Professional Documents
Culture Documents
Textbook Computer Algebra in Scientific Computing 16Th International Workshop Casc 2014 Warsaw Poland September 8 12 2014 Proceedings 1St Edition Vladimir P Gerdt Ebook All Chapter PDF
Textbook Computer Algebra in Scientific Computing 16Th International Workshop Casc 2014 Warsaw Poland September 8 12 2014 Proceedings 1St Edition Vladimir P Gerdt Ebook All Chapter PDF
https://textbookfull.com/product/computer-algebra-in-scientific-
computing-20th-international-workshop-casc-2018-lille-france-
september-17-21-2018-proceedings-vladimir-p-gerdt/
https://textbookfull.com/product/computer-algebra-in-scientific-
computing-22nd-international-workshop-casc-2020-linz-austria-
september-14-18-2020-proceedings-francois-boulier/
https://textbookfull.com/product/parameterized-and-exact-
computation-9th-international-symposium-ipec-2014-wroclaw-poland-
september-10-12-2014-revised-selected-papers-1st-edition-marek-
cygan/
Computer Algebra
in Scientific Computing
16th International Workshop, CASC 2014
Warsaw, Poland, September 8–12, 2014
Proceedings
123
Lecture Notes in Computer Science 8660
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen
Editorial Board
David Hutchison
Lancaster University, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Alfred Kobsa
University of California, Irvine, CA, USA
Friedemann Mattern
ETH Zurich, Switzerland
John C. Mitchell
Stanford University, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
Oscar Nierstrasz
University of Bern, Switzerland
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbruecken, Germany
Vladimir P. Gerdt Wolfram Koepf
Werner M. Seiler Evgenii V. Vorozhtsov (Eds.)
Computer Algebra
in Scientific Computing
16th International Workshop, CASC 2014
Warsaw, Poland, September 8-12, 2014
Proceedings
13
Volume Editors
Vladimir P. Gerdt
Joint Institute of Nuclear Research
Laboratory of Information Technologies (LIT)
Dubna, Russia
E-mail: gerdt@jinr.ru
Wolfram Koepf
Universität Kassel, Institut für Mathematik
Kassel, Germany
E-mail: koepf@mathematik.uni-kassel.de
Werner M. Seiler
Universität Kassel, Institut für Mathematik
Kassel, Germany
E-mail: seiler@mathematik.uni-kassel.de
Evgenii V. Vorozhtsov
Russian Academy of Sciences
Institute of Theoretical and Applied Mechanics
Novosibirsk, Russia
E-mail: vorozh@itam.nsc.ru
eigenvalue method with symmetry for solving polynomial systems arising in the
vibration analysis of mechanical structures with symmetry properties, applica-
tion of Gröbner systems for computing the (absolute) reduction number of poly-
nomial ideals, the application of cylindrical algebraic decomposition for solving
the quantifier elimination problems, certification of approximate roots of overde-
termined and singular polynomial systems via the recovery of an exact rational
univariate representation from approximate numerical data, and new parallel
algorithms for operations on univariate polynomials (multi-point evaluation, in-
terpolation) based on subproduct tree techniques.
Several papers are devoted to using CA for the investigation of various math-
ematical and applied topics related to ordinary differential equations (ODEs):
application of CAS Mathematica for the investigation of movable singularities
of the complex valued solutions of ODEs, the decidability problem for linear
ODE systems with variable coefficients, conversion of nonlinear ODEs to inte-
gral equations for the purpose of parameter estimation.
The invited talk by L. Plaskota deals with the information-based complexity
(IBC) as a branch of computational complexity that studies continuous prob-
lems, for which available information is partial, noisy, and priced. The basic ideas
of IBC are presented, and some sample results on optimal algorithms, complexity,
and tractability of such problems are given. The focus is on numerical integration
of univariate and multivariate functions.
A number of papers deal with applications of symbolic and symbolic-numeric
computations for investigating and solving partial differential equations (PDEs)
in mathematical physics: symbolic solution of algebraic PDEs, testing unique-
ness of analytic solutions of PDE with boundary conditions, construction of
high-order difference schemes for solving the Poisson equation, symbolic-numeric
solution of the Burgers and Korteweg–de Vries–Burgers equations at very high
Reynolds numbers, and derivation of new analytic solutions of the PDEs gov-
erning the motion of a special Cosserat rod-like fiber.
Application of symbolic and symbolic-numeric algorithms in mechanics and
physics is represented by the following themes: analytic calculations in Maple
for modeling smoothly irregular integrated optical waveguide structures, the
integrability of evolutionary equations in the restricted three-body problem with
variable masses, quantum tunneling problem of diatomic molecule through
repulsive barriers, symbolic-numeric solution of the parametric self-adjoint Sturm–
Liouville problem, obtaining new invariant manifolds in the classic and general-
ized Goryachev–Chaplygin problem of rigid body dynamics.
The invited talk by G. Regensburger, article written jointly with S. Müller,
addresses a recent extension of chemical reaction network theory (CRNT), called
generalized mass-action systems, where reaction rates are allowed to be power-
laws in the concentrations. In particular, the kinetic orders (the real exponents)
can differ from the corresponding stoichiometric coefficients. As with mass action
kinetics, complex balancing equilibria (determined by the graph Laplacian of the
underlying network) can be characterized by binomial equations and parameter-
ized by monomials. However, uniqueness and existence for all rate constants and
Preface VII
CASC 2014 was organized jointly by the Institute of Mathematics at the Uni-
versity of Kassel, Kassel, Germany, and the Faculty of Applied Informatics and
Mathematics, Warsaw University of Life Sciences (SGGW), Warsaw, Poland.
Program Committee
François Boulier (Lille) Georg Regensburger (Linz)
Hans-Joachim Bungartz (Munich) Eugenio Roanes-Lozano (Madrid)
Jin-San Cheng (Beijing) Valery Romanovski (Maribor)
Victor F. Edneral (Moscow) Markus Rosenkranz (Canterbury)
Dima Grigoriev (Lille) Doru Stefanescu (Bucharest)
Jaime Gutierrez (Santander) Thomas Sturm (Saarbrücken)
Sergey A. Gutnik (Moscow) Jan Verschelde (Chicago)
Jeremy Johnson (Philadelphia) Stephen M. Watt
Victor Levandovskyy (Aachen) (W. Ontario, Canada)
Marc Moreno Maza (London, Canada) Andreas Weber (Bonn)
Alexander Prokopenya (Warsaw) Kazuhiro Yokoyama (Tokyo)
Additional Reviewers
Parisa Alvandi Jorge Caravantes
Hirokazu Anai Francisco-Jesus Castro-Jimenez
Alexander Batkhin Changbo Chen
Johannes Blömer Colin Denniston
Charles Bouillaguet Gema M. Diaz-Toca
Jürgen Bräckle Matthew England
X Organization
Local Organization
Alexander Prokopenya (Chair) Luiza Ochnio
Ryszard Kozera Artur Wilinski
Publicity Chair
Andreas Weber (Bonn)
Website
http://wwwmayr.in.tum.de/CASC2014/
Table of Contents
1 Introduction
Linear ordinary differential systems with variable coefficients appear in various
areas of mathematics. Power series are very important objects in the represen-
tation of the solutions of such systems as well as of the systems themselves. The
representation of infinite series lies at the core of computer algebra. A general
formula that expresses the coefficients of a series is not always available and may
even not exist. One natural way to represent the series is the algorithmic one,
i.e., providing an algorithm which computes its coefficients. Such algorithmic
representation of a concrete series is not, of course, unique. This non-uniqueness
is one of the reasons for undecidability of the zero testing problem for such
computable series.
At first glance, it may seem that if we cannot decide algorithmically whether
a concrete coefficient of a system is zero or not, then we will not be able to solve
any more or less interesting problem related to the search of solutions. However,
this is not completely right: at least, if we know in advance that the system is of
full rank then some of the problems can still be solved. For example, we can find
Supported in part by the Russian Foundation for Basic Research, project no. 13-01-
00182-a. The first author thanks also Department of Mathematics and Informatics
of XLIM Institute of Limoges University for the hospitality during his visits.
V.P. Gerdt et al. (Eds.): CASC Workshop 2014, LNCS 8660, pp. 1–12, 2014.
c Springer International Publishing Switzerland 2014
2 S.A. Abramov and M.A. Barkatou
Laurent series [3] and regular [6] solutions. Some non-trivial characteristics can be
computed as well, e.g., the so called “width” of the system [3]. Nevertheless, many
of the problems are undecidable. For example, we cannot answer algorithmically
the following question: does a given full-rank system with power series coefficients
have a formal exponential-logarithmic solution which is not regular? We prove this
undecidability in the present paper. It is also shown that if exponential-logarithmic
solutions of a given full-rank system exist then there exists a basis of the space of
those solutions such that all the series which appear in the elements of the basis are
computable; the exact formulation is given in Proposition 7 of this paper.
So, we know that there exists a basis of the solution space which consists
of computable objects, but we are not able to find this basis algorithmically.
This is analogous to some facts of constructive mathematical analysis. In fact,
the notion of a constructive real number (computable point) is fundamental
in that discipline: “... an algorithm which finds the zeros of any alternating,
continuous, computable function is impossible. At the same time, there cannot
be a computable function that assumes values of different signs at the ends of
a given interval and does not vanish at any computable point of this interval
(a priori, it is impossible to rule out the existence of computable alternating
functions whose zeros are all ‘noncomputable’). These results are due to Tseitin
[21] ...” ([14, p. 5], see also [16, §24]).
We prove in the same direction that testing unimodularity, i.e., the invertibil-
ity of the corresponding operator and computing the dimension of the solution
space of an arbitrary system can be done algorithmically if and only if the zero
testing problem in the ground differential field is algorithmically decidable. As
a consequence, these problems are undecidable when the coefficients are power
series or Laurent series which are represented by arbitrary algorithms.
If the algorithmic way of series representation is used then some of the prob-
lems related to linear ordinary systems are decidable while others are not. Note
that the above mentioned algorithms for finding Laurent series solutions and reg-
ular solutions are implemented in Maple [23]. The implementation is described
in [3,6] and, is available at http://www.ccas.ru/ca/doku.php/eg.
The rest of the paper is organized as follows: After stating some prelimi-
naries in Section 2, we give in Section 3 a review of some results related to
systems whose coefficients belong to a field K of characteristic zero. The field K
is supposed to be a constructive differential field, i.e., there exist algorithms for
the field operations, differentiation, and for zero testing. The problems that are
listed in Section 3 can be solved algorithmically. On the other hand, we show
in Section 4 that the same problems are algorithmically undecidable, if the field
K is semi-constructive, i.e., there exist algorithms for the field operations and
differentiation but there is no algorithm for zero testing. Finally, we consider in
Section 5 semi-constructive fields of computable formal Laurent series in the role
of coefficient field of systems of linear ordinary differential systems.
The results of this paper supplement known results on the zero testing problem
and some algorithmically undecidable problems related to differential equations
(see, e.g., [10], [13]).
Computable Infinite Power Series 3
2 Preliminaries
The ring of m × m matrices with entries belonging to a ring R is denoted by
Matm (R). We use the notation [M ]i,∗ , 1 i m, for the 1 × m-matrix which is
the ith row of an m × m-matrix M . The notation M T is used for the transpose
of a matrix (vector) M .
If F is a differential field with derivation ∂ then Const (F ) = {c ∈ F | ∂c = 0}
is the constant field of F .
∂y = Ay, (1)
with A ∈ Matm (K) has a solution space of dimension m in Λm over Const (Λ).
If Const (K) is algebraically closed then there exists a unique (up to a dif-
ferential isomorphism) adequate differential extension Λ such that Const (Λ) =
Const (K) which is called the universal differential field extension of K [18,
Sect. 3.2]. For any differential field K of characteristic 0 there exists a differ-
ential extension whose constant field is algebraically closed. Indeed, this is the
algebraic closure K̄ with the derivation obtained by extending the derivation of
K in the natural way. In this case, Const (K̄) = Const (K) (see [18, Exercises
1.5, 2:(c),(d)]). Existence of the universal differential extension for K̄ implies
that there exists an adequate differential extension for K, i.e., for an arbitrary
differential field of characteristic zero.
In the sequel, we denote by Λ a fixed adequate differential extension of K,
and we suppose that the vector solutions of systems in the form (2) lie in Λm .
In addition to the first-order systems of the form (1), we also consider the
differential systems of arbitrary order r 1. Each of these systems can be
represented, e.g., in the form
Lij ∈ K [∂], i, j = 1, . . . , m, with maxi,j ord Lij = r. We say that the operator
L ∈ Matm (K [∂]) (as well as the system L(y) = 0) is of full rank, if the rows
(Li1 , . . . , Lim ), i = 1, . . . , m, of matrix (4) are linearly independent over K [∂].
The matrix Ar is the leading matrix of both the system L(y) = 0 and operator
L, regardless of representation form.
Let K0 be a subfield of the complex number field C and K be the field K0 ((x))
of formal Laurent series with coefficients in K0 , equipped with the derivation
d
∂ = dx . As it is well known [20, Sect. 110], if K0 is algebraically closed then the
universal differential field extension Λ is the quotient field of the ring generated
by expressions of the form
j = 0, 1, . . . , s.
In fact, system (1) has m linearly independent solutions b1 (x), . . . , bm (x) such
that
bi (x) = ePi (x) xγi Ψi (x), (7)
where the factor ePi (x) xγi is common for all components of bi , and
i = 1, . . . , m.
Suppose that K is a constructive field. Then the proof of the already men-
tioned theorem [9, Thm. 2.2] gives an algorithm for constructing U, L̆. We will
refer to this algorithm as RR (Row-Reduction).
Hence, when the field K is constructive we can apply algorithm RR, and
compute, by Proposition 2, the dimension of the solution space of a given full-
rank system.
Note that in the case when K is the field of rational functions of x over a field
d
of characteristic zero with ∂ = dx , some inequalities close to the formula given
in Proposition 2 can be derived from the results of [12].
6 S.A. Abramov and M.A. Barkatou
0 if u = 0,
dim VL =
1 if u = 0.
This implies that if we have an algorithm for computing the dimension then we
have an algorithm for the zero testing problem.
(b) As we have seen the operator L of the form (9) is unimodular if and only
if u = 0.
Remark 2. The field K|c is smaller than the field K because not every sequence
of coefficients can be represented algorithmically. Indeed, the set of elements of
K|c is countable (each of the algorithms is a finite word in some fixed alphabet)
while the cardinality of the set of elements of K is uncountable.
with elements of K|c , i.e., with computable Laurent series, we cannot compute
a−1 (x) for a given non-zero a(x) ∈ K|c , since the coefficient of x−1 of the
series a (x)a−1 (x) ∈ K|c is equal to val a(x), i.e., is equal to the value that
we are not able to find algorithmically knowing only Ξa . This means that a
suitable representation has to contain some additional information besides a
corresponding algorithm. The value val a(x) cannot close the gap, since we have
no algorithm to compute the valuation of the sum of two series. However, we
can use a lower bound of the valuation instead: observe that if we know that a
series a(x) is non-zero then using a valuation lower bound we can compute the
exact value of val a(x). Thus, we can use as the representation of a(x) ∈ K|c a
pair of the form
(Ξa , μa ), (10)
Observe that if the standard representation form is used for rational functions,
i.e., for elements in K0 (x), then the field K0 (x) is constructive.
For such an operator, we know in advance which of its coefficients are equal to
zero, but we do not know whether the power series u(x) is equal to zero. It is
easy to see that
0 if u(x) = 0,
dim VL =
1 if u(x) = 0.
Proof. (i) This follows from [3,6], as it was explained in the beginning of this
section.
(ii) A given L is unimodular if and only if the system (11) has no non-zero
formal exponential-logarithmic solution, and the claim follows from Proposition
5 (problem (b)).
(iii) A full-rank operator L is evidently unimodular if and only if it has no
regular solution and no exponential-logarithmic solution which is not regular. By
(i), we can test whether the system L(y) = 0 has no regular solution. Thus, if
we are able to test whether this system has no exponential-logarithmic solution
which is not regular then we can test whether L is unimodular or not. However,
this is an undecidable problem by Proposition 5 (problem (b)).
Proposition 7. Let m be an integer number, m 2, K0 be a constructive subset
of C. Let L(y) = 0 be a full-rank system of the form (11), and d = dim VL . Then
VL has a basis b1 (x), . . . , bd (x) consisting of exponential-logarithmic solutions
such that any Ψi (x) from (7) is of the form Ψi (x) = Φi (x1/qi ) where qi is a
non-negative integer,
Φi (x) ∈ ((K1 [[x]]) |c )m [ln x], (12)
and K1 is a simple algebraic extension of K0 . In addition to (12), γi ∈ K1 ,
Pi (x) ∈ K1 [x−1/qi ], i = 1, . . . , d.
Proof. It follows from, e.g., [4,5,8], that for any operator L of full rank there
exists an operator F such that the leading matrix of F L is invertible. The system
F L(y) = 0 is equivalent to a first order system of the form y = Ay, A ∈
Matms (K((x))), s = ord F L. It is known ([7]) that for a first-order system
there exists a simple algebraic extension K1 of K0 such that those γi and the
coefficients of Pi (x) which appear in its solutions of the form (7), belong to K1 .
The field K1 is constructive since K0 is. Obviously, qi ∈ N.
The substitution qi
x = tqi , y(tqi ) = z(t)ePi (t ) ,
Pi (tqi ) ∈ K1 [1/t], into the original system L(y) = 0 transforms it into a full-rank
system which can be represented as
d
L̃(z) = 0, L̃ ∈ Matm (K1 [[t]])|c .
dt
The Laurent series that appear in the regular solutions of this new system can
be taken to be computable, as it follows from [3,6] (see the beginning of this
section).
Thus, the series that appear in the representation of solutions are computable
(Proposition 7), but we cannot find them algorithmically (Proposition 6). In fact,
Proposition 7 guarantees existence. However, the operator F mentioned therein
cannot be constructed algorithmically.
Remark 5. In the case of first-order systems of the form (1), the questions
formulated in Proposition 6(ii, iii) are decidable. This follows from the fact that
Computable Infinite Power Series 11
References
1. Abramov, S.A., Barkatou, M.A.: On the dimension of solution spaces of full rank
linear differential systems. In: Gerdt, V.P., Koepf, W., Mayr, E.W., Vorozhtsov,
E.V. (eds.) CASC 2013. LNCS, vol. 8136, pp. 1–9. Springer, Heidelberg (2013)
2. Abramov, S.A., Barkatou, M.A.: On solution spaces of products of linear differential
or difference operators. ACM Communications in Computer Algebra (accepted)
3. Abramov, S.A., Barkatou, M.A., Khmelnov, D.E.: On full-rank differential systems
with power series coefficients. J. Symbolic Comput. (accepted)
4. Abramov, S.A., Khmelnov, D.E.: Desingularization of leading matrices of systems
of linear ordinary differential equations with polynomial coefficients. In: Inter-
national Conference “Differential Equations and Related Topics” Dedicated to
I.G.Petrovskii, Moscow, MSU, May 30-June 4, p. 5. Book of Abstracts (2011)
5. Abramov, S.A., Khmelnov, D.E.: On singular points of solutions of linear differen-
tial systems with polynomial coefficients. J. Math. Sciences 185(3), 347–359 (2012)
6. Abramov, S.A., Khmelnov, D.E.: Regular solutions of linear differential systems
with power series coefficients. Programming and Computer Software 40(2), 98–106
(2014)
7. Barkatou, M.A.: An algorithm to compute the exponential part of a formal fun-
damental matrix solution of a linear differential system. Applicable Algebra in
Engineering, Communication and Computing 8, 1–23 (1997)
8. Barkatou, M.A., El Bacha, C., Labahn, G., Pflügel, E.: On simultaneous row and
column reduction of higher-order linear differential systems. J. Symbolic Com-
put. 49(1), 45–64 (2013)
9. Beckermann, B., Cheng, H., Labahn, G.: Fraction-free row reduction of matrices
of Ore polynomials. J. Symbolic Comput. 41(5), 513–543 (2006)
10. Denef, J., Lipshitz, L.: Power series solutions of algebraic differential equations.
Math. Ann. 267, 213–238 (1984)
12 S.A. Abramov and M.A. Barkatou
11. Frölich, A., Shepherdson, J.C.: Effective procedures in field theory. Phil. Trans. R.
Soc. Lond. 248(950), 407–432 (1956)
12. Grigoriev, D.: NC solving of a system of linear differential equations in several
unknowns. Theor. Comput. Sci. 157(1), 79–90 (1996)
13. van der Hoeven, J., Shackell, J.R.: Complexity bounds for zero-test algorithms. J.
Symbolic Comput. 41(4), 1004–1020 (2006)
14. Kushner, B.A.: Lectures on Constructive Mathematical Analysis (Translations of
Mathematical Monographs) Amer. Math. Soc. (1984)
15. Lutz, D.A., Schäfke, R.: On the identification and stability of formal invariants for
singular differential equations. Linear Algebra and Its Applications 72, 1–46 (1985)
16. Martin-Löf, P.: Notes on Constructive Mathematics. Almquist & Wiskell, Stokholm
(1970)
17. Pflügel, E.: Effective formal reduction of linear differential systems. Applicable
Algebra in Engineering, Communication and Computation 10(2), 153–187 (2000)
18. van der Put, M., Singer, M.F.: Galois Theory of Linear Differential Equations.
Grundlehren der mathematischen Wissenschaften, vol. 328. Springer, Heidelberg
(2003)
19. Ryabenko, A.: On exponential-logarithmic solutions of linear differential systems
with power series coefficients (In preparation)
20. Schlesinger, L.: Handbuch der Theorie der linearen Differentialgleichungen, vol. 1.
Teubner, Leipzig (1895)
21. Tseitin, G.S.: Mean-value Theorems in Constructive Analysis. Problems of the
Constructive Direction in Mathematics. Part 2. Constructive Mathematical Anal-
ysis. Collection of Articles: Trudy Mat. Inst. Steklov, Acad. Sci. USSR 67, 362–384
(1962)
22. Turing, A.: On computable numbers, with an application to the Entscheidungs-
problem. Proc. London Math. Soc., Series 2 42, 230–265 (1936)
23. Maple online help, http://www.maplesoft.com/support/help/
Relation Algebra, RelView, and Plurality Voting
Rudolf Berghammer
Institut für Informatik, Universität Kiel, Olshausenstraße 40, 24098 Kiel, Germany
rub@informatik.uni-kiel.de
1 Introduction
Since centuries systematic experiments are an accepted means for doing science.
In the meantime they have also become important in formal fields, such as math-
ematics, scientific computing, and theoretical computer science. Here computer
support proved to be very useful, e.g., for computing results and for discover-
ing mathematical relationships by means of visualization and animation. Used
are general Computer Algebra systems, like Maple and Mathematica, but
frequently also tools that focus on specific mathematical objects and (in many
cases: algebraic) structures and particular applications. RelView (cf. [2,12,15])
is a tool of the latter kind. It can be regarded as specific purpose Computer Alge-
bra system and its main purpose is the mechanization of heterogeneous relation
algebra in the sense of [13,14] and the visualization of relations. Computational
tasks can be expressed by relational functions and programs. These frequently
consist of only a few lines that present the relation-algebraic specification of the
notions in question. In view of efficiency the implementation of relations via bi-
nary decision diagrams (BDDs) proved to be superior to many other well-known
implementations, like Boolean matrices or successor lists. Their use in RelView
led to an amazing computational power, in particular if hard problems are to
solve and this is done by the search of a huge set of objects.
In [4,6] it is demonstrated how relation algebra and RelView can be com-
bined to solve computational problems of voting systems. The papers consider
two specific systems, known as approval voting and Condorcet voting. Our paper
constitutes a continuation of this work. We concentrate on plurality voting, a
further popular voting system that simply selects the alternatives with the high-
est number of first place votes and is widely used through the world. After the
presentation of the relation-algebraic preliminaries we model two versions of this
V.P. Gerdt et al. (Eds.): CASC Workshop 2014, LNCS 8660, pp. 13–27, 2014.
c Springer International Publishing Switzerland 2014
14 R. Berghammer
kind of voting within relation algebra. Based on these models, we then present
relation-algebraic specifications for computing dominance relations and (sets of)
winners. Next, for the second model we show how relation-algebraically to solve
hard so-called control problems. For a control of an election the assumption is
that the authority conducting the election – usually called the chair – knows all
individual preferences of the voters and tries to achieve (in case of constructive
control ) or to avoid (in case of destructive control ) win for a specific alternative
by a strategic manipulation of the set of voters and alternatives, respectively.
The chair’s knowledge of all individual preferences and the ability to manipulate
by ‘dirty tricks’ (mistimed meetings, excuses like ‘to expensive’ or ‘legally not
allowed’, etc.) are worst-case assumptions that are not entirely unreasonable in
some settings, for instance, in case of commissions of political institutions. All
relation-algebraic specifications we will present can be transformed immediately
into the programming language of the Computer Algebra system RelView such
that this tool can be used to compute results and to visualize them. Finally, we
evaluate our approach in view of its advantages and its drawbacks.
2 Relation-Algebraic Preliminaries
In this section we recall some preliminaries of (typed, heterogeneous) relation
algebra. For more details, see [13,14], for example.
Given two sets X and Y we write [X ↔ Y ] for the set of all (binary) relations
with source X and target Y , i.e., for the powerset 2X×Y . Furthermore, we write
R : X ↔ Y instead of R ∈ [X ↔ Y ] and call then X ↔ Y the type of R. If the two
sets X and Y of R’s type are finite, then we may consider R as a Boolean matrix.
Since a Boolean matrix interpretation is well suited for many purposes and also
used by the RelView tool as the main possibility to visualize relations, in the
present paper we frequently use matrix terminology and notation. In particular,
we speak about the entries, rows and columns of a relation/matrix and write
Rx,y instead of (x, y) ∈ R or x R y. We assume the reader to be familiar with the
basic operations on relations, viz. RT (transposition), R (complement), R ∪ S
(union), R ∩ S (intersection), and R; S (composition), the predicates R ⊆ S
(inclusion) and R = S (equality), and the special relations O (empty relation), L
(universal relation), and I (identity relation). In case of O, L, and I we overload
the symbols, i.e., avoid the binding of types to them.
T
By syq(R, S) := RT ; S ∩ R ; S we define the symmetric quotient of two re-
lations R : X ↔ Y and S : X ↔ Z. From this definition we get the typing
syq(R, S) : Y ↔ Z and that for all y ∈ Y and z ∈ Z it holds
syq(R, S)y,z ⇔ ∀ x ∈ X : Rx,y ↔ Sx,z . (1)
Descriptions of the form (1) with relationships expressed by logical formulae and
indices are called point-wise. Such descriptions of symmetric quotients and the
constructs we introduce now for modeling sets and direct products will play a
fundamental role in the remainder of the paper. We will use them to get relation-
algebraic (or point-free) specifications, which then can be treated by RelView.
Relation Algebra, RelView, and Plurality Voting 15
Vectors are a first well-known relational means to model sets. For the purpose
of this paper it suffices to define them as relations r : X ↔ 1 (we prefer lower
case letters in this context) with a specific singleton set 1 = {⊥} as target.
Then relational vectors can be considered as Boolean column vectors. To be
consonant with the usual notation, we always omit the second subscript, i.e.,
write rx instead of rx,⊥ . Given r : X ↔ 1 and Y ∈ 2X we then define that
r models the subset Y of X :⇔ ∀ x ∈ X : x ∈ Y ↔ rx . (2)
A point is a specific vector with precisely one 1-entry if considered as a Boolean
column vector. Consequently, it models a singleton subset {x} of X and we then
say that it models the element x of X. In conjunction with a powerset 2X we
will also use the membership relation M : X ↔ 2X , point-wisely defined by
Mx,Y :⇔ x ∈ Y, (3)
for all x ∈ X and Y ∈ 2 . The following lemma shows how symmetric quotients
X
X the specific sets {x ∈ X | Rx,y } of Lemma 2.1 w.r.t. their sizes within the
language of relation algebra.
Lemma 2.2. Let relations Q : X ↔ Y and R : X ↔ Z be given, where X is
finite, and S : 2X ↔ 2X be a size-comparison relation. Furthermore, let
scomp(Q, R) := set(Q); ST ; set(R)T : Y ↔ Z.
Then we have scomp(Q, R)y,z iff |{x ∈ X | Qx,y }| ≥ |{x ∈ X | Rx,z }|, for all
y ∈ Y and z ∈ Z.
16 R. Berghammer
scomp(Q, R)y,z
⇔ (set(Q); ST ; set(R)T )y,z
⇔ ∃ U ∈ 2X : set(Q)y,U ∧ ∃ V ∈ 2X : ST T
U,V ∧ set(R)V,z
⇔ ∃ U ∈ 2X : U = {x ∈ X | Qx,y } ∧
∃ V ∈ 2X : |U | ≥ |V | ∧ V = {x ∈ X | Rx,z } Lemma 2.1, (4)
⇔ |{x ∈ X | Qx,y }| ≥ |{x ∈ X | Rx,z }|
for all z ∈ Z and u ∈ X×Y . Using the projection relations π : X×Y ↔ X and
ρ : X×Y ↔ Y , we also are able to define functions which establish an isomor-
phism between the Boolean lattices [X ↔ Y ] and [X×Y ↔ 1 ]. The direction from
[X ↔ Y ] to [X×Y ↔ 1 ] is given by the function vec, with vec(R) := (π; R ∩ρ); L,
and that from [X×Y ↔ 1 ] back to [X ↔ Y ] by the inverse function rel, with
rel(v) := π T ; (ρ∩v; LT ). In point-wise descriptions the definitions say that Ru1 ,u2
iff vec(R)u and vu iff rel(v)u1 ,u2 , for all u ∈ X×Y .
As from now we assume all constructions of Section 2 to be at hand. Except
the functions set, scomp, vec and rel they are available in the programming
language of RelView as or via pre-defined operations and the implementing
BDDs of relations are comparatively small. For example, a size-comparison re-
lation S : 2X ↔ 2X can be implemented by a BDD with O(|X|2 ) nodes and for
a membership relation M : X ↔ 2X even O(|X|) nodes suffice (for details, see
[11,12]). An implementation of set, scomp, vec, and rel in the tool is nothing else
than the translation of their relation-algebraic definitions into RelView-code;
the RelView-programs for set and scomp can be found in Section 5.
The accents of the Homeric hexameter are the soft rustle of a leaf
in the midday sun, the rhythm of matter; but the “Stabreim” likes
“potential energy” in the world-pictures of modern physics, creates a
tense restraint in the void without limits, distant night-storms above
the highest peaks. In its swaying indefiniteness all words and things
dissolve themselves—it is the dynamics, not the statics, of language.
The same applies to the grave rhythm of Media vita in morte sumus.
Here is heralded the colour of Rembrandt and the instrumentation of
Beethoven—here infinite solitude is felt as the home of the Faustian
soul. What is Valhalla? Unknown to the Germans of the Migrations
and even to the Merovingian Age, it was conceived by the nascent
Faustian soul. It was conceived, no doubt, under Classic-pagan and
Arabian-Christian impressions, for the antique and the sacred
writings, the ruins and mosaics and miniatures, the cults and rites
and dogmas of these past Cultures reached into the new life at all
points. And yet, this Valhalla is something beyond all sensible
actualities floating in remote, dim, Faustian regions. Olympus rests
on the homely Greek soil, the Paradise of the Fathers is a magic
garden somewhere in the Universe, but Valhalla is nowhere. Lost in
the limitless, it appears with its inharmonious gods and heroes the
supreme symbol of solitude. Siegfried, Parzeval, Tristan, Hamlet,
Faust are the loneliest heroes in all the Cultures. Read the wondrous
awakening of the inner life in Wolfram’s Parzeval. The longing for the
woods, the mysterious compassion, the ineffable sense of
forsakenness—it is all Faustian and only Faustian. Every one of us
knows it. The motive returns with all its profundity in the Easter
scene of Faust I.