condition numbers in nonarchimedean semidefinite
play

Condition numbers in nonarchimedean semidefinite programming . . . - PowerPoint PPT Presentation

Condition numbers in nonarchimedean semidefinite programming . . . and what they say about stochastic mean payoff games Xavier Allamigeon, St ephane Gaubert, Ricardo Katz, Mateusz Skomra INRIA and CMAP, Ecole polytechnique, CNRS January


  1. Definition (SDFP over Puiseux series) Given symmetric matrices Q (0) , Q (1) , . . . , Q ( n ) , denote Q ( x ) = Q (0) + x 1 Q (1) + · · · + x n Q ( n ) . Decide if the following spectrahedron is empty S = { x ∈ K n � 0 : Q ( x ) is positive semidefinite } Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 6 / 58

  2. Definition (SDFP over Puiseux series) Given symmetric matrices Q (0) , Q (1) , . . . , Q ( n ) , denote Q ( x ) = Q (0) + x 1 Q (1) + · · · + x n Q ( n ) . Decide if the following spectrahedron is empty S = { x ∈ K n � 0 : Q ( x ) is positive semidefinite } Proposition S � = ∅ iff for all t large enough, the following real spectrahedron is non-empty � 0 : Q (0) ( t )+ x 1 Q (1) ( t )+ · · · + x n Q ( n ) ( t ) is pos. semidef. } S ( t ) = { x ∈ R n Proof. K is the field of germs of univariate functions definable in a o-minimal structure. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 6 / 58

  3. Theorem (Allamigeon, SG, Skomra) There is a correspondence between nonarchimedean semidefinite programming problems and zero-sum stochastic games with perfect information. If the valuations of the matrices Q ( i ) are generic, feasibility holds iff Player Max wins the game. X. Allamigeon, S. Gaubert, and M. Skomra. “Solving Generic Nonarchimedean Semidefinite Programs Using Stochastic Game Algorithms”. In: Journal of Symbolic Computation 85 (2018), pp. 25–54. doi : 10.1016/j.jsc.2017.07.002 . eprint: 1603.06916 . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 7 / 58

  4. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 x 3 x 3 − 5 / 4 − 3 / 4 1 − 3 / 4 0 3 2 9 / 4 9 / 4 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  5. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 3 2 9 / 4 9 / 4 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  6. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  7. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  8. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  9. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  10. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  11. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  12. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  13. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 − 1 0 x 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  14. Take the spectrahedral cone − t 3 / 4 x 3   t x 3 − x 1 t − 1 x 1 + t − 5 / 4 x 3 − x 2  � 0 . Q ( x ) := − x 1 − x 3  − t 3 / 4 x 3 t 9 / 4 x 2 − x 3 We associate with Q ( x ) a stochastic game with x 3 x 3 perfect information. − 5 / 4 − 3 / 4 1 − 3 / 4 0 Circles: Min plays, Square: Max plays, Bullet: Nature flips coin, Payments made 3 2 9 / 4 9 / 4 by Min to Max 1 0 0 x 2 − 1 Max is winning implies that − 1 0 the cone is nontrivial, and x 1 yields a feasible point ( t 1 . 06 , t 0 . 02 , t 1 . 13 ). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 8 / 58

  15. Benchmark We tested our method on randomly chosen matrices Q (1) , . . . , Q ( n ) ∈ K m × m with positive entries on diagonals and no zero entries. We used the value iteration algorithm. ( n , m ) (50 , 10) (50 , 40) (50 , 50) (50 , 100) (50 , 1000) time 0.000065 0.000049 0.000077 0.000279 0.026802 ( n , m ) (100 , 10) (100 , 15) (100 , 80) (100 , 100) (100 , 1000) time 0.000025 0.000270 0.000366 0.000656 0.053944 ( n , m ) (1000 , 10) (1000 , 50) (1000 , 100) (1000 , 200) (1000 , 500) time 0.000233 0.073544 0.015305 0.027762 0.148714 ( n , m ) (2000 , 10) (2000 , 70) (2000 , 100) (10000 , 150) (10000 , 400) time 0.000487 1.852221 0.087536 19.919844 2.309174 Table: Execution time (in sec.) of Procedure CheckFeasibility on random instances. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 9 / 58

  16. Experimental phase transition for random nonarchimedean SDP n = # variables, m = size matrices 60 100 % 80 % 40 60 % m 40 % 20 20 % 0 % 200 400 600 800 1 , 000 n Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 10 / 58

  17. The present work on tropical condition numbers grew to explain this picture. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 11 / 58

  18. Valuation of Puiseux series ∞ � c k t α k x = x ( t ) = k =1 log | x ( t ) | val( x ) = lim = α 1 (and val(0) = −∞ ) . log t t →∞ Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 12 / 58

  19. Valuation of Puiseux series ∞ � c k t α k x = x ( t ) = k =1 log | x ( t ) | val( x ) = lim = α 1 (and val(0) = −∞ ) . log t t →∞ Lemma Suppose that x , y ∈ K n � 0 . Then x � y = ⇒ val( x ) � val( y ) val( x + y ) = max(val( x ) , val( y )) val( xy ) = val( x ) + val( y ) . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 12 / 58

  20. Valuation of Puiseux series ∞ � c k t α k x = x ( t ) = k =1 log | x ( t ) | val( x ) = lim = α 1 (and val(0) = −∞ ) . log t t →∞ Lemma Suppose that x , y ∈ K n � 0 . Then x � y = ⇒ val( x ) � val( y ) val( x + y ) = max(val( x ) , val( y )) val( xy ) = val( x ) + val( y ) . Thus, val is a morphism from K � 0 to a semifield of characteristic one, the tropical semifield T := ( R ∪ {−∞} , max , +). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 12 / 58

  21. Tropical spectrahedra Definition Suppose that S is a spectrahedron in K n � 0 . Then we say that val( S ) is a tropical spectrahedron. How can we study these creatures? Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 13 / 58

  22. A S ⊂ K n is basic semialgebraic if S = { ( x 1 , . . . , x n ) ∈ K n : P i ( x 1 , . . . , x n ) ⋄ 0 , ⋄ ∈ { >, = } , ∀ i ∈ [ q ] } where P 1 , . . . , P q ∈ K [ x 1 , . . . , x n ]. A semialgebraic set is a finite union of basic semialgebraic sets. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 14 / 58

  23. A S ⊂ K n is basic semialgebraic if S = { ( x 1 , . . . , x n ) ∈ K n : P i ( x 1 , . . . , x n ) ⋄ 0 , ⋄ ∈ { >, = } , ∀ i ∈ [ q ] } where P 1 , . . . , P q ∈ K [ x 1 , . . . , x n ]. A semialgebraic set is a finite union of basic semialgebraic sets. A set S ⊂ R n is basic semilinear if it is of the form S = { ( x 1 , . . . , x n ) ∈ R n : ℓ i ( x 1 , . . . , x n ) ⋄ h ( i ) , ⋄ ∈ { >, = } , ∀ i ∈ [ q ] } where ℓ 1 , . . . , ℓ q are linear forms with integer coefficients, h (1) , . . . , h ( q ) ∈ R . A semilinear set is a finite union of basic semilinear sets. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 14 / 58

  24. A S ⊂ K n is basic semialgebraic if S = { ( x 1 , . . . , x n ) ∈ K n : P i ( x 1 , . . . , x n ) ⋄ 0 , ⋄ ∈ { >, = } , ∀ i ∈ [ q ] } where P 1 , . . . , P q ∈ K [ x 1 , . . . , x n ]. A semialgebraic set is a finite union of basic semialgebraic sets. A set S ⊂ R n is basic semilinear if it is of the form S = { ( x 1 , . . . , x n ) ∈ R n : ℓ i ( x 1 , . . . , x n ) ⋄ h ( i ) , ⋄ ∈ { >, = } , ∀ i ∈ [ q ] } where ℓ 1 , . . . , ℓ q are linear forms with integer coefficients, h (1) , . . . , h ( q ) ∈ R . A semilinear set is a finite union of basic semilinear sets. Theorem (Alessandrini, Adv. in Geom. 2013) > 0 is semi-algebraic, then val( S ) ⊂ R n is semilinear and it is If S ⊂ K n closed. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 14 / 58

  25. A S ⊂ K n is basic semialgebraic if S = { ( x 1 , . . . , x n ) ∈ K n : P i ( x 1 , . . . , x n ) ⋄ 0 , ⋄ ∈ { >, = } , ∀ i ∈ [ q ] } where P 1 , . . . , P q ∈ K [ x 1 , . . . , x n ]. A semialgebraic set is a finite union of basic semialgebraic sets. A set S ⊂ R n is basic semilinear if it is of the form S = { ( x 1 , . . . , x n ) ∈ R n : ℓ i ( x 1 , . . . , x n ) ⋄ h ( i ) , ⋄ ∈ { >, = } , ∀ i ∈ [ q ] } where ℓ 1 , . . . , ℓ q are linear forms with integer coefficients, h (1) , . . . , h ( q ) ∈ R . A semilinear set is a finite union of basic semilinear sets. Theorem (Alessandrini, Adv. in Geom. 2013) > 0 is semi-algebraic, then val( S ) ⊂ R n is semilinear and it is If S ⊂ K n closed. Constructive version in Allamigeon, SG, Skomra arXiv:1610.06746 using Denef-Pas quantifier elimination in valued fields. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 14 / 58

  26. S := val( S ) is tropically convex max( α, β ) = 0 , u , v ∈ S = ⇒ sup( α e + u , β e + v ) ∈ S , where e = (1 , . . . , 1) ⊤ . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 15 / 58

  27. S := val( S ) is tropically convex max( α, β ) = 0 , u , v ∈ S = ⇒ sup( α e + u , β e + v ) ∈ S , where e = (1 , . . . , 1) ⊤ . Figure: Tropical spectrahedron. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 15 / 58

  28. Theorem (Semi-algebraic version of Kapranov theorem, Allamigeon, SG, Skomra arXiv:1610.06746) Consider a collection of m regions delimited by hypersurfaces: S i := { x ∈ K n � 0 | P − i ( x ) � P + i ( x ) } , i ∈ [ m ] i ,α x α ∈ K � 0 [ x ] , and let where P ± α p ± i = � S i := { x ∈ R n | max α (val p − α (val p + i ,α + � α, x � ) � max i ,α + � α, x � ) } Then � � � val( S i ) ⊂ val( S i ) ⊂ S i i ∈ [ m ] i ∈ [ m ] i ∈ [ m ] and the equality holds if � i ∈ [ m ] S i is the closure of its interior; in particular if the valuations val p ± i ,α are generic. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 16 / 58

  29. Example 1. S = { x ∈ K 3 > 0 | x 2 1 � tx 2 + t 4 x 2 x 3 } val S = { x ∈ R 3 | 2 x 1 � max(1 + x 2 , 4 + x 2 + x 3 ) } Example 2. Figure: This set is the closure of its interior. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 17 / 58

  30. The correspondence between stochastic mean payoff games and nonarchimedean spectrahedra explained Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 18 / 58

  31. Stochastic mean payoff games Two player, Min and Max, and a half-player, Nature, move a token on a digraph, alternating moves in a cyclic way: Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 19 / 58

  32. Stochastic mean payoff games Two player, Min and Max, and a half-player, Nature, move a token on a digraph, alternating moves in a cyclic way: If the current state i belongs to Player Min, this player chooses and arc i → j , and receives A ji from Player Max. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 19 / 58

  33. Stochastic mean payoff games Two player, Min and Max, and a half-player, Nature, move a token on a digraph, alternating moves in a cyclic way: If the current state i belongs to Player Min, this player chooses and arc i → j , and receives A ji from Player Max. The current state j now belongs to the half-player Nature, Nature throws a dice and next state becomes r with probability P jr . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 19 / 58

  34. Stochastic mean payoff games Two player, Min and Max, and a half-player, Nature, move a token on a digraph, alternating moves in a cyclic way: If the current state i belongs to Player Min, this player chooses and arc i → j , and receives A ji from Player Max. The current state j now belongs to the half-player Nature, Nature throws a dice and next state becomes r with probability P jr . The current state r now belongs Player Max, this player chosses an arc r → s , and receives B rs from Player Max. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 19 / 58

  35. Stochastic mean payoff games Two player, Min and Max, and a half-player, Nature, move a token on a digraph, alternating moves in a cyclic way: If the current state i belongs to Player Min, this player chooses and arc i → j , and receives A ji from Player Max. The current state j now belongs to the half-player Nature, Nature throws a dice and next state becomes r with probability P jr . The current state r now belongs Player Max, this player chosses an arc r → s , and receives B rs from Player Max. the current state s now belongs to Player Min, and so on. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 19 / 58

  36. If Min / Max play k turns according to strategies σ, τ , the payment of the game starting from state i ∈ [ n ] := { Min states } is denoted by R k i ( σ, τ ). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 20 / 58

  37. If Min / Max play k turns according to strategies σ, τ , the payment of the game starting from state i ∈ [ n ] := { Min states } is denoted by R k i ( σ, τ ). v k i is the value of the game in horizon k , starting from state i , and σ ∗ , τ ∗ are optimal strategies if E R k i ( σ ∗ , τ ) � v k i = E R k i ( σ ∗ , τ ∗ ) � E R k i ( σ, τ ∗ ) , ∀ σ, τ Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 20 / 58

  38. If Min / Max play k turns according to strategies σ, τ , the payment of the game starting from state i ∈ [ n ] := { Min states } is denoted by R k i ( σ, τ ). v k i is the value of the game in horizon k , starting from state i , and σ ∗ , τ ∗ are optimal strategies if E R k i ( σ ∗ , τ ) � v k i = E R k i ( σ ∗ , τ ∗ ) � E R k i ( σ, τ ∗ ) , ∀ σ, τ Theorem (Shapley) , v 0 ≡ 0 � v k s ∈ Min states ( B rs + v k − 1 � � i = min − A ji + P jr max ) s j ∈ Nature states r ∈ Max states v k = F ( v k − 1 ) , F : R n → R n Shapley operator Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 20 / 58

  39. If Min / Max play k turns according to strategies σ, τ , the payment of the game starting from state i ∈ [ n ] := { Min states } is denoted by R k i ( σ, τ ). v k i is the value of the game in horizon k , starting from state i , and σ ∗ , τ ∗ are optimal strategies if E R k i ( σ ∗ , τ ) � v k i = E R k i ( σ ∗ , τ ∗ ) � E R k i ( σ, τ ∗ ) , ∀ σ, τ Theorem (Shapley) , v 0 ≡ 0 � v k s ∈ Min states ( B rs + v k − 1 � � i = min − A ji + P jr max ) s j ∈ Nature states r ∈ Max states v k = F ( v k − 1 ) , F : R n → R n Shapley operator F ( x ) = ( − A ⊤ ) ⊙ min , + ( P × ( B ⊙ max , + x )) = A ♯ ◦ P ◦ B ( x ) Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 20 / 58

  40. The mean payoff vector k →∞ v k / k = lim k →∞ F k (0) / k ∈ R n ¯ v := lim does exist and it is achieved by positional stationnary strategies (coro of Kohlberg 1980). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 21 / 58

  41. The mean payoff vector k →∞ v k / k = lim k →∞ F k (0) / k ∈ R n ¯ v := lim does exist and it is achieved by positional stationnary strategies (coro of Kohlberg 1980). Mean payoff games: compute the mean payoff vector Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 21 / 58

  42. The mean payoff vector k →∞ v k / k = lim k →∞ F k (0) / k ∈ R n ¯ v := lim does exist and it is achieved by positional stationnary strategies (coro of Kohlberg 1980). Mean payoff games: compute the mean payoff vector We say that the mean payoff game with initial state i is (weakly) winning for Max if lim k v k i / k � 0. Gurvich, Karzanov and Khachyan asked in 1988 whether the determinisitic version is in P. Still open. Their argument implies membership in NP ∩ coNP, see also Zwick, Paterson. Same is true in the stochastic case (Condon). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 21 / 58

  43. Collatz-Wielandt property / winning certificates T := R ∪ {−∞} , Theorem (Akian, SG, Guterman IJAC 2912, coro of Nussbaum) max i ∈ n ¯ v i = cw ( R ) λ ∈ R | ∃ x ∈ T n , x �≡ −∞ : λ e + x � F ( x ) � � cw ( F ) := max Corollary Player Max has at least one winning state (i.e., 0 � max i ¯ v i ) iff ∃ x ∈ T n , x �≡ −∞ , x � F ( x ) Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 22 / 58

  44. Definition A square matrix is called a Metzler matrix if its off-diagonal entries are nonpositive. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 23 / 58

  45. Definition A square matrix is called a Metzler matrix if its off-diagonal entries are nonpositive. We suppose Q (1) , . . . , Q ( n ) ∈ K m × m are Metzler — the general case will reduce to this one. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 23 / 58

  46. Definition A square matrix is called a Metzler matrix if its off-diagonal entries are nonpositive. We suppose Q (1) , . . . , Q ( n ) ∈ K m × m are Metzler — the general case will reduce to this one. Want to decide whether Q ( x ) = x 1 Q (1) + · · · + x n Q ( n ) � 0 for some x ∈ K n � 0 , x � = 0. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 23 / 58

  47. If Q � 0 is a m × m symmetric matrix, then, the 1 × 1 and 2 × 2 principal minors of Q are nonnegative: Q ii � 0, Q ii Q jj � Q 2 ij . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 24 / 58

  48. If Q � 0 is a m × m symmetric matrix, then, the 1 × 1 and 2 × 2 principal minors of Q are nonnegative: Q ii � 0, Q ii Q jj � Q 2 ij . Is there a “converse”? Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 24 / 58

  49. If Q � 0 is a m × m symmetric matrix, then, the 1 × 1 and 2 × 2 principal minors of Q are nonnegative: Q ii � 0, Q ii Q jj � Q 2 ij . Is there a “converse”? Lemma Assume that Q ii � 0 , Q ii Q jj � ( m − 1) 2 Q 2 ij . Then Q � 0 . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 24 / 58

  50. If Q � 0 is a m × m symmetric matrix, then, the 1 × 1 and 2 × 2 principal minors of Q are nonnegative: Q ii � 0, Q ii Q jj � Q 2 ij . Is there a “converse”? Lemma Assume that Q ii � 0 , Q ii Q jj � ( m − 1) 2 Q 2 ij . Then Q � 0 . Proof. Can assume that Q ii ≡ 1 (consider diag ( Q ) − 1 / 2 Q diag ( Q ) − 1 / 2 ). Then, | Q ij | � 1 / ( m − 1), and so Q ii � � j � = i | Q ij | implies Q � 0. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 24 / 58

  51. If Q � 0 is a m × m symmetric matrix, then, the 1 × 1 and 2 × 2 principal minors of Q are nonnegative: Q ii � 0, Q ii Q jj � Q 2 ij . Is there a “converse”? Lemma Assume that Q ii � 0 , Q ii Q jj � ( m − 1) 2 Q 2 ij . Then Q � 0 . Proof. Can assume that Q ii ≡ 1 (consider diag ( Q ) − 1 / 2 Q diag ( Q ) − 1 / 2 ). Then, | Q ij | � 1 / ( m − 1), and so Q ii � � j � = i | Q ij | implies Q � 0. Archimedean modification of Yu’s theorem, that the image by the nonarchimedean valuation of the SDP cone is given by 1 × 1 and 2 × 2 minor conditions. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 24 / 58

  52. Let S := { x ∈ K n � 0 : Q ( x ) � 0 } Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 25 / 58

  53. Let S := { x ∈ K n � 0 : Q ( x ) � 0 } Let S out be defined by the 1 × 1 and 2 × 2 principal minor conditions Q ii ( x ) Q jj ( x ) � ( Q ij ( x )) 2 Q ii ( x ) � 0 , Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 25 / 58

  54. Let S := { x ∈ K n � 0 : Q ( x ) � 0 } Let S out be defined by the 1 × 1 and 2 × 2 principal minor conditions Q ii ( x ) Q jj ( x ) � ( Q ij ( x )) 2 Q ii ( x ) � 0 , Let S in be defined by the reinforced minor conditions Q ii ( x ) Q jj ( x ) � ( m − 1) 2 ( Q ij ( x )) 2 Q ii ( x ) � 0 , Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 25 / 58

  55. Let S := { x ∈ K n � 0 : Q ( x ) � 0 } Let S out be defined by the 1 × 1 and 2 × 2 principal minor conditions Q ii ( x ) Q jj ( x ) � ( Q ij ( x )) 2 Q ii ( x ) � 0 , Let S in be defined by the reinforced minor conditions Q ii ( x ) Q jj ( x ) � ( m − 1) 2 ( Q ij ( x )) 2 Q ii ( x ) � 0 , Theorem (Allamigeon, SG, Skomra) S in ⊆ S ⊆ S out and if Q is tropically generic (valuations of coeffs are generic), val( S in ) = val( S ) = val( S out ) . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 25 / 58

  56. Let S := { x ∈ K n � 0 : Q ( x ) � 0 } Let S out be defined by the 1 × 1 and 2 × 2 principal minor conditions Q ii ( x ) Q jj ( x ) � ( Q ij ( x )) 2 Q ii ( x ) � 0 , Let S in be defined by the reinforced minor conditions Q ii ( x ) Q jj ( x ) � ( m − 1) 2 ( Q ij ( x )) 2 Q ii ( x ) � 0 , Theorem (Allamigeon, SG, Skomra) S in ⊆ S ⊆ S out and if Q is tropically generic (valuations of coeffs are generic), val( S in ) = val( S ) = val( S out ) . We show that if X = ∩ k { x | P k ( x ) � 0 } , then val X = ∩ k val { x | P k ( x ) � 0 } if the polynomials P k are tropically generic Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 25 / 58

  57. Let S := { x ∈ K n � 0 : Q ( x ) � 0 } Let S out be defined by the 1 × 1 and 2 × 2 principal minor conditions Q ii ( x ) Q jj ( x ) � ( Q ij ( x )) 2 Q ii ( x ) � 0 , Let S in be defined by the reinforced minor conditions Q ii ( x ) Q jj ( x ) � ( m − 1) 2 ( Q ij ( x )) 2 Q ii ( x ) � 0 , Theorem (Allamigeon, SG, Skomra) S in ⊆ S ⊆ S out and if Q is tropically generic (valuations of coeffs are generic), val( S in ) = val( S ) = val( S out ) . We show that if X = ∩ k { x | P k ( x ) � 0 } , then val X = ∩ k val { x | P k ( x ) � 0 } if the polynomials P k are tropically generic (apply semi-algebraic version of Kapranov theorem) Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 25 / 58

  58. Can we describe combinatorially val S ? Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 26 / 58

  59. Suppose Q ii ( x ) � 0, write Q ii = Q + ii − Q − ii . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 27 / 58

  60. Suppose Q ii ( x ) � 0, write Q ii = Q + ii − Q − ii . Then val Q + ii ( x ) � val Q − ii ( x ) Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 27 / 58

  61. Suppose Q ii ( x ) � 0, write Q ii = Q + ii − Q − ii . Then val Q + ii ( x ) � val Q − ii ( x ) Moreover, if Q ii ( x ) Q jj ( x ) � ( Q ij ( x )) 2 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 27 / 58

  62. Suppose Q ii ( x ) � 0, write Q ii = Q + ii − Q − ii . Then val Q + ii ( x ) � val Q − ii ( x ) Moreover, if Q ii ( x ) Q jj ( x ) � ( Q ij ( x )) 2 then Q + ii ( x ) Q + jj ( x ) + Q − ii ( x ) Q − jj ( x ) � Q + ii ( x ) Q − jj ( x ) + Q − ii ( x ) Q + jj ( x ) + ( Q ij ( x )) 2 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 27 / 58

  63. Suppose Q ii ( x ) � 0, write Q ii = Q + ii − Q − ii . Then val Q + ii ( x ) � val Q − ii ( x ) Moreover, if Q ii ( x ) Q jj ( x ) � ( Q ij ( x )) 2 then Q + ii ( x ) Q + jj ( x ) + Q − ii ( x ) Q − jj ( x ) � Q + ii ( x ) Q − jj ( x ) + Q − ii ( x ) Q + jj ( x ) + ( Q ij ( x )) 2 and so val Q + ii ( x ) + val Q + jj ( x ) � 2 val Q ij ( x ) Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 27 / 58

  64. Tropical Metzler spectrahedra Theorem (tropical Metzler spectrahedra) For tropically generic Metzler matrices ( Q ( k ) ) k the set val( S ) is described by the tropical minor inequalities of order 1 and 2 , ( x k + val( Q ( k ) ( x l + val( Q ( l ) ∀ i , max ii )) � max jj )) Q ( k ) Q ( l ) ii > 0 jj < 0 and ( x k + val( Q ( k ) ( x k + val( Q ( k ) ∀ i � = j , max ii )) + max jj )) Q ( k ) Q ( k ) ii > 0 jj > 0 ( x l + val( Q ( l ) � 2 max ij )) . Q ( l ) ij < 0 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 28 / 58

  65. Tropical Metzler spectrahedra Theorem (tropical Metzler spectrahedra) For tropically generic Metzler matrices ( Q ( k ) ) k the set val( S ) is described by the tropical minor inequalities of order 1 and 2 , ( x k + val( Q ( k ) ( x l + val( Q ( l ) ∀ i , max ii )) � max jj )) Q ( k ) Q ( l ) ii > 0 jj < 0 and ( x k + val( Q ( k ) ( x k + val( Q ( k ) ∀ i � = j , max ii )) + max jj )) Q ( k ) Q ( k ) ii > 0 jj > 0 ( x l + val( Q ( l ) � 2 max ij )) . Q ( l ) ij < 0 Extends the characterization of val( SDPCONE ) by Yu. . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 28 / 58

  66. From spectrahedra to Shapley operators Lemma The set val( S ) can be equivalently defined as the set of all x such that for all k we have ij ) + 1 � − val( Q ( k ) (val( Q ( l ) � x k � min max ii ) + x l ) 2 Q ( k ) Q ( l ) ij < 0 ii > 0 �� (val( Q ( l ) + max jj ) + x l ) . Q ( l ) jj > 0 In other words, we have val( S ) = { x ∈ ( R ∪ {−∞} ) n : x � F ( x ) } , where F is a Shapley operator of a stochastic mean payoff game. We denote this game by Γ . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 29 / 58

  67. Reading the Game on the Shapley Operator ij ) + 1 � − val( Q ( k ) (val( Q ( l ) � x k � min max ii ) + x l ) 2 Q ( k ) Q ( l ) ij < 0 ii > 0 �� (val( Q ( l ) + max jj ) + x l ) . Q ( l ) jj > 0 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 30 / 58

  68. Reading the Game on the Shapley Operator ij ) + 1 � − val( Q ( k ) (val( Q ( l ) � x k � min max ii ) + x l ) 2 Q ( k ) Q ( l ) ij < 0 ii > 0 �� (val( Q ( l ) + max jj ) + x l ) . Q ( l ) jj > 0 MIN wants to show infeasibility, MAX feasibility Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 30 / 58

  69. Reading the Game on the Shapley Operator ij ) + 1 � − val( Q ( k ) (val( Q ( l ) � x k � min max ii ) + x l ) 2 Q ( k ) Q ( l ) ij < 0 ii > 0 �� (val( Q ( l ) + max jj ) + x l ) . Q ( l ) jj > 0 MIN wants to show infeasibility, MAX feasibility state of MIN, x k , 1 � k � n Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 30 / 58

  70. Reading the Game on the Shapley Operator ij ) + 1 � − val( Q ( k ) (val( Q ( l ) � x k � min max ii ) + x l ) 2 Q ( k ) Q ( l ) ij < 0 ii > 0 �� (val( Q ( l ) + max jj ) + x l ) . Q ( l ) jj > 0 MIN wants to show infeasibility, MAX feasibility state of MIN, x k , 1 � k � n MIN chooses { i , j } , 1 � i � = j � m or { i } with Q k ii < 0, MAX pays to MIN val Q ( k ) ij Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 30 / 58

  71. Reading the Game on the Shapley Operator ij ) + 1 � − val( Q ( k ) (val( Q ( l ) � x k � min max ii ) + x l ) 2 Q ( k ) Q ( l ) ij < 0 ii > 0 �� (val( Q ( l ) + max jj ) + x l ) . Q ( l ) jj > 0 MIN wants to show infeasibility, MAX feasibility state of MIN, x k , 1 � k � n MIN chooses { i , j } , 1 � i � = j � m or { i } with Q k ii < 0, MAX pays to MIN val Q ( k ) ij NATURE throws a dice to decide whether i or j is the next state Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 30 / 58

  72. Reading the Game on the Shapley Operator ij ) + 1 � − val( Q ( k ) (val( Q ( l ) � x k � min max ii ) + x l ) 2 Q ( k ) Q ( l ) ij < 0 ii > 0 �� (val( Q ( l ) + max jj ) + x l ) . Q ( l ) jj > 0 MIN wants to show infeasibility, MAX feasibility state of MIN, x k , 1 � k � n MIN chooses { i , j } , 1 � i � = j � m or { i } with Q k ii < 0, MAX pays to MIN val Q ( k ) ij NATURE throws a dice to decide whether i or j is the next state suppose next state of MAX, i , 1 � i � m , Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 30 / 58

  73. Reading the Game on the Shapley Operator ij ) + 1 � − val( Q ( k ) (val( Q ( l ) � x k � min max ii ) + x l ) 2 Q ( k ) Q ( l ) ij < 0 ii > 0 �� (val( Q ( l ) + max jj ) + x l ) . Q ( l ) jj > 0 MIN wants to show infeasibility, MAX feasibility state of MIN, x k , 1 � k � n MIN chooses { i , j } , 1 � i � = j � m or { i } with Q k ii < 0, MAX pays to MIN val Q ( k ) ij NATURE throws a dice to decide whether i or j is the next state suppose next state of MAX, i , 1 � i � m , MAX moves to x l such that Q ( l ) ii > 0, MIN pays to MAX val Q ( l ) ii . Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 30 / 58

  74. Main example revisited   0 − 1 0 x 3 Q (1) :=  , t − 1 − 5 / 4 − 5 / 4 − 1 0 − 3 / 4 1 1 0  0 0 0  0 0 0  Q (2) :=  , 0 − 1 0 3 2  9 / 4 9 / 4 t 9 / 4 0 0 1 0 0 x 2  − t 3 / 4  t 0 − 1 − 1 Q (3) :=  . t − 5 / 4 0 − 1 0  − t 3 / 4 x 1 − 1 0 Construction of Γ We construct Γ as follows: Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 31 / 58

  75. Main example revisited   0 − 1 0 x 3 Q (1) :=  , t − 1 − 5 / 4 − 5 / 4 − 1 0 − 3 / 4 1 1 0  0 0 0  0 0 0  Q (2) :=  , 0 − 1 0 3 2  9 / 4 9 / 4 t 9 / 4 0 0 1 0 0 x 2  − t 3 / 4  t 0 − 1 − 1 Q (3) :=  . t − 5 / 4 0 − 1 0  − t 3 / 4 x 1 − 1 0 Construction of Γ The number of matrices (here: 3) defines the number of states controlled by Player Min. Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 31 / 58

  76. Main example revisited   0 − 1 0 x 3 Q (1) :=  , t − 1 − 5 / 4 − 5 / 4 − 1 0 − 3 / 4 1 1 0  0 0 0  0 0 0  Q (2) :=  , 0 − 1 0 3 2  9 / 4 9 / 4 t 9 / 4 0 0 1 0 0 x 2  − t 3 / 4  t 0 − 1 − 1 Q (3) :=  . t − 5 / 4 0 − 1 0  − t 3 / 4 x 1 − 1 0 Construction of Γ The size of matrices (here: 3 × 3) defines the number of states controlled by Player Max (here: 3). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 31 / 58

  77. Main example revisited   0 − 1 0 x 3 Q (1) :=  , t − 1 − 5 / 4 − 5 / 4 − 1 0 − 3 / 4 1 1 0  0 0 0  0 0 0  Q (2) :=  , 0 − 1 0 3 2  9 / 4 9 / 4 t 9 / 4 0 0 1 0 0 x 2  − t 3 / 4  t 0 − 1 − 1 Q (3) :=  . t − 5 / 4 0 − 1 0  − t 3 / 4 x 1 − 1 0 Construction of Γ If Q ( k ) is negative, then Player Min can move from state k to state i . ii After this move Player Max receives − val( Q ( k ) ii ). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 31 / 58

  78. Main example revisited   0 − 1 0 x 3 Q (1) :=  , t − 1 − 5 / 4 − 5 / 4 − 1 0 − 3 / 4 1 1 0  0 0 0  0 0 0  Q (2) :=  , 0 − 1 0 3 2  9 / 4 9 / 4 t 9 / 4 0 0 1 0 0 x 2  − t 3 / 4  0 t − 1 − 1 Q (3) :=  . t − 5 / 4 0 − 1 0  − t 3 / 4 x 1 − 1 0 Construction of Γ If Q ( k ) is positive, then Player Max can move from state i to state k . ii After this move Player Max receives val( Q ( k ) ii ). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 31 / 58

  79. Main example revisited   0 − 1 0 x 3 Q (1) :=  , t − 1 − 5 / 4 − 5 / 4 − 1 0 − 3 / 4 1 1 0  0 0 0  0 0 0  Q (2) :=  , 0 − 1 0 3 2  9 / 4 9 / 4 t 9 / 4 0 0 1 0 0 x 2  − t 3 / 4  t 0 − 1 − 1 Q (3) :=  . t − 5 / 4 0 − 1 0  − t 3 / 4 x 1 − 1 0 Construction of Γ If Q ( k ) is nonzero, i � = j , then Player Min have a coin-toss move from ij state k to states ( i , j ) and Player Max receives − val( Q ( k ) ij ). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 31 / 58

  80. Main example revisited   0 − 1 0 x 3 Q (1) :=  , t − 1 − 5 / 4 − 5 / 4 − 1 0 − 3 / 4 1 1 0  0 0 0  0 0 0  Q (2) :=  , 0 − 1 0 3 2  9 / 4 9 / 4 t 9 / 4 0 0 1 0 0 x 2  − t 3 / 4  t 0 − 1 − 1 Q (3) :=  . t − 5 / 4 0 − 1 0  − t 3 / 4 x 1 − 1 0 Construction of Γ If Q ( k ) is nonzero, i � = j , then Player Min have a coin-toss move from ij state k to states ( i , j ) and Player Max receives − val( Q ( k ) ij ). Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 31 / 58

  81. Example There is only one pair of optimal 3 policies − 5 / 4 − 3 / 4 1 0 � � 3 → 1 , 3 , 2 → 1 . 3 2 9 / 4 1 0 2 − 1 0 1 Allamigeon, Gaubert, Katz, Skomra (Inria-X) Nonarchimedean SDP January 24, 2019, Birmingham 32 / 58

Recommend


More recommend