the resolvent operator
/ ri-ZOLL-vent /
When you study an operator, the most revealing thing to ask is: for which numbers lambda can you invert L - lambda? The answer sorts the complex plane into the points where the inversion fails (the spectrum — the operator's resonant frequencies) and the points where it succeeds. At every successful point the inverse itself is a well-defined operator. The resolvent is that family of inverses, one for each lambda, and tracking how it behaves is the master strategy of spectral theory.
Formally, the resolvent of L is R(lambda) = (L - lambda I)^(-1), defined for every lambda not in the spectrum. To solve (L - lambda I) u = f you simply apply it: u = R(lambda) f. The connection to Green's functions is direct and important: R(lambda) is an integral operator whose kernel is the Green's function of the shifted operator L - lambda I. From the spectral representation, R(lambda) has the explicit form (R(lambda) f)(x) = integral of G_lambda(x, y) f(y) dy with G_lambda(x, y) = sum over n of phi_n(x) phi_n(y) / (lambda_n - lambda). Stare at that denominator: as lambda approaches an eigenvalue lambda_n, the corresponding term blows up — the resolvent has a pole exactly at each eigenvalue. So the spectrum (where inversion fails) is read off as the singularities of the resolvent, and the residue at a pole picks out the projection onto that eigenspace.
Why the resolvent is central: it packages a whole problem family (one for every lambda) into a single analytic object, and the tools of complex analysis — poles, residues, contour integrals — then deliver the spectral decomposition, the time-evolution semigroup (via inverse Laplace transform of R(lambda)), and the Green's function at any energy. It is the rigorous backbone behind eigenfunction expansions. The honest subtlety is that for differential operators the spectrum can be more than isolated eigenvalues — it can include a continuous part (think of the free particle on the whole line) — and then the resolvent has a branch cut rather than only poles, and the clean sum becomes an integral over the continuous spectrum.
For L = -d^2/dx^2 on [0, pi] with zero ends, R(lambda) f = u solves -u'' - lambda u = f. Its kernel G_lambda(x, y) = sum over n of (2/pi) sin(n x) sin(n y) / (n^2 - lambda) has a pole whenever lambda = n^2 — exactly the eigenvalues.
The resolvent's poles sit exactly at the operator's eigenvalues.
For differential operators the spectrum need not be just isolated eigenvalues — it can have a continuous part (e.g. the free particle on the line), and then the resolvent carries a branch cut rather than only poles, with an integral replacing the eigenvalue sum.