(Thru OCA) : Dr. Alan D

GEORGIA INSTITUTE OF TECHNOLOGY
OFFICE OF CONTRACT ADMINISTRATION
Dr. Alan D.
SPONSORED PROJECT INITIATION

Date: May 11,1976
Project Title:
"Applications of Nonstandard Analysis to Mathematical Physics"

G-37-603
Project No:
Sloan
Project Director:
National Science Foundation
Sponsor:
From
Agreement Period:
Until
7/1/76
12/31/77*
*12 month budget period plus 6 months for submission of required reports, etc.
Type Agreement: Grant No. MCS76-07543
$5,800 NSF
1,803 GIT (3-37-313)
$7,603 'MAL
Amount:
Reports Required:
Annual Letter Technical, Final Report

Sponsor Contact Person (s):
Contractual Matters
Technical Matters
(thru OCA)
James L. Bostick
Grants Officer
Washington, D. C.
20550
Dr. William G. Rosen

Program Director
Mbdern Analysis and Probability
Washington, D. C. 20550
I
Defense Priority Rating: NONE
Assigned to:
(School/Laboratory)
Matheratics
COPIES TO:
Project Director
Library, Technical Reports Section
Division Chief (EES)
Office of Computing Services
School/Laboratory Director
Director, Physical Plant
Dean/DirectorEES
EES Information Office
Accounting Office
Project File (OCA)
Procurement Office
Project Code (GTRI)
Security Coordinator (OCA)
Other
1,11;orts Coordinator (OCA)
V
OFFICE OF CONTRACT ADMINISTRATION
SPONSORED PROJECT TERMINATION

Date:
Project Title:
Applications of Nonstandard AnalysisMathematical Physics
Project No: G-37-603

Project Director: Dr. Alan D. Sloan
Sponsor: National Science Foundation
Effective Termination Date:
12/31/79
12/31/79
ClearncofAutigChares:
Grant/Contract Closeout Actions Remaining:
Final Invoice and Closing Documents

_x Final Fiscal Report via FCTR
_x Final Report of Inventions
Govt. Property Inventory & Related Certificate
Classified Material Certificate
Other
( Scho ol/ UN:34MAX
Assigned to: Mathematics

COPIES TO:
Project Director
Library, Technical Reports Section
Division Chief (EES)
EES Information Office
School/Laboratory Director
Project File (OCA)
Dean/DirectorEES
Project Code (GTRI)
Accounting Office
Other
Procurement Office
Security Coordinator (OCA)
/Reports Coordinator (OCA)

CA- 4 (1/79)

ATLANTA GEORGIA 30332
June 28, 1977
MA1HEMATICS
James L. Bostick
Grants Offices
Washington, D.C. 20550
Dear Mr. Bostick:
This is the annual _teclanic,al__letter for NSF grant
number MCS 76-07543 awarded to Georgia Institute of Technology
with Dr. Alan D. Sloan as project director.
The main results fo the research conducted from 7/1/76
to 6/30/77 under the above grant can be found in two papers:
1. "An Application of the Nonstandard Trotter Product
Formula" submitted to the Journal of Mathematical Physics; and
2. "A Note on Exponentials of Distributions," in
preparation.
Preprints are available on request and reprints will
be sent when available.
The project director was invited to the Iowa Symposium
on Infinitesimals from 6/1/77 to 6/5/77 and delivered an
hour address entitled "Nonstandard Perturbations."
Sincerely.
Alan D. Sloan
ADS:mc

ATLANTA. GEORGIA 30332
IATHEMATICS
June 16, 1978
Mr. James L. Bostick

Grants Offices
Washington, D.C.
20550
Dear Mr. Bostick:
This is the annual technical letter for NSF grant
number MCS 76-07543 awarded to Georgia Institute of
Technology with Dr. Alan D. Sloan as project director.
1. Dr. Sloan's paper "An Application of the Nonstandard Trotter Product Formula" appeared in the Journal
of Mathematical Physics. Two reprints are enclosed.
2. Dr. Sloan completed work on a paper entitled
"A Note on Exponentials of Distributions" which was
accepted for publication in the Pacific Journal of Mathematics. Preprints are available on request.
3. Dr. Sloan received an invitation to an Oberwolfach
conference July 2-July 8 in Germany. He prepared a lecture
on nonstandard aspects of evolution equations in Hilbert
space.
4. Dr. Sloan investigated Sobolev inequalities
connected with nonstandard distribution theory.
Sincerely,
Alan D. Sloan
ADS:mc
Enclosure
An application of the nonstandard Trotter product formula

Alan Sloan')
School of Mathematics, Georgia Institute of Technology, Atlanta, Georgia 30332
(Received 21 March 1977)

The nonstandard Trotter product formula is used to extend the Feynman integral interpretation of
solutions to the Schrbdinger equation in the presence of a highly singular potential.
We seek an answer to the question of when a quantum

mechanical Hamiltonian H determines a system which
evolves according to a Feynman path integral. I Specifically, we want to know when there is a potential V so
that, upon setting k =1, and fixing t > 0
exp(- itH)u=limT,,, v (u),
(1)
u in L 2 (IR3 ),
where T" v (u) is defined as the integral
(T, vu)(x 0 )=-_- c, fro, exp[iS(xo , x, x 0 ;n,t, V)]
exp(- itH)uz exp[- it (H + V u.
xu(x n)dx dx,
(2)
and
co = (27rit/rnn)-3012, M>
,X77.1)
(6)
By the Trotter product formula and the transfer principle, for each K in *IN
exp[- t(Ho + VK)]=S-lina(exp( itH o/n) exp(- itV K/n)]".
n in *111
(7)
Combining (6) and (7) we find
and
S(X0,
X n ; n,
t, V) =
c,
J.1
112(
XX
"- 1) 2
2 (t/n)
An explic t computation' shows that
T 0, v [exp(- itHo/n) exp(- itV/n)r,
V(x j)- .
(3)
Ho = - A/ 2/12 on L 2 (1R3 )
(8)
i
(4)
For all finite n and K, (4) shows that
T 0, v lc = [exp(- itHo/n) exp(- it VK/n)] ",

( 5)
for V any real measurable function. Consequently,

Nelson2 was able to employ the Trotter product formula' to verify (1) in case H is the closure of the operator
sum, H0 + V, and V is chosen so that H0 + V is essentially self-adjoint. Nelson 2 and Faris.' extended this type
of formula to more singular potentials by including an
additional limiting operation after analytic continuation
in the mass parameter m. Although there has been
progress in extending Trotter's formula for exp[- t(H o
+ V)], 5-1 when V is singular relative to I/ 0 , analogous
results are not known for the unitary groups exp[it(H o
+ V)]. Through the use of elementary nonstandard techniques we shall find a formula of the type (1) valid for
singular potentials.
For details on nonstandard analysis and notation we
refer the reader to Refs. 4 and 8-10. Here wd adopt
the convention that Xc: *X for any set X. If X is a topological space, a in X, b in *X, we write a= b if and only
if b is in the monad of a. If
1) we say a is the standard part of b and that a and b are infinitely close. We
use the Euclidean topology on IR" and the norm topology
on L 2 (1R3 )
a) Supported
exp(- itH)u=lim [exp(- itHo/n) exp(- itV K/n)]".

n in *IN
where
2495
Suppose I/ is a self-adjoint operator on L 2 (IR3 ) which

can be approximated in the generalized strong sense by
bounded perturbations of H o ; i.e. , there is a sequence
V K of bounded self-adjoint operators on L 2 (IR3 ) such that
the resolvents of H o + VK converge strongly to those of
H. Then, (Ref. 11, 502), providing {H o + VK} is uniformly bounded below, it follows that exp(- itH) is the strong
limit of exp[it(H + V K)]. Thus, for all K in *IN- IN,
where N=-{1, 2, - }, and for all u in L 2 we have
in part by NSF MCS 76-97543.
Journal of Mathematical Physics, Vol. 18, No. 12, December 1977
(9)
so by the transfer principle (9) holds for all n, K in

*IN. Thus, for each fixed infinite K, (8) yields
exp(- itH)u- lim T 0 v K (u).
(10)
n n;
i
For each n, K in IN, (2) shows that T n, vic (u) is given

by an action integral. The transfer principle then implies that for all n, K in *N formulas (2) and (3) remain
valid with V K replacing V. Consequently, we may conclude that
exp(- itH)u( ) 0 lim
co f
exp[iS(.,x,x,pi,tV,)]
n in *IC
xu(x 0)dx dx n .
Lemma: Let X be a separable normed linear space.
Let f: X -*X be bounded linear operators which are
uniformly bounded in the sense that there is an M in IN
IN U {0}. If fo -f o pointsuch thatX Ia;1;2
f II _ for
in all n in *IN
*N, there is an /V in *IN such that
wise on
I
n >N impliesfo (x)'-- f 0 (x) for all x in X.
Proof: Let C be a countable dense subset of X. Let
X be a positive infinitesimal. Choose N 0 in *IN so that n
> N, implies II fn(c) - fo(c)11 < X. Let /V be an upper bound
for IN,: c in Cl; N exists by Ref. 5, p. 59. Let x in X
be arbitrary. FiX 6> 0 in R. There is a c in C so that
Copyright 1977 American Institute of Physics
2495
IIcxII < 5. Then for n>N

II fn(x) fo(x)II
II fn(x) fn(c)II
f n(c) fo(c)II +.11 fo(c) fo(x)II
<MIIx- cll +X+Mile xil <3.1V16.

Because 6 is arbitrary we conclude that f(x) 2=fo(x)
whenever n>N.
Q. E. D.
For the necessary approximation theorem in case (b)

see Ref. 12 where the VK 's are defined by truncation.
In cases (a) and (c) see Ref. 13 where the VK 's are given by regularization of V. See Ref. 13 for additional
examples
Similar techniques apply in dimensions other than 3
and for other H o 's.
Combining this lemma with the previous conclusion,

noting that exp( itH) and the T,,,v ic are uniformly
bounded by 1, yields the following:
Theorem: If VK are bounded self-adjoint operators
such that Ho + VK converge to H in the generalized strong
sense as n 00 in IN, then for each fixed positive infinite
integer K there is an N in *IN such that for all n> N and
for all u in L 2
exp(itH)u() --'' cn "R oo exp[iS( ,x, x;n, t, V K )]
xu(x n)dxdx,
(11)
in *L 2 .
To compare the types of potentials covered by formulas (1) and (11) we note that 2 (1) holds if V is in LP
p =2, whereas (11) holds whenever H is the generalized strong limit of bounded self-adjoint perturbations of H o . Examples of such H' s can be found by defining H to be the form sum of H o and V when either
(a) V is in L 5 + L , p >1;
(b) V> 0 is locally in L i outside a closed set of measure zero;
(c) V is a delta function distribution concentrated on
the surface of a compact C 1 hypersurface in Ie.
2496
J. Math. Phys., Vol. 18, No. 12, December 1977
Feynman and A. Hibbs, Quantum Mechanics and Path

Integrals (McGraw-Hill, New York, 1965),
1 B.
2 E.
Nelson, "Feynman Integrals and the SchrOdinger Equation," J. Math. Phys. 5, 332-43 (1964).
3 H. Trotter, "Approximation of Semi-groups of Operators,"
Pacific J. Math. 8, 887-919 (1958).
4 M. Davis, Applied Nonstandard Analysis (Wiley-Interscience,
New York, 1977).
5 P, Chernoff, "Product Formulas, Nonlinear Semigroups and
Addition of Unbounded Operators," Mem. Am, Math. Soc.
140, 1-121 (1974).
6 W. Faris, "The Product Formula for Semi-groups defined by
Friedrics Extensions," Pacific J. Math, 21, 47-70 (1967).
7 T. Kato, "Trotter's Product Formula for an Arbitrary Pair
of Selfadjoint Contraction Semigroups," Berkeley preprint.
6 A. Robinson, Nonstandard Analysis (North-Holland, Amsterdam, 1974).
3 A. Robinson and A. Lightstone, Non-Archimedian Fields
and Asymptotic Expansions (North-Holland, Amsterdam,
1975).
' 0 K. Stroyan and W. Luxemburg, Introduction to the Theory
of Infinitesimals (Academic, New York, 1976),
11 T, Kato, Perturbation Theory for Linear Operators
(Springer, Berlin, 1966).
12 W. Faris, Self Ad joint Operators (Springer, Berlin, 1975).
13 1. Herbst and A. Sloan, "Perturbation of Translation Invariant Positivity Preserving Semigroups on L 2 (EN)," to
appear in Trans. Am. Math, Soc.
-
Alan Sloan
2496
EY-37-6
ATLANTA. GEORGIA 30332
MATHEMATICS
June 18, 1979
Mr. James L. Bostick

Grants Offices
Washington, D. C. 20550
Dear Mr. Bostick:
This is the annual technical letter for NSF grant
MCS 76-07543 awarded to Georgia Institute of Technology
with Dr. Alan D. Sloan as project director. For the
period 7/1/78 to 6/30/7? .
1. Dr. Sloan delivered an invited talk at Oberwolfach,
Germany, July 2-July 8 on nonstandard aspects of evolution
equations;
2. Dr. Sloan attended the International Congress of
Mathematicians in Helsinki in August travelling on a Mathematics Research Council Travel Grant and discussed his work
on nonstandard analysis;
3. Dr. Sloan's paper "A Note on Exponentials of Distributions" appeared in the Pacific Journal of Mathematics,
Vol. 79, No. 1;
4. Dr. Sloan prepared and submitted the manuscript
"Strong Convergence of Schrodinger Propagators" for publication;
5. Dr. Sloan continued work on nonstandard evolution
equations.
6. Dr. Sloan initiated investigations into explicit
solutions of partial differential equations using nonstandard techniques.
Sincerely,
Alan Sloan
cm
6-3'7-603
PACIFIC JOURNAL OF MATHEMATICS
Vol. 79, No. 1, 1978
A NOTE ON EXPONENTIALS OF DISTRIBUTIONS

ALAN SLOAN
Nonstandard analysis is used to discuss nonlinear functions of distributions. An application is given to obtain a
generalized Trotter product formula. The strong resolvent
topology is discussed from a nonstandard point of view.
Enlarging the real number system to include infinite and infinitesimal quantities enabled Laugwitz [5] to view the delta function
distribution as a point function. Independently Robinson [7] demonstrated that distributions could be viewed as generalized polynomials.
Luxemburg [6] presented an alternate picture of distributions as
generalized functions within the context of Robinson's theory of
nonstandard analysis and it is a special case of this point of view
that we take here. Once one accepts distributions as generalized
functions, the composition map provides a natural method of defining nonlinear functions of distributions. Unfortunately, the difficulties in the standard attempt to define a function, f, of a
distribution, r, (by first writing r as a limit, in some appropriate
sense, of smooth standard functions r % , next composing these
approximations with f, and finally taking the limit of f o r%) still
remain in the nonstandard theory. In particular, this procedure
may not lead to a distribution and even when it does the distribution obtained may depend on the representation of the original
distribution as a generalized function. Thus, at present, there is
no comprehensive theory of nonlinear functions of distributions.
The development of such a theory may well proceed along
alternate tracks depending on the applications intended. One use
for distributions occurs in the study of perturbations of selfadjoint
operators in Hilbert space. In some perturbation problems the
pathology of obtaining compositions which do not represent distributions may be avoided by considering only bounded functions of
distributions. For example, in the Trotter product formula one
works with bounded exponentials. Even though, in this case, the
exponentials of distributions may be identified with standard distributions it is the conclusion of this paper that such an identification
should not be made. It is the author's view that functions of

distributions should be regarded as generalized functions but not
distributions. The utility of this view is illustrated by a nonstandard version of the Trotter product formula.
For an introduction to nonstandard analysis and its relation to
distributions, see [10]. In [11] a square root for the delta function
207
ALAN SLOAN
208
was defined nonstandardly.

Let *X be a 3-incomplete ultrapower of a structure X containing the real numbers, R. We shall use the convention that Ac*
for all sets A considered. If T is a topological space with u in T
and v in * T then we shall write u v if and only if v is in
fl Y-1': 21 is an open set containing u}. In this case we say u is
the standard part of v, relative to the given topology and that v
is near standard. We write u = st(v). We always assume that
RP has the Euclidean topology and all Hilbert spaces have the norm
topology.
If f and g are Lebesgue measurable on *S for some standard
fgdx, proset S which is measurable, we shall write <f, g> for
s
viding the integral exists. If, in fact, f and g are standard funcfgdx, by the transfer principle.
tions then <f, g>
Let W be an open subset of R. Let B. be an increasing
sequence of relatively compact subsets of W whose union is W and
whose closures are in W. If _gr
C (B.) is given the inductive limit topology then the elements of
the dual of _0", are
the distributions on W. In order to define a function of a distribution we first regularize the distribution.
Let p be a Cr (RP) function satisfying 0 5 p < 1, p(x) ---- 1 if
II S 1 and p(x) = 0 if II x II 2. Set
7,,(x) = P(npx)(S p(n ot)dt)
where no is chosen so that

distance (B,13:+1) > 2/n0
Choose R. to be a Cr (R') function satisfying R. = 1 on B_ 1 , 0 S
13. < 1 and support (SO c
If x is in B. and 7(x y) # 0 then Itx y II < 2/n o so that y
y) then
If 7(x ) denotes the function y
is in
on
W define
if
x
e
B..
For
any
distribution
T
C:'(B,,,,)
.) is in
a function T. by
T.(x) = f3.(x)T(7,.(x
))
for n in N. Then T. is in C,-(RP) and support (T) c B. By the

transfer principle T is in *C:* (R') and support (T.) c *B for n in
co in N it follows that
*N. Further, since T. T in _0' as n
(1)
T(f) T(f)
for all n in *N N and f in .0 where T.(f) = <T, f>.
209
The T,,, are called the regularizations of T.

DEFINITION 1. Fix n infinite in *N. Let g be a complex valued
function of a complex variable defined on the range of T. Then
we define g(T) to be g o T.
Thus, g(T) is a generalized function in the sense that it is an
internal *C-valued function defined on * W. Since TA (x) = 0 if x is
not in we find that g(T)(x) = g(0) if x is not in *B..
Certain generalized functions h: * j 1.' *C define distributions r
according to the rule
(2)
r(f) = st(<11,, f>)
for f in 1.. Here, st (x,) is the standard part of X,, if X, is a finite

hypercomplex number, and undefined otherwise. Thus not every
function of a distribution will determine a distribution according
to (2). If g is a bounded measurable function then g(T) always
defines a distribution by (2).
If T is given by a locally
function F on W, (i.e., if T(f)
f) for all f in _0) then g of T is naturally defined within
standard distribution theory as the distribution f <g o F, f> providing the composition is defined. A simple example suffices to
show that the distribution determined by g(T) need not be g o F.
EXAMPLE 2. Let E be the complement of the Cantor set, in
[0, 1], of measure 1/2 obtained by repeated deletions of 2n -1 open
intervals of length 2 -2", n = 1, 2, from [0, 1]. Let F be the
characteristic function of E and define T(f) = <F, f>. Let W =
(-2, 2) in R 1 . Choosing B, so that [0, 1] c B, gives
TT(x) = T(7.(x ))
<F, 7.(x )>
`(y)7.(x y)dy >
0 for all n in N
and all x in [0, 1]. By the transfer principle 71,(x) > 0 for all n in
*N and x in *[0, 1]. Let g(x) = 1 if x
0 and g(0) = 0. Then
g(T)(x) = 1 if x is in 10, 1] while g o F(x) = 1 if x is in E and
g o F(x) = 0 if x is not in E. Choosing f in Cr( 2, 2) with f = 1
on [0, 1] shows g(T)(f) = <g(T), f> = 1 while
<9 o F, f> = Lg(F(x))f(x)dx
= Eg(F(x))f(x)dx
= Lf(x)dx =
measure (E) = 1/2 .
ALAN SLOAN
El0
Thaw
g F(f) = <g F, f>
<g(T), f> = g(T)(f) .
The necessity of considering only bounded "smooth" g's is now

demonstrated, if one is to obtain a generalization of composition of
functions. In Lemma 5 we obtain such a generalization.
LEMMA 3.
Let f be in * Li(K, dx) where K is a subset of R.

Then M = {x in * f(x) r 0) is contained in an
Let iIf II 1
internal measurable subset P of *K and ?measure (P) 0.
0.
Proof.
Let A m = {x in *K; I f (x)I > 1/m). Then
U. , A..
Since
>
dx > meas (A m)/m ,

dm
we find that meas (A m) P,$ 0 for each m in N. By definition A i c A,c

A,
-.
Extend {A m} to an !internal sequence. Let I = {m in *N: 2

m implies meas (A k ) < 1/k and A k _, c A k ). I is internal and
contains N. Since N is external there is an w in (*N N) n I.
Choose P = A,,. Then meas (P) 0 and U. A,,, c P.
LEMMA 4. Let F be a locally L' function on W. Let T be
the distribution T(f) = <F, f>. For every compact K in W and
every infinite n in *N
IIT Fll,
where
II IL
Proof.
is the L' norm on *K.
Choose an open set Q with compact closure such that

KcQcQc W.
For n in N sufficiently large, R. = 1 on Q, and x in K implies

support (7(x -))c Q. - Consequently, for n in N sufficiently large
Tm (x) = ("fm *F)(x). But 7.*F F in Li(K) so for n infinite IIF
7.*F IIl 0 while by definition T. = 7.*F so }IF T.II, F.& 0.
LEMMA 5. Let g be a bounded uniformly continuous function
defined on a closed interval containing 0 and the range of a locally
L' function F. Let T be the distribution defined by F: T(f) =
<F, f) for f in .0c. Then g o F is the distribution determined by
g(T)
A NOTE ON EXI'ONENTIALS OF DISTRIBUTIONS
211
Proof. We first must show g(T) is defined. There is a closed

interval on which g is defined and which contains the range of F.
Since 0 < f3 < 1 and 7dx=1 it follows that the range of R,(77,F)--=71 is in the closed interval and so the range of T is in the domain
of g for all n in N. By transfer g(T) is defined for all n in *N.
Fix n infinite and define g(T) = g(T).
We are to prove that for every f in 2'
<g(T), f>
F, f> .
Let K be a compact set containing the support of f. Let A be

an internal subset of *K containing {x in *K: T(x) F(x)} and
having measure X 1= 0. Since g is uniformly continuous, x in *K
A implies g(T(x)) g(F(x)). Let M in N be a bound for g. Let E
be a positive real number. Let in be the Lebesgue measure of K.
Then
I <g(T), F> <g 0 F, f>1
A
!g(T) g(F)I f dx +
< 2.111 ;! f IlcoX
Since
Blif
<
2c
A I g(T) g(F)I f dx
KA
Ilf 11-nt
is arbitrary, this completes the proof.
EXAMPLE 6. Let o be the delta function o(f)=f(0), f in _"(11 1).

Choose no = n and B = (n, n). Then, 3(x) = 7(x). For all n in
N, support (3)c(-2/n, 2/n) and 0 3, < 0.9. Thus for t > 0 in R,
0 < Cu* < 1 and if Ix I > 2/n then 6- ""(x) = 1. By the transfer
principle choosing n in *N N and setting g(T) = g(T) we find
0 < g(T) <1 and g(T)(x) =1 if ix! >2/n. Thus
(g(T), f>
<1, f>, f in .0 so
1 is the distribution determined by c".

Next we wish to discuss exponentials of distributions in connection with the Trotter product formula. This requires a discussion
of the topology of strong resolvent convergence, which we now
give. More details on this topology are to be found in the appendix.
DEFINITION. Let SA(K) be the set of all selfadjoint operators
on a Hilbert space K. A neighborhood basis of a selfadjoint operator
S in the topology of strong resolvent convergence is given by the
collection
{G(S, E, 6): E is a finite subset of K, 0 < e < 00)
211
ALAN SLOAN
where G(S, E, e)
(T in SA(K): 11(T + i)- 'h (S + i) -1101 < e, h in
E}. We refer to the elements of *SA(K) as the internal selfadjoint
operators on *K. Information about *SA(K) may be obtained by
transfer. For example, "For all T in *SA(K), A, in *C, X#0, 11(T+
x) -i H < I im (x) 1 --"' is obtained by transfer from "For all S in
SA(K), X in C, # 0, 11(S + x) -1 11 < 1 Im (X)
LEMMA 7. An internal selfadjoint operator T is near standard
in the topology of generalized strong convergence if and only if
there is an S in SA(K) such that
(T + i) - 'h (S + i) - 'hil
0 for all h in K .
Proof. S = st(T) if T is in
{
n *G(S, E, r): E is finite, 0 < r <}.
But T is in *G(S, E, r) if
11(T + i)- 'h (S +
< r for all h in E, since E is finite.
LEMMA 8. Let S belong to SA(K) and T to *SA(K). Then

T S if and only if
(3)
eitrh
e itsh
for all finite t in *R and h in K.

If T and S are bounded from below by some finite number.
Then (3) may be replaced with
(4)
e -trh
e -ts
for all finite nonnegative t in *R and h in K.

Proof. We give the proof that T S implies (3) for t positive
and finite. For any P in *SA(K) let R(P) + 1)-1 and
E(P) = es". Let h be an arbitrary element of K. Then, by definition of T S, one has R(T)h R(S)h. Since R(T), R(S) are
contractions it follows that R(T)a R(S)b for all near standard
a, b in *K with a P.1 b. In particular, since s -+ E(sS) is strongly
continuous we conclude that E(sS)h is near standard for all finite
s in *R, so that 11 (R(T) R(S))E(sS)hll ftf 0. Consequently, for all
positive real X, 11E((t s)T)(R(T) R(S))E(sS)hil < "X. Transferring equation (2.27) of Reference 3, page 501, gives
R(T)(E(tT) E(tS))R(S) = E((t s)T)(R(T) R(S))E(sS)ds
0
A NOTE ON EXPONENTIALS OF DIS'T'RIBUTIONS
213
Xt for all
and provides the estimate IIR(T)(E(tT) E(tS))R(S)hil
positive real X so that R(T)E(tT)h F,e E(tS)R(S)h. For all t in *R,
E(tS)R(S) = R(S)E(tS), follows by transferring the corresponding
statement with t in R. Since E(tS)h is near standard we conclude
R(T)E(tT)h R(T)E(tS)h .
(5)
Next we argue
E(tT)R(S)h E(tT)R(T)h = R(T)E(tT)h R(T)E(tS)h

R(S)E(tS)h = E(tS)R(S)h
Since the range of R(S) is dense in K and since E(tS), E(tT) have
finite norms we conclude that E(tT)h E(tS)h.
If t is finite and negative the proof follows as above upon
replacing T and S by T and S respectively.
For the proof of the converse we assume e"Theush for all finite
t in *R and h in K. From Lemma 12, appendix, it suffices to prove
(iT + 1) - 'h (iS + 1) - 'h
for all h in K. This is done via the transferred Laplace transform

formula (iT + 1) - 'h = e- te"rhdt. Let X, E > 0 be arbitrary real
numbers. First choose a positive real number c so that e'dt<e.
11( iT + 1)lh (iS + 1 )'11 11
2E
c`;! (e' enh11 dt
+ e- tXdt < + .
Thus, (iT +1) -'h ry (iS + 1) - 'h for all h in K.

The proof for the case of T and S finitely bounded below
follows similarly.
REMARK. The proof above closely parallels Kato's proof [3, p.
502] of an analogous standard result.
REMARK. The restriction of (3) to finite t in *R is not superfluous. Let n be a positive infinite integer, T = 1/n, S = 0, and
t = n. Then T 1%4 S but
e"rh
eih h = eitsh
EXAMPLE 8. Let ho: R 9 [0, 00) be continuous and suppose cuo

is in V and is positive definite. Let Ho = FM(ho)F' where F is
the unitary Fourier transform on L2 (RP) and M(h o) is the selfadjoint
214
ALAN SLOAN
operator given as multiplication by the function h o . Examples of

such operators are given by ho(k) = 'kr when Ho = 4 and ho(k)
> 0. For a complete discussion of such operators,
VIkr + 9n2 ,
see [2]. Let r be a real distribution on R 0 which satisfies for some
0 < a < 1 and b > 0
I 2)
a(H,f,
+ b Ilf 112
for all f in C,'(1V). Such a r is called a small form perturbation

of Ho. As shown in [2] the form sum of Ho and the continuous
extension of r to the form domain of H, is selfadjoint and bounded
below. We denote this standard operator by H, + 7. As concrete
examples choose Ho = 4 and r the delta function concentrated on
the surface of a sphere in p > 1 dimensions and r = 6, the delta
function in p = 1 dimensions. Also when p = 1, 7 = e' cos (e') is
a small form perturbation of 4. See [2] for more examples.
Let 7. denote the regularizations of r. Then, defining Ho + r
as the operator sum, it follows from the strong convergence of the
resolvents of Ho + v to those of Ho + r, [2], that
Fk:$
H, + r
for all positive infinite integers, rt.

EXAMPLE 9. Let Ho, V be nonnegative selfadjoint operators
on a Hilbert space K. Suppose D((Hor) fl D((V)") is dense. Define
the truncations of V by
V =
V if V 5 n
lo if V > n
via the spectral theorem. Let H. + V denote the standard selfadjoint operator defined as the form sum of Ho and V in [1]. Then,
since (H, + + i)' converges strongly to (H, + V + i)', [1], we
conclude
+ V Rd Ho +
for all positive infinite integers, n. See [1] and [4] for concrete
examples.
We now proceed to discuss the Trotter product formula for
given form sums. A discussion of form sums may be found in
[1]. Here H. is a selfadjoint operator on a Hilbert space K and H
is the selfadjoint operator on K given as the form sum of H, and
V in the following three cases:
4
Case 1. V is a selfadjoint operator and the operator sum H,+
215
V is essentially selfadjoint.
Case 2. Ho and V are bounded below selfadjoint operators
with dense form domain intersection; and
4
Case 3.
V is a small Hermitian form perturbation of Ho .
In Case 1, H is just the closure of Ho + V and the Trotter

product formula holds [12]:
(6)
e" = strong limit (e"o'"e''''''')
uniformly for t in compact subsets of [0, c-.), where x =

If
H V are bounded below then (6) also holds with X = 1.
In Case 2 Kato has shown [4] that (6) holds for x = 1 but
in general (6) is not known for X --- i and in fact there has been
essentially no progress in extending (6) for X = i beyond Case 1.
In Case 3, not only is Trotter's product formula unknown for
= i and X = 1 but in general it makes no sense. For
example, if Ho = d2/dx 2 and V = 8, the Dirac delta function on
Lz(RI), then e'7" is undefined in any standard sense. In some
examples of Case 3 such as Ho = d2/de and V(x) = ez cos (ex) the
formula (6) makes sense but its validity is unknown for x = 1.
Elementary nonstandard analysis provides a convenient framework for obtaining a type of product formula in the cases discussed
above.
THEOREM 10. A Product Formula:
Let Ho, H be in SA(K) and V in *SA(K) where K is a separable Hilbert space. Suppose the closure [Ho + V] of the operator
sum of H o and V is in *SA(K). Also assume H d [Ho + V].
Then there is an N in *N such that for all n > N, h in K and
nonnegative finite t in *R:
e 1tNh
(7 )
e 2 2 1-1 0/se it isyth
where X =
If also, H, Ho, V > c for some c in R then (7) holds with X= 1.
Proof.
(8)
By Lemma 8
e
ltHh F=4, enuro+vjh
for all h in K and all finite t in *R.

Fix a to be a positive infinitesimal. Fix h in K and M in N.
216
ALAN SLOAN
It follows from the transferred Trotter product formula of Case 1

that there is a Q(h, M) in *N, (we suppress the a dependence), such
that q > Q(h, M) implies
(9)
Ile :itplo+V]h
( e 2fIlc, , q e ;0 11q)9h I
<a
for all t in *[0, M]. Let Q(h) be an upper bound for {Q(//, M): M
is in N}, which exists by [8, page 59]. Then q > Q(h) implies
e itIlh ( e 1t110 , g e )tV/Th
(10)
for all finite t in 10,
Let C be a countable basis for K. Let Q be an upper bound
for {Q(h): h is in C}, which again exists by [8]. Let q > Q, t in
be finite and h in K be arbitrary. Given 13 in (0,
choose
Then
k in C so that I! h k II <
(11)
1<
Me n h (e" H oge '1 9 9hil
k)) 1 I
(e 2tH c ,, e lir , 9)9k1I
+ Il(en'e"7/ 9q (k
< 313 ,
since all operators appearing are contractions and the 2nd term
being infinitesimal is infact less than R. As )6 is arbitrary in (0, 00)
the difference in (11) is infinitesimal and the proof is complete.
REMARKS. (a) The theorem applies to Example (8) with X = +i
and in case the distribution is nonnegative, such as with the delta
functions, X may also be taken as 1. The theorem also applies
1. H is the form sum and in these
to Example (9) with X =
is
either
a
truncation
or a regularization with
examples, the V
positive infinite integral index and so is a bounded element of
*SA(K). Consequently [H., + V] = H, + V.
(b) In case X = 1 it suffices to assume Ho , V > C as we will
see in the appendix, Example 6.
(c) If V is in SA(K) rather than *SA(K) then the formula (7)
is equivalent to the standard formula (6).
(d) In [9] we used a preliminary form of this product formula
to express the dynamics of a singularly perturbed quantum system
in terms of a nonstandard Feynman path integral.
(e) If K is not separable, one still obtains a product formula
but then the N in *N depends on the h in K.
EXAMPLE 11. In Example 6 we observed that c" could be
identified with 1. However, this identification may not be made in
EXPONENTIALS OF DISTRIBUTIONS
217
the present context, for we have just seen that

e - - - d 2 /dx 2 -4-15)h
( e t(edx 2 :n) e -fit/tt)
lh
or all n sufficiently large. Had we also identified

would have obtained
e -td ; ,.
with 1 we
e ,(e;d.r2;h
which implies d 2/dx 2 + o = d2/dx 2 and this is false.

APPENDIX. The Strong Resolvent Topology.
In this appendix T will denote an element of *SA(K) and S an
element of SA(K), for K a complex Hilbert, space. Conditions for
and consequences of T S will be investigated. There are closely
related standard results which are obtained by considering a sequence, {T}, in SA(K) and its (possible) limit S, in the strong
resolvent topology. One could obtain corresponding nonstandard
results by transferring to T = T, n infinite and such a procedure
would be sufficient for the examples considered in this paper: T.=.
Ho + V. However, in this appendix we are not assuming that T
in *SA(K) has this special form and so, in general simple transfer
arguments are not necessarily sufficient to establish the desired
results.
DEFINITION. The finite resolvent set of T, denoted p(T), is the
set of all finite X in *C such that T X is 1 1, onto and 11 (T
X) -1 11 is finite.
EXAMPLES. By transfer of the resolvent equation
(a) (finite x,s*C: Im (X) 9.6 0) c p( T);
(b) If T z 0 then
(finite Xs*R: X < 0, X 96 0) c p(T) .
LEMMA 12. Let p denote the intersection of the finite resolvent

sets of S and T. Then
(i ) T i%s S if there is a X in p so that
(12)
(T x) -lh, (S x) -'h 11 P&0
for all h in K.
( ii) If T S then (12) holds for all X. in p.
The proof follows from a resolvent equation as in [3,

page 429] together with the fact that operators with finite norm
take infinitesimal vectors into infinitesimal vectors.
Proof.
218
ALAN SLOAN
DEFINITION. Let D be a subspace of the domain of S, D(S).

D is core for S if S restricted to D has exactly one selfadjoint
extension.
LEMMA 13. Let D be a core for S. If D cD(T) and IISf
Tf
0 for all f in D then T
Pz'
S.
Proof. E = (S + i)(D) is dense in K.
Let g in K be arbitrary
and let o be a positive real number. Choose f in !E so that 1;f
gl: < 3/3. Then
11(T+
g (S +
g
+
II(T + i) - '(g
+
(S + i)-1 1f Il
+ II(S +
g)II
11g
+ ii(T+ i) - '(T S)(S +
f H+
WV'
<
3
.
23/3 + 11(T S)(S +
gll
REMARK. See [3, page 424] for analogous standard result.

EXAMPLE 14. Let Ho, V be in SA(K). Let V be the truncations of V as in Example 9. If the operator sum Ho + V is essentially selfadjoint then Ho + [H, + V], the closure of Ho + V,
for all positive infinite ii. This follows directly from Lemma 13
because if f e D(V) then Vf Vf, n e *N N, by dominated
convergence. Thus,
(H0 + V.)f [Ho + V]f
for all f in the core D(H, + V) for [Ho + V].

REMARK. It may happen that T S but that there is no core
D for S on which T and S nearly agree. For example let V be a
nonnegative locally L' function on R 1 which is not 1,2 on any
interval. Let V, be the truncation of V: V.(x) = V(x) if V(x) < n
and V(x) = 0 otherwise. Let S = dO/dx= and T = the form sum
of S and V, n infinite. If f is in D(T) = D(S) and in K then f is
continuous and unless f is identically zero, If 1 is bounded -away
from zero on some interval. Consequently II V,if II is infinite. Since
11Sf 11 is finite it follows that
f e K fl D(T)
Tf fl infinite.
Thus, there is no converse to the previous lemma. On the

other hand we do have
LEMMA 15. If T S then to each h in D(S) there is an h' in
I.
D(T) with h'
h and T(h')
219
S(11).
Proof. Define h' = (T + i)- '(S + i)h. Then h' e D(T) and h'
h + ((T + i)' (S + i)- `)(S + i)h h. Further, (T + i)h' =- (S + i)h
so that Th' Sh.
EXAMPLE 16. Let c belong to R. Let Ho e SA(K), V E *SA(K)

be * bounded and H,, V z c. Suppose H e SA(K) and H H, + V.
By Lemma 15, for each h in D(H) there is an h' in D(H, + V)
such that Hh (Ho + V)h' and h h'. Thus
(Hh, h)
(Hh, h')
((H, + V)h', h')
11 2
OW
Since (Hh, h) and cjj h 1r are in R we conclude H c.

For bounded operators there are no domain problems so we
expect
LEMMA 17. Let S be bounded and let T be everywhere defined
with finite norm.
(a) If Th Sh for all h in a dense subspace D of K, then
TES.
(b) If T S then Th Fke Sh for all h in K.
Proof. Since D is dense and S is bounded, D is a core for S
so part (a) follows from Lemma 13.
For part (b), assume T S so that (T + i)- 'h fke (S + i) - 'h for
all h in K. Since T has finite norm h (T + i)(S + h for all h
in K. Let g = (S + i)'h. Then (S + i)g 0%4 (T + i)g for all g in
(the range of (S + i)") = K.
One of the reasons for the utility of the topology of strong

resolvent convergence in perturbation theory is
LEMMA 18. Let 55 be a bounded Boni function on the real line
which is continuous except on a closed set of S spectral measure
zero. If T S then 55(T)h (S)h, for all h in K.
With only minor modifications, (replacing certain "="'s
with "P., "'s), the proof of Faris, [1, pages 40-42] for an analogous
standard result works here.
Proof.
LEMMA 19. Let T be near standard in the strong resolvent
ALAN SLOAN
220
topology. Then for all h in K and all X.

e"Th
h.
If also T > c, c in R, then
Pr of.
e'h h .
By Lemma 8
e i2Th , e nsh , h
by strong continuity of t e"s.

EXAMPLE 20. Let T = n, n a positive, infinite integer. Let
h so T is not near standard even
F=e 0. Then eurh =
though e"Th is near standard for all t in *R.
Similarly (1 +
h while from Lemma 19 and the Laplace
transform formula
(1 +
Ve - te - "`Glidt
0
it follows that (1 +
h for all near standard G. Thus,
(T c) - 'h near standard for all h in K does not imply that T is
near standard in *SA(K). We next discuss when (T c) - 'h near
standard does imply T is near standard. We shall use the following hybrid notation.
For a. in *R we let "as 0, n in N" indicate that for every
real 0 < c < co there is an N in N such that n > N implies !a!<e.
LEMMA 21.
If h. is near standard in *K, if h is in *K and if

Ilh. hll
0, n in N
then h is near standard.

Proof. Though this lemma is a special case of [10, 8.4.29].
We include the following elementary proof:
Let h. = k. + E., k. in K, s.
O. Then II k. h II
II k. h.II +
n in N, so that Ilk.
< Ilk. hll + Ilh
1Ih.
) 0 11
as n,
co in N. Thus {k4,} is a Cauchy sequence in K. Let
Then k is in K and for every 6 in (0, cc') Il h kll
Ilh h.11 + II h. kll + Ilk. k I I < 8, by choosing n sufficiently
large. Thus, h 4=5 k.
LEMMA 22. Suppose that for some finite X in *C with Im (1),4
0, (T X,) - 'h is near standard for all h in K. Then, for all finite
NOTE ON EXI'ONENTIALS OF DISTRIBUTIONS
221
0 and sign (Im (X.)) = sign (Im (p)) and for

p in *C with Im (p)
all h in K, (T p) -1 h is near standard.
Proof. It suffices to consider those p's satisfying 1p
<
Im (X)I and I p
XI 0 I Im (X) I. For such p's there is the formula
(T p) - ' =
(p x)k(T X) - '"". The lemma then follows from
the observation that since (T X) - ' maps standard vectors into
near standard vectors and since 11(T X) - ' 1I is finite, (T X) - ' in
fact maps near standard vectors into near standard vectors so that
(T ;,)-(''') maps standard vectors into near standard ones.
REMARK. See Kato [3, page 427] for an analogous standard
result.
THEOREM 23. In order that T be near standard it is necessary
and sufficient that both
(a) for some finite, positive and noninfinitesimal c, (Tie) - 'h
be near standard for all h in K; and
(b) (1 i G T) - 'h h
for all positive infinitesimal e and all h in K, be true.
Proof. The necessity follows from Lemma 12 and the discussion
in Example 20. For sufficiency, define, by Lemma 22,
R(ia)h = st((T ia)- 'h)
for all positive, finite noninfinitesimal a and all h in K.

Then
(R(x) R(y))h
((T + x) - ' (T
= (y x)(T +
y) - ')h
+ y)-1h
But
II (T + x) - '(T + y)-'h
R(x)R(y)hll
II(T + x) -TT +
R(y))hII
0
+ II((T + x)-' R(x))R(y)hII
because 11(T + x)-1 II has finite norm. Thus (R(x) R(y))h %+ (y

x)R(x)R(y)h and since both sides are standard the pseudo-resolvent
equation results:
R(x) R(y) = (y x)R(x)R(y)
The proof may now be completed as in Kato's proof [3, page 503]
of an analogous standard result.
I
222
ALAN SLOAN
THEOREM 24. In order that T be near standard it is necessary

and sufficient that both
(13)
rh
for all positive infinitesimal s; and
e Th
;tt
(14)
is near standard for all finite t 0 in *R

be true for all h in K with both X = +i and X =
If T z c for some c in R and if (13) and (14) both hold with
= 1 then T is near standard.
We give the proof in case X i. The case of X= 1

follows similarly. The necessity follows from Lemmas 8 and 19.
For the proof of sufficiency, fix h in K, finite c > 0 in *R,
C fje 0. The Riemann integrals on *[0, c)
Proof.
o'''e- `te -"Thdt
are defined by the strong continuity of the integrand.

Fix E > 0 in R. Choose b > 0 in R so that
e- "dt < el211hIl
which satisfy
Let I be the collection of z's in *[0,
t, 8 in [0, b], It sl < z

imply
< ec/(2(1 C a.))
(eur
I is internal. The semigroup property and (13) show that I contains all positive infinitesimals. Consequently, there is a a in I n
(0, 00).
Let a k = k for k = 0, 1, , n, where n in N is chosen so that
n8 <b while (n + 1)a z b. Set a. +, = b. Let
ak+i e dclt .
h. = 1-1 eikarhi
k=0
at
h, is near standard being a finite sum of near standard vectors.

c, e- aeitThdt h,11
ck+le
k=0 ak
-et
(e"r einhlidt
TE
223
EIXPONENTIALS OF DISTRIBUTIONS
h dt
+
Lc.
0 (e - sc /2(1
"))dt +
By Lemma 21,
0
e - "e"Thdt
is near standard. From the Laplace transform formulas

(T
= --T-
dt
we conclude that (T ic)- 'h is near standard. We have thus verified (a) of Theorem 23. We next verify (b).
Let s > 0, s a 0. Choose ,63 > 0 to be infinite in such a way
that e
0. Then
+ ie Trih hll
e+"'Thdt h
0
H e"`Th hH dt
+ 2HI
dt
a0
since 0 < t < 13 k.f

P 0
etTit R. h.
Thus (1 +
this theorem follows from Theorem 23.
Is
and
See Example 20 for a remote T. Notice that (13) fails, in

other words, the semigroup is not strongly S-continuous.
REFERENCES
1.
2.
W. Faris, Self-Adjoint Operators, Springer-Verlag, (1975).

I. Herbst and A. Sloan, Perturbation of translation invariant positivity preserving semigroups on L 2(RN), Trans. Amer. Math. Soc., 236 (1978), 325-360.
3. T. Kato, Perturbation Theory for Linear Operators, Springer-Verlag, (1966).
, Trotter's product formula for an arbitrary pair of selfadjoint contrac4.
tion semigroups, Berkeley preprint.
5. D. Laugwitz, Eine einfuhrung der 3-funktionen, S.-B. Bayerische Akad. Wiss., 4
(1958), 41-59.
6. W. Luxemburg, Nonstandard Analysis. Lectures on A. Robinson's Theory of
Infinitesimal and Infinitely Large Numbers, Mimeographed Notes, California Institute of Technology, (1962).
ALAN SLOAN
224
A. Robinson, Nonstandard Analysis, North-Holland, (1974).

A. Robinson and A. Lightstone, Non-Archimedian Fields and Asymptotic Expansions, North-Holland, (1975).
9. A. Sloan, An application of the nonstandard Trotter product formula, J. Math.
Phys., 18 (1977), 2495-2996.
10. K. Stroyan and W. Luxemburg, Introduction to the Theory of Infinitesimal-8,
Academic Press, (1976).
11. J. Thurber and J. Katz, Applications of Fractional Powers of Delta Functions,
in "Victoria Symposium on Nonstandard Analysis", A. Hurd and P. Loeb, eds.,
Springer-Verlag, (1974).
12. H. Trotter, Approximation of semigroups of operators, Pacific J. Math., 8 (1958),
887-919.
7.
8.
Received August 4, 1977 and in revised form January 11, 1978. Supported in part
by NSF MCS 76-07543.
ATLANTA, GA
30332
NATIONAL SCIENCE FOUNDATION'`

Washington, D.C. 20550
FINAL PROJECT REPORT

NSF FORN1 98A
PLEASE READ INSTRUCTIONS ON REVERSE BEFORE COMPLETING

PART IPROJECT IDENTIFICATION INFORMATION
2. NSF Program Math&Comp.Sci
1. Institution and Address
Georgia Institute of Technology
Modern
MCS
Analysis
7/1/76
From
-707543
5. Cumulative Award Amount
4. Award Period
225 North Avenue,NW,Atlanta,Ga.
3. NSF Award Number
12/31/75
To
$19,000
6. Project Title
Applications of Nonstandard Analysis to Mathematical Physics

PART IISUMMARY OF COMPLETED PROJECT (FOR PUBLIC USE)
The Mathematical theory of infinitesimals is developed and used

to analyze mathematical problems arising in quantum physics.
1. In"An Application of the Nonstandard Trotter Product Formula"
a new formulation of the Feynman path integral is given.
2. In"A Note on Exponentials of Distributions" it is proposed
that instead of attempting to treat nonlinear functions of
distributions within distribution theory on should go beyond to
nonstandard analysis.
It was noted that the exponential of the
negative Dirac delta function could be defined within the context
of distributions to be identically one whereas it is not one
in the nonstandard framework and this leads to the correct
perturbation
formula. This paper also contains the nonstandard
..
. .
analysis of strong resolvent convergence.
.
3.In "The Strong Convergence of Schrodinger Propagators" time
dependent perturbation formulas are developed for highly singular
potentials in quantum mechanics using,techniques_of_the
infinitesimal approach.
.
,
-
.;:p-. 1.
-;-..
'
PART IIITECHNICAL INFORMATION (FOR PROGRAM MANAGEMENT USES)
I.
.
ITEM (Check appropriate blocks)
NONE
a. Abctracts of Theses
ATTACHED
PREVIOUSLY
FURNISHED
TO BE FURNISHED
SEPARATELY TO PROGR.
Check ()
Approx. Date
b. Publication Citations
c. Data on Scientific Collaborators
d. Information on Inventions
e. Technical Description of Project and Results
C Other (specify)
2.. Principal Investigator/Project Director Name (typed)
3.
Prinr-in2I Invectiontnr/Prn;rrt Ilirertor
civriature
4. Date
.
Alan D. Sloan
-,
3/14/81)
-
An application of the nonstandard Trotter product formula

Alan Sloan a)
School of Mathematics, Georgia Institute of Technology, Atlanta,
'(Received 21 March 1977)
Georgia 30332
The nonstandard Trotter product formula is used to extend the Feynman integral interpretation of
solutions to the Schrddinger equation in the presence of a highly singular potential.
We seek an answer to the question of when a quantum

mechanical Hamiltonian H determines a system which
evolves according to a Feynman path integral.' Specifically, we want to know when there is a potential V SO
that, upon setting 1-1=1, and fixing 1> 0
v (u), u in L 2 (R3 ),
exp(- itH)u
(1 )
where T o. v (u) is defined as the integral
(T,0)(A-0)=c fn3o exp[inv o , x, x; n, t, V)]
and
c=(27it,Inin)-3" 12 ,
(XI X2,
By the Trotter product formula and the transfer principle, for each K in *IN
exp[- t(Ho+VK)]=s- lim[exp(- itHo/n) exp(n in *V
117 > 0,
Xn-1),
VK,/n)14.
(7)
Combining (6) and (7) we find
and
S(X0, X, X,,; 71, t,V)=i-I, (1)1(x x

t/71)`
J.1
An explicit computation2 shows that
T,v=[exp(- itHo/n)exp(- it V/17)1",
vtx
k 11.)t
(3)
Ho
A/2m on L2(1R3),
[exp(- itHo/n) exp(- it VK/n)] ".
exp(-
(8)
ri;4711
(4)
where
(5)
for V any real measurable function. Consequently,

Nelson 2 was able to employ the Trotter product formula' to verify (1) in case H is the closure of the operator
sum, Ho + V, and V is chosen so that Ho + V is essentially self-adjoint Nelson 2 and Faris" extended this type
of formula to more singular potentials by including an
additional limiting operation after analytic continuation
in the mass parameter in. Although there has been
progress in extending Trotter's formula for exp[- t(Ho
5-7 when V is singular relative to H o , analogous +V)],
results are not known for the unitary groups exp[it(H o
+V)] Through the use of elementary nonstandard techniques we shall find a formula of the type (1) valid for
singular potentials.
For details on nonstandard analysis and notation we
refer the reader to Refs. 4 and 8-10. Here we adopt
the convention that Xc *X for any set X. If X is a topological space, a in X, b in *X, we write a= b if and only
if b is in the monad of a. If a= b we say a is the standard part of b and that a and b are infinitely close. We
use the Euclidean topology on IR" and the norm topology
on L2(R3),
Supported in part by NSF .TICS 76-97543.

2495
(6)
exp(- itH)uz exp[- it (H + V Au.
(2)
X u(x)dx dx,
Suppose H is a self-adjoint operator on L 2 (1R3 ) which

can be approximated in the generalized strong sense by
bounded perturbations of H o ; e. , there is a sequence
VK of bounded self-adjoint operators on L 2 (R3 ) such that
the resolvents of Ho + VK converge strongly to those of
H. Then, (Ref. 11, 502), providing {1/ 0 + V K } is uniformly bounded below, it follows that exp(- itH) is the strong
limit of exp[it(H + V A. Thus, for all K in *IN- IN,
where IN ={l, 2, and for all u in L 2 we have
Journal of Mathematical Physics, Vol. 18, No. 12, December 1977
For all finite n and K, (4) shows that

T,vic= [exp(- itHo/n) exp(- itVKA)]",
(9)
so by the transfer principle (9) holds for all n, K in

*N. Thus, for each fixed infinite K, (8) yields
exp(-
lira T o, vic (u).
(10)
n in ''14
For each n, K in IN, (2) shows that T, vic (u) is given

by an action integral. The transfer principle then implies that for all n, K in *N formulas (2) and (3) remain
valid with VK replacing V. Consequently, we may conclude that
lim c,,111 exp[iS( , x, x o ; 72, /VA
exp(- itH)u( ) =n3o
ni;
xu(x)dx dx.
Lemma: Let X be a separable normed linear space.
Let f: X *X be bounded linear operators which are
uniformly bounded in the sense that there is an M in N
such that lif11 -. cm for all n in *N UK If
fo pointwise on X as n - -.0 in *IN, there is an N in *N such that
n>N implies f(x)Pzf o (x) for all x in X.
Proof: Let C be a countable dense subset of X. Let
X be a positive infinitesimal. Choose N, in *IN so that n
>N, implies Ilf(c)-fo(c)ii < X. Let N be an upper bound
for INc : c in Cl; N exists by Ref. 5, p. 59. Let x in X
be arbitrary. Fix 6> 0 in R. There is a c in C so that
Copyright 1977 American Institute of Physics
2495
tic:Ail< 6. Then for n
I I fn(x) -- fo(x) I I
Il f.(x) in (c)I I + II fri(c) fo(c) I
+ f0(c) f0(x) I I
<mllxcll+A+MIlcxll <34116.
Because 6 is arbitrary we conclude that f(x)"f0(x)
Q. E. D.
whenever n >N.
For the necessary approximation theorem in case (b)

see Ref. 12 where the VK 's are defined by truncation.
In cases (a) and (c) see Ref. 13 where the V K 's are given by regularization of V. See Ref. 13 for additional
examples.
Similar techniques apply in dimensions other than 3
and for other H o 's.
Combining this lemma with the previous conclusion,

noting that exp( itH) and the T,v K are uniformly
bounded by 1, yields the following:
Theorem: If VK are bounded self-adjoint operators
such that H o + VK converge to H in the generalized strong
sense as n co in IN, then for each fixed positive infinite
integer K there is an N in *111 such that for all n >N and
for all u in L 2
exp(itH)u( ')'cn fion exp[iS( , x, x;n ,t, Vx)]
(11)
xu(x,,)dxdx,
in *L 2 .
To compare the types of potentials covered by formulas (1) and (11) We note that 2 (1) holds if V is in LP
p > 2, whereas (11) holds whenever H is the generalized strong limit of bounded self-adjoint perturbations of H o . Examples of such H's can be found by defining H to be the form sum of H o and V when either
(a) V is in LP +L', p >
(b) V+ 0 is locally in L l outside a closed set of mea-
sure zero;
(c) V is a delta function distribution concentrated on
the surface of a compact C I hypersurface in IR 3 .
Feynman and A. Hibbs, Quantum Mechanics and Path

Integrals (McGraw-Hill, New York, 1965).
1 R.
2 E.
Nelson, "Feynman Integrals and the SchrOdinger Equation," J. Math. Phys. 5, 332-43 (1964).
3 H. Trotter, "Approximation of Semi-groups of Operators,"
Pacific J. Math. 8, 887-919 (1958).
4 M. Davis, Applied Nonstandard Analysis (Wiley-Interscience,
New York, 1977).
5 P. Chernoff, "Product Formulas, Nonlinear Semigroups and
Addition of Unbounded Operators," Mem. Am. Math. Soc.
140, 1-121 (1974).
6 W. Faris, "The Product Formula for Semi-groups defined by
Friedrics Extensions," Pacific J. Math. 21, 47-70 (1967).
7 T. Kato, "Trotter's Product Formula for an Arbitrary Pair
of Selfadjoint Contraction Semigroups," Berkeley preprint.
8 A, Robinson, Nons tandard Analysis (North-Holland, Amsterdam, 1974).
9 A. Robinson and A. Lightstone, Non-Archimedian Fields
and Asymptotic Expansions (North-Holland, Amsterdam,
1975).
I0 K. Stroyan and W. Luxemburg, Introduction to the Theory
of Infinitesimals (Academic, New York, 1976).
II T. Kato, Perturbation Theory for Linear Operators
(Springer, Berlin, 1966).
12T.V!r. Faris, Self-Adjoint Operators (Springer, Berlin, 1975).
13 1. Herbst and A. Sloan, "Perturbation of Translation Invariant Positivity Preserving Semigroups on L 2 (101)," to
appear in Trans. "Am. Math. Soc.
2496
J. Math. Phys., Vol. 18, No. 12, Decembe 1977
Alan Sloan
2496
The Strong Convergence of Schrodinger Propagators
Alan D. Sloan
Georgia Institute of Technology
Atlanta, Georgia 30329
Accepted for Publication in Transactions of AMS
AMS(MOS) subject classifications (1970). Primary 3 5 B25, 35J10,

47D99, 81A10.
Key words and phrases. Evolution equations, propagators, perturbations, Schrodinger equation.
Research supported in part by NSF under Grant MCS76-07543.
Abstract
Time dependent versions of the Trotter-Kato theorem are

discussed using nonstandard analysis. Both standard and nonstandard results are obtained. In particular, it is shown that
if a sequence of generators converges in the strong resolvent
topology at each time to a limiting generator and if the sequence
of generators and limiting generator uniformly satisfy Kisynski
type hypotheses then the corresponding Schrodinger propagators
converge strongly. The results are used to analyze time dependent, form bounded perturbations of the Laplacian.
I.
Introduction.
The time dependent Schrodinger equation
dx (t)
dt
x(t)
A(t
(1)
x(s) = x
may be advantageously discussed in terms of unitary propagators.

Equation (1) is set in a complex Hilbert space, H, and the time
variables, t,s , are to range in some closed and possibly infinite interval, J, of real numbers. For each such t, A(t) is a
4
selfadjoint operator on H while x(t) is an element of
H as is
...4
x
. A() is called the generator of (1).
Definition: A jointly strongly continuous map (t,$)
U(t,$) from
J x J into the unitary operators on H is a unitary propagator
providing
(a)
U(t,t) = I
i l j. )U(s,r) = U(t,$)
t Ut,s
hold for all (r,s,t
Given a generator A the relevant propagator is expected to
have the property that x(t) = U(t,$)x s is the "solution" to
equation (1). The precise sense in which x() is a solution and

the exact collection of initial states x
for which such la solu-
tion exists, remain to be specified later. At times, to avoid

ambiguity, we will denote the propagator related to equation (1)
by UA .
The most satisfactory results are known in case A(t) = A for
each t in J where A is some fixed selfadjoint operator. This is
the time independent case. The related propagator may be explicitly given as U A (t,$) = e i(t-s)A . Then x(t) = U(t,$)x is difs
ferentiable in norm and satisfies equation (1) for all x s in D(A),
the domain of A.
Once a solution is known to exist, an analysis of stability
properties of the equation may begin; this is perturbation theory.
The basic question in this theory is: If two generators are close,
are the corresponding solutions close? In the time independent
case, this question is affirmatively answered by the Trotter-Kato
Theorem 1 [5, page 5021:
If A n is a sequence of selfadjoint operators whose resolvents
converge strongly to the resolvent of a selfadjoint operator A,
(t,$) converges strongly to U A (t,$) for each t,s in J.
n
The situation is more complicated in the time dependent case.
n
For example, for t in IR 1 , let f n (t)= 1 + 2 X n (t), where X n is
-(11+1)
the characteristic function of the interval [n - 2
then U
-1
+ 2
(n+1)
] for n = 1,2,... . Let A n (t)= f n (t)I and A = I.
Then A (t) converges in the strong resolvent topology to A but

n
UA (t,0) does not converge strongly to U A (t,0) for.any t > 0, even
though the An 's and A are bounded.
In II we present a Trotter-Kato type theorem for the case of

bounded generators. In III a similar result is presented for a
certain class of unbounded generators. For unbounded and time
dependent generators, A, there is a general technique for constructing UA . First one approximates A by bounded generators, A n ,
next one constructs U
the U
An
explicitly, and finally one proves that
converge to a propagator which is then defined to be UA .
Yoshida, [15], developed an approximation procedure which has

proved to be very durable. He defined A n (t)to be A(t)[1+A(t)/n )-1.
This method has recently been applied in [16] to study the "hyperbolic case" of evolution equations. In addition to its continued
applicability, another indication of the strength of the Yoshida
approximation may be described in the framework of nonstandard
analysis. In the next paragraph we informally discuss the ideas
involved.
Given some set S one can form S, a new set containing not
only a copy of S but also ideal elements, whenever S is infinite.
For example if IN is the set of positive integers then ]N contains
infinite integers while for ]R the reals, ]R contains both infinite and infinitesimal numbers. The map V
V of subsets of S to
subsets of S is a Boolean algebra isomorphism into. Consequently,

all the concepts of standard analysis on S may be transferred to
S. In particular there is a natural order on ]R and an induced
definition of convergence: for a
,a in M2, n
]N, an
a means
for every c > 0 in IR there is an N in lq such that n > N implies

la n -al < c. Given such a sequence a n one may also ask if a n
in the standard sense that for every c > 0 in 71,1 3N in 7N such

that n > N implies Ian-al < c. These two types of convergence
are not equivalent. For example, if m is an infinite integer then
a
n = m/n + a = 0 in the transferred sense, but not the standard
sense.
In III we consider the class, K, of generators introduced by
Kisynski, [8], and studied also by Simon [11]. The aforementioned
authors showed that whenever A is in K, then U A converges strongly
n
to U where A is the Yoshida approximation to A. Transferring
A
n
this result we obtain the fact that whenever B is in K then U
converges strongly to UB . A priori, we may not assert that UB

n
converges to UB in the standard sense. Nevertheless we are able
to prove that this additional convergence takes place, giving an
additional indication of the strength of the Yoshida approximation.
Moreover, the general techniques of nonstandard analysis may be
used to show that in order to obtain a standard Trotter-Kato type
theorem in K (i.e., whenever C n ,C are in K and C n (t) + C(t) in
the strong resolvent topology for each t, then U c (t,$) converges
n
strongly to Uc (t,$)) the standard convergence of U B (t,$) to U B (t,$)
*
for B in K is sufficient. Thus we are able to present a new standard result obtained first using nonstandard techniques. Our
methods also revealed a standard proof which we present. The nonstandard results do have interesting consequences in IV:
In IV we are particularly interested in generators of the form
2
A(t) = -A + V(t), selfadjoint on L (1Rn ) for each t. The corresponding propagator describes the evolution of the quantum system
5
with Hamiltonian A(t) at time t. If v(t) is a quadratic form then
the usual formulas of quantum physics do not have an obvious
meaning; see for example the Dyson expansion, Theorem 3, which is
typically the starting point for time dependent perturbation theory. Nonetheless, if one regularizes the V(t) it is possible to
obtain a nonstandard bounded W(t) so that UA and U B are infinitely
close, where B(t) = -A + W(t). Thus, the usual perturbation formulas, which are valid for U B , may be used for UA with only infinitesimal errors. Consequently, we obtain a nonstandard Dyson expansion for propagators of quantum mechanical systems of particles
in potentials given as distributions. These and other applications
are discussed further in IV.
Finally, we remark that the reader interested in perturbation
theory but not nonstandard analysis may read the remark following
Theorem 7, the introduction to III through the discussion of Theorem 8, Corollary 10, Theorems 11, 12, Corollary 13 and Examples
14 and 15 as these standard results are self contained.
II.
Stability of Solutions for Bounded Generators.

Let X be an adequate ultrapower of a structure X containing
the real numbers, 1R 1 .

For A selfadjoint and B an internal selfadjoint operator we
write A 11,1; B if B is in the monad of A relative to the strong resolvent topology. In this case we say that B is near standard and
that A is the standard part of B. If P is a bounded linear operator and Qan internal finitely bounded linear operator write P ti Q
in case Ph ',11," Qh for all h in H. In case P and Q are additionally
selfadjoint the two definitions of Q; coincide. Here we have written h k to mean Ilh-k
Z: 0 for h,k in H. This coincides with

h being in the monad of k, in case k E H, relative to the norm topology on H. The weak topology relation will be denoted by h
k.
This means (h,v)
(k,v) for all v in H. On 11R
to the ordinary Euclidean topology.
is with respect
6
In [13] it was shown that a standard technique of proving
the Trotter-Kato theorem, (see [5]), could be used to prove a
nonstandard version:
Theorem 2: If A is selfadjoint and B is internal selfadjoint

then A ',1", B implies U A (t,$) ti U B (t,$) for all finite t,s in ]R1 .
See [13] for a discussion of the strong resolvent topology
from . a nonstandard point of view. For an introduction to nonstandard analysis the reader is referred to [9] and [14].
Definition: A D-generator is a strongly continuous map of J

into the bounded selfadjoint operators.
An existence theorem for solutions of the Schrodinger equation defined by a D-generator is proved by explicit construction
via the Dyson expansion.
Theorem 3: [10] Let A be a D-generator. Then defining U(t,$)

CO
by I +
n=1
(-i) n T (A,t,$) for t > s in J, where

n
T (A,t,$) =
n
ft
t,
S S
t
A(t
S
)A(t ) dt dt
n
n
1
and by U (s,t) for s > t in J, gives a unitary propagator. Here

the integrals are strong while the sum converges in operator
norm. x(t) = U(t,$)x s gives a norm differentiable function which
is a solution to equation (1) for all x s in H.
The main ingredient in the proof of theorem 3 is an application of the uniform boundedness principle to prove
Lemma 4 [10]: Let A be a D-generator and K a compact interval
in J. Then sup! IA(t) I I < m
-LEK
Lemma 5: Let A be a bounded linear operator and B an internal
finitely bounded linear operator. If g,h E H,g ti h and Ag,%'Bg

then Ah ti Bh.
Proof:
1 lAh-Bh1 1 <
11A(h-g)11 + 11(A-B)g11 + 11B(g-h)11

< 11All 11h - g11 + 11(A-B)gll + 11B11
11g -hi' ,T, 0.

Q.E.D.
Lemma 6: Let A be a D-generator. Then for all finite t,s in
J, t "t" s implies A(t) ti A(s). In particular A(t) is near

*
standard for each finite t iin J.
Proof: If t,s in J are finite there is an N in 14, the non-
negative integers, such that It! < N and Is! < N. By lemma 4
sup 11A(t)11 = c is finite. By transfer, 11A(t)11 < c for all
tEIR
ItI<N
t E * IR , It! < N. If s
ti
t, then s
ti
By strong continuity, A(s)h ti A(r)h
r ti t, r E IR and 1r1 < N.
ti
Since A(s),A(r),A(t)
A (t)h.
are all finitely bounded we obtain A(s) ti A(r)
A(t).
Q.E.D.
8
Theorem 7: Let A be a D-generator and B an internal D
Suppose additionally
enerator.
is finite for all f nite t
at
*
in J.
Then A(t) ti B(t) for all finite t in J implies
U (t,$) ti U (t,$) for all finite t,s in J.
A
B
Proof: Fix t,s finite in J. Let K be a compact interval in J

*
so that t,s are iin K. It suffices to consider the case t > s.
is internal so {N E
*
I IB (r) I I
< N dr E
K} is internal
As 24-DI is external we conclude that
and contains
there is a c in IN such that
I IB (r) I I
< c for all r E K.
Write T (0) for T (0,t,$) and U. for U.(t,$).

n
n
a
n T
-(I+
(-i)
Choose a in IN -IN . Observe that
n
n=1
co
co
cn(t_ s) n ,,, 0.
1
Tn(IIBII) _
Similary,
,I,
n!
< n=a+1
n=a 1
a
(-i) n T n (A))II ,'-'', 0. Consequently, we obtain
lUA-(I+
n=1
I IU B
(B))II <
Il[uB -uA]hIl rt,<

(2)
.,
1 11[B(t )---B(t n )-A(t )-..A(t )1hIldt -..dt

n
1
n
1
1
*
for all h in ft and a in IN
B(t )---B(tn ) - A(t 1 )-.-A(tn )

(3)
n
B( -t.
=
i=1
)--.A(t )
)[B(t i )-A(t i )]A(t. 13(.
n
11 +1
9
If the t i 's are all finite and if h is near standard then by
lemmas 5 and 6 A(t i+i )A(t )h is near standard. By lemma 5
n
1
1,n
i nfinites-
imal norm, providing n is finite. Let S n = maxfl I h.

1,n
i= 1,2, n1
0, for n finite. From (3) we conclude
11[B(t 1 )--B(tn )-A(t 1 ).--A(tn )hll

n
<
1=1
.
1-1
Ilhi,nll
< S n (1-c n )/(1-c). Then
11[B(t1)---B(tn)-A(t1)--.A(tn)]hil
(4)
ti 0
for all n EDT and t. finite.

1
Fix h E H.
Let I
E Tq: (n<mdt
1<m,--,It 1<m)
n
(Ilm[B(t 1 ).B(tn )-A(t1 ).-.A(t n )]hll < 1)1. I 0 is internal

and contains, by (4) above, the external set IN . Thus y E I 0
*
for some y iin 24-Tg. By (2) there now follows

1 (t )n
II(U -U )11II <
n!
L y
B
i=1
that U A (t,$)
(t,$)
e(t-s)
y
o. Thus UhUh so
' A
B
for all finite t,s in J.

Q.E.D.
Remark: If B
n is a sequence of standard generators with
sup suplIBn (t)11 < co for all compact K, then the Dyson expansions
n tEK
for
converge to U
(t,$) uniformly in n for each t,s.
10
Consequently, if B n (r) converges strongly to A(r) then T m (Bn ,t,$)

converges to T m (A,t,$) for each m as n
co by dominated conver-
gence and one may conclude UB (t,$) converges to U A (t,$) strongly.

The previous proof follows this argument.
ll
III. Stability of Solutions for Unbounded Generators.

Definition: A K-generator is a map A of J into the selfadjoint
operator on U for which there exists a selfadjoint operator
H >1
and a positive real number c such that
c-1(H+1) < A(t) < c(H+1)
G(t) -
d(A(t) -1 )
dt
exists in norm
1/ 2 11 < c
I IA(t) 1/2 G (t) A(t)
(5)
(6)
(7)
hold for all t in J. H and c are the related constants.

In [11] Simon presents an existence theorem for equations
with K-generators. We have included the relevant part of this
result as Theorem 8 below. Simon points out that, in fact, this
theorem is a special case of a result due to Kisynski,- [8]. In
this theorem there is no hypothesis concerning the independence
of D(A(t)) from t. Thus it is particularly useful in treating
problems of singular perturbations.
11
Theorem 8: Let A be a K-generator. There is a unitary propaga-
tor U such that defining a(t) = U(t,$)x s in D(H 1/2 ) gives a weak
solution to Schrodinger's equation (1) in the sense that
1/2
1/2
-4 (f,a(t)) = -i(A
(t)f,A
(t)a(t)) for all f in D(H 1/2 ).
dt
Some Remarks on Theorem 8:
Although the theorem only gives a weak solution it follows

III
by a related uniqueness result that if a strong solution exists,

it is the one given in theorem 8.
Both Simon and Kisynski use a modification of techniques invented by Yoshida. One defines
A (t) = A(t)(1+n -1 A(t)) -1

n
(8)
for n =. Observing that t An (t) is a D-generator and

that
liAn (t)II < n
(9)
for all t in J, one obtains a strong solution a n (t) to the

Schrodinger equation (1) with generator A
n'
using theorem 3. One
then proves
lim (f,an (t)) = (f,a(t))

no-co
for all f in H and t in J.
(10)
12
Defining IIgIl_i = II(H+ 1 ) -112 II

11914, = II(H+1)1/2
gII
it can be shown that for any compact interval T containing s,

there is a constant K
such that
an (t)
II
at
I Ia n (t)
and
for all t in T. Here if
KT
T <- (c+1)
I I -1 <
+1
KT
KT
(12)
is the largest t in T, then
2 e 3c(T-s)/211 x
II sII+1
The proofs of all these numbered facts may be found in [11].
Theorem 9: Let A be a K-generator with constants H,c. Let B be
an internal K-generator with constants H,c.

If A(t) r,1; B(t) for all finite t in J then UA (t,$) ti UB (t,$)
for all finite t,s in J.
Corollary 10: Let
for n =
An , A be K-generators on J with constants (H,c)
. Suppose An (t) converges in the strong resolvent
topology to A(t) for each t in J. Then UA (t,$) converges strongly

n.
s) for every t and s in J.
to UA (t
'
4
To obtain corollary 10 from theorem 9 one may proceed as follows. Choose B(t)=A n (t) for some positive infinite integer n.
Equation (7) implies that
1 1dB
-1
dt
_ c 2 for all t in J. The
-1
-1
norm mean value theorem then yields 11B
(t)-B
(s)II
0 when-1
s similarly for A . Since we are assuming A(s)
ever t
B(s)
-1
-1
for each s in J we arrive at the conclusion B
(t)
B
(s)
A
-1
- -1
(s) - A
(t) whenever s is in J and t
s, using the fact that
J is closed. Since, however, corollary 10 is an interesting standard result, such a proof is presented, Both the proof of theorem
9 and the corollary are based on the following standard computation.
Technical Lemma: Let T be any compact interval of 1R and let
KT be as in equations (10), (11), (13). Then for all A in K(H,c),
t > s in T, and x in D(H
1/2
1A -1/2 (t (u (st)-U rn (s t))xlf

'
) A
A
n
2 c (t-s) -1) (
1 +-1 )/c.
8K (e
T
m n
Here An is defined as in (8).

Proof: Fix s in T and for t > s in T define a n (t) = U A (s,t)x.
n
Fix positive integers m and n. Define
p(t) = a n (t) - am (t)
w(t)=(p(t),A -1 (t)p(t))
and
-1
dA
(t) = G(t).
dt
In general we denote dt by f so that p = -iAn an + iAmam and
-1
-1.
(PIA w= p) + (P.GP)
(P,A
-1
1
Since A = A
we may write
n
14
-1
w = (-iA n a n +iA m a , A 1 P)+(P,GP)+(P,A
(-iA n a n +iA a ))
m m
1
= i (A n an' (An - 1- ) p)-i (A a (A -1 - m1 ) p)+(p,Gp)
m
m
n
-1
1
1 1
+ i (p, (Am - ) Am am ) - i (p, (A n -
n ) A n p)
= i (a n , P) -
(An a n ,p)
(a m , P)
+Ilri
( Amam ,p)
(p,Gp)
(p,Amam)
- i(P,an ) + 1411 (p,An an ) +i(p,am) - Il
-Tl
= i(a n -a ,p) - i(p,a -a ) + (p,Gp) + 1 (p,-iA a )
m
n m
m m
- 1 (-iA n a p) + -1 (-iA a ,p) - 1 (p,-iA a )
n'
m m
n n
(An ,p) + 11n (Lm ,P) - 11z (P, 11 )
= i(p,P) - i(P,P) + (P GP) -
1 (p,a m )
+ 171
Thus
w = (p,Gp) + 12n Re(Am ,p)
- 121 Re(ki ,p)
(14)
Next observe that since
l(P,GP)1 = I I (A
1/2 p,A 1/2GA1/2_-1/2 .1

A
p ji
II A1/2 GA1/2I1
IIA-1/2 131 12
= 11A 1/2 GA1/2 11(p,A-1 p)
we obtain
l(P,GP)- 1 < cw
(15)
15
Also, for r in 11 ,
-1/2 ar
I =
1(L r' P)-
IlLr 11_1 1101 +1
(H+1) 1/2 p)1
11.;r 11_ 1 ( 1 1 an 11 +1
I la.11+1 )
so that from (11), (12) and (13) there follows
2K 2
l( r (t),P(t))1 <
- T
(16)
Substituting (15) and (16) into (14) we find
It
dt
I < cw(t) + K (
(17)
for all t > s in T. Here K is the finite constant 8K
T .
To finish the proof we observe that w(s) = (p(s),A -1 (s)p(s)) =
(an (s)-am (s),A 1 (s)p(s)) = (x-x,A -1 (s)p(s) = 0.
Since w > 0, (17)
implies.
K 1
1
0 < w(t) < ( +
- c n in
c(t-s)
K 1
1
- c( n m
)
Q.E.D.
Proof of Theorem 9: Fix a finite s in

Define a
J and x in D(H
1/2
as in the proof of the technical lemma and a as in theo-
rem 8. Similarly, define B
-1 '
(t) = B(t)(1+n 1 B(t)) so that B is
an internal D-generator with
11Bn.(t)11 < n .
(18) .
16
Let b
n be the solution to the Schrodinger equation (1) with gen-
erator B
and initial state x. Since
An (t)=
-1
-1
1
-1
1
-1
A(t)
+ '\ B(t)
+
= B (t)
n
n
it follows from bounded stability, theorem 7, that
a (t) ti b n(t) , d n E IN, V finite t in J.

n
Let I
= {mE
n<M
*J i
I
(19)
is internal and by (15) I
lt' < n
lin(an (t)-b n (t))1! < 1)).
contains N. Since 1\T is external
there is an w in I 0 n ( N - IN ) such that
an
rt:
ti, (t) bn (t)' V n < w, Vfinite t in J .
(20)
Let p(t) = bn (t) - bm (t) and
w(t) = (p(t),B
-1
(t)p(t)),
for n,m infinite. Since every finite t in J is in T for,:some com-
pact interval T with s in T it follows transferring the technical

lemma that
w(t) ti0 for all finite t in J .
(21)
17
Now suppose g is in D(H 1/2 ). Then
I(P(t),g)1 =
I(B-1/2 (t)P(t),B 1/2 (t)g)I
< 11B -1/2 (t)p(t)I1 I1B 1/2 (t)gll

< w(t) 1/2 c((H+1)g,g) 1/2 . Therefore from (21) there follows
(p(t),g) ti 0
(22)
for all finite t in J and g in D(H 1/2 ).

Let h in U be given. Let e be an arbitrary positive real
1/2
1/2
number. D(H
) such
) is dense in H, so there is a g in D(H
that Ilg - h11 < e. Then
I(p(t),h1 < l(P(t) , g)1
< I(p(t),(h-g))I
11P(t)11+1
by (22)
Ilh-g 1 1- 1
11P (t) 11+1
Since
l(P(t),h - g)1
<I
lp(t)
11 +1 Ilh-gli.
is finite for all finite t in J we conclude
that l(p(t),h)l < 2K T e where T is any compact interval chosen so
that T contains t and s.
Since s is an arbitrary positive real
number (p(t),h) 'rt 0. That is
(23)
bn (t) wti bm (t)
for all infinite n and m and finite t in J.
By
*
rem 8 and its proof we know that for each t in J
weak limit
n in
bn (t) = b(t).
transfer of theo-
18
In particular if E q", 0, E > 0 in
H there is an n(t,h,E) in
IR, t is finite in
J, h is in
such that for n > n(t,h,E) we have
(b (t) ,h) - (b n (t) ,h) I <
If m in 3N is infinite we argue, using (23) that

1(bm (t) - b(t),h)I
1(bm (t) -bn (t),h)I + 1(b n (t)-b(t),h)I
ti
ti
0.
The conclusion is
b (t) ti b(t)
m
(24)
*
for all infinite m and finite t in J. An analogous argument
shows
am (t) rk, a(t)
(25)
for all finite m and finite t in J.

Combining (16), (24) and (25) we may write
a(t)
an (t)b (t)
n
b(t)
for all finite t in J and infinite, though sufficiently small n.

In other words
19
U
for all finite t,s in
(t s)x s
'
(t,$)x
J and x s in D(H
(26)
1/2 ).
If y is arbitrary in H, choose c > 0 arbitrary in 7R and x

in D(H
1/2 ) such that Ilx-yll < E . Then for g in H, using (26)
I(U Ay-U B y,g)I

< I(U A (y-x),g)I
I(U A -U B )x,g)I
1U (x - Y)/g)I
2Il y - x1
since U
11g 1 1 < 2 El i g ll
and U B are unitary. As e is arbitrary we conclude
UA(t,$)y ti U p.(t,S)y
w
(27)
for all finite t,s in J and y in H. Again, since the U,Is and
U B I s are unitary we may argue that
11 2
IlUAy-UByll = (UAy-UBy,UAy-U By)
I ,
112
= IlUAll 2 + IlUByli - (UBY,UAY) - (UAY,UBY)
= 21!y11 2 - (UBy,UA y) - (UAy,U By)

r?:211Y112
= 2
1IY11 2
(UAy,UA y) - (UAy,UA y)
=
, using (27).
Thus
UA (t,$)y
ti
UB(t,$)y
20
for all finite t,s in J and y in H.
we conclude U A
11-,(t,$)
Since
IIUA II
= 1
= 11%11
B (t,$) for all finite t,s in J.
Q.E.D.
Proof of Corollary 10: Fix E > 0, t > s and x in
D(H 1/2 ).
Let T
be a compact interval containing t and s. Choose K T as in (11),

(12) and (13). Choose N to be an integer larger than
c(t-s)
8K(e
-1)/2Ec. Let m > N. For any n, let B (n) be the m th
m
th
Yoshida approximation to A and C the m
Yoshida approximation
A. Then by the Technical Lemma and the triangle

I IA 1/2 (t) (UA (s,t)-UA (s,t) )x I <2E + I IA-1/2 (t) (U B
to
Since for any fixed m, IIB m (n)
II
< m,
I ICI I
inequality
(n)
-U )x lI
C
< m and Bm (n) con-
verges strongly to C as n 00 it follows from theorem 7 and the

remark which follows it that for n sufficiently large
Since
is dense it follows that U
all
xE
D(A 1/2 (E))
D(H 1/2 )
(s,t)x
U A (s,t)x weakly as n
00 for
n
D(H 1/2 ). Since, in general, weak convergence on a dense
A
set implies strong convergence of unitary operators the proof is

completed.
Q.E.D.
Remarks. 1.
In the last section of this paper we apply the above
results to the analysis of form bounded perturbations, Proof similar to those just given lead to results useful in studying operator bounded perturbations. Here we just state these results.
21
Definition: A Y-generator on J is a function A() from J into
the selfadjoint operators on H such that there are positive
numbers E and M satisfying
(i) A(t) > E + 1 ;
(ii)
(A(t)+E) -1 is strongly differentiable with deri-
vative G(t) ;
(iii)
(iv)
G is strongly continuous; and

II(A(t)+E)B(t)I1 < M.
Then according to a theorem of Yoshida [15, 429] there is a unitary propagator UA such that for all x in D(H(s)), defining
a(t) = U(t,$)x gives a solution to the Schrodinger equation in
the sense that
da
dt
(t) = -iA(t)a(t).
The techniques given above lead to the following theorem and

corollary.
Theorem: Let A be a Y-generator with constants E and M. Let B

be an internal Y-generator with constants E and M. If
(A(t)+E) -1 :1;(B(t)+E) -1 for all finite t in * J then
UA (t,
11,(ts) U
(t, s) for all finite t,s in J.
Corollary: Let A,A n be Y-generators with constants E,M for

n =
. Let An (t) converge to A(t) in the strong resol-
vent topology for each t in J.Then U A (t,$) converges strongly

n
UA (t
' s) for each t,s in J.
2. In corollary 10 and the last corollary, the convergence
of progators is uniform in any finite interval of J.
itrtivigsrws
22
IV. Perturbations of a Time Independent Generator.

In this section H
will be a non-negative selfadjoint oper0

ator on H. If F is a Hermitian form Q(F) will denote the form
1/2
domain of F. In particular Q(H ) = D(H
).
0
0
Definition: A K-perturbation of H
is a Hermitian form valued

0
function, V, defined on J with the properties
(33)
)C Q (V(t) ) .
(i)
Q (II
(ii)
There are constants 0 < a < 1, 0 < b < co such that
IV(t)(f,f)I
< a(H 10 /2 c _1/2 f)

" n0
(34)
blIf112
for all f in Q(H 0 ).

(iii) There is a Hermitian form, V(t), which is the derivative
of V in the sense that (H
+1)
-1/2 V(t)(H0 +1) -1/2 iis norm differ-
entiable and its derivative is (H
+1)
0
and
(iv)
11../(t)(f,f)1 < a(1136/2f,HV2f)
-1/2
1.7(t)(H0 t)
blIf112
-1/2
(35)
for all f in Q(11 0 ).

(a,b) are the H 0 -bounds of V.
The following theorem is due to Simon. He states it for
the case where each V(t) is a selfadjoint operator but the proof
he gives works just as well for each V(t) being a Hermitian form.
Theorem 11 [11, pg. 66]:
Let V be a K-perturbation of H 0 . There is a constant E
such that A(t) = H 0 + V(t) + E is a K-generator with constants
(H o ,c) for some real number c.
23
Theorem 12: Let V,W n be K-perturbations of H 0 , for n =
Suppose the H o -bounds of W n ,(a n ,bn ) satisfy a n + bn
0 as
n co. Then there is a'constant E such that for n sufficiently

large An (t) = H 0 + V(t) + W n (t) + E is a K generator and U A (t,$)
converges strongly to U
0
+V+E(t's)
for all t,s in J.
Proof: First note that

Q(H 0 ) C 4(V(t))1)Q(W n (t)) = 4(V(t)+W (t)). Next observe that
n
I(V(t)+Wn (t))(f,f)1 < IV(t)(f,f)I +
IWn (t)(f,f)I
1/2
1/2
2
< a(H 1/2
o f,H 1/2
0 f) + b(f,f) + a n (H o f,H 0 f) + b n Ilf11
(a+1)
(H
1 /2
1/2
2
f,H
f) + 2b11fII
0
for all f in Q(H 0 ) and n sufficiently large. Also, letting

K = (H +1)
0
-1/2
, it is true that K(V(t)+Wn (t))K
= KV(t)K + KWn (t)K and so di K(V(t))+Wn (t)K exists in norm

and equals 1077(t)K + KW n (t)K. As above,
I(V(t)+Wn(t))(f,f)1
< a+1(H/2f'H
11_.
0/2r) + 2b11f11 2
- 2
0
By theorem 11, there is a constant E such that for n

sufficiently large A n (t) is a K-generator.
1(KWn (t)Kf,f)I < (an+bn)11f112
so IIKW (t)K11
n
0 as n
co,uniformly in t. Consequently, by
the Neumann series for the resolvent, one obtains that

(1+K(V(t)+E)K+KW (t)K)
n
-1
24
converges in norm to (1+K(V(t)+E)K) -1 as n -4- co.
We conclude
that (A (t)+1) -1 = K(1+K(V(t)+E)K + KW K) -1 K converges in norm,

n
n
-1
and hence strongly, to (H o +V(t)+E+1) . We are done by corollary 10.
Q.E.D.
As a special case of the previous theorem we note
Corollary 13: Let V,W be K-perturbations of H

constant E so that A
(t) = H
. There is a
0 + V(t) +EW(t) + E is a K-generator
for all E sufficiently small and U A (t,$) converges strongly to

U (t,$) for all t,s in J where A = H + V(t) + E as 6
0
A
O.
Example 14: Let V be a small Hermitian form perturbation of

1 /2
H 0 , i.e., Q(H 0 )C=Q(V) and IV(f,f)1 < a(H 1/2
o
f,H 0
f) + bI If
for all f EQ(H
l I2
) where 0 < a < 1 and 0 < b.< co. Let f be a
bounded C l function with bounded derivative. Let

A (t) = H
H +
0
ef(t)v
( ,$) as E
Example 15: Let H
+ E. Then U
(t , s) converges strongly to
0.
here denote
2
n
A on L (IR ). Let
be a real
distribution on 3e1. T is called an H 0 -small distribution if

there are numbers 0 < a < 1, 0 < b < 0. such that
1,11'2
f) + bill!
Ta f) < a(H 1/2
0 f,H 1/2
0
(36)
25
for all f in C 00 (R n )
(a,b) are the H
-bounds of
T.
Given such
a distribution one can define a Hermitian form T(f,g) = T(fg)

which can be extended to a Hermitian form
IT(f,f)I < a(H
on Q(H 0 ) satisfying
1 /2
1/2
f
f) + b(f,f)
0
' H
(37)
for all f in Q(11 0 ).

Let p be a C c (1R
) function which satisfies p(x) = 1 for

1
HI
Ilxil < 1 and 0 < p < 1. Set P. k (x) = p(x/k) and

yn (x) = p(kx)(fp(kx)dx) -1 . Define Vk (f,g) = T(y n *Ok fg)), so
co
n
that each V k is a C(IR ) function. The V 's are the regulark
izations of
Moreover H
+ V
the form sum H
and converge to T in the sense of distributions.

k
converges in the strong resolvent topology to
T.
See [3] for details.
Suppose f is C l on J. Define
II d
i
c ft-I LI.
ilifill
Let A(t) = H
f(t)V and f(t)V
Suppose I I I fl I I <
+ f(t)T and A (t) = H + f(t)Vk . Then

k
0
are K-perturbations of H
and by the preceding
remarks Ak (t) converges in the strong resolvent topology to A(t)

for each t on J. By theorems 11, 9 and corollary 10 U
(t,$)
k
converges strongly to U A (t,$) for each t,s in J while

U
t,$) ti U (t,$) for each infinite k in DT and every finite
A
A(
k
*
t,s in J.
26
For specific examples of
which may be chosen we refer
T'S
the reader to [3]. Here we note that
P
may be in L (ER) + L(IR n )
where p > 1 if n = 1, p > 1 if n = 2 and p > n/2 if n > 3. Also

T
may be a delta function concentrated on a C
compact hyper-
surface or the distributional directional derivative of a bounded

function.
If
and
are two H O -small distributions then for suitable
f's V(t) = f(t)
T1
(1 - f(t))T 2
is a K-perturbation of H 0 .
Choosing f so that f = 0 for t < 0 and f = 1 for t > 1 one can

treat the problem of switching from one potential to another.
By choosing f(t) = sin wt one obtains periodically varying
potentials.
One obtains considerable technical advantage in continuing
the study of time dependent singular perturbations of -A by
introducing infinite regularizations and by accepting infinitesimal errors. This naturally occurs because U
generator H
while U
H +V+E
0
+ V
H +V +E
0 k
+ E defined as an operator sum with V
has the generator H
For example, since
has the
* cQ
in C
k
c
+ V + E defined as a form sum.
vk e * L Pn )
for all p > 1, the product
formula of Faris [2] is valid and one can express the solution
u of the Schrodinger equation (1) as
m-1( it
A
e ra
u (t) eri,
j=0
it V ()
jt
m
us
(38)
27
with u(0) = u
for u
in Q(H 0 ),for m sufficiently large in IN.
2
n
Since L (IR ) i
is separable one m can be chosen so that (38)
holds for all u
in Q(H 0 ) and all finite t (see [13] for details
of this argument). Following Faris, (38) leads to a Feynman

path integral:
* 2n
For each pair of points (x 0 ,xm ) in IR we may view
* n (m-2)
as a space of "polygonal paths" connecting x
and x
These paths are not standard, but if x 0 and xm are finite and if
w
is any standard path connecting x
and x ; i.e.
m
w : [0,t] + AP and w(0) = st(x 0 ),w(t) = st(xm ), then there

are polygonal paths w
rk, I
so that w(s) rk, w (s) for all s in [0,t].
There are many more polygonal paths than standard paths. Since
*
for each t,Vk (t) is a C c function we let (Vk (t))(y) = Vk (y,t).
For each polygonal path x = (x 1 ,,xm _ 1 ) in
2(t/m)
IR
n(m-2)
let
(x.-x.
S(x o ,x,xm;m,t,V) =X j=1
t
V (x. jt)
-- .
k 3 'm m
Let u(,t)E L 2 (IRP) be the state at time t of the quantum

mechanical system with time dependent Hamiltonian H 0 + V(t)
with state at time 0 given by u(-,0)E Q(H o )

(U H
Thus
(t' 0))(u(-,0)) = u(-,t). Define the amplitude of the
corresponding quantum mechanical particle to go from x m at time

0 to x
at time t by
4211 74t90..47t
401,111,M,
K((x
,t),(x
m'
28
0)) =
(39)
iS(x,x,x
e
1R
-m,t,V )
m'
k
u(x ,O)dx
(m-2)-n
m
Since e
. 2
isA .
. -3n/2 ly /4s .
n
is convolution by (4ffis)
iI n IR , (38)
implies
u(.,t) ti
K((.,t),(x m ,0))u(x n ,O)dxm

*
IR
(40)
in * 2
n
n L (IR ).
(39) and (40) are a formulation of the Feynman path integral
for particles moving in singular and time dependent potentials.
2
Since adding a delta function to -A =- d 2 in L 2 (23) is
dy
equivalent to imposing a boundary condition on the maximally
defined
d2
, it follows that the problem of smoothly
dy 2
changing boundary conditions in time can be given a Feynman path
integral interpretation.
We conclude the discussion of this example by noting that,
in general, the solution to equation (1) is shown to exist by
proving certain approximate solutions converge weakly. However,
by accepting infinitesimal errors one can explicitly construct
the propagator. For this purpose we enter the interaction
itH 00
representation defining V(t) = e
-itH
V (t)e
k
where k remains
an infinite positive integer. Then t V(t) is an internal

D-generator so that
29
t t
(-i) P f
U(t,$) = 1 +
p=1
f ...f
s s
p-1
v(t 1
)dt ...dt
1
P P
converges in norm to a unitary propagator and

-itH
isH
0
5(t,$)e
Ua (t,$)= e0
is almost the propagator for equation
(1) in the sense that

U(t,$)x s rrt`; U a (t,$)x s
for all x s in Q(H ) and t,s finite. For the standard details,
0
see [10]. Thus we have defined time ordered exponentials of
certain quadratic form valued functions.
30
References
1.
P. Chernoff, Product Formulas, Nonlinear Semigroups and

Addition of Unbounded Operators, Mem. Am. Math. Soc.
140 1-121 (1974).
2.
W. Faris, Product Formulas for Perturbations of Linear

Propagators, J. Functional Analysis, 1, 93-108 (1907).
3.
I. Herbst and A. Sloan, Perturbation of Translation Invariant

2
Positivity Preserving Semigroups on L (at'), Trans.
Am. Math. Soc., 236, 325-360 (1978).
4.
T. Kato, Integration of the Equation of Evolution in a

Banach Space, J. Math. Soc. Japan, 5, 208-234 (1953).
5.
T. Kato, Perturbation Theory for Linear Operators, Springer,

Berlin, 1966.
6.
T. Kato, On Linear Differential Equations in Banach Spaces,

Comm. Pure and Applied Math., IX., 479-486 (1956).
7.
T. Kato, Linear Evolution Equations of Hyperbolic Type, J.

Fac. Sci. Univ. Tokyo, I, 17, 241-258 (1970).
8.
J. Kisynski, Sur les Operateurs de Green des Problems de

Cauchy Abstraits, Studia Math. XXIII, 285-328 (1964).
9.
E. Nelson, Internal Set Theory: A - New Approach to Non -standard Analysis, Bull. Am. Math. Soc. 83, 1165-1198
(1977).
10.
M. Reed and B. Simon, Methods of Modern Mathematical Physics

II Fourier Analysis, Self-Adjointness, Academic Press,
New York (1975).
11.
B. S imon, Quantum Mechanics for Hamiltonians Defined as

Quadratic Forms, Princeton Un. Press, Princeton, (1971).
12.
A. Sloan, An Application of the Nonstandard Trotter Product

Formula, J. Math. Phys. 18, 2495-2496 (1977).
13.
A. Sloan, A Note on Exponentials of Distributions, to appear,

Pacific J. of Math.
14.
K. Stroyan and W. Luxemburg, Introduction to the Theory of

Infinitesimals, Academic Press, New York, 1976.
15.
K. Yoshida, Functional Analysis Springer-Verlag, Berlin 1965.
16.
DaPrato, Giuseppe and Iannelli, Mimmo, On a Method for Studying

Abstract Evolution Equations in the Hyperbolic Case, Comm.
in Partial Differential Equations, 1(6), 585-608 (1976).

(Thru OCA) : Dr. Alan D

Enviado por

Dados do documento

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

(Thru OCA) : Dr. Alan D

Enviado por

Direitos autorais:

Formatos disponíveis

GEORGIA INSTITUTE OF TECHNOLOGY

OFFICE OF CONTRACT ADMINISTRATION

SPONSORED PROJECT INITIATION

"Applications of Nonstandard Analysis to Mathematical Physics"

National Science Foundation

Annual Letter Technical, Final Report

Dr. William G. Rosen

Library, Technical Reports Section

Division Chief (EES)

Office of Computing Services

Director, Physical Plant

EES Information Office

Project File (OCA)

Project Code (GTRI)

Security Coordinator (OCA)

1,11;orts Coordinator (OCA)

SPONSORED PROJECT TERMINATION

Applications of Nonstandard AnalysisMathematical Physics

Project No: G-37-603

Effective Termination Date:

Grant/Contract Closeout Actions Remaining:

Final Invoice and Closing Documents

( Scho ol/ UN:34MAX

Assigned to: Mathematics

Library, Technical Reports Section

Division Chief (EES)

EES Information Office

Project File (OCA)

Project Code (GTRI)

/Reports Coordinator (OCA)

GEORGIA INSTITUTE OF TECHNOLOGY

June 28, 1977

GEORGIA INSTITUTE OF TECHNOLOGY

June 16, 1978

Mr. James L. Bostick

An application of the nonstandard Trotter product formula

(Received 21 March 1977)

We seek an answer to the question of when a quantum

where T" v (u) is defined as the integral

(T, vu)(x 0 )=-_- c, fro, exp[iS(xo , x, x 0 ;n,t, V)]

exp(- itH)uz exp[- it (H + V u.

xu(x n)dx dx,

Combining (6) and (7) we find

An explic t computation' shows that

T 0, v [exp(- itHo/n) exp(- itV/n)r,

For all finite n and K, (4) shows that

T 0, v lc = [exp(- itHo/n) exp(- it VK/n)] ",

for V any real measurable function. Consequently,

exp(- itH)u=lim [exp(- itHo/n) exp(- itV K/n)]".

Suppose I/ is a self-adjoint operator on L 2 (IR3 ) which

in part by NSF MCS 76-97543.

Journal of Mathematical Physics, Vol. 18, No. 12, December 1977

so by the transfer principle (9) holds for all n, K in

For each n, K in IN, (2) shows that T n, vic (u) is given

IIcxII < 5. Then for n>N

f n(c) fo(c)II +.11 fo(c) fo(x)II

<MIIx- cll +X+Mile xil <3.1V16.

For the necessary approximation theorem in case (b)

Combining this lemma with the previous conclusion,

J. Math. Phys., Vol. 18, No. 12, December 1977

Feynman and A. Hibbs, Quantum Mechanics and Path

June 18, 1979

Mr. James L. Bostick