x, t1, &&&, tm1 is Strue for all xS, i&e&, if and only if H0t, t 1, &&&, tm1 is true in 5 for all constant terms t in 5& Gy the induction assumption, H0t, t 1, &&&, tm1 is true in 5, if and only if T2 U- H0t, t1, &&&, tm1& +et us pro#e that A0t1, &&&, tm1 is false in 5, if and only if T2 U- xfH0x, t1, &&&, tm1& 9irst, let us assume that A0t1, &&&, tm1 is false in 5& Then H0t, t 1, &&&, tm1 is false in 5 for some constant term t in 5& Gy the induction assumption, and by the abo#e note, T2 U- fH0t, t1, &&&, tm1& 6s in the <ase /, let us recall the axiom +1(: fH0t, t1, &&&, tm1`xfH0x, t1, &&&, tm1& 7n T2, all the axioms of the classical lo!ic are adopted, hence, by 5%, T2 U- xfH0x, t1, &&&, tm1& ?o , let us assume that T2 U- xfH0x, t1, &&&, tm1& 6s in the <ase /, let us recall the axiom xfH0x, t1, &&&, tm1`fH0cfH, t1, &&&, tm1, here cfH is an ob>ect constant& Hence, by 5%, T2 U- fH0cfH, t1, &&&, tm1, i&e& T2 does not pro#e H0cfH, t1, &&&, tm1& Then, by the induction assumption, H0c fH, t1, &&&, tm1 is false in 5, i&e& xH0x, t1, &&&, tm1 is false in 5, i&e A0t1, &&&, tm1 is false in 5& Thus, e kno that A0t1, &&&, tm1 is true in 5, if and only if T 2 does not pro#e

1/( xfH0x, t1, &&&, tm1& 'ince T2 is a complete theory, A0t1, &&&, tm1 is true in 5, if and only if T2 U- fxfH0x, t1, &&&, tm1& ?o , let us recall from 'ection (&2, Table (&2, Aroup 7, V+1-+11, +12-+1/, 5%, AenW: U- fxfGaxG& 7n T 2, all the axioms of the classical lo!ic are adopted, hence, T 2 U- fxfH0x, t1, &&&, tm1, if and only if T2 U- xH0x, t1, &&&, tm1, i&e& A0t1, &&&, tm1 is true in 5, if and only if T2 U- A0t1, &&&, tm1& D&@&D& This completes the proof of the 5odel @xistence Theorem& D&@&D& 3ttention= nonBconstructive reasoning6 The abo#e construction of the model 5 seems to be Salmost constructi#eS& The domain D 5 consists of all constant terms from the lan!ua!e of T0& The axiom set of T1 is al!orithmically sol#able 0#erify;1& The interpretations of function constants are computable functions 0#erify;1& Gut the interpretations of predicate constants8 2e interpreted each predicate constant p as the relation pB such that pB0t 1, &&&, tn1 is true, if and only if T2 pro#es p0t1, &&&, tn1& This relation ould be, in !eneral, not al!orithmically sol#able, e#en if the axiom set of T2 ould be sol#able; Gut, in !eneral, the axiom set of theory T2 0obtained by means of +indenbaumBs +emma1 is not al!orithmically sol#able; Thus, our construction of the model 5 is essentially non-constructi#e& !7ercise *.).- 0for smart students1& Determine the Sde!ree of nonconstructi#enessS of the 5odel @xistence Theorem& 0Hint: pro#e that it is m2, i&e& that all the necessary functions are Scomputable in the limitS& 6 function f0x1 is called computable in the limit, if and only if there is a computable g ( x , y) 1 function !0x,y1 such that, for all x, f ( x )=lim y onse>uences of 5MdelJs om"leteness 9heorem 4otion of Logical onse>uence 6s noted abo#e 0@xercise .&1&-1, some formula A is a Slo!ical conseQuenceS of the formulas 61, &&&, 6n, if and only if the formula 6 1, &&&, 6n`A is lo!ically #alid, hence, by AFdelBs <ompleteness Theorem 3 if and only if, 5 can be derived from 3', ..., 3n by using the a7ioms and rules of inference of the classical logic. This completes the formali:ation of the some hat mystical notion of Slo!ical conseQuenceS, and allo s to consider reasonin! as a process that could be performed by usin! computers 0see belo 1&

1/.

onsistency and ,atisfiability 6 set of formulas 91, &&&, 9n is called inconsistent, if and only if a contradiction 0i&e& a formula B7 B 1 can be deri#ed from it& 9or example, the set XG, G`<, <`fGY is inconsistent 0#erify1& The 5odel @xistence Theorem allo s to connect the notions of consistency and satisfiability& !7ercise *.)./. Verify, that a set of formulas in a predicate lan!ua!e: a1 is consistent in the classical lo!ic, if and only if it is satisfiable, b1 is inconsistent in the classical lo!ic, if and only if it is unsatisfiable& 0Hint: use the result of @xercise .&1&11& om"utational om"le7ity of the Problem orollary *.).*. 7n any "redicate lan!ua!e the set of all lo!ically #alid formulas is al!orithmically enumerable& 7&e& !i#en a lan!ua!e +, e can rite an al!orithm that 0 orkin! a+ infinit$m1 prints out all the lo!ically #alid formulas of + 0and only these formulas1& Proof. 7mmediately from @xercise 1&1&. and AFdelBs <ompleteness Theorem& This makes AFdelBs <ompleteness Theorem #ery si!nificant: it sho s that the Sdoubly non-constructi#eS notion of lo!ically #alid formula is at least /0T constructi#e 3 semiBconstructive; 'emi-feasible for computers; 'till, unfortunately, this notion appears to be not 100T constructi#e& 7n 1,(-, 6lon:o <hurch pro#ed that at least some predicate lan!ua!es do not allo an al!orithm determinin!, is a !i#en formula lo!ically #alid or not 0i&e& an al!orithm sol#in! the famous Entschei+$ngsprob%em 3 the decision problem1:
3. hurch. 6 note on the @ntscheidun!sproblem& SHournal of 'ymb& +o!icS, 1,(-, #ol&1, pp&.0-.1

6fter this, +as:lo $almar in


L. Kalmar. Die Lurueckfuehrun! des @ntscheidun!sproblems auf den 9all #on 9ormeln mit einer ein:i!en, binlren 9unktions#ariablen& S<ompositio 5ath&S, 1,(-, Vol&., pp&1(=-1..

impro#ed <hurchBs result: hurchBKalmar 9heorem. 7f a predicate lan!ua!e contains at least one "redicate constant that is at least binary, then this lan!ua!e does not allo an al!orithm determinin!, is a !i#en formula of this lan!ua!e lo!ically #alid or not& Thus, none of serious predicate lan!ua!es allo s such an al!orithm 0lan!ua!es of %6 and L9 included1& 9or details, 5endelson V1,,=W&

1// 'ometimes, this fact 0the /0T constructi#eness of the notion of the lo!ical #alidity1 is expressed a follo s: the lo!ical #alidity of predicate formulas is semiBdecidable& orollary *.).-& 7f a predicate lan!ua!e contains at least one predicate constant that is at least binary, then this lan!ua!e does not allo an al!orithm determinin!, does some formula A of this lan!ua!e follo from some other formulas 61, &&&, 6n& 7n other ords 3 the task of reasoning in such a language is not algorithmically solvable. !7ercise *.).0. Verify, this& hurchBKalmar 9heorem and Kno8ledge %ases 7f e ill build our kno led!e base by usin! some predicate lan!ua!e, then, in !eneral, the situation ill be as follo s: a1 2e ill ha#e some set of constants re!istered in the kno led!e base: ob@ect constants: c1, c2, &&&, ck, function constants and "redicate constants: p1, p2, &&&, pm 0 ith ar!ument numbers specified1& VThe losed 1orld 3ssum"tion: in the orld, there exist only ob>ects denoted by our constants c1, c2, &&&, ck& 7n fact, this assumption should be represented as an axiom x 0x^c1 # x^c2 # &&& # x^ck1& The ?"en 1orld 3ssum"tion: in the orld, there exist more ob>ects than denoted by our ob>ect constants&W b1 :acts, conce"ts and rules 0Mla sN1 are stored in the set of formulas 91, 92, &&&, 9n& 9acts are represented ithout or ith ne!ation, that do not contain #ariables: fpi0c>1, c>2, &&&, c>s1& 9acts build up a kind of Mdatabase rules may ser#e as integrity con+itions& kno led!e base as a as atomic formulas, pi0c>1, c>2, &&&, c>s1, or tablesN& 'ome of the

VThe losed 1orld 3ssum"tion: the predicate pi0c>1, c>2, &&&, c>s1 is re!arded as true, if and only if the formula pi0c>1, c>2, &&&, c>s1 is stored in the kno led!e base& 7f there is no such formula in the kno led!e base, then p i0c>1, c>2, &&&, c>s1 is assumed to be false& The ?"en 1orld 3ssum"tion: the predicate pi0c>1, c>2, &&&, c>s1 is re!arded as true or false, if and only if, correspondin!ly, the formula pi0c>1, c>2, &&&, c>s1, or the formula fpi0c>1, c>2, &&&, c>s1 is stored in the kno led!e base&W c1 6 >uery is simply another formula 8A& 6ns erin! of such a Query means that the 4$ery processor of the kno led!e base must determine, does A 0or,

1/maybe, fA1 follo base& from the formulas 91, 92, &&&, 9n, stored in the kno led!e

7f A contains a free #ariable x, then the Query 8A0x1 means the follo in!: return all the ob>ect constants ci, for hich the formula A0ci1 follo s from 91, 92, &&&, 9n& Thus, to build the Query processor of our kno led!e base, e must use some al!orithm allo in! to determine 0as fast as possible1, !i#en the formulas 91, 92, &&&, 9n, A, does A follo from 91, 92, &&&, 9n, or not& +etBs call this task the reasoning task& 6ccordin! to AFdelBs <ompleteness Theorem, A follo s from 91, 92, &&&, 9n, if and only if V+1-+1/, 5%, AenW: 91, 92, &&&, 9n U- A, i&e& if A can be deri#ed from91, 92, &&&, 9n in the classical predicate lo!ic& This makes the reasonin! task at least semi-feasible for computers 0in the sense of <orollary .&(&.1& Ho e#er, orollary 0of the <hurch-$almar theorem, <orollary .&(&/1& 7f, hen buildin! a kno led!e base, e ill use the full po er of some predicate lan!ua!e 0containin! at least one predicate constant that is at least binary1, then the reasonin! task ill not be al!orithmically sol#able, and 3 for such a kno led!e base 3 8e 8ill fail to build a universal >uery "rocessor& Thus, to build a really usable kno led!e base, e must restrict someho our predicate lan!ua!e to make the reasonin! task sol#able& 9or a successful attempt to do this see the so-called description lo!ics& ,kolemJs Parado7 7nitially, the 5odel @xistence Theorem as pro#ed in a eaker form in 1,1/ 0by +eopold +F enheim1 and 1,1, 0by Thoralf 'kolem1: if a first order theory has a model, then it has a finite or countable model 0the famous LM8enheimB ,kolem theorem1& %roof 0after 1,.,1: if T has a model, then T is consistent, i&e& T has a finite or countable model&
L. LM8enheim. *eber 5F!lichkeiten im )elati#kalkuel& S5athematische 6nnalenS, 1,1/, Vol&=-, pp&..=-.=0& 9h. ,kolem. +o!isch-kombinatorische *ntersuchun!en gber die @rfgllbarkeit und Ge eisbarkeit mathematischen 'lt:e nebst einem Theoreme gber dichte 5en!en& Ji+ens&absa&a+emiet i Cristiania, S&rifter 7, ?o& ., 1,20, pp& 1-(-&

+F enheim-'kolem theorem 0and the 5odel @xistence theorem1 is steadily

1/= pro#okin! the so-called ,kolemJs Parado7, first noted by 'kolem in his address before the /th <on!ress of 'candina#ian 5athematicians 0Huly .-=, 1,221:
9h. ,kolem. @ini!e Gemerkun!en :ur axiomatischen Ge!ruendun! der 5en!enlehre& .atemati&er&ongressen i He%singfors +en K>L /$%i ;M99, ,en femte s&an+ina'is&a matemati&er&ongressen, )e+ogNre%se, 6kademiska Gokhandeln, Helsinki, 1,2(, pp& 21=-2(2&

'kolem called the effect Srelati#ity of set-theoretic notionsS& 7n all formal set theories 0for example, in L91 e can pro#e the existence of uncountable sets& 'till, accordin! to the 5odel @xistence theorem, if our formal set theory consistent, then there is a countable model here all its axioms and theorems are true& 7&e& a theory pro#es the existence of uncountable sets, yet it has a countable model; 7s this possible8 Does it mean that all formal set theories are inconsistent8 %latonists put it as follo s: any consistent axiomatic set theory has countable models, hence, no axiom system can represent our SintendedS set theory 0i&e& the %latonist S orld of setsS1 adeQuately& 9or a formalist, 'kolemBs %aradox is not a paradox at all& 7 ould rather call it 'kolemBs @ffect E like as the photo-effect, it is simply a striking "henomenon& 7ndeed, let H be a countable model of our formal set theory& 7n this theory, e can pro#e that the set r of all real numbers is uncountable, i&e& ff 0f is 1-1 function from r into 1, 011 here is the set of all natural numbers& 2hat is the meanin! of this theorem in the countable model H8 7nterpretations of r and are subsets of the domain DH, i&e& they both are countable sets, i&e& f 0f is 1-1 function from rH into 7nterpretation of 011 in H is ff00 f ,/ 1 and 0f is 1-1 function from rH into
H11& H1&

021

Hence, the mappin! f of 021 does exist, yet it exists outside the model H; Do you think that f of 021 SmustS be located in the model8 2hy8 7f you are li#in! 0as an Sinternal obser#erS1 ithin the model H, the set rH seems uncountable to you 0because you cannot find a 1-1 function from r H into H in your orld H1& 'till, for me 0an Sexternal obser#erS1 your uncountable r H is countable E in my orld 7 ha#e a 1-1 function from rH into H; Hence, indeed, 'kolemBs %aradox represents simply a stri&ing phenomenon& 7t is orth of kno in!, yet there is no dan!er in it&
6dded 9ebruary ,, 200=& The inter-relationship of the <ompleteness Theorem and 5odel @xistence Theorem can be

1/C
represented in the follo in! #ery !eneral ay& +et us replace: - %redicate lan!ua!e + E by any set ' of SformulasS& - 9irst order theory T E by any SformulaS of ' 0assume, T contains only a finite number of axioms, and take the con>uction of them1& - The notion of interpretation E by any set 5 and a SpredicateS T0m, 91 0 here m is member of 5, and 9 E a formula of '1& 7f you ish, you may read T0m, 91 as Sm makes 9 trueS, i&e& m is a SmodelS of 9& - The notion of pro#ability in the classical lo!ic E by a SpredicateS %091 0 here 9 is a formula of '1& 7f you ish, you may read %091 as S9 is pro#able in the classical lo!icS& 6ssume, for a a set of SformulasS ', e ha#e any set 5 and any t o SpredicatesS T0m, 91 and %091 0 here m is a member of 5, and 9 E a formula of '1 such that only the follo in! simple principles hold: a1 9or all 9, F S 5 7 F S b1 9or all m . 0i&e& ' is closed under ne!ation1& and F S , T0m, 91 afT0m, f91&

c1 9or all F S , ff%0ff91 ` % 091& < om"leteness 9heorem<& 9or all 9, m T0m, 91 ` %091& <Model !7istence 9heorem<& 9or all 9, f%0f91 ` m T0m, 91& 9heorem. 7f a, b, c1 hold, then the abo#e StheoremsS are eQui#alent& Proof& 11 6ssume m T0m, 91 ` %091 for all 9& Then f%091 ` fm T0m, 91, and by a1 also, f%0f91 ` fm T0m, f91 ` mfT0m, f91 ` m T0m, 91 by b1& D&@&D& 21 6ssume f%0f91 ` m T0m, 91 for all 9&Then fm T0m, 91 ` ff%0f91, and by a1 also fm T0m, f91 ` ff%0ff91& Gy b1, m T0m, 91 ` m fT0m, f91 ` fm T0m, f91 ` ff%0ff91 ` %091 by c1& D&@&D&

*.*. onstructive Pro"ositional Logic E Kri"ke ,emantics


'aul 6aron $ripke 0born 1,.01& S6merican lo!ician and philosopher 'aul $ripke is one of todayBs leadin! thinkers on thou!ht and its manifold relations to the orld& His name is attached to ob>ects in se#eral fields of lo!ic from $ripke-%latek axioms in hi!her recursion theory to the SGrou er-$ripke schemeS in intuitionistic mathematics& $ripke models for modal lo!ic, a disco#ery he made in his teena!e years, became part of the standard #ocabulary of mathematical lo!icians after his first article appeared in 1,-(, hen he as >ust 2( years old& $ripke models and the results that depend upon them are cited today not only in

1/, philosophy and lo!ic, but also in lin!uistics and computer science&&&S 0The Aa:ette& The ne spaper of the Hohn Hopkins *ni#ersity, 5ay 12, 1,,=, Vol&2-, ? (.1 ,. Kri"ke 01,-(1& 'emantical <onsiderations on 5odal +o!ic, Acta Phi%osophica Fennica '/: C(-,.& ,. Kri"ke 01,-(1& 'emantical analysis of modal lo!ic& 7& ?ormal modal propositional calculi& F( .ath( Logi& Gr$n+%( .ath(, ,:-=-,-, 1,-(& ,. Kri"ke 01,-/1& 'emantical analysis of intuitionistic lo!ic& 7n: /( "( Cross%ey, .( A( E( ,$mmet e+s(!, Forma% systems an+ rec$rsi'e f$nctions & 6msterdam, ?orth Holland, 1,-/, pp&,2-12,& 6s usual, let us assume, the formula 9 has been built of SatomicS formulas G 1, G2, &&&, Gn by usin! propositional connecti#es only& 7nstead of simply computin! truth #alues of 9 from truth #alues of G 1, G2, &&&, Gn, $ripke proposed to consider the behavior of 9 hen the truth #alues of G 1, G2, &&&, Gn are chan!in! !radually from false to true accordin! to some SscenarioS& Thus, $ripke proposed to replace the classical semantics 0interpretation1 of the propositional connecti#es 0defined by the classical truth tables1 by a more complicated SdynamicS semantics& 7nstead of simply sayin! that f9 is true, iff 9 is false, let us say that, at some point in a scenario, f9 is true, if and only if, at this point, 9 is false and remains false, hen truth #alues of G1, G2, &&&, Gn are chan!in! accordin! to the scenario& +et o stand for implication, con>unction or dis>unction& 7nstead of simply sayin! that 9oA is true, if and only if, 9oA is true accordin! to the classical truth tables, let us say that, at some point in a scenario, 9oA is true, if and only if, at this point, it is true and remains true, hen truth #alues of G 1, G2, &&&, Gn are chan!in! accordin! to the scenario& !7am"le *.*.'. +et us consider the beha#ior of the classical axiom + 11: B7 B in the scenario, here, at first, G is false, and at the next step it becomes true: 0 -------------- 1 6t the startin! point, G is false, fG also is false 0here, for fG to be true, G must remain false at the next step, but it doesnBt1& This means that, at the startin! point, B7 B is false& 6t the next step: G is true, hence, fG is false, but, of course, B7 B is true& Thus, in $ripke scenarios, B7 B is not al ays true& 'urprisin!ly, some time later 0+emma .&.&(1, e ill deri#e from

1-0 this simple fact that G#fG cannot be pro#ed in the constructi#e lo!ic 0 e already kno this from 'ection 2&C1& !7am"le *.*... +et us consider the beha#ior of the 0only1 classically pro#able half of the 9irst de 5or!an +a : 7 ( A B ) 5 7 A7 B in the scenario, here, at first 6 and G are both false, and at the next step, t o branches appear in the scenario: in the first branch: 6 remains false, and G becomes true, in the second branch: 6 becomes true, and G remains false: U---01 00-U---------U---10 6t the startin! point: 6 is false, f6 3 also is false 0for f6 to be true, 6 must remain false at the next step, but in the second branch it doesnBt1& 'imilarly, at the startin! point: G is false, fG 3 also false 0for fG to be true, G must remain false at the next step, but in the first branch it doesnBt1& This means that, at the startin! point, 7 ( A B ) is true 0because A B is false, and it remains false in both branches1, and 7 A7 B is false, hence, 7 ( A B ) 5 7 A7 B is false& Thus, in $ripke scenarios, 7 ( A B ) 5 7 A7 B is not al ays true& 'urprisin!ly, some time later 0+emma .&.&(1, e ill deri#e from this simple fact that the this half of the 9irst de 5or!an +a cannot be pro#ed in the constructi#e lo!ic& 2e failed to do this in 'ection 2&C; !7ercise *.*.'. 7n#esti!ate, in appropriate scenarios, the beha#ior of the follo in! 0only1 classically pro#able formulas: 77 ( A B ) 5 77 A7 7 B , 06`G1`00f6`G1`G1, ( A 5 B )( B 5 A) , and #erify that, in $ripke scenarios, these formulas are not al ays true& 'ome time later 0+emma .&.&(1, e ill deri#e from this simple fact that these formulas cannot be pro#ed in the constructi#e lo!ic& 2e failed to do this in 'ection 2&C; 0Hint: try the most simple scenarios first: 00--01, 00-10, 00-11, etc&1 5ore precisely, the definition of the Kri"ke semantics for the propositional lo!ic is as follo s& 6ssume, the formula 9 has been built of SatomicS formulas G1, G2, &&&, Gn by usin! propositional connecti#es only& 7nstead of considerin! truth #alues of 9 for all the possible tuples of truth #alues of G 1, G2, &&&, Gn, let us consider the beha#ior of 9 in all the possible $ripke scenarios, defined as follo s& Definition of Kri"ke scenarios& @ach scenario s is a triple 0b, c, t1 of the follo in! ob>ects& 9irst, b is a finite set of ob>ects called nodes 0or, states1&

1-1 The second member c is a partial orderin! relationship bet een the nodes, i&e& for all x , y , 3 b : xcy ` 0yc: ` xc:1 0transiti#ity1& The third member t of the tripple is a function 0t means StrueS1& 7t associates ith each node x a S!ro in!S set t0x1 of SatomicS formulas, i&e& a subset of XG1, G2, &&&, GnY in such a ay that for all x , y b : x y 5 t ( x ) t ( y ) & 4ote. 7n some other textbooks, $ripke scenarios are called Kri"ke models, or Kri"ke structures& +et us say that Gi is true at the node x, if and only if G i is in the set t0x1& 2e ill denote this fact by x U^ G i 0Sat x, Gi is trueS, or Sx forces GiS1& 'ince t is monotonic, if x U^ Gi , then y U^ Gi for all y after x, i&e& for all y b such that ydx& 7&e& if Gi is true at some node x, then Gi remains true at all nodes after x& +et us define x U^ 9 0S9 is true at xS, or Sx forces 9S1 for any formula 9 that has been built of SatomicS formulas G 1, G2, &&&, Gn by usin! propositional connecti#es only& 1& ?e!ation& 'uppose, the truth #alue of x U^ 9 is already defined for all x b & Then x U^ f9 is defined to be true, if and only if, for all y O x b , y U^ 9 is false 0i&e& f0y U^ 91 is true accordin! to the classical truth table of the ne!ation connecti#e1& 2& 7mplication, con>unction or dis>unction& 'uppose, the truth #alues of x U^ 9 and x U^ A are already defined for all x b & Then x U^ 9oA is defined to be true, if and only if, for all y O x b , 0y U^ 91o0y U^ A1 is true accordin! to the classical truth table of the implication, con>unction or dis>unction connecti#e respecti#ely& Lemma *.*.'. 9or any formula 9, any $ripke scenario 0b, c, t1, and any node x b : if x U^ 9, then y U^ 9 for all y b such that ydx& 7&e& if, in a $ripke scenario, a formula becomes true at some node, then it remains true fore#er after this node& Proof. Gy induction& Induction base& 'ee abo#e: if x U^ Gi , then y U^ Gi for all y after x, i&e& for all y b such that ydx& Induction ste". '. 4egation. 6ssume, x U^ f9, i&e&, accordin! to the classical truth table, not y U ^ 9 for all y O x b & 7f ydx, then is y U^ f9 true or false8 Gy definition, y U^ f9 ould be true, if and only if not : U^ 9 for all 3 O y b & Gy transiti#ity of c, if :dy and ydx, then :dx& Thus, by our assumption, if :dy, then not : U^ 9& Hence, y U^ f9& D&@&D&

1-2 .. Im"lication, con@unction or dis@unction. 6ssume, x U^ 9oA, i&e&, accordin! to the correspondin! classical truth table, 0y U^ 91o0y U^ A1 is true for all y O x b & 7f ydx, then is y U^ 9oA true or false8 Gy definition, y U^ 9oA ould be true, if and only if 0: U^91o0: U^ A1 ould be true for all 3 O y b & Gy transiti#ity of c, if :dy and ydx, then :dx& Thus, by our assumption, if :dx, then 0: U^ 91o0: U^ A1 is true& Hence, y U^ 9oA& D&@&D& !7ercise *.*... Verify that if x is a maximal node in a scenario 0b, c, t1, then x U^ 9, if and only if 9 is true at x accordin! to the classical truth tables& $ripke established that a formula is "rovable in the constructive "ro"ositional logic, if and only if it is true at all nodes in all Kri"ke scenarios. 9heorem *.*.. C,. Kri"ke, com"leteness of the constructive "ro"ositional logicD. 6 formula 9 is pro#able in the constructi#e propositional lo!ic 0i&e& V+ 1+10, 5%W U- 91, if and only if 9 is true at all nodes in all $ripke scenarios& 6s usual, the hard part of the proof is establishin! that Strue is pro#ableS, i&e& if 9 is true at all nodes in all $ripke scenarios, then V+ 1-+10, 5%W U- 9 0see <orollary .&.&= belo 1& The easy part of the proof is, as usual, the soundness lemma: Lemma *.*.). 7f V+1-+10, 5%W U- 9, then 9 is true at all nodes in all $ripke scenarios& This lemma ill follo from Lemma *.*.*. 7f 9 is any of the constructi#e axioms + 1-+10, then, for any $ripke scenario 0b, c, t1, and any node x b : x U^ 9& 7&e& the constructi#e axioms are true at all nodes in all $ripke scenarios& and Lemma *.*.-. 7f, in a $ripke scenario 0b, c, t1, at the node x b : x U^ 9 and x U^ 9`A, then x U^ A& Hence, if 9 and 9`A are true at all nodes in all $ripke scenarios, then so is A& Proof of Lemma *.*.)& 7ndeed, by +emma .&.&., all the constructi#e axioms +1-+10 are true at all nodes in all scenarios, and, by +emma .&.&/, the 5odus %onens rule preser#es the property of bein! Strue at all nodes in all scenariosS& D&@&D& 4ote& +et us return to the abo#e @xample .&.&2 and @xercise .&.&1& 2e established that formulas 7 ( A B ) 5 7 A 7 B 4 77 ( A B ) 5 77 A7 7 B 4

1-( 06`G1`00f6`G1`G1 are not true at all nodes in all scenarios& Hence, by +emma .&.&(, these formulas cannot be pro#ed in the constructi#e lo!ic V+ 1-+10, 5%W& 2e failed to pro#e this in 'ection 2&C; Proof of Lemma *.*.-. 2e kno that x U^ 9`A means that 0y U^ 91`0y U^ A1 is true 0accordin! to the classical truth table1 for all y O x b & Gy +emma .&.&1, e kno that y U^ 9 for all y O x b & Hence, if y U^ A ould be false, then 0y U^ 91`0y U^ A1 also ould be false& Hence, x U^ A& D&@&D& Proof of Lemma *.*.*. L'= %FC F%D x U^ G`0<`G1 is true, if and only if 0y U^ G1`0y U^ <`G1 is true for all ydx& x U^ G`0<`G1 is false, if and only if 0y U^ G1`0y U^ <`G1 is false for some ydx& Ho could 0y U^ G1`0y U^ <`G1 be false for some ydx8 6ccordin! to the classical implication truth table, this could be only, if and only if y U^ G is true, and y U^ <`G is false& y U^ <`G is true, if and only if 0: U^ <1`0: U^ G1 is true for all :dy& y U^ <`G is false, if and only if 0: U^ <1`0: U^ G1 is false for some :dy& Ho could 0: U^ <1`0: U^ G1 be false for some :dy8 6ccordin! to the classical implication truth table, this could be, if and only if : U^ < is true, and : U^ G is false& 'ummary: x U^ G`0<`G1 is false if and only if ydx 0y SW % is true and y U^ <`G is false1 if and only if :dy 0: U^ < is true and ; SW % is false1 Hence, if x U^ G`0<`G1 is false, then there are y and : such that: xcyc:, y SW % is true, : U^ < is true, and ; SW % is false& Gy +emma .&.&1, if yc: and y U^ G is true, then : U^ G is true& <ontradiction ith S: U^ G is falseS& Thus, x U^ G`0<`G1 is true& L'I= R%FC%F D x U^ fG`0G`<1 is false, if and only if 0y U^ fG1`0y U^ G`<1 is false for some ydx, i&e& if and only if y U^ fG is true, and y U^ G`< is false& y U^ fG is true, if and only if : U^G is false for all :dy&

1-. y U^ G`< is false, if and only if 0: U^ G1`0: U^ <1 is false for some :dy, i&e& if and only if : U^ G is true, and : U^ < is false& 'ummary: x U^ fG`0G`<1 is false if and only if ydx 0y U^ fG is true and y U^ G`< is false1 if and only if if and only if :dy 0; SW% is false1 :dy 0; SW % is true and : U^ < is false1 Hence, if x U^ fG`0G`<1 is false, then there is ydx such that: a1 :dy 0 ; SW% is false1, and b1 :dy 0; SW % is trueD. <ontradiction& Thus, x U^ fG`0G`<1 is true& L)= B C B x U^ BC 5 B is false if and only if ydx 0y U^ BC is true and y SW % is false1 if and only if :dy C; SW% is true and : U^ < is true1 Hence, there is y such that xcy and y U^ G is false& 9rom :dy C: U^G is true1 e obtain that y U^ G is true& <ontradiction& Thus, x U^ BC 5 C is true& L*=
B C C

'imilarly& L-= B (C BC ) x U^ B 5 ( C 5 BC ) is false if and only if ydx 0y SW% is true and y U^ C 5 BC is false1 if and only if :dy C; SW is true and ; SW B C is false1 Hence, there are y, : such that xcyc:, y U^ G is true, and : U^ < is true, and ; SW BC is false& Then, by +emma .&.&1, 0u U^ G is true1and0u U^ <1 for all ud:& 7&e& ; SW BC is true& <ontradiction& Thus, x U^ B 5 ( C 5 BC ) is true& L/= B BC x U^ B 5 BC is false if and only if ydx 0y SW% is true and y U^ BC is false1 if and only if

1-/ :dy 0; SW % is false and : U^ < is false1 Hence, there are y, : such that xcyc:, y U^ G is true, and : U^ G is false& Gy +emma .&.&1, this is a contradiction& Thus, x U^ B 5 BC is true& L0= C BC 'imilarly& LG= ( B D ) (( C D) ( BC D)) x U^ ( B 5 , ) 5 (( C 5 , ) 5 ( B C 5 ,)) is false if and only if ydx 0y SW%FD is true and y U^ ( C 5 ,) 5 ( BC 5 , ) is false1 if and only if :dy 0; SW FD is true and : U^ BC 5 , is false1 if and only if ud: 0u U^ BC is true and u SW D is false1 Hence, there are y, :, u such that xcyc:cu, y U^ G`D is true, : U^ <`D is true, and u U^ D is false& Gy +emma .&.&1, u U^ G`D is true, and u U^ <`D is true& Thus, if u U^ G ould be true, then u U^ D also ould be true& Hence, u U^ G is false& 'imilarly, u U^ < also is false& Hence, u U^ BC is false& Gut e kno that it is true& <ontradiction& Thus, x U^ +C is true& L.= C%FC FDDDFCC%F DFC%FDDD x U^ 0G`0<`D11`00G`<1`0G`D11 is false if and only if ydx 0y U^ G`0<`D1 is true and y U^ 0G`<1`0G`D1 is false1 if and only if if and only if :dy 00: U^ G1`0: U^ <`D11 :dy 0: SW G`< is true and : U^ G`D is false1 if and only if if and only if ud: 00u U^ G1`0u U^ <11 ud: 0u U^ G is true and u U^ D is false1 Hence, there are y, :, u such that xcyc:cu, u U^ G is true and u SW D is false& 9rom ud: 00u U^ G1`0u U^ <11 e obtain that u U^< also is true, and from :dy 00: U^ G1`0: U^ <`D11 3 that : U^ <`D is true& Then, by +emma .&.&1, u U^ <`D also is true, i&e& #du 00# U^ <1`0# U^ D11, in particular, 0u U^ <1`0u U^ D1& Hence, uSW D is true& <ontradiction& Thus, x U^ +2 is true& L(= C%F DFCC%FR DFR%D x U^ 0G`<1`00G`f<1`fG1 is false if and only if ydx 0y U^ G`< is true and y U^ 0G`f<1`fG is false1

1-if and only if ;Xy CC; SW %DFC; SW DD if and only if :dy 0: U^ G`f< is true and : U^ fG is false1 if and only if if and only if uX; CCu SW %DFCu SW R DD ud: 0u SW % is true1

Hence, there are y, :, u such that xcyc:cu , and u SW % is true& 9rom ;Xy CC; S W %DFC; SW DD e obtain that u U^ < is true& 9rom uX; CCu SW %DFCu SW R DD e obtain that u U^ f< is true, i&e& # U^ < is false for some #du& Gy +emma .&.&1, if u U^ < is true, then # U^ < is true& <ontradiction ith S# U^ < is falseS& Hence, x U^ +, is true& !7ercise *.*.). Verify that, in the abo#e recursi#e definition of x U^ 9, the item 2& 7mplication, con>unction or dis>unction: x U^ 9oA is defined to be true, if and only if, accordin! to the classical truth tables, 0y U^ 91o0y U^ A1 is true for all y O x b & could be replaced by 2a& 7mplication 0Snon-monotonicS connecti#e1: x U^ 9`A is defined to be true, if and only if, accordin! to the classical truth tables, 0y U^ 91`0y U^ A1 is true for all y O x b & 2b& <on>unction or dis>unction 0SmonotonicS connecti#es1: x U^ 9oA is defined to be true, if and only if, accordin! to the classical truth tables, 0x U^ 91o0x U^ A1 is true& @nd of @xercise .&.&(& 9he $ard Part of the Proof ?o , let us try pro#in! that, if 9 is true at all nodes in all $ripke scenarios, then 9 is pro#able in the constructi#e propositional lo!ic1& 2e ill follo the paper by Kudith Under8ood. 6 constructi#e <ompleteness %roof for 7ntuitionistic %ropositional <alculus& T)-,0-11=,, December 1,,0, ,epartment of Comp$ter Science, Corne%% 2ni'ersity& based on the contructions from Melvin :itting& 7ntuitionistic +o!ic, 5odel Theory and 9orcin!& ?orthHolland, 6msterdam, 1,-, The smart idea is to generali;e the "roblem in the follo in! ay& 7nstead of considerin! constructi#e pro#ability of sin!le formulas, let us consider the constructi#e pro#ability of D1, D2, &&&, Dm U- C 1C 2&&& C n for arbitrary formulas D1, D2, &&&, Dm, <1, <2, &&&, <n, i&e& let us consider ordered pairs of

1-= sets 0XD1, D2, &&&, DmY, X<1, <2, &&&, <nY1& +et us call such pairs se>uents& 7f '1, '2 are sets of formulas 0'1 may be empty1, let us call the seQuent 0' 1, '21 constructi#ely pro#able, if and only if V+1-+10, 5%W: '1 U- V'2, here V'2 denotes the dis>unction of formulas contained in '2& 5oreo#er, let us consider sets of se>uents& This ill allo to carry out a specific induction ar!ument 0considerin! sin!le formulas or sin!le seQuents does not allo such an ar!ument;1& +et us say that a $ripke scenario 0b, c, t1 contains a countere7am"le for the seQuent 0'1, '21, if and only if the seQuent is false at some node in the scenario 0or, more precisely, if and only if there is x b such that x U^ 9 for all formulas F S 1 and not x U^ A for all formulas G S 2 1& 6dditionally, let us use <orollary -&1&20b1 of Theorem -&1&1 to replace all ne!ations f9 by 9`f, here f is an atomic formula, hich is Sal ays falseS, i&e& hich, in a seQuent 0' 1, '21, ne#er belon!s to '1& Thus, formulas mentioned in the proof of the follo in! Theorem .&.&- do not contain ne!ations 0but they may contain the specific atomic formula f1& 9heorem *.*./. 9or any set ' of seQuents, either some seQuent of ' is constructi#ely pro#able, or there is a $ripke scenario 0b, c, t1, hich contains counterexamples for each seQuent in '& Proof. +et us start cases: ith a "roof overvie8& 2e ill consider the follo in!

ase '. ' contains 0'1, '21 such that A B S 1( A S 1 B S 1) & +et us consider the set 'B obtained from ' by addin! the Smissin!S formulas 6, G to '1, i&e& by replacin! 0'1, '21 by 0 S 1{ A , B } , '21& +et us #erify that if Theorem is true for 'B, then it is true for '&&& ase .. ' contains 0'1, '21 such that A B S 2( A S 2 B S 2 ) & +et us consider the follo in! t o sets: a1 'B 3 obtained from ' by addin! the formula 6 to '2, i&e& by replacin! 0'1, '21 by 0'1, S 2{ A } 1& b1 'BB 3 obtained from ' by addin! the formula G to '2, i&e& by replacin! 0'1, '21 by 0'1, S 2{ B } 1& +et us #erify that if Theorem is true for 'B and 'BB, then it is true for '&&& ase ). ' contains 0'1, '21 such that A B S 1( A S 1 B S 1) & +et us consider the follo in! t o sets: a1 'B 3 obtained from ' by addin! the formula 6 to '1, i&e& by replacin! 0'1, '21 by 0 S 1{ A } , '21& b1 'BB 3 obtained from ' by addin! the formula G to '1, i&e& by replacin! 0'1, '21 by 0 S 1{ B } , '21& +et us #erify that if Theorem is true for 'B and 'BB, then it is true for '&&&

1-C ase *. ' contains 0'1, '21 such that A B S 2( A S 2 B S 2) & +et us consider the set 'B obtained from ' by addin! the Smissin!S formulas 6, G to '2, i&e& by replacin! 0'1, '21 by 0'1, S 2{ A , B } 1& +et us #erify that if Theorem is true for 'B, then it is true for '&&& ase -. ' contains 0'1, '21 such that A 5 B S 1( A S 2 B S 1 ) & +et us consider the follo in! t o sets: a1 'B 3 obtained from ' by addin! the formula 6 to '2, i&e& by replacin! 0'1, '21 by 0'1, S 2{ A } 1& b1 'BB 3 obtained from ' by addin! the formula G to '1, i&e& by replacin! 0'1, '21 by 0 S 1{ B } , '21& +et us #erify that if Theorem is true for 'B and 'BB, then it is true for '&&& ase /. ' contains 0'1, '21 such that A 5 B S 2 and for e#ery seQuent ( T 1, T 2 ) S , ( S 1T 1 A T 1 B T 2) & +et us consider the set 'B obtained from ' by addin! the seQuent 0 S 1{ A } , G1 to it& +et us #erify that if Theorem is true for 'B, then it is true for '&&& ase 0. ?one of the abo#e cases hold for '& Then, Theorem is true for ' 3 easy to #erify&&& The first six cases represent the induction ar!ument: pro#in! of Theorem for a seQuent set ' is reduced to pro#in! it for some other sets 3 'B and 'S& Gy iteratin! this reduction, e al ays arri#e happily to the <ase =, here Theorem is easy to #erify& 7ndeed, let us denote by $ni'erse ( S 1, S 2 ) the set of all formulas and subformulas 0of the formulas1 contained in S 1 S 2 & +et us denote by $ni'erse ( S ) the union of the uni#erses of seQuents from '& !7ercise *.*.*& Verify that: a1 2hen, in the <ases 1-/, the seQuent 0'1, '21 is replaced by some other seQuent 0T1, T21, then
$ni'erse (T 1, T 2 )$ni'erse ( S 1, S 2) &

b1 2hen, in the <ase -, because of the seQuent 0' 1, '21, the seQuent ( S 1{ A } , B ) is added to ', then $ni'erse ( S 1{ A } , B ) $ni'erse ( S 1, S 2) & c1 9or a !i#en $ni'erse ( S ) , there exist no more than " = 2$ni'erse ( S )+1 different seQuents 0'1, '21 such that $ni'erse ( S; , S9) P $ni'erse ( S ) & 6nd, no more than 2? different sets of seQuents& Thus, any chain of iterated <ases 1-- cannot be lon!er than 2 ?\1 3 either e ill arri#e at a set of seQuents already built at a pre#ious step, or e ill arri#e

1-, at the <ase =& ?o 3 the proof as it should be& ase '. ' contains 0'1, '21 such that A B S 1( A S 1 B S 1) & +et us consider the set 'B obtained from ' by addin! the Smissin!S formulas 6, G to '1, i&e& by replacin! 0'1, '21 by ( S 1{ A , B } , S 2) & +et us #erify that if Theorem is true for 'B, then it is true for '& 6ssume, some seQuent of 'B is constructi#ely pro#able, then it is ( S 1{ A , B } , S 2) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able& 7f ( S 1{ A , B } , S 2) is constructi#ely pro#able, then so is 0'1, '21& 7ndeed, if S 1{ A , B } U- V'2 is constructi#ely pro#able, ho to pro#e '1 U- V'28 'ince '1 contains A B , by axioms +( and +( e can deri#e 6 and G& 6fter this, e can apply the proof of S 1{ A , B } U- V'2& Hence, '1 U- V'2 is constructi#ely pro#able& Rn the other side, if there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'B, then it contains also a counterexample for each seQuent in '& 7ndeed, a seQuent in ' is either 0'1, '21, or some other seQuent& 7f it is some other seQuent, then it belon!s to 'B, i&e& 0b, c, t1 contains a counterexample for it& Does 0b, c, t1 contain a counterexample also for 0' 1, '218 2e kno that it contains a counterexample for ( S 1{ A , B } , S 2) , i&e& for some x b , x U^ 9 for all formulas F S 1{ A , B } and not x U^ A for all formulas G S 2 & Hence, 0b, c, t1 contains a counterexample also for 0' 1, '21& D&@&D& ase .. ' contains 0'1, '21 such that A B S 2( A S 2 B S 2 ) & +et us consider the follo in! t o sets: a1 'B 3 obtained from ' by addin! the formula 6 to ' 2, i&e& by replacin! 0'1, '21 by ( S 1 , S 2 { A}) & b1 'BB 3 obtained from ' by addin! the formula G to ' 2, i&e& by replacin! 0'1, '21 by ( S 1 , S 2 {B }) & +et us #erify that if Theorem is true for 'B and 'BB, then it is true for '& 6ssume, some seQuent of 'B and some seQuent of 'BB is constructi#ely pro#able& The seQuent of 'B is ( S 1 , S 2 { A}) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able&

1=0 The seQuent of 'BB is ( S 1 , S 2 {B }) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able& 'o, let us consider the situation, hen ( S 1 , S 2 { A}) and ( S 1 , S 2 {B }) both are constructi#ely pro#able& 7f '1 U- A S 2 and '1 U- B S 2 both are constructi#ely pro#able, ho pro#e '1 U- V'2 0 e kno that '2 contains A B 18 Gy Theorem 2&(&1, con>unction is distributi#e to dis>unction: V+1-+C, 5%W: U- ( A B )C 6 ( A C )( B C ) & Hence, V+1-+C, 5%W: ( A S 2)( B S 2 ) 5 ( A B )S 2 & 'o, let us mer!e the proofs of '1 U- A S 2 and '1 U- B S 2 , and let us append the proof of Theorem 2&(&1& Thus, e ha#e obtained a proof of '1 U- ( A B)S 2 & 9rom 'ection 2&( e kno that in V+1-+C, 5%W dis>unction is associati#e, commutati#e and idempotent& 6nd, by )eplacement +emma 10e1: V+1-+C, 5%W 6aG U- AC 6 B C & 'ince '2 contains A B , these facts allo , from a proof of '1 U- ( A B)S 2 , to deri#e a proof of '1 U- V'2& Rn the other side, if there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'B, then it contains also a counterexample for each seQuent in '& 7ndeed, a seQuent in ' is either 0'1, '21, or some other seQuent& 7f it is some other seQuent, then it belon!s to 'B, i&e& 0b, c, t1 contains a counterexample for it& Does 0b, c, t1 contain a counterexample also for 0' 1, '218 2e kno that it contains a counterexample for ( S 1 , S 2 { A}) , i&e& for some x b , x U^ 9 for all formulas F S 1 and not x U^ A for all formulas G S 2{ A} & Hence, 0b, c, t1 contains a counterexample also for 0' 1, '21& D&@&D& 7f there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'BB, then it contains also a counterexample for each seQuent in '& The ar!ument is similar to the abo#e& ase ). ' contains 0'1, '21 such that A B S 1( A S 1 B S 1) & +et us consider the follo in! t o sets: a1 'B 3 obtained from ' by addin! the formula 6 to ' 1, i&e& by replacin! 0'1, '21 by ( S 1{ A } , S 2 ) & b1 'BB 3 obtained from ' by addin! the formula G to ' 1, i&e& by replacin! 0'1, '21 by ( S 1{ B } , S 2 ) & to

1=1 +et us #erify that if Theorem is true for 'B and 'BB, then it is true for '& 6ssume, some seQuent of 'B and some seQuent of 'BB is constructi#ely pro#able& The seQuent of 'B is ( S 1{ A } , S 2 ) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able& The seQuent of 'BB is ( S 1{ B } , S 2 ) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able& 'o, let us consider the situation, hen ( S 1{ A } , S 2 ) and ( S 1{ B } , S 2 ) both are constructi#ely pro#able& +et us recall the @xercise 2&(&2 V+1, +2, +C, 5%W: if 61, 62, &&&, 6n, G U- D, and 61, 62, &&&, 6n, < U- D, then 61, 62, &&&, 6n , BC U- D& Thus, if S 1{ A } UV'2 and S 1{ B } U- V'2 both are constructi#ely pro#able, then 0since ' 1 contains A B 1 so is '1*XGYU- V'2& Rn the other side, if there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'B, then it contains also a counterexample for each seQuent in '& 7ndeed, a seQuent in ' is either 0'1, '21, or some other seQuent& 7f it is some other seQuent, then it belon!s to 'B, i&e& 0b, c, t1 contains a counterexample for it& Does 0b, c, t1 contain a counterexample also for 0' 1, '218 2e kno that it is contains counterexample for ( S 1{ A } , S 2 ) , i&e& for some x b , x U^ 9 for all formulas F S 1{ A} and not x U^ A for all formulas G S 2 & Hence, 0b, c, t1 contains a counterexample also for 0' 1, '21& D&@&D& 7f there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'BB, then it is also contains counterexample for each seQuents in '& The ar!ument is similar to the abo#e& ase *. ' contains 0'1, '21 such that A B S 2( A S 2 B S 2) & +et us consider the set 'B obtained from ' by addin! the Smissin!S formulas 6, G to '2, i&e& by replacin! 0'1, '21 by ( S 1 , S 2 { A , B }) & +et us #erify that if Theorem is true for 'B, then it is true for '& 6ssume, some seQuent of 'B is constructi#ely pro#able, then it is ( S 1 , S 2{ A , B }) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able& 7f ( S 1 , S 2{ A , B }) is constructi#ely pro#able, then so is 0'1, '21& 7ndeed, if '1U- ( A B )S 2 is constructi#ely pro#able, ho to pro#e '1 U- V'2 0 here '2 contains A B 18 9rom 'ection 2&( e kno that in V+1-+C, 5%W dis>unction is associati#e,

1=2 commutati#e and idempotent& 6nd, by )eplacement +emma 10e1: V+1-+C, 5%W 6aG U- AC 6 B C & 'ince that '2 contains 6#G, these facts allo , from a proof of '1 U- ( A B )S 2 , to deri#e a proof of '1 U- V'2& Rn the other side, if there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'B, then it contains also a counterexample for each seQuent in '& 7ndeed, a seQuent in ' is either 0'1, '21, or some other seQuent& 7f it is some other seQuent, then it belon!s to 'B, i&e& 0b, c, t1 contains a counterexample for it& Does 0b, c, t1 contain a counterexample also for 0' 1, '218 2e kno that it contains a counterexample for ( S 1 , S 2{ A , B }) , i&e& for some x b , x U^ 9 for all formulas F S 1 and not x U^ A for all formulas G S 2{ A , B } & Hence, 0b, c, t1 contains a counterexample also for 0'1, '21& D&@&D& 7f there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'BB, then it contains also a counterexample for each seQuent in '& The ar!ument is similar to the abo#e& ase -. ' contains 0'1, '21 such that A 5 B S 1( A S 2 B S 1 ) & +et us consider the follo in! t o sets: a1 'B 3 obtained from ' by addin! the formula 6 to ' 2, i&e& by replacin! 0'1, '21 by ( S 1 , S 2 { A}) & b1 'BB 3 obtained from ' by addin! the formula G to ' 1, i&e& by replacin! 0'1, '21 by ( S 1{ B } , S 2 ) & +et us #erify that if Theorem is true for 'B and 'BB, then it is true for '& 6ssume, some seQuent of 'B and some seQuent of 'BB is constructi#ely pro#able& The seQuent of 'B is ( S 1 , S 2{ A}) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able& The seQuent of 'BB is ( S 1{ B } , S 2 ) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able& 'o, let us consider the situation, hen ( S 1 , S 2{ A}) and ( S 1{ B } , S 2 ) both are constructi#ely pro#able& 2e ha#e t o proofs: '1 U- A S 2 and '1, G U- V'2, and contains 6`G& Ho to deri#e a proof of '1 U- V'28 e kno that '1

'ince '1 contains 6`G, e ha#e a proof of ' 1, 6 U- G& To!ether ith '1, G UV'2 this yields a proof of '1, 6 U- V'2& Rf course, V'2 U- V'2& ?o , let us

1=( recall the @xercise 2&(&2 V+1, +2, +C, 5%W: 7f 61, 62, &&&, 6n, G U- D, and 61, 62, &&&, 6n, < U- D, then 6 1, 62, &&&, 6n , BC U- D& Thus, '1, A S 2 U- V'2& 'ince e ha#e a proof of '1 UA S 2 , e ha#e also a proof of '1 U- A S 2 & Rn the other side, if there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'B, then it contains also a counterexample for each seQuent in '& 7ndeed, a seQuent in ' is either 0'1, '21, or some other seQuent& 7f it is some other seQuent, then it belon!s to 'B, i&e& 0b, c, t1 contains a counterexample for it& Does 0b, c, t1 contain a counterexample also for 0' 1, '218 2e kno that it contains a counterexample for ( S 1 , S 2{ A}) , i&e& for some x b , x U^ 9 for all formulas F S 1 and not x U^ A for all formulas G S 2{ A} & Hence, 0b, c, t1 contains a counterexample also for 0' 1, '21& D&@&D& 7f there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'BB, then it contains also a counterexample for each seQuent in '& The ar!ument is similar to the abo#e& ase /. ' contains 0'1, '21 such that A 5 B S 2 and for e#ery seQuent ( T 1, T 2 ) S , ( S 1T 1 A T 1 B T 2) . +et us consider the set 'B obtained from ' by addin! the seQuent ( S;2 A , B ) to it& +et us #erify that if Theorem is true for 'B, then it is true for '& 6ssume, some seQuent of 'B is constructi#ely pro#able, then it is ( S 1{ A } , B ) or some other seQuent& 7f it is some other seQuent, then it belon!s to ', i&e& some seQuent of ' is constructi#ely pro#able& 7f ( S 1{ A } , B ) is constructi#ely pro#able, then so is 0'1, '21& 7ndeed, if '1, 6 U- G is constructi#ely pro#able, then, by Deduction Theorem 1, ' 1 U- 6`G, and '1 U- V'2 0since '2 contains 6`G1& Rn the other side, if there is a $ripke scenario 0b, c, t1, hich contains a counterexample for each seQuent in 'B, then, since ' is a subset of 'B, this scenario contains also a counterexample for each seQuent in '& ase 0. ?one of the abo#e cases hold for '& Hence, for e#ery seQuent ( S 1 , S 2 ) S the follo in! holds: 11 7f A B S 1 , then A S 1 B S 1 , 21 7f (1 7f A B S 2 , then A S 2 B S 2 , A B S 1 , then A S 1 B S 1 ,

1=. .1 7f A B S 2 , then A S 2 B S 2 , /1 7f A 5 B S 1 , then A S 2 B S 1 , -1 7f A 5 B S 2 , then there is

( T; , T9 ) S such that S 1T 1 A T 2 B T 2 &


9or this kind of seQuent sets e ha#e a #ery simple situation: a1 7f, in some seQuent ( S 1 , S 2 ) S the sets '1, '2 contain the same formula 6, then from +-: A 5 A B e can deri#e easily that V+ 1-+C, 5%W: '1 UV'2& b1 7f the sets '1, '2 are dis>oint for all seQuents ( S 1 , S 2 ) S , then e must 0and ill1 build a scenario, containin! a counterexample for each seQuent in '& 'o, let us suppose that the sets '1, '2 are dis>oint for all seQuents ( S 1, S 2 ) S , and let us define the follo in! $ripke scenario 0b, c, t1: b ^ ', xcy must be defined for e#ery t o members x, y of b, i&e& for e#ery t o seQuents 0'1, '21 and 0T1, T21 in '& +et us define 0'1, '21 c 0T1, T21, if and only if S 1T 1 & Rf course, G G is a partial orderin! of b& t must be a monotonic mappin! from members of b to sets of atomic formulas& +et us define t0'1, '21 as the set of all atomic formulas in ' 1& Rf course, t is monotonic for G G & 06nd, of course, f 3 our atomic SfalseS, ne#er belon!s to t0'1, '211& Thus, 0b, c, t1 is a $ripke scenario& +et us pro#e that it contains a counterexample for each seQuent in '& 7n fact, e ill pro#e that for each seQuent ( S 1 , S 2 ) S , and each formula 9: 7f F S 1 , then 0'1, '21 U^ 9& 7f F S 2 , then not 0'1, '21 U^ 9& This ill mean that, 0'1, '21 represents a counterexample for 0'1, '21& Rf course, our proof ill be by induction alon! the structure of the formula 9& a1 9 is an atomic formula& 7f F S 1 , then F t (T 1, T 2 ) for '21c0T1, T21& Hence, 0'1, '21 U^ 9& e#ery ( T 1 , T 2 ) S such that 0'1,

1=/ 7f F S 2 , then, since '1 and '2 are dis>oint sets, F S 1 , and F t ( S 1, S 2 ) , i&e& not 0'1, '21 U^ 9& b1 9 is A B & 7f F S 1 , then, by 011, A S 1 B S 1 & Hence, by induction assumption, 0'1, '21 U^ 6 and 0'1, '21 U^ G, i&e&, by @xercise .&.&(, 0'1, '21 U^ A B & 7f F S 2 , then, by 021, A S 2 B S 2 & 7f A S 2 , then, by induction assumption, not 0'1, '21 U^ 6, i&e&, by @xercise .&.&(, not 0' 1, '21 U^ A B & 7f B S 2 3 the ar!ument is similar& c1 9 is A B & 7f F S 1 , then, by 0(1, A S 1 B S 1 & 7f A S 1 , then, by induction assumption, 0'1, '21 U^ 6, i&e&, by @xercise .&.&(, 0' 1, '21 U^ A B & 7f B S 1 3 the ar!ument is similar& 7f F S 2 , then, by 0.1, A S 2 B S 2 & Gy induction assumption, not 0'1, '21 U^ 6 and not 0'1, '21 U^ G, i&e&, by @xercise .&.&(, not 0'1, '21 U^ A B & d1 9 is 6`G& d11 F S 1 & 2e must pro#e that 0'1, '21 U^ 6`G, i&e& that 0T1, T21 U^ 6`G for each ( T 1 , T 2 ) S such that 0'1, '21c0T1, T21& 'o, let us assume that not 0T1, T21 U^ 6`G, i&e& that 0*1, *21 U^ 6 and not 0*1, *21 U^ G for some ( 2 1, 2 2) S such that 0T1, T21c 0*1, *21& 'ince A 5 B S 1 , then also A 5 B 2 1 , and, by 0/1, A 2 2 B 2 1 & Gy induction assumption, this means that not 0* 1, *21 U^ 6 or 0*1, *21 U^ G& <ontradiction, hence, 0'1, '21 U^ 6`G& d21 F S 2 & 2e must pro#e that not 0'1, '21 U^ 6`G, i&e& that there is ( T 1, T 2 ) S such that 0'1, '21c0T1, T21 and 0T1, T21 U^ 6 and not 0T1, T21 U^ G& 'ince A 5 B S 2 , by 0-1, there is ( T 1, T 2 ) S such that 0'1, '21c0T1, T21 and A T 1 and B T 2 & Gy induction assumption, this means that 0T1, T21 U^ 6 and not 0T1, T21 U^ G& D&@&D& This completes the proof of Theorem .&.&-& 4ote. The abo#e proof contains an algorithm allo in! to find, for each set ' of seQuents, either a constructi#e proof of some seQuent of ', or a $ripke

1=scenario containin! counterexamples for each seQuent of '& orollary *.*.0. 7f a formula 9 is true at all nodes in all scenarios, then V+1-+10, 5%W U- 9 0i&e& 9 is pro#able in the constructi#e propositional lo!ic1& 7ndeed, let us consider the set of seQuents X00, X9Y1Y consistin! of a sin!le seQuent 00, X9Y1, here 0 is empty set& Gy Theorem .&.&-, either the seQuent 00, X9Y1 is constructi#ely pro#able, or there is a $ripke scenario 0b, c, t1, hich contains a counterexample for 00, X9Y1& 'ince 9 is true at all nodes in all $ripke scenarios, it cannot ha#e counterexamples4 hence, the seQuent 00, X9Y1 0i&e& the formula 91 is constructi#ely pro#able& To!ether ith +emma .&.&( this <orollary implies the abo#e Theorem .&.&2 3 Kri"keJs theorem on the com"leteness of the constructive "ro"ositional logic: a formula 9 is true at all nodes in all $ripke scenarios, if and only if 9 is pro#able in the constructi#e propositional lo!ic& orollary *.*.G Cdecidability of the constructive "ro"ositional logicD. There is an algorithm allo in! to determine for any formula 9, is this formula pro#able in the constructi#e propositional lo!ic V+1-+10, 5%W, or not& Aerhard Aent:en established this fact in 1,(.: 5. 5ent;en& *ntersuchun!en gber das lo!ische 'chliessen 77& .athematische Feitschrift, 1,(., Vol& (,, pp& .0/-.(1& orollary *.*.(& 7f F G is true at all nodes in all scenarios, then 9 is true at all nodes in all scenarios, or A is true at all nodes in all scenarios& Proof. 6ssume, there is a scenario 0b1, c1, t11 such that x1 U^ 9 is false for some x 1 b1 , and a scenario 0b2, c2, t21 such that x2 U^ A is false for some x 2 b2 & 2e may assume that the 0node1 sets b 1 and b2 do not intersect& +et us mer!e these scenarios by addin! a ne common startin! node x 0, here all Gi are false& Then, x0 U^ 9 is false 0+emma .&.&11, and x0 U^ A is false 0similarly1& Hence, accordin! to the classical dis>unction truth table, x 0 U^ F G is false& Gut, x U^ F G is al ays true& Hence, x U^ 9 is al ays true, or x U^ A is al ays true& D&@&D& 9heorem *.*.'I. 0AFdel V1,(2W1& 7f V+1-+10, 5%W: U- BC , then V+1-+10, 5%W: U- G or V+1-+10, 5%W: U- <& 07&e& if the dis>unction BC is constructi#ely pro#able, then one of the formulas G, < also is constructi#ely pro#able&1 Proof. 7f V+1-+10, 5%W: U- BC , then, by $ripkeBs <ompleteness Theorem

1== .&.&2, BC is true at all nodes in all scenarios& Then, by <orollary .&.&,, so is G or so is <& Gy $ripkeBs <ompleteness Theorem .&.&2, this means that one of the formulas G, < is constructi#ely pro#able& D&@&D& +et us recall the constructi#e interpretation of dis>unction from 'ection 1&(: - To pro#e BC constructi#ely, you must pro#e G, or pro#e <& To pro#e BC classically, you may assume ( BC ) as a hypothesis, and deri#e a contradiction& Ha#in! only such a Sne!ati#eS proof, you may be unable to determine, hich part of the dis>unction BC is true 3 G, or <, or both& 6ccordin! to Theorem .&.&10, the constructi#e propositional lo!ic V+ 1-+10, 5%W supports the constructi#e interpretation of dis>unction& $&AFdel established this fact in 1,(2: K. 5Mdel& Lum intuitionistischen 6ussa!enkalkgl& A&a+emie +er Bissenschaften in Bien, .athematisch> nat$r-issenschaft%iche C%asse, An3eiger, 1,(2, Vol&-,, pp&-/---& !7ercise *.*.- 0for smart students1& Gy addin! the schema ( B 5 C )( C 5 B ) to the axioms of the constructi#e lo!ic, e obtain the so-called GN+e%>,$mmett %ogic& Verify, that a propositional formula 9 is pro#able in AFdel-Dummett lo!ic, if and only if 9 is true at all nodes in all linear $ripke scenarios 0i&e& in the scenarious that do not allo branchin!1& 'ee also 7ntuitionistic +o!ic by Hoan 5oscho#akis in 'tanford @ncyclopedia of %hilosophy, and 5ichael Dummett in 7nternet @ncyclopedia of %hilosophy&

1=C

-. 4ormal :orms. #esolution Method


7n this section, e ill try to produce a practical method allo in! to deri#e conseQuences and pro#e theorems by usin! computers& 7n !eneral, this task is not feasible because of its enormous computational complexity 0see 'ection .&(1& 'till, for problems of a Spractical si:eS 0arisin!, for example, in deducti#e databases and other artificial intelli!ence systems, or, tryin! to formali:e real mathematical proofs1, such methods are possible and some of them are already implemented successfully& This field of research is called automated reasonin!, or automated theorempro#in!& 1arning6 The principal results of this 'ection are #alid only for the classical logic; Main Ideas 7f 91, &&&, 9n is the set of our assumptions 0facts, rules, axioms, hypotheses etc&1, does the assertion A follo from this set8 Rne of the ell kno n approaches to pro#in! theorems in mathematics 3 and especially con#enient for computers 3 are the so-called refutation "roofs 0re+$ctio a+ abs$r+$m1 3 proofs by deri#in! a contradiction: assume fA, and deri#e a contradiction& 7&e& pro#e that 91, &&&, 9n, fA is an inconsistent set of assumptions& Idea V'= let us deri#e conseQuences and pro#e theorems only in this 8ay& +et us try de#elopin! the best possible method of deri#in! contradictions from inconsistent sets of assumptions& This 0at first !lance 3 tri#ial1 decision is one of the most important steps in the hole story 3 it ill allo 0see 'ection /&2 belo 1 con#ersion of the formulas 91, &&&, 9n, fA into a form that does not contain existential Quantifiers& 6nd after this, ha#in! uni#ersal Quantifiers only, e may simply drop them at all, and continue orkin! ith Quantifier-free formulas 0see 'ection /&.1& Idea V.: let us Snormali:eS our assumptions as far as possible& The first step 0idea V.a1 is here reducin! to the so-called "rene7 normal form 3 mo#in! all the Quantifiers to left& 9or example, the formula V0xG0x1 ` x<0x11 ` xD0x1W ` x90x1 is eQui#alent 0in the classical lo!ic;1 to the follo in! formula in prenex normal form:

1=, x1x2x(x.VV0G0x11 ` <0x211 `D0x(1W ` 90x.1W& 02hen mo#in! Quantifiers to left, some of them must be chan!ed from to , or from to , see 'ection /&1 belo &1 The second step 0idea V.b, due to Thoralf 'kolem1 allo s elimination of existential Quantifiers& 7ndeed, x1x2 means that x2^f0x11, and x1x(x. means that x.^!0x1, x(1, here f and ! are some functions 0see 'ection /&21& 7n this ay e obtain the so-called ,kolem normal form, containin! uni#ersal Quantifiers only: x1x(VV0G0x11 ` <0f0x1111 `D0x(1W ` 90!0x1, x(11W& ?ote that a formula and its 'kolem normal form are not e>uivalent 0e#en in the classical lo!ic;1, they are only a kind of Ssemi-eQui#alentS: a set of formulas is inconsistent, if and only if so is the set of their 'kolem normal forms& ?o , since, our formulas contain uni#ersal Quantifiers only, e may dro" these >uantifiers 0simply by assumin! that all free #ariables are uni#ersally Quantified1: V0G0x11 ` <0f0x1111 `D0x(1W ` 90!0x1, x(11& The third step 0idea V.c1 3 reduction of Quantifier-free formulas to the socalled con@unctive normal form 0a con>unction of dis>unctions of atomic formulas 3 ith or ithout ne!ations, see 'ection /&(1& 9or example, the abo#e formula can be reduced to the follo in! form:
( 7 B ( x 1)C ( f ( x 1)) F ( g ( x 1, x ( )))( 7 , ( x( ) F ( g ( x1, x( ))) &

Gy assumin! that a set of formulas is eQui#alent to their con>unction, drop the con>unction0s1 obtainin! a set of the so-called clauses: 7 B ( x 1)C ( f ( x 1)) F ( g ( x 1, x ( )) 4
7 , ( x () F ( g ( x 1, x ()) &

e can

@ach clause is a dis>unctions of atomic formulas 3 ith or ithout ne!ations& 7n this ay, instead of our initial set of assumptions 9 1, &&&, 9n, fA, e obtain a set of clauses 0Mlar!e cloud of simple dis>unctionsN1, hich is inconsistent, if and only if so is the set 91, &&&, 9n, fA& The last step 3 ho to ork ith a set of clauses 0Mlar!e cloud of simple dis>unctionsN18 Idea V) 0due to Hohn 6lan )obinson, see 'ection /&/ and /&=1 3 a set of clauses is inconsistent, if and only if a contradiction can be deri#ed from it by usin! term substitution and the so-called #obinsonJs #esolution

1C0 rule: F C , C G & F G <ontinue readin!&&& 3lternative method: the so-called 5ethod of 6nalytic Tableaux&

-.'. Prene7 4ormal :orm


1arning6 The principal results of this 'ection are #alid only for the classical logic; +et us consider an interpretation H of some predicate lan!ua!e +, such that the domain DH contains an infinite set of Sob>ectsS& *nder such interpretation, the Smeanin!S of formulas containin! Quantifiers may be more or less nonconstructi#e, or, at least, Sconstructi#ely difficultS& 9or example, the formula xG0x1 ill be true, if G0x1 ill be true for all Sob>ectsS x in the 0infinite;1 set D H& Thus, it is impossible to #erify directly 0i&e& SempiricallyS1, is xG0x1 true or not& 'ayin! that xy0x\y^y\x1 is true under the standard interpretation of first order arithmetic, does not mean that e ha#e #erified this fact empirically 3 by checkin! x\y^y\x for all pairs of natural numbers x, y& Then, ho do e kno that xy0x\y^y\x1 is true8 Rf course, e either postulated this feature of natural numbers directly 0i&e& deri#ed it from Sempirical e#idenceS1, or pro#ed it by usin! some set of axioms 0i&e& deri#ed it from other postulates1& Gut, in !eneral, formulas ha#in! the form xG0x1, are Sconstructi#ely difficultS& The formula xy<0x, y1 may be e#en more difficult: it ill be true, if for each x in DH e ill be able to find y in D H such that <0x, y1 is true& Thus, thinkin! constructi#ely, e could say that xy<0x, y1 is true, only, if there is an al!orithm, hich, for each x in DH can find y in DH such that <0x, y1 is true& 9or example, under the standard interpretation of first order arithmetic, the formula x y ( x < y prime ( y )) is true 0i&e& Sthere are infinitely many prime numbersS1& Ho do e kno this8 This fact as pro#ed in V7 century G<& Gut the 0similarly Quantified1 formula x y ( x < y prime ( y ) prime ( y +2 )) , i&e& the famous t in prime con>ecture, is it true or not8 *ntil no , nobody

1C1 kno s the ans er& !7ercise -.'.'. Verify that the Smeanin!S of xy:D0x, y, :1 and xy:u90x, y, :, u1 may be e#en more non-constructi#e& Gut ho about the formula xA0x1`yH0y18 7s it constructi#ely more difficult than xy<0x, y1, or less8 7n !eneral, e could pro#e that xA0x1`yH0y1 is true, if e had an al!orithm, hich, for each x , / such that A0x1 is true, could find y ,# such that A0 y1 is true, i&e& if xy0A0x1`H0y11 ould be true& 2e ill establish belo , that, in the classical lo!ic, if A does not contain y, and H does not contain x, then the formula xA0x1`yH0y1 is eQui#alent to xy0A0x1`H0y11& Thus, in !eneral, the formula xA0x1`yH0y1 is constructi#ely as difficult as is the formula xy<0x, y1; To !enerali:e this approach to comparin! Sconstructi#e difficultyS of formulas, the so-called "rene7 normal forms ha#e been introduced: a1 7f a formula does not contain Quantifiers, then it is in the prenex normal form& b1 7f x is any #ariable, and the formula 9 is in the prenex normal form, then x9 and x9 also are in the prenex normal form& c1 07f you ish so,1 there are no other formulas in the prenex normal form& 7&e& a formula is in the prenex normal form, if and only if it has all its Quantifiers !athered in front of a formula that does not contain Quantifiers& 7t appears, that in the classical lo!ic, each formula cane be SreducedS to an appropriate eQui#alent formula in the prenex normal form& To obtain this normal form, the follo in! +emmas /&1&1-/&1&( can be used& Lemma -.'.'. 7f the formula A does not contain x as a free #ariable, then: a1 V+1, +2, +/, +12, +1., 5%, AenW: 0A`x90x11 a x0A`90x11& b1 V+1, +2, +/, +12-+1/, 5%, AenW: 0x90x1`A1 a x090x1`A1& 2hat does it mean precisely8 c1 V+1-+11, +12-+1/, 5%, AenW: 0A`x90x11 a x0A`90x11& 5ore precisely: V+1-+11, +12-+1/, 5%, AenW: 0A`x90x11 ` x0A`90x11& This formula cannot be pro#ed constructi#ely; 2hy8 'ee 'ection .&/& Gut the con#erse formula can be pro#ed constructi#ely: V+1, +2, +1(-+1/, 5%, AenW: x0A`90x11 ` 0A`x90x11& d1 V+1-+11, +12-+1/, 5%, AenW: 0 x90x1`A1 a x090x1`A1& 2hat does it mean precisely8 5ore precisely:

1C2 V+1-+11, +12-+1/, 5%, AenW: 0 x90x1`A1 ` x090x1`A1& This formula cannot be pro#ed constructi#ely; 2hy8 'ee 'ection .&/& Gut the con#erse formula can be pro#ed constructi#ely: V+1, +2, +1(-+1/, 5%, AenW: x090x1`A1 ` 0 x90x1`A1& Proof. 9irst, let us note that 0a1j is an instance of the axiom +1.: x0A`90x11`0A`x90x11, and that 0b1j is an instance of the axiom +1/& %ro#e 0a1` and 0b1` as the @xercise /&1&2 belo & +et us pro#e 0c1j: x0A`90x11`0A`x90x11& 011 021 0(1 0.1 0/1 0-1 A`90x1 A 90x1 x90x1 0A`90x11`0A`x90x11 x00A`90x11`0A`x90x111 x0A`90x11`0A`x90x11 Hypothesis& Hypothesis& Gy 5%& Gy 6xiom +1(: 90x1`x90x1& Gy Deduction Theorem 1& Gy Aen& Gy 6xiom +1/: x090x1`A1`0x90x1`A1, since A`x90x1 does not contain x as a free #ariable&

0=1

+et us pro#e 0d1j: x090x1`A1 `0 x90x1`A1& 011 021 0(1 0.1 0/1 0-1 0=1 90x1`A x90x1 90x1 A 090x1`A1`0 x90x1`A1 Hypothesis& Hypothesis& Gy 6xiom +12: x90x1`90x1& Gy 5%& Gy Deduction Theorem 1&

x0090x1`A1`0 x90x1`A11 Gy Aen& x090x1`A1 `0 x90x1`A1 Gy 6xiom +1/:

1C( x090x1`A1`0x90x1`A1, since x90x1`A does not contain x as a free #ariable& ?o , let us pro#e 0c1`: 0A`x90x11 ` x0A`90x11 in the classical lo!ic 0a constructi#e proof is impossible, see 'ection .&/1& 9irst, let us pro#e: 011 021 0(1 0.1 fA`0A`90x11 0A`90x11`x0A`90x11 fA`x0A`90x11 fA ` 00A`x90x11 ` x0A`90x111 ?o , let us pro#e: 0/1 0-1 0=1 0C1 0,1 A A`x90x1 x90x1 90x1`0A`90x11 x090x1`0A`90x111 R5 ` 00A`x90x11`x0A`90x111 6xiom +10& 6xiom +1(: 90x1`x90x1& 9rom 011 and 021& Gy 6xiom +1: G`0<`G1& 5 ` 00A`x90x11`x0A`90x111 Hypothesis& Hypothesis& 9rom 0/1 and 0-1& 6xiom +1: G`0<`G1& Gy Aen& Gy Theorem (&1&10b1, V+1, +2, +12+1/, 5%, AenW Ux0G`<1`0xG`x<1& 9rom 0=1 and 0101& Gy Deduction Theorem 2 0x is not a free #ariable in A and A`x90x1& 9rom 0.1 and 0121, by 6xiom +C& The total is V+1, +2, +C, +10, +12-+1/, 5%, AenW Gy 37iom L'': A#fA&

0101 x90x1`x0A`90x11 0111 x0A`90x11 0121 A ` 00A`x90x11` x0A`90x111

5vR5 ` 00A`x90x11` 01(1 x0A`90x111 01.1 0A`x90x1` x0A`90x11

1C. 9inally, let us pro#e 0d1`: 0 x90x1`A1 ` x090x1`A1 in the classical lo!ic0a constructi#e proof is impossible, see 'ection .&/1& +et us denote this formula by H& 9irst, let us pro#e: 011 x90x1 021 x90x1`A 0(1 A 0.1 90x1`A 0/1 x090x1`A1 0-1 x90x1 `H ?o , let us pro#e: 0/1 f90x1 0-1 f90x1`090x1`A1 0=1 90x1`A 0C1 x090x1`A1 0,1 0x90x1`A1 ` x090x1`A1 0101 f90x1`H 0111 7R:C7D`H 7:C7D`H Hypothesis& Hypothesis& 9rom 011 and 021& Gy 6xiom +1: G`0<`G1& Gy 6xiom +1(: 90x1`x90x1& Gy Deduction Theorem 2& 7R:C7D`H Hypothesis& 6xiom +10& 9rom 0/1 and 0-1& Gy 6xiom +1(: 90x1`x90x1& Gy 6xiom +1: G`0<`G1& Gy Deduction Theorem 2& Gy Aen and 6xiom +1/: x0f90x1`H1` 0xf90x1`H1& Gy 'ection (&2, 777-.& V+1-+11, +1(, +1., 5%, AenW: Ufx90x1`xf90x1& 37iom L'' is used here6 9rom 0.1 and 0121, by 6xiom +C& Gy 6xiom

0121 f7:C7D`H

01(1 7:C7D v R7:C7D ` H 01(1 H

1C/ +11: 7:C7D v R7:C7D D&@&D& !7ercise -.'... a1 %ro#e 0a1` of +emma /&1&1, V+1, +2, +/, +12, +1., 5%, AenW: 0A`x90x11 ` x0A`90x11& b1 %ro#e 0b1` of +emma /&1&1, V+1, +2, +/, +12-+1/, 5%, AenW: 0x90x1`A1 ` x090x1`A1& Lemma -.'... 7f the formula A does not contain x as a free #ariable, then a1 V+1-+/, +12-+1/, 5%, AenW: xF ( x )G 6 x ( F ( x )G ) & b1 V+1-+/, +12, +1., 5%, AenW: xF ( x )G 6 x ( F ( x )G ) & c1 V+1, +2, +/, +--+C, +12-+1/, 5%, AenW: xF ( x )G 6 x ( F ( x )G ) & d1 V+1-+11, +12, +1., 5%, AenW: xF ( x )G 6 x ( F ( x )G ) & 5ore precisely: V+1, +2, +/, +--+C, +12, +1., 5%, AenW: xF ( x )G 5 x ( F ( x )G ) , i&e& this part of the eQui#alence can be pro#ed constructi#ely& Gut, V+1-+11, +12, +1., 5%, AenW: x ( F ( x )G ) 5 xF ( x )G & This formula cannot be pro#ed constructi#ely; 2hy8 'ee 'ection .&/& Proof. %ro#e 0a, b, c1 as the @xercise /&1&( belo & +et us pro#e 0d1`: xF ( x )G 5 x ( F ( x )G ) & 011 021 0(1 0.1 0/1 0-1 0=1 F ( x ) 5 F ( x )G x ( F ( x ) 5 F ( x )G ) x0G`<1`0xG`x<1 xF ( x ) 5 x ( F ( x )G ) G 5 F ( x )G x ( G 5 F ( x )G ) G 5 x ( F ( x )G ) 6xiom +-& Gy Aen& Theorem (&1&10a1 V+1, +2, +12, +1., 5%, AenW& 9rom 021 and 0(1& 6xiom +=& Gy Aen& Gy 6xiom +1.&

1C-

0C1

xF ( x )G 5 x ( F ( x )G ) 9rom 0.1 and 0=1, by 6xiom +C&

9inally, let us pro#e 0d1j: x ( F ( x )G ) 5 xF ( x )G in the classical lo!ic 0a constructi#e proof is impossible, see 'ection .&/1& 011 021 0(1 0.1 x ( F ( x )G ) F ( x )G G F ( x ) fA Hypothesis& Gy 6xiom +12& 9rom 021& Hypothesis& Gy Theorem 2&/&10b1 V+1, +2, +C, +10, 5%W: UA B 5 ( 7 A5 B ) Gy Aen& Gy 6xiom +-& Gy Deduction Theorem 2 0x is not free #ariable in x ( F ( x )G ) & 6xiom +=& Gy 6xiom +1: G`0<`G1&

0.1

90x1

0/1 0-1

x90x1 xF ( x )G 7G 5 ( x ( F ( x )G ) 5 xF ( x )G ) G 5 xF ( x )G G 5 ( x ( F ( x )G ) 5 xF ( x )G )

0=1

0C1 0,1 0101 0111 D&@&D

9rom 0=1 and 0,1, by G7 G 5 ( x ( F ( x )G ) 5 xF ( x )G ) 6xiom + & C x ( F ( x )G ) 5 xF ( x )G Gy 37iom L'': G7 G &

!7ercise -.'.). %ro#e +emma /&1&20a, b, c1& Lemma -.'.). a1 V+1-+10, +12-+1/, 5%, AenW: fx90x1 a xf90x1&

1C= b1 V+1-+11, +12-+1/, 5%, AenW: f x90x1 a xf90x1& 5ore precisely: V+1-+11, +1(, +1., 5%, AenW: f x90x1 ` xf90x1& This formula cannot be pro#ed constructi#ely; 2hy8 'ee 'ection .&/& Gut, V+1-+10, +1(, +1., 5%, AenW: xf90x1 ` f x90x1& Proof. a1 'ee 'ection (&2, Aroup 7V& b1`& This is exactly 'ection (&2, 777-.& b1j& 'ee 'ection (&2, Aroup 777& D&@&D& +et us recall that a formula is in the prenex normal form, if and only if it has all its Quantifiers !athered in front of a formula that does not contain Quantifiers& 9heorem -.'.*. 7n the classical lo!ic, each formula is eQui#alent to an appropriate formula in the prenex normal form& 5ore precisely, if 9 is a formula, then, follo in! a simple al!orithm, a formula 9B can be constructed such that: a1 9B is in a prenex normal form, b1 9B has the same free #ariables as 9, c1 V+1-+11, +12-+1/, 5%, AenW: 9a9B& Proof. +et us start by an example: xA0x1`yH0y1& 7f H did not contain x as a free #ariable, then, by +emma /&1&10b1: x90x1`A a x090x1`A1, i&e& this formula ould be eQui#alent to x0A0x1`yH0y11& ?o , let us consider the sub-formula A0x1`yH0y1& 7f A did not contain y as a free #ariable, then, by +emma /&1&10c1: A`x90x1 a x0A`90x11, the subformula ould be eQui#alent to y0A0x1`H0y11& Hence, by )eplacement Theorem 2, x0A0x1`yH0y11 ould be eQui#alent to xy0A0x1`H0y11& Gut, if H ould contain x as a free #ariable, andPor A ould contain y as a free #ariable8 Then our Sshiftin! Quantifiers upS ould be ron! 3 the formula xy0A0x1`H0y11 ould not be eQui#alent to xA0x1`yH0y1& To a#oid this problem, let us use )eplacement Theorem (, hich says that the meanin! of a formula does not depend on the names of bound #ariables used in it& Thus, as the first step, in xA0x1, let us replace x by another #ariable x 1 that does not appear neither in A, nor in H& Then, by )eplacement Theorem (, xA0x1 is eQui#alent to x1A0x11, and by )eplacement Theorem 2, xA0x1`yH0y1 is eQui#alent to x1A0x11`yH0y1& ?o ,

1CC x10A0x11`yH0y11 is really eQui#alent to x1A0x11`yH0y1& 6s the next step, in yH0y1, let us replace y by another #ariable y 1 that does not appear neither in A, nor in H& Then, by )eplacement Theorem (, yH0y1 is eQui#alent to y1H0y11, and by )eplacement Theorem 2, A0x 11`y1H0y11 is eQui#alent to y10A0x11`H0y111& 6nd, finally, xA0x1`yH0y1 is eQui#alent to x1y10A0x11`H0y111& ?o , e can start the !eneral proof& 7n a formula 9, let us find the leftmost Quantifier ha#in! a propositional connecti#e o#er it& 7f such a Quantifier does not exist, the formula is in the prenex normal form& 7f such a Quantifier exists, then 9 is in one of the follo in! forms: DQDQ&&&DQ0&&&0fDxA1&&&1, or DQDQ&&&DQ0&&&0DxAooH1&&&1, or DQDQ&&&DQ0&&& 0AooDxH1&&&1, here DQDQ&&&DQ are the Quantifiers Salready in prefixS, D is the Quantifier in Question, and oo is the propositional connecti#e standin! directly o#er D& 7n the first case, by +emma /&1&(, fDxA is eQui#alent to DBxfA, here DB is the Quantifier opposite to D& Gy )eplacement Theorem 2, D QDQ&&&DQ0&&& 0fDxA1&&&1 is then eQui#alent to DQDQ&&&DQ0&&&0DBxfA1&&&1, i&e& DB has no one propositional connecti#e less o#er it 0 0than had D1& 7n the second case, as the first step, in DxA, let us replace x by another #ariable x1 that does not appear in the entire formula 9 at all& Then, by )eplacement Theorem (, DxA is eQui#alent to Dx1A1, and by )eplacement Theorem 2, DQDQ&&&DQ0&&&0DxAooH1&&&1 is eQui#alent to DQDQ&&&DQ0&&& 0Dx1A1ooH1&&&1& ?o , e can apply the appropriate case of +emma /&1&1 or +emma /&1&2, obtainin! that Dx1A1ooH is eQui#alent to DBx10A1ooH1, here DB is the Quantifier determined by the lemma applied& Then , by )eplacement Theorem 2, DQDQ&&&DQ0&&&0Dx1A1ooH1&&&1 is eQui#alent to DQDQ&&&DQ0&&& 0DBx10A1ooH11&&&1, i&e& DB has no one propositional connecti#e less o#er it 0than had D1& 7n the third case, the ar!ument is similar& Gy iteratin! this operation a finite number of times, e arri#e at a formula 9B hich is in the prenex normal form, and hich is 0in the classical lo!ic1 eQui#alent to 9& D&@&D& 4ote. 5ost formulas admit many different prenex normal forms& 9or example, the abo#e formula xA0x1`yH0y1 is eQui#alent not only to

1C, x1y10A0x11`H0y111, but also to y1x10A0x11`H0y111 0#erify1& 6s an example, let us obtain a prenex normal form of the follo in! formula: xB ( x ) xC ( x ) 5 x, ( x )( 7 xF ( x )) & 9irst, assi!n uniQue names to bound #ariables:
x 1 B ( x 1) x 2 C ( x 2) 5 x ( , ( x ()( 7 x . F ( x .)) &

%rocess dis>unction: x 1 x 2 ( B ( x 1 )C ( x 2 )) 5 x ( , ( x ( )( 7 x . F ( x .)) & %rocess ne!ation 0-1:


x 1 x 2 ( B ( x 1 )C ( x 2 )) 5 x ( , ( x ( ) x . 7 F ( x .) &

%rocess con>unction: x 1 x 2 ( B ( x 1 )C ( x 2 )) 5 x ( x . ( , ( x ( )7 F ( x . )) & %rocess implication premise 0-, - 1:


x 1 x 2 ( B ( x 1 )C ( x 2 ) 5 x ( x . ( , ( x ()7 F ( x .))) &

%rocess implication conseQuent: x 1 x 2 x ( x . ( B ( x 1)C ( x 2 ) 5 , ( x ()7 F ( x . )) & The last t o steps could be performed in the re#erse order as ell& !7ercise -.'.*. Transform each of the follo in! formulas into a prenex normal form& 2rite do n e#ery sin!le step of the process& 0Hint: the al!orithm is explained in the proof of Theorem /&1&.&1 a1 xB ( x ) 5 ( xC ( x ) 5 x, ( x )) , b1 x yB ( x , y ) xC ( x ) 5 y x, ( x , y ) , c1 xB ( x , y , 3 ) 5 xC ( x , y ) y, ( y , 3 ) , d1 xB ( x ) 5 ( xC ( x ) 5 ( x, ( x ) 5 xF ( x ))) , e1 (( xB ( x ) 5 xC ( x )) 5 x, ( x )) 5 xF ( x ) & 4ote. 9rom a pro!rammerBs point of #ie , prenex normal forms are, in a sense, a cra:y in#ention& 7n computer pro!rammin!, you al ays try to reduce loop bodies, not to extend them as much as possible; !7ercise -.'.-& 2e may use reduction to prenex normal forms in proofs& 5ore precisely, let us try extendin! the classical lo!ic by introducin! of the follo in! additional inference rule 0let us call it P4:Brule1: !i#en a formula 9, replace it by some its prenex normal form 9B& Verify, that, in fact, this rule does not extend the classical lo!ic, i&e& if there is a proof of 91, 92, &&&, 9n U- A in V+1-

1,0 +1/, 5%, Aen, %?9-ruleW, then there is a proof of the same in V+ 1-+1/, 5%, AenW& 07n some other texts, such rules are called admissible rules& Thus, the %?9-rule is an admissible rule in the classical lo!ic&1
The notion of prenex normal forms and a #ersion of Theorem /&1&. '& %eirce in 1CC/: ere kno n to

<harles

. ,. Peirce. Rn the al!ebra of lo!ic: 6 contribution to the philosophy of notation& American /o$rna% of .athematics, 1CC/, #ol&=, pp&1C0-202& 6s noted by 3lasdair Ur>uhart at http:PP &cs&nyu&eduPpipermailPfomP200=HulyP011=20&html: SRn pa!e 1,- of that article, he !i#es a brief sketch of con#ersion to prenex normal form, remarkin! that it Scan e#idently be done&SS&

-... ,kolem 4ormal :orm


This normal form 1,2C: as first introduced by Thoralf 'kolem 01CC=-1,-(1 in

9h.,kolem. nber die mathematische +o!ik& S?orsk matematisk tidsskriftS, 1,2C, #ol&10, pp&12/-1.2&

1arning6 The principal results of this 'ection are #alid only for the classical logic; The first #ery important idea as proposed by 'kolem already in 1,20:
9h. ,kolem. +o!isch-kombinatorische *ntersuchun!en gber die @rfgllbarkeit und Ge eisbarkeit mathematischen 'lt:e nebst einem Theoreme gber dichte 5en!en& Ji+ens&absa&a+emiet i Cristiania, S&rifter 7, ?o& ., 1,20, pp& 1-(-&

?amely, accordin! to 'kolemBs idea, further Snormali:ationS becomes possible, if e drop the reQuirement that the Snormal formS must be eQui#alent to the initial formula, and replace it by the reQuirement: <normal form< must be logically validY if and only if the initial formula is logically valid & 7t appears, that in this ay e can SreduceS any closed formula to a closed formula containin! only one kind of Quantifiers: x1x2&&&xnH0x1, x2, &&&, xn1, here H does not contain Quantifiers at all 0see Theorem /&2&. belo 1& 'till, in his ori!inal formulation, instead of lo!ical #alidity, 'kolem as interested in a more technical notion 3 satisfiability& +et us recall that, in a predicate lan!ua!e +, a formula 9 is called satisfiable, if and only if there is an interpretation of the lan!ua!e + such that 9 is true for some #alues of its free #ariables& 9or our current purpose 3 refutation "roofs 0to pro#e that 91, &&, 9n

1,1 U- A, e assume fA and try to deri#e a contradiction1 satisfiability orks as ell as does lo!ical #alidity& 7ndeed 0#erify, see @xercise .&1&11, a set of formulas is inconsistent, if and only if it is unsatisfiable& Thus, if, in a refutation proof, e replace some formula H by an SeQually satisfiableS formula HB 0i&e& HB is satisfiable, if and only if so is H1, then the refutation proof remains #alid& 7&e& if, this ay, e deri#e a contradiction from 9 1, &&, 9n,fA, then this set of fomulas is, indeed, unsatisfiable, i&e& A lo!ically follo s from 91, &&, 9n 0for a more precise #ersion of this ar!ument see @xercises /&2&.1& 'kolemBs second main idea 0proposed in his 1,2C paper1: allo8 introduction of ne8 ob@ect constants and function constants & 7t can be demonstrated on the follo in! example: ho could e SsimplifyS the formula xy 90x, y18 7t asserts that for each x there is y such that 90x, y1 is true& Thus, it asserts, that there is a function !, hich selects for each #alue of x a #alue of y such that 90x, y1 is true& Thus, in a sense, xy 90x, y1 is SeQui#alentS to x 90x, !0x11& 7n hich sense8 7n the sense that xy 90x, y1 is satisfiable, if and only if x 90x, !0x11 is satisfiable& 7ndeed, 1& 7f xy 90x, y1 is satisfiable, then there is an interpretation H here it is true, i&e& for each #alue of x there is a #alue of y such that 90x, y1 is true& This allo s us to define the follo in! interpretation of the function constant !: !0x1 is one of y-s such that 90x, y1 is true in H& 7f e extend H by addin! this interpretation of the function constant !, e obtain an interpretation HB, here x 90x, !0x11 is true, i&e& this formula is satisfiable& 2& 7f x 90x, !0x11 is satisfiable, then there is an interpretation H here it is true, i&e& for each #alue of x the formula 90x, !0x11 is true& Hence, in this interpretation, for each #alue of x there is a #alue of y 0namely, !0x11 such that 90x, y1 is true in H& Thus, xy 90x, y1 is true in H, i&e& this formula is satisfiable& 4ote& 7n the first part of this proof, to define the function !, e need, in !eneral, the 6xiom of <hoice& 7ndeed, if there is a non-empty set K x of y-s such that 90x, y1 is true, to define !0x1, e must choose a sin!le element of K x& 7f e kno nothin! else about the interpretation H, e are forced to use the 6xiom of <hoice& Gut, if e kno that the interpretation H has a countable domain, then e can define !0x1 as the SleastS y from the set K x& 7n this ay e can a#oid the 6xiom of <hoice& The third idea is e#en simpler: the formula x 90x1 asserts that there is x such that 90x1 is true, so, let us denote by 0an ob>ect constant1 c one of these x-s,

1,2 thus obtainin! 90c1 as a Snormal formS of x 90x1& Rf course 0#erify1, x 90x1 is satisfiable, if and only if 90c1 is satisfiable& These t o ideas allo Sreducin!S of any Quantifier prefix Dx 1Dx2&&&Dxn to a seQuence of uni#ersal Quantifiers only: 9heorem -...' C9h. ,kolemD& +et + be a predicate lan!ua!e& There is an al!orithm allo in! to construct, for each c%ose+ formula 9 of this lan!ua!e, a c%ose+ formula 9B 0in a lan!ua!e +B obtained from + by addin! a finite set of ne ob>ect constants and ne function constants 3 dependin! on 91 such that: a1 9B is satisfiable, if and only if 9 is satisfiable, b1 9B is in form x1x2&&&xnA, here nd0, and A does not contain Quantifiers& 7f a formula is in form x1x2&&&xnA, here nd0, and A does not contain Quantifiers, let us call it ,kolem normal form& Thus, each closed formula can be reduced to a 'kolem normal form in the follo in! sense: for each closed formula 9 of a lan!ua!e + there is a 'kolem normal form U9U 'k 0in the lan!ua!e + extended by a finite set of 'kolem constants and 'kolem functions1, hich is satisfiable, if and only if so is 9& 4ote& 7n computer science slan!, the reduction procedure leadin! to 'kolem normal form is called Sskolemi;ationS& 4ote. Theorem /&2&1 does not assert that a formula and its 'kolem normal form are eQui#alent& 7t asserts only that the satisfiability "roblem of the first formula is eQui#alent to the satisfiability problem of the second formula& 6s already mentioned abo#e, this is enou!h to allo usin! of 'kolem reduction in refutation proofs& Thus, if e are interested in determinin! the satisfiability of formulas, then reducin! to 'kolem normal forms is a promisin! method& 7ndeed, formulas x1x2&&&xnA 0 here A does not contain Quantifiers1 are, perhaps, easier to analy:e than more complicated combinations of Quantifiers& Proof of 9heorem -...' 9irst, let us obtain a prenex normal form 9 1 of the formula 9 0see 'ection /&11& 7ndeed, by Theorem /&1&., there is a simple al!orithm, allo in! to construct a closed formula 91 such that 91 is a prenex normal form, and, in the classical lo!ic, U- 9a91& Rf course, 91 is satisfiable4 if and only if so is 9& 7f the Quantifier prefix of 91 starts ith a seQuence of existential Quantifiers 0&&&&&&1, e ill need the follo in! lemma to SreduceS these Quantifiers: Lemma -.... & 6 closed formula x1x2&&&xn H0x1, x2, &&&, xn1 is satisfiable, if

1,( and only if H0c1, c2, &&&, cn1 is satisfiable, constants that do not appear in H& here c1, c2, &&&, cn are ne ob>ect

6fter this operation, e ha#e a closed prenex formula H0c 1, c2, &&&, cn1 0in a lan!ua!e obtained from + by addin! a finite set of ne ob>ect constants, called ,kolem constants1, hich is satisfiable, if and only if so is 9 1 0and 91& The the Quantifier prefix of H0c1, c2, &&&, cn1 0if any1 starts ith a seQuence of uni#ersal Quantifiers 0 &&&&&&1& To proceed, e ill need the follo in! Lemma -...)& 6 closed formula x1x2&&&xny$0x1, x2, &&&, xn, y1 is satisfiable, if and only if x1x2&&&xn$0x1, x2, &&&, xn, !0x1, x2, &&&, xn11 is satisfiable, here ! is a ne n-ary function constant 0called ,kolem function1, hich does not appear in $& Gy iteratin! this lemma, e can SreduceS the entire Quantifier prefix of H0c1, c2, &&&, cn1 to a seQuence of uni#ersal Quantifiers only 0 &&&1& 9or example, the formula t xy:u 90t, x, y, :, u, 1 is satisfiable, if and only if so is xyu 90c, x, y, !0x, y1, u, 1 0 here c is a 'kolem constant that does not appear in 91, and, if and only if so is xyu 90c, x, y, !0x, y1, u, 1, and, if and only if so is the 'kolem normal form: xyu 90c, x, y, !0x, y1, u, h0x, y, u11, here ! and h are 'kolem functions that do not appear in 9& !7ercise -...'. a1 %ro#e +emma /&2&2& b1 %ro#e +emma /&2&(& Ho many ne ob>ect constants and ne function constants 0'kolem constants and functions1 do e need to obtain the final formula 9B8 The number of ne symbols is determined by the number of existential Quantifiers in the Quantifier prefix of the prenex formula 9 1& 7ndeed, a1 the number of ne ob>ect constants is determined by the number of existential Quantifiers in front of the prefix, and b1 the number of ne function constants is determined by the number of existential Quantifiers that follo after the uni#ersal ones& This completes the proof of Theorem /&2&1& !7ercise -....& Rbtain 'kolem normal forms of the formulas mentioned in @xercise /&1&.&

1,. 'ee also: S'kolemi:ationS from The 2olfram Demonstrations %ro>ect& <ontributed by: Hector Lenil& 'till, if e are interested in determinin! the lo!ical #alidity of formulas, then e should apply the result of @xercise .&1&1 to!ether ith Theorem /&2&1: 9 is lo!ically #alid, if and only if f9 is not satisfiable, if and only if a 'kolem normal form of f9 is not satisfiable, if and only if x1x2&&&xnA 0 here nd0, and A does not contain Quantifiers1 is not satisfiable, if and only if fx1x2&&&xnA is lo!ically #alid, if and only if x1x2&&&xnfA is lo!ically #alid& Thus e ha#e pro#ed the follo in! 9heorem -...*& +et + be a first order lan!ua!e& There is an al!orithm allo in! to construct, for each closed formula 9 of this lan!ua!e, a closed formula 9B 0in a lan!ua!e +B obtained from + by addin! a finite set of ne ob>ect constants and ne function constants 3 dependin! on 91 such that: a1 9B is lo!ically #alid 0or, pro#able in the classical lo!ic1, if and only if 9 is lo!ically #alid 0or, pro#able in the classical lo!ic1, b1 9B is in form x1x2&&&xnA, here nd0, and A does not contain Quantifiers& ,kolem 4ormal :orm of a ,et of :ormulas $no led!e bases are, as a rule, lar!e sets of c%ose+ formulas 91, 92, &&&, 9n, i&e&, in fact, lar!e con>unctions F 1 F 2&&& F n of closed formulas& <ould e obtain a 'kolem normal form of this con>unction simply by reducin! to 'kolem normal form each formula separately8 6ssume that durin! the entire process of reducin! the formulas 91, 92, &&&, 9n to their 'kolem normal forms 9B1, 9B2, &&&, 9Bn , these formulas are Mkept separatedN, i&e& the name of each ne 'kolem constant and 'kolem function is chosen as Mcompletely ne N ith respect to the entire process& Gy examinin! carefully the proof of Theorem /&2&1, one can see that this is enou!h to !uarantee that the con>unction F G 1 F G 2&&& F G n is satisfiable, if and only if so is F 1 F 2&&& F n & !7ercise -...) 0for smart students1& 7n his abo#e-mentioned 1,20 paper, for Quantifier elimination, 'kolem proposed introduction of ne8 "redicate constants 0to the idea that function constants ill do better, he arri#ed only in

1,/ the 1,2C paper1& Do not read neither 'kolemBs papers, nor the abo#ementioned online comments, and pro#e yourself that by introduction of ne predicate constants, the satisfiability problem of any closed formula can be reduced to the satisfiability problem of a formula ha#in! the form x1x2&&&xmy1y2&&&ynA, here m, nd0, and A does not contain Quantifiers& Thus, function constants S ill do betterS 3 see Theorem /&2&1& !7ercise -...* 0compare ith @xercise /&1&/1& 'ince, in !eneral, 'kolem normal form is not eQui#alent to the initial formula, e cannot use reduction to 'kolem normal forms in the usual 0Spositi#eS, or affirmati#e1 proofs& Gut e may use it in Sne!ati#eS 0or, refutation1 proofs, i&e& in proofs aimed at deri#in! a contradiction; 5ore precisely, let us try extendin! the classical lo!ic by introducin! of the follo in! additional inference rule 0let us call it ,4:Brule1: !i#en a formula 9, replace it by some its 'kolem normal form 9B 0such that the ne ly introduced ob>ect constants and function constants do not appear in the proof before 9B1& Verify, that, in fact, this rule does not extend the classical lo!ic for refutation proofs, i&e& if, from a set of formulas 9 1, 92, &&&, 9n, one can deri#e a contradiction by usin! V+1-+1/, 5%, Aen, '?9-ruleW, then one can do the same by usin! V+1-+1/, 5%, AenW& 0Thus, the '?9-rule is admissible for refutation "roofs in the classical lo!ic&1

-.). on@unctive and Dis@unctive 4ormal :orms


1arning6 The principal results of this 'ection are #alid only for the classical logic; +et us continue the Snormali:ationS process that e started in 'ection /&1 by reducin! formulas to their prenex normal forms, here all Quantifiers are !athered in front of a formula that does not contain Quantifiers& Ho could e further Snormali:eS this Sformula that does not contain QuantifiersS8 ,te" '= eliminate e>uivalence 9irst of all, e can eliminate all eQui#alence connecti#es because Ga< is only a shortcut for ( B 5 C )( C 5 B ) & 2hy should e8 Gecause, pro#in! of Ga< consists of pro#in! of G`< and pro#in! of <`G& *sin! the shortcut simplifies the appearance of the formula, not its proof& ,te" .= eliminate im"lication 6fter this, our formula ill contain only implication, con>unction, dis>unction and ne!ation connecti#es& 6s the next step, e could try to eliminate one 0or

1,t o81 of these connecti#es& The classical lo!ic allo s to do that& 9or example, by Theorem 2&-&.0b1, V+1-+11, 5%W: U- ( A 5 B ) 6 7 A B & Gy usin! this eQui#alence, e can eliminate implication connecti#es& 9or example, the formula G`0<`D1 is eQui#alent 0in the classical lo!ic only;1 to 7 B ( 7 C , ) & Gut, instead of implications, con>unctions as ell& 7ndeed, a1 U- ( A 5 B ) 6 7 ( A7 B ) & b1 U- ( A B ) 6 ( 7 A 5 B ) & c1 U- ( A B ) 6 7 ( 7 A7 B ) & d1 U- ( A B ) 6 7 ( A 5 7 B ) & e1 U- ( A B ) 6 7 ( 7 A7 B ) & 09or smart students1 Determine, hich parts of these eQui#alences can be pro#ed in the constructi#e lo!ic V+1-+10, 5%W& @nd of @xercise /&(&1& Gy usin! these results, e could eliminate from our formulas any one 0or any t o1 of the three connecti#es 3 implication, con>unction, or dis>unction& Ho e#er, the best decision ould be eliminatin! only implications& 2hy8 Gecause con>unction and dis>unction are associati#e and commutati#e operations 3 and #ery much like addition 0dis>unction1 and multiplication 0con>unction1; 9or example, after reducin! the formula G`0<`G1 to 7 B ( 7 C B ) , e can further transform it to 7 B 7C B and ( 7 B B )C 3 and conclude that it is Strue and pro#ableS 0no surprise, it is 6xiom +11& ,te" )= move negations do8n to atoms Thus, after 'tep 2, our formula contains only con>unction, dis>unction and ne!ation connecti#es& ?o , let us recall the t o de 5or!an +a s: Theorem 2&-&(, V+1-+11, 5%W: U- 7 ( A B ) 6 7 A7 B & Theorem 2&.&100b1, V+1-+,, 5%W U- 7 ( A B ) 6 7 A7 B & Gy usin! these eQui#alencies, e can shift ne!ations do n 3 until the atoms of the formula& 9or example, let us transform the formula (( A5 B ) 5 C ) 5 BC & e could try eliminatin! dis>unctions or

!7ercise -.).'& 7n the classical lo!ic V+1-+11, 5%W, pro#e the follo in!:

1,= 9irst, eliminate implications: 7 (( A5 B ) 5 C )( B C ) , 7 ( 7( A 5 B)C )( B C ) , 7 ( 7( 7 A B )C )( B C ) & 6pply de 5or!an +a s: ( 77 (7 A B )7 C )( B C ) , ( 7 ( 77 A7 B)7C )( BC ) , (( 7 77 A77 B )7C )( B C ) & ?o , let us recall the Double ?e!ation +a : Theorem 2&-&1, V+1-+11, 5%W: U- ff6 a 6& 7t allo s droppin! the excessi#e ne!ations 3 ffG 3 by G: e can replace fff6 by f6 and

(( 7 A B)7C )( BC ) & 4ote. This form of formulas is called negation normal form& ?amely, a formula is in ne!ation normal form, if it is built of atoms ith or ithout ne!ations by usin! con>unctions and dis>unctions only& 7&e& a formula in ne!ation normal form contains only con>unctions, dis>unctions and ne!ations, and ne!ations are located at the atoms only& 6s e see, in the classical lo!ic, any propositional formula can be reduced 0is eQui#alent1 to some formula in ne!ation normal form& ?e!ation normal form is the startin! point for an alternati#e 0to the )esolution method described in this 'ection /1 method of automated theorem-pro#in! 3 the so-called 5ethod of 6nalytic Tableaux& Rne does not use skolemi:ation here, one simply obtains the ne!ation normal form of the formula 0 ith Quantifiers inside1 and after this, applies a specific tree al!orithm of the Tableaux method& 05y exposition for students, in +at#ian: Tablo al!oritms&1 !nd of 4ote. ,te" *= algebra 6fter 'tep (, our formula is built up by usin!: a1 atoms, b1 atoms preceded by a ne!ation, c1 con>unction and dis>unction connecti#es& <on>unction and dis>unction are associati#e and commutati#e operations& Gy the beha#ior of Struth #aluesS, con>unction is a kind of multiplication:
0 0= 0,0 1=1 0=0, 11=1 ,

1,C and dis>unction 3 a kind of addition: 0 0= 0,0 1=1 0=1, 11 =1 & Ho e#er, for these operations t8o distributi#e la s are #alid 0Theorem 2&(&11 3 con>unction is distributi#e to dis>unction, and dis>unction is distributi#e to con>unction: V+1-+C, 5%W: U- ( A B )C 6 ( A C )( B C ) , V+1-+C, 5%W: U- ( A B )C 6 ( A C )( B C ) & Thus, both of the t o decisions could be >ustified: 11 0Rur first Sal!ebraS1 +et us treat con>unction as multiplication and dis>unction 3 as addition 0\1& Then the abo#e formula (( 7 A B)7C )( BC ) takes the form 006B\G1<B1\G< 0let us replace f6 by the Smore al!ebraicS 6B1& 6fter this, the usual al!ebraic transformations yield the formula 6B<B\G<B\G<& 21 0Rur second Sal!ebraS1 +et us treat con>unction as addition 0\1 and dis>unction 3 as multiplication& Then the abo#e formula (( 7 A B)7C )( BC ) takes the form 06BG\<B10G\<1& 6fter this, the usual al!ebraic transformations yield the formula 6BGG\6BG<\<BG\<B<& 6dditional rules can be applied in these Sal!ebrasS& 9irst rule 3 con>unction and dis>unction are idempotent operations: V+1- +/, 5%W: U- A A 6 A 0see 'ection 2&21& V+1, +2, +/, +--+C, 5%W: U- A A 6 A 0@xercise 2&(&10c11& Thus, in both of our Sal!ebrasS: 3Z3 W 33 W 3& 'econd rule 3 A7 A 0i&e& SfalseS1 is a kind of S:eroS in the first Sal!ebraS, and a kind of SoneS 3 in the second Sal!ebraS: V+1-+10, 5%W: U- B( A7 A) 6 B 0@xercise 2&/&10a11, V+1-+10, 5%W: U- (( A7 A) B )C 6C 0@xercise 2&/&10b11& 7ndeed, in the first Sal!ebraS, these formulas mean G\66B ^ G and 66BG\< ^ <, i&e& e may think that 66B^0, G0^0, <\0^<& 7n the second Sal!ebraS, these formulas mean G06\6B1 ^ G and 06\6B\G1< ^ <, i&e& e may think that 6\6B^1, G1^G, <\1^1& Third rule 3 A7 A 0i&e& StrueS1 is a kind of SoneS in the first Sal!ebraS, and a kind of S:eroS 3 in the second Sal!ebraS: V+1-+11, 5%W: U- B( A7 A) 6 B 0@xercise 2&-&20a11,

1,, V+1-+11, 5%W: U- (( A7 A) B )C 6C 0@xercise 2&-&20b11&

7ndeed, in the first Sal!ebraS, these formulas mean G06\6B1 ^ G and 06\6B\G1< ^ <, i&e& e may think that 6\6B^1, G1^1, <\1^1& 7n the second Sal!ebraS& these formulas mean G\66B ^ G and 66BG\< ^ <, i&e& e may think that 66B^0, G0^0, <\0^<& Thus, in both algebras, 33JWI, %IWI, ZIW , 3Z3JW', %'W%, Z'W'. 'o, let us continue our example 11 0The first Sal!ebraS1 The formula 6B<B\G<B\G< is eQui#alent to 6B<B\G0<B\<1 ^ 6B<B\G, or, if e return to lo!ic: ( 7 A7C ) B & 'uch dis>unctions consistin! of con>unctions are called dis@unctive normal forms 0D?9s1& 7n a D?9, each con>unction contains each atom no more than once 3 either ithout ne!ation, or ith it& 7ndeed, if it contains some atom J t ice, then: a1 replace JJ by J, or b1 replace JBJB by JB, or c1 replace JJB by 0 0in the latter case 3 drop the entire con>unction from the expression1& 7n this ay, for some formulas, e may obtain S:eroS, i&e& an em"ty D4:& Rf course, such formulas take only false #alues 0SfalseS is S:eroS in the first Sal!ebraS1& 6nd for some formulas, e may obtain SoneS, i&e& a kind of <full< D4:& 'uch formulas take only true #alues 0StrueS is SoneS in the first Sal!ebraS1& 21 0The second Sal!ebraS1 The formula 6BGG\6BG<\<BG\<B< is eQui#alent to 6BG\6BG<\G<B ^ 6BG01\<1\G<B ^ 6BG\G<B, or, if e return to lo!ic: ( 7 A B )( B7 C ) & 'uch con>unctions consistin! of dis>unctions are called con@unctive normal forms 0<?9s1& 7n a <?9, each dis>unction contains each atom no more than once 3 either ithout ne!ation, or ith it& 7ndeed, if it contains some atom J t ice, then: a1 replace JJ by J, or b1 replace JBJB by JB, or c1 replace JJB by 0 0in the latter case 3 drop the entire dis>unction from the expression1& 7n this ay, for some formulas, e may obtain S:eroS, i&e& an em"ty 4:& Rf course, such formulas take only true #alues 0StrueS is S:eroS in the second Sal!ebraS1& 6nd for some formulas, e may obtain SoneS, i&e& a kind of <full< 4:& 'uch formulas take only false #alues 0SfalseS is SoneS in the second Sal!ebraS1& Thus, e ha#e pro#ed the follo in! 9heorem -.).'. 7n the classical lo!ic, e#ery propositional formula can be reduced to D?9 and to <?9& 5ore precisely, assume, the formula 9 has been built of formulas G1, G2, &&&, Gn by usin! propositional connecti#es only& Then: a1 There is a formula 91, hich is in a 0possibly empty or full1 dis>uncti#e normal form o#er G1, G2, &&&, Gn such that V+1-+11, 5%W: U- 9 a 91&

200 b1 There is a formula 92, hich is in a 0possibly empty or full1 con>uncti#e normal form o#er G1, G2, &&&, Gn such that V+1-+11, 5%W: U- 9 a 92& !7ercise -.)..& a1 Guild D?9s and <?9s of the follo in! formulas& 0Hint: the al!orithm is explained in the abo#e 'teps 1-.&1 7 ( A B 5 C ) , ( A 5 B )6 (C 5 ,) , A B 6C , , A B 6C , & b1 Guild D?9s and <?9s of the follo in! formulas: 7 ( A 7 A ) , (( A5 B ) 5 A )5 A , ( A 5 B ) 5 (( 7 A 5 B) 5 B ) &
The notion of dis>uncti#e normal form as kno n in 1CC( to Rscar Ho ard 5itchell 01C/11CC,1: ?scar $o8ard Mitchell& Rn a ?e 6l!ebra of +o!ic& 7n: St$+ies in Logic by .embers of the /ohns Hop&ins 2ni'ersity, 1CC/, pp& =2-10-&

-.*. lause :orm


1arning6 The principal results of this 'ection are #alid only for the classical logic; lause :orms of Pro"ositional :ormulas 2hich form is more SnaturalS 3 D?9, or <?98 Rf course, <?9 is more natural& 7ndeed, a D?9 , 1 , 2&&& , m asserts that one 0or more1 of the formulas Di is true& This is a #ery complicated assertion 3 sometimes D1 is true, sometimes D2 is true, etc& Gut, if e ha#e a <?9 instead 3 C 1C 2&&& C n 8 7t asserts that all the formulas <i are true, i&e& e can replace a lon! formula C 1C 2&&& C n by a set of shorter formulas < 1, <2, &&&, <n& 9or human readin! and for computer processin!, a set of shorter formulas is much more con#enient than a sin!le lon! formula& +et us return to our example formula (( A5 B ) 5 C ) 5 BC of 'ection /&(, for hich e obtained a D?9 ( 7 A7C ) B and a <?9: ( 7 A B )( B7 C ) &

201 2ithout a transformation, D?9 may be hard for readin! and understandin!& The <?9 is more con#enient 3 it says simply that 7 A B is true and B7 C is true& 6s another step, makin! the formulas easier to understand, e could apply the follo in! eQui#alences: V+1-+11, 5%W: U- 7 A B 6 A 5 B , V+1-+11, 5%W: U- 7 A7 B C 6 A B 5 C , V+1-+11, 5%W: U- 7 A BC 6 A5 B C , V+1-+11, 5%W: U- 7 A7 B C , 6 A B 5C , , etc& !7ercise -.*.'. Verify these eQui#alences by pro#in! that, !enerally 0in the classical lo!ic1, V+1-+11, 5%W: U- 7 A1 7 A2&&&7 Am B 1 B2 &&& B n a
( A1 A2&&& Am 5 B1 B 2&&& B n) &

Thus, e can replace our set of t o formulas 7 A B , B7C by the set A 5 B ,C 5 B & The con>unction of these t o formulas is eQui#alent to the initial formula (( A5 B ) 5 C ) 5 B C & 9ormulas ha#in! the form A1 A2&&& Am 5 B1 B 2&&& B n , or, alternati#ely, 7 A1 7 A2 &&&7 Am B 1 B2&&& Bn , here 61, 62, &&& , 6m, G1, G2, &&& , Gn are atoms, are called clauses& <lauses are ell suited for computer processin!& 7ndeed, in the computer memory, e can represent the abo#e formula simply as a pair of sets of atoms 3 the ne!ati#e set X61, 62, &&& , 6mY and the positi#e set XG1, G2, &&& , GnY& 2hat, if one 0or both1 of these sets is 0are1 empty8 7f, in the formula 7 A1 7 A2 &&&7 Am B 1 B2&&& Bn , e ha#e m ^ 0 and n b 0, then, of course, this formula asserts simply that B1 B 2&&& Bn , i&e& Scon#ertin!S it into an implication ith empty premise 5 B1 B 2&&& Bn leads us to the follo in! definition: the clause 5 B1 B 2&&& Bn means the same as B1 B 2&&& Bn &

202 7f, in the formula 7 A1 7 A2 &&&7 Am B 1 B2&&& Bn , e ha#e m b 0 and n ^ 0, then, of course, this formula asserts simply that 7 A1 7 A2&&&7 Am , i&e& Scon#ertin!S it into an implication ith empty conseQuence A1 A2&&& Am 5 leads us to the follo in! definition: the clause A1 A2&&& Am 5 means the same as ( A1 A2&&& Am) & 7f m^n^0, then, as an empty dis>unction, the clause must be Qualified as false& 4ote. <lauses are similar to seQuents 3 pairs of sets of formulas 0' 1, '21, used in the proof of Theorem .&.&/ 0completeness of the constructi#e propositional lo!ic1 in 'ection .&.& 7n a seQuent 0'1, '21, the sets '1, '2 could contain arbitrary formulas, but, in a clause, '1, '2 are sets of atoms& 'ets 0i&e& con>unctions1 of clauses are called clause forms 0in some texts 3 c%a$sa% forms1& Gy Theorem /&(&1, e#ery propositional formula can be reduced to a 0possibly empty, i&e& true1 <?9& 'ince e#ery con>uction member of a <?9 represents, in fact, a clause, e ha#e established the follo in! 9heorem -.*.'& 7n the classical lo!ic, e#ery propositional formula can be reduced to a clause form& 5ore precisely, assume, the formula 9 is built of formulas G1, G2, &&&, Gn by usin! propositional connecti#es only& Then there is a 0possibly empty1 clause form 9B 0i&e& a set of clauses1 o#er G1, G2, &&&, Gn such that V+1-+11, 5%W: 9 a con>09B1, here con>09B1 denotes the con>unction of the clauses contained in the set 9B& 9or example, as e established abo#e, the set 7 A B , B 7C 0or, alternati#ely, A 5 B ,C 5 B 1 is a clause form of the formula (( A5 B ) 5 C ) 5 BC & !7ercise -.*..& Rbtain clause forms of the formulas mentioned in the @xercise /&(&2& <lause forms 0in a sense, Mclouds of simple dis>unctionsN1 are computer processin!& 7n the computer memory, e#ery clause
7 A1 7 A2 &&&7 Am B 1 B2&&& Bn

ell suited for

can be represented as a pair of sets of atoms: 0EX61, 62, &&&, 6mY, \XG1, G2, &&&, GnY1, and e#ery clause form 3 as a set of such pairs 3 i&e& it means less character strin! processin! and less expression parsin!;

20(

lause :orm of a ,et of :ormulas 7n the kno led!e base, the set of formulas 91, 92, &&&, 9k is assertin! the con>uction F 1 F 2&&& F & & Hence, the clause form of this set can be obtained simply as the union of clause forms of separate formulas 9i& lause :orms of Predicate :ormulas Rf course 0unfortunately1, if e ould insist that the clause form must be eQui#alent to the initial formula, then nothin! comparable to clause forms ould be obtained for predicate formulas& 'till, reducin! of predicate formulas to Sclause formsS becomes possible, if e drop this reQuirement, and replace it by the reQuirement that the Sclause formS must be satisfiable, if and only if the initial formula is satisfiable& 6nd 3 if e allo 'kolemBs style extendin! of the lan!ua!e by addin! ne ob>ect constants and ne function constants& Then, by 'kolemBs Theorem 0Theorem /&2&11, for each closed formula 9, e can obtain a ,kolem normal form x1x2&&&xk A, here kd0, the formula A does not contain Quantifiers, and this form is satisfiable, if and only if so is 9& 6s the next step, by Theorem /&.&1, let us con#ert A into a 4:, and then 3 into a clause form AB, i&e into a set of clauses 0 ith atomic sub-formulas of A playin! the role of atoms G1, G2, &&&, Gn1& 'ince con>0AB1 is eQui#alent to A, the formula x1x2&&&xkcon>0AB1 is satisfiable, if and only if so is 9& Rne more step is necessary to separate clauses completely 3 renamin! of #ariables in such a ay that no t8o clauses contain common variables& 9or the set of clauses AB ^ X< 1, <2, &&&, <kY, the formula x1x2&&&xncon>0AB1 is eQui#alent to the formula
( x 1 x 2 &&& x n C 1)( x 1 x 2 &&& x n C 2)&&& ( x 1 x 2 &&& x n C & ) &

6ccordin! to the )eplacement Theorem (, e ill obtain an eQui#alent formula, if e ill rename the #ariables xi in such a ay that no t o clauses contain common #ariables& 6fter this separation of clauses #ia renamin! of #ariables, e drop Quantifiers entirely, and the set AB is then called a clause form of the formula 9& 9or predicate formulas, clauses are built as dis>unctions of atomic formulas 0 ithout, or ith ne!ation1, i&e& the formulas ha#in! the form p0t1, &&&, tm1, here p is a predicate constant, and t 1, &&&, tm are terms 0possibly, containin! #ariables1&

20. Thus, e ha#e pro#ed the follo in! 9heorem -.*..& +et + be a predicate lan!ua!e& There is an al!orithm allo in! to construct, for each c%ose+ formula 9 of this lan!ua!e, a clause form ', i&e& a finite set of clauses 0in a lan!ua!e +B obtained from + by addin! a finite set of ne ob>ect constants and ne function constants 3 dependin! on 91 such that 9 is satisfiable, if and only if so is the 0closed1 formula x1x2&&&xncon>0'1, here con>0'1 denotes the con>unction of the clauses contained in ', and x 1, x2, &&&, xn are all the #ariables appearin! in the clauses contained in '& 4ote. 7n most texts, the closed formula x1x2&&&xncon>0'1 0i&e& here all the #ariables appearin! in con>0'1 are uni#ersally Quantified1 is called the universal closure of con>0'1& 6s an example, let us consider the formula assertin! that there are infinitely many prime numbers: prime ( x ) : x >1 7 y 3 ( y >1 3 >1 x = y 3 ) , $ x ( x >$ prime ( x )) , $ x ( x >$ x >1 7 y 3 ( y >1 3 >1 x = y3 )) <on#ert it into a prenex normal form: $ x ( x >$ x >1 y 3 7 ( y >1 3 >1 x = y 3 )) , $ x y 3 ( x > $ x >17 ( y >1 3 >1 x = y 3 )) & )eplace ux by u by introducin! a 'kolem function !: $ y 3 ( g ($ )>$ g ( $ )>17 ( y >1 3 >1 g ( $ )= y3 )) & 7n this 'kolem normal form, con#ert the Quantifier-free part into a con>uncti#e normal form: $ y 3 ( g ( $ )>$ g ( $ )>1 ( 7 ( y >1 )7 ( 3 >1)7( g ( $ )= y3 ))) & 9his formula is satisfiable, if and only if so is the initial formula C'D. The last step: since the last formula is eQui#alent to the con>unction of three formulas: $ y 3 ( g ($ )>$ ) , $ y 3 ( g ( $ )>1) , $ y 3 ( 7 ( y >1 )7 ( 3 >1 )7 ( g ( $ )= y 3 )) , e can rename the #ariables in such a common #ariables: ay that no t o clauses contain 011

$ 1 ( g ( $1 )>$ 1) , $ 2 ( g ( $ 2)>1 ) , $ ( y 3 ( 7( y >1)7( 3 >1 )7 ( g ( $( )= y 3 )) &

20/ Thus, e ha#e obtained a set of ( clauses: g ( $1 )>$ 1 , g ( $2 )>1 , 7 ( y >1 )7 ( 3 >1 )7 ( g ( $( )= y 3 ) & or, alternati#ely,
5 g ( $1)>$ 1 , 5 g ( $2 )>1 , y >1, 3 >1, g ( $( )= y 3 5 &

These sets of ( formulas are clause forms of the formula 011& !7ercise -.*.)& Rbtain clause forms of the formulas mentioned in the @xercise /&1&. 0assume that G, <, D, 9 are predicate constants1& lause :orm of a ,et of :ormulas $no led!e bases are, as a rule, lar!e sets of c%ose+ formulas 91, 92, &&&, 9n, i&e&, in fact, lar!e con>unctions F 1 F 2&&& F n of closed formulas& The clause form of this set can be obtained, simply as the union of clause forms of separate formulas 9i& Ho e#er, each formula must be Mkept separatedN durin! the entire process: a1 hen reducin! to 'kolem normal forms, the name of each ne 'kolem constant and 'kolem function must be chosen as Mcompletely ne N ith respect to the entire process 0for details, see the end of 'ection /&214 b1 hen renamin! clause #ariables, one must !uarantee that no t o clauses of the entire process contain common #ariables&

$orn lauses 6lfred Horn 01,1C-20011& 7n, in a clause A1 A2&&& Am 5 B1 B 2&&& B n , or, alternati#ely,
7 A1 7 A2 &&&7 Am B 1 B2&&& Bn ,

e ha#e n^1 or n^0, then it is called Horn clause& 7&e&, A1 A2&&& Am 5 B , or, alternati#ely,

207 A1 7 A2&&&7 Am B & There are formulas that cannot be reduced to Horn clauses 0#erify1&
http:PP ombat&doc&ic&ac&ukPfoldocPfoldoc&c!i8Horn\clause The name SHorn <lauseS comes from the lo!ician 6lfred Horn, ho first pointed out the si!nificance of such clauses in 1,/1, in the article SRn sentences hich are true of direct unions of al!ebrasS, Hournal of 'ymbolic +o!ic, 1-, 1.-21& http:PP &cs&ucsd&eduPusersP!o!uenPcoursesP2(0Ps11&html 6s a footnote, 6lfred Horn, for hom Horn clauses are named, had nothin! to do ith lo!ic pro!rammin!4 he as a professor of lo!ic at *<+6 ho in 1,/1 rote paper usin! the sentences that no bear his name for reasons ha#in! little to do ith computer science& 6s a second footnote, it seems to me rather misleadin! to call %rolo! a Slo!ic pro!rammin!S lan!ua!e, since it departs rather far from lo!ic4 7 ould rather ha#e had it called a Srelational pro!rammin!S lan!ua!e, because it is the use and manipulation of relations that is most characteristic of its pro!rammin! style& http:PP &cs&fit&eduPfryanPstudyPbiblio!raphy&html Horn, 6lfred& kkRn sentences hich are true of direct unions of al!ebras&BB Hournal of 'ymbolic +o!ic, #olume 1-, number 1, 5arch 1,/1, pa!es 1.-21& This paper has #ery little to do ith Horn clauses& To be continued&

-.-. #esolution Method for Pro"ositional :ormulas


1arning6 The principal results of this 'ection are #alid only for the classical logic; )emember, that e are sol#in! the problem of determinin!, does the formula A follo from the formulas 9 1, 92, &&&, 9n& 7f does so, if and only if the set of formulas 91, 92, &&&, 9n, fA is unsatisfiable& 6ssume, e ha#e obtained a clause form ' of the formula F 1 F 2 &&& F n7G & Then ' is unsatisfiable, if and only if so is the set 91, 92, &&&, 9n, fA& Ho to determine, is ' unsatisfiable, or not8 7n a sense, ' represents Ma cloud of simple dis>unctionsN& Ho to ork ith such a cloud effecti#ely8 $istory
K. 3. #obinson. Theorem-pro#in! on the computer& SHour& 6ssoc& <omput& 5ach&S, #ol&10, ?2, 1,-(, pp&1-(-1=. K. 3. #obinson. 6 machine-oriented lo!ic based on the resolution principle, SHour& 6ssoc& <omput& 5ach&S, #ol&12, ?1, Hanuary 1,-/, pp&2(-.1 0a#ailable online, )ussian translation

20=
a#ailable: S$ib& sbornik 0no#aya seriya1S, =, 1,=0, pp&1,.-21C1 Kohn 3lan #obinson= SGorn in Korkshire in 1,(0, )obinson came to the *nited 'tates in 1,/2 ith a classics de!ree from <ambrid!e *ni#ersity& He studied philosophy at the *ni#ersity of Rre!on before mo#in! to %rinceton here he recei#ed his %hD in philosophy in 1,/-& Temporarily kkdisillusioned ith philosophy,kk he ent to ork as an operations research analyst for Du %ont, here he learnt pro!rammin! and tau!ht himself mathematics& )obinson mo#ed to )ice *ni#ersity in 1,-1, spendin! his summers as a #isitin! researcher at the 6r!onne ?ational +aboratoryBs 6pplied 5athematics Di#ision& 7ts then Director, 2illiam 9& 5iller, pointed )obinson in the direction of theorem pro#in!&&& 5iller sho ed )obinson a 1,-0 paper by 5artin Da#is and Hilary %utnam 0coincidentally, the latter had been )obinsonBs %hD super#isor1 proposin! a predicate-calculus proof procedure that seemed potentially superior to AilmoreBs, but hich they had not yet turned into a practical computer pro!ram& 5iller su!!ested that )obinson use his pro!rammin! skills to implement Da#is and %utnamBs procedure on the 6r!onne 7G5 =0.& )obinson Quickly found that their procedure remained #ery inefficient& Ho e#er, hile implementin! a different procedure also su!!ested in 1,-0 by Da! %ra it:, )obinson came to see ho the t o sets of ideas could be combined into a ne , far more efficient, automated proof procedure for firstorder predicate lo!ic: SresolutionS&&&S 06ccordin! to Donald 5ac$en:ie, The 6utomation of %roof: 6 Historical and 'ociolo!ical @xploration, S7@@@ 6nnals of the History of <omputin!S, #ol&1=, ?(, 1,,/, pp& =-2,1&

S7n retrospect, unification and resolution seem rather ob#ious ideas, hich arise ine#itably hen one asks hat must be syntactically true of a set of clauses hich possesses the semantic property of ha#in! no Herbrand models&S
0H&6&)obinson, S*nification and )esolution in )etrospectS, 1,,=, see at http:PP orleans&frP'<7@?<@'P+79RP5anifestationsPHfplcZ*nifZ,=P>fplcPin#ite-francais&html1& &uni#-

4ote. 6lmost at the same time hen H&6&)obinson in#ented the resolution method, 'er!ei 5aslo# in#ented his inverse method, hich has a similar ran!e of applications:
,. [u. Maslov& 6n in#erse method of establishin! deducibilities in the classical predicate calculus, S'o#iet 5athematics, DokladyS, 1,-., ?/, pp&1.20-1.2(( 'ee also: 5aslo# '& K& 01,(,-1,C21, human ri!hts acti#ist in @?<K<+R%6@D76 R9 '67?T %@T@)'G*)A& 6bout the history of the problem see: K. 3. #obinson. <omputational +o!ic: 5emories of the %ast and <hallen!es for the 9uture& Comp$tationa% Logic Q CL 9===, First #nternationa% Conference, Lon+on, 2C, 9K>9R /$%y, 9===, Procee+ings, 'prin!er, +ecture ?otes in <omputer 'cience, 2000, Vol& 1C-1, pp& 1-2. 0online copy1& M. Davis. The @arly History of 6utomated Deduction&7n: Han+boo& of A$tomate+ )easoning, ed& by 6& )obinson and 6& Voronko#, @lse#ier 'cience, 2001, #ol& 7, pp& (-1/ 0online postscript1

9he Method

20C 6!ain, ho to ork ith Ma cloud of simple dis>unctionsN effecti#ely8 6ssume that, in a set of clauses, t o clauses are contained such that an atom < appears as a positi#e member in the first clause, and as a ne!ati#e member in the second one: 7 A1 7 A2 &&&7 Am B 1 B2&&& Bn C ,
C 7 ,17 , 2&&&7 , p E 1 E 2&&& E 4 ,

011 021

or, simply, F C , C G & 01a1 02a1

7f < is false, then 01a1 yields 9, and, if < is true, then 02a1 yields A& Thus, from 01a1 and 02a1 e ha#e deri#ed F G & 7&e& deri#in! of F G from F C and C G is Slo!ically correctS, and it is called #obinsonJs #esolution rule 0H&6&)obinson proposed it in the abo#e 1,-( paper1: F C , C G & F G Takin! into account the rule 0of the classical lo!ic1 7 A B 6 ( A5 B ) , can obtain an alternati#e form of the )esolution rule: F C ,C G & F G 7n the classical lo!ic, this form is eQui#alent to the La8 of ,yllogism 0transiti#ity of implication1& 7f 9 is empty, then this form deri#es A from <, <`A, i&e& #esolution rule includes Modus Ponens as a special case& 7f A is empty, then from 7 F C ,7 C 0i&e& 9`<, f<1, the #esolution rule deri#es f9, i&e& it includes Modus 9ollens as a special case& !7ercise -.-.'& Deri#e the )esolution rule in the constructi#e lo!ic, i&e& pro#e that V+1-+10, 5%W: C F , 7C G U- F G & Verify that it cannot be pro#ed in the minimal lo!ic V+1-+,, 5%W& 0Hint: in the positi#e part 3 use Theorem 2&/&10b1 V+1, +2, +C, +10, 5%W: F C , 7C U- F & 7n the ne!ati#e part 3 #erify that in the minimal lo!ic, the )esolution rule allo s pro#in! of +10, see 'ection 2&/1& Thus, from the clauses 011 and 021, )obinsonBs )esolution rule allo s deri#in! of the follo in! clause:
7 A17 A2 &&&7 Am 7 ,1 7 ,2 &&&7 , p B1 B 2&&& B n E 1 E 2&&& E 4

20, 6t first !lance, this approach leads to nothin!, because this formula seems to be much lon!er than 011, and than 021& 'till, this is not 100T true, because, additionally, e can reduce the repeatin! atoms, and, finally, the set of different atoms, used in a clause form, is fixed; 7f, in our set of clauses, there are ? different atoms, then none of the clauses 0initial, or !enerated by resolution1 ill contain more than ? atoms 0each ith or ithout ne!ation1& 6nd the total number of different clauses ill ne#er exceed ( ? 0missing, -itho$t negation, -ith negation1& Thus, repeated applications of the )esolution rule ill SrotateS ithin this restricted Ssearch spaceS& The smart idea behind )obinsonBs )esolution rule is as follo s: it is a uni#ersal tool for deriving contradictions from inconsistent sets of clauses ; ?o other axioms and rules of inference are necessary; 5ore precisely, it is uni#ersal, if used to!ether ith the follo in! tri#ial rules of inference: F C D G 0permutation1, F DC G F C C G 0reduction1& F C G The permutation rule allo s arbitrary reorderin! of atoms in a clause 0for example, mo#in! < to ri!ht, and mo#in! f< to left1& The reduction rule allo s reduction of repeatin! identical atoms& !7ercise -.-..& Deri#e these inference rules in the minimal lo!ic V+1-+,, 5%W& 9heorem -.-.' CK.3.#obinsonD& 7n the classical propositional lo!ic V+1-+11, 5%W, a finite set of propositional clauses is inconsistent, if and only if )obinsonBs )esolution rule 0to!ether ith permutation and reduction rules1 allo s for deri#in! of a contradiction from it& 4ote& 7n some other texts, this fact is called Sthe refutation-completeness of the )esolution ruleS for the propositional lo!ic& Proof& 1& 6s you ha#e pro#ed in the @xercises /&/&1 and /&/&2, all the formulas, deri#ed from a set of formulas $1, $2, &&& , $s by usin! the %ermutation, )esolution and )eduction rules are conseQuences of $ 1, $2, &&& , $s& Hence, if these rules allo deri#in! a contradiction from this set of formulas, then it 0the set1 is inconsistent& 2& ?o , let us assume that a set of propositional clauses $ 1, $2, &&& , $s is inconsistent, i&e& a contradiction A7 A can be deri#ed from it: V+1-+11, 5%W: $1, $2, &&& , $s U- A7 A & Then, under the classical truth tables, the con>unction C 1 C 2 &&& C s takes

210 only false #alues 0#erify;1& +et us mark one of the atoms 0the atom <1 in it& +et us denote: - by C F i 3 the clauses containin! < ithout ne!ation, - by 7C G A 3 the clauses containin! < ith ne!ation, - by Hk 3 the clauses that do not contain <& 6ll the formulas 9i, A>, Hk are dis>unctions of atoms 0 ith or ne!ations1 that do not contain the atom <& Thus C 1 C 2 &&& C s is eQui#alent to
conA ( C F i) conA ( 7 C G A ) conA ( H & ) &

ithout

0.1

+et us apply 0the stran!e1 one of the distribution rules 0Theorem 2&(&11: V+1-+C, 5%W U- ( A B)C 6 ( AC )( BC ) & Hence, C 1 C 2 &&& C s is eQui#alent to ( C conA ( F i))( 7C conA ( G A ))conA ( H & ) & 7f < is false, then this formula is eQui#alent to conA ( F i )conA ( H & ) , i&e& conA ( F i )conA ( H & ) takes only false #alues& 7f < is true, then it is eQui#alent to conA ( G A )conA ( H & ) , i&e& conA ( G A )conA ( H & ) takes only false #alues& Thus the dis>unction ( conA ( F i ) conA ( H & ))( conA ( G A )conA ( H & )) 0/1 also takes only false #alues& ?o , let us, apply 0the SnormalS1 one of the distribution rules 0Theorem 2&(&11: V+1-+C, 5%W U- ( A B )C 6 ( A C )( B C ) , obtainin! that 0/1 is eQui#alent to ( conA ( F i ) conA ( G A)) conA ( H & ) & 0-1 7&e& this formula also takes only false #alues& 6nd 3 important note; 3 it does not contain the atom <& 9inally, by applyin!, a!ain, 0the stran!e1 one of the distribution rules 0Theorem 2&(&11 e can conclude that 0-1 is eQui#alent to conA (conA ( F i G A ))conA ( H & ) , i&e& to the set of clauses F i G A and Hk 0 here i, >, k run o#er their initial ran!es1& 2hat does this achie#ement mean8 7f the set of propositional clauses $1, $2, &&& , $s is inconsistent, then there is a set of clauses F i G A and Hk

211 0 here i, >, k run o#er their initial ran!es1, hich is inconsistent as hich contains one atom less than $1, $2, &&& , $s& ell, but

?o , ima!ine, that, in the clause form 0.1, e ha#e applied the )esolution rule for the atom < in all the "ossible 8ays 0before applyin!, apply the permutation rule to reorder atoms mo#in! < to ri!ht, and f< 3 to left1: F iC , C G A & F i G A 6fter this, apply the permutation and reduction rules to reduce identical atoms& 7n this ay e ha#e obtained exactly the abo#e-mentioned inconsistent set of clauses F i G A and Hk 0 here i, >, k run o#er their initial ran!es1& Thus, if some set of propositional formulas $ 1, $2, &&& , $s is inconsistent, then the )esolution rule 0to!eher ith the permutation and reduction rules1 allo s to deri#e from it another inconsistent set of propositional formulas, hich contains one atom less& Gy iteratin! this process, at the end of it, e ill ha#e an inconsistent set of propositional formulas built of a sin!le atom G& 7n a clause form, there can be only one such set 3 the set G, fG& This set represents a contradiction& D&@&D& 6s an example, let use )obinsonBs )esolution rule to pro#e that
BC , C 5 B , B 5 , U- B , &

+et us add 7 ( B ,) to the premises BC , C 5 B , B 5 , & 2e must pro#e that this set of . formulas is inconsistent& 9irst, let us obtain clause forms:
BC in clause form is BC , C 5 B in clause form is 7C B , B 5 , in clause form is 7 B , ,

7 ( B ,) is eQui#alent to 7 B 7 , & ?o , let us apply resolution to deri#e a contradiction from this set of . clauses: G#<, f<#G, fG#D, fG#fD: 9rom BC , C B e deri#e G, and ha#e no / clauses: e deri#e fG, and ha#e no - clauses: BC , 7C B , 7 B , , 7 B 7 , , B & 9rom 7 B D ,7 B D BC , 7C B , 7 B , , 7 B 7 , , B ,7 B &

212 2e ha#e deri#ed a contradiction: G, fG& This pro#es that the formula B , follo s from BC , C 5 B , B 5 , & D&@&D& !7ercise -.-.)& *se the )esolution rule to pro#e the follo in!: a1 6`G, f6`G U- G& b1 06`G1`6 U- 6 0%eirceBs +a 1& c1 G`0<`D1, G`< U- G`D 06xiom +21& d1 G`D, <`D U- BC 5 , & 06xiom +C1& e1 A BC , B 5 AC , A5 C U- <& :rom a ProgrammerJs Point of Vie8 Rf course, hen implementin! the )esolution rule in a computer pro!ram, e do not need decorations like the permutation and reduction rules& 7n a pro!ram, e ill represent each clause 7 A1 7 A2&&&7 Am B 1 B2 &&& B n as a pair of sets: ne!ati#e atoms, ? ^ X61, 62, &&& , 6mY, and positi#e atoms, % ^ XG1, G2, &&& , GnY& Rf course, the sets ?, % do not intersect 0if they do, then this clause contains 7C C &&& , i&e& it can be dropped as Snon-informati#eS1& #esolution rule CnonBrefined versionD. 7f there are t o clauses ?1, %1 and ?2, %2 such that %1 and ?2 0or ?1 and %21 contain a common atom <, then e can deri#e the clause " 1 " 2{C } , P 1 P 2 {C } & Rf course, the set union operation includes reduction of identical members automatically& The condition S%1 and ?2 0or ?1 and %21 contain a common atom <S can be expressed as C ( P 1 " 2)( P 2 " 1 ) & 7f, in the resultin! clause, the sets " 1 " 2{C } , P 1 P 2{C } intersect, then e should i!nore such result& 9ortunately, this can be detected in ad#ance& 7ndeed, ( " 1 " 2)( P 1 P 2 ) ^ ( " 1 P 1 )( " 1 P 2)( " 2 P 1)( " 2 P 2) ^ ( P 1 " 2 )( P 2 " 1 ) , because " 1 P 1 , " 2 P 2 are empty sets& The set ( P 1 " 2 )( P 2 " 1 ) is exactly the set of all atoms < allo in! application of the )esolution rule to clauses ?1, %1 and ?2, %2& Hence, the sets " 1 " 2{C } , P 1 P 2{C } ill not intersect, if and only if the set ( P 1 " 2 )( P 2 " 1 ) contains e7actly one atom <, i&e&, if and only if there is exactly one atom allo in! application of the )esolution rule&

21( #esolution rule Crefined versionD. 7f there are t o clauses ? 1, %1 and ?2, %2 such that the set ( P 1 " 2 )( P 2 " 1 ) contains exactly one atom <, then e can deri#e the clause " 1 " 2{C } , P 1 P 2{C } & ?o , let us try to desi!n a pro!ram implementin! the last step of Spro#in! by resolutionS 3 suppose, e ha#e already the initial list of clauses, and e ish to apply the )esolution rule tryin! to deri#e a contradiction& The main data stora!e ill be a !ro in! list of clauses 0the main list1: 0?1, %11, 0?2, %21, &&&, 0?k, %k1, &&& 7t ill start as the initial list, and each application of the )esolution rule append a ne clause to it& ill

To !uarantee a success, e must apply the )esolution rule in all the "ossible 8ays, i&e& e must scan all pairs of clauses 0? i, %i10?>, %>1, here i ^ 1, 2, &&&4 > ^ i\1, i\2, &&& To achie#e this, let us use the follo in! pair enumeration process: 0?1, %110?2, %21 3 first, scan all pairs 0i, >1 ith @W., i_>& 0?1, %110?(, %(1, 0?2, %210?(, %(1 3 after this, scan all pairs 0i, >1 ith @W), i_>& 0?1, %110?., %.1, 0?2, %210?., %.1, 0?(, %(10?., %.1 3 after this, scan all pairs 0i, >1 ith @W*, i_>& @tc& The process ill stop, hen e ill arri#e at the le#el >, and the main list ill contain less than > 0in fact, >-11 clauses& 9or a set of n atoms, there are only ( n different clauses& 9or example, for t o atoms 6, G, there are , different clauses: 7 A7 B , 7 A B , A7 B , A B , 7 A , A ,7 B , B , and the empty clause 0representin! contradiction1& 7&e&, if e ill prohibit duplicate clauses in the main list, then our process ill al ays stop& Thus, the follo in! pseudo-code parsin! necessary;1: ill do 0no strin! processin!, no expression

function propositional resolution 0initial list1 X of clauses Y begin if initial list contains contradiction then return T)*@ X contradiction found Y main list ^ eliminate duplicates 0initial list1 for > ^ 2 by 1 begin - if count 0main list1 _ > then return 96+'@ X no contradiction deri#ed Y - else - for i ^ 1 to >E1 by 1

21. -- X consider i-th and >-th clauses in the main list: 0?i, %i1, 0?>, %>1 Y -- if ( " i P A )( P i " A) contains exactly one element < then -- begin --- Xapply resolutionY --- if 0 " i " A{C } , P i P A{C } 1 not in main list then --- begin ---- add it to main list ---- if main list contains contradiction then return T)*@ X contradiction deri#ed Y -- end --- end end end !7ercise -.-.*& De#elop a computer pro!ram implementin! the abo#e pseudocode& 4ote& 'ee my #ersion of such a pro!ram in <\\: header file, implementation1& 1arning6 Despite its beauty, resolution method cannot o#ercome the !eneral complexity problem, mentioned at the end of 'ection .&2: in the classical propositional lo!ic, the task of reasonin! is Mco-?%-completeN& 6nd a closer analysis sho s that, in the 8orst "ossible case, !i#en a set of formulas of total len!th n, the time reQuired by resolution method ill be exponentional 3 about 2Cn seconds& Gut in many "ractical situations, experience sho s that resolution method sol#es its task, and 3 in acceptable time& 7n particular, %rolo! interpreters are usin! resolution, and are sol#in! many practical tasks in acceptable time;

-./. $erbrandJs 9heorem


1arning6 The principal results of this 'ection are #alid only for the classical logic; HacQues Herbrand 01,0C-1,(11 S&&& 6fter lea#in! AFttin!en, Herbrand decided on a holiday in the 6lps before his intended return to 9rance& Ho e#er he as ne#er to complete his plans for he died in a mountaineerin! accident in the 6lps only a fe days after his holiday be!an& His death at the a!e of 2( in one of the tra!ic losses to mathematics&S 0accordin! to 5acTutor History of 5athematics archi#e1&

21/ Herbrand pro#ed his famous theorem in 1,2,: K.$erbrand. )echerches sur la thiorie de la dimonstration& %h&D& Thesis, *ni#ersity of %aris, 1,(0 0appro#ed in 6pril 1,2,1& *nlike the proof presented belo , the ori!inal proof of HerbrandBs Theorem does not depend on AFdelBs <ompleteness Theorem 0or 5odel @xistence Theorem1& Herbrand completed his %h&D& thesis in 1,2,& 7n the same 1,2, AFdel completed his doctoral dissertation about completeness 0see 'ection .&(1& 7n fact, HerbrandBs method allo s pro#in! of AFdelBs <ompleteness Theorem, but he 0Herbrand1 Sdid not notice itS& 2hy8 'ee
,amuel #. %uss. Rn HerbrandBs Theorem& <+ecture ?otes in <omputer 'cienceS, Vol&,-0, 1,,/, 'prin!er-Verla!, pp&1,/-20, 0a#ailable online1&

The fla#our of this famous theorem can be best presented in its simplest #ersion& 7n this #ersion, 90x1 is a Quantifier-free formula containin! only one #ariable x& Then, HerbrandBs Theorem says: The formula x90x1 is lo!ically #alid, if and only if there is a finite set of constant terms t1, &&&, tn such that the dis>unction F ( t 1 )&&& F ( t n) is lo!ically #alid& Rr, eQui#alently 0#ia AFdelBs <ompleteness Theorem1, The formula x90x1 is pro#able in the classical lo!ic, if and only if there is a finite set of constant terms t1, &&&, tn such that the dis>unction F ( t 1 )&&& F ( t n ) is pro#able in the classical lo!ic& 6s e ill see in the proof, HerbrandBs Theorem is ScausedS by the simple SfactS that in any proof of x90x1 only a finite set of terms could be used& ?o , more precisely& +et + be a predicate lan!ua!e, containin! at least one ob>ect constant, and let 9 be a Quantifier-free formula& 7dea [1 0author81& The formula p ( c 1) 4 ( c 2 , f ( x )) is Quantifie-free 0c1, c2 are ob>ect constants, f 3 a function constant, p, Q 3 predicate constants1& 7n a sense, any SclosedS interpretation domain for this formula must contain ob>ects denoted by the terms c1, c2, f0c11, f0c21, f0f0c111, f0f0c211,&&& 'o, let us define the so-called $erbrandJs universe of the formula 9 0let us denote it by H*91 as the minimum set of all constant terms such that: a1 7f c is an ob>ect constant occurrin! in 9, then c is in H*9& b1 7f 9 does not contain ob>ect constants, then one of the constants of the lan!ua!e + is in H*9&

21c1 7f terms t1, &&&, tk are in H*9, and f is a k-ary function constant occurrin! in 9, then the term f0t1, &&&, tk1 is in H*9& !7ercise -./.'. Verify that H*9 is a non-empty finite or countable set 0pro#ide an al!orithm !eneratin! the members of H*91& 9heorem -./.' CHerbrandBs Theorem 3 the simplest case1& +et + be a predicate lan!ua!e, containin! at least one ob>ect constant, and let 90x1 be a Quantifierfree formula containin! only one free #ariable x& Then the formula x90x1 is logically valid Ci.e. "rovable in the classical "redicate logicD, if and only if there is a finite set of terms t 1, &&&, tn from H*9 such that the dis>unction F ( t 1 )&&& F ( t n) is logically valid Ci.e. "rovable in the classical "redicate logicD& Proof. +et us assume the contrary 3 that none of the dis>unctions F ( t 1 )&&& F ( t n) is lo!ically #alid 0ti-s are terms from H*91& 7dea [2 3 then the follo in! theory T is consistent: T ^ X f90t1 U t is a term from H*9Y& 7ndeed, if T ould be inconsistent, then there ould be a T-proof of some formula GIfG& 7n this proof, only a finite set of the axioms f90t1 ould be used, i&e& for some terms t1, &&&, tn from H*9: V+1-+1/, 5%, AenW: f90t11, &&&, f90tn1 U- B7 B & Hence, by Deduction Theorem 2 0it is applicable here, because 90x1 contains only one free #ariable, and ti-s are constant terms, i&e& e#ery f90ti1 is a closed formula1: V+1-+1/, 5%, AenW: U- 7 F ( t 1)&&& 7 F (t n) ` B7 B , V+1-+1/, 5%, AenW: U- ( F ( t 1 )&&& F ( t n )) ` B7 B , and thus, V+1-+1/, 5%, AenW: U- F ( t 1 )&&& F ( t n) & 7&e&, F ( t 1 )&&& F ( t n) is lo!ically #alid& This contradicts our assumption, that none of the dis>unctions F ( t 1 )&&& F ( t n) is lo!ically #alid& Hence, T is a consistent theory& 7dea [( 3 if T is consistent, then, by the 5odel @xistence Theorem, there is a model H of T& 7n this model, all the axioms of T are true, i&e& so are all the formulas f90t1 ith t from the set H*9&

21= 7dea [. 3 let us restrict the domain of the model H to those elements of it, hich are interpretations of terms from the set H* 9, and let us restrict the entire interpretation correspondin!ly& +et us denote this ne interpretation by H1& Then, a1 6ll the formulas f90t1 0 ith t from the set H* 91 are true in H1& 7ndeed, f90t1 contains only constant terms from H*9 0idea [1 orkin!;1, and all of them ha#e the same interpretations in H1 that they had in H& Thus, if f90t1 as true in H, it remains true in H1& b1 Hence, the formula xf90x1 is true in H1 0because the domain of H1 consists only of those elements, hich are interpretations of terms from the set H*91& c1 Hence, the formula x90x1 is false in H1& This contradicts the lo!ical #alidity of x90x1& D&@&D& !7ercise -./... )epeat the abo#e proof, pro#in! a more !eneral form of HerbrandBs Theorem: 9heorem -./.. 0HerbrandBs Theorem 3 the simplest case1& +et + be a predicate lan!ua!e, containin! at least one ob>ect constant, and let 90x 1, &&&, xm1 be a Quantifier-free formula containin! only m free #ariables x 1, &&&, xm& The formula x1&&&xm90x1, &&&, xm1 is logically valid, if and only if there is a finite set of m-tuples tt1, &&&, ttn of terms from H*9 such that the dis>unction F ( tt 1)&&& F ( tt n) is logically valid& 6s you #erified it in the @xercise .&1&1, any formula A is lo!ically #alid, if and only if fA is unsatisfiable& Thus, x1&&&xm90x1, &&&, xm1 is lo!ically #alid, if and only if x1&&&xmf90x1, &&&, xm1 is unsatisfiable& Rn the other hand, F ( tt 1)&&& F ( tt n) is lo!ically #alid, if and only if 7 F ( tt 1)&&&7 F ( tt n ) is unsatisfiable& ?o , let us replace 9 by f9, and e ha#e pro#ed 9heorem -./.) 0HerbrandBs Theorem 3 a more useful alternati#e form1& +et + be a predicate lan!ua!e, containin! at least one ob>ect constant, and let 90x1, &&&, xm1 be a Quantifier-free formula containin! only m free #ariables x1, &&&, xm& The formula x1&&&xm90x1, &&&, xm1 is unsatisfiable 0i&e& inconsistent in the classical lo!icD, if and only if there is a finite set of mtuples tt1, &&&, ttn of terms from H*9 such that the con>unction

21C 7 F ( tt 1)&&&7 F ( tt n ) is unsatisfiable 0i&e& inconsistent in the classical lo!icD& 4ote. 6s you #erified it in the @xercise .&(&-, a set of formulas is inconsistent in the classical lo!ic, if and only if it is unsatisfiable& 2hy is this form Smore usefulS8 +et us try applyin! this form of HerbrandBs Theorem to sets of formulas in clause form& 11 The Smeanin!Sof any set of closed formulas 91, &&& , 9k is represented by their con>unction F 1 &&& F & & 21 6 clause is any dis>unction of atomic formulas or their ne!ations& 9or example, 7 p ( c1 ) p ( c 2 )4 ( x , f ( y )) , or p ( x )7 4 ( y , f ( 3 )) & The Smeanin!S of a set of clauses is represented by their uni#ersally Quantified con>unction& 9or example, x y 3 ([7 p ( c 1) p ( c 2) 4 ( x , f ( y ))][ p ( x )7 4 ( y , f ( 3 ))]) & (1 6s e kno from the pre#ious 'ection /&., the set 91, &&& 9k can be reduced to a clause from, i&e& there is a set of clauses ' such that 9 1, &&& , 9k is unsatisfiable, if and only if ' is unsatisfiable& ?o , let us apply the abo#e form of HerbrandBs Theorem 0Theorem /&-&(1& 7f ' contains m #ariables 0of course, all of them are uni#ersally Quantified1, then ' is unsatisfiable, if and only if there is a finite set of m-tuples tt 1, &&&, ttn of terms from H*' such that the con>unction S ( tt 1 )&&& S ( tt n) is unsatisfiable& 7f e take a clause from ', and substitute some terms from H* ' for all its #ariables, then e obtain a 0so-called1 ground clause of '& 9or example, if ' ^ X 7 p ( c1 ) p ( c 2)4 ( x , f ( y )) 4 p ( x )7 4 ( y , f ( 3 )) Y, then the substitution X c1 P x4 c2 P y4 f0c21 P : Y yields the follo in! t o !round clauses: 7 p ( c1 ) p ( c 2 )4 ( f ( c 1) , f ( c 2)) ,
p ( c 1)7 4 ( c 2, f ( f ( c 2 ))) &

Rf course, the con>unction S ( tt 1 )&&& S (tt n) is a set of !round clauses& Thus, if , is unsatisfiable, then there is an unsatisfiable finite set of ground clauses of ,. 6nd con#ersely8 7f there is an unsatisfiable finite set < ^ X<1, &&&, <nY of !round clauses of ', then each <i is !enerated by some substitution, hich can be represented as an m-tuple tti of terms from H*'& 7f X<1, &&&, <nY is unsatisfiable, then X'0tt11, &&&,

21, '0ttn1Y 3 as a super-set of the former, is unsatisfiable, too 0Se#en more unsatisfiableS1& ?o , if ' ould be satisfiable, then 0because all the #ariables of ' are meant uni#ersally Quantified1 so ould be the formula S ( tt 1 )&&& S (tt n) & <ontradiction& Thus, e ha#e pro#ed another form of HerbrandBs Theorem& 9heorem -./.* 0HerbrandBs Theorem 3 the most useful form& 6uthor 3 Herbert G&@nderton81& +et + be a predicate lan!ua!e, containin! at least one ob>ect constant, and let 91, &&&, 9k be a set of closed formulas in +& Then this set is unsatisfiable, if and only if its clause form allo s an unsatisfiable finite set of !round clauses& 2hy is this form Sthe most usefulS8 Gecause 0let us i!nore performance problems1, a1 The clause form of 91, &&&, 9k is a finite set ', !enerated by a simple 0but a #ery slo 1 al!orithm 0see 'ections /&1-/&.1& b1 HerbrandBs uni#erse H*' is a finite or infinite set of constant terms, !enerated by a simple al!orithm 0see @xercise /&-&11& c1 Thus, all the possible finite sets of !round clauses of ' can be !enerated by a simple combination of the abo#e t o al!orithms& d1 *nsatisfiability of each finite set of !round clauses can be detected by a simple 0but a #ery slo 1 al!orithm 0see +emma /&-&/ belo 1& Thus, e ha#e here a sim"le algorithm 0but a #ery slo "rovability in the classical "redicate logic. one1 for checking

Lemma -./.-& 6 finite set of !round clauses is unsatisfiable, if and only if the con>unction of these clauses is unsatisfiable under the classical truth tables& Proof& 7n the abo#e example of !round clauses:
7 p ( c1 ) p ( c 2)4 ( f ( c1 ) , f ( c 2 )) ,

p ( c 1)7 4 ( c 2, f ( f ( c 2))) , e ha#e / different atoms: p0c11, p0c21, Q0f0c11, f0c211, Q0c2, f0f0c2111& +et us denote these atoms by D1, D2, D(, D.& Thus e obtain the follo in! propositional formula
( 7S 1S 2S ()( S 17 S . ) &

1& 7f this formula cannot be satisfied under the classical truth tables, then e cannot assi!n truth #alues to predicates p, Q in a ay makin! all the

220 correspondin! clauses true& 7&e& then the correspondin! set of !round clauses also cannot be satisfied& D&@&D& 2& 7f this formula can be satisfied under the classical truth tables, then find a truth #alue assi!nement makin! it true, for example: D1^false 0this makes the first dis>unction true1, D.^false 0this makes the second dis>unction true1& ?o , e can define the follo in! interpretation H makin! the correspondin! !round clauses true: DH ^ X c1, c2, f0c11, f0c21, f0f0c21 Y 0the set of all terms appearin! in the clauses, i&e& a subset of the Herbrand uni#erse14 p0c11^false, Q0c2, f0f0c211^false 0these assi!nements make both !round clauses true1& 6ll the other truth #alues are irrele#ant, so, as follo s: e can define them, for example, e can

p0c21^true, p0f0c111^true, p0f0c211^true, p0f0f0c211^true4 Q0x, y1^true, if x is not c2, or y is not f0f0c21& D&@&D& &&& To be continued& &&& 9urther readin!: Michael 5enesereth. <omputational +o!ic 0see at http:PPlo!ic&stanford&eduPclassesPcs1/=P200/fallPcs1/=&html1&

-.0. #esolution Method for Predicate :ormulas


1arning6 The principal results of this 'ection are #alid only for the classical logic; 7f e are interested only in deri#in! contradictions from inconsistent sets of formulas, then e can note that a set of closed predicate formulas is

221 inconsistent 0i&e& allo s deri#in! a contradiction in the classical lo!ic1, if and only if the con>unction of these formulas is unsatisfiable 0@xercise .&(&-1& Thus, instead of the initial set, e can analy:e the set of clause forms of these formulas& 7ndeed, if e deri#e a contradiction from the set of clause forms, then this set is unsatisfiable, i&e&, by Theorem /&.&2, so is the initial set, and hence, the initial set is inconsistent& 6nd con#ersely, if the initial set of formulas is consistent, then it is satisfiable, i&e& so is the set of clause forms, i&e& e ill be not able to deri#e a contradiction from it& The next step for ard 3 in clause forms, e can dro" all the universal >uantifiers& 7ndeed, if e deri#e a contradiction from a set uni#ersally Quantified clause forms, then e can deri#e it from the correspondin! nonQuantified set 0 e can apply the Aen inference rule 90x1 U- x90x1 to obtain the Quantified forms from the non-Quantified ones1& 6nd con#ersely, if e deri#e a contradiction from a set of non-Quantified clause forms, then e can deri#e it from the correspondin! uni#ersally Quantified set 0apply the 6xiom + 12: x90x1 ` 90x1 to obtain non-Quantified forms from the Quantified ones1& 6fter droppin! Quantifiers, sets of clause forms become simply sets of clauses 0con>unction of con>unctions is eQui#alent to a S>ointS con>unction1& Thus, e can concentrate on sets of clauses that do not contain Quantifiers, like as the one obtained in 'ection /&.:
5 g ( $1)>$ 1 , 5 g ( $2 )>1 , y >1, 3 >1, g ( $( )= y 3 5 &

?ote that clauses consist of atomic formulas only, and no t8o clauses contain common variables& Thus, clauses are completely separated, and this separation ill !reatly simplify processin! of clauses by means of substitution 0see belo 1& 2ill the )obinsonBs )esolution rule remain a uni#ersal tool for deri#in! contradictions also from inconsistent sets of predicate formulas 0i&e& sets of non-Quantified clauses, consistin! of atomic formulas18 +et us ima!ine, e ha#e deri#ed the follo in! t o formulas 0p is a unary predicate constant, 0 3 an ob>ect constant1: p ( x 1) F ( x 1 , y 1) , 7 p ( 0 )G ( x 2 , y 2) & To apply the )obinsonBs )esolution rule, for x1: e must first, in p0x 11, substitute 0

p (0 ) F ( 0, y 1) , 7 p ( 0 )G ( x 2 , y 2) &

?o , e can apply the )esolution rule, obtainin! the formula

222 F ( 0, y 1 )G ( x 2 , y 2 ) & 'urprisin!ly, this simple idea of Sunification by substitutionS appears to be sufficient to make )obinsonBs )esolution rule a uni#ersal tool for deri#in! contradictions also from inconsistent sets of predicate formulas; 6nd, in !eneral, the necessary substitutions are not much more complicated than in the abo#e simplest example& The substitution rule allo s, in some clause <, replacin! of all occurrences of some #ariable x by any term t& 9heorem -.0.' CK. 3. #obinsonD& 7n the classical predicate lo!ic V+1-+11, +12+1/, 5%, AenW, a set of predicate clauses is inconsistent, if and only if )obinsonBs #esolution rule 0to!ether ith permutation, reduction and substitution rules1 allo s deri#in! a contradiction from it& 4ote& 7n some other texts, this fact is called Sthe refutation-completeness of the )esolution ruleS& Proof& 1& 6ll the formulas, deri#ed from a set of clauses $ 1, $2, &&& , $s by usin! permutation, reduction, substitution and )esolution rules, are conseQuences of $1, $2, &&& , $s& Hence, if these rules allo deri#in! a contradiction from this set of clauses, then it 0the set1 is inconsistent& 2& ?o , let us assume that the set of clauses ' ^ X$ 1, $2, &&& , $sY is inconsistent& Then it is unsatisfiable 0@xercise .&(&-1& 6nd then, by HerbrandBs Theorem, it allo s a finite unsatisfiable set of !round clauses < 1, &&&, <n& @ach <i of these !round clauses is obtained from some clause in ' by means of some substitution subi 0of terms from the Herbrand uni#erse H*'1, i&e& by applyin! the substitution rule& Gy +emma /&-&/, the set <1, &&&, <n is unsatisfiable, if and only if the con>unction C 1&&&C n is unsatisfiable under the classical truth tables, i&e&, if and only if the set <1, &&&, <n is inconsistent& 6nd, by Theorem /&/&1, a finite set of propositional clauses is inconsistent4 if and only if )obinsonBs )esolution rule 0to!ether ith permutation and reduction rules1 allo s deri#in! a contradiction from it& D&@&D& #efinements E ,te" ' C:irst of the 98o ,mart IdeasD +et us examine once more the part t o of the proof of Theorem /&=&1, here a specific 0hopeless;1 Sproof strate!yS is used& 9irst, since t o clauses $i do not contain common #ariables, e can think that

22( each of the substitutions sub> is applied to a sin!le clause, i&e& e can think, in fact, of a 0finite1 set of substitutions subi>, here each subi> is applied only to the clause $i& +et us denote by 9&sub the result of application of the substitution sub to the formula 9& 'econd, to deri#e a contradiction from X$1, $2, &&& , $sY, e may apply, first, all the necessary substitutions 0sta!e 1 3 substitutions only;1, and, after this, all the necessary permutations, reductions and resolutions 0sta!e 2 3 no more substitutions;1& This is exactly the abo#e-mentioned specific 0hopeless;1 Sproof strate!yS& 2hy hopeless8 Gecause, before applyin! the substitutions sub i>, e must find them among all the "ossible substitutions of terms from the infinite set H*'& This is a performance problem that does not affect our abo#e theoretical considerations, but could make their result useless& The smart ideas [1 and [2 introduced belo , allo to restrict the substitution search space considerably& 7ma!ine one of the resolutions of sta!e 2, here <1 is an atomic formula: F 1C 1 , C 1G1 & F 1 G 1 7f both premises F 1 C 1 , 7C 1G 1 are comin! directly from sta!e 1, then they ha#e been obtained from some initial clauses F C , 7 ,G by t o substitutions sub1 and sub2 such that: 91 is 9&sub1, <1 is <&sub1, f<1 is fD&sub2, A1 is A&sub2& 2e can call such pair of substitutions a unifier, because <&sub1 and D&sub2 represent the same atomic formula 0compare the example before the text of Theorem /&=&11& 7f one 0or both1 of the premises does not come directly from sta!e 1, then it is either an initial clause, or the result of a pre#ious resolution& Gy puttin! an empty substitution 0 hich does no chan!e formulas1 instead of sub 1 or sub2 0or both1 e can still think of the premises as obtained by a unification& 6nd, finally, if, to deri#e a contradiction G, fG from $ 1, $2, &&& , $s, e do not need resolution at all, then e need, ne#ertheless, unifyin! substitutions, con#ertin! t o clauses G1 and fG2 into G and fG& Thus 0smart idea [11, to derive contradictions, 8e can do 8ith one s"ecific kind of the substitution rule E the unification rule: a1 Take t o clauses, mark a positi#e atom < in the first clause, and a ne!ati#e

22. atom fD in the second one& Thus, and 7 , G & e are considerin! t o clauses: F C

b1 Try to find t o substitutions sub 1 and sub2 such that <&sub1 and D&sub2 represent the same atom <1& 6nd you do not need to introduce #ariables of the other clauses; 7f you succeed, you ha#e obtained t o clauses: F 1 C 1 , 7C 1G1 , here <1 is <&sub1 0^D&sub21, 91 is 9&sub1 and A1 is A&sub2& 'ince clauses do not contain common #ariables, the union sub1sub 2 is a substitution 0a unifier of < and D1& c1 6pply resolution, obtainin! the clause F 1 G1 & 2e ha#e pro#ed the follo in! refined #ersion of Theorem /&=&1: 9heorem -.0.. C K.3.#obinsonD& 7n the classical predicate lo!ic V+1-+11, +12+1/, 5%, AenW, a set of predicate clauses is inconsistent4 if and only if )obinsonBs #esolution rule 0to!ether ith permutation, reduction and unification rules1 allo s deri#in! a contradiction from it& 2hy is this refinement important8 Gecause no , instead of tryin! out all the possible substitutions 0of terms from H*' for clause #ariables1, e can concentrate on substitutions that unify t o clauses& This allo s to restrict the substitution search space considerably& #efinements E ,te" . C,econd of the 98o ,mart IdeasD ,ubstitution <3lgebra< 7n !eneral, each substitution in#ol#es a list of distinct #ariables x 1, &&&, xk and a list of terms t1, &&&,tk& 6ll occurrences of the #ariable xi are replaced by the term ti& Thus, this operation can be most naturally represented by the set of pairs X t1 P x1, &&&, tk P xk Y& The order of pairs ti P xi is irrele#ant because of the follo in! Santi-cascadin!S condition: the ne occurrences of the #ariables x 1, &&&, xk created by the substitution, are not replaced& The result of application of some substitution sub to some expression 0term or formula1 9, is usually denoted by 9&sub& 9or example, if 9 is p0x, f0y11 and sub ^ X f0:1 P x, : P y Y, then 9&sub is p0f0:1, f0:11& The empty set of pairs XY represents the so-called empty substitution& Rf course, 9&XY ^ 9, for any expression 9& 7f the #ariable sets of t o substitutions sub 1 and sub2 do not intersect, and the terms of sub1 do not contain the #ariables of sub2, and the terms of sub2 do not

22/ contain the #ariables of sub1, then the union sub1sub 2 0of t o sets of pairs1 defines a substitution& 'till, the most important operation on substitutions is com"osition& 7f sub1 and sub2 are t o substitutions, then sub1&sub2 denotes the composed substitution Sapply first sub1, and after this, apply sub2S& 9or example, if sub1 ^ X f0:1 P x, : P y Y and sub2 ^ X f0 1 P : Y, then sub1&sub2 ^ X f0f0 11 P x, f0 1 P y, f0 1 P : Y& !7ercise -.0... a1 Verify that the substitution composition is associati#e and non-commutati#e 0pro#ide a counter-example1, and that the empty substitution is the only Sunit elementS 0i&e& XY&sub ^ sub&XY ^ sub for any substitution sub1& b1 7s there any al!ebraic correlation bet een composition and union of substitutions8 Most 5eneral Unifiers Ho do beha#e unifiers in the substitution Sal!ebraS8 6ssume, sub 1 and sub2 are t o different unifiers of the same pair of expressions 9 and A& 7&e& 9&sub1 ^ A&sub1, 9&sub2 ^ A&sub2& 7f there ould be a substitution sub such that sub 2^sub1&sub, then e could say that sub1 is a no less general unifier than sub2& 9or example, let us try to unify the first members of the follo in! t o formulas: p ( x 1) F ( x 1 , y 1) , 7 p ( f ( x 2))G ( x 2, y 2 ) & 7t ould be natural to use the substitution sub1 ^ X f0:1 P x1, : P x2 Y, obtainin! p ( f ( 3 )) F ( f ( 3 ) , y 1 ) , 7 p ( f ( 3 ))G ( 3 , y 2) & Gut, in principle, one could use also the substitution sub 2 ^ X f0f0:11 P x1, f0:1 P x2 Y, obtainin!
p ( f ( f ( 3 ))) F ( f ( f ( 3 )) , y 1) , 7 p ( f ( f ( 3 )))G ( f ( 3 ) , y 2 ) &

Rf course, sub1 is SbetterS, because sub2 ^ sub1&X f0:1 P : Y& 2hy8 7f our purpose as unifyin! p0x11 ith p0f0x211, then sub1 performs this 0as ell as sub21, but it Slea#es more spaceS for subseQuent substitutions 0than sub 21& 7ndeed, to continue after sub1, instead of sub2 ^ sub1&X f0:1 P : Y, e can choose also sub( ^ sub1&X !0:1 P : Y etc& Thus, usin! a more !eneral unifier is preferable&

22'o, let us call a unifier sub of t o expressions 9 and A a most general unifier CmguD of 9 and A, if and only if it is no less !eneral than any other unifier of 9 and A 0i&e& if and only if, for any other unifier subB of 9 and A, there is a substitution subBB such that subB ^ sub&subBB1& Lemma -.0.). 7f t o expressions lists 99 and AA are unifiable, then there exists an m!u of 99 and AA& Proof 0lon!, but easy1& +et us define the total length of an expression list as follo s: a1 0atomic expressions1 the total len!th of a constant or of a #ariable is 1, b1 the total len!th of the expression list e 1, &&&, en is the sum of the total len!ths of the members e1, &&&, en, c1 0composite expressions1 the total len!th of the expression f0t1, &&&, tn1 0 here f is function constant or predicate constant1, is the total len!th of the expression list t1, &&&, tn plus 1& +et us pro#e our +emma by induction usin! min0totalZlen!th0991, totalZlen!th0AA11 as the induction parameter& 11 Induction base. The total len!th of 99 or AA is 1& +et us assume totalZlen!th0991^1& a1 99 is a constant c& Then 99 and AA are unifiable, if and only if AA is the same constant c& Then, empty substitution is the only possible m!u 0#erify1& b1 99 is a #ariable x& Then, 99 and AA are not unifiable, if: b11 AA is a list of more than one expression, or, b21 AA is a composite expression that contains x 0then any substitution of t for x makes AA lon!er than t1& 6nd, 99 and AA are unifiable, if and only if AA is a #ariable, or AA is a composite expression that does not contain x& 7f AA is the #ariable x, then the empty substitution is the only possible m!u 0#erify1& 7f AA is a #ariable y 0other than x1, then all unifications of 99 and AA ha#e the form X t P x, t P y, &&& Y, here t is any term& 6mon! them, m!u-s are X : P x, : P y Y, here : is any #ariable 0#erify1& 7f AA is a composite expression that does not contain x, then all unifications of 99 and AA ha#e the form X AA&sub P x, &&& Y * sub, here sub is any substitution that does not substitute for x 0#erify1& 6mon! them, m!u-s are X AA&sub P xY* sub, here sub substitutes distinct #ariables for distinct #ariables 0#erify1& This completes the induction base&

22= 21 Induction ste". 6ssume, min0totalZlen!th0991, totalZlen!th0AA11^n, here nb1& 7f 99 and AA are unifiable, then, as lists, they contain the same number of members& 2a1 99 and AA contain are sin!le expressions& 'ince min0totalZlen!th0991, totalZlen!th0AA11b1, both are composite expressions 3 suppose, 99 is f0s 1, &&&, sm1 0 here f is function constant or predicate constant, and s 1, &&&, sm are terms1, and AA is !0t1, &&&, tn1 0 here ! is function constant or predicate constant, and t1, &&&, tn are terms1& 99 and AA are unifiable, if and only if a1 f and ! represent the same constant, and b1 the lists s 1, &&&, sm and t1, &&&, tn are unifiable& Thus, the unifiers of 99 and AA coincide ith the unifiers of lists& 'ince min0totalZlen!th0s1, &&&, sm1, totalZlen!th0t1, &&&, tn11_n, by the induction assumption, +emma /&=&( holds for the lists, i&e& it holds also for 99 and AA& 2b1 99 and AA contain t o or more members& 7f 99 and AA are unifiable, then so are their first members 0SheadsS1 91 and A1& +et us denote by 992 and AA2 the rests of lists 0StailsS1& 'ince min0totalZlen!th0911, totalZlen!th0A111_n, by the induction assumption, there exists at least one m!u of 91 and A1& The same is true also for 992 and AA2& +et us denote by m!u1 an arbitrary m!u of 91 and A1 ?o , let us consider an arbitrary unifier u of 99 and AA& 7t must unify also 9 1 ith A1, and 992 ith AA2& Hence, u ^ m!u1&sub1, here sub1 is some substitution& 2e kno that 91&m!u1 ^ A1&m!u1& Gut hat about 992&m!u1 and AA2&m!u18 +et us apply sub1 to both: 992&m!u1&sub1 ^ 992&u AA2&m!u1&sub1 ^ AA2&u 'ince u unifies 992 ith AA2, 992&m!u1&sub1 ^ AA2&m!u1&sub1, i&e& sub1 unifies 992&m!u1 ith AA2&m!u1& +et us denote by m!u 12 an arbitrary m!u of 992&m!u1 and AA2&m!u1& Then, sub1 ^ m!u12&sub12, here sub12 is some substitution, and m!u1&m!u12&sub12 ^ m!u1&sub1^u& Thus, e ha#e established that for an arbitrary unifier u of 99 and AA there is

22C a substitution sub12 such that m!u1&m!u12&sub12 ^ u& Rf course, the composition m!u1&m!u12 unifies 99 ith AA 0since it unifies 9 1 ith A1, and 992 ith AA21& Hence, m!u1&m!u12 is an m!u of 99 and AA& D&@&D& Unification 3lgorithm Ho could e determine, can t o atomic formulas < and D be unified, or not8 This problem can be sol#ed by the follo in! simple pseudo-code Get.ostGenera%2nifier, hich follo s the abo#e proof of +emma /&=&(, and here e7"ression lists are defined in the +7'% style: 11 @ach #ariable, constant, function constant and predicate constant is an expression list 0consistin! of a sin!le member1& 21 7f s1, &&&, sn are expression lists, then the list of s 1, &&&, sn is an expression list 0consistin! of members s1, &&&, sn1& The first member s1 is called the head of the list, and the list of s2, &&&, sn 3 the tail of the list& Thus, instead of, for example, f0t1, &&, tn1, e use simply the 0+7'% style1 list f, t1, &&, tn& This simplifies the recursion interface& This pro!ram detects, are t o expression lists unifiable, or not, and, if they are, it returns one of their most !eneral unifiers& function Aet5ostAeneral*nifier 0expressionZlist1, expressionZlist21 begin if len!th0expressionZlist11 b 1 and len!th0expressionZlist21 b 1 then begin --- h1 ^ head0expressionZlist114 --- h2 ^ head0expressionZlist214 --- subH ^ Aet5ostAeneral*nifier0h1, h214 --- if subH ^ false then return false4 Xunification impossibleY --- t1 ^ tail0expressionZlist11&subH4 --- t2 ^ tail0expressionZlist21&subH4 --- subT ^ Aet5ostAeneral*nifier0t1, t214 --- if subT ^ false then return false4 Xunification impossible, note that subH is a m!u;Y --- return subH&'ubT4 Xthis composition unifies expressionZlist1 and expressionZlist2Y end Xno , expressionZlist1, or expressionZlist2 consists of a sin!le member: m1 or m2Y if len!th0expressionZlist11 ^ 1 and m1 is #ariable then

22, begin --- if m1 ^ expressionZlist2 then return XY4 Xempty substitutionY --- if m1 occurs in expressionZlist2 then return false4 Xunification impossible 3 #erify;Y --- return XexpressionZlist2 P m1Y4 Xsubstitute expressionZlist2 for m1Y end if len!th0expressionZlist21 ^ 1 and m2 is #ariable then begin --- if m2 ^ expressionZlist1 then return XY4 Xempty substitutionY --- if m2 occurs in expressionZlist1 then return false4 Xunification impossible 3 #erify;Y --- return XexpressionZlist1 P m2Y4 Xsubstitute expressionZlist1 for m2Y end Xno , expressionZlist1, or expressionZlist2 consists of a sin!le member that is not #ariableY if expressionZlist1 ^ expressionZlist2 then return XY4 Xempty substitutionY return false4 Xunification impossible 3 #erify;Y end !7ercise -.0.). Verify that this pro!ram detects, are t o expression lists unifiable, or not, and, if they are, it returns one of their m!u-s& 0Hint: repeat the proof of +emma /&=&(&1 ,mart idea V.= 9o derive contradictions, 8e can do 8ith even more s"ecific kind of the unification rule E the mguBrule: a1 Take t o clauses, mark a positi#e atom < in the first clause, and a ne!ati#e atom fD in the second one& Thus, e are considerin! t o clauses: 9#< and fD#A& b1 Try to find any m!u of < and D& 7f you succeed, you ha#e obtained t o clauses: F(mg$ C 1 ,7 C 1G(mg$ , here <1 is <&m!u 0^D&m!u1& c1 6pply resolution, obtainin! the clause F(mg$ G(mg$ & 9heorem -.0.* CK. 3. #obinsonD& 7n the classical predicate lo!ic V+1-+11, +12+1/, 5%, AenW, a set of predicate clauses is inconsistent4 if and only if )obinsonBs #esolution rule 0to!ether ith permutation, reduction and mgurules1 allo s deri#in! a contradiction from it& 2hy is this 0second;1 refinement important8 Gecause no , instead of tryin! out all the possible unifications, e can concentrate on m!u-s that unify t o clauses& This allo s to further restrict the substitution search space 0 hen compared ith Theorem /&=&21&

2(0 The hard part of the proof is in#entin! of the follo in! Lemma -.0.-. 6ny proof $1, $2, &&& , $s U- $ 0all $-s are clauses1, here only permutation, reduction, substitution and )esolution rules are used, can be con#erted into a proof $1, $2, &&& , $s U- $B such that: a1 in the proof, only permutation, reduction, mgu and )esolution rules are used4 b1 $ can be obtained from $B by a sin!le 0possibly empty1 substitution, follo ed by a chain of permutations and reductions& Proof of 9heorem -.0.*. 6ssume, the set of clauses $1, $2, &&& , $s is inconsistent& Then, by Theorem /&=&1, there are t o proofs $1, $2, &&& , $s U- G, $1, $2, &&& , $s U- fG, here here only permutation, reduction, substitution and )esolution rules are used& 9rom clauses, these rules allo deri#in! only of clauses& Hence, G is an atomic formula& Gy +emma /&=&/, both proofs can be con#erted into proofs $ 1, $2, &&& , $s UG1, $1, $2, &&& , $s U- fG2 such that: a1 in the proofs, only permutation, reduction, mgu and )esolution rules are used4 b11 G can be obtained from G1 by a sin!le 0possibly empty1 substitution 0permutations and reductions do not apply to atomic formulas1, b21 G can be obtained from G2 by a sin!le 0possibly empty1 substitution& Thus, G1 and G2 are unifiable& +et us take their m!u, and apply it& 6s the result, e obtain a contradiction GB, fGB, here GB is G 1&m!u 0^ G2&m!u1& 6nd e ha#e obtained this contradiction from the clauses $1, $2, &&& , $s by usin! only permutation, reduction, mguB and )esolution rules& D&@&D& Proof Lemma -.0.-.& 7nduction by the Shei!ht of the resolution treeS 0see belo 1& 1& 7nduction base 3 no resolutions applied in the proof $ 1, $2, &&& , $s U- $& Then $ is obtained from some $i by a chain of permutations, reductions and substitutions& 6dd to this fact an SemptyS proof $1, $2, &&& , $s U- $i& 6nd let us compose all the substitutions into a sin!le substitution& D&@&D& 2& 7nduction step& 6ssume, e ha#e the proof $1, $2, &&& , $s U- $, containin! at least one resolution& 7ma!ine the last resolution in this proof 0< is an atomic formula1: F C , C G & F G Then $ is obtained from the formula F G by a chain of permutations,

2(1 reductions and substitutions& The proofs of the formulas F C , 7C G possess a Shei!ht of the resolution treeS less than the one of the proof $ 1, $2, &&& , $s U- $& Thus, by induction assumption, e can con#ert these proofs into permutation-reductionmguBresolution proofs of some formulas F 1 C 1 F 2 , G17 C 2G 2 such that: a1 F C can be obtained from F 1 C 1 F 2 by a sin!le 0possibly empty1 substitution sub1, follo ed by a chain of permutations and reductions& *nder sub1, the atomic formula <1 is con#erted into <& b1 7C G can be obtained from G 17 C 2G 2 by a sin!le 0possibly empty1 substitution sub2, follo ed by a chain of permutations and reductions& *nder sub2, the atomic formula <2 is con#erted into <& 'ince the clauses F 1 C 1 F 2 , G17 C 2G 2 do not contain common #ariables, the substitutions sub1 and sub2 do not intersect, hence, their union sub1*sub2 is a substitution sub 0a unifier of <1 and <21 such that: a11 9 can be obtained from ( F 1 F 2) &sub by a chain of permutations and reductions& b11 A can be obtained from ( G1G 2) & sub by a chain of permutations and reductions& 6s e kno from the abo#e, the atomic formulas <1 and <2 are unifiable& +et us take their m!u, and apply it to the formulas F 1 C 1 F 2 , G17 C 2G 2 & +et us denote by <B the formula < 1&m!u 0it is eQual to <2&m!u1& Thus, e ha#e t o formulas 91&m!u # <B # 92&m!u and A1&m!u # f<B # A2&m!u, and, by permutation and resolution, e can obtain the formula
( F 1 F 2) & mg$(G 1G 2) & mg$ &

Thus, for the formula ( F 1 F 2 ) & mg$ ( G 1G2 ) & mg$ , e ha#e a permutation-reduction-mguBresolution proof& 7t remains to sho that, from this formula, F G can be obtained by a sin!le substitution, follo ed by a chain of permutations and reductions& 'ince the substitution sub is a unifier of < 1 and <2, then, by the definition of m!u, sub^m!u&subB, here subB is some substitution& Hence, a21 9 can be obtained from ( F 1 F 2) & mg$ by the substitution subB, follo ed by a chain of permutations and reductions&

2(2 b21 A can be obtained from ( G1G 2) & mg$ by the substitution subB, follo ed by a chain of permutations and reductions& Thus, F G can be obtained from ( F 1 F 2 ) & mg$ ( G 1G2 ) & mg$ by the substitution subB, follo ed by a chain of permutations and reductions& D&@&D& 1arning6 Despite its beauty, the resolution method cannot o#ercome the !eneral complexity problem, mentioned at the end of 'ection .&(: by <hurch-$almar Theorem, in the classical predicate lo!ic, the task of reasonin! is not al!orithmically sol#able& 6nd a closer analysis sho s that all computer pro!rams implementin! resolution method run into loop in many situations, hen the formula to be pro#ed is, in fact, unpro#able& Gut in many "ractical situations, experience sho s that resolution method sol#es its task, and 3 in acceptable time& 7n particular, %rolo! interpreters are usin! resolution, and are sol#in! many practical tasks in acceptable time; 9urther readin!: +o!ic& %art 2 by 5iorgio Ingargiola #a@@an ,hinghal& 9ormal <oncepts in 6rtificial 7ntelli!ence& 9undamentals& <hapmanIHall, 1,,2, --- pp& Handbook of 6utomated )easonin!, ed& by H& 6& )obinson and 6& Voronko#, @lse#ier and 57T %ress, 2001, #ol& 7, 77& +arry 2osBs home pa!e 6bout the ubiQuity of the abo#e-mentioned unification operation in human and computer reasonin!: Kohn :. ,o8a, 3run K. Ma@umdar& 6nalo!ical )easonin!& 7n: Concept$a% Str$ct$res for Cno-%e+ge Creation an+ Comm$nication , %roceedin!s of 7<<' 200(, +?67 2=.-, 'prin!er-Verla!, Gerlin, 200(, pp& 1--(-& 0a#ailable online1&

2((

/. Miscellaneous
/.'. 4egation as ontradiction or 3bsurdity
The idea behind this approach is as follo s: let us define fG 0i&e& SG is falseS1 as SG implies absurdityS& 'o, let us add to our first order lan!ua!e a predicate constant f 0meanin! SfalseS, or SabsurdityS1, and let us replace all ne!ation expressions f9 by 9`f& Then, the three ne!ation axioms ill take the follo in! forms: L(: 0G`<1`00G`f<1`fG1, L(B: 0G`<1`00G`0<`f11`0G`f11, L'I: fG`0G`<1, L'IB: 0G`f1`0G`<1, L'': B7 B , L''B: B( B 5 f ) & 6fter this, surprisin!ly, the axiom +,B becomes deri#able from +1-+2; 7ndeed, 011 021 0(1 0.1 0/1 0-1 G`< G`0<`f1 G <`f < f Hypothesis& Hypothesis& Hypothesis& Gy 5%, from 021 and 0(1 Gy 5%, from 011 and 0(1 Gy 5%, from 0.1 and 0/1 V+1, +2, 5%W U- 0G`<1`00G`0<`f11`0G`f11& 'econd obser#ation& The axiom +10B: 0G`f1`0G`<1 can be replaced simply by f`<& 7ndeed, if e assume f`<, then +10B becomes deri#able:

Hence, by Deduction Theorem 1,

2(. 011 021 0(1 0.1 0/1 G`f G f f`< < Hypothesis& Hypothesis& Gy 5%, from 011 and 021 f`< Gy 5%, from 0(1 and 0.1

Hence, by Deduction Theorem 1, V+1, +2, f`<, 5%W U- 0G`f1`0G`<1& Third obser#ation& 6s e kno from Theorem 2&.&,: V+ 1, +2, +,, 5%W UfG`0G`f<1, in the minimal lo!ic e can pro#e /0T of + 10: S<ontradiction implies that all is ron!S& 6fter our replacin! ne!ations by G`f the formula 0G`f1`0G`0<`f1 becomes deri#able from +1-+2& 7ndeed, 011 G`f 021 G 0(1 f 0.1 f`0<`f1 0/1 <`f Hypothesis& Hypothesis& Gy 5%, from 011 and 021 6xiom +1 Gy 5%, from 0(1 and 0.1

Hence, by Deduction Theorem 1, V+1, +2, 5%W U- 0G`f1`0G`0<`f11& Thus, e see that +1 0and not +,;1 is responsible for the pro#ability of the /0T Scra:yS formula fG`0G`f<1& 7s +1 /0T as Scra:yS as +108 Kes; +et us compare: +10: fG`0G`<1 states that S<ontradiction implies anythin!S& +1: G`0<`G1 states that S7f G is true, then G follo s from anythin!S& +et us recall our Sar!umentS for + 10 from 'ection 1&(: S&&& e do not need to kno , ere < StrueS or not, if fG and G ere StrueS simultaneously& Gy assumin! that Sif fG and G ere true simultaneously, then anythin! ere trueS e !reatly simplify our lo!ical apparatus&S ?o , similarly: if G is 0unconditionally1 true, then e do not need to kno , follo s G from < or not& Gy assumin! that Sif G is true, then G follo s from anythin!S e !reatly simplify our lo!ical apparatus&

2(/ 7n a sense, the axiom +, SdefinesS the ne!ation of the minimal lo!ic, the axioms +, and +10 SdefineS the ne!ation of the constructi#e lo!ic, and +,-+11 SdefineS the ne!ation of the classical lo!ic& 7s our definition of fG as G`f eQui#alent to these SdefinitionsS8 Kes; 9heorem /.'.'. 9or any formula 9, let us denote by 9B the formula obtained from 9 by replacin! all sub-formulas fA by A`f& Then, for any formulas G1, &&&, Gn, <: V+1-+,, 5%W: G1, &&&, Gn U- <, if and only if V+1-+C, 5%W: GB1, &&&, GBn U- <B& Proof. 11 `& +et us consider a proof of V+1-+,, 5%W: G1, &&&, Gn U- <& 7n this proof: E let us replace each formula A by its StranslationS AB, E before each instance of +,, let us insert a proof of the correspondin! instance of +B, in V+1, +2, 5%W 0see abo#e1& 7n this ay e obtain a proof of V+1-+C, 5%W: GB1, &&&, GBn U- <B& 7ndeed, a1 7f some formula G is an instance of + 1-+C, then GB is an instance of the same axiom 0#erify;1& b1 0G`D1B is GB`DB, hence, if the initial proof contains a conclusion by 5% from G and G`D to D, then, in the deri#ed proof, it is con#erted into a conclusion by 5% from GB and GB`DB to DB& c1 7f the initial proof contains an instance of + ,, then the deri#ed proof contains the correspondin! instance of +B , preceded by its proof in V+1, +2, 5%W& D&@&D& 21 j& +et us recall the abo#e translation operation: for any formula 9, e denoted by 9B the formula obtained from 9 by replacin! all sub-formulas fA by A`f& ?o , let us introduce a kind of a con#erse operation 3 the re-translation operation: for any formula 9, let us denote by 9S the formula obtained from 9: a1 by replacin! all sub-formulas A`f by fA, and after this, b1 by replacin! all the remainin! fBs 0f means SfalseS;1 by f0a`a1, here a is some closed formula of the lan!ua!e considered& Rf course, for any formula 9, 09B1S is 9 0#erify1&

2(4ote& )eplacin! f by a formula preceded by ne!ation, is crucial 3 it ill allo applyin! of Theorem 2&.&,: V+1-+,, 5%W: fG`0G`f<1 instead of the 6xiom +10: fG`0G`<1& ?o , let us consider a proof of V+1-+C, 5%W: GB1, &&&, GBn U- <B& 7n this proof, let us replace each formula A by its re-translation AS& Then <B becomes <, and GB1, &&&, GBn become G1, &&&, Gn, but hat about the remainin! formulas contained in the proof8 a1 7nstances of the axioms +1-+C& L': G`0<`G1 7f G is not f, then 0G`0<`G11S is GS`0<S`GS1, i&e& re-translation yields a!ain an instance of +1& 7f G is f, then 0f`0<`f11S is f0a`a1`f<S& This formula is pro#able in V+ 1+,, 5%W& 7ndeed, 011 021 0(1 0.1 f0a`a1 U- f0a`a1`00a`a1`f<S1 U- a`a f<S Hypothesis& Theorem 2&.&,, V+1-+,, 5%W& Theorem 1&.&1 V+1-+2, 5%W& Gy 5%, from 011, 021 and 0(1&

Thus, re-translation of any instance of +1 is pro#able in V+1-+,, 5%W& L.: 0G`0<`D11`00G`<1`0G`D11 7f < and D are not f, then re-translation yields a!ain an instance of +2& 7f < is f, and D is not, then re-translation yields 0GS`0f0a`a1`DS11`0fGS`0GS`DS11& This formula is pro#able in V+1-+,, 5%W& 7ndeed, 011 GS`0f0a`a1`DS1 021 fGS 0(1 GS 0.1 f0a`a1`DS Hypothesis& Hypothesis& Hypothesis& Gy 5%, from 011 and 0(1&

2(= 0/1 U- fGS`0GS`f0a`a11 0-1 f0a`a1 0=1 DS Hence, by Deduction Theorem 1, V+1-+,, 5%W U- 0GS`0f0a`a1`DS11`0fGS`0GS`DS11& 7f D is f, and < is not, then re-translation yields 0GS`f<S1`00GS`<S1`fGS1& This formula is pro#able in V+1-+,, 5%W& 7ndeed, 011 021 0(1 GS`f<S GS`<S fGS Hypothesis& Hypothesis& Gy 5%, from 6xiom +,& V+1-+,, 5%W U-0GS`f<S1`00GS`<S1`fGS1& 7f < and D both are f, then re-translation yields 0GS`ff0a`a11`0fGS`fGS1& This formula is pro#able in V+1-+,, 5%W& 7ndeed, 011 021 0(1 U- fGS`fGS Theorem 1&.&1 V+1-+2, 5%W& Theorem 2&.&, V+1-+,, 5%W& Gy 5%, from 021, 0(1 and 0/1& Gy 5%, from 0.1 and 0-1&

Hence, by Deduction Theorem 1,

U0fGS`fGS1`0J`0fGS`fGS 6xiom +1, J is GS`ff0a`a1& 11 U- J`0fGS`fGS1 Gy 5%, J is GS`ff0a`a1&

Thus, re-translation of any instance of +2 is pro#able in V+1-+,, 5%W& L): BC 5 B 7f G is not f, then re-translation yields a!ain an instance of +(& 7f G is f, then re-translation yields #ia 7 ( f C ) the formula 7 ( 7( a 5 a )C ) & This formula is pro#able in V+1-+,, 5%W& 7ndeed,

2(C 011 021 0(1 0.1 0/1 7 ( a 5 a )C 5 7 ( a 5 a ) 77 ( a 5 a ) 5 7 ( 7( a 5 a )C ) 0a`a1`ff0a`a1 a`a 7 ( 7( a 5 a )C ) 6xiom +(& 9rom 011, by the <ontraposition +a & Theorem 2&.&.: V+1, +2, +,, 5%W U6`ff6 Theorem 1&.&1 V+1-+2, 5%W& Gy 5%, from 0(1, 0.1 and 021&

Thus, re-translation of any instance of +( is pro#able in V+1-+,, 5%W& L*: BC 5 C 'imilarly to +( 3 re-translation of any instance of +. is pro#able in V+1-+,, 5%W& L-: B 5 ( C 5 BC ) )e-translation yields a!ain an instance of +/& L/: B 5 BC )e-translation yields a!ain an instance of +-& L0: C 5 B'C )e-translation yields a!ain an instance of +=& LG: ( B 5 , ) 5 (( C 5 , ) 5 ( BC 5 , )) 7f D is not f, then re-translation yields a!ain an instance of +C& 7f D is f, then re-translation yields 7 B 5 ( 7C 57 ( BC )) & Gy Theorem 2&.&100b1, this formula is pro#able in V+1-+,, 5%W & Thus, re-translation of any instance of +C is pro#able in V+1-+,, 5%W& Hence, re-translations of all 0i&e& +1-+C1 axiom instances are pro#able in V+1+,, 5%W& 2hat about applications of 5% in the initial proof8 7f the initial proof contains a conclusion by 5% from G and G`D to D, then the follo in! situations are possible: a1 7f G and D are not f, then, in the deri#ed proof, this conclusion is con#erted into a conclusion by 5% from GS and GS`DS to DS&

2(, b1 7f G is f, and D is not, then, in the deri#ed proof, this conclusion is con#erted into a conclusion by 5% from f0a`a1 and f0a`a1`DS to DS& c1 7f D is f, and G is not, then, in the deri#ed proof, this conclusion is con#erted into three formulas: GS, fGS, f0a`a1& To deri#e f0a`a1 from GS and fGS, e can use 5% and Theorem 2&.&,: V+1-+,, 5%W U- fGS`0GS`f0a`a11& d1 7f G and D are both f, then, in the deri#ed proof, this conclusion is con#erted into three formulas: f0a`a1, ff0a`a1, f0a`a1& 'imply drop the third formula from the proof& Thus, the re-translation operation, hen applied to all formulas of a proof of V+1-+C, 5%W: GB1, &&&, GBn U- <B, yields a seQuence of formulas that are pro#able in V+1-+,, 5%W from hypotheses G1, &&&, Gn& Hence, so is <& D&@&D& This completes the proof of Theorem -&1&1& orollary /.'..& a1 6 formula < is pro#able in the minimal propositional lo!ic V+1-+,, 5%W, if and only if V+1-+C, 5%W U- <B& b1 6 formula < is pro#able in the constructi#e propositional lo!ic V+ 1-+10, 5%W, if and only if V+1-+C, f`G, 5%W U- <B& c1 6 formula < is pro#able in the classical propositional lo!ic V+ 1-+11, 5%W, if and only if V+1-+C, f`G, +B11, 5%W U- <B& Proof. a1 <onsider an empty set of hypotheses in Theorem -&1&1& b1 7f V+1-+10, 5%W U- <, then V+1-+,, 5%W: G1, &&&, Gn U- <, are instances of the axiom +10& Gy Theorem -&1&1, V+1-+C, 5%W: GB1, &&&, GBn U- <B& 6s established abo#e, GB1, &&&, GBn can be pro#ed by usin! the axiom schema f`G, i&e& V+1-+C, f`G, 5%W U- <B& D&@&D& ?o , if V+1-+C, f`G, 5%W U- <B, then, c1 7f V+1-+11, 5%W U- <, then V+1-+,, 5%W: G1, &&&, Gn U- <, here hypotheses are instances of the axioms +10 and +11& )eturn to case 0b1& D&@&D& orollary /.'.)& a1 6 formula < is pro#able in the minimal predicate lo!ic V+1-+,, +12-+1/, 5%, AenW, if and only if V+1-+C, +12-+1/, 5%, AenW U- <B& here hypotheses

2.0 b1 6 formula < is pro#able in the constructi#e predicate lo!ic V+ 1-+10, +12+1/, 5%, AenW, if and only if V+1-+C, f`G, +12-+1/, 5%, AenW U- <B& c1 6 formula < is pro#able in the classical predicate lo!ic V+1-+11, +12-+1/, 5%, AenW, if and only if V+1-+C, f`G, +11B, +12-+1/, 5%, AenW U- <B& !7ercise /.'.'. %ro#e the <orollary -&1&(&

Você também pode gostar