Fast algorithm for finding the eigenvalue distribution of very large ...

PRE 62 4371FAST ALGORITHM FOR FINDING THE EIGENVALUE ...FIG. 3. Energy top and specific heat bottom of the XY model see Eq. 39, with 1 and h0. Left: L6; middle: L10; right:L15. Solid lines: exact result; crosses: simulation data using S5 samples; squares: simulation data using S20 samples. Error bars: Onestandard deviation.exact results and, equally important, the estimate for the errorcaptures the deviation from the exact result very well. Wealso see that in general the error decreases with the systemsize. Both the imaginary- and real-time methods seem towork very well, yielding accurate results for the energy andspecific heat of quantum spin systems with modest amountsof computational effort.X. CONCLUSIONSThe theoretical analysis presented in this paper gives asolid justification of the remarkable efficiency of the realtimeequation-of-motion method for computing the distributionof all eigenvalues of very large matrices. The real-timemethod can be used whenever the more conventional,Lanczos-like, sparse-matrix techniques can be applied:Memory and CPU requirements for each iteration time-stepare roughly the same depending on the actual implementationfor both approaches.We do not recommend using the real-time method if oneis interested in the smallest or largest eigenvalue only.Then the Lanczos method is computationally more efficientbecause it needs less iterations time steps than the real-timeapproach. However, if one needs information about all ei-FIG. 4. Energy top and specific heat bottom of the Ising model in a transverse field see Eq. 39 with 0 and h0.75J. Left:L6; middle: L10; right: L15. Solid lines; exact result; crosses: simulation data using S5 samples; squares: simulation data usingS20 samples. Error bars: one standard deviation.