Minimal & Characteristic Polynomials

characteristic polynomial (operator)

In Vol I you computed the characteristic polynomial of a matrix as det(xI - A). The danger is that A lives in coordinates — so does this polynomial somehow depend on the basis you happened to pick? Reassuringly, no. Similar matrices have the same characteristic polynomial, so the formula attaches to the operator T itself, not to any particular matrix for it.

Define it basis-free as chi_T(x) = det(xI - T), where det is the coordinate-free determinant of an operator. It is a monic polynomial of degree exactly n = dim V. Its roots, counted with multiplicity, are the eigenvalues of T. Two pieces fall out of its coefficients for free: the coefficient of x^(n-1) is minus the trace, and the constant term is (-1)^n det(T).

It matters because it manufactures eigenvalues. Setting chi_T(x) = 0 is the characteristic equation, and solving it is how you find the spectrum. Over an algebraically closed field like C it always factors completely into linear pieces (x - lambda_1)...(x - lambda_n), so the operator always has n eigenvalues counted with multiplicity — the multiplicity of a root here is the algebraic multiplicity.

One honest limitation: the characteristic polynomial knows the eigenvalues and their algebraic multiplicities, but it can be FOOLED about the operator's finer structure. The matrices [2, 1; 0, 2] and 2I share chi(x) = (x-2)^2 yet behave completely differently. To see past that you need the minimal polynomial, which is why these two travel together.

A = [0, -2; 1, 3] chi_A(x) = det(xI - A) = x^2 - 3x + 2 = (x-1)(x-2) trace = 3 (= sum of roots), det = 2 (= product of roots)

The trace and determinant read straight off the characteristic polynomial's coefficients, and its roots 1 and 2 are the eigenvalues.

Convention warning: some texts define the characteristic polynomial as det(A - xI), which differs from det(xI - A) by a factor of (-1)^n. The det(xI - A) convention keeps it monic, which is why most modern operator-theory texts prefer it.

Also called
char poly特征多项式 chi_T(x)