Noncomplete graphs and the two largest adjacency eigenvalues #
A missing edge gives a vanishing 2x2 principal submatrix of the adjacency
matrix. Courant-Fischer on that coordinate plane yields lambda2 G ≥ 0 when G
is not complete, and therefore F(A_G) = lambda1 G ^ 2 + lambda2 G ^ 2. The
largest eigenvalue is nonnegative for every finite graph, complete or not.
SP13 — A nonedge gives a zero principal 2x2 #
The principal 2x2 submatrix of the adjacency matrix on a nonedge is identically zero.
The adjacency quadratic form vanishes on vectors supported on a nonedge.
SP14 — Rayleigh on a 2-dimensional coordinate subspace #
Euclidean inner product on coordinate space. Avoids the BollobasNikiforov.inner name clash.
Equations
- BollobasNikiforov.euclInner x y = inner ℝ x y
Instances For
The Euclidean quadratic form expands in the ordered eigenbasis.
On the orthogonal complement of a top eigenvector, the Rayleigh form is at
most lambdaSecond.
A unit vector in span {e i, e j} orthogonal to a top eigenvector.
Rayleigh values of unit vectors in span {e i, e j}.
Equations
Instances For
SP14. Courant-Fischer comparison on a coordinate 2-plane:
lambdaSecond is at least the infimum of the Rayleigh form on span {e i, e j}.
If the Rayleigh form is nonnegative on the coordinate 2-plane, then
0 ≤ lambdaSecond.
If the quadratic form vanishes on span {e i, e j}, then 0 ≤ lambdaSecond.
SP15 — lambda2 G ≥ 0 if G is not complete #
SP15. A noncomplete graph on at least two vertices has lambda2 ≥ 0.
SP17 — lambda1 G ≥ 0 #
SP17. The largest adjacency eigenvalue is nonnegative.
SP16 — F(A_G) = lambda1^2 + lambda2^2 for noncomplete G #
SP16. On a noncomplete graph the positive-part squares in F are
lambda1 ^ 2 and lambda2 ^ 2.