clique-stable set separationaulagout/data/uploads/slides... · 2020. 11. 30. · introduction...
Post on 06-Mar-2021
3 Views
Preview:
TRANSCRIPT
Introduction Results Relationship with other problems Conclusion
Clique-Stable set Separation
Aurélie Lagoutte
LIP, ENS Lyon
Princeton Discrete Mathematics Seminar - Oct.15, 2015
Joint work with N. Bousquet, S. Thomassé and T. Trunck
1/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
Do the clique and the stable set intersect?
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
Do the clique and the stable set intersect?
Yescertificate: xx
log(n) bits
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
Do the clique and the stable set intersect?
Yescertificate: xx
log(n) bits
I agreeI agree
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
Do the clique and the stable set intersect?
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
Do the clique and the stable set intersect?
Nocertificate: ?
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
Do the clique and the stable set intersect?
certificate: CNo
log(k) bits
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
G
Alice Bob
Prover
Do the clique and the stable set intersect?
certificate: CNo
log(k) bits
I agreeI agree
2/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
Goal (Yannakakis 1991)Find a CS-separator : a family of cuts that can separate all thepairs Clique-Stable set.
Non-det communication complexity↔ min. size of a CS-Separator.
Main question (Yannakakis 1991)Do perfect graphs admit polynomial-size CS-Separator?
Upper Bound: there exists a CS-separator of size O(nlog n).Lower bound in perfect graphs? Lower bound in general?Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? Or for which classes of graphs does it exist?
3/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
Goal (Yannakakis 1991)Find a CS-separator : a family of cuts that can separate all thepairs Clique-Stable set.
Non-det communication complexity↔ min. size of a CS-Separator.
Main question (Yannakakis 1991)Do perfect graphs admit polynomial-size CS-Separator?
Upper Bound: there exists a CS-separator of size O(nlog n).Lower bound in perfect graphs? Lower bound in general?Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? Or for which classes of graphs does it exist?
3/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
Goal (Yannakakis 1991)Find a CS-separator : a family of cuts that can separate all thepairs Clique-Stable set.
Non-det communication complexity↔ min. size of a CS-Separator.
Main question (Yannakakis 1991)Do perfect graphs admit polynomial-size CS-Separator?
Upper Bound: there exists a CS-separator of size O(nlog n).
Lower bound in perfect graphs? Lower bound in general?Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? Or for which classes of graphs does it exist?
3/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
Goal (Yannakakis 1991)Find a CS-separator : a family of cuts that can separate all thepairs Clique-Stable set.
Non-det communication complexity↔ min. size of a CS-Separator.
Main question (Yannakakis 1991)Do perfect graphs admit polynomial-size CS-Separator?
Upper Bound: there exists a CS-separator of size O(nlog n).Lower bound in perfect graphs?
Lower bound in general?Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? Or for which classes of graphs does it exist?
3/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
Goal (Yannakakis 1991)Find a CS-separator : a family of cuts that can separate all thepairs Clique-Stable set.
Non-det communication complexity↔ min. size of a CS-Separator.
Main question (Yannakakis 1991)Do perfect graphs admit polynomial-size CS-Separator?
Upper Bound: there exists a CS-separator of size O(nlog n).Lower bound in perfect graphs? Lower bound in general?
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? Or for which classes of graphs does it exist?
3/33
Introduction Results Relationship with other problems Conclusion
Clique vs Independent Set Problem
Goal (Yannakakis 1991)Find a CS-separator : a family of cuts that can separate all thepairs Clique-Stable set.
Non-det communication complexity↔ min. size of a CS-Separator.
Main question (Yannakakis 1991)Do perfect graphs admit polynomial-size CS-Separator?
Upper Bound: there exists a CS-separator of size O(nlog n).Lower bound in perfect graphs? Lower bound in general?Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? Or for which classes of graphs does it exist?
3/33
Introduction Results Relationship with other problems Conclusion
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)?
What about random graphs?Theorem [Bousquet, L., Thomassé]
Random graphs admit a CS-Separator of size O(n7) asymptoticallyalmost surely.
Lower Bound:
(Huang, Sudakov 2012): we need Ω(n 65 ) cuts for some graphs.
(Amano, Shigeta 2013): we need Ω(n2−ε) for some graphs.(Göös 2015): we need nΩ(log0.128 n) cuts for some graphs.
For which classes of graphs does there exist a polynomialCS-Separator?
4/33
Introduction Results Relationship with other problems Conclusion
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)?What about random graphs?
Theorem [Bousquet, L., Thomassé]
Random graphs admit a CS-Separator of size O(n7) asymptoticallyalmost surely.
Lower Bound:
(Huang, Sudakov 2012): we need Ω(n 65 ) cuts for some graphs.
(Amano, Shigeta 2013): we need Ω(n2−ε) for some graphs.(Göös 2015): we need nΩ(log0.128 n) cuts for some graphs.
For which classes of graphs does there exist a polynomialCS-Separator?
4/33
Introduction Results Relationship with other problems Conclusion
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)?What about random graphs?Theorem [Bousquet, L., Thomassé]
Random graphs admit a CS-Separator of size O(n7) asymptoticallyalmost surely.
Lower Bound:
(Huang, Sudakov 2012): we need Ω(n 65 ) cuts for some graphs.
(Amano, Shigeta 2013): we need Ω(n2−ε) for some graphs.(Göös 2015): we need nΩ(log0.128 n) cuts for some graphs.
For which classes of graphs does there exist a polynomialCS-Separator?
4/33
Introduction Results Relationship with other problems Conclusion
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)?What about random graphs?Theorem [Bousquet, L., Thomassé]
Random graphs admit a CS-Separator of size O(n7) asymptoticallyalmost surely.
Lower Bound:
(Huang, Sudakov 2012): we need Ω(n 65 ) cuts for some graphs.
(Amano, Shigeta 2013): we need Ω(n2−ε) for some graphs.(Göös 2015): we need nΩ(log0.128 n) cuts for some graphs.
For which classes of graphs does there exist a polynomialCS-Separator?
4/33
Introduction Results Relationship with other problems Conclusion
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)?What about random graphs?Theorem [Bousquet, L., Thomassé]
Random graphs admit a CS-Separator of size O(n7) asymptoticallyalmost surely.
Lower Bound:
(Huang, Sudakov 2012): we need Ω(n 65 ) cuts for some graphs.
(Amano, Shigeta 2013): we need Ω(n2−ε) for some graphs.
(Göös 2015): we need nΩ(log0.128 n) cuts for some graphs.
For which classes of graphs does there exist a polynomialCS-Separator?
4/33
Introduction Results Relationship with other problems Conclusion
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? No!What about random graphs?Theorem [Bousquet, L., Thomassé]
Random graphs admit a CS-Separator of size O(n7) asymptoticallyalmost surely.
Lower Bound:
(Huang, Sudakov 2012): we need Ω(n 65 ) cuts for some graphs.
(Amano, Shigeta 2013): we need Ω(n2−ε) for some graphs.(Göös 2015): we need nΩ(log0.128 n) cuts for some graphs.
For which classes of graphs does there exist a polynomialCS-Separator?
4/33
Introduction Results Relationship with other problems Conclusion
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? No!What about random graphs?Theorem [Bousquet, L., Thomassé]
Random graphs admit a CS-Separator of size O(n7) asymptoticallyalmost surely.
Lower Bound:
(Huang, Sudakov 2012): we need Ω(n 65 ) cuts for some graphs.
(Amano, Shigeta 2013): we need Ω(n2−ε) for some graphs.(Göös 2015): we need nΩ(log0.128 n) cuts for some graphs.
For which classes of graphs does there exist a polynomialCS-Separator?
4/33
Introduction Results Relationship with other problems Conclusion
Does there exist for all graph G on n vertices a CS-separatorof size poly(n)? No!What about random graphs?Theorem [Bousquet, L., Thomassé]
Random graphs admit a CS-Separator of size O(n7) asymptoticallyalmost surely.
Lower Bound:
(Huang, Sudakov 2012): we need Ω(n 65 ) cuts for some graphs.
(Amano, Shigeta 2013): we need Ω(n2−ε) for some graphs.(Göös 2015): we need nΩ(log0.128 n) cuts for some graphs.
For which classes of graphs does there exist a polynomialCS-Separator?
4/33
Introduction Results Relationship with other problems Conclusion
Motivation
Stable Set polytopeSTAB(G) = conv(χS ∈ Rn | S ⊆ V is a stable set of G)where χS denotes the characteristic vector of S ⊆ V .
v1
v2
v3
v1 v2 v3S = v1, v3 χS = ( 1 0 1 )S = v1, v2 χS = ( 1 1 0 )S = v1 χS = ( 1 0 0 )S = v2 χS = ( 0 1 0 )S = v3 χS = ( 0 0 1 )S = ∅ χS = ( 0 0 0 )
v1
v2
v3
5/33
Introduction Results Relationship with other problems Conclusion
Motivation
Stable Set polytopeSTAB(G) = conv(χS ∈ Rn | S ⊆ V is a stable set of G)where χS denotes the characteristic vector of S ⊆ V .
v1
v2
v3
v1 v2 v3S = v1, v3 χS = ( 1 0 1 )S = v1, v2 χS = ( 1 1 0 )S = v1 χS = ( 1 0 0 )S = v2 χS = ( 0 1 0 )S = v3 χS = ( 0 0 1 )S = ∅ χS = ( 0 0 0 ) v1
v2
v3
5/33
Introduction Results Relationship with other problems Conclusion
Motivation
Stable Set polytopeSTAB(G) = conv(χS ∈ Rn | S ⊆ V is a stable set of G)where χS denotes the characteristic vector of S ⊆ V .
v1
v2
v3
v1 v2 v3S = v1, v3 χS = ( 1 0 1 )S = v1, v2 χS = ( 1 1 0 )S = v1 χS = ( 1 0 0 )S = v2 χS = ( 0 1 0 )S = v3 χS = ( 0 0 1 )S = ∅ χS = ( 0 0 0 ) v1
v2
v3
5/33
Introduction Results Relationship with other problems Conclusion
Motivation
Stable Set polytopeSTAB(G) = conv(χS ∈ Rn | S ⊆ V is a stable set of G)where χS denotes the characteristic vector of S ⊆ V .
v1
v2
v3
v1 v2 v3S = v1, v3 χS = ( 1 0 1 )S = v1, v2 χS = ( 1 1 0 )S = v1 χS = ( 1 0 0 )S = v2 χS = ( 0 1 0 )S = v3 χS = ( 0 0 1 )S = ∅ χS = ( 0 0 0 ) v1
v2
v3
5/33
Introduction Results Relationship with other problems Conclusion
We like polytopes with a small number of facets.
P: polytope in R2 we want to optimize on (8 facets)Q: polytope in R3 which projects to P (6 facets)⇒ Easier to optimize on Q and project the solution!
6/33
Introduction Results Relationship with other problems Conclusion
Extended formulation
Extension complexity of PMinimum number of facets of apolytope Q that projects onto P.
In perfect graphs:we can describe STAB(G) with clique inequalities.
0 ≤ xv ≤ 1 ∀v ∈ V∑v∈K xv ≤ 1 ∀ clique K
Extension complexity of STAB(G) ≥ min. size of a CS-Separator
7/33
Introduction Results Relationship with other problems Conclusion
Extended formulation
Extension complexity of PMinimum number of facets of apolytope Q that projects onto P.
In perfect graphs:we can describe STAB(G) with clique inequalities.
0 ≤ xv ≤ 1 ∀v ∈ V∑v∈K xv ≤ 1 ∀ clique K
Extension complexity of STAB(G) ≥ min. size of a CS-Separator
7/33
Introduction Results Relationship with other problems Conclusion
Extended formulation
Extension complexity of PMinimum number of facets of apolytope Q that projects onto P.
In perfect graphs:we can describe STAB(G) with clique inequalities.
0 ≤ xv ≤ 1 ∀v ∈ V∑v∈K xv ≤ 1 ∀ clique K
Extension complexity of STAB(G) ≥ min. size of a CS-Separator
7/33
Introduction Results Relationship with other problems Conclusion
Reminder: definition of a CS-SeparatorA set of cuts such that for every clique K and stable set S disjointfrom K , there is a cut that separates K from S.Its size is the number of cuts.
In which classes of graphs do we have a polynomial CS-separator?
An easy example: if the clique number ω is bounded, say by 3:
For every subset T of size ≤ 3, take the cut (T ,V \ T )⇒ CS-separator of size O(n3).
8/33
Introduction Results Relationship with other problems Conclusion
Reminder: definition of a CS-SeparatorA set of cuts such that for every clique K and stable set S disjointfrom K , there is a cut that separates K from S.Its size is the number of cuts.
In which classes of graphs do we have a polynomial CS-separator?An easy example: if the clique number ω is bounded, say by 3:
For every subset T of size ≤ 3, take the cut (T ,V \ T )⇒ CS-separator of size O(n3).
8/33
Introduction Results Relationship with other problems Conclusion
Reminder: definition of a CS-SeparatorA set of cuts such that for every clique K and stable set S disjointfrom K , there is a cut that separates K from S.Its size is the number of cuts.
In which classes of graphs do we have a polynomial CS-separator?An easy example: if the clique number ω is bounded, say by 3:
For every subset T of size ≤ 3, take the cut (T ,V \ T )⇒ CS-separator of size O(n3).
8/33
Introduction Results Relationship with other problems Conclusion
Reminder: definition of a CS-SeparatorA set of cuts such that for every clique K and stable set S disjointfrom K , there is a cut that separates K from S.Its size is the number of cuts.
In which classes of graphs do we have a polynomial CS-separator?An easy example: if the clique number ω is bounded, say by 3:
For every subset T of size ≤ 3, take the cut (T ,V \ T )⇒ CS-separator of size O(n3).
8/33
Introduction Results Relationship with other problems Conclusion
Reminder: definition of a CS-SeparatorA set of cuts such that for every clique K and stable set S disjointfrom K , there is a cut that separates K from S.Its size is the number of cuts.
In which classes of graphs do we have a polynomial CS-separator?An easy example: if the clique number ω is bounded, say by 3:
For every subset T of size ≤ 3, take the cut (T ,V \ T )⇒ CS-separator of size O(n3).
8/33
Introduction Results Relationship with other problems Conclusion
The following classes of graphs admit poly-size CS-Separator:
1 chordal graphs (linear number of max. cliques)2 comparability graphs (Yannakakis 1991)3 C4-free graphs (Conseq. of Alekseev 1991)4 P5-free graphs (Conseq. of Lokshtanov, Vatchelle, Villanger 2014)
Even stronger: 1© 2© and 4© have polynomial extension complexity.
9/33
Introduction Results Relationship with other problems Conclusion
The following classes of graphs admit poly-size CS-Separator:1 chordal graphs (linear number of max. cliques)2 comparability graphs (Yannakakis 1991)3 C4-free graphs (Conseq. of Alekseev 1991)4 P5-free graphs (Conseq. of Lokshtanov, Vatchelle, Villanger 2014)
Even stronger: 1© 2© and 4© have polynomial extension complexity.
9/33
Introduction Results Relationship with other problems Conclusion
The following classes of graphs admit poly-size CS-Separator:1 chordal graphs (linear number of max. cliques)2 comparability graphs (Yannakakis 1991)3 C4-free graphs (Conseq. of Alekseev 1991)4 P5-free graphs (Conseq. of Lokshtanov, Vatchelle, Villanger 2014)
Even stronger: 1© 2© and 4© have polynomial extension complexity.
9/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
10/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
11/33
Introduction Results Relationship with other problems Conclusion
Split-free
Comparability graphs [Yannakakis 1991]Every comparability graph has a CS-separator of size O(n2).
Split-free [Bousquet, L., Thomassé]Let H be a split graph. Then every H-free graph has a CS-separatorof size O(ncH ).
12/33
Introduction Results Relationship with other problems Conclusion
Split-free
Comparability graphs [Yannakakis 1991]Every comparability graph has a CS-separator of size O(n2).
Split-free [Bousquet, L., Thomassé]Let H be a split graph. Then every H-free graph has a CS-separatorof size O(ncH ).
12/33
Introduction Results Relationship with other problems Conclusion
Split-free
Comparability graphs [Yannakakis 1991]Every comparability graph has a CS-separator of size O(n2).
Split-free [Bousquet, L., Thomassé]Let H be a split graph. Then every H-free graph has a CS-separatorof size O(ncH ).
12/33
Introduction Results Relationship with other problems Conclusion
Split-free
Comparability graphs [Yannakakis 1991]Every comparability graph has a CS-separator of size O(n2).
Split-free [Bousquet, L., Thomassé]Let H be a split graph. Then every H-free graph has a CS-separatorof size O(ncH ).
12/33
Introduction Results Relationship with other problems Conclusion
Split-free
Split graphA graph (V ,E ) is split if V can be partitioned into a clique and astable set.
Split-free [Bousquet, L., Thomassé]Let H be a split graph. Then every H-free graph has a CS-separatorof size O(ncH ).
12/33
Introduction Results Relationship with other problems Conclusion
Split-free
Split graphA graph (V ,E ) is split if V can be partitioned into a clique and astable set.
Split-free [Bousquet, L., Thomassé]Let H be a split graph. Then every H-free graph has a CS-separatorof size O(ncH ).
12/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.
Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
K S
GH-free
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
K S
GH-free
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
K S
GH-free
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
K S
GH-free
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
K S
GH-free
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
K S
GH-free
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
K S
GH-free
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Let H be a split graph and t ≈ 64|V (H)|2.Key Lemma (using VC-dimension)Let G be a H-free graph, K be a clique of G and S be a stable setdisjoint from K . Then one of the following holds:
∃S ′ ⊆ S s. t. |S ′| = t and S ′ dominates K , or∃K ′ ⊆ K s. t. |K ′| = t and K ′ antidominates S.
K S
GH-free
Union of t neighborhoods contains Kand is disjoint from S.
For every set of size ≤ t:
13/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
14/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
14/33
Introduction Results Relationship with other problems Conclusion
(Pk , Pk)-free graphs
CS-Sep. in (Pk ,Pk)-free graphs [Bousquet, L., Thomassé]Fix k. Every (Pk ,Pk)-free graph has a O(nck ) CS-Separator.
Strong Erdős-Hajnal prop. - (Pk ,Pk)-free graphsFix k. Every (Pk ,Pk)-free graph has a linear-size biclique or com-plement biclique (A,B).
A B
G
|A| ≥ ck.n |B| ≥ ck.nor
A B
G
|A| ≥ ck.n |B| ≥ ck.n
15/33
Introduction Results Relationship with other problems Conclusion
(Pk , Pk)-free graphs
CS-Sep. in (Pk ,Pk)-free graphs [Bousquet, L., Thomassé]Fix k. Every (Pk ,Pk)-free graph has a O(nck ) CS-Separator.
Strong Erdős-Hajnal prop. - (Pk ,Pk)-free graphsFix k. Every (Pk ,Pk)-free graph has a linear-size biclique or com-plement biclique (A,B).
A B
G
|A| ≥ ck.n |B| ≥ ck.nor
A B
G
|A| ≥ ck.n |B| ≥ ck.n
15/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof
Theorem (Rödl 1986, Fox, Sudakov 2008)∀k, every graph G satisfies one of the following:
G induces all graphs on k vertices.G contains a sparse induced subgraph of linear size.G contains a dense induced subgraph of linear size.
G : (Pk ,Pk)-free on n vertices.
First item does not hold.Extract a sparse induced subgraph of linear size.Extract a connected induced subgraph of linear size, withmaximum degree ≤ ε · n.
16/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof
Theorem (Rödl 1986, Fox, Sudakov 2008)∀k, every graph G satisfies one of the following:
G induces all graphs on k vertices.G contains a sparse induced subgraph of linear size.G contains a dense induced subgraph of linear size.
G : (Pk ,Pk)-free on n vertices.
First item does not hold.Extract a sparse induced subgraph of linear size.Extract a connected induced subgraph of linear size, withmaximum degree ≤ ε · n.
16/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof
Theorem (Rödl 1986, Fox, Sudakov 2008)∀k, every graph G satisfies one of the following:
G induces all graphs on k vertices.G contains a sparse induced subgraph of linear size.G contains a dense induced subgraph of linear size.
G : (Pk ,Pk)-free on n vertices.
First item does not hold.
Extract a sparse induced subgraph of linear size.Extract a connected induced subgraph of linear size, withmaximum degree ≤ ε · n.
16/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof
Theorem (Rödl 1986, Fox, Sudakov 2008)∀k, every graph G satisfies one of the following:
G induces all graphs on k vertices.G contains a sparse induced subgraph of linear size.G contains a dense induced subgraph of linear size.
G : (Pk ,Pk)-free on n vertices.
First item does not hold.Extract a sparse induced subgraph of linear size.
Extract a connected induced subgraph of linear size, withmaximum degree ≤ ε · n.
16/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof
Theorem (Rödl 1986, Fox, Sudakov 2008)∀k, every graph G satisfies one of the following:
G induces all graphs on k vertices.G contains a sparse induced subgraph of linear size.G contains a dense induced subgraph of linear size.
G : (Pk ,Pk)-free on n vertices.
First item does not hold.Extract a sparse induced subgraph of linear size.Extract a connected induced subgraph of linear size, withmaximum degree ≤ ε · n.
16/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.
LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
v
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
v
N(v) G \N [v]small
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
v
N(v) G \N [v]small
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
v
N(v) G \N [v]small
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
v
N(v)small
G \N [v]
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
v
N(v)small
v′
G \N [v]
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
v
N(v)small
v′
G \N [v]
17/33
Introduction Results Relationship with other problems Conclusion
Sketch of proof (continued)Connected Pk -free subgraph G ′ on n′ vertices, max degree ≤ ε · n′.LemmaFor all k, if G is a connected graph where:
There is no vertex of degree ≥ εk · n, orThere does not exist an antibiclique of size ≥ ck · n.
Then for every vertex v , there exists a Pk starting at v ,
v
N(v)small
G \N [v]
v′
17/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
18/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
18/33
Introduction Results Relationship with other problems Conclusion
Main step in the proof of the Strong Perfect Graph Theorem:Decomposition [Chudnovsky, Robertson, Seymour, Thomas 2002]
If a graph is Berge, then for G or G , one of the following holds :It is a basic graph: bipartite, line graph of bip., or double split.There is a 2-join.There is a balanced skew partition.
A1
B1
C1
A2
B2
C2
A1
A2
B1 B2
2-Join Skew Partition
Theorem [L., Trunck]Let G be a perfect graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
19/33
Introduction Results Relationship with other problems Conclusion
Main step in the proof of the Strong Perfect Graph Theorem:Decomposition [Chudnovsky, Robertson, Seymour, Thomas 2002]
If a graph is Berge, then for G or G , one of the following holds :It is a basic graph: bipartite, line graph of bip., or double split.There is a 2-join.There is a balanced skew partition.
A1
B1
C1
A2
B2
C2
A1
A2
B1 B2
2-Join Skew Partition
Theorem [L., Trunck]Let G be a perfect graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
19/33
Introduction Results Relationship with other problems Conclusion
Strong Erdős-Hajnal property [L., Trunck]Every perfect graph with no balanced skew partition admits a bi-clique or complement biclique of size at least n/148.
A B
G
|A| ≥ n148 |B| ≥ n
148
A B
G
|A| ≥ n148 |B| ≥ n
148
Not hereditary class ⇒ cannot directly deduce the poly CS-sep.
But there exist perfect graphs where the Strong Erdős-Hajnalproperty does not hold [Fox, Pach 2009] ⇒ Evidence of somespecial structure.
20/33
Introduction Results Relationship with other problems Conclusion
Strong Erdős-Hajnal property [L., Trunck]Every perfect graph with no balanced skew partition admits a bi-clique or complement biclique of size at least n/148.
A B
G
|A| ≥ n148 |B| ≥ n
148
A B
G
|A| ≥ n148 |B| ≥ n
148
Not hereditary class ⇒ cannot directly deduce the poly CS-sep.
But there exist perfect graphs where the Strong Erdős-Hajnalproperty does not hold [Fox, Pach 2009] ⇒ Evidence of somespecial structure.
20/33
Introduction Results Relationship with other problems Conclusion
Strong Erdős-Hajnal property [L., Trunck]Every perfect graph with no balanced skew partition admits a bi-clique or complement biclique of size at least n/148.
A B
G
|A| ≥ n148 |B| ≥ n
148
A B
G
|A| ≥ n148 |B| ≥ n
148
Not hereditary class ⇒ cannot directly deduce the poly CS-sep.
But there exist perfect graphs where the Strong Erdős-Hajnalproperty does not hold [Fox, Pach 2009] ⇒ Evidence of somespecial structure.
20/33
Introduction Results Relationship with other problems Conclusion
Strong Erdős-Hajnal property [L., Trunck]Every perfect graph with no balanced skew partition admits a bi-clique or complement biclique of size at least n/148.
A B
G
|A| ≥ n148 |B| ≥ n
148
A B
G
|A| ≥ n148 |B| ≥ n
148
Not hereditary class ⇒ cannot directly deduce the poly CS-sep.
But there exist perfect graphs where the Strong Erdős-Hajnalproperty does not hold [Fox, Pach 2009] ⇒ Evidence of somespecial structure.
20/33
Introduction Results Relationship with other problems Conclusion
Let G be a Berge graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
Proof by induction:
For basic graphs: direct proof.For a graph G with a 2-join :
From G , we build two "half" graphs G1 and G2, eachcorresponding to a side of the 2-join + a gadget.Check that G1 and G2 are still Berge with no balanced skewpartition [Chudnovsky, Trotignon, Trunck, Vušković 2012]Get CS-separators for G1 and G2 by induction hypothesisTransform them into a CS-separator for G .
21/33
Introduction Results Relationship with other problems Conclusion
Let G be a Berge graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
Proof by induction:
For basic graphs: direct proof.For a graph G with a 2-join :
From G , we build two "half" graphs G1 and G2, eachcorresponding to a side of the 2-join + a gadget.Check that G1 and G2 are still Berge with no balanced skewpartition [Chudnovsky, Trotignon, Trunck, Vušković 2012]Get CS-separators for G1 and G2 by induction hypothesisTransform them into a CS-separator for G .
21/33
Introduction Results Relationship with other problems Conclusion
Let G be a Berge graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
Proof by induction:
For basic graphs: direct proof.
For a graph G with a 2-join :
From G , we build two "half" graphs G1 and G2, eachcorresponding to a side of the 2-join + a gadget.Check that G1 and G2 are still Berge with no balanced skewpartition [Chudnovsky, Trotignon, Trunck, Vušković 2012]Get CS-separators for G1 and G2 by induction hypothesisTransform them into a CS-separator for G .
21/33
Introduction Results Relationship with other problems Conclusion
Let G be a Berge graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
Proof by induction:
For basic graphs: direct proof.For a graph G with a 2-join :
From G , we build two "half" graphs G1 and G2, eachcorresponding to a side of the 2-join + a gadget.Check that G1 and G2 are still Berge with no balanced skewpartition [Chudnovsky, Trotignon, Trunck, Vušković 2012]Get CS-separators for G1 and G2 by induction hypothesisTransform them into a CS-separator for G .
21/33
Introduction Results Relationship with other problems Conclusion
Let G be a Berge graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
Proof by induction:
For basic graphs: direct proof.For a graph G with a 2-join :
From G , we build two "half" graphs G1 and G2, eachcorresponding to a side of the 2-join + a gadget.
Check that G1 and G2 are still Berge with no balanced skewpartition [Chudnovsky, Trotignon, Trunck, Vušković 2012]Get CS-separators for G1 and G2 by induction hypothesisTransform them into a CS-separator for G .
21/33
Introduction Results Relationship with other problems Conclusion
Let G be a Berge graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
Proof by induction:
For basic graphs: direct proof.For a graph G with a 2-join :
From G , we build two "half" graphs G1 and G2, eachcorresponding to a side of the 2-join + a gadget.Check that G1 and G2 are still Berge with no balanced skewpartition [Chudnovsky, Trotignon, Trunck, Vušković 2012]
Get CS-separators for G1 and G2 by induction hypothesisTransform them into a CS-separator for G .
21/33
Introduction Results Relationship with other problems Conclusion
Let G be a Berge graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
Proof by induction:
For basic graphs: direct proof.For a graph G with a 2-join :
From G , we build two "half" graphs G1 and G2, eachcorresponding to a side of the 2-join + a gadget.Check that G1 and G2 are still Berge with no balanced skewpartition [Chudnovsky, Trotignon, Trunck, Vušković 2012]Get CS-separators for G1 and G2 by induction hypothesis
Transform them into a CS-separator for G .
21/33
Introduction Results Relationship with other problems Conclusion
Let G be a Berge graph with no balanced skew partition, then thereexists a CS-separator for G of size O(n2).
Proof by induction:
For basic graphs: direct proof.For a graph G with a 2-join :
From G , we build two "half" graphs G1 and G2, eachcorresponding to a side of the 2-join + a gadget.Check that G1 and G2 are still Berge with no balanced skewpartition [Chudnovsky, Trotignon, Trunck, Vušković 2012]Get CS-separators for G1 and G2 by induction hypothesisTransform them into a CS-separator for G .
21/33
Introduction Results Relationship with other problems Conclusion
? ?
A1
B1
C1
A2
B2
C2
C1
A1
B1
a2
b2
C1
A1
B1 b2
c2
a2
22/33
Introduction Results Relationship with other problems Conclusion
? ?
A1
B1
C1
A2
B2
C2
C1
A1
B1
a2
b2
C1
A1
B1 b2
c2
a2
22/33
Introduction Results Relationship with other problems Conclusion
? ?
A1
B1
C1
A2
B2
C2
C1
A1
B1
a2
b2
A2
B2
C2
a1
b1
22/33
Introduction Results Relationship with other problems Conclusion
? ?
A1
B1
C1
A2
B2
C2
C1
A1
B1
a2
b2
A2
B2
C2
a1
b1
22/33
Introduction Results Relationship with other problems Conclusion
? ?
A1
B1
C1
A2
B2
C2
C1
A1
B1
a2
b2
A2
B2
C2
a1
b1
22/33
Introduction Results Relationship with other problems Conclusion
? ?
A1
B1
C1
A2
B2
C2
A1
B1
C1
a2
b2
A2
B2
C2
a1
b1
22/33
Introduction Results Relationship with other problems Conclusion
Red=What is puton the left(clique side)Green=What is puton the right(stable set s.)
A1
B1
C1
A2
B2
C2
A1
B1
C1
a2
b2
A2
B2
C2
a1
b1
23/33
Introduction Results Relationship with other problems Conclusion
Red=What is puton the left(clique side)Green=What is puton the right(stable set s.)
A1
B1
C1
A2
B2
C2
A1
B1
C1
a2
b2
A2
B2
C2
a1
b1
23/33
Introduction Results Relationship with other problems Conclusion
Red=What is puton the left(clique side)Green=What is puton the right(stable set s.)
A1
B1
C1
A2
B2
C2
A1
B1
C1
a2
b2
A2
B2
C2
a1
b1
23/33
Introduction Results Relationship with other problems Conclusion
Red=What is puton the left(clique side)Green=What is puton the right(stable set s.)
A2
B2
C2C1
A1
B1
A1
B1
C1
a2
b2
A2
B2
C2
a1
b1
23/33
Introduction Results Relationship with other problems Conclusion
Red=What is puton the left(clique side)Green=What is puton the right(stable set s.)
A2
B2
C2C1
A1
B1
a2
b2
C1
A1
B1
A2
B2
C2
a1
b1
23/33
Introduction Results Relationship with other problems Conclusion
Red=What is puton the left(clique side)Green=What is puton the right(stable set s.)
A2
B2
C2C1
A1
B1
a2
b2
A1
B1
C1C1
A1
B1
A2
B2
C2
a1
b1
23/33
Introduction Results Relationship with other problems Conclusion
Red=What is puton the left(clique side)Green=What is puton the right(stable set s.)
A2
B2
C2
B2
C2
A1
B1
C1C1
A1
B1
A2
a2
b2
A1
B1
C1C1
A1
B1
A2
B2
C2
a1
b1
23/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
24/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
24/33
Introduction Results Relationship with other problems Conclusion
Another toolGeneralization of "having a simplicial vertex".Def: k-easy-neighborhood property in a class C∀G ∈ C, ∃v ∈ V (G) s.t. G [N(v)] admits a O(|N(v)|k) CS-sep.
If C is hereditary and the k-easy-neighborhood property holds, thenevery G ∈ C has a O(nk+1) CS-Separator.
25/33
Introduction Results Relationship with other problems Conclusion
Another toolGeneralization of "having a simplicial vertex".Def: k-easy-neighborhood property in a class C∀G ∈ C, ∃v ∈ V (G) s.t. G [N(v)] admits a O(|N(v)|k) CS-sep.
If C is hereditary and the k-easy-neighborhood property holds, thenevery G ∈ C has a O(nk+1) CS-Separator.
v
N(v)
CS-sep O(nk)V2
25/33
Introduction Results Relationship with other problems Conclusion
Another toolGeneralization of "having a simplicial vertex".Def: k-easy-neighborhood property in a class C∀G ∈ C, ∃v ∈ V (G) s.t. G [N(v)] admits a O(|N(v)|k) CS-sep.
If C is hereditary and the k-easy-neighborhood property holds, thenevery G ∈ C has a O(nk+1) CS-Separator.
v
N(v)
CS-sep O(nk)V2
ind. hyp
v in stable set side
Cuts every (K,S) with v ∈ S
25/33
Introduction Results Relationship with other problems Conclusion
Another toolGeneralization of "having a simplicial vertex".Def: k-easy-neighborhood property in a class C∀G ∈ C, ∃v ∈ V (G) s.t. G [N(v)] admits a O(|N(v)|k) CS-sep.
If C is hereditary and the k-easy-neighborhood property holds, thenevery G ∈ C has a O(nk+1) CS-Separator.
v
N(v)
CS-sep O(nk)V2
CS-sep O(nk)
v in clique side
V2 in stable set side
Cuts every (K,S) with v ∈ K
25/33
Introduction Results Relationship with other problems Conclusion
Windmill-free graphs
A 4-windmill
Let G be a k-windmill-free graph.For every vertex v , the graph G ′ = G [N(v)] is kK2-free.⇒ G ′ has O(|V (G ′)|2k−2) maximal stable sets (Alekseev 1991).⇒ G ′ has a CS-Separator, hence v has an easy neighborhood.
Fix k. Every k-windmill-free graph has a O(n2k−1) CS-Separator.
26/33
Introduction Results Relationship with other problems Conclusion
Windmill-free graphs
A 4-windmill
Let G be a k-windmill-free graph.
For every vertex v , the graph G ′ = G [N(v)] is kK2-free.⇒ G ′ has O(|V (G ′)|2k−2) maximal stable sets (Alekseev 1991).⇒ G ′ has a CS-Separator, hence v has an easy neighborhood.
Fix k. Every k-windmill-free graph has a O(n2k−1) CS-Separator.
26/33
Introduction Results Relationship with other problems Conclusion
Windmill-free graphs
A 4-windmill
Let G be a k-windmill-free graph.For every vertex v , the graph G ′ = G [N(v)] is kK2-free.
⇒ G ′ has O(|V (G ′)|2k−2) maximal stable sets (Alekseev 1991).⇒ G ′ has a CS-Separator, hence v has an easy neighborhood.
Fix k. Every k-windmill-free graph has a O(n2k−1) CS-Separator.
26/33
Introduction Results Relationship with other problems Conclusion
Windmill-free graphs
A 4-windmill
Let G be a k-windmill-free graph.For every vertex v , the graph G ′ = G [N(v)] is kK2-free.⇒ G ′ has O(|V (G ′)|2k−2) maximal stable sets (Alekseev 1991).
⇒ G ′ has a CS-Separator, hence v has an easy neighborhood.
Fix k. Every k-windmill-free graph has a O(n2k−1) CS-Separator.
26/33
Introduction Results Relationship with other problems Conclusion
Windmill-free graphs
A 4-windmill
Let G be a k-windmill-free graph.For every vertex v , the graph G ′ = G [N(v)] is kK2-free.⇒ G ′ has O(|V (G ′)|2k−2) maximal stable sets (Alekseev 1991).⇒ G ′ has a CS-Separator, hence v has an easy neighborhood.
Fix k. Every k-windmill-free graph has a O(n2k−1) CS-Separator.
26/33
Introduction Results Relationship with other problems Conclusion
Windmill-free graphs
A 4-windmill
Let G be a k-windmill-free graph.For every vertex v , the graph G ′ = G [N(v)] is kK2-free.⇒ G ′ has O(|V (G ′)|2k−2) maximal stable sets (Alekseev 1991).⇒ G ′ has a CS-Separator, hence v has an easy neighborhood.
Fix k. Every k-windmill-free graph has a O(n2k−1) CS-Separator.
26/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
27/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
27/33
Introduction Results Relationship with other problems Conclusion
Tools and results about the CS-Separation
Comparability graphsLP to compute maximumweighted stable set.
When α or ω is bounded,chordal graphs, C4-free, ....Polynomially many maximalcliques or stable sets.
P5-free graphsIn-depth study of potentialmaximal cliques.
H-free graphs when H is splitCan define cuts with a con-stant nb of neighborhoods.
(Pk ,Pk)-free graphsStrong Erdős-Hajnal property.
Perfect graphs with no BSPDecomposition by 2-joins.
k-windmill-freeEasy neighborhood property.
Homogeneous set
27/33
Introduction Results Relationship with other problems Conclusion
Relationship with otherproblems
28/33
Introduction Results Relationship with other problems Conclusion
Erdős-Hajnal propertyCan we always find a large clique or a large stable set?
Ramsey TheoryEvery graph has a clique or a stable set of logarithmic size.
⇒ Logarithmic order is best possible (Erdős 1947).
Definition: The Erdős-Hajnal propertyA class C of graphs has the Erdős-Hajnal property if there existsβ > 0 such that:every G ∈ C has a clique or a stable set of size at least |V (G)|β.
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.
29/33
Introduction Results Relationship with other problems Conclusion
Erdős-Hajnal propertyCan we always find a large clique or a large stable set?
Ramsey TheoryEvery graph has a clique or a stable set of logarithmic size.
⇒ Logarithmic order is best possible (Erdős 1947).
Definition: The Erdős-Hajnal propertyA class C of graphs has the Erdős-Hajnal property if there existsβ > 0 such that:every G ∈ C has a clique or a stable set of size at least |V (G)|β.
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.
29/33
Introduction Results Relationship with other problems Conclusion
Erdős-Hajnal propertyCan we always find a large clique or a large stable set?
Ramsey TheoryEvery graph has a clique or a stable set of logarithmic size.
⇒ Logarithmic order is best possible (Erdős 1947).
Definition: The Erdős-Hajnal propertyA class C of graphs has the Erdős-Hajnal property if there existsβ > 0 such that:every G ∈ C has a clique or a stable set of size at least |V (G)|β.
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.
29/33
Introduction Results Relationship with other problems Conclusion
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.True when C is the class of:
perfect graphsH-free graphs, when H has at most 4 verticesbull-free graphs (Chudnovsky, Safra 2008)
Next interesting cases: P5-free or C5-free graphs.
Towards P5-free graphs?(P5,P5)-free graphs (Fouquet 1993)(P5,P6)-free graphs (Chudnovsky, Zwols 2012)(P5,P7)-free graphs (Chudnovsky, Seymour 2012)
30/33
Introduction Results Relationship with other problems Conclusion
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.True when C is the class of:
perfect graphs
H-free graphs, when H has at most 4 verticesbull-free graphs (Chudnovsky, Safra 2008)
Next interesting cases: P5-free or C5-free graphs.
Towards P5-free graphs?(P5,P5)-free graphs (Fouquet 1993)(P5,P6)-free graphs (Chudnovsky, Zwols 2012)(P5,P7)-free graphs (Chudnovsky, Seymour 2012)
30/33
Introduction Results Relationship with other problems Conclusion
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.True when C is the class of:
perfect graphsH-free graphs, when H has at most 4 vertices
bull-free graphs (Chudnovsky, Safra 2008)
Next interesting cases: P5-free or C5-free graphs.
Towards P5-free graphs?(P5,P5)-free graphs (Fouquet 1993)(P5,P6)-free graphs (Chudnovsky, Zwols 2012)(P5,P7)-free graphs (Chudnovsky, Seymour 2012)
30/33
Introduction Results Relationship with other problems Conclusion
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.True when C is the class of:
perfect graphsH-free graphs, when H has at most 4 verticesbull-free graphs (Chudnovsky, Safra 2008)
Next interesting cases: P5-free or C5-free graphs.
Towards P5-free graphs?(P5,P5)-free graphs (Fouquet 1993)(P5,P6)-free graphs (Chudnovsky, Zwols 2012)(P5,P7)-free graphs (Chudnovsky, Seymour 2012)
30/33
Introduction Results Relationship with other problems Conclusion
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.True when C is the class of:
perfect graphsH-free graphs, when H has at most 4 verticesbull-free graphs (Chudnovsky, Safra 2008)
Next interesting cases: P5-free or C5-free graphs.
Towards P5-free graphs?(P5,P5)-free graphs (Fouquet 1993)(P5,P6)-free graphs (Chudnovsky, Zwols 2012)(P5,P7)-free graphs (Chudnovsky, Seymour 2012)
30/33
Introduction Results Relationship with other problems Conclusion
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.True when C is the class of:
perfect graphsH-free graphs, when H has at most 4 verticesbull-free graphs (Chudnovsky, Safra 2008)
Next interesting cases: P5-free or C5-free graphs.
Towards P5-free graphs?(P5,P5)-free graphs (Fouquet 1993)(P5,P6)-free graphs (Chudnovsky, Zwols 2012)(P5,P7)-free graphs (Chudnovsky, Seymour 2012)
30/33
Introduction Results Relationship with other problems Conclusion
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.True when C is the class of:
perfect graphsH-free graphs, when H has at most 4 verticesbull-free graphs (Chudnovsky, Safra 2008)
Next interesting cases: P5-free or C5-free graphs.
Towards P5-free graphs?
(P5,P5)-free graphs (Fouquet 1993)(P5,P6)-free graphs (Chudnovsky, Zwols 2012)(P5,P7)-free graphs (Chudnovsky, Seymour 2012)
30/33
Introduction Results Relationship with other problems Conclusion
The Erdős-Hajnal ConjectureEvery strict hereditary class C has the Erdős-Hajnal property.True when C is the class of:
perfect graphsH-free graphs, when H has at most 4 verticesbull-free graphs (Chudnovsky, Safra 2008)
Next interesting cases: P5-free or C5-free graphs.
Towards P5-free graphs?(P5,P5)-free graphs (Fouquet 1993)(P5,P6)-free graphs (Chudnovsky, Zwols 2012)(P5,P7)-free graphs (Chudnovsky, Seymour 2012)
30/33
Introduction Results Relationship with other problems Conclusion
Erdős-Hajnal prop. - (Pk ,Pk)-free [Bousquet, L., Thomassé]
There exists βk > 0 such that every (Pk ,Pk)-free graph G has aclique or a stable set of size |V (G)|βk .
Strong Erdős-Hajnal prop. - (Pk ,Pk)-free [Bousquet, L., Thomassé]
For every k, every graph G with no Pk nor Pk has a linear-sizebiclique or antibiclique (A,B).
A B
G
|A| ≥ ck.n |B| ≥ ck.nor
A B
G
|A| ≥ ck.n |B| ≥ ck.n
31/33
Introduction Results Relationship with other problems Conclusion
Erdős-Hajnal prop. - (Pk ,Pk)-free [Bousquet, L., Thomassé]
There exists βk > 0 such that every (Pk ,Pk)-free graph G has aclique or a stable set of size |V (G)|βk .
Strong Erdős-Hajnal prop. - (Pk ,Pk)-free [Bousquet, L., Thomassé]
For every k, every graph G with no Pk nor Pk has a linear-sizebiclique or antibiclique (A,B).
A B
G
|A| ≥ ck.n |B| ≥ ck.nor
A B
G
|A| ≥ ck.n |B| ≥ ck.n
31/33
Introduction Results Relationship with other problems Conclusion
Alon-Saks-Seymour conjecture
Alon-Saks-Seymour conjecture (1991)For every graph G , χ(G) ≤ bp(G) + 1.
bp(G)= min. nb of complete bipartite graphs to partition E (G).
True when G = Kn (Graham-Pollack theorem, 1972)Disproved in 2012 by Huang and Sudakov: χ(G) ≥ bp(G)6/5
for some graphs.
"Polynomial Alon-Saks-Seymour conjecture"For every graph G , χ(G) ≤ poly(bp(G)).
⇔Poly CS-Separation for every graphEvery graph G admits a CS-Separator of size poly(|V (G)|).
Disproved by Göös, 2015.
32/33
Introduction Results Relationship with other problems Conclusion
Alon-Saks-Seymour conjecture
Alon-Saks-Seymour conjecture (1991)For every graph G , χ(G) ≤ bp(G) + 1.
bp(G)= min. nb of complete bipartite graphs to partition E (G).True when G = Kn (Graham-Pollack theorem, 1972)
Disproved in 2012 by Huang and Sudakov: χ(G) ≥ bp(G)6/5
for some graphs.
"Polynomial Alon-Saks-Seymour conjecture"For every graph G , χ(G) ≤ poly(bp(G)).
⇔Poly CS-Separation for every graphEvery graph G admits a CS-Separator of size poly(|V (G)|).
Disproved by Göös, 2015.
32/33
Introduction Results Relationship with other problems Conclusion
Alon-Saks-Seymour conjecture
Alon-Saks-Seymour conjecture (1991)For every graph G , χ(G) ≤ bp(G) + 1.
bp(G)= min. nb of complete bipartite graphs to partition E (G).True when G = Kn (Graham-Pollack theorem, 1972)Disproved in 2012 by Huang and Sudakov: χ(G) ≥ bp(G)6/5
for some graphs.
"Polynomial Alon-Saks-Seymour conjecture"For every graph G , χ(G) ≤ poly(bp(G)).
⇔Poly CS-Separation for every graphEvery graph G admits a CS-Separator of size poly(|V (G)|).
Disproved by Göös, 2015.
32/33
Introduction Results Relationship with other problems Conclusion
Alon-Saks-Seymour conjecture
Alon-Saks-Seymour conjecture (1991)For every graph G , χ(G) ≤ bp(G) + 1.
bp(G)= min. nb of complete bipartite graphs to partition E (G).True when G = Kn (Graham-Pollack theorem, 1972)Disproved in 2012 by Huang and Sudakov: χ(G) ≥ bp(G)6/5
for some graphs.
"Polynomial Alon-Saks-Seymour conjecture"For every graph G , χ(G) ≤ poly(bp(G)).
⇔Poly CS-Separation for every graphEvery graph G admits a CS-Separator of size poly(|V (G)|).
Disproved by Göös, 2015.
32/33
Introduction Results Relationship with other problems Conclusion
Alon-Saks-Seymour conjecture
Alon-Saks-Seymour conjecture (1991)For every graph G , χ(G) ≤ bp(G) + 1.
bp(G)= min. nb of complete bipartite graphs to partition E (G).True when G = Kn (Graham-Pollack theorem, 1972)Disproved in 2012 by Huang and Sudakov: χ(G) ≥ bp(G)6/5
for some graphs.
"Polynomial Alon-Saks-Seymour conjecture"For every graph G , χ(G) ≤ poly(bp(G)).
⇔Poly CS-Separation for every graphEvery graph G admits a CS-Separator of size poly(|V (G)|).
Disproved by Göös, 2015.
32/33
Introduction Results Relationship with other problems Conclusion
Alon-Saks-Seymour conjecture
Alon-Saks-Seymour conjecture (1991)For every graph G , χ(G) ≤ bp(G) + 1.
bp(G)= min. nb of complete bipartite graphs to partition E (G).True when G = Kn (Graham-Pollack theorem, 1972)Disproved in 2012 by Huang and Sudakov: χ(G) ≥ bp(G)6/5
for some graphs.
"Polynomial Alon-Saks-Seymour conjecture"For every graph G , χ(G) ≤ poly(bp(G)).
⇔Poly CS-Separation for every graphEvery graph G admits a CS-Separator of size poly(|V (G)|).
Disproved by Göös, 2015.32/33
Introduction Results Relationship with other problems Conclusion
Perspectives
Polynomial CS-Separation in perfect graphs?
Extending results on CS-Separation to extension complexity?Find polynomial CS-Separators in some more classes ofgraphs?
Explore relationship between different properties of classes:CS-Separationχ-boundednessErdős-Hajnal propertycomputing α is polynomial-time solvable...
Thank you for your attention!
33/33
Introduction Results Relationship with other problems Conclusion
Perspectives
Polynomial CS-Separation in perfect graphs?Extending results on CS-Separation to extension complexity?
Find polynomial CS-Separators in some more classes ofgraphs?
Explore relationship between different properties of classes:CS-Separationχ-boundednessErdős-Hajnal propertycomputing α is polynomial-time solvable...
Thank you for your attention!
33/33
Introduction Results Relationship with other problems Conclusion
Perspectives
Polynomial CS-Separation in perfect graphs?Extending results on CS-Separation to extension complexity?Find polynomial CS-Separators in some more classes ofgraphs?
Explore relationship between different properties of classes:CS-Separationχ-boundednessErdős-Hajnal propertycomputing α is polynomial-time solvable...
Thank you for your attention!
33/33
Introduction Results Relationship with other problems Conclusion
Perspectives
Polynomial CS-Separation in perfect graphs?Extending results on CS-Separation to extension complexity?Find polynomial CS-Separators in some more classes ofgraphs?
Explore relationship between different properties of classes:CS-Separationχ-boundednessErdős-Hajnal propertycomputing α is polynomial-time solvable...
Thank you for your attention!
33/33
Introduction Results Relationship with other problems Conclusion
Perspectives
Polynomial CS-Separation in perfect graphs?Extending results on CS-Separation to extension complexity?Find polynomial CS-Separators in some more classes ofgraphs?
Explore relationship between different properties of classes:CS-Separationχ-boundednessErdős-Hajnal propertycomputing α is polynomial-time solvable...
Thank you for your attention!
33/33
Introduction Results Relationship with other problems Conclusion
Perspectives
Polynomial CS-Separation in perfect graphs?Extending results on CS-Separation to extension complexity?Find polynomial CS-Separators in some more classes ofgraphs?
Explore relationship between different properties of classes:CS-Separationχ-boundednessErdős-Hajnal propertycomputing α is polynomial-time solvable...
Thank you for your attention!33/33
top related