TheoremComplete

Gershgorin Circle Theorem

Theorem6.1Gershgorin Circle Theorem

Let A=(aij)Cn×nA = (a_{ij}) \in \mathbb{C}^{n \times n}. Every eigenvalue of AA lies in at least one of the Gershgorin discs Di={zC:zaiiRi}D_i = \{z \in \mathbb{C} : |z - a_{ii}| \leq R_i\} where Ri=jiaijR_i = \sum_{j \neq i} |a_{ij}| is the deleted absolute row sum. Moreover, if mm of the nn discs form a connected component that is isolated from the remaining nmn - m discs, then exactly mm eigenvalues (counted with multiplicity) lie in that component.


Proof

Proof

Part 1: Eigenvalues lie in the union of discs. Let λ\lambda be an eigenvalue with eigenvector x=(x1,,xn)Tx = (x_1, \ldots, x_n)^T. Let xix_i be the entry with maximum absolute value: xi=x>0|x_i| = \|x\|_\infty > 0.

From Ax=λxAx = \lambda x, the ii-th equation gives: j=1naijxj=λxi\sum_{j=1}^n a_{ij} x_j = \lambda x_i.

Isolating the diagonal term: (λaii)xi=jiaijxj(\lambda - a_{ii}) x_i = \sum_{j \neq i} a_{ij} x_j.

Taking absolute values: λaiixijiaijxjjiaijxi=Rixi|\lambda - a_{ii}| |x_i| \leq \sum_{j \neq i} |a_{ij}| |x_j| \leq \sum_{j \neq i} |a_{ij}| |x_i| = R_i |x_i|.

Dividing by xi>0|x_i| > 0: λaiiRi|\lambda - a_{ii}| \leq R_i, so λDi\lambda \in D_i.

Part 2: Counting eigenvalues in connected components. Consider the family of matrices A(t)=D+t(AD)A(t) = D + t(A - D) for t[0,1]t \in [0, 1], where D=diag(a11,,ann)D = \mathrm{diag}(a_{11}, \ldots, a_{nn}). At t=0t = 0, the eigenvalues are a11,,anna_{11}, \ldots, a_{nn} with the ii-th eigenvalue in the degenerate disc {aii}\{a_{ii}\}. At t=1t = 1, we recover AA.

The Gershgorin discs for A(t)A(t) are Di(t)={z:zaiitRi}D_i(t) = \{z : |z - a_{ii}| \leq tR_i\}. As tt increases from 0 to 1, the discs grow continuously, and the eigenvalues (as continuous functions of tt) move continuously within the union of discs.

If mm discs form a connected component Ω(t)\Omega(t) that remains isolated from the other nmn - m discs for all t[0,1]t \in [0, 1] (i.e., Ω(t)Dj(t)=\Omega(t) \cap D_j(t) = \emptyset for jj outside the component), then no eigenvalue can enter or leave Ω(t)\Omega(t) as tt varies. At t=0t = 0, exactly mm eigenvalues (the mm diagonal entries) lie in Ω(0)\Omega(0). By continuity, exactly mm eigenvalues lie in Ω(1)\Omega(1). \square


Applications

ExampleEigenvalue Localization

For A=(1010.50.250.30.10.21)A = \begin{pmatrix} 10 & 1 & 0.5 \\ 0.2 & 5 & 0.3 \\ 0.1 & 0.2 & 1 \end{pmatrix}: D1={z:z101.5}D_1 = \{z: |z - 10| \leq 1.5\}, D2={z:z50.5}D_2 = \{z: |z - 5| \leq 0.5\}, D3={z:z10.3}D_3 = \{z: |z - 1| \leq 0.3\}. The three discs are disjoint (intervals [8.5,11.5][8.5, 11.5], [4.5,5.5][4.5, 5.5], [0.7,1.3][0.7, 1.3]), so each contains exactly one eigenvalue. Column Gershgorin (applied to ATA^T) can give tighter localization.

RemarkDiagonal Scaling Refinement

For any nonsingular diagonal D=diag(d1,,dn)D = \mathrm{diag}(d_1, \ldots, d_n), the matrices AA and D1ADD^{-1}AD have the same eigenvalues, but different Gershgorin discs. Optimizing over DD 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.