mcmc_steps = 10000
minimization_steps = 50
burn_in_steps = minimization_steps
particle_number = 10
max_delta = 1.0
cutoff = 0.98 * (upper - lower) / 2
potential = LennardJonesOpt(sigma=1., epsilon=3.0, cutoff=cutoff)
p_group = ParticleGroup.random_square(number=particle_number,
lower=lower,
upper=upper,
dimension=2,
kind='p',
use_cpp=False)
p_group.attach_pairwise_potential(potential)
propagator = Propagator(
p_group, termo_properties=['potential_energy', 'trajectory'])
#####
propagator.minimize(minimization_steps)
propagator.propagate_mcmc_nvt_onep(steps=mcmc_steps,
temperature=temperature,
max_delta=max_delta)
print(propagator.last_run_time)
propagator.burn_in(burn_in_steps)
#propagator.dump_to_xyz('test_lj.xyz')
acceptances.append(propagator.get_acceptance_percentage())
print(f"Acceptance is {propagator.get_acceptance_percentage()}")
propagator.clear_properties()
pickle.dump((temperatures, acceptances),
open('temperatures_acceptances.pkl', 'wb'))
import chem_mcmc
import numpy as np
import matplotlib.pyplot as plt
from chem_mcmc.staging import Propagator, ParticleGroup
from chem_mcmc.potentials import LogGaussianOpt
from chem_mcmc.utils import get_running_mean, get_running_std, PDensity
from chem_mcmc import potentials
from chem_mcmc import plotting
from chem_mcmc import constants
# EXAMPLE 1 do this once
temperature = 600
lower = 0.
upper = 12.
# ~ 80k is converged (10 s) , ~ 800k is very well converged
mcmc_steps = 80000
#potential = LogGaussian(A=[2., 0.8], mu=[4, 8])
potential = LogGaussianOpt(A=[2., 0.8], mu=[4, 8])
p_group = ParticleGroup.random_square(number=1,
lower=lower,
upper=upper,
dimension=1,
kind='r')
p_group.attach_external_potential(potential)
propagator = Propagator(p_group,
termo_properties=['potential_energy', 'trajectory'])
propagator.propagate_mcmc_nvt(steps=mcmc_steps,
temperature=temperature,
max_delta=1.0)
# Simulation setup #
temperature = 10000
pressure = 2 # units should be kcal/(mol ang^2)
# an approximate idea given by the ideal gas law is that T*N * R * 4.184/upper^2
# \approx P
lower = 0.
upper = 12.
mcmc_steps = 50000
potential = HardSpheresStep(sigma1=0.5, sigma2=1.0, epsilon=2)
p_group = ParticleGroup.random_square(number=20,
lower=lower,
upper=upper,
dimension=2,
kind='p')
p_group.attach_pairwise_potential(potential)
propagator = Propagator(
p_group,
termo_properties=['potential_energy', 'trajectory', 'bound_sizes'])
propagator.minimize(50)
propagator.propagate_mcmc_npt(steps=mcmc_steps,
temperature=temperature,
pressure=pressure,
max_delta_coord=1.0)
propagator.burn_in(50)
propagator.dump_to_xyz('npt_low_fixed.xyz')
#bound_sizes = propagator.get_bound_sizes()
#np.set_printoptions(threshold=3000)
#print(bound_sizes)
#####
from chem_mcmc.potentials import LogGaussian from chem_mcmc.utils import get_running_mean, get_running_std, PDensity from chem_mcmc import potentials from chem_mcmc import plotting from chem_mcmc import constants from time import time # EXAMPLE 1 do this once temperature = 600 lower = 0. upper = 12. # ~ 80k is converged (10 s) , ~ 800k is very well converged mcmc_steps = 20000 potential = LogGaussian(A=[2., 0.8], mu=[4, 8]) p_group = ParticleGroup.random_square(number=1, lower=lower, upper=upper, dimension=1, kind='r') p_group.attach_external_potential(potential) propagator = Propagator(p_group, termo_properties=['potential_energy', 'trajectory']) propagator.propagate_mcmc_nvt(steps=mcmc_steps, temperature=temperature, max_delta=1.0) propagator.burn_in(100) position = propagator.get_trajectory()[:,0,0] energy = propagator.get_potential_energy() pdensity = PDensity(temperature, potential, lower, upper) plotting.plot_analysis(position, pdensity, lower, upper, potential, bounds_potential = [-5, 40]) runmean_position = get_running_mean(position) runstd_position = get_running_std(position) runmean_energy = get_running_mean(energy) runstd_energy = get_running_std(energy) properties = [get_running_mean(position), get_running_std(position), get_running_mean(energy), get_running_std(energy)] titles = [r'Mean position ($\AA$)', r'Std. position ($\AA$)', r'Mean potential energy (kcal/mol)', r'Std. potential energy (kcal/mol)'] titlex = r'MCMC steps'
def replica_remc(random_square_kwargs, propagate_mcmc_nvt_onep_kwargs,
potential, termo_properties, minimization_steps,
burn_in_steps, mcmc_steps, replica_idx, connections):
p_group = ParticleGroup.random_square(**random_square_kwargs)
p_group.attach_pairwise_potential(potential)
propagator = Propagator(p_group, termo_properties=termo_properties)
propagator.minimize(minimization_steps)
iterations = mcmc_steps // propagate_mcmc_nvt_onep_kwargs['steps']
replica_acceptance = 0
for iteration_idx in range(iterations):
propagator.propagate_mcmc_nvt_onep(**propagate_mcmc_nvt_onep_kwargs)
# Only returns true if the parities are different, for even
# iterations this has to be done by odd replicas
# for odd iterations this has to be done by even replicas
if iteration_idx % 2 != replica_idx % 2:
# for even iterations I have to exchange 0-1 2-3 4-5 6-7 so the
# processes in charge of the exchange will be 0 2 3 4 6 ... etc
# This means I will only do the calculation if the process index is
# even, but I first need to receive a value from the processes with
# odd process index
# odd indices send data to even indices
right_energy = propagator.get_potential_energy()[-1]
right_temperature = propagate_mcmc_nvt_onep_kwargs['temperature']
connections['left'].send(
(right_energy, right_temperature, replica_idx))
# after doing this I try to receive some coordinates, if I receive
# something then the exchange was successful! if I receive None then
# the exchange was not successful
left_particle_group = connections['left'].recv()
if left_particle_group is not None:
# if the exchange was successful I do this
connections['left'].send(propagator.particle_group)
propagator.particle_group = left_particle_group
propagator.store_termo(temperature=right_temperature)
replica_acceptance += 1
else:
# if the exchange was unsuccessful I store the same
# conformation again and continue
propagator.store_termo(temperature=right_temperature)
# Only returns true if the parities are equal, for even
# iterations this has to be done by even replicas
# for odd iterations this has to be done by odd replicas
# This is the code that actually manages the exchange calculations
if iteration_idx % 2 == replica_idx % 2:
# Here I receive data and calculate the mcmc exchange
# probability
left_energy = propagator.get_potential_energy()[-1]
left_temperature = propagate_mcmc_nvt_onep_kwargs['temperature']
left_beta = 1 / (constants.kb * left_temperature)
# this blocks until there is something to receive
right_energy, right_temperature, right_replica_idx = connections[
'right'].recv()
right_beta = 1 / (constants.kb * right_temperature)
mc_factor = np.exp(
(left_beta - right_beta) * (left_energy - right_energy))
if propagator.prng.uniform(low=0., high=1.) < mc_factor:
# first I send the particle group through the pipe
connections['right'].send(propagator.particle_group)
# afterwards I receive the particle right particle_group
# from the pipe I change the coordinates of all the
# particles to be those of the particle group and I store
# the termo properties as if the temperature was the
# temperature of this specific replica
propagator.particle_group = connections['right'].recv()
propagator.store_termo(temperature=left_temperature)
replica_acceptance += 1
else:
# I just send an "exchange rejected signal" and store the
# termo properties again
connections['right'].send(None)
propagator.store_termo(temperature=left_temperature)
replica_acceptance_rate = replica_acceptance / iterations