TOPICS
Search

Lovász Number


The Lovász number theta(G) of a finite simple graph G with at least one graph vertex, sometimes also called its theta function, was introduced by Lovász (1979) to estimate its Shannon capacity. For a graph G that is not a complete graph, let A be the family of nonzero real symmetric matrices A=(a_(ij)) such that a_(ij)=0 if i=j or i and j are adjacent in G, with the other entries unconstrained. Write the eigenvalues as lambda_1(A)>=lambda_2(A)>=...>=lambda_n(A), where n is the vertex count. Then

 theta(G)=max_(A in A)[1-(lambda_1(A))/(lambda_n(A))]
(1)

(Lovász 1979, Knuth 1994, Brimkov et al. 2000).

An equivalent definition, also valid for a complete graph, considers the family B of real symmetric matrices B=(b_(ij)) such that b_(ij)=1 if i=j or i and j are nonadjacent in G, with the other entries unconstrained. Then

 theta(G)=min_(B in B)lambda_1(B).
(2)

Let theta(G) be the Lovász number of a graph G. Then

 omega(G)<=theta(G^_)<=chi(G),
(3)

where omega(G) is the clique number and chi(G) is the chromatic number of G. This is the sandwich theorem. It can be rewritten by changing the role of graph complements, giving

 omega(G^_)<=theta(G)<=chi(G^_),
(4)

which can be written using omega(G^_)=alpha(G) with alpha the independence number and theta(G)=chi(G^_) the clique covering number as

 alpha(G)<=theta(G)<=theta(G).
(5)

For a perfect graph,

 theta(G)=omega(G^_)=chi(G^_).
(6)

Despite much work on the problem of determining theta(G), finding explicit values for interesting special cases of graphs remains an open problem (Brimkov et al. 2000). However, explicit formulas are known for theta(G) for several families of simple graphs. For example, for the cycle graph C_n with n>=3 and odd,

 theta(C_n)=(ncos(pi/n))/(1+cos(pi/n))
(7)

(Lovász 1979, Brimkov et al. 2000).

Self-complementary vertex-transitive graphs-including the Paley graphs--have theta(G)=sqrt(V(G)), where V(G) is the vertex count. A Kneser graph K(n,r), for positive integers n and r with n>=2r, has

 theta(K(n,r))=(n-1; r-1)
(8)

(Lovász 1979).

Fung (2011, p. 36) gives the Lovász numbers of the Keller graphs G_1, G_2, ..., as 4, 6, 28/3, 2^4, 2^5, ....

For a random graph G sampled from G(n,1/2), Potechin and Xu (2026) proved that, with high probability,

 theta(G)=(1+o(1))sqrt(n)
(9)

as n->infty. This gives the asymptotic value up to vanishing relative error.

A related asymptotic concerns the expectation value in a dependent-edge model. For odd n, independently include each cyclic distance j=1, ..., (n-1)/2 with probability 1/2, and join vertices of Z/nZ whose cyclic distance is included. For the resulting random circulant graph G_n, Chen et al. (2026) proved

 Etheta(G_n)=(1+o(1))sqrt(n)
(10)

as n->infty through odd integers, including composite orders. This is an expectation value result and does not assert concentration with high probability.

The following table gives some special cases.

Brimkov et al. (2000) determined the additional closed forms for quartic circulant graphs, namely

theta(Ci_(1,2)(n))=n{1-(1/2-cos(ax)-cos[x(a+1)])/([cos(ax)-1][cos[x(a+1)]-1])}
(11)
theta(Ci_(1,3)(n))=n{1-(cos^2y-cosycos(bx)+cos^2(bx)-1)/((cosy+1)[cos(bx)-1][1-cosy+cos(bx)])}
(12)

for odd n>=5 in the first formula and odd n>=7 in the second, where x=2pi/n, y=pi/n, a=|_n/3_|, and b=[(n-3)/6]. Here |_z_| is the floor function, and [z] is the ceiling function. These are special cases of the formula for integers n and j with n odd and 1<j<=(n-1)/2

 theta(Ci_(1,j)(n))=min_(x,y)max_(0<=i<=(n-1)/2)f_i(x,y),
(13)

where

f_0(x,y)=n+2x+2y
(14)
f_i(x,y)=2xcos((2pii)/n)+2ycos((2piij)/n)
(15)

for i=1, 2, ..., (n-1)/2.

The Lovász number satisfies

 theta(G□AdjustmentBox[x, BoxMargins -> {{-0.65, 0.13913}, {-0.5, 0.5}}, BoxBaselineShift -> -0.1]H)<=theta(G)theta(H),
(16)

where G□AdjustmentBox[x, BoxMargins -> {{-0.65, 0.13913}, {-0.5, 0.5}}, BoxBaselineShift -> -0.1]H denotes the graph strong product (Lovász 1979). In addition, if G is vertex-transitive, then

 theta(G)theta(G^_)=n,
(17)

where G^_ denotes the graph complement of G and n is its vertex count.

The Haemers number sometimes gives a better bound on the Shannon capacity than does the Lovász number.


See also

Clique Number, Coloring, Haemers Number, Sandwich Theorem, Shannon Capacity

Explore with Wolfram|Alpha

References

Brimkov, V. E.; Codenotti, B.; Crespi, V.; and Leoncini, M. "On the Lovász Number of Certain Circulant Graphs." In Algorithms and Complexity. Papers from the 4th Italian Conference (CIAC 2000) Held in Rome, March 1-3, 2000 (Ed. G. Bongiovanni, G. Gambosi, and R. Petreschi). Berlin, Germany: Springer-Verlag, pp. 291-305, 2000. https://doi.org/10.1007/3-540-46521-9_24.Chen, X.; Hu, H.; Hu, Y.; Wang, W.; and Yang, L. "Sharp Asymptotics for the Lovász Number of Random Circulant Graphs." 16 Oct 2026. https://arxiv.org/abs/2610.22971.Fung, M. "The Lovász Number of the Keller Graphs." Master's thesis. Leiden, Netherlands: Mathematisch Instituut, Universiteit Leiden, 2011. https://math.leidenuniv.nl/scripties/1FungMaster.pdf.Knuth, D. E. "The Sandwich Theorem." Elec. J. Combin. 1, A1, 1994. https://doi.org/10.37236/1193.Lovász, L. "On the Shannon Capacity of a Graph." IEEE Trans. Inform. Th. IT-25, 1-7, 1979.Potechin, A. and Xu, J. "Sharp Lovász-Theta Bounds on Random Graphs." 24 Sep 2026. https://arxiv.org/abs/2609.30064.Schrijver, A. "A Comparison of the Delsarte and Lovász Bounds." IEEE Trans. Inform. Th. 25, 425-429, 1979.Shannon, C. E. "The Zero-Error Capacity of a Noisy Channel." IRE Trans. Inform. Th. 2, 8-19, 1956.Skarakis, C. "The Sandwich Theorem." §4.2.3 in "Convex Optimization: Theory and Practice." Master's dissertation. York, UK: Department of Mathematics, University of York, pp. 43-46, Aug. 22, 2008. https://keithbriggs.info/documents/Skarakis_MSc.pdf.

Referenced on Wolfram|Alpha

Lovász Number

Cite this as:

Weisstein, Eric W. "Lovász Number." From MathWorld--A Wolfram Resource. https://mathworld.wolfram.com/LovaszNumber.html

Subject classifications