The American Mathematical Monthly

ISSN: 0002-9890 (Print) 1930-0972 (Online) Journal homepage: https://tandfonline.com/loi/uamm20

Condorcet’s Paradox and Ultrafilters

Timothy Trujillo

To cite this article: Timothy Trujillo (2020) Condorcet’s Paradox and Ultrafilters, The American
Mathematical Monthly, 127:2, 169-174, DOI: 10.1080/00029890.2020.1680200

To link to this article: https://doi.org/10.1080/00029890.2020.1680200

Published online: 06 Jan 2020.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://tandfonline.com/action/journalInformation?journalCode=uamm20
 Condorcet’s Paradox and Ultrafilters
Timothy Trujillo

Abstract. We prove a version of Arrow’s impossibility theorem that at its heart is essentially
Condorcet’s voting paradox.

1. INTRODUCTION. Two of the most iconic results of voting theory are Con-
dorcet’s paradox and Arrow’s impossibility theorem. We give a proof of a linearized
version of Arrow’s impossibility theorem that uses Condorcet’s paradox as the crux
of the argument. Along the way ultrafilters and partition regularity play a prominent
role. We conclude with a proof, using Ramsey theory, that the axiom of determinacy
implies an extension of Arrow’s theorem that applies when there are countably many
voters, finite or infinite. For a more in-depth introduction to ultrafilters, voting theory,
or Ramsey theory, see the references [7], [13], or [14], respectively.
When an infinite set is partitioned into finitely many parts, then at least one of
the parts must also be infinite. Given a set X, a collection of subsets P Ď PpXq is
partition regular if whenever X is partitioned into finitely many parts, one of the
parts is in P . The infinite subsets of an infinite set are the prototype example of a
partition regular collection. An ultrafilter adds the requirement that there is a unique
part in P .
Definition. Given a set X, an ultrafilter1 on X is a collection of subsets U Ď PpXq
such that whenever X is partitioned into finitely many parts, one and only one of the
parts is in U .
The next lemma shows that we can replace the word “finitely” in the definition of
an ultrafilter with “three.” Somewhat surprisingly, “finitely” cannot be replaced with
“two.” Let X be a set with an odd number of elements and M be the collection of sub-
sets of X that contain more than half of the elements of X. Whenever X is partitioned
into two parts, one and only one of the parts is in M. However, there are partitions of
X into finitely many parts such that none of the parts are in M.
A similar characterization of ultrafilter was used to prove the Gibbard–Satterthwaite
theorem [12], the second most important result in social choice theory after Arrow’s
impossibility theorem. The significance of the next lemma for social choice theory is
evidenced by the fact that the two characterizations, obtained independently, have been
used to prove the two most important results of social choice theory.
Lemma 1. U Ď PpXq is an ultrafilter on X if whenever X is partitioned into three
parts, one and only one of the parts is in U .
Proof. No two elements of U are disjoint. If two elements were disjoint, then we could
partition X into three parts with more than one part in U , namely, the two disjoint sets
and the complement of their union.
Assume U is not an ultrafilter and choose a partition X0 , . . . , Xn´1 witnessing this
failure with n as small as possible. The minimality of n shows that the union of any
doi.org/10.1080/00029890.2020.1680200
MSC: Primary 03E05
1 This is not the standard definition of an ultrafilter. In [3], Galvin and Horn show that the definition provided

here is equivalent to the standard definition of an ultrafilter.

February 2020] NOTES 169

two of these sets is in U . If n were larger than three, then lumping the first two together
and the last two together would yield a partition of n ´ 2 sets with two in U .

For each x P X, the set of all subsets of X that contain x is an ultrafilter—if we

partition X into finitely many parts then one and only one of the parts contains x.
These ultrafilters are known as the principal ultrafilters.

2. MATHEMATICAL VOTING THEORY. For a finite set A of alternatives, an A-

ballot is a relation on A that is complete and transitive. Intuitively, it is a list of the
alternatives from best to worst with possible ties. A linear A-ballot is an A-ballot that
is also antisymmetric, i.e., a list without any ties.
Definition. A collection of linear A-ballots indexed by the set of all voters V is called
a linear A-election. A linear social welfare function2 is a procedure that returns a
societal ordering of A that is complete, transitive, antisymmetric, and can accommo-
date elections with at least three alternatives, i.e., |A| ě 3.
If alternative a is ranked higher than alternative b in the societal ranking, then we
say that society prefers a to b. One desirable property of a linear social welfare func-
tion is that the procedure treats all alternatives the same.
Definition. A linear social welfare function is neutral if for all alternatives a, b P A,
the following conditional statement holds: If society prefers a to b and alternative a is
interchanged with alternative b on all ballots, then society prefers b to a.
In this note, we are only concerned with linear social welfare functions where soci-
ety’s preference between two alternatives depends only on the voters’ relative prefer-
ences for the two alternatives and nothing more.
Definition. A linear social welfare function is independent3 if for all alternatives
a, b P A, the following conditional statement holds: If society prefers a to b and a
third alternative is altered on some ballots (without altering the relative order of the
alternatives a and b), then society prefers a to b.
A dictatorship is a linear social welfare function where the societal ordering is
always obtained from a single voter’s ballot. The voter that determines the outcome of
the election is called the dictator. If a dictator were to prefer a to b and all other voters
rank b above a, then society would still prefer a to b. The next definition extends this
idea to subsets of voters.
Definition. A subset of the voters is called a decisive coalition if for all alternatives
a, b P A, the following conditional statement holds: If the voters in the coalition are
the only voters that rank a above b, then society prefers a to b. A linear social welfare
function is Pareto if the set of all voters is a decisive coalition.
Lemma 2. An independent linear social welfare function is Pareto only if the welfare
function is neutral.

Proof. Suppose that a, b P A and consider an election where society prefers a to b. Let
c be a third alternative different from a and b. Alter each ballot by placing c directly
2 We use “linear” to mean “without ties.” To simplify our analysis we only consider the context of linear

ballots and linear societal orderings. The more general notion of a social welfare function allows for ties in
both the ballots and the societal ordering. The definitions of “independent” and “Pareto” given here can be
extended to social welfare functions by simply deleting the adjective “linear” from the definitions.
3 This property is commonly referred to as “independence of irrelevant alternatives” in voting theory.

170 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 127
above alternative a on all ballots. Since our function is Pareto and independent, society
prefers c to a and a to b, respectively. In particular, society prefers c to b. By similar
arguments we can swap b with a and then swap c with b obtaining an election where
society prefers b to a. Now place c back into its original position. Since the function is
independent, society still prefers b to a. The net effect of the swaps is an interchange
of a and b on all ballots. Since society also interchanges its preference for a and b, the
social welfare function is neutral.

Lemma 3. An independent linear social welfare function is Pareto only if the following
two statements are true:
1. No two decisive coalitions are disjoint.
2. Every subset of the voters or its complement is a decisive coalition.

Proof. 1. Suppose that V0 Ď V is a decisive coalition and V1 Ď V is disjoint from V0 .

Let V2 “ V zpV0 Y V1 q. Consider an election where the only voters that rank a higher
than b are those in V1 . Let c be a third alternative not equal to a or b. Alter each ballot
in V0 Y V1 by placing c directly below b and then alter those in V2 by placing c directly
below a. Since V0 is a decisive coalition, society prefers c to a. Since the function is
Pareto, society prefers b to c. Thus society prefers b to a. Since the only voters that
prefer a to b are those in V1 , we find that V1 cannot be a decisive coalition.
2. Toward a contradiction suppose that neither V0 Ď V nor V1 “ V zV0 are decisive
coalitions. For i “ 0, 1, since Vi is not a decisive coalition, there are two alternatives
ai , bi P A and an election where the voters in Vi are the only voters that rank ai higher
than bi but society prefers bi to ai . By the previous lemma, without loss of general-
ity, we can assume a0 “ a1 and b0 “ b1 . Since our function is independent, we can,
without changing society’s preference for bi over ai in either election, move all other
alternatives to the bottom of all ballots in alphabetical order. However, this is a contra-
diction by the previous lemma.

According to Taylor [13], the next result with the word “linear” removed “should be
properly attributed to Hansson (1976) and Kirman and Sondermann in (1972) although
their work was proceeded by earlier observations of Blau, unpublished, Fishburn
(1970) and Guilbaud (1952)” (see [5], [6], [2], and [4], respectively). The term ultra-
filter does not explicitly appear in Guilbaud’s paper; however, the analogy between
Arrow’s theorem and Condorcet’s paradox does appear in [4]. Arrow was made aware
of the connection by Guilbaud, who coined the expression of an “effet Condorcet” for
it (see [10]).
Lemma 4. A linear social welfare function is independent and Pareto if and only if its
collection of decisive coalitions is an ultrafilter on the set of voters.

Proof. (ñ) Toward a contradiction assume the collection of decisive coalitions is not
an ultrafilter. By Lemma 1 and Lemma 3, there is a partition of the voters V0 , V1 ,
V2 into three parts such that their complements V1 Y V2 , V0 Y V2 , and V0 Y V1 are
all decisive coalitions. Consider a linear A-election where a, b and c are ranked as
follows:
a b c
b c a
c a b
.. .. ..
. . .
V0 V1 V2

February 2020] NOTES 171

Note that voters in V1 Y V2 prefer c to a, voters in V0 Y V2 prefer a to b, and voters in
V0 Y V1 prefer b to c. Since each of these sets is a decisive coalition, society prefers c
to a, a to b, and b to c. However, this is inconsistent with a linear societal ordering of
A. This contradiction is known as Condorcet’s paradox.
(ðq Let a, b P A be any two alternatives. Consider an election where society
prefers a to b. Let V0 be those voters that rank a above b. Since V0 and V zV0 form a
partition of V into two parts, V0 must be a decisive coalition. Now alter a third alter-
native different from a and b on some ballots. Since the only voters that rank a above
b in the new election are still those in the decisive coalition V0 , society still prefers a
to b. Therefore the linear social welfare function is independent.
Consider the partition of V into three parts V0 “ V , V1 “ H, and V2 “ H. Since
V1 “ V2 , we have that V0 “ V is a decisive coalition. That is, the linear social welfare
function is Pareto.

The next theorem, among other things, shows that not all voters are treated equally
when a linear social welfare function is Pareto and independent. The result with the
word “linear” removed is the celebrated Arrow’s impossibility theorem [1].

Theorem 1. Suppose that there is a linear social welfare function with a finite deci-
sive coalition.4 The welfare function is independent and Pareto if and only if it is a
dictatorship.

Proof. (ñ) Partition the finite decisive coalition into singleton sets, i.e., sets with one
element. These singleton sets and the set of all voters not in the finite coalition form a
partition of the voters into finitely many parts. By Lemma 4, one of the singleton sets
is a decisive coalition.
Now consider an election where the voter in the singleton decisive coalition ranks
a above b. Let V0 Ď V denote those voters that agree and also rank a above b. Since
the complement of V0 is disjoint from the singleton, by Lemma 3, V0 is a decisive
coalition. Therefore society prefers a to b. Thus the voter in the singleton is a dictator.
(ð) Clearly every dictatorship is both independent and Pareto.

3. HIGHER-DIMENSIONAL RAMSEY THEORY. Litak [8] has shown, with a

strategy-stealing argument, that the axiom of determinacy5 implies that an independent
social welfare function with countably many voters is Pareto if and only if it is a
dictatorship. At the end of this section, we derive a linearized version of the same
result using some Ramsey theory.
For a set X and a natural number n P N, we use the following notation: rXsn is the
set of all n-element subsets of X, and rXs8 is the set of all infinite subsets of X. The
n-dimensional Ramsey theorem states that for the natural number n, the collection of
subsets of rNsn that contain rXsn for some infinite subset X of N is partition regular.
The axiom of choice implies that the infinite-dimensional analogue of the Ramsey
theorem is false. In other words, the collection of subsets of rNs8 that contain rXs8
for some infinite subset X of N is not partition regular.

Definition. A partition of rNs8 into finitely many parts X0 , X1 , . . . , Xi is Ramsey if

there exists X P rNs8 and j ď i such that rXs8 Ď Xj .
4 Since the linear welfare function is Pareto this includes the case where the set of voters is finite.
5 The axiom of determinacy states that for any set A of infinite sequences of natural numbers,
Dx0 .@x1 .Dx2 . . . pxi q P A Ø @x0 .Dx1 .@x2 . . . pxi q R A. The axiom of determinacy is inconsistent with the
axiom of choice.

172 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 127
 A partition that is not Ramsey is a counterexample to the infinite-dimensional ana-
logue of the Ramsey theorem. The next theorem is a translation of a result of Baum-
gartner concerning ultrafilters (in a correspondence with Mathias [9]) to the setting of
linear social welfare functions. Baumgartner showed that the existence of a nonprinci-
pal ultrafilter on N implies the existence of a partition of rNs8 that is not Ramsey.
Theorem 2 (Baumgartner [9]). Assume there are countably many voters. An inde-
pendent linear social welfare function that is not a dictatorship is Pareto only if there
exist a partition of rNs8 that is not Ramsey.

Proof. By Theorem 1 we can assume that no finite set of voters forms a decisive coali-
tion. In particular, there are infinitely many voters.
For each X P rNs8 , let EX denote the linear A-election where voter n prefers o to e
if and only if X X t0, 1, . . . , nu has an odd number of elements (all other alternatives
are ranked below e and o in alphabetical order). Let Xe denote those X where society
prefers e to o. Let Xo denote those X where society prefers o to e. Note that Xe and
Xo together form a partition of rNs8 .
For a given X P rNs8 , let V0 “ tn P N : n ă minpXqu. Note that V0 is the set of
voters whose preferences do not change between EX and EXztminpXqu . By Lemma 4,
since V0 is finite, society’s preference for o and e changes between EX and EXztminpXqu .
Notice that XztminpXqu P rXs8 . Therefore, for all X P rNs8 , rXs8 Ę Xe and
rXs8 Ę Xo . That is, the partition is not Ramsey.
The axiom of determinacy implies that all finite partitions of rNs8 are Ramsey, a
result of Pirkry [11]. Thus, by the previous theorem, the axiom of determinacy implies
an independent linear social welfare function with countably many voters is Pareto
only if it is a dictatorship. If not, we could use the axiom of determinacy to construct
a non-Ramsey set, a contradiction. For the converse, clearly every dictatorship is both
independent and Pareto.

ACKNOWLEDGMENT. The author was partially supported by an internal grant from Sam Houston State
University OSRP.

REFERENCES

[1] Arrow, K. (1963). Social Choice and Individual Values, 2nd ed. New Haven, CT: Yale Univ. Press.
[2] Fishburn, P. (1970). Arrow’s impossibility theorem: Concise proof and infinite voters. J. Econ. Theory.
2(1): 103–106.
[3] Galvin, F., Horn, A. (1970). Operations preserving all equivalence relations. Proc. Amer. Math. Soc.
24(3): 521–523.
[4] Guilbaud, G. (1952). Les théories de l’intérêt général et le problème logique de l’agrégation. Econom.
Appl. 5: 501–584.
[5] Hansson, B. (1969). Group preferences. Econometrica. 37(1): 50–54.
[6] Kirman, A., Sondermann, D. (1972). Arrow’s theorem, many agents, and invisible dictators. J. Econ.
Theory. 5(2): 267–277.
[7] Komjath, P., Totik, V. (2008). Ultrafilters. Amer. Math. Monthly. 115(1): 33–44.
[8] Litak, T. (2018). Infinite populations, choice and determinacy. Studia Logica. 106(5): 969–999.
[9] Mathias, A. (1975). A remark on rare filters. In: Hajnal, A., Rado, R., Sós, V. T., eds. Infinite and
Finite Sets (Colloq., Keszthely, 1973), Vol. III. Colloq. Math. Soc. János Bolyai, Vol. 10. Amsterdam:
North-Holland, pp. 1095–1097.
[10] Monjardet, B. (2005). Social choice theory and the “Centre de Mathématique Sociale”: Some historical
notes. Soc. Choice Welf. 25(2/3): 433–456.
[11] Prikry, K. (1976). Determinateness and partitions. Proc. Amer. Math. Soc. 54(1): 303–306.
[12] Rao, S., Basile, A., Rao, K. P. S. B. (2018). On the ultrafilter representation of coalitionally strategy-
proof social choice functions. Econ. Theory Bull. 6(1): 1–13.

February 2020] NOTES 173

[13] Taylor, A. (2005). Social Choice and the Mathematics of Manipulation. New York, NY: Cambridge
Univ. Press.
[14] Todorcevic, S. (2010). Introduction to Ramsey Spaces. Annals of Mathematics Studies, Vol. 174. Prince-
ton, NJ: Princeton Univ. Press.

Sam Houston State University, Huntsville, TX 77340

[email protected]

Special Properties of Power Functions

We give a proof that power functions uniquely exhibit two interesting properties. Let
f be an increasing continuous function on r0, 8q with f p0q “ 0. Let b ě 0 be a
constant. Then the following are equivalent:
(a) For every t ą 0, f ptq is pb ` 1q times the average value of f over r0, ts.
Ś
(b) For every t ą 0, the area of the portion of the rectangle r0, ts r0, f ptqs above
the graph of f is b times the area of the portion below.
(c) f pxq “ ax b , for some constant a.
That (c) ùñ (a) follows from computation with the power rule:
ż
b`1 t b
ax dx “ at b .
t 0

Furthermore, (a) ùñ (b) since

żt
b`1
f ptq “ f pxq dx (1)
t 0

can be rearranged as
żt żt
tf ptq ´ f pxq dx “ b f pxq dx, (2)
0 0
Ś
where the left-hand side is the area of the portion of r0, ts r0, f ptqs above the
graph of f and the right-hand side is b times the area of the portion below. To close
the loop, we show (b) ùñ (c). Obtaining (1) from (2), we see that (b) implies f is
differentiable at every t ą 0. Differentiating through (2) with respect to t yields the
differential equation tf 1 ptq “ bf ptq. The general solution on p0, 8q is f ptq “ at b ,
where a is an arbitrary constant, so (c) follows.
—Submitted by Seth Borgo, Seattle, Washington
doi.org/10.1080/00029890.2020.1680203
MSC: Primary 26A09

174 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 127