9. Convergence and Monte Carlo Errors.ppt

9. Convergence and Monte Carlo Errors Measuring Convergence to Equilibrium Variation distance Eigenvalue Problem Consider the matrix S defined by [S]ij = pi? W(i-j) pj-? then S is real and symmetric and eigenvalues of S satisfy |?n| ≤ 1 One of the eigenvalue must be ?0=1 with eigenvector pj?. Spectrum Decomposition Then we have UTSU = Λ, or S = U Λ UT where Λ is a diagonal matrix with diagonal elements ?k and U is orthonormal matrix, U UT = I. W can be expressed in U, P, and Λ as W = P-?UΛUTP? Evolution in terms of eigen-states Pn= P0Wn = P0 P-?UΛUTP? P-?UΛUTP?… = P0 P-?UΛnUTP? In component form, this means Pn(j) = ∑iP0(i) pi-?pj?∑k?kn uikujk Discussion In the limit n goes to ∞, Pn(j) ≈ ∑iP0(i) pi-?pj? ui0uj0 = pj The leading correction to the limit is Pn(j) ≈ pj + a ?1n = pj + a e-n/? Exponential Correlation Time We define ? by the next largest eigenvalue ? = - 1/log ?1 This number characterizes the theoretical rate of convergence in a Markov chain. Measuring Error Let Qt be some quantity of interest at time step t, then sample average is QN = (1/N) ∑t Qt We treat QN as a random variable. By central limit theorem, QN is normal distributed with a mean QN=Q and variance σN2 = QN2-QN2. Confidence Interval The chance that the actual mean Q is in the interval [ QN – σN, QN + σN ] is about 68 percents. σN cannot be computed (exactly) in a single MC run of length N. Estimating Variance Error Formula The above derivation gives the famous error estimate in Monte Carlo as: where var(Q) = Q2-Q2 can be estimated by sample variance of Qt. Time-Dependent Correlation function and integrated correlation time We define and Circular Buffer for Calculating f(t) An Example of f(t) Efficient Method for Computing ?int Exponential and integrated correlation times where ?1 1 is the second largest eigenvalue of W matrix. This result says that exponential correlation time ? (=-1/log?1) is related to the largest integrated correlation time. C