minimal polynomial
Pick an operator T on a finite-dimensional space and start feeding it to polynomials: T, then T^2, then T^3, and so on. Because the space of operators is itself finite-dimensional, these powers cannot stay independent forever — eventually some combination c_0 I + c_1 T + ... + c_k T^k equals the zero operator. The minimal polynomial is the SIMPLEST such relation: the monic polynomial m(x) of least degree for which m(T) = 0.
Precisely, m(x) is the unique monic polynomial of smallest degree with m(T) = 0. Monic means its leading coefficient is 1, which pins down a single representative (otherwise you could scale any annihilator). Its key property: m(x) divides EVERY polynomial p(x) with p(T) = 0. So it is not just one annihilator among many — it is the one that controls all the others by division.
Why care? The minimal polynomial is a complete fingerprint of the operator's algebraic behavior. Its roots are exactly the eigenvalues; whether those roots are repeated tells you if T is diagonalizable; its degree bounds how complicated T can be. Two operators that are similar share a minimal polynomial, so it is a similarity invariant you can read off without choosing coordinates.
A caveat worth holding onto: the minimal polynomial can have SMALLER degree than the characteristic polynomial. For the identity I on an n-dimensional space the characteristic polynomial is (x-1)^n, but the minimal polynomial is just x-1, since I - I = 0 already. The minimal polynomial throws away the inflated multiplicities and keeps only what the operator genuinely needs.
Two matrices with the same characteristic polynomial. A is a single Jordan block (defective), so its minimal polynomial keeps the square; B is already 2I, so its minimal polynomial drops to degree 1. The minimal polynomial sees the difference the characteristic polynomial misses.
The minimal polynomial divides the characteristic polynomial (a consequence of Cayley-Hamilton), and they share the same set of roots — but their multiplicities can differ. The minimal polynomial is the gentler, leaner record of the two.