Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

A Unifying Framework for Manipulation Problems

Extended Abstract

Dušan Knop∗ Martin Koutecký† Matthias Mnich‡

Dept. of Informatics, Univ. of Bergen Technion - Israel Institute of Universität Bonn
Charles University, Prague Technology Maastricht University
[email protected] [email protected] [email protected]

ABSTRACT Conference on Autonomous Agents and Multiagent Systems (AAMAS 2018),

Manipulation models for electoral systems are a core research theme Stockholm, Sweden, July 10–15, 2018, IFAAMAS, 9 pages.
in social choice theory; they include bribery (unweighted, weighted,
swap, shift, . . . ), control (by adding or deleting voters or candidates), 1 INTRODUCTION
lobbying in referenda and others. Problems of manipulation, bribery and control constitute a funda-
We develop a unifying framework for manipulation models with mental part of computational social choice. Many such problems
few types of people, one of the most commonly studied scenarios. are known to be NP-hard (or worse). However, their input can nat-
A critical insight of our framework is to separate the descriptive urally be partitioned into several parts, like the number of voters,
complexity of the voting rule R from the number of types of peo- the number of candidates, and others. This motivates the study of
ple. This allows us to finally settle the computational complexity such problems by the powerful tools of parameterized complexity.
of R-Swap Bribery, one of the most fundamental manipulation A by-now classical example in this direction is the R-Swap
problems. In particular, we prove that R-Swap Bribery is fixed- Bribery problem, which takes as input an election consisting of
parameter tractable when R is Dodgson’s rule and Young’s rule, a set C of candidates and a set V of voters with their individual
when parameterized by the number of candidates. This way, we re- preference lists ≻v (for v ∈ V ), which are total orders over C.
solve a long-standing open question from 2007 which was explicitly Additionally, for each voter v ∈ V and each pair of consecutive
asked by Faliszewski et al. [JAIR 40, 2011]. candidates c ≻v c ′ , there is some cost σ v (c, c ′ ) ∈ Z of swapping
Our algorithms reveal that the true hardness of bribery problems the order of c and c ′ in ≻v . The objective is to find a minimum-cost
often stems from the complexity of the voting rules. On one hand, set of swaps of consecutive candidates in the preference lists in
we give a fixed-parameter algorithm parameterized by number order to make a designated candidate c ⋆ ∈ C the winner of the
of types of people for complex voting rules. Thus, we reveal that thus-perturbed election under a fixed voting rule R.
R-Swap Bribery with Dodgson’s rule is much harder than with This problem was introduced by Elkind et al. [8] and has since
Condorcet’s rule, which can be expressed by a conjunction of linear been studied for many classical voting rules R [6, 11, 18, 23]. In
inequalities, while Dodson’s rule requires quantifier alternation particular, its computational complexity has been thoroughly ana-
and a bounded number of disjunctions of linear systems. On the lyzed with respect to the number of candidates |C |. The observation
other hand, we give an algorithm for quantifier-free voting rules that |C | is often small motivated the search for fixed-parameter al-
which is parameterized only by the number of conjunctions of the gorithms for R-Swap Bribery parameterized by |C |, which are
voting rule and runs in time polynomial in the number of types algorithms that run in time f (|C |) · nO (1) for some computable
of people. This way, our framework explains why Shift Bribery function f , here n denotes the size of the input election; if such an
is polynomial-time solvable for the plurality voting rule, making algorithm exists, we then say that the problem is fixed-parameter
explicit that the rule is simple in that it can be expressed with tractable with respect to the parameter |C |.
a single linear inequality, and that the number of voter types is Despite the problem’s importance, for a long time, only the “uni-
polynomial. form cost” case of R-Swap Bribery was known to be fixed-parame-
ter tractable for various voting rules, parameterized by the number
KEYWORDS of candidates; here, uniform cost refers to the special case that all
Swap bribery; Dodgson’s rule; Young’s rule; fixed-parameter algo- voters have the same cost function, that is, σ v ≡ σ for all v ∈ V .
rithms This is a fundamental result due to Dorn and Schlotter [6], who
ACM Reference Format:
showed that R-Swap Bribery with uniform cost can be solved in
|C |O (1)
Dušan Knop, Martin Koutecký, and Matthias Mnich. 2018. A Unifying time 22 · nO (1) for all voting rules R that are “linearly describ-
Framework for Manipulation Problems. In Proc. of the 17th International able”. Many classical voting rules are indeed linearly describable,
like any scoring protocol, Copelandα , Maximin, or Bucklin.
∗ Authorsupported by grant projects P202/12/G061 of GA ČR and NFR MULTIVAL.
† Author Recently, Knop et al. [18] gave the first fixed-parameter algo-
supported by Technion postdoctoral fellowship.
‡ Author supported by ERC Starting Grant 305465 (BeyondWorstCase). rithms for R-Swap Bribery for general cost functions for most vot-
ing rules R studied in the literature (scoring protocol, Copelandα ,
Proc. of the 17th International Conference on Autonomous Agents and Multiagent Systems Maximin, Bucklin etc.), thereby removing the uniform cost assump-
(AAMAS 2018), M. Dastani, G. Sukthankar, E. André, S. Koenig (eds.), July 10–15, 2018, tion. This way, they resolved a long-standing open problem. More-
Stockholm, Sweden. © 2018 International Foundation for Autonomous Agents and O (1)
Multiagent Systems (www.ifaamas.org). All rights reserved. over, their algorithm runs in time 2 |C | · nO (1) for many rules R,

256
Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

and thus improves the double-exponential run time by Dorn and Thus, it is not clear how to solve R-Swap Bribery even for uniform
Schlotter. Their key idea was to reduce the problem to so-called cost with any fixed-parameter algorithm for R being the Dodg-
n-fold integer programming, which allowed them to solve the prob- son rule or the Young rule. These complications led Faliszewski
lem efficiently for bounded number of candidates despite their et al. [10] to explicitly ask for the complexity of R-Swap Bribery
integer program having an unbounded number of variables. Their parameterized by the number of candidates under these rules.
approach also solved R-Swap Bribery for R being the Kemeny
rule, even for general cost functions, though the Kemeny rule is 1.2 Our Contributions
not known to be linearly describable (cf. [10, p. 338]). However, this We start by making a key observation about the majority of fixed-
does not apply for Dodgson’s and Young’s rules. parameter algorithms for R-Swap Bribery when |C | is small. A
typical such result is an algorithm for R-Swap Bribery for R being
1.1 The Challenge Condorcet’s voting rule. That algorithm uses two key ingredients:
Even so, there are some notable voting rules R for which the com- (1) There are at most |C |! preference orders of C, and hence each
plexity of R-Swap Bribery remained open even in the uniform cost voter falls into one of |C |! types; thus, an input election is ex-
case. This includes the Dodgson rule and the Young rule. Those pressible as a society s = (s 1 , . . . , s |C |! ), where si is the number
rules are based on the notion of Condorcet winner, which is a can- of voters of type i.
didate who beats any other candidate in a head-to-head contest. (2) Expressing that a candidate c ⋆ is a Condorcet winner is possible
The Condorcet voting rule is very natural and dates back to the using a conjunction of |C | − 1 linear inequalities in terms of s.
18th century; however, clearly there exist elections without a Con- As those key properties hold almost universally for voting rules R,
dorcet winner. In such a situation one proclaims those candidates one might be tempted to think that if there are many types of voters,
as winners who are “closest” to being a Condorcet winner; different the R-Swap Bribery problem must be hard, and if there are few
notions of closeness then yield different voting rules: types of voters, the problem must be easy. However, two points arise
• Closeness measured as the of number of swaps in voter’s prefer- as counter-evidence. First, very recently, Knop et al. [18] showed
ence orders defines the Dodgson rule. that even if there are many types of voters who differ by their cost
• Closeness measured as the number of voter deletions defines the functions, the R-Swap Bribery problem remains fixed-parameter
Young rule. tractable for a wide variety of voting rules R. Second, as already
Thus, a candidate c is a Dodgson winner if s/he can be made a mentioned, it was open since 2007 whether R-Swap Bribery with
Condorcet winner by a minimum number of swaps in the voter’s Dodgson’s and Young’s voting rule are fixed-parameter tractable
preference orders over all candidates; analogously for the Young for few candidates, even for uniform cost functions.
rule and voter deletions. From voters and candidates to societies. Here, we take a novel
When considering R-Swap Bribery, the Dodgson rule and the perspective. We observe that the two key ingredients (1) and (2)
Young rule are much more complicated to handle than other rules; apply much more widely than for R-Swap Bribery; namely, they
the reasons are several. First, for many voting rules R, the winner of are also present in many other manipulation, bribery and control
an election can be found in polynomial time, and solving this winner problems. We therefore abstract away the specifics of such problems
determination problem is certainly a necessary subtask when solv- and introduce general notions of “society”, “moves in societies”, and
ing R-Swap Bribery. However, for R ∈ {Dodgson, Young}, already “winning conditions”. Let τ ∈ N be the number of types of people
winner determination is NP-hard, and so even verifying that c ⋆ is (e.g., voters in an election or a referendum). A society s is simply a
indeed the winner of the perturbed election is intractable. In fact, non-negative τ -dimensional integer vector encoding the numbers
winner determination for these voting rules is complete for parallel of people of each type. A move m is a τ 2 -dimensional integer vector
access to NP [13, 14, 22], denoted PNP||
-complete1 . Second, for more whose elements sum up to zero; it encodes how many people move
than 25 years the winner determination problem for the Dodgson from one type to another. A change ∆ is a τ -dimensional vector
rule and Young rule was only known to be solvable by an ILP- (typically associated with a move) encoding the effect of a move
based algorithm [1] with doubly-exponential dependence in |C |; on a society, such that s + ∆ is again a society. Finally, a winning
a single-exponential algorithm is only known since recently [18]. condition Ψ(s) is a predicate encoding some desirable property of
This provides a sharp contrast to the Kemeny rule, for which simply a society, such as that a preferred candidate has won or that a
enumerating all |C |! possible preference orders suffices to determine preferred agenda was selected in a referendum. Specifically, we
the winner. Faliszewski et al. [9] describe these difficulties: study winning conditions which are describable by formulas in
It is interesting to consider which features of Kemeny elections Presburger Arithmetic (PA). PA is a logical language whose atomic
allow us to employ the above [ILP-based] attack, given that the formulas are linear inequalities over the integers, which are then
same approach does not seem to work for either Dodgson or Young joined with logical connectives and quantifiers. Thus, winning
elections. One of the reasons is that the universal quantification conditions describable by PA formulas widely generalize the class
implicit in Dodgson and Young elections is over an exponentially of linearly describable voting rules by Dorn and Schlotter [6].
large search space, but the quantification in Kemeny is, in the Our main technical contribution informally reads as follows:
case of a fixed candidate set, over a fixed number of options. Theorem 1.1 (informal). Deciding satisfiability of PA formu-
1 Theclass PNP contains all problems solvable in polynomial time by a deterministic
las with two quantifiers is fixed-parameter tractable with respect to
||
Turing machine which has access to an NP oracle, but must ask all of its oracle queries the dimension and length of formula, provided its coefficients and
at once (i.e., the queries can not depend on each other). constants are given in unary.

257
Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

The importance of Theorem 1.1 arises from its applicability to active/latent bit; thus the number of potential types is 2|V |. Simi-
the following general manipulation problem that we introduce here. lar arguments also work for Support Bribery where we change
This general manipulation problem, which we call Minimum Move, voters’ approval counts, and with a more intricate argumentation
captures that many manipulation problems can be cast as finding a also for some voting rules and Bribery and $Bribery. In this sense,
minimum move with respect to some objective function; in particu- the fact that we need to consider |C |! potential voter types in R-
lar, it encompasses the well-studied R-Swap Bribery problem. We Swap Bribery almost seems like an anomaly, rather than a rule. In
study Minimum Move for linear objective functions and winning summary, the complexity of Minimum Move depends primarily on
conditions Ψ expressible with PA formulas of the form “∃∀”. For the descriptive complexity of the winning condition Ψ, because in
all such Ψ, with the help of Theorem 1.1, we show that Minimum many cases the number of types of people is polynomially bounded.
Move is fixed-parameter tractable for combined parameter the de- Another consequence of Theorem 1.1 are the first fixed-
scriptive complexity (length) of the winning condition Ψ and the parameter algorithms for two important manipulation problems
number τ of “types of people”, that is, it is fixed-parameter tractable beyond R-Swap Bribery. The Resilient Budget problem asks, for
for parameter lengthe of Ψ plus τ . As an important special case, a given society whether allocating budget B is sufficient in order to
we obtain the first fixed-parameter algorithm for R-Swap Bribery repel any adversary move of cost at most Ba with a counter-move
for R the Dodgson rule and the Young Rule with uniform costs. of cost at most B (so that the winning condition is still satisfied).
To this end, we model the winning condition of the Dodgson rule Similarly, Robust Move asks for a move of cost at most B which
and Young rules as a PA formula. For intuition, consider the Young causes the winning condition to be satisfied even after any adver-
rule: a candidate c ⋆ is a Young winner (with score d) if there ex- sary move of cost at most Ba . For formal definitions and results, cf.
ists a set V ⋆ ⊆ V of at most d voters such that c ⋆ is a Condorcet Sect. 4.2.
winner of the election (C, V \ V ⋆ ), and for all sets V ′ ⊆ V of at
most d − 1 voters any other candidate c , c ⋆ is not a Condorcet 1.3 Interpretation of Results
winner of the election (C, V \ V ′ ). This formula has one quantifier
Intuitively, the results obtained with Theorem 1.1 can be interpreted
alternation, and its length (for a fixed score d) is bounded by some
as follows. Dodgson-Swap Bribery is fixed-parameter tractable pa-
function of |C |; finally we have to take a disjunction of such formu-
rameterized by |C |; however, this comes with at least two limitations
las over all possible scores d. For a candidate set C, the number τ of
as compared to prior work for simpler voting rules R. First, our
types of people is bounded by |C |!. Consequently, we finally settle
methods do not extend beyond the uniform cost scenario, and this
the long-standing open question about the complexity of R-Swap
remains a major open problem. Second, our result requires the
Bribery for R ∈ {Dodgson, Young}, that was explicitly raised by
input election to be given in unary, while prior work allows it to
Faliszewski [10]:
be given in binary (this is sometimes called the succinct case [9]).
Theorem 1.2. R-Swap Bribery with uniform cost is fixed-pa- This is easily explained by the different descriptive complexities of
rameter tractable parameterized by the number of candidates for R the respective voting rules: for example, while Condorcet’s voting
being the Dodgson rule or the Young rule; it can be solved in time rule can be formulated as a quantifier-free PA formula, formulating
f (|C |) · |V |O (1) for some computable function f . Dodgson’s rule requires a long disjunction of formulas which use
two quantifiers and a bounded number of disjunctions.
Beyond this fundamental problem, we show that a host of other Theorem 1.4 lets us discuss more specifically the complexity of
well-studied manipulation problems are captured by our fixed- various voting rules. For example, the Plurality voting rule can be
parameter algorithm for Minimum Move: expressed with a single linear inequality encoding that a preferred
candidate obtained more points than the remaining candidates alto-
Corollary 1.3. For R ∈ {Dodgson, Young}, the following prob-
gether. Thus, all problems which can be modeled as Minimum Move
lems are fixed-parameter tractable for uniform costs when parameter-
are polynomial-time solvable with the Plurality voting rule. This
ized by the number |C | of candidates: R-$Bribery, R-CCDV/CCAV,
interprets the result of Elkind et al. [8, Theorem 4.1] that Plurality-
R-Possible Winner, and R-Extension Bribery.
Shift Bribery is polynomial-time solvable: the number of potential
Let us turn our attention to the parameter “number of types of voter types is polynomial, and Plurality has a simple description.
people” τ . Our main contribution here is the following: Continuing, we may compare R=Borda with R=Copeland. The
winning condition for R=Borda can be described with |C | − 1 in-
Theorem 1.4 (informal). For any quantifier-free winning con- equalities, while R=Copeland requires O(|C | 2 ) inequalities. Thus
dition Ψ, Minimum Move can be solved in time polynomial in the 2
Borda-Swap Bribery is solvable in time |C |O ( |C | ) log |V |, while
number of types and exponential only in the number of linear in- 4
Copeland-Swap Bribery requires time |C |O ( |C | ) log |V |. Finally,
equalities of Ψ.
all descriptions of Kemeny’s voting rule we are aware of require
Note that in many models of bribery and control, the number |C |! inequalities, and thus result in Kemeny-Swap Bribery being
O (1)
of potential types of people (i.e., types that can occur in any feasi- solvable in time |C |!( |C |!) log |V |. We do not claim these com-
ble solution) is polynomial in the number of people on input. For plexities to be best possible, but conjecture the existence of lower
example, in Shift Bribery, every voter can be bribed to change bounds separating the various voting rules; in particular, we believe
their preferences order to one of |C | − 1 orders; thus the number that Kemeny-Swap Bribery requires double-exponential time.
of potential types is (|C | − 1)|V |. Similarly, in CCAV / CCDV (con- Finally, our work provides a natural next step in unifying the
structive control by adding or deleting voters), every voter has an many different models that have been proposed for voting, bribing

258
Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

and manipulation problems. In this direction, Faliszewski et al. [10] Hence, we believe that ours are the first fixed-parameter algorithms
study what happens when multiple bribery and manipulation ac- for any Bribery-like problem for R ∈ {Dodgson, Young}.
tions can occur in an election; e.g., CCAV asks for constructive
control by adding voters while CCDV by deleting voters; similarly 2 PRELIMINARIES
for CCAC and CCDC for adding/deleting candidates. Faliszewski et Let m, n be integers. We define [m, n] := {m, m + 1, . . . , n} and
al. unify those various (up to that point separately studied) attacks. [n] := [1, n]. Throughout, we reserve bold face letters (e.g. x, y)
Similarly, Knop et al. [18] formulate the R-Multi Bribery problem, for vectors. For a vector x its i-th coordinate is x i .
which also incorporates swaps and perturbing approval counts. The Next, we provide notions and notations for R-Swap Bribery.
problem we put forward in this paper, Minimum Move, in some
sense generalizes and simplifies all those “meta”-problems. Elections. An election (C, V ) consists of a set C of candidates and
a set V of voters, who indicate their preferences over the candidates
1.4 Related Work in C, represented via a preference order ≻v which is a total order
over C. We often identify a voter v with their preference order ≻v .
We have reviewed most of the relevant computational social choice
Denote by rank(c, v) the rank of candidate c in ≻v ; v’s most pre-
work already. However, there seems to be some confusion in the
ferred candidate has rank 1 and their least preferred candidate has
literature that deserves clarification. The paper of Faliszewski et
rank |C |. For distinct candidates c, c ′ ∈ C, write c ≻v c ′ if voter v
al. [9] pioneering the concept of bribery in elections indeed consid-
prefers c over c ′ .
ers the voting rules Kemeny, Dodgson and Young, and provides a
fixed-parameter algorithm for Kemeny-Bribery. There are three Swaps. Let (C, V ) be an election and let ≻v ∈ V be a voter. For
features of their paper that we wish to discuss. candidates c, c ′ ∈ C, a swap s = (c, c ′ )v means to exchange the
First, turning their attention to Dodgson-Bribery, they write: positions of c and c ′ in ≻v ; denote the perturbed order by ≻vs . A
swap (c, c ′ )v is admissible in ≻v if rank(c, v) = rank(c ′, v) − 1. A
Applying the integer programming attack for the case of bribery
set S of swaps is admissible in ≻v if they can be applied sequentially
within Dodgson-like election systems [...] is more complicated.
in ≻v , one after the other, in some order, such that each one of
These systems involve a more intricate interaction between bribing
them is admissible. Note that the perturbed vote, denoted by ≻vS , is
the voters and then changing their preferences. For Dodgson elec-
independent from the order in which the swaps of S are applied. We
tions, after the bribery, we still need to worry about the adjacent
extend this notation for applying swaps in several votes and denote
switches within votersâĂŹ preference lists that make a particular
it V S . We specify v’s cost of swaps by a function σ v : C × C → Z.
candidate a Condorcet winner. [...] This interaction seems to be
too complicated to be captured by an integer linear program, but Voting rules. A voting rule R is a function that maps an election
building on the flavor of the Bartholdi et al. [1] ILP attack we (C, V ) to a subset W ⊆ C, called the winners. A candidate c ∈ C
can achieve the following: Instead of making p a winner, we can is a Condorcet winner if any other c ′ ∈ C \ {c} satisfies {≻v ∈ V |
attempt to make p have at most a given Dodgson or Young score. c ≻v c ′ } > {≻v ∈ V | c ′ ≻v c} ; then we say that c beats c ′
They call this problem DodgsonScore-Bribery and provide positive in a head-to-head contest. The Young score of c ∈ C is the size of
results for it. Notice, however, that finding a bribery which makes c ⋆ the smallest subset V ′ ⊆ V such that c is a Condorcet winner in
have a certain Dodgson score does not prevent another candidate to (C, V \ V ′ ). Analogously, the Dodgson score of c ∈ C is the size of
have a lower score and winning the bribed election. Thus, solving the smallest admissible set of swaps S such that c is a Condorcet
DodgsonScore-Bribery can be very far from the desired result. winner in (C, V S ). Then, c is a Young (Dodgson) winner if it has
Second, the authors then observe that a brute force approach minimum Young (Dodgson) score.
enumerating all |V | |C |! briberies solves the Dodgson-Bribery prob- We aim to solve the following problem:
lem in polynomial time for constantly many candidates; however, R-Swap Bribery
theirs is not a fixed-parameter algorithm for parameter |C |. Input: An election (C, V ), a designated candidate c ⋆ ∈ C
Third, they then introduce another voting system called and swap costs σ v : C × C → Z for v ∈ V .
Dodgson ′ , which is similar to Dodgson, and provide a fixed- Find: A set S of admissible swaps of minimum cost so that
parameter algorithm for winner determination. However, as in the c ⋆ wins the election (C, V S ) under the rule R.
case of Dodgson-Bribery, they do not provide a fixed-parameter
algorithm for Dodgson ′ -Bribery.
The issue is then that a subsequent paper of Falisezwski et al. [10] 3 MOVES IN SOCIETIES AND PRESBURGER
claims that the Dodgson rule is “integer-linear-program imple- ARITHMETIC
mentable” and that this implies a certain election control problem Let τ ∈ N be the number of types of people. A society is a non-
generalizing Bribery to be fixed-parameter tractable [10, Theorem negative τ -dimensional integer vector s = (s 1 , . . . , sτ ). In most
6.2]. We believe the authors do not sufficiently differentiate between problems, we are interested in modifying a society by moving
determining the winner with one ILP, as is the case for most simple people between types. A move is a vector m = (m 1,1 , . . . , mτ ,τ ) ∈
2
voting rules, and with multiple ILPs, as is the case for Dodgson. Thus, Zτ . Intuitively, mi, j is the number of people of type i turning type j.
we believe there is no evidence that the Dodgson rule is “integer-
linear-program implementable”. Yet, this may be possible and this Definition 3.1. A change is a vector ∆ = (∆1 , . . . , ∆τ ) ∈ Zτ
question still deserves attention. Whatever the reason, we are con- whose elements sum up to 0. We say that ∆ is the change associated
with a move m if ∆i = τj=1 m j,i − mi, j , and we write ∆ = ∆(m).
Í
vinced that their [10, Theorem 6.1] does not hold for R=Dodgson.

259
Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

A change ∆ is feasible with respect to society s if s + ∆ ≥ 0, i.e., if ∃/∀x1

applying the change ∆ to s results in a society. ..
.
One more useful notion is that of a “move costs vector”. A move ∃/∀xk
2
costs vector is a vector c = (c 1,1 , . . . , cτ ,τ ) in (N ∪ {+∞})τ which
(∨), δ
satisfies the triangle inequality, i.e., c i,k ≤ c i, j + c j,k for all distinct
i, j, k. Let k ∈ N and c be a move costs vector. A move m is a
(c, k)-move if c⊺ m ≤ k. (∧), γ (∧), γ ... (∧), γ
Finally, we want to check that the society (e.g., resulting from Φ
applying some moves) satisfies a certain desired condition. This a1 · (x0 , . . . , xk ) ≤ b1
condition depends on the problem we are modeling: in variants of .. ≤γ
bribery, it says that a preferred candidate is elected as a winner or .
ap · (x0 , . . . , xk ) ≤ bp
to be a part of a committee under a given voting rule; in the context
of lobbying, it says that a preferred agenda was selected. To allow ≤α ≤β
large expressibility, we make a very broad definition: A winning
condition of width τ is a predicate Ψ(s) with τ free variables. Figure 1: Visualization of PA formula Ψ(x0 ) in DNF.

3.1 Presburger Arithmetic

• feasible(s, ∆) ≡ (s + ∆(m) ≥ 0 ∧ 1⊺ ∆ = 0) ∈
For two formulas Φ and Ψ, we denote their equivalence by Φ  Ψ.
P0,(2τ ),0,τ ,2,0 , and,
Definition 3.2 (Presburger Arithmetic). Let • (c, k)-move(m) ≡ c⊺ m ≤ k ∈ P0,(τ ),0,1, ∥c ∥∞,k .
P̂0,(n0 ),δ,γ ,α, β = Ψ̂(x0 ) We note that, for elections, our definition of winning condition

generalizes the notion of “linearly-definable voting rules” by Dorn
be the set of quantifier-free Presburger Arithmetic (PA) formulas
and Schlotter [6]. Precisely, those rules belong to ∃P with k = 1;
with n 0 free variables x0 which are a disjunction of at most δ con-
we will show that Dodgson and Young are in ∃P with k = 2. Thus,
junctions of linear inequalities a⊺ x0 ≤ b, each of length at most γ ,
our winning conditions capture an extensive set of voting rules.
where ∥a∥∞ ≤ α and |b | ≤ β for each inequality. Then, let
n o
P0,(n0 ),δ,γ ,α, β = Ψ(x0 ) | ∃Ψ̂(x0 ) ∈ P̂0,(n0 ),δ,γ ,α, β : Ψ̂  Ψ 3.2 Modeling Problems as Minimum Move
We model moves in societies by the following general problem:
be the set of PA formulas equivalent to some DNF formula from
P̂0,(n0 ),δ,γ ,α, β . Finally, let Minimum Move
2
Pk,n,δ,γ ,α, β = {Ψ(x0 ) ≡ ∃/∀x1 ∃/∀x2 · · · ∃/∀xk : Φ(x0 , x1 , . . . , xk )} Input: A society s, an objective function f : Zτ → Z, a
winning condition Ψ.
be the set of PA formulas with quantifier depth k, n 0 free variables x0 , Find: A move m minimizing f (m) s.t. Ψ(s + ∆(m)) ∧
and dimension n = (n 0 , n 1 , . . . , nk ); here, x i ∈ Zni for each i and feasible(s, ∆(m)).
Φ(x0 , . . . , xk ) ∈ P0,(n),δ,γ ,α, β with n = ki=1 ni . The length of
Í
Ψ(x0 ) is the number of symbols it contains, which is polynomi- It models many well-studied problems:
ally bounded in (n, γ , δ ). By ∃P and ∀P we denote the sets of PA Multi bribery. Knop et al. [18] introduce a generalization of vari-
formulas whose leading quantifier is ∃ or ∀, respectively. ous bribery problems called R-Multi Bribery. Informally, we are
given an election (C, V ) where each voter further has an approval
Example. A simple example of PA is the following formula.
count, and is either active or latent the status of which can be
Ψ(y) ≡ ∀x 1x 2 ∃z 1z 2z 3 : (x 1 + y = z 3 ∧ y ≥ 0) ∨ changed at certain cost; likewise, there are costs for perturbing
3x 1 + 10y − 3z 1 ≤ 13 ∧ 2x 2 + 5y − z 2 ≤ 11 their preference order or approval count. This problem general-
izes Bribery, $Bribery, Swap Bribery, Shift Bribery, Support
∧ x 1 + 1y − z 3 ≥ 9 ∧ z 1 − z 2 + 2z 3 ≤ 6


Bribery, Extension Bribery, Possible Winner, Constructive
Here k = 2, n = (1, 2, 3) , δ = 2, γ = 4, α = 10, and β = 13. Control by Adding/Deleting Voters and other problems.
Notice that there are at most |C |! possible preference orders, at
We study winning conditions Ψ(s) expressible in PA, and state
most |C | possible approval counts, and 2 states “active” or “latent”.
our complexity results with respect to the descriptive complexity
Thus, there are at most τ ≤ 2|C | · |C |! potential types of voters,
of Ψ, which is its number of variables, quantifiers, logical connec-
and we can express the input election as a society s. A move costs
tives, and unary encoding length of coefficients and constants.
vector c describing the costs of moving a voter from one type to
Vocabulary. We express relevant definitions by simple PA for- another is obtained by calculating (possibly using a shortest path
mulas over integral variables and with integer coefficients and algorithm) the least costs based on the given cost functions. Let Ψ(s)
constants: be a PA formula which is satisfied if the preferred candidate wins
• society(s) ≡ s ≥ 0 ∈ P0,(τ ),0,τ ,1,0 , under the voting rule R in a society s. Then, a bribery of minimum
• move(m) ∈ P0,(τ 2 ),0,0,0,0 , cost in a R-Multi Bribery instance can be modeled as solving
• ∆ = ∆(m) is a linear map ∆i = τj=1 m j,i − mi, j ; thus if we Minimum Move with f (m) = c⊺ m. This modeling, combined with
Í

let Ψ(m, ∆) ≡ ∆ = ∆(m), then Ψ(m, ∆) ∈ P0,(τ +τ 2 ),0,τ ,1,0 Theorem 1.2 and Corollary 4.6, yields Corollary 1.3.

260
Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

Multiwinner elections. Bredereck et al. [4] study the complexity i.e., the set of feasible solutions does not change by restricting all
of Shift Bribery in committee elections, that is, in elections with quantifiers and free variables to the corresponding box of size B.
multiple winners. The modeling is exactly the same as above, except Moreover, we say that any Ψ ∈ Pk,n,δ,γ , B, B is B-small if it is
for the winning condition Ψ which will be a long disjunction over B-bounded, that is, its coefficients and constants are bounded by B.
all committees which include the preferred candidate.
A special case are ILPs which are B-small; they correspond to
Lobbying in referenda. Bredereck et al. [3] study the complex- PA formulas with k = 0 and δ = 0; for such formulas we show:
ity of Lobbying in referenda. There, voters cast ballots with their
“yes”/“no” answers to issues. The task is to push an agenda, i.e., a Lemma 4.4 (ILP disjunctions [folklore]). Let Ai x ≤ bi for
certain outcome. Again, voters fall into groups according to their i ∈ [d] be B-small ILPs with Ai ∈ Zm×n for each i ∈ [d]. Then,
a (B 2n)-small system Ax ≤ b with A ∈ Z(md +d+1)×(n+d) can be
ballots, the costs of changing their opinions forms a move costs vec- 
tor, and a winning condition Ψ expresses that the selected agenda
Íd
constructed in time O dm + n + i=1 ⟨Ai , bi ⟩ such that
succeeded. Ü
∃(x, y) ∈ Zn+d : A(x, y) ≤ b ⇐⇒ ∃x ∈ Zn : Ai x ≤ bi .
4 SENTENCES WITH TWO QUANTIFIERS i ∈[d]
We shall now introduce the building blocks of our proof of Theo-
Proof. Let M = B 2n, let yi for i ∈ [d] be binary variables, and
rem 1.1. Woods [24] gives an algorithm that efficiently converts any
consider the following system:
quantifier-free PA formula Φ into an equivalent DNF formula Φ̂ of
bounded length: d
Õ Û
yi = 1 yi ≥ 0, Ai x ≤ bi + M(1 − yi ) for all i ∈ [d] .
Lemma 4.1 (Woods [24, Proposition 5.1]). Let Φ(x) be a quanti- i=1
fier-free PA formula with x ∈ Zd containing N inequalities, whose Assume it has an integer solution y. Then there is an index i ∈ [d]
coefficients and right-hand sides are bounded in absolute value by α such that yi = 1 and thus Ai x ≤ bi + 0 holds; thus, the system
and β, respectively. Then Φ(x) can be converted into an equivalent DNF Ai x ≤ bi has an integer solution. In the other direction, assume
formula Φ̂(x) with at most δ = N O (d) disjunctions, each containing that the system Ai x ≤ bi has a solution x; then let yi = 1. We shall
at most N conjunctions with the same bound on α and β. prove that Ai ′ x ≤ bi ′ + M holds for all i ′ , i. Since Ai ′ x ≤ bi ′ is
It is often useful for the quantifiers of a PA formula to range over B-small, each of its row sums has n terms which are a multiple of
integer points of polyhedra, e.g. ∀x ∈ Q (we do not write Q ∩ Zn for two numbers, each bounded by B, and thus is at most B 2n. Moreover,
brevity, as we assume everything to be integer); again, our definition since yi = 1, we have yi ′ = 0 and thus the right hand side is bi ′ + M
is not restrictive by the fact that we can always rearrange: and every assignment of x feasible for Ai x ≤ bi satisfies it. Clearly,
the new system has n + d variables, md + d + 1 inequalities, is
Ψ(x0 ) ≡ ∃x1 ∈ Q 1 · · · ∀/∃xk ∈ Q k : Φ(x0 , x1 , . . . , xk ) ≡ B-bounded and ∥A∥∞ = B 2n and thus it is (B 2n)-small, and can be

∃x1 · · · ∀/∃xk : Φ(x0 , x1 , . . . , xk ) constructed in the claimed time. □
Û Ü  We now prove Theorem 1.1; we restate it in formal terms here:
(x1 ∈ Q 1 ∧ x3 ∈ Q 3 · · · )∧ (x2 < Q 2 ∨ x4 < Q 4 · · · )
Theorem 4.5 (formal version of Theorem 1.1). Let Ψ be a
Parametric ILP. A special case of PA are parametric ILPs, which β-small P2,n,δ,γ ,α, β sentence (i.e., without free variables and thus
can be viewed2 as deciding the sentence n 0 = 0). Then Ψ can be decided in time д(n, δ, γ ) poly(α, β) for some
∀x ∈ Zp : Ax ≤ b ∃y ∈ Zn : B(x, y) ≤ e, computable function д.
where A ∈ Zℓ×p and B ∈ Zm×n are integer matrices. A consequence Proof. Let Ψ ≡ ∃x1 ∀x2 : Φ(x1 , x2 ). Clearly, to decide Ψ we can
of an algorithm of Eisenbrand and Shmonin [7] is the following: instead decide ¬Ψ ≡ ∀x1 ∃x2 : ¬Φ(x1 , x2 ). Consider the formula
Corollary 4.2 ([7, Theorem 4.2], [5, Corollary 1]). Any para- ¬Φ(x1 , x2 ): by Lemma 4.1, there exists an equivalent β-small DNF
metric ILP whose entries of A, B, b and e are given in unary, is fixed- formula Φ̄(x1 , x2 ) such that the number of its disjunctions and
parameter tractable when parameterized by n, m and p. conjunctions is a function of just the original δ and γ .
Thus, from now on focus on the case Ψ ≡ ∀x1 ∃x2 : Φ(x1 , x2 ).
ILP and disjunctions. We shall use a folklore result about imple- Our next task is to construct an instance of Parametric ILP equiv-
menting disjunctions in ILP when the domains of variables can be alent to deciding Ψ. To do that, replace Φ(x1 , x2 ) with Ξ(x1 , x2 ) ≡
bounded. For that, we need another definition. ∃x3 : A(x1 , x2 , x3 ) ≤ b, where x3 is of dimension δ and the sys-
tem of linear inequalities A(x1 , x2 , x3 ) ≤ b has bounded length,
Definition 4.3 (B-bounded, B-small PA formula). Let Ψ(x0 ) ≡ coefficients and right sides. Now we use Lemma 4.4. Assume that
∃/∀x1 · · · ∃/∀xk : Φ(x0 , . . . , xk ) be a PA formula, and let ΨB (x0 ) ≡ Φ(x1 , x2 ) is a disjunction of δ linear systems, each of at most γ
∃/∀x1 ∈ [−B, B]n1 · · · ∃/∀xk ∈ [−B, B]nk : Φ(x0 , . . . , xk ). Then conjunctions, and by the assumptions of the theorem we know
we say that Ψ(x0 ) is B-bounded if that Ψ is β-small. Plugging into Lemma 4.4, we have B = β,
x ∈ Zn0 | ΨB (x) ∩ [−B, B]n0 = x ∈ Zn0 | Ψ(x) , d = δ , m = γ and n = n 1 + n 2 , and we obtain a formula
 

2 Parametric
∃x3 : A(x1 , x2 , x3 ) ≤ b equivalent to Φ(x1 , x2 ). Thus, we are left
ILPs are typically viewed as ILPs with a varying right hand side, that is,
deciding the sentence ∀b∃x : Ax ≤ b; it is known that our formulation is equivalent, with deciding ∀x1 ∃(x2 , x3 ) : A(x1 , x2 , x3 ) ≤ b with the following
as shown by Crampton et al. [5], who call it ILP Resiliency. parameters:

261
Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

• The coefficients and right-hand sides ∥(A, b)∥∞ are bounded by a bribery which, for some d ∈ N, makes c ⋆ be in distance d or
some computable function f (n, δ, γ ) · (α β)2 . less from being a Condorcet winner, and, simultaneously, making
• The dimensions of A are bounded by f (n, δ, γ ). every other candidate c , c ⋆ be in distance at least d from being
Thus, we are in the setting of Corollary 4.2 and we can decide the a Condorcet winner.
above sentence by a fixed-parameter algorithm. □ Let us fix d ∈ N; later we will argue that we can go through all D
relevant choices of d. First, we will express the condition that in a
Corollary 4.6. Let Ψ(x0 ) be a disjunction of D many β-small society s, candidate c beats candidate c ′ in a head-to-head contest:
∃P2,(τ ,n1,n2 ),δ,γ ,α, β formulas. Then Minimum Move with objec- Õ Õ
tive f (x0 ) = c⊺ x0 and winning condition Ψ can be solved in time beats(c, c ′, s) ≡ si > si ,
д(τ , n 1 , n 2 , δ, γ ) poly(D, α, β, c) for some computable function д. i:c ≻i c ′ i:c ′ ≻i c

Proof. Let Ψ ≡ Ψ1 ∨ · · · ∨ ΨD , and let Ξi (m) ≡ Ψi (s + ∆(m)) ∧ where ≻i is the preference order shared by all voters of type i. Then,
(c, B)-move(m)∧feasible(∆(m), s), for each i ∈ [D]. Now we need it is easy to express that c ⋆ is a winner in s under Condorcet’s rule:
to perform binary search with parameter B on the sentence
Û
ΦCondorcet (c, s) ≡ beats(c, c ′, s) .
∃mΨ(s + ∆(m) ∧ (c, B)-move(m) ∧ feasible(∆(m), s) . c ′ ,c

This sentence is obviously equivalent to Finally, we express that c ⋆ is a Dodgson-winner with score d by
∃mΞ1 (m) ∨ · · · ∨ ∃mΞD (m), ΨDodgson,d as follows:

which can be decided by D application of Theorem 4.5. This is ΨDodgson,d (s) ≡ ∃m : (cswap , d)-move(m) ∧ feasible(∆(m), s)∧
possible in the claimed time, as, for each i ∈ [D], ∃mΞi (m) has ΦCondorcet (c ⋆ , s + ∆(m)) ∧


quantifier depth 2, is β ′ -small for β ′ = poly(β, τ + n 1 + n 2 , ∥c∥∞ )
and has dimension n = (0, τ +n 1 , n 2 ). A minimum move can then be ∀m : (cswap , d − 1)-move(m) ∧ feasible(∆(m), s)∧
Û
constructed by coordinate-wise binary search; cf. [19, Thm. 4.1]. □ ¬ΦCondorcet (c, s + ∆(m))

c,c ⋆
4.1 Application: Swap Bribery for the Dodgson
Then, ΨDodgson ≡ ΨDodgson,1 ∨ · · · ∨ ΨDodgson, D . Clearly, ΨYoung,d
Rule and Young Rule is obtained simply by replacing cswap with cdel .
We now prove Theorem 1.2 by giving our fixed-parameter algorithm Complexity. The number D of relevant choices of d is bounded
for R-Swap Bribery with R the Dodgson Rule or the Young Rule. by |C | 2 |V |: at most |C | 2 swaps suffice for any bribery of a single
Proof of Theorem 1.2. Fix an instance (C, V , c ⋆ , σ ) of R-Swap voter, and there are at most |V | voters, thus D ≤ |C | 2 |V | is a bound
Bribery. As there are |C |! possible total orders on C, each voter on the Dodgson score of a candidate. For the Young score, D ≤ |V |.
has one of these orders. Thus we view the election as a society Since for each d ∈ [D], ΨDodgson,d ≡ ∃m : Ξ1 (m) ∧ ∀m : Ξ2 (m)
s = (s 1 , . . . , s |C |! ). Let cswap be a move costs vector defined as can be equivalently rewritten as ∃m∀m ′ : Ξ1 (m) ∧ Ξ2 (m ′ ), it has
c i, j =“swap distance between types i and j” (this is simply the num- quantifier depth 2, and thus belongs to ∃P2,n,δ,γ ,α, β . Let us deter-
ber of inversions between the permutations i and j [8, Proposition mine the parameters:
3.2]); observe that a society s is in swap distance at most d from s ′ • n is the vector of dimensions; thus n 0 = τ = |C |! are the dimen-
if s ′ = s + ∆(m) and m is a feasible (cswap , d)-move for s. Moreover, sions of a vector encoding a society, and n 1 = τ 2 = |C |!2 are the
let cdel be defined as c i,0 = 1 for every type i and c i, j = +∞ for dimensions of a vector encoding a move.
every two types i, j , 0, where 0 is a type for latent voters; observe • δ is the number of disjunctions, and it is polynomial in |C |,
that s is in voter deletion distance at most d from s ′ if s ′ = s + ∆(m) • γ is the number of conjunctions, and it is polynomial in τ = |C |!,
and m is a (cdel , d)-move. • α is the largest coefficient, which is ∥cswap ∥∞ ≤ |C | 2 ,
Our plan is to express the winning condition for R using a PA for- • β is the largest right-hand side, which is d ≤ |C | 2 |V |.
mula ΨDodgson (s) which is a disjunction of polynomially many (in Thus, n, δ, γ are functions of the parameter |C | and α, β are polyno-
|V | and |C |) formulas from ∃P2,n,δ,γ ,α, β , and then solve Minimum mial in the size of the input election (C, V , {≻v | v ∈ V }), as required
Move with f (m) = σ ⊺ m and Ψ = ΨDodgson (s) using Corollary 4.6; by Corollary 4.6. Analogous analysis applies to ΨYoung,d . □
recall that since the instance is uniform, we have σ v ≡ σ for all
v ∈ V , and we let σ be the move costs vector obtained from σ . Replacing cswap with 1⊺ produces the winning condition for
For ΨYoung , we proceed analogously. In the end, we will verify that Dodgson ′ as introduced by Faliszewski et al. [9]. Furthermore, it
n, δ, γ are bounded by a function of |C |, and that α, β are polynomial is interesting to consider voting rules obtained by replacing Con-
in the input size. dorcet’s rule in the definition of Dodgson’s and Young’s rule. For
Expressing ΨDodgson and ΨYoung . The winning condition for can- example, the Majority rule also might not produce a winner, and
didate c ⋆ in both the Dodgson and Young rule can be viewed as c ⋆ most rules (Copelandα , Scoring protocol etc.) allow ties. Let the
being closest to being a Condorcet winner with respect to some score of a candidate be their distance (swap, deletion, etc.) from
distance measure. Specifically, for the Dodgson rule, this distance being a (unique) winner under rule R. We remark that if we replace
is the number of swaps in the preference orders, and in the Young Condorcet’s rule with R with a “simple” PA desciription, it corre-
rule, the distance is the number of voter deletions. For this reason, sponds to replacing ΦCondorcet in the proof above, thus yielding
finding a bribery which makes c ⋆ a winner corresponds to finding fixed-parameter tractable algorithms for all such rules as well.

262
Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

4.2 Application: Resilient and Robust Moves be the cost of moving from i to the cheapest equivalent of j, and let
Theorem 1.1 allows us to develop fixed-parameter algorithms for c ij,k = +∞ for any i , j, k ∈ [C]. Let c̄ = (c1 , . . . , cτ )
problems related to moves in society. Problem Robust Move asks Then, consider the following ILP with variables x =
for a move that is robust to any adversary move of cost at most Ba : (x1 , . . . , xτ ) ∈ Zτ C ; we obtain the minimum move m from its
optimal solution by taking mi, j = x ji ′ where j ′ ∼ j:
Robust Move
Input: A society s, move costs vectors c and ca , a winning
Õ Õ
min c̄⊺ x s.t. Āxi ≤ b, x ji = si ∀i ∈ [τ ] .
condition Ψ, budgets B, Ba ∈ N i=1, ...,τ j=1, ...,C
Find: A (c, B)-move m such that for every adversary
(ca , Ba )-move ma , Ψ(s + ∆(ma ) + ∆(m)) holds. This ILP is a combinatorial pre-n-fold IP; by Knop et al. [16] (detailed
in [17, Corollary 23]), it can be solved in the claimed time. □
The second problem, Resilient Budget, asks if a budget B suf-
fices to counter any adversary move of cost at most Ba . Crampton et 6 DISCUSSION
al. [5] consider a specialization of this problem for R-Swap Bribery.
We raise three important questions which naturally arise from
Resilient Budget this work. First, we ask whether our fixed-parameter algorithm for
Input: A society s, move costs vectors c and ca , a winning {Dodgson,Young}-Swap Bribery with uniform cost extends to gen-
condition Ψ, budgets B, Ba ∈ N eral cost functions. This parallels [2, Challenge #2]. For much sim-
Find: Does for every adversary (ca , Ba )-move ma exist a pler voting rules an analogous result was shown only recently [18].
(c, B)-move m such that Ψ(s + ∆(ma ) + ∆(m))? We believe that if the answer is positive, proving it would require
providing new powerful integer programming tools, in particular,
Theorem 4.7. Robust Move and Resilient Budget with Ψ ∈ some analogue of Theorem 1.1 for n-fold integer programming,
P0,(n0 ),δ,γ ,α, β can be solved in time д(τ , n 0 , δ, γ )(α + β + B + Ba + which is the engine behind the recent progress [18].
∥c∥∞ + ∥ca ∥∞ )O (1) , that is, FPT parameterized by τ + n 0 + δ + γ . Second, can the run time of our algorithm be improved? This
analogously recalls [2, Challenge #1]. The run time of Corollary 4.6
Proof. We apply Theorem 1.1 to decide the following formulas, is double-exponential in the dimension O(τ ); thus Theorem 1.2
which is clearly equivalent to deciding the problems at hand: shows that Dodgson-Swap Bribery is solvable in triple-exponential
time in parameter |C |. We believe it can be improved, but we are
Ψ(s)RB ≡∃m : feasible(s, ∆(m)) ∧ (c, B)-move(m) ∧
sceptical that it could be made single-exponential, and thus ask if
∀ma : feasible(s + ∆(m), ∆(ma )) ∧ (ca , Ba )-move(ma ) ∧ a double-exponential lower bound holds. A related question is to
Ψ(s + ∆(m) + ∆(ma )) show lower bounds separating the complexity of R-Swap Bribery
Ψ(s)RM ≡∀ma : feasible(s, ∆(ma )) ∧ (ca , Ba )-move(ma ) ∧ for different voting rules such as Borda, Copeland and Kemeny.
Third, we note that, unlike most previous results, the run time of
∃m : feasible(s + ∆(ma ), ∆(m)) ∧ (c, B)-move(m) ∧
our algorithm depends polynomially on the number |V | of voters.
Ψ(s + ∆(m) + ∆(ma )) □ Many previous results solve the succinct variant of the problem
(cf. Falizewski et al. [9]) and depend polynomially only on log |V |.
5 POLYNOMIALLY MANY TYPES Thus we ask whether {Dodgson,Young}-Swap Bribery is fixed-pa-
We now prove Theorem 1.4, which formally reads: rameter tractable also in the succinct variant.
Finally, we discuss our usage of Presburger arithmetic as a gen-
Theorem 5.1. Let Ψ(s) ∈ P0,(τ ,n1 ),δ,γ ,α, β be a winning condition. eralization of ILPs. Solving ILPs amounts to deciding ∃x : Ax ≤ b;
Minimum Move can be solved in time д(γ , α)(τ +δ )O (1) log(s) for any by Lenstra’s algorithm [20] and its improvements by Kannan [15]
linear function f (m) = c⊺ m, where д is some computable function. and Frank and Tardos [12], this task is fixed-parameter tractable
parameterized by the dimension of x even for unbounded ∥A, b∥∞
Proof. Since k = 0, Φ(y) is a disjunction of δ linear systems and if A has polynomially (in the length of the input) many rows.
Ai y ≤ bi , with ∥Ai ∥∞ ≤ α and with at most γ rows, for every In 1990, Kannan claimed to show that Parametric ILP, which
i ∈ [δ ]. Thus we instead solve δ instances of Minimum Move with amounts to deciding ∀b ∈ Q ∃x : Ax ≤ b for some polyhedron Q,
Φi (y) ≡ Ai y ≤ bi and pick the best solution among them. So from is fixed-parameter tractable parameterized by the dimension of x;
now on assume that Φ(y) ≡ Ay ≤ b with ∥A∥∞ ≤ α and A ∈ Zγ ×τ . here, ∥A∥∞ must be bounded by a polynomial and the number of
Observe that A can have at most α O (γ ) different columns. For rows of A also has to be a parameter. However, Kannan’s result
two types i, j ∈ [τ ], we say they are equivalent and write i ∼ j if relies on Kannan’s Partitioning Theorem (KPT), which was recently
the columns Ai and A j are identical. Thus, the τ types of people disproved by Nguyen and Pak [21]. Nguyen and Pak [21, Theorem
fall into C ≤ α O (γ ) equivalence classes. For every type i ∈ [τ ], let 1.9] state that Woods [24] gave a polynomial-time algorithm for de-
C[i] = {j ∈ [τ ] | j ∼ i} be the equivalence class containing i, and ciding ∀y∃x : Φ(x, y) when the dimensions of x and y are constant;
let Ā ∈ Zγ ×C be a matrix with, for every i ∈ [C], Āi = A j where however, it is unclear if this is a fixed-parameter algorithm. For
j ∈ C[i]. Now, for every type i ∈ [τ ], we shall create a reduced this reason, as we have the dimensions of x and y as (non-constant)
2
custom move costs vector ci ∈ NC . For every j ∈ [C], i , j, let parameter, we chose to prove Theorem 1.1 as a slightly weaker
result (still sufficient for our purposes) but using only elementary
c i,i j = min c i, j ′
j ′ ∈ C[j] techniques.

263
Session 8: Social Choice on Networks AAMAS 2018, July 10-15, 2018, Stockholm, Sweden

REFERENCES
[1] John J. Bartholdi III, Craig A. Tovey, and Michael A. Trick. 1989. Voting
schemes for which it can be difficult to tell who won the election. Soc.
[13] Edith Hemaspaandra, Lane A. Hemaspaandra, and Jörg Rothe. 1997.
Choice Welfare 6, 2 (1989), 157–165.
Exact analysis of Dodgson elections: Lewis Carroll’s 1876 voting system
[2] Robert Bredereck, Jiehua Chen, Piotr Faliszewski, Jiong Guo, Rolf Nie-
is complete for parallel access to NP. J. ACM 44, 6 (1997), 806–825.
dermeier, and Gerhard J. Woeginger. 2014. Parameterized algorithmics
[14] Edith Hemaspaandra, Holger Spakowski, and Jörg Vogel. 2005. The
for computational social choice: Nine research challenges. Tsinghua Sci.
complexity of Kemeny elections. Theoret. Comput. Sci. 349, 3 (2005),
Tech. 19, 4 (2014), 358–373.
382–391.
[3] Robert Bredereck, Jiehua Chen, Sepp Hartung, Stefan Kratsch, Rolf Nie-
[15] Ravi Kannan. 1987. Minkowski’s convex body theorem and integer
dermeier, Ondřej Suchý, and Gerhard J. Woeginger. 2014. A multivariate
programming. Math. Oper. Res. 12, 3 (Aug. 1987), 415âĂŞ440.
complexity analysis of lobbying in multiple referenda. J. Artificial Intelli-
[16] Dušan Knop, Martin Koutecký, and Matthias Mnich. 2017. Combinatorial
gence Res. 50 (2014), 409–446.
n -fold integer programming and applications. In Proc. ESA 2017 (Leibniz
[4] Robert Bredereck, Piotr Faliszewski, Rolf Niedermeier, Piotr Skowron,
Int. Proc. Informatics), Vol. 87. 54:1–54:14.
and Nimrod Talmon. 2016. Complexity of shift bribery in committee
[17] Dušan Knop, Martin Koutecký, and Matthias Mnich. 2017. Combinatorial
elections. In Proc. AAAI 2016. 2452–2458.
n -fold integer programming and applications. Technical Report. https:
[5] Jason Crampton, Gregory Gutin, Martin Koutecký, and Rémi Watrigant.
//arxiv.org/abs/1705.08657.
2017. Parameterized resiliency problems via integer linear programming.
[18] Dušan Knop, Martin Koutecký, and Matthias Mnich. 2017. Voting and
In Proc. CIAC 2017 (Lecture Notes Comput. Sci.), Vol. 10236. 164–176.
bribing in single-exponential time. In Proc. STACS 2017 (Leibniz Int. Proc.
[6] Britta Dorn and Ildikó Schlotter. 2012. Multivariate complexity analysis
Informatics), Vol. 66. 46:1–46:14.
of swap bribery. Algorithmica 64, 1 (2012), 126–151.
[19] Matthias Köppe, Maurice Queyranne, and Chris T. Ryan. 2010. Parametric
[7] Friedrich Eisenbrand and Gennady Shmonin. 2008. Parametric integer
integer programming algorithm for bilevel mixed integer programs. J.
programming in fixed dimension. Math. Oper. Res. 33, 4 (2008).
Optim. Theory Appl. 146, 1 (2010), 137–150.
[8] Edith Elkind, Piotr Faliszewski, and Arkadii Slinko. 2009. Swap bribery.
[20] Hendrik W. Lenstra, Jr. 1983. Integer programming with a fixed number
In Proc. SAGT 2009 (Lecture Notes Comput. Sci.), Vol. 5814. 299–310.
of variables. Math. Oper. Res. 8, 4 (1983), 538–548.
[9] Piotr Faliszewski, Edith Hemaspaandra, and Lane A. Hemaspaandra.
[21] Danny Nguyen and Igor Pak. 2017. Complexity of short Presburger
2009. How hard is bribery in elections? J. Artificial Intelligence Res. 40
arithmetic. In Proc. STOC 2017. 812–820.
(2009), 485–532.
[22] Jörg Rothe, Holger Spakowski, and Jörg Vogel. 2003. Exact complexity
[10] Piotr Faliszewski, Edith Hemaspaandra, and Lane A. Hemaspaandra.
of the winner problem for Young elections. Theory Comput. Syst. 36, 4
2011. Multimode control attacks on elections. J. Artificial Intelligence
(2003), 375–386.
Res. 40 (2011), 305–351. Issue 1.
[23] Ildikó Schlotter, Piotr Faliszewski, and Edith Elkind. 2017. Campaign
[11] Piotr Faliszewski, Yannick Reisch, Jörg Rothe, and Lena Schend. 2014.
management under approval-driven voting rules. Algorithmica 77, 1
Complexity of manipulation, bribery, and campaign management in
(2017), 84–115.
Bucklin and fallback voting. In Proc. AAMAS 2014. 1357–1358.
[24] Kevin Woods. 2015. Presburger arithmetic, rational generating functions,
[12] András Frank and Éva Tardos. 1987. An application of simultaneous Dio-
and quasi-polynomials. J. Symbolic Logic 80, 2 (2015), 433–449.
phantine approximation in combinatorial optimization. Combinatorica 7,
1 (1987), 49–65.

264