-
Notifications
You must be signed in to change notification settings - Fork 29
/
CliqueNumber.m
69 lines (62 loc) · 3.19 KB
/
CliqueNumber.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
%% CLIQUENUMBER Bounds the clique number (i.e., maximum size of a clique) of a graph
% This function has one required input argument:
% A: the adjacency matrix of a graph
%
% UB = CliqueNumber(A) is an upper bound on the clique number of the
% graph with adjacency matrix A. This upper bound is computed via
% the sum-of-squares hierarchy and thus makes use of semidefinite
% programming.
%
% This function has two optional input arguments:
% K (optional, default 0): a non-negative integer that indicates the
% level of the hierarchy used to bound the optimal value
% MODE (optional, default 'sos'): either 'sos' or 'nosdp', indicating
% whether the upper bound should be computed via the sum-of-squares
% hierarchy or a no-SDP alternate hierarchy. The 'nosdp' hierarchy
% is less accurate for a given level of the hierarchy, but is much
% faster and less memory-intensive to run.
%
% [UB,LB] = CliqueNumber(A,K,MODE) gives an upper bound (UB) and lower
% bound (LB) on the clique number of the graph with adjacency
% matrix A, computed via the K-th level of the hierarchy
% specified by MODE. Some other easily-computed bounds are
% incorporated into these bounds too.
%
% URL: http://www.qetlab.com/CliqueNumber
% author: Nathaniel Johnston ([email protected])
% package: QETLAB
% last updated: August 4, 2023
function [ub,lb] = CliqueNumber(A,varargin)
% set optional argument defaults: K=0, MODE='sos'
[k,mode] = opt_args({ 0, 'sos' },varargin{:});
do_sos = strcmpi(mode,'sos');% true means SOS, false means SDP-free alternative
n = length(A);% number of vertices
d = 2;% max clique is modelled by a degree-4 homogeneous polynomial, we use d = 4/2 = 2 is half the degree
p = CopositivePolynomial(A);% the degree-4 polynomial associated with A
if(nargout > 1)% only compute a lower bound if it was requested, since it takes extra computational time
if(do_sos)
[ub,lb] = PolynomialSOS(p,n,d,k,'max');
else
[ub,lb] = PolynomialOptimize(p,n,d,k,'max');
end
lb = ceil(1/(1-lb) - 0.0000000001);% the 0.0000000001 is there for numerical reasons: we do not want to round to an incorrect max clique bound as a result of numerical imprecision
else
if(do_sos)
ub = PolynomialSOS(p,n,d,k,'max');
else
ub = PolynomialOptimize(p,n,d,k,'max');
end
end
if(ub >= 1)
ub = Inf;
else
ub = floor(1/(1-ub) + 0.0000000001);% the 0.0000000001 is there for numerical reasons: we do not want to round to an incorrect max clique bound as a result of numerical imprecision
end
% Simple upper bound based on number of edges in the graph
m = round(sum(sum(A))/2);% number of edges
ub = min(ub,floor((sqrt(8*m+1)+1)/2 + 0.0000000001));% this formula comes from inverting triangular numbers; the 0.0000000001 is there for numerical reasons
% Simple lower bounds based on number of edges and other easily-computable properties of the graph
if(nargout > 1)
lb = max(lb,ceil(2*m/(2*m - max(abs(eig(A)))^2) - 0.0000000001));% Nikiforov bound
end
end