WorksPhysics and Astronomy
Efficient Finite-Temperature Krylov Methods Based on Imaginary-Time Evolution
Abstract
We develop two finite-temperature Krylov methods that construct Krylov subspaces using imaginary-time evolution rather than direct applications of the Hamiltonian. The finite-temperature imaginary-time Lanczos method (FTiTL) applies the Lanczos method to the imaginary-time evolution operator, whereas the finite-temperature imaginary-time Krylov method (FTiTK) constructs a generalized eigenvalue problem using a nonorthogonal Krylov basis. We evaluate the accuracy and computational efficiency of these methods for the square-lattice transverse-field Ising model, using exact diagonalization for and quantum Monte Carlo results for as references. With appropriately chosen imaginary-time steps, both methods achieve higher accuracy than the conventional finite-temperature Lanczos method (FTLM) with fewer local operator applications in the ferromagnetic regime and near the quantum critical point. For , FTiTL requires approximately half as many local operator applications as FTLM to achieve a root-mean-square error of or less in the specific heat . Furthermore, FTiTK can be implemented using only one state vector of full Hilbert-space dimension, halving the memory required for storing state vectors relative to FTLM.