Gershgorin Circle Theorem
Let . Every eigenvalue of lies in at least one of the Gershgorin discs where is the deleted absolute row sum. Moreover, if of the discs form a connected component that is isolated from the remaining discs, then exactly eigenvalues (counted with multiplicity) lie in that component.
Proof
Part 1: Eigenvalues lie in the union of discs. Let be an eigenvalue with eigenvector . Let be the entry with maximum absolute value: .
From , the -th equation gives: .
Isolating the diagonal term: .
Taking absolute values: .
Dividing by : , so .
Part 2: Counting eigenvalues in connected components. Consider the family of matrices for , where . At , the eigenvalues are with the -th eigenvalue in the degenerate disc . At , we recover .
The Gershgorin discs for are . As increases from 0 to 1, the discs grow continuously, and the eigenvalues (as continuous functions of ) move continuously within the union of discs.
If discs form a connected component that remains isolated from the other discs for all (i.e., for outside the component), then no eigenvalue can enter or leave as varies. At , exactly eigenvalues (the diagonal entries) lie in . By continuity, exactly eigenvalues lie in .
Applications
For : , , . The three discs are disjoint (intervals , , ), so each contains exactly one eigenvalue. Column Gershgorin (applied to ) can give tighter localization.
For any nonsingular diagonal , the matrices and have the same eigenvalues, but different Gershgorin discs. Optimizing over can tighten the localization significantly. The intersection of Gershgorin regions over all diagonal scalings gives the optimal Gershgorin set, which can be much smaller than the standard discs.