Você está na página 1de 15

Lake Wobegon Dice

David G. Storka,b and Jorge Moraledaa

Both authors contributed equally to this work


a
Ricoh Innovations, 2882 Sand Hill Road Suite 115, Menlo Park CA 94025 USA
b
Department of Statistics (Consulting Professor), Stanford University, Stanford CA 94305 USA

Abstract
We present sets of n non-standard dice—Lake Wobegon dice—having the following paradoxical property: On
every (random) roll of a set, each die is more likely to roll greater than the set average than less than the set
average; in a specific statistical sense, then, each die is “better than the set average.”
We define the Lake Wobegon Dominance of a die in a set as the probability the die rolls greater than the
set average minus the probability the die rolls less than the set average. We further define the Lake Wobegon
Dominance of the set to be the dominance of the set’s least dominant die and prove that such paradoxical
dominance is bounded above by (n − 2)/n regardless of the number of sides s on each die and the maximum
number of pips p on each side. A set achieving this bound is called Lake Wobegon Optimal. We give a constructive
proof that Lake Wobegon Optimal sets exist for all n ≥ 3 if one is free to choose s and p. We also show how to
construct minimal optimal sets, that is, that set that requires the smallest range in the number of pips on the
faces.
We determine the frequency of such Lake Wobegon sets in the n = 3 case through exhaustive computer search
and find the unique optimal n = 3 set having minimal s and p. We investigate symmetry properties of such sets,
and present equivalence classes having identical paradoxical dominance. We construct inverse sets, in which on
any roll each die is more likely to roll less than the set average than greater than the set average, and thus each
die is “worse than the set average.”We show the unique extreme “worst” case, the Lake Wobegon pessimal set.

Keywords: Lake Wobegon dice, nontransitive dice, paradoxical dominance, non-zero-sum game

1 Background
Efron introduced nontransitive dice, in which when rolled in pairs, most frequently die A beats die B, B beats C,
and C beats A. This is a statistical analogy of the game Rock, paper, scissors (also called rochambeau, roshambo,
and jan-ken-pon), in which each element or action beats another and is beaten by yet another.[?] In such a set there
is no “best” or dominant element. Of course the scalar mean values of the dice cannot exhibit nontransitivity; it
is only in the statistical case of rolls of different pairs of dice where nontransitivity can arise. Such nontransitive
dominance or paradoxical dominance can occur in preference for artworks and for political candidates, as well as
in dominance among warriors avatars in some computer games, some athletes, and among teams, for instance the
common claim that “Patriots beat Colts, Colts beat Broncos, and Broncos beat Patriots.”[?, ?] Figure 1 shows a set
of three nontransitive dice.
We generalize Efron’s notion to the case in which on each roll of the full set of n dice each die is more likely to
roll greater than the current average of the set than to roll less than the current average of the set. Informally, where
Efron studied the case where each die need beat merely one other die, in our case each die must beat the average
of all the dice. We use the term Lake Wobegon to refer to this paradoxical property in honor of the children in the
mythical Minnesota town Lake Wobegon chronicled by humorist Garrison Keillor, “Where all the women are strong,
all the men are good looking, and all the children are above average”[?].
We begin in Sect. 2 with our notation and definitions, particular our formal definition of Lake Wobegon Dominance
as a type of paradoxical dominance. Section 3 contains our main theoretical results. We first present two symmetry
properties of Lake Wobegon dice and then our most significant results: the concept of Lake Wobegon Optimality as
the maximum of Lake Wobegon Dominance and a proof that this optimality is bounded above by (n − 2)/n regardless
of the number of sides s on each die and the maximum number of pips p on each side. We give a constructive proof

1
2 1 3
4 2 4 6 1 6 5 3 5
9 8 7
A B C
9 8 7

Figure 1: A set of nontransitive dice (after Efron), where in fair rolls of pairs of dice, on average A beats B, B beats C,
and C beats A. In our convention, dice are denoted by a non-decreasing list of their pip values: A = {2, 2, 4, 4, 9, 9},
B = {1, 1, 6, 6, 8, 8} and C = {3, 3, 5, 5, 7, 7}, and the spatial arrangements on the physical dice are irrelevant. In this
set, P [A > B] = P [B > C] = P [C > A] = 5/9 = 55.56%.

that Lake Wobegon Optimal sets exist for all n if one is free to choose s and p, and we show how to construct minimal
optimal sets, that is, ones that require the smallest range in the number of pips on different sides.
In Sect. ?? we investigate minimal Lake Wobegon Optimality. We find there is a set of Lake Wobegon Optimal
dice that is minimal in three different senses: number of dice, number of sides and maximum number of pips on
a side. In Sect. ?? we report statistical results on the prevalence of Lake Wobegon sets for the n = 3 case, found
through exhaustive computer search. We mention two properties of Lake Wobegon dice in Sect. ?? and conclude in
Sect. ?? with some possible applications and speculations.

2 Notation and definitions


We have been informally referring to dice, appealing to the reader’s intuition. We now formally define this term in
our framework.
Definition 1 (die, side, pip, coin, roll). We formally use the terms s-sided die/dice to refer to integer valued
multinomial random variables where the probability of each value is of the form k/s, s ∈ N, k ∈ {1, 2, . . . , s}. For any
set of s-sided dice, we can produce another dice with ms, m ∈ N sides with corresponding repetition of the pip labels
to yield an equivalent set of dice. We use the term coin to refer to a 2-sided die. Furthermore, the integer value on
each side—the number of its pips—is in the set {1, 2, . . . , p}, and is possibly the same on different sides. Without loss
of generality (cf., Theorem 1), we demand that a 1 appear in every set of dice. We denote the number of pips on the
sides in non-decreasing order, for instance two s = 6, p = 7 dice are: {1, 2, 3, 4, 4, 7} and {2, 2, 4, 5, 5, 6}. Occassionally
we use the alternative notation of listing the unique number pips followed by the probability  with  which  they  occur
1 1 1 1 1

in parenthesis.
 1 Using this notation, the two dice above would be denoted 1 6 , 2 6 , 3 6 , 4 3 , 7 6 and
1 1 1
  
2 3 ,4 6 ,5 3 ,6 6 .
Note that the familiar 6-sided fair die, {1, 2, 3, 4, 5, 6}, naturally meets our definitions. Since we consider physical
dice as random variables, we consider them distinguishable if and only if their set of pip values differ; that is, we do
not care about the arrangement of pips on the physical die since only the probabilities a side emerge in a roll are
important here. The number of distinguishable dice, according to our convention, is the number of different sets of
p objects chosen s at a time with replacement, that is, s+p−1 s . A roll of a set of dice is an independent uniformly
distributed random selection of the integer pip value of each fair die in the set, or equivalently according to the
probabilities listed in our alternate representation. We let M denote the set mean on a given roll.
In the reminder of this section, let χ = {X1 , X2 , . . . , Xn } be a set of n ∈ N real-valued random variables. Note
that some of the following definitions and theoretical results in the paper apply to finite sets of real-valued random
variables, and not just to finite sets of dice.
Definition 2 (Dominance of a variable in a set). For each variable Xi ∈ χ, we define its paradoxical dominance or
Lake Wobegon Dominance, Fi , as the probability that Xi is greater than the average of variables X1 , X2 , . . . , Xn on
a roll minus the probability that it is smaller:

Fi = P [Ei > 0] − P [Ei < 0], (1)


where

2
n
X
Ei = nXi − Xj . (2)
j=1

Not that, much as with Efron’s nontransitive dice, we do not care by how much a die is above or below the set mean,
just the probability of those events.
Definition 3 (Lake Wobegon Dominance of a set of variables). We define the paradoxical dominance or Lake Wobegon
Dominance of the set χ, Fχ , as the minimum of the Lake Wobegon Dominance of the variables in the set:
n
Fχ = min Fi . (3)
j=1

Definition 4 (Lake Wobegon set). The set χ is said to be Lake Wobegon if and only if its Lake Wobegon Dominance
Fχ is greater than zero. Thus, a set of dice is Lake Wobegon if the probability that each die in the set rolls greater
than the set’s average on that roll, M , is greater than the probability that it rolls less than M .
A simple illustration of a Lake Wobegon set for n = 3, s = 3 and p = 2 is χ = {{1, 2, 2}, {1, 2, 2}, {1, 2, 2}}, where
the probability each die (i = 1, 2, 3) exceeds the group average on a roll is P [Ei > 0] = 10/27, and the probability it
is less than the average is P [Ei < 0] = 8/27 (and the probability it equals the average is P [Ei = 0] = 9/27), as can
be seen by enumerating the sn = 27 combinations of faces. Thus the dominance of each die is Fi = 2/27 > 0, and
thus the paradoxical dominance of this set χ is Fχ = 2/27 > 0. Hence this set is Lake Wobegon.
The Lake Wobegon property does not imply nontransitive dominance: the set {{1, 2, 2}, {1, 2, 2}, {1, 2, 2}} is
Lake Wobegon, but clearly not nontransitive, since the dice are identical. Conversely, a nontransitive set need not
be Lake Wobegon, as illustrated by the set {{1, 1, 6, 6, 8, 8}, {3, 3, 5, 5, 7, 7}, {2, 2, 4, 4, 9, 9}}, shown in Fig. 1.

3 Theoretical results
This section contains our main theoretical results. We first presents two symmetry properties of Lake Wobegon dice
and then our most significant results: The concept of Lake Wobegon Optimality as the maximum of Lake Wobegon
Dominance, a proof of several theorems associated with optimal sets, and a procedure to construct minimal optimal
sets.

3.1 Lake Wobegon Symmetry


First we show a partition of the set of sets of dice into equivalence classes, where all sets in the same equivalence
class are related by an affine transformation on the number of pips showing on their sides. Then we present other,
inverse, sets, in which on any roll each die is more likely to roll less than the set average than greater than the set
average.
Theorem 1 (Affine equivalence). The affine transformation x → xa + b : a ∈ R+ , b ∈ R on each of the values
that each of the variables in any set χ of n multinomial real-valued random variables preserves the Lake Wobegon
Dominance of each of the variables, as well as the Lake Wobegon Dominance of the set. Such transformations, then,
yield a natural partition of the set of finite sets of multinomial random variables into equivalence classes.
Proof. It is straightforward to verify that the variables or dice that are above (below) the average in a roll Ei > 0
(Ei < 0) in Eq. 2 continue to be above (below) the average if all sides of all the variables have been transformed using
the above affine transformation. Of course, the transformation does not change any probabilities of appearance on
a roll. In short, the Lake Wobegon dominance is unchanged by the transformation.
Theorem 2 (Inverted Lake Wobegon). Performing an affine transformation x → xa + b : a ∈ R, a < 0, b ∈ R on
each of the values that each of the variables in any set χ of n multinomial real valued random variables produces an
inverted Lake Wobegon set where the Lake Wobegon Dominance of each variable has opposite sign as in the original
set. Thus if the original set was Lake Wobegon, the resulting set will be Inverted Lake Wobegon: The probability
each die will be less than the group mean is greater than the probability it will be greater than the group mean.
Proof. This proof is analogous to that of theorem 1.

3
The minimum value of any pip is 1 (by our convention) and the maximum over all dice in a set is pmax . An
inversion transformation of particular interest that produces an Inverted Lake Wobegon set from a Lake Wobegon set
is pnew ← pmax + 1 − p, as it preserves the minimum and maximum value of pips, as well as transforming integers to
integers. We define Fˆχ , the natural symmetric measure to Fχ of Eq. 3 as:
n
Fˆχ = − max −Fi . (4)
j=1

Any theorem for Lake Wobegon Dominance leads to a corresponding complementary claim for inverted Lake
Wobegon sets, where F̂ = F .

3.2 Lake Wobegon Optimality


This section contains our main results: We begin by showing that Fχ = (n − 2)/n is an upper bound of the Lake
Wobegon Dominance Fχ of a set χ comprised of n dice. Then we show how to construct for all n ∈ N a set of n
dice that achieves this bound as long as we are free to choose both the maximum number of pips on any side p, and
the number of sides on the dice s. Finally we show that the set constructed is minimal in the sense of requiring the
minimum number of pips in the side showing the most pips.
Theorem 3 (Optimality Bound). The tightest upper bound of the paradoxical dominance of a set of n real-valued
random variables is Fχ = (n − 2)/n.

Proof. In every roll at least one die needs to be below average, unless all dies are equal to the average. (The latter
case does not improve the Lake Wobegon Dominance of any die above 0, so it cannot improve the Lake Wobegon
Dominance of the set above 0 either.) Thus an upper bound of the Lake Wobegon Dominance of the set occurs when
in every roll exactly one die is below average and n − 1 are above average and each die is below average in exactly
1/n of the rolls. This yields an upper bound for the Lake Wobegon Dominance of the set
n−1 1 n−2
Fχ ≤ − = . (5)
n n n
A more formal proof of this bound is given in the Appendix, Theorem ??. Note that in the limit, lim Fχ → 1, that
n→∞
is, in the limit of large number of dice, every die is almost always above the set average.
The next result shows that Lake Wobegon Optimality is achieved for any number n of dice (i.e., independent
variables). Contrast this result with the one in Theorem ?? where as a side result of formally proving the optimality
bound, it was shown that Lake Wobegon Optimality is achieved for all n if variables are allowed to be correlated.

Theorem 4 (Optimality existence). For all n ∈ Z, n ≥ 3 the set of n dice



vmax − 1 (1)
1 1
  
vmax 2 , vmax − n 2
2 1
  
vmax 3 , vmax − n(n − 1) 3
··· (6)
i−1 1
  i−2

vmax i , vmax − n(n − 1) i
···
n−1 1
 
vmax − n(n − 1)n−2

vmax n , n ,
where

vmax = 1 + n(n − 1)n−2 , (7)


is Lake Wobegon Optimal.
Proof. As shown above, a necessary and sufficient condition for Lake Wobegon Optimality is that every die be below
the set’s mean M with probability 1/n, and above M otherwise. Let Hi be the event where die i rolls its lowest side
and all dice k > i roll their highest side. Notice that events Hi , i ∈ 1, . . . , n are mutually exclusive and collectively
exhaustive. We prove every die is below or equal to average with probability 1/n by showing that:

1. event Hi occurs with probability 1/n and

4
2. die i is below the average of the roll if and only if event Hi occurs, and is above the average of the roll otherwise.
Consider the first claim. Since dice are independent variables, the probability that the ith die rolls its lowest side
and all dice k > i roll their highest side is the product of the probabilities of each die rolling the given side, given in
parentheses in Eq. 6,
 1 i n−1 1
P vi = vmax − n AND ∀k > i : vk = vmax − n(n − 1)n−2 =

··· = . (8)
i i+1 n n
We prove the second claim in two parts. First we remark that if event Hi occurs then the ith die is showing the
smallest value of all dice. Since they are not all equal, it is below the average of the roll. Second we show that all
other dice are above the average of the roll.
Let j < i, j 6= 1 be the die showing the next lowest value in the roll besides i. Showing that die j is above average
proves that all dice except for i are above average. The largest mean that this roll can have occurs when all dice
except for i and j are showing their largest side. This mean value is:

1
(vmax − 1) + vmax − n(n − 1)j−2 + vmax − n(n − 1)i−2 + (n − 3)vmax =
  
M=
n (9)
1
= vmax − (n − 1)j−2 − (n − 1)i−2 − .
n
We subtract the value that die j is showing, vmax − n(n − 1)j−2 , from the value of the average, M , and confirm that
die j is indeed above the average of the roll:
 
1
vmax − n(n − 1)j−2 − M = vmax − n(n − 1)j−2 − vmax − (n − 1)j−2 − (n − 1)i−2 −

n
1
= −n(n − 1)j−2 + (n − 1)j−2 + (n − 1)i−2 + + n(n − 1)j−2
n (10)
j−1 i−2 1
= −(n − 1) + (n − 1) +
n
1
≥ .
n
(The special cases when i = 1 or j = 1 are analogous.)
We show next that the Lake Wobegon Optimal set produced in Theorem 4 for a given value of n is minimal in
the sense that no other Lake Wobegon Optimal set can be constructed for the same value of n without at least one
side showing a value greater or equal than vmax
Theorem 5 (Minimal pips). For all n any Lake Wobegon Optimal set has at least one side showing the (pip) value
vmax = 1 + n(n − 1)n−2 or greater.
Proof. We first introduce some notation: Let set χn comprised of n dice be Lake Wobegon Optimal. Let li , ui , for i ∈
{1, . . . , n} be the minimum (lower) and maximum (upper) value on die i’s side, respectively, thus ∀ i : li ≤ ui .
Moreover let the dice be sorted such that i < j =⇒ li ≤ lj . For each die i, we define its spread, ∆i , as

∆i = 1 + ln − li . (11)
Our proof proceeds as follows: First we show that max ui > ln . This proves that the spread of a die is a lower
i
bound on the difference between the maximal number of pips in any side on the set χn and the number of pips in
the smallest side of die i. Afterwards, proving that ∆1 ≥ vmax − 1 will complete our proof.
To verify max ui > ln consider the roll {u1 , . . . , un−1 , ln }. Since χn is Lake Wobegon Optimal, only one die can
i
be below the average of the roll. Assume die j 6= n is the one die below the average. This would imply uj < ln which
would, in turn, imply that in any roll, die n would always be higher than die j, and thus die n could never be the
sole die below the average. Such a case is not possible because χn is Lake Wobegon Optimal, thus die n must be the
one below the average and u1 , . . . , un−1 > ln . Note that we can achieve an equality bound by adding 1 because all
values are integer. We will use this result several times.

min{u1 , . . . , un−1 } ≥ 1 + ln . (12)


We complete our proof by computing a lower bound on ∆1 . This is done recursively. First we compute a lower
bound on ∆n−1 as an initial case followed by a lower bound on ∆i , i ∈ {1, . . . , n − 2} as a function of ∆i+1 for the
recursion step.

5
Initial step. Bound on ∆n−1
Consider the roll {u1 , . . . , un−2 , ln−1 , ln }. The only die below average is die n − 1 since ln−1 ≤ ln < u1 , . . . , un−1 .
Thus we have

u1 + · · · + un−2 + ln−1 + ln < nln Die n is above average


(n − 2)(1 + ln ) + ln−1 + ln < nln Substituting ui by 1 + ln according to Eq. 12
−1 − ln + ln−1 < 1 − n Substracting n − 1 + nln on both sides (13)
∆n−1 > n − 1 Multiplying both sides times -1 and using definition of ∆i in Eq. 11
∆n−1 ≥ n Equality bound on ∆n−1 obtained because all variables are integers

Recursive step. Bound on ∆i , i ∈ {1, 2, . . . , n − 2} as a function of ∆i+i


Consider the roll {u1 , . . . , ui−1 , li , li+1 , ui+2 , . . . , un }. The only die below average is die i since li ≤ li+1 ≤ ln ≤
u1 , . . . , un . Thus we have

u1 + · · · + ui−1 + li + li+1 + ui+2 + · · · + un < nli+1 Die i + 1 is above average


(n − 3)(1 + ln ) + li + ln < (n − 1)li+1 Simplifying li+1 terms and substituting
ui according to Eq. 12 and ln ≤ un
(n − 2)(1 + ln ) + li ≤ (n − 1)li+1 Equality bound is attained by adding
one because all variables are integers
−1 − ln + li ≤ (n − 1)(li+1 − 1 − ln ) Substracting (n − 1)(1 + ln )
∆i ≥ (n − 1)∆i+1 Using the definition of ∆ from Eq. 11
(14)
Solving the recursion for ∆ defined by Eqs. ?? and ?? yields

∆i ≥ n(n − 1)n−i−1 , (15)


and thus, using the definition of vmax in Eq. 7, we conclude

∆1 ≥ n(n − 1)n−2 = vmax − 1. (16)

Figure ?? shows F and pmax derived in Theorem 6. At large n, the Lake Wobegon dominance approaches 1.0,
meaning that for large n, every die is nearly always above average.

1. 1010
0.9 109
0.8 108
0.7 107
0.6 106
pmax
F

0.5 105
0.4 104
0.3 103
0.2 102
0.1 101
0. 100
3 4 5 6 7 8 9

Figure 2: The maximum value of the Lake Wobegon Dominance of a set F = (n − 2)/n (left, linear axis) and the
maximum number of pips on any side in the set, pmax = 1 + n(n − 1)n−2 (right, logarithmic axis) as given in Eqs. 5
and 7.

Notice that the probabilities involved in the minimal optimal set of Lake Wobegon dice given in Theorem 4 imply
the dice in the set need to have a number of sides equals to the LCM[1, 2, . . . , n]. Table ?? in the Appendix shows

6
the Lake Wobegon optimal set of dice produced using the above results for 3 ≤ n ≤ 9. Figures ?? and ?? show the
n = 3 (cube) and n = 4 (dodecahedron) Lake Wobegon optimal dice.

6 7 1
6 6 6 4 4 4 7 1 7
6 7 7
6 7 7

dice_photo.pdf

Figure 3: Lake Wobegon optimal set of dice for n = 3, s = 6, and p = 7. The probability each die is above the group
average on a roll of all three dice is 2/3 and the probability it is less than the group mean is 1/3; thus F = 1/3.
Moreover, on every roll of these dice, exactly two dice are guaranteed to be above the group mean M . This set is
minimal in three distinct senses: n, s and p. It is the unique optimum set out of the 115,264,382 distinguishable sets.

6 7 7 7
36 3 6 33 3 3 37 3 5 37 3 1
36 3 37 3 37 2 37
36 36 37 33 25 37 1 37
36 36 33 37 37 37 37 37
36 3 33 3 25 3 1 3
36 3 6 37 3 7 25 3 7 37 3 7
6 3 7 7

Figure 4: Lake Wobegon optimal set of dice for n = 4, s = 12, and p = 37. The probability each die is above the
group average on a roll of all three dice is 3/4 and the probability it is less than the group mean is 1/4; thus Fχ = 1/2.
Moreover, on every roll of these dice, exactly three dice are guaranteed to be above the group mean.

If it is not required that all dice in a set have the same number of sides, then having one die with one side,
another with two sides, up to one die with n sides is sufficient to have a Lake Wobegon Optimal set that is minimal.
Another way to obtain the required probabilities is with rotating game wheels, where the angles of the numbered
sectors correspond to probabilities. Figure ?? shows optimal disks or game wheels for 5 ≤ n ≤ 7.

Corollary 5.1 (Low sides are distinct). The value in the lowest side of each die in a Lake Wobegon Optimal set
appears in no other die of the same set.
Proof. This results follows immediately from eq. ?? which precisely indicates the minimal difference between the
lowest sides of two dice in a Lake Wobegon Optimal set.

7
n=5

LakeWobegonOptimalDisks5.pdf

n=6

LakeWobegonOptimalDisks6.pdf

n=7

LakeWobegonOptimalDisks7.pdf

Figure 5: Lake Wobegon Optimal game wheels, for 5 ≤ n ≤ 7. Here each wheel is divided into one or two sectors,
numbered as shown. Each player spins the disk (randomly) selecting the value pointing in a given direction, say up.
The probabilities of the dark-colored sections are 11 , 12 , 13 , . . . , n1 .

4 Minimal Lake Wobegon Optimality


In this section we investigate Minimal Lake Wobegon Optimality. In particular we show that:
Theorem 6. The optimal set of dice {{6, 6, 6, 6, 6, 6}, {4, 4, 4, 7, 7, 7}, {1, 1, 7, 7, 7, 7}} of Fig. ?? is minimal in three
senses:

1. n = 3 is minimal: No set with fewer than than three dice is Lake Wobegon (and hence Lake Wobegon Optimal).
2. p = 7 is minimal: No set of dice with fewer than seven pips per side is Lake Wobegon Optimal.
3. s = 6 is minimal: No set of three dice with fewer than six sides is Lake Wobegon Optimal.

Proof. Claim 1: n = 3 is minimal


With only two variables, n = 2, the Lake Wobegon Dominance is at most zero because on every roll either both
variables are equal to the mean, or one variable is above the mean and one variable below the mean. Thus the sum of
the Lake Wobegon Dominance of both variables is zero. Thus the Lake Wobegon Dominance of the least dominant
variable is non-positive, and hence the Lake Wobegon Dominance of the set is non-positive.

Claim 2: p = 7 is minimal
The fact that n = 3 is minimal together with Theorem 5 show that p = 1 + n(n − 1)n−2 = 7 is minimal.

Claim 3: s = 6 is minimal for sets of 3 dice


This result is proved by considering the various cases of sets of three dice where s < 6 and demonstrating they are
not Lake Wobegon Optimal. These cases are explored in the results that follow, thus the proof of claim 3 is given at
the end of the section.
Theorem 7 (No Lake Wobegon coins). A set comprised of n ∈ N fair 2-sided dice—or coins—cannot be Lake
Wobegon.

8
Proof. Let the coins have lower and upper pip values li and ui , respectively, for i = 1, . . . , n. The probability of each
value emerging on a roll is 1/2. Then the set of n coins is:

χ = {{l1 , u1 }, {l2 , u2 }, ..., {ln , un }} where ∀i : li ≤ ui . (17)


Assume the set χ is Lake Wobegon. Note that not all sides of all coins show the same number as otherwise the set
is trivially not Lake Wobegon, that is,

@r s.t. ∀ i : li = ui = r. (18)
Call i∗ the coin that has the lowest side of all coins (∀j : li ∗ ≤ lj ). In a roll, coin i∗ will be below the average
every time it shows its lower side (which occurs with probability 1/2) unless all coins have identical lower sides (i.e.,
unless ∀i, j : li = lj ). Thus a necessary condition for set χ to be Lake Wobegon is that all coins show the same
number, L, on their lower side:

∀i, j : li = lj = L. (19)
Remark also that another necessary condition for set χ to be Lake Wobegon is that no coin shows the same
number in both sides:

∀j : uj > L, (20)
because a coin i+ for which ui+ = L can never be above the average.
In a roll, a coin from a set χ exhibiting the three properties shown in Eqs. ??,?? and ?? will be below the average
n−1
at least every time it shows L unless all the other coins are also showing L. This occurs with probability 2 2n−1 .
Consider coin i× with the lowest high side such that ∀j : ui× ≤ uj . Coin i× will be above average at most every
n−1
time it shows ui× unless all other coins are also showing their high sides. This occurs with probability 2 2n−1 . Thus
the Lake Wobegon Dominance of coin i× is at most 0, and thus the Lake Wobegon Dominance Fχ of set χ is at most
0 and thus by definition the set is not Lake Wobegon.
Lemma 1 (N3, S3 non-optimality). It is impossible for three 3-sided dice to be Lake Wobegon Optimal.
Proof. Let the the three dice be D1 , D2 and D3 such that all their sides are strictly positive integers, and at least
one of the sides of D1 is equal to 1. This can always be achieved because of Theorem 1 (Affine Equivalence). Assume
D1 , D2 and D3 are Lake Wobegon Optimal and let U be the number of rolls where D2 < D3 when dice D2 and D3
are rolled (total of 9 rolls).
Lake Wobegon Optimality implies that when D1 rolls a 1, D2 and D3 both roll a number larger than one since
only one die can be below average in a roll. Since optimality implies each die being below average exactly 31 of the
time, exactly one side of D1 can have a one, and D1 will be below average every time it rolls a 1.
Of the reminder 18 possible rolls (those where D1 > 1), exactly 2U will have D2 < D3 . Since U is integer, there
cannot be exactly 9 rolls (1/3 of the time) with D2 being below average, which contradicts the hypothesis that this
set of dice is Lake Wobegon Optimal.
Lemma 2 (N-S necessary condition). There exist no Lake Wobegon Optimal set of n s-sided dice for sn > 0 mod n.
Proof. If a set of dice is Lake Wobegon Optimal then exactly a fraction (n − 1)/n of all roll combinations must have
each die being above average. The total number of roll combinations is sn . Thus (n − 1)sn /n must be integer. Since
n − 1 and n cannot have any common factors, this yields the desired result.
Thus s = 6 is minimal for three sided dice since the Lake Wobegon Dominance is trivially 0 for s = 1. Theorem ??
(No Lake Wobegon Coins) shows it to be zero as well for s = 2. Lemma ?? (N3,S3 non-optimality) shows that Lake
Wobegon Optimality is not achived for n = 3, s = 3, and Lemma ?? (N-S necessary condition) when applied to n = 3
shows Lake Wobegon Optimality is not achieved for s = 4 or s = 5.
Note that applying the results of Theorem 2 (Inverted Lake Wobegon) to the minimal Lake Wobegon Optimal set
{{6, 6, 6, 6, 6, 6}, {4, 4, 4, 7, 7, 7}, {1, 1, 7, 7, 7, 7}} (with F = 1/3) produces the minimal Lake Wobegon Pessimal set
{{2, 2, 2, 2, 2, 2}, {1, 1, 1, 4, 4, 4}, {1, 1, 1, 1, 7, 7}} shown in figure ??.

9
2 1 7
2 2 2 4 4 4 1 7 1
2 1 1
2 1 1

Figure 6: Minimal Lake Wobegon pessimal dice for (n = 3, s = 6, p = 7), derived from the minimal Lake Wobegon
optimal dice in Fig. ??. On every roll of the set, the probability each die is below the group mean is 2/3. Moreover,
on every roll of these dice, exactly two dice are guaranteed to be below the group mean. Thus, according to Eq. 4,
this set has F̂ = 1/3.

5 Occurrence statistics
We explored the frequency of occurrence of Lake Wobegon and Lake Wobegon Optimal sets in a computationally
tractable case. Let k = s+p−1

s be the number of distinguishable dice of s sides each of which can have between
1 and p pips. Then the total number of different sets of n such dice is k+n−1

n . (Here we temporarily relax our
constraint and do not demand a 1 appear within every set.) For n = 3, s = 6, p = 7, the total number is 131,908,700,
as shown on the bottom row of Table ??. For each such set, the roughly sn possible rolls must be computed and
F computed, which takes less than a day on a workstation. (For n = 4, s = 6, p = 7, the number of distinguishable
sets is 30, 569, 841, 225 and exhaustive search is prohibitive even on a large cluster of workstations, even exploiting
symmetries.)

maximum number of pips pmax 1 2 3 4 5 6 7


number of distinguishable sets 1 84 4060 102,340 1,565,620 16,542,064 131,908,700
number of distinguishable equivalent sets 1 83 3894 98,116 1,459,223 14,972,305 115,264,382

Table 1: The number of distinguishable sets of n = 3 dice each having s = 6 sides, as a function of the number of
pips p. The bottom row is the number of equivalence classes. For instance, when pmax = 2 the set consisting of three
identical {1, 1, 1, 1, 1, 1} is equivalent by the symmetry transformation (Theorem 1) to {2, 2, 2, 2, 2, 2} by either the
addition of 1 to each pip, or multiplication by 2 to each pip.

We performed an exhaustive computer search and computed the paradoxical dominance F for each set for n =
3, s = 6, p = 7. We found that 623,105 combinations of the 115,264,382 distinct combinations of three such dice up
to symmetry transformations (see Theorem 1) are Lake Wobegon, or 0.541%. Moreover, there was but a single Lake
Wobegon Optimal set, as shown in Fig. ??.

6 Other properties
We mention in passing some other properties of Lake Wobegon dice.

• Each die in the Lake Wobegon set {{1, 6, 6, 6, 7, 7}, {1, 6, 6, 6, 7, 7}, {4, 4, 4, 7, 7, 7}} has mean 11/2 and para-
doxical dominance 7/36, so the set has Lake Wobegon dominance 7/36. Exhaustive search shows that this set
has the largest Lake Wobegon Dominance among all sets of three 6-sided dice with equal means and up to 7
pips per side.
• The Lake Wobegon Dominance of a set χ of n identical s-sided dice of the form {1, P, P, ..., P }, where P > 1,
is:

s − 2 (s − 1)n − 1
Fχ = − . (21)
s sn
We remark that this set has the property that any subset of it with more than three dice is also Lake Wobegon.
Anecdotically, for a given number of sides s, the number of dice n that maximizes the Lake Wobegon Dominance
Fχ of the set is the floor or ceiling of

10
K
105 æ æ
æ æ æ
æ
æ
æ ææ æ
æ æ æ
104 æ æ æ
æ æ æ
æ æ
æ
æ æ æ
3 æ
10 æ
æ æ
æ æ
æ
æ
æ
102
æ

æ æ
æ
æ æ æ
æ
1
10
æ æ
æ
æææ æ æ F
19 29 13

Figure 7: A histogram of the number K of distinguishable set of n = 3, s = 6, p = 7 sets that are Lake Wobegon, as
a function of the paradoxical dominance F of Eq. 3. The binning increment along the abscissa is 1/s3 = 1/72. Note
the logarithmic ordinate as well as the lone optimal set at F = 1/3, the set shown in Fig. ??.

log(log(s)) − log(log(s) − log(s − 1))


n∗ = . (22)
log(s − 1)
It remains an open question whether the minimal optimal sets shown in equation 6 also have this property.

7 Conclusions
In conclusion, we have defined a novel type of paradoxical dominance and we have identified a new class of sets of
nonstandard dice having such paradoxical dominance, that is, in which each die individually is statistically superior to
the current average of the entire set. We proved that under general conditions, the maximum paradoxical dominance
of a set of n dice is bounded by F = (n − 2)/n; in the large n limit, every die rolls nearly always above the set
average. We showed how to construct sets of dice that achieve this bound for all values of n. Furthermore we have
shown that the sets constructed by our method are minimal in the maximum number of pips per side p required.
An outstanding problem is to prove the minimal optimal sets in the sense of the smallest number of sides s
required for a given n. We conjecture these sets may indeed be the same ones that are minimal in the p sense and
that require s = LCM[1, 2, . . . , n] sides, the minimal according to our construction.
Lake Wobegon dice, and Lake Wobegon distributions more generally, could provide models for non-cooperative,
non-zero-sum games.[?] We can imagine that some colluding economic agents might engage in policies that each
agent properly predicts will provide individual benefit above the group average.
Whether such policies can be found in real economic systems, and how they work, is future work.

Acknowledgements
The authors would like to thank Brad Efron for words of encouragement.

Appendix
The following theorem and proof show more formally than theorem 3 that (n − 2)/n is an upper bound of the Lake
Wobegon Dominance of a set, and it shows, furthermore, that this upper bound is achieved if one allows for the
variables to be correlated.

11
Let a1 , a2 , ..., an ∈ {0, 1}n be indicator variables such that ai = 1 iff Ei > 0 and ai = 0 iff Ei < 0, where Ei was
defined in Eq. 2. For the joint distribution of n variables we define the 2n following probability masses:
" ( #
n Ei > 0, if ai = 1
p2a1 ,a2 ,...,an = P AND . (23)
i=1 Ei < 0, if ai = 0
(Note that there should be no confusion between our use of p in this Appendix for a probability and p in the body
of the paper for pip value.) For example, for three variables, p21,0,1 represents the probability that variables one and
three are above the average of a roll and variable two is below the average of a roll. Note that the p2a1 ,a2 ,...,an variables
are mutually exclusive but not collectively exhaustive, since the cases where a variable is equal to the average of the
roll (Ei = 0) are not included.
Theorem 8 (Formal Optimality Bound). The maximum Lake Wobegon Dominance Fχ of a set χ comprised of n
real-valued random variables is Fχ = (n − 2)/n. Moreover, this value is achieved when:
 n
 1 , if P a = n − 1
i
p2a1 ,a2 ,...,an = n i=1 (24)
0, otherwise.

This equation states that the bound is achieved if in every roll exactly n − 1 of the variables are above the average
and exactly one variable is below the average, and the probability of each variable being the one below the average is
1/n.
Proof. Note that the Lake Wobegon Dominanceof the set, Fχ is less than the average Lake Wobegon Dominanceof
the variables:
n
n 1X
Fχ = min Fi ≤ Fi . (25)
j=1 n i=1

By the definition of Lake Wobegon Dominance and of p2a1 ,...,an , the Lake Wobegon Dominance of each variable, Fi ,
can be written as follows:
X X
Fi = p2a1 ,...,an − p2a1 ,...,an . (26)
a1 ,...,an ∈ a1 ,...,an ∈
{0,1}i−1 ,1,{0,1}n−i {0,1}i−1 ,0,{0,1}n−i

We substitute Eq. ?? into ?? and find:


n−1
1X X
Fχ ≤ (2j − n) p2a1 ,...,an . (27)
n j=1 a1 ,...,an ∈
{0,1}n | n
P
a =n
i=1 i

We
P next use Lagrange multipliers to solve for the 2n variables of the form p2a1 ,...,an while imposing the condition that
p2a1 ,...,an ≤ 1. This leads to the following necessary condition for the above expression to reach a maximum:
n
X n
X
p2a1 ,...,an 6= 0 AND p2b1 ,...,bn 6= 0 → ai = bi . (28)
i=1 i=1

Thus, given that p21,1,...,1 = p20,0,...,0 = 0 because the events they denote are impossible (at least one variable must
be below average and at least one above average), we obtain the desired bound:

n
1X n−1 2j − n X
max F i = max p2
p2a1 ,...,an
n i=1
j=1 n a1 ,...,an ∈ a1 ,...,an
Pn
{0,1}n | a =j
i=1 i

2j − n
n−1 (29)
≤ max
j=1 n
n−2
= .
n

12
Corollary 8.1 (Optimality equivalence). Optimality is preserved if Lake Wobegon Dominance is defined as P [Ei >
0], instead of P [Ei > 0] − P [Ei < 0] in Eq. 3.
Proof. This result follows from the fact that the optimal solution found in Theorem ?? with the original definition
of Lake Wobegon Optimality has P [Ei = 0] = 0 for all i.

13
Dice Fχ Fχ
{{6, 6, 6, 6, 6, 6}, {4, 4, 4, 7, 7, 7}, {1, 1, 7, 7, 7, 7}} 72/216 0.333
{ {6, 6, 6, 6, 6, 7} , {4, 4, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 56/216 0.259
{ {4, 6, 6, 6, 6, 7} , {4, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 54/216 0.250
{ {5, 6, 6, 6, 6, 7} , {4, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 54/216 0.250
{ {6, 6, 6, 6, 6, 7} , {4, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 54/216 0.250
{ {6, 6, 6, 6, 6, 6} , {4, 4, 4, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 54/216 0.250
{ {5, 6, 6, 6, 6, 6} , {3, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 50/216 0.231
{ {6, 6, 6, 6, 7, 7} , {1, 4, 7, 7, 7, 7} , {1, 4, 7, 7, 7, 7} } 50/216 0.231
{ {5, 5, 5, 5, 5, 5} , {3, 3, 4, 6, 6, 6} , {1, 1, 6, 6, 6, 6} } 48/216 0.222
{ {5, 5, 5, 5, 6, 6} , {4, 4, 4, 6, 6, 6} , {1, 1, 6, 6, 6, 6} } 48/216 0.222
{ {5, 5, 5, 5, 5, 5} , {1, 1, 6, 6, 6, 6} , {3, 3, 4, 6, 6, 7} } 48/216 0.222
{ {5, 6, 6, 6, 6, 6} , {1, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {1, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {5, 6, 6, 6, 6, 6} , {2, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {2, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {3, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {5, 6, 6, 6, 6, 6} , {4, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {3, 6, 6, 6, 6, 7} , {4, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {1, 6, 6, 6, 7, 7} , {4, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {1, 4, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {2, 4, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {3, 4, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {4, 4, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 7, 7} , {5, 5, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 48/216 0.222
{ {5, 6, 6, 6, 6, 6} , {4, 4, 4, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {4, 4, 5, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 7, 7} , {5, 5, 5, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 48/216 0.222
{ {6, 6, 6, 6, 6, 6} , {4, 4, 5, 7, 7, 7} , {1, 3, 7, 7, 7, 7} } 48/216 0.222
{ {1, 6, 6, 6, 6, 6} , {4, 4, 4, 5, 7, 7} , {1, 1, 6, 7, 7, 7} } 47/216 0.218
{ {4, 6, 6, 6, 6, 6} , {1, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 46/216 0.213
{ {1, 3, 6, 6, 6, 6} , {1, 3, 6, 6, 6, 6} , {5, 5, 5, 5, 5, 7} } 45/216 0.208
{ {5, 5, 5, 5, 5, 6} , {3, 4, 4, 6, 6, 6} , {1, 1, 6, 6, 6, 6} } 44/216 0.204
{ {5, 5, 5, 5, 6, 6} , {1, 3, 6, 6, 6, 6} , {1, 3, 6, 6, 6, 6} } 44/216 0.204
{ {1, 6, 6, 6, 6, 6} , {4, 4, 4, 4, 7, 7} , {1, 1, 6, 7, 7, 7} } 44/216 0.204
{ {1, 6, 6, 6, 6, 6} , {4, 4, 4, 5, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {5, 6, 6, 6, 6, 7} , {4, 4, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {3, 6, 6, 6, 7, 7} , {4, 4, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 6, 7} , {4, 5, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {5, 6, 6, 6, 7, 7} , {4, 5, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 7, 7} , {4, 5, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 6, 7} , {4, 4, 6, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {5, 6, 6, 6, 7, 7} , {4, 4, 6, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 7, 7} , {4, 5, 6, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 44/216 0.204
{ {5, 6, 6, 6, 6, 7} , {4, 4, 5, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 6, 7} , {4, 4, 5, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 6, 7} , {4, 5, 5, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 6, 7} , {4, 4, 6, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 6, 7} , {4, 5, 5, 7, 7, 7} , {1, 3, 7, 7, 7, 7} } 44/216 0.204
{ {6, 6, 6, 6, 7, 7} , {1, 4, 7, 7, 7, 7} , {2, 4, 7, 7, 7, 7} } 44/216 0.204
{ {4, 5, 5, 5, 5, 5} , {3, 3, 3, 6, 6, 6} , {1, 1, 6, 6, 6, 6} } 42/216 0.194
{ {5, 5, 5, 5, 6, 6} , {4, 4, 4, 6, 6, 6} , {1, 2, 6, 6, 6, 6} } 42/216 0.194
{ {4, 5, 5, 5, 5, 5} , {1, 1, 6, 6, 6, 6} , {3, 3, 3, 6, 6, 7} } 42/216 0.194
{ {1, 6, 6, 6, 7, 7} , {1, 6, 6, 6, 7, 7} , {4, 4, 4, 7, 7, 7} } 42/216 0.194
{ {5, 5, 6, 6, 6, 6} , {1, 4, 4, 7, 7, 7} , {1, 1, 6, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 6, 6} , {1, 4, 4, 7, 7, 7} , {1, 1, 6, 7, 7, 7} } 42/216 0.194
{ {5, 5, 6, 6, 6, 6} , {2, 4, 4, 7, 7, 7} , {1, 1, 6, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 6, 6} , {2, 4, 4, 7, 7, 7} , {1, 1, 6, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 6, 6} , {3, 4, 4, 7, 7, 7} , {1, 1, 6, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 6, 6} , {4, 4, 4, 7, 7, 7} , {1, 1, 6, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 6, 6} , {2, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 6, 6} , {3, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 6, 6} , {4, 4, 4, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 7, 7} , {4, 5, 5, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 42/216 0.194
{ {4, 6, 6, 6, 7, 7} , {4, 4, 6, 7, 7, 7} , {1, 1, 7, 7, 7, 7} } 42/216 0.194
{ {6, 6, 6, 6, 6, 6} , {1, 4, 5, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 42/216 0.194
{ {6, 6, 6, 6, 6, 6} , {2, 4, 5, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 42/216 0.194
{ {6, 6, 6, 6, 6, 6} , {3, 4, 5, 7, 7, 7} , {1, 2, 7, 7, 7, 7} } 42/216 0.194
{ {6, 6, 6, 6, 7, 7} , {5, 5, 5, 7, 7, 7} , {1, 3, 7, 7, 7, 7} } 42/216 0.194
{ {6, 6, 6, 6, 6, 6} , {4, 4, 5, 7, 7, 7} , {1, 4, 7, 7, 7, 7} } 42/216 0.194
{ {6, 6, 6, 7, 7, 7} , {4, 4, 7, 7, 7, 7} , {1, 6, 7, 7, 7, 7} } 42/216 0.194
{ {4, 4, 4, 5, 5, 7} , {1, 1, 6, 6, 6, 7} , {1, 1, 6, 6, 6, 7} } 41/216 0.190
{ {5, 6, 6, 6, 7, 7} , {1, 3, 7, 7, 7, 7} , {1, 3, 7, 7, 7, 7} } 41/216 0.190
{ {5, 6, 6, 6, 7, 7} , {1, 3, 7, 7, 7, 7} , {1, 4, 7, 7, 7, 7} } 41/216 0.190
{ {5, 6, 6, 6, 7, 7} , {2, 3, 7, 7, 7, 7} , {1, 4, 7, 7, 7, 7} } 41/216 0.190

Table 2: The top 74 sets of dice and their paradoxical dominance, Fχ , for n = 3, s = 6, and Pmax = 7. All paradoxical
dominances are integer multiples of 1/s3 = 1/216. The bold face entry at the top is the Lake Wobegon optimal set.

14
Fχ = Sn = vmax = pl = pu =
n (n − 2)/n LCM [1, . . . , n] 1 + n(n − 1)n−2 l sl sl /Sn u su su /Sn
6 6 1 – – –
3 1/3 6 7 4 3 1/2 7 3 1/2
(cube) 1 2 1/3 7 4 2/3
36 12 1 – – –
4 1/2 12 37 33 6 1/2 37 6 1/2
(dodecahedron) 25 4 1/3 37 8 2/3
1 3 1/4 37 9 3/4
320 321 1 – – –
316 30 1/2 321 30 1/2
5 3/5 60 321 301 20 1/3 321 40 2/3
241 15 1/4 321 45 3/4
1 12 1/5 321 48 4/5
3750 60 1 – – –
3745 30 1/2 3751 30 1/2
6 2/3 60 3751 3721 20 1/3 3751 40 2/3
3601 15 1/4 3751 45 3/4
3001 12 1/5 3751 48 4/5
1 10 1/6 3751 50 5/6
54,432 420 1 – – –
54,426 210 1/2 54,433 210 1/2
54,391 140 1/3 54,433 280 2/3
7 5/7 420 54,433 54,181 105 1/4 54,433 315 3/4
52,921 84 1/5 54,433 336 4/5
45,361 70 1/6 54,433 350 5/6
1 60 1/7 54,433 360 6/7
941,192 840 1 – – –
941,185 420 1/2 941,193 420 1/2
941,137 280 1/3 941,193 560 2/3
8 3/4 840 941,193 940,801 210 1/4 941,193 630 3/4
938,449 168 1/5 941,193 672 4/5
921,985 140 1/6 941,193 700 5/6
806,737 120 1/7 941,193 720 6/7
1 105 1/8 941,193 735 7/8
18,874,368 2520 1 – – –
18,874,360 1260 1/2 18,874,369 1260 1/2
18,874,297 840 1/3 18,874,369 1680 2/3
18,873,793 630 1/4 18,874,369 1890 3/4
9 7/9 2520 18,874,369 18,869,761 504 1/5 18,874,369 2016 4/5
18,837,505 420 1/6 18,874,369 2100 5/6
18,579,457 360 1/7 18,874,369 2160 6/7
16,515,073 315 1/8 18,874,369 2205 7/8
1 280 1/9 18,874,369 2240 8/9

Table 3: The minimal Lake Wobegon Optimal sets of dice for 3 ≤ n ≤ 9. The minimum number of sides, according
to our construction, is Sn = LCM [1, . . . , n], and the maximum pip value vmax . For each die, there are at most two
values, the lower and upper, l and u, appearing on the integer number of sides sl and su , and thus appearing with
probabilities pl and pu . The n = 3 and n = 4 sets can be created in Platonic solids, and hence fair physical dice.
Non-Platonic side-fair dice can be constructed for arbitrary n = 2k (k a positive integer) by merging two pyramids
having k isosceles triangular faces and a regular k-gon base, as for instance in the k = 4 octahedron.

15

Você também pode gostar