Linear Algebra Notes Thomas Goller September 4, 2011 1 Rational and Jordan Canonical Forms For the rational form (over any field), the cyclic decomposition theorem guarantees the existence of non-zero vectors α1 , . . . , αr with respective T -annihilators p1 , . . . , pr such that V = Z(α1 ; T ) ⊕ · · · ⊕ Z(αr ; T ) and pk |pk−1 , k = 2, . . . , r, with p1 = p the minimal polynomial of T , p1 · · · pr = f the characteristic polynomial, and deg pi = dim Z(αi ; T ). Then in the basis {α1 , T α1 , . . . , T deg p1 −1 α1 , α2 , T α2 , . . . }, T is composed of blocks 0 0 0 1 0 0 0 1 0 .. .. .. . . . 0 0 0 ··· ··· ··· 0 0 0 .. . −a0 −a1 −a2 .. . ··· 1 −ak−1 , where pi = xk + ak−1 xk−1 + · · · + a0 . The Jordan form (over an algebraically closed field like C) is obtained by combining the primary decomposition theorem with the rational form, as follows. Given A with characteristic polynomial f = (x − c1 )d1 · · · (x − ck )dk and minimal polynomial p = (x − c1 )r1 · · · (x − ck )rk , with 1 ≤ ri ≤ di for each i, we use the primary decomposition theorem to write V = W1 ⊕ · · · ⊕ Wk , where Wi is the null space of (T − ci I)ri , and Wi has dimension di . We choose a basis for Wi corresponding to the cyclic decomposition of the nilpotent operator 1 Ni := T − ci I on Wi . Since Ni has minimal polynomial xri and T = Ni + ci I, the matrix for T in the chosen basis is composed of blocks of the form ci 0 · · · 0 0 1 ci · · · 0 0 .. .. .. .. , . . . . ci 0 0 · · · 1 ci decreasing in size, with the first of size ri and the sum of the sizes equal to di . Putting these result together for all the Wi gives the Jordan form. 2 Special Matrices Let V be a finite-dimensional inner product space. • For every self-adjoint linear operator T on V , there is an orthonormal basis of V consisting of characteristic vectors of T . • It follows that Hermitian matrices are diagonalizable by unitary matrices, and real symmetric matrices are diagonalizable by real orthogonal matrices. • Even stronger: if T is a normal operator on V , there is an orthonormal basis of V consisting of characteristic vectors of T . Hence every normal matrix is diagonalizable by a unitary matrix. 2