variational eigenvalue estimate

An exact eigenvalue is often out of reach, but a guaranteed bound on it is not. The variational eigenvalue estimate is the practical recipe that turns the Rayleigh quotient's minimum principle into numbers: choose a family of trial functions, optimize over them, and read off a rigorous upper bound for the lowest eigenvalue. It trades exact solvability for a controlled, honest approximation that you can sharpen at will.

The minimum principle says lambda_1 equals the minimum of the Rayleigh quotient R[u] over all admissible u. You cannot search all functions, so you restrict to a finite-dimensional trial space — say u = a_1 phi_1 + ... + a_N phi_N for chosen basis functions phi_k and free coefficients a_k. Minimizing R[u] over the coefficients turns the calculus-of-variations problem into a finite generalized matrix eigenvalue problem A a = mu B a, where A and B are the stiffness and mass matrices (the numerator and denominator integrals between basis functions). Its smallest eigenvalue mu_1 is your estimate of lambda_1, and it is provably at least lambda_1: more (or better) trial functions can only lower the estimate toward the truth. This is exactly the Rayleigh-Ritz method, and enlarging the trial space until the estimate stops moving is the conceptual seed of the finite element method.

Because the bound is one-sided and monotone, the variational estimate is trustworthy in a way many approximations are not — you always know you are above the true value, and you can watch the estimate descend as you add basis functions. It is how engineers find buckling loads and vibration frequencies, and how quantum chemists estimate ground-state energies (the variational principle there guarantees the computed energy never dips below the true ground state). The catch is that the clean upper-bound guarantee applies to the lowest eigenvalue; bounding higher modes requires orthogonality constraints or the more delicate min-max characterization.