Algebraic and Geometric Multiplicities of Eigenvalues Let: T : V → V be a linear operator on a finitedimensional K-vector space V. Definition: The algebraic multiplicity of λ (as an eigenvalue of T ) is its multiplicity as a root of chT ; we denote it by mλ(T ) = multλ(chT ) = max{m : chT (x) = (x−λ)mh(x), for some h(x) ∈ K[x]}. The geometric multiplicity of λ (as an eigenvalue of T ) is gλ(T ) := dim Eλ(T ) = dim(V ) − rank(T − λIV ). The algebraic and geometric multiplicities of λ with respect to A ∈ Mn(K) are defined by mλ(A) = mλ(TA) and gλ(A) = gλ(TA). Theorem 17: If B is any basis of V , then mλ([T ]B ) = mλ(T ) and gλ([T ]B ) = gλ(T ). 2 Corollary: If A ∼ B are two similar matrices, i.e. if A = P BP −1 for some invertible matrix P , then mλ(A) = mλ(B) and gλ(A) = gλ(B). Remark: By the Corollary of Theorem 16 we have that X (1) gλ(T ) ≤ dim(V ). λ∈spec(T ) Analogously we have the inequality X (2) mλ(T ) ≤ dim(V ). λ∈spec(T ) Q ∗ Indeed, write chT (x) = λ∈spec(T )(x − λ)mλ(T ). Then we have that chT (x) = ch∗T (x)h(x), where h(x) ∈ K[x] is a (monic) polynomial with no roots in K. From (3) the inequality (2) follows by taking degrees. Moreover, we thus see that (3) Equality holds in (2) ⇔ ch∗T (x) = chT (x) ⇔ chT (x) splits completely over K. 3 Theorem 18: We have that (4) gλ(T ) ≤ mλ(T ), for all λ ∈ spec(T ). Theorem 19 (Diagonalization Theorem): The following conditions are equivalent: (a) T is diagonable; (b) equality holds in (1); (c) mλ(T ) = gλ(T ), for all λ ∈ spec(T ) and chT (x) splits completely over K. Remark: The splitting condition is missing from the statement of Theorem 4.27 of the text. It seems that the book assumes here that K = C (so all diagonalization takes place over C), but this is not stated explicitly. h i Example: Let A = 01 −10 . Since chA(λ) = λ2 + 1 has no real roots, A has no real eigenvalues and A is not diagonable over R. However, chA(λ) = (λ − i)(λ + i) in C[λ], so A can be diagonalized over C. Explicitly: h i h ih ih i−1 A = 01 −10 = −i1 1i i0 −i0 −i1 1i .