Revision as of 16:07, 10 September 2013

Equilibration

Rule of thumb: All Monte Carlo simulations have to be equilibrated before taking measurements.

Example: Classical Monte Carlo (local updates) simulations

As an example, we will implement a classical Monte Carlo simulation implemented in the Ising model on a finite square lattice of size 48².

Using command line

The parameter file parm1a:

LATTICE="square lattice"
T=2.269186
J=1
THERMALIZATION=10000
SWEEPS=50000  
UPDATE="local"
MODEL="Ising"
{L=48;}

We first convert the input parameters to XML and then run the application spinmc:

parameter2xml parm1a
spinmc --Tmin 10 --write-xml parm1a.in.xml

Add in timeseries analysis here after Python

Based on the timeseries, the user will then judge for himself/herself whether the simulation has reached equilibration.

Using Python

The following describes what is going on within the script file tutorial1a.py.

The headers:

import pyalps

Set up a python list of parameters (python) dictionaries:

parms = [{
  'LATTICE'         : "square lattice",
  'MODEL'           : "Ising",
  'L'               : 48,
  'J'               : 1.,
  'T'               : 2.269186,
  'THERMALIZATION'  : 10000,
  'SWEEPS'          : 50000,
}]

Write into XML input file:

input_file = pyalps.writeInputFiles('parm1a',parms)

and run the application spinmc:

pyalps.runApplication('spinmc', input_file, Tmin=10, writexml=True)

We first get the list of all result files via:

files = pyalps.getResultFiles(prefix='parm1a')

and then extract, say the timeseries of the |Magnetization| measurements:

ts_M = pyalps.loadTimeSeries(files[0], '|Magnetization|');

We can then visualize graphically:

import matplotlib.pyplot as plt
plt.plot(ts_M)
plt.show()

Based on the timeseries, the user will then judge for himself/herself whether the simulation has reached equilibration.

A convenient tool: pyalps.checkNonSteadyState

ALPS Python provides a convenient tool to check whether a measurement observable(s) has (have) not reached steady state equilibrium. Read [http//alps.comp-phys.org/mediawiki/index.php/Documentation:Monte_Carlo_Equilibration#link_equilibration_theory here]

Here is one example (observable: |Magnetization|, confidence interval = 95%):

pyalps.checkSteadyState(files[0], '|Magnetization|', confidenceInterval=0.95)

and another one:

pyalps.checkSteadyState(files[0], ['|Magnetization|', 'Energy'], confidenceInterval=0.95)

Description
1. steady_state_check first performs a linear fit on the timeseries, and decides whether the measurement observable has reached steady state equilibrium based on the gradient/slope of the fitted line.

2. The optional arguments of steady_state_check are:

argument	default	remark
confidenceInterval	0.01
simplified	False	shall we combine the checks of all observables as 1 final boolean answer?
includeLog	False	shall we print the detailed log?

3. To see the complete log for instance:

pyalps.checkSteadyState(files[0], ['|Magnetization|', 'Energy'], confidenceInterval=0.95, includeLog=True)

Using Vistrails

To run the simulation in Vistrails open the file mc-01b-equilibration.vt.