TOPICS
Search

Graph Likelihood


The likelihood of a simple graph is defined by starting with the set S_1={(K_11)}. The following procedure is then iterated to produce a set of graphs G_n of order n. At step n, randomly pick an integer k from the set {0,1,...,n-1}. Now randomly pick one of graphs in S_(n-1) (keeping the probability that it was constructed in step n-1) and from it add a new vertex that is connected to all of k randomly selected of its existing vertices. Now merge any isomorphic graphs produced by this procedure by totalling the their probabilities. The likelihood of a graph G on n vertices is then defined as the probability that G appears in S_n.

GraphLikelihood

The nth iteration of this procedure produces every possible graph on n nodes,. The results for graphs on n=1 to 4 nodes are illustrated above. Likelihoods for all simple graphs of size up to 10 nodes have been computed by E. Weisstein (Dec. 23, 2013).

L(G^_)=L(G), where G^_ is the graph complement of G. K_n and K^__n are therefore co-likely.

Since the values are probabilities, the sum of likelihoods over all n-node graphs is 1 and individual likelihoods satisfy

 0<L(G)<=1,
(1)

with L(G)=1 holding only for G=K_1. L(G) also satisfies the stronger inequality

 1/(|Aut(G)|product_(i=1)^(n)(i-1; |_(n-1)/2_|))<=L(G)<=1/(|Aut(G)|),
(2)

where |Aut(G)| is the order of the automorphism group of G (Banerji et al. 2014).

The following table summarizes the likelihoods for members of a number of special classes.

graphOEISvalues
Andrásfai graphA000000/A000000
antiprism graphA000000/A00000013/21600, 1909/2540160000, ...
barbell graphA000000/A00000097/129600, 79/282240000, ...
cocktail party graph K_(n×2)A000000/A0000001/2, 1/36, 13/21600, 11/1587600, ...
complete graph K_nA000000/A0000001, 1/2, 1/6, 1/24, 1/120, 1/720, ...
crown graphA000000/A00000029/64800, 11/40642560, ...
cycle graph C_nA000000/A0000001/2, 1/270, 1909/2540160000, ...
empty graph K^__nA000000/A0000001, 1/2, 1/6, 1/24, 1/120, 1/720, ...
hypercube graph Q_nA000000/A0000001, 1/2, 1/36, 11/40642560, ...
ladder graphA000000/A0000001/2, 1/36, 61/43200, 20299/2540160000, ...
ladder rung graphA000000/A0000001/2, 1/36, 13/21600, 11/1587600, ...
Möbius ladder M_nA000000/A00000023/259200, 1909/2540160000, ...
path graph P_2A000000/A0000001, 1/2, 1/3, 1/9, 29/1080, 2/405, 2509/3402000, 1889/20412000, ...
prism graph Y_nA000000/A00000029/64800, 11/40642560, ...
star graph S_nA000000/A0000001, 1/2, 1/3, 5/72, 17/1440, 77/43200, 437/1814400
sun graphA000000/A00000059/25920, 101/9072000, ...
triangular graphA000000/A0000001, 1/6, 13/21600, ...
wheel graph W_nA000000/A0000001/24, 13/720, 203/129600, 2393/18144000, ...

Classes with known closed form values include

L(K_n)=1/(n!)
(3)
L(K^__n)=1/(n!)
(4)
L(S_n)={1/2 for n=2; n/((n!)^2)sum_(k=0)^(n-1)k! otherwise
(5)
=((-1)^nn!!(-n-1))/(n!)-(n!(-1))/((n!)^2)
(6)
L(nP_2)=1/((2n)!)sum_(2<=i_1<i_2<...<i_n<=2n)product_(j=1)^(n)(i_j-(2j-1))/(i_j-1),
(7)

where K_n is a complete graph, K^__n is an empty graph, S_n is a star graph, nP_2 is a ladder rung graph, n! is a factorial, and !n is a subfactorial. In addition, there is a relationship between L(C_n) for a cycle graph and L(P_(n-1)) for a path graph given by

 L(P_(n-1))=n(n-1; 2)L(C_n)
(8)

(Banerji et al. 2014).

In general, a graph on n vertices with s isolated edges has likelihood

L(G_s)=1/(n!)sum_(2<=i_1<i_2<...<i_s<=n)product_(j=1)^(s)(i_j+1-2j)/(i_j-1)
(9)
=1/(n!)sum_(i_1=2)^(n-2+1)sum_(i_2=i_1+1)^(n-s-2)sum_(i_3=i_2+1)^(n-s+3)...sum_(i_s=i_(s-1)+1)^(n)(i_2-3)/(i_2-1)(i_3-5)/(i_3-1)...(i_s-(2s-1))/(i_s-1),
(10)

giving special cases

L(G_1)=(n-1)/(n!)
(11)
L(G_2)=((n-1)(n-6)+4H_(n-1))/(2n!),
(12)

where H_n is a harmonic number.

GraphLikelihoods

Values of L(G) for n-nodes graphs are plotted above.

For all values of n<=10 except n=1, 3, and 5 (for which the minima occur for K_1, K_3, and C_5, respectively), the minimum value of L(G) occurs for the complete bipartite graph K_(|_n/2_|,[n/2]) and its graph complement. The minimum values of L(G_n) for n=1, 2, ... are 1, 1/2, 1/6, 1/36, 1/270, 23/259200, 319/54432000, 319/15240960000, ... (OEIS A234234 and A234235).

The situation for maximum L(G) as a function of n is less clear, with maxima occurring for n=1, 2, ... for K_1, P_2, P_3, paw graph, dart graph, ... and their complements. The corresponding maximum values are 1, 1/2, 1/3, 13/72, 307/4320, 1927/86400, 39211/6804000, 27797639/22861440000, ... (OEIS A234236 and A234237).


See also

Likelihood

Explore with Wolfram|Alpha

References

Banerji, C. R. S.; Mansour, T.; and Severini, S. "A Notion of Graph Likelihood and an Infinite Monkey Theorem." J. Phys. A: Math. Theor. 47, 035101 (8 pp.), 2014.Sloane, N. J. A. Sequences A000088/M1253 A234234, A234235, A234236, and A234237 in "The On-Line Encyclopedia of Integer Sequences."

Cite this as:

Weisstein, Eric W. "Graph Likelihood." From MathWorld--A Wolfram Web Resource. https://mathworld.wolfram.com/GraphLikelihood.html