The Multicanonical Approach

The entropic sampling version of the multicanonical algorithm, as applied to a discrete lattice model of a chain with attracting interactions, is of interest and is described here. Let the number of different energies of the system be L, and let i andbe chain conformations, with the degeneracy of energy E denoted by (compare with Eqs. [3], [13], and [14]). The multicanonical probability of conformation i is 

The problem of course lies in the fact that the entropies are not known a priori. The entropic sampling version provides a simple prescription enabling one to build, in a recursive way, a function J(E) that is proportional to S(E). More specifically, the process consists of several separate simulations carried out with transition probabilities 

The next simulation is carried out with the new set of /( ) using Eq. [83], where more bins are visited, and the histogram H(E) is calculated and used to update 

If the total number of configurations ft is known, at the end of the process one can obtain the absolute entropies. 

In any case, the final set of J(E) enables one to estimate the ensemble average of any function X at any temperature T, 

---

However, since and -5 asymptote to the same function, one might approximate (U) = S dJ) in (3.57) so that the acceptance probability is a constant.3 The procedure allows trial swaps to be accepted with 100% probability. This general parallel processing scheme, in which the macrostate range is divided into windows and configuration swaps are permitted, is not limited to density-of-states simulations or the WL algorithm in particular. Alternate partition functions can be calculated in this way, such as from previous discussions, and the parallel implementation is also feasible for the multicanonical approach [34] and transition-matrix calculations [35],... 

One of the drawbacks of the multicanonical method is that, during the simulations tc derive the weight factor, the energy distribution in H(E) can oscillate rather than steadilj approaching a limiting distribution. Another drawback is that it can fail to properlj... 

Not surprisingly, the essential component of flat-histogram algorithms is the determination of the weights, r/, or the thermodynamic potential, e.g., / or /. There exist a number of techniques for accomplishing this task. The remainder of this section is dedicated to reviewing a small but instructive subset of these methods, the multicanonical, Wang-Landau, and transition-matrix approaches. We subsequently discuss their common and sometimes subtle implementation issues, which become of practical importance in any simulation. 

Transition matrix estimators have received less attention than the multicanonical and Wang-Landau methods, but have been applied to a small collection of informative examples. Smith and Bruce [111, 112] applied the transition probability approach to the determination of solid-solid phase coexistence in a square-well model of colloids. Erring ton and coworkers [113, 114] have also used the method to determine liquid-vapor and solid-liquid [115] equilibria in the Lennard-Jones system. Transition matrices have also been used to generate high-quality data for the evaluation of surface tension [114, 116] and for the estimation of order parameter weights in phase-switch simulations [117]. 

Fig. 3. Scaling of round-trip times for a random walk in energy space sampling a flat histogram open squares) and the optimized histogram solid circles) for the two-dimensional fully frustrated Ising model. While for the multicanonical simulation a power-law slowdown of the round-trip times 0 N L ) is observed, the round-trip times for the optimized ensemble scale like 0([A ln A ] ) thereby approaching the ideal 0(A )-scaling of an unbiased Markovian random walk up to a logarithmic correction...

In this section we describe several methods not pertaining to the techniques described earlier. We discuss the multicanonical method of Berg as applied to macromolecules, " " - and the adiabatic switching procedure of Rein-hardt, which is related to the thermodynamic integration approach but is based on different grounds. For completeness, we mention four additional techniques, three of which were developed originally for spin models. 

We will now take a closer look at the adsorption transition in the phase diagram (Fig. 13.12) and we do this by a microcanonical analysis [307, 308]. As we have discussed in detail in Section 2,7, the microcanonical approach allows for a unique identification of transition points and a precise description of the energetic and entropic properties of structural transitions in finite systems. The transition bands in canonical pseudophase diagrams are replaced by transition lines. Figure 13.15 shows the microcanonical entropy per monomer s e)=N lng e) as a function of the energy per monomer e=EfN for a polymer with N=, 20 monomers and a surface attraction strength = 5, as obtained from multicanonical simulations of the model described in Section 13.6. 

Among the methods discussed in this book, FEP is the most commonly used to carry out alchemical transformations described in Sect. 2.8 of Chap. 2. Probability distribution and TI methods, in conjunction with MD, are favored if there is an order parameter in the system, defined as a dynamical variable. Among these methods, ABF, derived in Chap. 4, appears to be nearly optimal. Its accuracy, however, has not been tested critically for systems that relax slowly along the degrees of freedom perpendicular to the order parameter. Adaptive histogram approaches, primarily used in Monte Carlo simulations - e.g., multicanonical, WL and, in particular, the transition matrix method - yield superior results in applications to phase transitions,... 

---