mc

The mc keyword is used to carry out Monte Carlo (MC) trial steps, usually in combination with a MD simulation. Three MC ensembles are supported, including the canonical, the semi-grand canonical (SGC), and the variance-constrained semi-grand canonical (VCSGC) [Sadigh2012a] [Sadigh2012b] ensemble.

This keyword requires an ordinary energy NEP (model nep) in the nep.txt format.

Syntax

mc canonical <md_steps> <mc_trials> <T_i> <T_f> [group <grouping_method> <group_id>]
mc sgc <md_steps> <mc_trials> <T_i> <T_f> <num_species> {<species_0> <mu_0> <species_1> <mu_1> ...} [group <grouping_method>  <group_id>]
mc vcsgc <md_steps> <mc_trials> <T_i> <T_f> <num_species> {<species_0> <phi_0> <species_1> <phi_1> ...} <kappa> [group <grouping_method>  <group_id>]

Parameters

For all three forms, mc_trials trials are performed every md_steps MD steps. The MC target temperature changes linearly from T_i to T_f during the run.

  • The listed species must be supported by the NEP model.

  • For all the MC ensembles, there is an option to specify the grouping method grouping_method and the group ID group_id in the given grouping method, after the parameter group. The functionality is illustrated in the example section below.

  • There must be at least one listed species in the initial model system or specified group. For example, if you list Au and Cu for doing SGC MC, the system or the specified group must have some Au or Cu atoms (or both); otherwise the MC trial cannot get started.

canonical

Canonical sampling exchanges atomic species while keeping their counts fixed.

sgc

SGC sampling uses species transmutations with chemical-potential differences.

num_species specifies the number of species that are to be included in the sampling. It must be no less than 2 and no larger than 4. After specifying the number of species, one needs to specify their chemical symbols (species_i) and chemical potentials (mu_i) in units of eV. The species can be listed in arbitrary order. Note that only the differences between the chemical potentials matter.

For a trial transmutation from species \(a\) to species \(b\), GPUMD uses the acceptance probability

\[P_{a\to b}=\min\left[1,\exp\left(-\frac{\Delta U+\mu_b-\mu_a}{k_{\rm B}T}\right)\right],\]

where \(\Delta U\) is the change in potential energy. Thus, increasing the input \(\mu_b\) makes conversion into species \(b\) less favorable. For a convention based on \(U-\sum_s\mu_s^{\rm phys}N_s\), the corresponding input values are \(\mu_s=-\mu_s^{\rm phys}\) up to a common additive constant.

vcsgc

VCSGC sampling adds a concentration bias.

num_species specifies the number of species that are to be included in the sampling. It must be no less than 2 and no larger than 4. After specifying the number of species, one needs to specify their chemical symbols (species_i) and dimensionless concentration-bias parameters (phi_i = \(\phi_i\)). The species can be listed in arbitrary order. Only differences between the \(\phi_i\) parameters enter the acceptance criterion. Next one needs to specify the (dimensionless) kappa parameter (\(\kappa\)).

The \(\phi\) and \(\kappa\) parameters constrain the average and variance of the species concentrations, respectively. For a binary system, a symmetric scan can use \(\phi_A=-\phi\) and \(\phi_B=\phi\), varying the common scan parameter \(\phi\) approximately from -1.2 to +1.2. This changes \(\phi_B-\phi_A\) from -2.4 to +2.4; it is distinct from varying only one species parameter while fixing the other. The resulting compositions depend on the interactions and temperature and should be checked from the sampled concentrations. Typically a \(\kappa\) value of 100 is suitable. If the concentration fluctuations are too large (e.g., deep with miscibility gaps) one should increase this value.

The parameter normalization differs from that used in the original papers [Sadigh2012a] [Sadigh2012b]; see also [Rahm2021].

Output

Monte Carlo statistics are written to mcmd.out.

Examples

Example 1

An example for sampling in the canonical ensemble is:

mc canonical 100 200 500 100 group 1 3

This means

  • Perform 200 MC trials after every 100 MD steps.

  • Change the temperature for the MC simulation linearly from 500 to 100 K.

  • Only the atoms in group 3 of grouping method 1 will be considered during MC sampling.

Example 2

Here is an example for MC sampling the SGC ensemble:

mc sgc 100 1000 300 300 2 Cu 0 Au 0.6

This means

  • Perform 1000 MC trials after every 100 MD steps.

  • The temperature for the MC ensemble will be kept at 300 K.

  • Only the Cu and Au atoms are involved in the MC process. The input parameter for Au is 0.6 eV higher than that for Cu, so the parameter contribution to the acceptance exponent favors Au-to-Cu transmutations.

Example 3

Here is an example for sampling in the VCSGC ensemble:

mc vcsgc 200 1000 500 500 2 Al -2 Ag 0 10000

This means

  • Perform 1000 MC trials after every 200 MD steps.

  • The temperature for the MC ensemble will be kept at 500 K.

  • Only the Al and Ag atoms are involved in the MC process. The dimensionless \(\phi\) parameters for Al and Ag are -2 and 0, respectively. The dimensionless \(\kappa\) parameter is 10000.

Notes

None.