combinatorialadventures inanalysis,algebra, andtopologycombinatorialadventures inanalysis,algebra,...
TRANSCRIPT
Combinatorial Adventuresin Analysis, Algebra,and TopologyStephen Melczer, Marni Mishna,and Robin Pemantle
The authors of this piece are organizers of the AMS2020 Mathematics Research Communities summer
conference Combinatorial Applications ofComputational Geometry and Algebraic Topology,one of five topical research conferences offered thisyear that are focused on collaborative research and
professional development for early-careermathematicians. Researchers with experience in oneor more of the areas of combinatorics, computationalalgebra, analysis, algebraic topology, and singularitytheory are welcomed. Additional information can be
found at https://www.ams.org/programs/research-communities/2020MRC-CompGeom.Applications are open until February 15, 2020.
Capturing large-scale behavior of combinatorial objectscan be difficult but very revealing. Consider Figure 1, forinstance, which illustrates the behavior of random objectsof large size in two combinatorial classes inspired by quan-tum computing and statistical mechanics.
Stephen Melczer is a CRM-ISM postdoctoral fellow at the Universite du QuebecΓ Montreal. His email address is [email protected] Mishna is a professor of mathematics at Simon Fraser University. Heremail address is [email protected] Pemantle is the Merriam Term Professor of Mathematics at the Universityof Pennsylvania. His email address is [email protected].
For permission to reprint this article, please contact:[email protected].
DOI: https://doi.org/10.1090/noti2031
Figure 1. ACSV has been used to determine the shape of largerandom quantum random walks [3] (left) and cube groves [5](right).
Perhaps surprisingly, complex analysis, algebraic ge-ometry, topology, and computer algebra are importantingredients in such results, combining to form the rela-tively new field of analytic combinatorics in several variables(ACSV). ACSV builds on classic generating function tech-niques in combinatorics to provide new tools for the studyof sequences and discrete objects. Applications previouslystudied in the literature include queuing systems, bioinfor-matics, data structures, random tilings, special functions,and integrable systems. Ongoing research efforts from thecombinatorial side concern an evolving study of latticepaths [8] and objects from representation theory and al-gebraic combinatorics [14].
The purpose of this note is to give a brief overview ofACSV and its history to drive interest among those in anyof the diverse areas touched by the topic. A large collectionof worked examples can be found in [15] and a detailedtreatment of ACSV in [16].
262 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 67, NUMBER 2
Analytic CombinatoricsThe use of analytic techniques in combinatorics and prob-ability theory has a long history, dating back to theeighteenth-century work of Laplace [12] and contempo-raries such as Stirling [18] and de Moivre [9]. In mod-ern times, the use of real and complex analysis to deriveasymptotic behavior is the domain of analytic combina-torics [11], a field which finds application in many areas ofmathematics, computer science, and the natural sciences.The key idea behind such results is that important prop-erties of a sequence (ππ) = (π0, π1, β¦ ) of real or complexnumbers are easily deduced from the generating function
πΉ(π§) = βπβ₯0
πππ§π = π0 + π1π§ +β― .
Although much can be deduced treating πΉ only as a for-mal series [20], when ππ grows at most exponentially as πgoes to infinity then πΉ defines a complex-analytic functionat the origin. The powerful methods of complex analysiscan then be used to study properties of ππ, including itsasymptotic behavior.
Indeed, the Cauchy integral formula implies that for ev-ery natural number π the coefficient ππ can be representedby a contour integral
ππ =12ππ β«π
πΉ(π§)π§βπβ1ππ§,
where π is a positively oriented circle sufficiently close tothe origin. The singularities of the generating function playa crucial role in determining sequence asymptotics: the do-main of integration π can be deformed without changingthe value of the Cauchy integral as long it does not crossany singularities of πΉ. Setting some technicalities aside,each singularity of πΉ gives a βcontributionβ to asymptoticsof ππ depending on its location in the complex plane andthe local behavior of πΉ near the singularity. This approachis powerful enough to be completely automated for largeclasses of generating functions, such as rational and al-gebraic functions, and is flexible enough to extend as re-quired by applications. A thorough development of thisunivariate theory and a survey of its applications can befound in the fantastic text of Flajolet and Sedgewick [11].
ACSVIn contrast to the well-developed univariate theory, untilrecently there was no general framework for the study ofmultivariate generating functions. To rectify this gap in theliterature, over the last two decades the theory of ACSV hasbeen developed [16]. Here we focus on a rational functionπΉ(π³) = πΉ(π§1, β¦ , π§π) in π > 1 variables with power series
expansion
πΉ(π³) = βπ’ββπ
ππ’π³π’ = βπ’ββπ
ππ1,β¦,πππ§π11 β―π§πππ .
Given π« β βπ, the π«-diagonal of πΉ(π³) is the sequence (πππ«).Manymathematical sequences arise naturally as diagonalsof explicit rational functions. Furthermore, coefficient se-ries for all algebraic functions may be embedded as gen-eralized diagonals of rational functions [17]. It is conjec-tured that the same is true of all globally bounded D-finitefunctions [7, Conjecture 4].
Example 1. A key step in Aperyβs proof [19] of the ir-rationality of π(3) is determining asymptotics of the se-quence
ππ =πβπ=0
(ππ)2
(π + ππ )
2
.
Because it is a sum of nonnegative terms, saddle point ap-proximations will produce the desired asymptotics.
Such sums with mixed signs are notoriously difficult toestimate. However, any binomial sum sequence regardlessof signs can automatically be written as the π-diagonal ofa rational function [6]. For instance, Aperyβs sum is theπ-diagonal of the four-variable rational function
11 β π‘(1 + π₯)(1 + π¦)(1 + π§)(1 + π¦ + π§ + π¦π§ + π₯π¦π§) .
The methods of ACSV determine asymptotics of ππ com-pletely automatically [13].
Although the π«-diagonal of a fixed rational function isdefined only when π« has integer coordinates, ACSV showsthat asymptotics along most1 directions π« β βπ
>0 can bewell defined by a limiting procedure.
Multivariate Analysis and Singularity TheoryAs in the univariate setting, the methods of ACSV startfrom a Cauchy integral formula
πππ« =1
(2ππ)π β«ππΉ(π³)π³βππ«βπππ§1β―ππ§π ,
where π is now a product of positively oriented circles suf-ficiently close to the origin. The singularities of πΉ(π³) againplay a crucial role in asymptotics. If πΉ(π³) = πΊ(π³)/π»(π³) is arational function with coprime numerator πΊ and denomi-nator π», then the singularities of πΉ form the set π± = {π³ ββπ βΆ π»(π³) = 0}.
Once the dimension is greater than 1, the singular setπ±is no longer discrete but is an algebraic variety of positivedimension. In the geometrically simplest case, when π±1Asymptotics of the π«-diagonal vary smoothly with π« inside open cones of βπ
>0.The boundaries of these cones form (π β 1)-dimensional sets where asymptoticscan sharply change.
FEBRUARY 2020 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 263
forms a smooth manifold, one searches for a (genericallyfinite) set of critical points near which the Cauchy integralcan be approximated by saddle-point integrals which areeasy to estimate asymptotically. When some of these crit-ical points lie on the boundary of the generating functiondomain of convergence and other mild conditions hold,asymptotics can be computed using explicit formulas. Inthe absence of critical points on the boundary of conver-gence or when the geometry of this domain is difficult toestablish, topological techniques may be applied to deter-mine which critical points are responsible for the coeffi-cient asymptotics. This was carried out in two dimensionsin [10]; generalizing to higher dimensions is an open prob-lem.
Without restricting the geometry of π±, recent work [4]has shown how stratified Morse theory helps determinethe deformations of π inβπ⧡π± which yield integrals of thetype analyzed in classic works on singularity theory [1, 2].The techniques of ACSV thus rely on an interesting mixof computer algebra, singularity theory, algebra, geome-try, and topology. By asking different questions, ACSV pro-motes new work in these areas. For example, any advancesin the automatic computation of singular integrals can beapplied back to the wavelike partial differential equationsfor which purpose singularity theory was developed in the1970s and 1980s. Indeed this developmentmay already beripe for a resurgence as it couples with modern computeralgebra methods.
AMS Math Research CommunityBeing a new and rapidly growing field which incorporatestechniques from diverse corners ofmathematics, ACSV hasa rich collection of problems accessible to early researchersin many areas. Some of these problems are computationalin nature, including the development of computer algebratools for asymptotics under varying assumptions. Othersare topological at heart, such as computing intersectionand linking numbers of attachment cycles which arise inthe Morse theoretic constructions discussed above. Manyrelate to asymptotic computations of saddle-point asymp-totics in degenerate settings which appear in combinato-rial applications. Of course, after developing such toolsthere is also a need for combinatorialists to apply them toproblems of differing scope. Most of the advances in ACSVhave come from combinatorial problems just beyond thereach of existing techniques.
We thus encourage early-career mathematicians with ex-perience in any one of the areas of combinatorics, com-putational algebra, analysis, algebraic topology, singular-ity theory, or hyperbolic partial differential equations toapply to our upcoming Math Research Community. Ourhope is to have participants from a wide variety of fields
interacting with each other to develop this exciting area ofmathematics.
References[1] Arnolβ²d VI, GuseΔ±n-Zade SM, Varchenko AN. Singularities
of Differentiable Maps. Vol. II, Monodromy and asymptoticsof integrals, translated from the Russian by Hugh Porteous,translation revised by the authors and James Montaldi,Monographs in Mathematics, vol. 83, BirkhΓ€user Boston,Inc., Boston, MA, 1988. MR966191
[2] Atiyah MF, Bott R, Garding L. Lacunas for hyperbolicdifferential operators with constant coefficients. I, ActaMath. (124):109β189, 1970, DOI 10.1007/BF02394570.MR470499
[3] Baryshnikov Y, Brady W, Bressler A, Pemantle R.Two-dimensional quantum random walk, J. Stat. Phys.,no. 1 (142):78β107, 2011, DOI 10.1007/s10955-010-0098-2. MR2749710
[4] Baryshnikov Y, Melczer S, Pemantle R. Critical pointsat infinity for analytic combinatorics, Submitted, arxiv.org/abs/1905.05250, 2019.
[5] Baryshnikov Y, Pemantle R. Asymptotics of multivariatesequences, part III: Quadratic points, Adv. Math., no. 6(228):3127β3206, 2011, DOI 10.1016/j.aim.2011.08.004.MR2844940
[6] Bostan A, Lairez P, Salvy B. Multiple binomial sums,J. Symbolic Comput., part 2 (80):351β386, 2017, DOI10.1016/j.jsc.2016.04.002. MR3574517
[7] Christol G. Globally bounded solutions of differen-tial equations, Analytic Number Theory (Tokyo, 1988),Springer, Berlin, 1990, 45β64, DOI 10.1007/BFb0097124MR1071744
[8] Courtiel J, Melczer S, Mishna M, Raschel K. Weighted lat-tice walks and universality classes, J. Combin. Theory Ser.A (152):255β302, 2017, DOI 10.1016/j.jcta.2017.06.008.MR3682734
[9] deMoivre A.Miscellanea analytica de seriebus et quadraturis,J. Tomson and J. Watts, London, 1730.
[10] DeVries T, van der Hoeven J, Pemantle R. Effectiveasymptotics for smooth bivariate generating functions,On-line J. Anal. Comb. (7):24, 2012.
[11] Flajolet P, Sedgewick R. Analytic Combinatorics, Cam-bridge University Press, Cambridge, 2009. MR2483235
[12] Laplace P-S. Memoire sur la probabilite des causespar les evΓ¨nemens, Memoires de mathematique et dephysique (Tome VI):621β656, 1774. gallica.bnf.fr/ark:/12148/bpt6k3482c/f685
[13] Melczer S, Salvy B. Effective coefficient asymptotics ofmultivariate rational functions via semi-numerical algo-rithms for polynomial systems, arxiv.org, 2019.
[14] Mishna M, Rosas M, Sundaram S. An elementary ap-proach to the quasipolynomiality of the kronecker coeffi-cients, https://arxiv.org/abs/1811.10015, 2018.
[15] Pemantle R, Wilson MC. Twenty combinatorial exam-ples of asymptotics derived from multivariate generatingfunctions, SIAM Rev., no. 2 (50):199β272, 2008, DOI10.1137/050643866. MR2403050
264 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 67, NUMBER 2
[16] Pemantle R,WilsonMC.Analytic Combinatorics in SeveralVariables, Cambridge Studies in Advanced Mathematics,vol. 140, Cambridge University Press, Cambridge, 2013.MR3088495
[17] Safonov KV. On power series of algebraic and rationalfunctions in ππ, J. Math. Anal. Appl., no. 2 (243):261β277,2000, DOI 10.1006/jmaa.1999.6667. MR1741523
[18] Stirling J. Methodus differentialis: sive tractatus de summa-tione et interpolatione serierum infinitarium, London, 1730.
[19] van der Poorten A. A proof that Euler missedβ¦Aperyβsproof of the irrationality of π(3), An informal report,Math. Intelligencer, no. 4 (1):195β203, 1978/79, DOI10.1007/BF03028234. MR547748
[20] Wilf HS. generatingfunctionology, 3rd ed., A K Peters, Ltd.,Wellesley, MA, 2006. MR2172781
Stephen Melczer Marni Mishna
Robin Pemantle
Credits
Figure 1 and photo of Robin Pemantle are courtesy of RobinPemantle.
Photo of Stephen Melczer was taken by Celia Laur.Photo of Marni Mishna is courtesy of Marni Mishna. www.ams.org/authors
AMS AUTHOR RESOURCE CENTERThe Author Resource Center is a collec-tion of information and tools available to assist you to successfully write, edit, illustrate, and publish your mathematical works.
To begin utilizing these important resources, visit:
FEBRUARY 2020 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 265