bad news
play

Bad News on Combinatorial Lower Bounds for the Extension - PowerPoint PPT Presentation

Bad News on Combinatorial Lower Bounds for the Extension Complexity of the Spanning Tree Polytope Kaveh Khoshkhah Dirk Oliver Theis Aussois Combinatorial Optimization Workshop Jan 9-13, 2017 Outline Extended Formulations for Spanning


  1. Extension Complexity of Spanning Tree Polytope Ω( n 2 ) = extension complexity(spanning tree n ) = O ( n 3 ) Progress on Upper Bounds Only for special graphs ( not complete!), e.g., ◮ Fiorini-Huynh-Pashkovich-Joret 2017 ◮ Kolman-Kouteck´ y-Tiwary 2016 ◮ Williams 2002 Progress on Lower Bounds

  2. Extension Complexity of Spanning Tree Polytope Ω( n 2 ) = extension complexity(spanning tree n ) = O ( n 3 ) Progress on Upper Bounds Only for special graphs ( not complete!), e.g., ◮ Fiorini-Huynh-Pashkovich-Joret 2017 ◮ Kolman-Kouteck´ y-Tiwary 2016 ◮ Williams 2002 Progress on Lower Bounds

  3. Extension Complexity of Spanning Tree Polytope Ω( n 2 ) = extension complexity(spanning tree n ) = O ( n 3 ) Progress on Upper Bounds Only for special graphs ( not complete!), e.g., ◮ Fiorini-Huynh-Pashkovich-Joret 2017 ◮ Kolman-Kouteck´ y-Tiwary 2016 ◮ Williams 2002 Progress on Lower Bounds

  4. Long-Term Goals:

  5. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough,

  6. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n );

  7. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n ); ◮ Improving lower bound to anything ≫ n 2 would be a breakthrough,

  8. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n ); ◮ Improving lower bound to anything ≫ n 2 would be a breakthrough, ◮ e.g., n 2 · log log( n ).

  9. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n ); ◮ Improving lower bound to anything ≫ n 2 would be a breakthrough, ◮ e.g., n 2 · log log( n ). This Talk’s Results:

  10. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n ); ◮ Improving lower bound to anything ≫ n 2 would be a breakthrough, ◮ e.g., n 2 · log log( n ). This Talk’s Results: ◮ Negative results for the lower bounds

  11. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n ); ◮ Improving lower bound to anything ≫ n 2 would be a breakthrough, ◮ e.g., n 2 · log log( n ). This Talk’s Results: ◮ Negative results for the lower bounds ◮ The convenient “combinatorial” lower bounds give:

  12. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n ); ◮ Improving lower bound to anything ≫ n 2 would be a breakthrough, ◮ e.g., n 2 · log log( n ). This Talk’s Results: ◮ Negative results for the lower bounds ◮ The convenient “combinatorial” lower bounds give: ◮ “Fooling set” bound: O ( n 2 )

  13. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n ); ◮ Improving lower bound to anything ≫ n 2 would be a breakthrough, ◮ e.g., n 2 · log log( n ). This Talk’s Results: ◮ Negative results for the lower bounds ◮ The convenient “combinatorial” lower bounds give: ◮ “Fooling set” bound: O ( n 2 ) ◮ “Nondeterministic communication complexity” bound: O ( n 2 log n )

  14. Long-Term Goals: ◮ Improving upper bound to anything ≪ n 3 would be a breakthrough, ◮ e.g., extended formulation of size n 3 / log log( n ); ◮ Improving lower bound to anything ≫ n 2 would be a breakthrough, ◮ e.g., n 2 · log log( n ). This Talk’s Results: ◮ Negative results for the lower bounds ◮ The convenient “combinatorial” lower bounds give: ◮ “Fooling set” bound: O ( n 2 ) ◮ “Nondeterministic communication complexity” bound: O ( n 2 log n ) ◮ “Double negative” results: upper bounds on lower bounds.

  15. Outline Extended Formulations for Spanning Tree: Status Quo Lower Bounds for Extension Complexity 101 What We Did Glimpse At the Details

  16. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity Successful for

  17. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity ◮ “Combinatorial” means: consider vertex-facet incidences only. Successful for

  18. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity ◮ “Combinatorial” means: consider vertex-facet incidences only. ◮ As opposed to: consider distance of vertex from facet hyperplane (non-combinatorial). Successful for

  19. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity ◮ “Combinatorial” means: consider vertex-facet incidences only. ◮ As opposed to: consider distance of vertex from facet hyperplane (non-combinatorial). ◮ Nondeterministic communication complexity bound is the strongest combinatorial bound. Successful for

  20. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity ◮ “Combinatorial” means: consider vertex-facet incidences only. ◮ As opposed to: consider distance of vertex from facet hyperplane (non-combinatorial). ◮ Nondeterministic communication complexity bound is the strongest combinatorial bound. ◮ “Nondet. Comm. Complexity bound” = 2 nondet. comm. complexity Successful for

  21. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity ◮ “Combinatorial” means: consider vertex-facet incidences only. ◮ As opposed to: consider distance of vertex from facet hyperplane (non-combinatorial). ◮ Nondeterministic communication complexity bound is the strongest combinatorial bound. ◮ “Nondet. Comm. Complexity bound” = 2 nondet. comm. complexity Successful for ◮ TSP

  22. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity ◮ “Combinatorial” means: consider vertex-facet incidences only. ◮ As opposed to: consider distance of vertex from facet hyperplane (non-combinatorial). ◮ Nondeterministic communication complexity bound is the strongest combinatorial bound. ◮ “Nondet. Comm. Complexity bound” = 2 nondet. comm. complexity Successful for ◮ TSP ◮ Max-Cut

  23. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity ◮ “Combinatorial” means: consider vertex-facet incidences only. ◮ As opposed to: consider distance of vertex from facet hyperplane (non-combinatorial). ◮ Nondeterministic communication complexity bound is the strongest combinatorial bound. ◮ “Nondet. Comm. Complexity bound” = 2 nondet. comm. complexity Successful for ◮ TSP ◮ Max-Cut ◮ Bipartite Matching

  24. Combinatorial Lower Bounds Some Combinatorial Bounds fooling set bound ≤ fractional rectangle covering bound ≤ 2 nondeterministic communication complexity ≤ extension complexity ◮ “Combinatorial” means: consider vertex-facet incidences only. ◮ As opposed to: consider distance of vertex from facet hyperplane (non-combinatorial). ◮ Nondeterministic communication complexity bound is the strongest combinatorial bound. ◮ “Nondet. Comm. Complexity bound” = 2 nondet. comm. complexity Successful for ◮ TSP ◮ Max-Cut ◮ Bipartite Matching ◮ . . .

  25. Nondeterministic Communication Protocols “Nondeterministic”: No false positive outputs; If f ( S , T ) = true , ∃ certificate s.t. output is true

  26. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol

  27. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol Link to Extensions of Polytopes Fix polytope P

  28. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol Link to Extensions of Polytopes Fix polytope P ◮ Alice gets a facet S

  29. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol Link to Extensions of Polytopes Fix polytope P ◮ Alice gets a facet S ◮ Bob gets a vertex T

  30. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol Link to Extensions of Polytopes Fix polytope P ◮ Alice gets a facet S ◮ Bob gets a vertex T � true , if vertex is off facet; ◮ f ( S , T ) = false , if vertex on facet.

  31. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol Link to Extensions of Polytopes Fix polytope P ◮ Alice gets a facet S ◮ Bob gets a vertex T � true , if vertex is off facet; ◮ f ( S , T ) = false , if vertex on facet. ◮ 2 ndCC( f ) ≤ extension complexity of P

  32. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol Link to Extensions of Polytopes Fix polytope P ◮ Alice gets a facet S ◮ Bob gets a vertex T � true , if vertex is off facet; ◮ f ( S , T ) = false , if vertex on facet. ◮ 2 ndCC( f ) ≤ extension complexity of P Careful!

  33. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol Link to Extensions of Polytopes Fix polytope P ◮ Alice gets a facet S ◮ Bob gets a vertex T � true , if vertex is off facet; ◮ f ( S , T ) = false , if vertex on facet. ◮ 2 ndCC( f ) ≤ extension complexity of P Careful! ◮ “Nondeterministic Communication Complexity”: # bits (in best protocol)

  34. ◮ “Nondeterministic Communication Complexity” of f : Cost of best protocol Link to Extensions of Polytopes Fix polytope P ◮ Alice gets a facet S ◮ Bob gets a vertex T � true , if vertex is off facet; ◮ f ( S , T ) = false , if vertex on facet. ◮ 2 ndCC( f ) ≤ extension complexity of P Careful! ◮ “Nondeterministic Communication Complexity”: # bits (in best protocol) ◮ “Nondeterministic Communication Complexity bound ” = 2 nondet. comm. complexity

  35. Back to Spanning Tree:

  36. Back to Spanning Tree: Polytope:

  37. Back to Spanning Tree: Polytope: ◮ P := spanning tree polytope

  38. Back to Spanning Tree: Polytope: ◮ P := spanning tree polytope ◮ Vertices ˆ = trees

  39. Back to Spanning Tree: Polytope: ◮ P := spanning tree polytope ◮ Vertices ˆ = trees ◮ Facets: � x e ≤ | S | − 1 e ⊂ S ∀ S � [ n ] , | S | > 1

  40. Back to Spanning Tree: Polytope: ◮ P := spanning tree polytope ◮ Vertices ˆ = trees ◮ Facets: � x e ≤ | S | − 1 e ⊂ S ∀ S � [ n ] , | S | > 1 ◮ Ignore x ≥ 0 inequalities.

  41. Back to Spanning Tree: Polytope: Communication complexity part ◮ P := spanning tree polytope ◮ Vertices ˆ = trees ◮ Facets: � x e ≤ | S | − 1 e ⊂ S ∀ S � [ n ] , | S | > 1 ◮ Ignore x ≥ 0 inequalities.

  42. Back to Spanning Tree: Polytope: Communication complexity part ◮ P := spanning tree polytope ◮ Alice gets set S (facet) ◮ Vertices ˆ = trees ◮ Facets: � x e ≤ | S | − 1 e ⊂ S ∀ S � [ n ] , | S | > 1 ◮ Ignore x ≥ 0 inequalities.

  43. Back to Spanning Tree: Polytope: Communication complexity part ◮ P := spanning tree polytope ◮ Alice gets set S (facet) ◮ Vertices ˆ ◮ Bob gets a tree T (vertex) = trees ◮ Facets: � x e ≤ | S | − 1 e ⊂ S ∀ S � [ n ] , | S | > 1 ◮ Ignore x ≥ 0 inequalities.

  44. Back to Spanning Tree: Polytope: Communication complexity part ◮ P := spanning tree polytope ◮ Alice gets set S (facet) ◮ Vertices ˆ ◮ Bob gets a tree T (vertex) = trees ◮ Facets: ◮ f ( S , T ) = � true , if S disconnected in T ; � x e ≤ | S | − 1 false , if S connected in T . e ⊂ S ∀ S � [ n ] , | S | > 1 ◮ Ignore x ≥ 0 inequalities.

  45. Outline Extended Formulations for Spanning Tree: Status Quo Lower Bounds for Extension Complexity 101 What We Did Glimpse At the Details

  46. ✑ Our Results Theorem Fooling set bound = O ( n 2 )

  47. Our Results Theorem Fooling set bound = O ( n 2 ) ✑ arXiv:1701.00350

  48. Our Results Theorem Fooling set bound = O ( n 2 ) ✑ arXiv:1701.00350 Theorem Nondet. communication complexity bound = O ( n 2 log n )

  49. Our Results Theorem Fooling set bound = O ( n 2 ) ✑ arXiv:1701.00350 Theorem Nondet. communication complexity bound = O ( n 2 log n ) Previously best bounds:

  50. Our Results Theorem Fooling set bound = O ( n 2 ) ✑ arXiv:1701.00350 Theorem Nondet. communication complexity bound = O ( n 2 log n ) Previously best bounds: ◮ O ( n 8 / 3 log n ) for fractional rectangle covering bound (Weltge ’15)

  51. Our Results Theorem Fooling set bound = O ( n 2 ) ✑ arXiv:1701.00350 Theorem Nondet. communication complexity bound = O ( n 2 log n ) Previously best bounds: ◮ O ( n 8 / 3 log n ) for fractional rectangle covering bound (Weltge ’15) ◮ O ( n 3 ) for nondet. communication complexity bound (trivial)

  52. How to prove an upper bound for Nondet. CC

  53. How to prove an upper bound for Nondet. CC ◮ Alice has set S

  54. How to prove an upper bound for Nondet. CC ◮ Alice has set S ◮ Bob has tree T

  55. How to prove an upper bound for Nondet. CC ◮ Alice has set S ◮ Bob has tree T ◮ f ( S , T ) = whether S is disconnected in T

  56. How to prove an upper bound for Nondet. CC ◮ Alice has set S ◮ Bob has tree T ◮ f ( S , T ) = whether S is disconnected in T ◮ Need: Protocol with short certificates

  57. ◮ Alice has a set S ,

  58. ◮ Alice has a set S , ◮ Bob has a tree T ,

  59. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T .

  60. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T . The “obvious” certificate Protocol:

  61. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T . The “obvious” certificate Protocol: ◮ Prover sends ( a , b , x ) ∈ [ n ] 3

  62. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T . The “obvious” certificate Protocol: ◮ Prover sends ( a , b , x ) ∈ [ n ] 3 ◮ Alice:

  63. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T . The “obvious” certificate Protocol: ◮ Prover sends ( a , b , x ) ∈ [ n ] 3 ◮ Alice: ◮ Check “ a , b ∈ S , x / ∈ S ”?

  64. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T . The “obvious” certificate Protocol: ◮ Prover sends ( a , b , x ) ∈ [ n ] 3 ◮ Alice: ◮ Check “ a , b ∈ S , x / ∈ S ”? ◮ Output answer

  65. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T . The “obvious” certificate Protocol: ◮ Prover sends ( a , b , x ) ∈ [ n ] 3 ◮ Alice: ◮ Check “ a , b ∈ S , x / ∈ S ”? ◮ Output answer ◮ Bob:

  66. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T . The “obvious” certificate Protocol: ◮ Prover sends ( a , b , x ) ∈ [ n ] 3 ◮ Alice: ◮ Check “ a , b ∈ S , x / ∈ S ”? ◮ Output answer ◮ Bob: ◮ Check “ a ⌢ x ⌢ b in T ?” (i.e., x on path in T between a and b )

  67. ◮ Alice has a set S , ◮ Bob has a tree T , ◮ They want to decide if S is disconnected in T . The “obvious” certificate Protocol: ◮ Prover sends ( a , b , x ) ∈ [ n ] 3 ◮ Alice: ◮ Check “ a , b ∈ S , x / ∈ S ”? ◮ Output answer ◮ Bob: ◮ Check “ a ⌢ x ⌢ b in T ?” (i.e., x on path in T between a and b ) ◮ Output answer

Recommend


More recommend