How to overcome free energy barriers in grandcanonical simulations Peter Virnau, M. Müller, B. Mognetti, L. Yelash, K. Binder Leipzig, 11/2007 Overview & Credits Grandcanonical Simulations • (How to determine critical points?) • How to overcome free energy barriers? Successive Umbrella Sampling • (How to calculate interface tension?) Applications • Systematic approach to model and coarse-grain small molecules (Noble gases, Alkanes, CO2, benzene) G.M. Torrie and J.P. Valleau, J. Comput. Phys. 23, 187 (1977) B. Berg and T. Neuhaus, Phys. Rev. Letters 68, 9 (1992) F. Wang and D.Landau, Phys. Rev. Letters 86, 2050 (2001) Grandcanonical Simulations P(n) liquid gas particle number n Problem: Regions of small probability (free energy barriers) algorithms -> applications Free energy barriers P n exp (-H(n)/kT) modify Hamiltonian: weight function: P(n) H´(n) = H(n) + kT w(n) ideal: w(n)=ln P(n) good estimate for w(n) -> “flat” histogram particle number n algorithms -> applications How to generate a good w(n) ? 1. Histogram-reweighting (extrapolate data from -> limited range, tedious P(n) previous simulation) 2. Adjust w(n) during simulation i.e., Wang-Landau (-> violates detailed balance) algorithms -> applications particle number n Idea: - simulate windows successively - generate w(n) for the next window by extrapolation P(n) P(1) P(2) P(n) ... P(0) P(0) P(1) P(n - 1) P(n) Successive umbrella sampling Adv: - works everywhere - efficient P. Virnau and M.Müller, J. Chem. Phys. (2004) algorithms -> applications particle number n Successive umbrella sampling Analysis: no violation of detailed balance -> errors are controlled and can be calculated: P(n) P(1) D n D n D0 P(0) P(0) Errors are independent of window size single window: t O(n2) window size one: t n O(1) = O(n)??? D0 n t n t n O(n) = O(n2) algorithms -> applications Noble gases Potential: Lennard-Jones Strategy: Equate critical points of simulation and experiment e and s How well is the rest of the diagram described? algorithms -> applications Alkanes Mapping 3CH2=1 bead Potential Lennard-Jones + FENE algorithms -> applications Carbon dioxide Mapping: CO2= 1 LJ bead + quadrupole moment algorithms -> applications Benzene Mapping: C6H6= 1 LJ bead + quadrupole moment algorithms -> applications Take home messages How to overcome free energy barriers? Successive umbrella sampling Modeling Equate critical points of simulation and experiments to obtain simulation parameters very good agreement with phase change data Critical points Binder cumulants U4 = <M4>/<M2>2 M = r-<r> algorithms -> applications Free energy Relation btw. P(n) und F: DF/kT F = -kT ln(Zcan) = -kT lnP(n) + const. Interface tension: g DF/2L2 particle number n algorithms -> applications
© Copyright 2025 Paperzz