(Nearly)-tight bounds on the linearity and contiguity of cographs

Transcription

1 Bordeaux Graph Workshop 21/11/ Bordeaux (Nearly)-tight bounds on the linearity and contiguity of cographs Christophe Crespelle LI Université de Lyon INRIA hilippe Gambette LIGM Université aris-est Marne-la-Vallée

2 Outline Contiguity and linearity Cographs and cotrees A min-max theorem on the rank of a tree Upper bounds with caterpillar decompositions Lower bounds with claws erspectives

3 Outline Contiguity and linearity Cographs and cotrees A min-max theorem on the rank of a tree Upper bounds with caterpillar decompositions Lower bounds with claws erspectives

4 Contiguity a graph G a b c f h d e g a closed-k-interval model of G a d b c e f h g order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ closed contiguity of G, cc(g): smallest k / G has a closed-k-interval model

5 Contiguity a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ N[a]: a d b c e f h g closed contiguity of G, cc(g): smallest k / G has a closed-k-interval model

6 Contiguity a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ a d b c e f h g N[a]: k 2 closed contiguity of G, cc(g): smallest k / G has a closed-k-interval model

7 Contiguity a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ a d b c e f h g N[b]: k 2 closed contiguity of G, cc(g): smallest k / G has a closed-k-interval model

8 Contiguity a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ a d b c e f h g N[c]: k 2 closed contiguity of G, cc(g): smallest k / G has a closed-k-interval model

9 Contiguity a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ... a d b c e f h g k = 2 cc(g) 2 closed contiguity of G, cc(g): smallest k / G has a closed-k-interval model

10 Contiguity a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ... a d b c e f h g k = 2 cc(g) 2 A min-max parameter: cc(g) = min max cc G,σ (v) σ v closed contiguity of G, cc(g): smallest k / G has a closed-k-interval model

11 Contiguity a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ... a d b c e f h g k = 2 cc(g) 2 A min-max parameter: cc(g) = min max cc G,σ (v) σ v closed contiguity of G, cc(g): smallest k / G has a closed-k-interval model compact representation of G

12 Contiguity and consecutives ones a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ a d b c e f h g a d b c e f h g adjacency matrix A of G G has closed contiguity k if the lines and columns of the adjacency matric A of G can be reordered so that they contain at most k blocks of consecutives ones.

13 Contiguity and consecutives ones a graph G a b c f h d e g a closed-k-interval model of G order σ where the closed neighborhood of each vertex is the union of at most k intervals of σ a d b c e f h g a d b c e f h g cc(g) 2 G has closed contiguity k if the lines and columns of the adjacency matric A of G can be reordered so that they contain at most k blocks of consecutives ones.

14 Contiguity a graph G a b c f h d e g an open-k-interval model of G order σ where the open neighborhood of each vertex is the union of at most k intervals of σ a d b c e f h g N(a): k 2 open contiguity of G, oc(g): smallest k / G has an open-k-interval model

15 Contiguity a graph G a b c f h d e g an open-k-interval model of G order σ where the open neighborhood of each vertex is the union of at most k intervals of σ N(h): a d b c e f h g k 3 oc(g) 3 open contiguity of G, oc(g): smallest k / G has an open-k-interval model

16 Contiguity a graph G a b c f h d e g an open-k-interval model of G order σ where the open neighborhood of each vertex is the union of at most k intervals of σ N(h): a d b c e f h g k 3 oc(g) 3 Remark: oc(g) cc(g)+1 ; cc(g) oc(g)+1 open contiguity of G, oc(g): smallest k / G has an open-k-interval model

17 Linearity a graph G a b c f h d e g a closed-k-line model of G k orders where the closed neighborhood of each vertex is the union of one interval per order N[a]: a d b c e f h g k 2 a d b c e f g h closed linearity of G, cl(g): smallest k / G has a closed-k-line model

18 Linearity a graph G a b c f h d e g an open-k-line model of G k orders where the open neighborhood of each vertex is the union of one interval per order N(h): a d b c e f h g k 2 a d b c e f g h open linearity of G, ol(g): smallest k / G has an open-k-line model

19 Linearity a graph G a b c f h d e g an open-k-line model of G k orders where the open neighborhood of each vertex is the union of one interval per order N(h): a d b c e f h g k 2 a d b c e f g h Remark: ol(g) cl(g)+1 ; cl(g) ol(g)+1 cl(g) cc(g) (replicate same order) open linearity of G, ol(g): smallest k / G has an open-k-line model

20 Computing the contiguity/linearity cc(g) = 1 cl(g) = 1 G unit interval graph oc(g) = 1 ol(g) = 1 G biconvex graph Given a fixed k 2, cc(g) = k? oc(g) = k? N-complete Wang, Lau & Zhao, DAM, 2007 Bounds: For any graph G, cl(g) cc(g) n/4+o( n log n ) Gavoille & eleg, SIAM JoDM, 1999 There exist interval graphs and permutation graphs with n vertices and with closed contiguity at least O(log n) / closed linearity at least O(log n / log log n) Crespelle & Gambette, IWOCA 2009

21 Computing the contiguity/linearity cc(g) = 1 cl(g) = 1 G unit interval graph oc(g) = 1 ol(g) = 1 G biconvex graph Given a fixed k 2, cc(g) = k? oc(g) = k? N-complete Wang, Lau & Zhao, DAM, 2007 Bounds: For any graph G, cl(g) cc(g) n/4+o( n log n ) Gavoille & eleg, SIAM JoDM, 1999 There exist interval graphs and permutation graphs with n vertices and with closed contiguity at least O(log n) / closed linearity at least O(log n / log log n) Crespelle & Gambette, IWOCA 2009 upper bound?

22 Outline Contiguity and linearity Cographs and cotrees A min-max theorem on the rank of a tree Upper bounds with caterpillar decompositions Lower bounds with claws erspectives

23 Cographs cograph G = graph without induced 4 graph built by series composition (series operation) and disjoint union (parallel operation) cotree T T b c G a b c a Two vertices adjacent in G iff their lowest common ancestor in T is a series node

24 Cographs cograph G = graph without induced 4 graph built by series composition (series operation) and disjoint union (parallel operation) cotree T T S c d b G a b c d a Two vertices adjacent in G iff their lowest common ancestor in T is a series node

25 Cographs cograph G = graph without induced 4 graph built by series composition (series operation) and disjoint union (parallel operation) cotree T d c T S S a b G a b c d e f g e g f Two vertices adjacent in G iff their lowest common ancestor in T is a series node

26 Cographs cograph G = graph without induced 4 graph built by series composition (series operation) and disjoint union (parallel operation) cotree T d r c S T G b S S a h a b c d e f g h e g f Two vertices adjacent in G iff their lowest common ancestor in T is a series node

27 Cographs cograph G = graph without induced 4 graph built by series composition (series operation) and disjoint union (parallel operation) cotree T d r c S T G b S S a h a b c d e f g h Useful properties: e g f Complement: For any graph G, cc(g) cc(g)+1 Series & parallel operation: For any graphs G and H, cc(s(g,h) max(cc(g),cc(h))+1, cc(p(g,h) max(cc(g),cc(h)) cl(s(g,h) max(cl(g),cl(h))+1, cl(p(g,h) max(cl(g),cl(h))

28 Cographs cograph G = graph without induced 4 graph built by series composition (series operation) and disjoint union (parallel operation) cotree T d r c S T G b S S a h a b c d e f g h Useful properties: e g f Complement: For any graph G, cc(g) cc(g)+1 Series & parallel operation: For any graphs G and H, cc(s(g,h) max(cc(g),cc(h))+1, cc(p(g,h) max(cc(g),cc(h)) cl(s(g,h) max(cl(g),cl(h))+1, cl(p(g,h) max(cl(g),cl(h))

29 Contiguity of cographs Useful properties: Complement: For any graph G, cc(g) cc(g)+1 Series & parallel operation: For any graphs G and H, cc(s(g,h) max(cc(g),cc(h))+1, cc(p(g,h) max(cc(g),cc(h)) cl(s(g,h) max(cl(g),cl(h))+1, cl(p(g,h) max(cl(g),cl(h)) cc(g) height(t) Cograph with caterpillar cotree: cc(g) 2 T S S r σ = d h a b c e f g a b c d e f g h

30 Contiguity of cographs Useful properties: Complement: For any graph G, cc(g) cc(g)+1 Series & parallel operation: For any graphs G and H, cc(s(g,h) max(cc(g),cc(h))+1, cc(p(g,h) max(cc(g),cc(h)) cl(s(g,h) max(cl(g),cl(h))+1, cl(p(g,h) max(cl(g),cl(h)) cc(g) height(t) Cograph with caterpillar cotree: cc(g) 2 T S S r σ = d h a b c e f g Combine both to get an upper bound for general cographs? a b c d e f g h

31 Outline Contiguity and linearity Cographs and cotrees A min-max theorem on the rank of a tree Upper bounds with caterpillar decompositions Lower bounds with claws erspectives

32 Rank of a tree rank(t) = maximal height of a complete binary tree T' obtained from T by edge contractions T

33 Rank of a tree rank(t) = maximal height of a complete binary tree T' obtained from T by edge contractions T T' rank(t)=2

34 Rank and path partition of a tree For any rooted tree T, rank(t) = maximum height of its path partitions A path partition { 1, 2, 3, 4, 5, 6 } of T A path partition tree of T T height

35 Outline Contiguity and linearity Cographs and cotrees A min-max theorem on the rank of a tree Upper bounds with caterpillar decompositions Lower bounds with claws erspectives

36 Upper bound on the contiguity / linearity For a cograph G and its cotree T, cc(g) 2 rank(t) (log n) Idea of the proof: Consider a root-path decomposition of T i [1..p], rank(t i ) +1 rank(t) S T S T p-4 S T p-3 T p-2 T p-1 T p T 6 T 1 T 2 T 4 T 3 T 5

37 Upper bound on the contiguity / linearity For a cograph G and its cotree T, cc(g) 2 rank(t) (log n) Idea of the proof: Consider a root-path decomposition of T i [1..p], rank(t i ) +1 rank(t) S T S T p-4 S T p-3 T p-2 T p-1 T p T 6 T 1 T 2 T 4 T 3 T 5 2 X 1 X 2 X 3 X 6 X p-2 X p-1 X 4 X 5 X p-4 X p Build the order cc(g) 2 + max cc(g[x i ]) i [1..p] X p-3

38 Upper bound on the contiguity / linearity For a cograph G and its cotree T, cc(g) 2 rank(t) (log n) Idea of the proof: Consider a root-path decomposition of T i [1..p], rank(t i ) +1 rank(t) S T S T p-4 S T p-3 T p-2 T p-1 T p T 6 T 1 T 2 T 4 T 3 T X 1 X 2 X 3 X 6 X p-2 X p-1 X 4 X 5 X p-4 X p Build the order cc(g) 2 + max cc(g[x i ]) i [1..p] Refine the order by recursively treating each T i X p-3

39 Outline Contiguity and linearity Cographs and cotrees A min-max theorem on the rank of a tree Upper bounds with caterpillar decompositions Lower bounds with claws erspectives

40 Lower bound on the contiguity For a cograph G with complete binary cotree, cc(g) ((log n)-5)/4 Idea of the proof: Base case: Induction: Vertices x y & t all need to be adjacent to z in σ so that their neighborhood is one interval cc(k 1,3 ) 2 4(k+1) T S u S u 1 u 2 S r t claw K 1,3 z x y x y t z v S cotree(k 1,3 ) 4k v 1 v S S 2 v 3 S S Tv T T v3 1 v 2 X 1 x X 2 y X 3 t z

41 Lower bound on the linearity For a cograph G with complete binary cotree, cl(g) O((log n)/(log log n)) Idea of the proof: Base case is star K 1,2k+1 (bigger than K 1,3 ) need a bigger complete binary cotree, of height 2k log(2k+1) +1

42 Tightness of the bounds For a cograph G with complete binary cotree, cc(g) = (log n)/2 +1 Idea of the proof: Oreste Manoussakis, 2012 Careful analysis of the result of the root-path decomposition algorithm for the upper bound. Analysis based on C 4 -cycles for the lower bound. Linearity open: O(log n) or O((log n)/(log log n))?

43 Tightness of the bounds Idea of the proof: For any cograph G, there is a linear time constant-factor approximation algorithm to compute its contiguity. Crespelle & Gambette, WALCOM 2013 Approximate value given by the root-path decomposition algorithm. Lower and upper bounds on the contiguity depending on the height of the biggest complete binary tree which is a minor of the cotree T of G, i.e. the rank of T: cc(g) 2 rank(t) + 1 cc(g) (rank(t) 7) / 4 approximation ratio 23

44 Outline Contiguity and linearity Cographs and cotrees A min-max theorem on the rank of a tree Upper bounds with caterpillar decompositions Lower bounds with claws erspectives

45 erspectives Open problems Linearity of cographs? Gap between linearity and contiguity? Linearity or contiguity of graphs classes generalizing cographs ractical applications of linearity and contiguity practical approaches to get upper bounds? use in algorithmic contexts? Solving problems on graphs with bounded linearity or contiguity. use for some graph classes arising from applications: express a complexity value for phylogenetic networks (min. spread) Asano, Jansson, Sadakane, Uehara & Valiente, 2010

46 Thank you for your attention Any questions? Work partially supported by the ES-C1 project Christophe Crespelle & hilippe Gambette (2009), Efficient Neighbourhood Encoding for Interval Graphs and ermutation Graphs and O(n) Breadth-First Search, IWOCA'09, LNCS 5874, p Christophe Crespelle & hilippe Gambette (2013), Linear-time Constant-ratio Approximation Algorithm and Tight Bounds for the Contiguity of Cographs, WALCOM'13, LNCS, to appear.