1 Independence densities of hypergraphs Anthony Bonato Ryerson University 2014 CMS Summer Meeting Independence densities - Anthony Bonato
1
Independence densities of hypergraphs
Anthony BonatoRyerson University
2014 CMS Summer Meeting
Independence densities - Anthony Bonato
Paths
Independence densities - Anthony Bonato
2
…• number of
independent sets = F(n+2)
- Fibonacci numberPn
Stars
Independence densities - Anthony Bonato
3
• number of independent sets = 1+2n
• independence density
= 2-n-1+½
K1,n
…
Independence density
• G order n
• i(G) = number of independent sets in G (including ∅)
– Fibonacci number of G
• id(G) = i(G) / 2n
– independence density of G
Independence densities - Anthony Bonato
4
Properties
• if G is a spanning subgraph of H, then
i(H) ≤ i(G)
• i(G U H) = i(G)i(H)
• if G is subgraph of H, then id(H) ≤ id(G)– G has an edge, then: id(G) ≤ id(K2) = 3/4
• id(G U H) = id(G)id(H)
Independence densities - Anthony Bonato
5
Infinite graphs?
• view countably a infinite graph as a limit of chains
• extend definition by continuity:
• well-defined?• possible values?
Independence densities - Anthony Bonato
6
Chains
• let G be a countably infinite graph
• a chain C in G is a set of induced subgraphs Gi such that:– for all i, Gi is an induced subgraph of Gi+1 and
• write id(G, C) =
Independence densities - Anthony Bonato
7
Existence and uniqueness
Theorem (B,Brown,Kemkes,Pralat,11)
Let G be a countably infinite graph.
1. For each chain C, id(G, C) exists.
2. For all chains C and C’ in G,
id(G, C)=id(G, C’).
Independence densities - Anthony Bonato
8
Examples
• stars: id(K1,∞) = 1/2
• one-way infinite path: id(P∞) = 0
Independence densities - Anthony Bonato
9
Bounds on id
• if G contains and infinite matching, then id(G) = 0• matching number of G, written µ(G), is the supremum of
the cardinalities of pairwise non-intersecting edges in G
Theorem (B,Brown,Kemkes,Pralat,14)
If µ(G) is finite, then:
• in particular, id(G) = 0 iff µ(G) is infinite
Independence densities - Anthony Bonato
10
Rationality
Theorem (BBKP,11)
Let G be a countable graph.
1. id(G) is rational.
2. The closure of the set
{id(G): G countable}
is a subset of the rationals.
Independence densities - Anthony Bonato
11
Aside: other densities
• many other density notions for graphs and hypergraphs:– upper density– homomorphism density– Turán density– co-degree density– cop density, …
Independence densities - Anthony Bonato
12
Question
• hereditary graph class X: closed under induced subgraphs
• egs: X = independent sets; cliques; triangle-free graphs;
perfect graphs; H-free graphs
• Xd(G) = proportion of subsets which induce a graph in X– generalizes to infinite graphs via chains
• Is Xd(G) rational?
Independence densities - Anthony Bonato
13
Hypergraphs
• hypergraph H = (V,E), E = hyperedges• independent set: does not contain a
hyperedge
• id(H) defined analogously– extend to infinite hypergraphs by continuity– well-defined
Independence densities - Anthony Bonato
14
1
2 3
4
Examples
∅,{1},{2},{3},{4},{1,2},{1,3},{2,3},{1,4},{3,4},{1,3,4}
id(H) = 11/16H
Examples, cont
Independence densities - Anthony Bonato
16
…
id(H) = 7/8
Hypergraph id’s
examples:
1. graph, E = subsets of vertices containing a copy of K2
– recovers the independence density of graphs
2. graph, fix a finite graph F; E = subsets of vertices containing a copy of F– F-free density (generalizes (1)).
3. relational structure (graphs, digraphs, orders, etc); F a set of finite structures; E = subsets of vertices containing a member of F – F-free density of a structure (generalizes (2))
Independence densities - Anthony Bonato
17
Bounds on id
• matching number of hypergraph H, written µ(H), is the supremum of the cardinalities of pairwise non-intersecting hyperedges in H
Theorem (B,Brown,Mitsche,Pralat,14)
Let H be a hypergraph whose hyperedges have cardinality bounded by k > 0. If µ(H) is finite, then:
• sharp if k = 1,2; not sure lower bound is sharp if k > 2
Independence densities - Anthony Bonato
18
Rationality
• rank k hypergraph: hyperedges bounded in cardinality by k > 0– finite rank: rank k for some k
Theorem (BBMP,14): If H has finite rank,
then id(H) is rational.
Independence densities - Anthony Bonato
19
Sketch of proof
• notation: for finite disjoint sets of vertices A and B
idA,B(H) = density of independent sets containing A and not B
• analogous properties to id(H) = id ,∅ ∅(H)
Independence densities - Anthony Bonato
20
Properties of idA,B(H)
1. For a vertex x outside A U B:
.
2. For a set W outside A U B:
Independence densities - Anthony Bonato
21
Out-sets
• for a given A, B, and any hyperedge S such that S∩B = ∅, the set S \ A is the out-set of S relative to A and B – example: A B
S• notation: idr
A,B(H) denotes that every out-set has cardinality at most r
• note that: idk,∅ ∅(H) = id(H)
Independence densities - Anthony Bonato
22
Claims
1. If A is not independent, then idrA,B(H) = 0.
2. If A is independent and there is an infinite family of disjoint out-sets, then idr
A,B(H) = 0.
3. If A is independent, then id0A,B(H)=2-(|A|+|B|).
4. Suppose that A is independent and there is no infinite family of disjoint out-sets. If O1, O2,…, Os is a maximal family of disjoint out-sets, then for all r > 0,
where W is the union of the Oi.
Independence densities - Anthony Bonato
23
Final steps…
• start with idk,∅ ∅(H)
• by (2), assume wlog there are finitely many out-sets• as A and B are empty, the out-sets are disjoint
hyperedges with union W• by (4):
• apply induction using (1)-(3); each term is rational, or a finite sum of terms where we can apply (1)-(3)
• process ends after k steps
Independence densities - Anthony Bonato
24
Unbounded rank
Independence densities - Anthony Bonato
25
…
H
Theorem (BBMP,14) 1. The independence density of H is
.
2. The value is S is irrational.
• proof of (2) uses Euler’s Pentagonal Number Theorem
Any real number
• case of finite, but unbounded hyperedges• Hunb = {x: there is a countable hypergraph
H with id(H) = x}
Theorem (BBMP,14) Hunb = [0,1].
• contrasts with rank k case, where there exist gaps such as (1-1/2k,1)
Independence densities - Anthony Bonato
26
Independence polynomials
• H finite, independence polynomial of H wrt x > 0
i(H,x) = – for eg: id(H) = i(H,1)/2n
• example– i(Pn,x) = i(Pn-1,x) + xi(Pn-2,x),
i(P1,x)=1+x, (P2,x)=1+2x
Independence densities - Anthony Bonato
27
Independence densities at x
• (,x): defined in natural way for fixed x ≥ 0– may depend on chain if x ≠ 1– (,1) = id(H)
• examples: – (,x) = 0 for all x
• generalizes to chains with βn = o(n)
– (H,x) = 0 if x < 1:• bounded above by ((1+x)/2)n =o(1)
Independence densities - Anthony Bonato
28
Examples, continued
• +
• with chain (Pn: n ≥ 1), we derive that:
(,x) =
Independence densities - Anthony Bonato
29
Examples, continued
• for each r > 1, can choose chain C such that
) =
• r is a jumping point
Independence densities - Anthony Bonato
30
Future directions
• classify gaps among densities for given hypergraphs
• rationality of closure of set of id’s for rank k hypergraphs
• which hypergraphs have jumping points, and what are their values?
Independence densities - Anthony Bonato
31
General densities
• d: hypergraph function satisfying:– multiplicative on disjoint unions– monotone increasing on subgraphs
• d(H) well-defined for infinite hypergraphs
• properties of d(H)? for eg, when rational?
Independence densities - Anthony Bonato
32