Symbolic-Numeric Tools for Multivariate Asymptotics

Transcription

1 Symbolic-Numeric Tools for Multivariate Asymptotics Bruno Salvy Joint work with Stephen Melczer to appear in Proc. ISSAC 2016 FastRelax meeting May 25, 2016

2 Combinatorics, Randomness and Analysis From simple local rules, a global structure arises. A quest for universality in random discrete structures: probabilistic complexity of structures and algorithms. Quantitative results using complex analysis. 2 / 26

3 Overview Ex.: binary trees Equations over combinatorial structures Generating functions 1X F (z) = f n z n n=0 Complex analysis f n, n!1 Aim: several vars, automatically. B = Z [ B B B(z) =z + B(z) 2 B n 4n 1 n 3/2 p 3 / 26

4 I. Families of Generating Functions 4 / 26

5 Rational Generating Functions (regular languages, linear recurrences with const. coeffs.) Example Generating function of 1,2,4,8, is z Example 1 1 z z 2 Fibonacci numbers 1,1,2,3,5,8, have gf. Example (Simple lattice walks) F (x, y) = 1 1 x y = X i,j c i,j x i y j = X i,j i + j Here c i,j counts the number of ways to walk from (0,0) to (i,j) using i steps (0,1) and j steps (1,0). i x i y j =1+x +2xy + x 2 + y 2 + x 3 + y 3 +3x 2 y +3xy 2 +6x 2 y / 26

6 Rational Generating Functions (regular languages, linear recurrences with const. coeffs.) Example (Restricted factors in words) Here c i,j counts the number of binary words with i zeroes and j ones that do not contain or / 26

7 Algebraic Generating Functions Plane trees with n nodes and given arity set; Bracketings of a word of length n; Paths from (0,0) to (n,0) using steps (1,k) and (1,-1); Dissections of a convex n-gon (x(xxx)) ~~~~~ ~~~~~ 7 / 26

8 Diagonals of Rational Functions Input: Rational function z z z with power series z iz i Goal: Asymptotics of the diagonal sequence i N as Example (Apéry) F (a, b, c, z) = 1 1 z(1 + a)(1 + b)(a + c)(1 + b + c + bc + abc) Here determines Apéry s sequence, related to his celebrated proof of the irrationality of.

9 Univariate Generating Functions D-FINITE Aim: asymptotics of coefficients of diagonals, automatically. DIAGONAL ALGEBRAIC RATIONAL

10 II. Analytic Combinatorics 10 / 26

11 Cauchys formula Thm. If f = X i 1,...,i n 0 neighborhood of 0, then c i1,...,i z i 1 n 1 z i n n is convergent in the c i1,...,i n = 1 2 i n Z T f(z 1,...,z n ) dz 1 dz n z i 1+1 z i n+1 1 n for any small torus T ( z j = re i j ) around 0. Asymptotics: deform the torus to pass where the integral concentrates asymptotically. 11 / 26

12 Coefficients of Univariate Rational Functions a n = 1 2 i Z f(z) z n+1 dz F 1 =1= 1 2 i I 1 dz 1 z z 2 z 2 = = As n increases, the smallest singularities dominate. F n = n n / 26

13 Conway s sequence 1,11,21,1211,111221, Generating function for lengths: f(z)=p(z)/q(z) with deg Q=72. Smallest singularity: δ(f) ρ=1/δ(f) n ρ n ρ Res(f,δ(f)) remainder exponentially small 13 / 26

14 Coefficients of Multivariate Rational Functions Def. F(z 1,,z n ) is combinatorial if every coefficient is 0. Assuming F=G/H is combinatorial + generic conditions, asymptotics are determined by the points in such that z z z partial derivatives z z Critical Points z z is coordinate-wise minimal in Aim: find these points, (1) is an algebraic condition. (2) is a semi-algebraic condition (can be expensive). in good complexity.

15 III. Symbolic-Numeric Computation 15 / 26

16 Kronecker Representation Algebraic part: ``compute the solutions of the system Suppose z z z Under genericity assumptions, in algorithm to find: bit ops there is a prob. History and Background: see Castro, Pardo, Hägele, and Morais (2001) Giusti, Lecerf, Salvy (2001) Schost (2001) System reduced to a univariate polynomial.

17 Example (Lattice Path Model) The number of walks from the origin taking steps {NW,NE,SE,SW} and staying in the first quadrant has One can calculate the Kronecker representation Next, find the minimal critical points, that lie on the boundary of the domain of convergence. so that the critical points are given by:

18 Testing Minimality Semi-algebraic part: In the combinatorial case, it is enough to examine only the positive real points on the variety to determine minimality. Thus, we add the equation for a new variable and select the positive real point(s) z with no from a new Kronecker representation: P (v) =0 P 0 (v)z 1 Q 1 (v) =0 P 0 (v)z n Q n (v) =0 P 0 (v)t Q t (v) =0.. This is done numerically, with enough precision. Fast univariate solving: Sagraloff and Mehlhorn (JSC 2016)

19 How many digits do we need? Suppose are roots of square-free poly of Then For the Kronecker Representation, we have polynomials of If so we need a precision are the of positive real roots to rigorously of then decide the roots signs, with find modulus exact zeroes, can be and found identify using equal coordinates. precision

20 First Complexity Results for ACSV Theorem (M. and Salvy, 2016) Under generic conditions, and assuming z is combinatorial, the points contributing to dominant diagonal asymptotics can be determined in bit operations. Each contribution has the form and can be found to precision in bit ops. The genericity assumptions heavily restrict the form of the asymptotic growth. Removing more of these assumptions is ongoing work.

21 Example 1 Example (Apéry) F (a, b, c, z) = 1 1 z(1 + a)(1 + b)(a + c)(1 + b + c + bc + abc) Here a Kronecker representation is There are two real critical points, and one is positive. After testing minimality, one has proven asymptotics

22 Example 2 Example (Restricted Words in Factors) Using the Kronecker representation gives where are polynomials with largest coeff around There are two points with positive coordinates, adding allows one to show which is minimal, and that

23 Conclusion We Give the first complexity bounds for methods in analytic combinatorics in several variables Combine strong symbolic results on the Kronecker representation with fast algorithms on univariate polynomials Work in progress: extend beyond some of the assumptions.