Optimal Linear Inequalities for the Edge and Stability Numbers - - PDF document

optimal linear inequalities for the edge and stability
SMART_READER_LITE
LIVE PREVIEW

Optimal Linear Inequalities for the Edge and Stability Numbers - - PDF document

Optimal Linear Inequalities for the Edge and Stability Numbers Jean-Paul Doignon Service de G eom etrie, Combinatoire et Th eorie des Groupes Universit e Libre de Bruxelles For a graph G = ( V, E ), denote by n its number of nodes


slide-1
SLIDE 1

Optimal Linear Inequalities for the Edge and Stability Numbers Jean-Paul Doignon Service de G´ eom´ etrie, Combinatoire et Th´ eorie des Groupes Universit´ e Libre de Bruxelles For a graph G = (V, E), denote by n its number of nodes, m its number of links, (thus |V | = n and |E| = m), α = α(G) its stability number (the largest number of two by two nonadjacent nodes in G). 1

slide-2
SLIDE 2

The Petersen Graph has n = 10, m = 15, α = 4. Adding some new link to a graph will leave unchanged or decrease α, deleting a link from a graph will leave unchanged or increase α. On 10 nodes, can we force α ≤ 4 with less than 15 links? 2

slide-3
SLIDE 3

Yes: still α = 4. But if we further delete any link: α = 5 > 4. What is the least number of links on 10 nodes that force α ≤ 4? 3

slide-4
SLIDE 4

The answer is: m = 8 with a unique solution graph: α = 4 This is the Tur´ an Graph T(10, 4). It is not connected.

  • Theorem. (Tur´

an, 1941). For n nodes and stability number α, any graph with the minimum possible number of links is a disjoint union of α balanced cliques (Tur´ an graph T(n, α)). Balanced means

  • f size

n α

  • r

n α

  • .
  • Problem. (Ore, 1962). Same minimizing problem for connected graphs.

Apparently open . . . until recently. 4

slide-5
SLIDE 5

We consider connected graphs on n nodes, with m links and stability number α. Some examples of valid linear inequalities on m and α (where we consider n as a parameter): m ≥ n − 1 α ≥ 1 α ≤ n − 1 2α + m ≤ (n − 2)(n − 3) 2 + 2n (n − 2)α + m ≤ 1 + (n − 2)n nα + 2m ≤ (n − 2)(n − 3) 2 + n2 + 1 . . . There are infinitely many of them! How can we master this infinite family of linear inequalities? Find a suitable geometric setting! 5

slide-6
SLIDE 6

For n fixed, plot in the plane all possible values of (α, m). With n=10:

11 9 14 17 21 24 30 35 39 42 44 45 1 2 3 4 5 6 7 8 9

α m 6

slide-7
SLIDE 7

Take the convex hull.

11 9 14 17 21 24 30 35 39 42 44 45 1 2 3 4 5 6 7 8 9

The edges

  • f the resulting polygon

correspond to

  • ptimal linear inequalities.

7

slide-8
SLIDE 8

Given k graph invariants β1, β2, . . . , βk, and a class G of graphs, for a fixed number n of nodes, we define in Rk the polytope of graph invariants P n

β1, β2, ..., βk(G).

When the polytope is full-dimensional, it admits a unique description by a minimum system of linear inequalities. These are the optimal linear inequalities. They are finite in number:

  • ptimal inequality

↔ facet of P n

β1, β2, ..., βk(G).

Any linear inequality among the invariants, valid for the class G, is a consequence of the optimal inequalities, in the sense that it is a positive combination of optimal inequalities.

  • Remark. n (the number of nodes) is a parameter.

8

slide-9
SLIDE 9

A team consisting of Julie Christophe Sophie Dewez Jean-Paul Doignon Sourour Elloumi Gilles Fasbender Philippe Gr´ egoire David Huygens Martine Labb´ e Hadrien M´ elot Hande Yaman is investigating this polyhedral approach to linear inequalities among graph invariants and has submitted a first paper. 9

slide-10
SLIDE 10
  • Example. (α, m) for connected graphs.

Here for n = 10:

11 9 14 17 21 24 30 35 39 42 44 45 1 2 3 4 5 6 7 8 9 α

m The leftmost edges The rightmost edges The horizontal edge 10

slide-11
SLIDE 11

The horizontal edge: xm ≥ n − 1. The rightmost vertices: for a given stability number α, find the largest possible number of links in a connected graph. Trivial answer: α(n − α) + n − α 2

  • .

The rightmost edges: k α + m ≤ n − k 2

  • + k n,

for k = 1, 2, . . . , n − 2. The leftmost vertices: for a given stability number α, find the least possible number of links. Not an easy exercise . . . . . . it is Ore Problem. The rightmost edges: . . . ? Okay, first find the vertices. 11

slide-12
SLIDE 12

Ore Problem (1962): What is the least number m of links in connected graphs on n nodes having stability number α? Definitions. The Tur´ an graph T(n, α) is the disjoint union of α balanced cliques (of size n α

  • ).

The Tur´ an number t(n, α) is the number of links of T(n, α). Tur´ an Theorem rephrased:

  • Theorem. (Tur´

an, 1941). Any graph G with n nodes and stability number α has at least t(n, α) links; if G has t(n, α) links, then G is (isomorphic to) the Tur´ an graph T(n, α). Quick conjecture for a “connected Tur´ an Theorem” solving Ore Problem: we need add only α − 1 links to the Tur´ an graph T(n, α) to make the graph connected. 12

slide-13
SLIDE 13

Example. n = 10, α = 4: m = 8 + 3 Not unique: etc. 13

slide-14
SLIDE 14

. . . and even worst: With n = 10, α = 4,

  • ther connected graphs which minimize the number of links:

m = 11 (= 8 + 3 ?) 14

slide-15
SLIDE 15

Theorem. Any connected graph on n nodes with stability number α has at least t(n, α) + α − 1 links. This solves Ore’s Problem, and improves on Harant and Schiermeyer (2001). The proof is more involved than the ones for Tur´ an Theorem. Other proofs have been given recently by Gitler and Valencia and by Bougard and Joret who also explain how to generate all critical graphs. 15

slide-16
SLIDE 16

Back to the polygon of the invariants α and m for connected graphs: The leftmost points seem to be of the form pk =

  • k, t(n, k) + k − 1
  • for k = 1, 2, . . . ,

n+1

2

  • .

We need to check: (i) Can “convexity” be broken by one of these points?

1 2 3 4 5 6 7 8 9

(i)? (ii)?

α m (ii) Can one point lies on the segment joining two other points? (i) No! and (ii) Yes! Investigate how the number of links evolves when k changes in the Tur´ an graph T(n, k), more precisely when T(n, k) transforms into T(n, k − 1) or T(n, k + 1). 16

slide-17
SLIDE 17

Lemma. pk ∈

  • pk−1, pk+1

  • n

k − 1

  • =
  • n

k + 1

  • + 1.

That is: n k − 1 and n k + 1 lie between two consecutive natural numbers.

  • Example. Take n = 24.

Mark the values of 24 k :

1 2 3 4 5 6 7 8 9 10 11 12 .

24/k k = 12 7 5 3 2 Values of k for which pk is a vertex: 1, 2, 3, 4, 5, 6, 8, 12; but not 7, 9, 10, 11. 17

slide-18
SLIDE 18
  • Theorem. The optimal linear inequalities in α and m for connected

graphs are m ≥ n − 1; k α + m ≤ n − k 2

  • + k n,

for k = 1, 2, . . . , n − 2; m − t(n, k) − (k − 1) ≥

  • t(n, k) − t(n, k − 1) + 1
  • (α − k),

for k = 2, 3, . . . , n + 1 2

  • with
  • n

k − 1

  • =
  • n

k + 1

  • + 1.
  • Remark. (e.g. Berge, 1983)

t(n, k) = n k

  • − 1
  • ·
  • n − k

2 n k

  • .

18

slide-19
SLIDE 19

Our investigations are computer supported: for small values of n, GraPHedron generates the optimal inequalities, then we formulate conjectures and (hopefully) prove them. Other automated systems exist for investigating invariants of graphs: GRAPH

  • f Cvetkovi´

c et al. (1980’s); INGRID

  • f Brigham and Dutton (1980’s);

Graffiti

  • f Fajtlowicz et al. (late 1980’s);

AutoGraphiX

  • f Caporossi and Hansen (1990’s).

Some systems just provide help to the user,

  • thers even build proofs!

GraPHedron provides reports for small values of n. It calls geng (Mckay) and (porta (Christof) or cdd (Fukuda)) as sub- routines. 19

slide-20
SLIDE 20

Two other examples investigated by our group, with three invariants: (i) α, m, and ∆ (the maximum degree); (ii) ∆, D (the diameter), and Irr (the irregularity), where Irr =

  • {i,j}∈E

| deg(i) − deg(j)|. We may also change the class G of graphs, for instance taking the class Ck of all k-connected graphs.

  • Problem. What is the minimum number of links in a k-connected graph
  • n n nodes, having stability number α ?

Solved for k = 2 by Bougard and Joret,

  • pen for k ≥ 3.

20

slide-21
SLIDE 21
  • Example. The case of

α, m, and ∆ (the maximum degree) for connected graphs on n ≥ 6 vertices. Several optimal inequalities have been obtained, e.g.: m ≥ n − 1 ∆ ≤ n − 1 2m ≤ n∆ m ≤ (n − 3)(n − 2 − α) + 2∆ ∆ − m − α ≤ 1 − n . . . We still need to work in order to obtain the full list. Here is a Schlegel diagram for the polytope of the invariants α, m, and ∆, when n = 10. 21

slide-22
SLIDE 22

Conclusions: a notion of optimal linear inequalities for graph invariants; a full list of these inequalities in the stability number and the number of edges for the class of connected graphs; a solution to Ore’s Problem; partial lists of optimal inequalities for cases with three invariants. 22

slide-23
SLIDE 23

A question. Does anybody know a reference for the following problem? Let X = { (x, y) | x, y natural numbers and x · y ≥ n }. What are the vertices of conv X ? Of course, integral points lying on the hyperbola x · y = n are vertices, but there can be other. 23