analytic algorithms for null cone membership
play

Analytic algorithms for null cone membership Ankit Garg Microsoft - PowerPoint PPT Presentation

Analytic algorithms for null cone membership Ankit Garg Microsoft Research India July 12, 2019 Overview Null cone membership: fundamental problem in invariant theory . Connections to several areas of computer science, mathematics and


  1. Analytic algorithms for null cone membership Ankit Garg Microsoft Research India July 12, 2019

  2. Overview � Null cone membership: fundamental problem in invariant theory . � Connections to several areas of computer science, mathematics and physics. Geometric complexity theory – asymptotic Quantum information theory– one-body vanishing of Kronecker coefficients. quantum marginal problem. Functional analysis – Brascamp-Lieb Optimization– Geodesic convexity. Captures inequalities. general linear programming. Complexity theory and derandomization – Special cases of polynomial identity testing. � Analytic algorithms for algebraic problems. � Non-convex optimization problems but geodesically convex .

  3. Outline � Invariant theory and null cone � Geometric invariant theory � Algorithms: a sample � Open problems

  4. Invariant theory and null cone

  5. Linear actions of groups � . Group acts linearly on vector space group homomorphism. invertible linear map . � � � and . � � � � � � �� Example � by permuting coordinates . � acts on �(�) . � � � �(�) Example acts on by conjugation . � � �� . �

  6. Objects of study Group acts linearly on vector space . Invariant polynomials: Polynomial functions on • invariant under action of . s.t. for all . • Orbits: Orbit of vector , . • Orbit-closures: Orbits may not be closed. Take their closures. Orbit-closure of vector .

  7. Null cone Group acts linearly on vector space . Null cone: Vectors s.t. lies in the orbit-closure of . � . Sequence of group elements � s.t. . � � � �→� Problem: Given , decide if it is in the null cone. Captures many interesting questions. [Hilbert 1893; Mumford 1965]: in null cone iff for all homogeneous invariant polynomials . � One direction clear (polynomials are continuous). � Other direction uses Nullstellansatz and some algebraic geometry.

  8. Example acts on by permuting coordinates. . Null cone = . No closures.

  9. Example acts on by conjugation . . � Invariants: generated by . � Null cone: nilpotent matrices.

  10. Example acts on by left-right multiplication. . • Invariants: generated by . • Null cone: Singular matrices.

  11. Example : acts on by left-right multiplication. . • Invariants: generated by . • Null cone: perfect matching. is in null cone iff has no perfect matching.

  12. Example : Linear programming : acts on by scaling variables. , . . . . Null cone Linear Programming not in null cone conv . In null cone (membership) problem is a non-commutative analogue of linear programming .

  13. Example acts on by simultaneous left-right multiplication. . � Invariants [DW , DZ , SdB , ANS ]: generated by . � Null cone: Non-commutative singularity. Captures non- commutative rational identity testing. [GGOW , DM , IQS ]: Deterministic polynomial time algorithms.

  14. Geometric invariant theory: certification of null cone

  15. GIT: computational perspective What is complexity of null cone membership? GIT puts it in (morally). � Hilbert-Mumford criterion: how to certify membership in null cone. � Kempf-Ness theorem: how to certify non- membership in null cone.

  16. Kempf-Ness Group acts linearly on vector space . How to certify not in null cone? Exhibit invariant polynomial s.t. . Not feasible in general. Invariants hard to find, high degree, high complexity etc. Kempf-Ness provides another (efficient) way.

  17. An optimization perspective Finding minimal norm elements in orbit-closures! Group acts linearly on vector space . . Null cone: s.t. .

  18. Moment map Group acts linearly on vector space . . Moment map : gradient of at . How much norm of decreases by infinitesimal action around . Much more general. Moment momentum . Fundamental in symplectic geometry and physics .

  19. Example acts on . . . Moment map: consider action of , . .

  20. Example acts on . . � � : . Directional derivative: action of , . , s.t. � � vectors of row and column norms of . .

  21. Kempf-Ness Group acts linearly on vector space . [Kempf, Ness 79]: not in null cone iff non-zero in orbit- closure of s.t. . certifies not in null cone. One direction easy. not in null cone. Take vector of minimal norm in orbit- � closure of . non-zero. minimal norm in its orbit. Norm does not decrease by � infinitesimal action around . . � Global minimum local minimum.

  22. Kempf-Ness Other direction: local minimum global . Some “ convexity” . � Commutative group actions – Euclidean convexity (change of variables) [exercise]. � Non-commutative group actions: geodesic convexity .

  23. Algorithms

  24. Algorithms Group acts linearly on vector space . � . � � �∈� Lots of work on Euclidean convex optimization. Few algorithms for geodesically convex case. � First order: gradient descent, alternating minimization. � Second order: Box constrained Newton’s method. No known generalization of interior point methods, ellipsoid method.

  25. Alternating minimization Widely used heuristic in machine learning and optimization. Optimizing over several variables or constraints. Optimizing/satisfying over an individual variable or constraint easy . Alternately optimize/satisfy over variables/constraints. Lot of work on understanding conditions for convergence and convergence rates. Very few cases in which provably converge in small number of iterations.

  26. Tensor scaling acts on naturally. Tensor . : require tristochasticity. Slices orthogonal (in all directions).

  27. Tensor scaling: alternating minimization Operations: Group action i.e. basis change (in all directions). Algorithm: Alternate basis change for steps. [BGOWW 17]: not in null cone -convergence in steps. GIT: in null cone no convergence. time algorithm solves null cone membership. � � For some problems (e.g. left-right action ) suffices [GGOW 16]. � Second order algorithm gets run time for such cases (e.g. left-right action ) [AGLOW 18, BFGOWW 19].

  28. Analysis using invariants Potential function: invariant polynomial . � Homogeneous degree . � �/� � �/� � �/� . � � � � not in null cone invariant s.t. . �/� . -step Analysis • [Lower bound]: Initially . [Progress per step]: If -far from tristochasticity, one step increases by a • factor of . Consequence of a robust AM-GM inequality and invariance. • [Upper bound]: . ��� � steps suffice. [BGOWW ]: � . �� . � ��

  29. Open problems

  30. Open problems � Null cone membership in ? � Polynomial time algorithms for null cone membership ? � Ellipsoid/interior point methods for geodesically convex problems. running time. � More applications ?

  31. Lectures and videos � IAS workshop videos: https://www.math.ias.edu/ocit2018 � Avi’s CCC tutorial: http://computationalcomplexity.org/Archive/2017/tutorial.p hp

  32. Thank You

Recommend


More recommend