Quickstart¶
ESPEI has two different fitting modes: singlephase and multiphase fitting. You can run either of these modes or both of them sequentially.
To run either of the modes, you need to have a phase models file that describes the phases in the system using the standard CALPHAD approach within the compound energy formalism.
You also need to describe the data that ESPEI should fit to.
You will need singlephase and multiphase data for a full run.
Fit settings and all datasets are stored as JSON files and described in detail at the Making ESPEI datasets page.
All of your input datasets should be validated by running espei checkdatasets myinputdatasets
, where myinputdatasets
is a folder of all your JSON files.
The main output result is going to be a database (defaults to out.tdb
), an array of the steps in the MCMC trace (defaults to trace.npy
), and the an array of the logprobabilities for each iteration and chain (defaults to lnprob.npy
).
Singlephase only¶
If you have only heat capacity, entropy and enthalpy data and mixing data (e.g. from firstprinciples), you may want to see the starting point for your MCMC calculation.
Create an input file called espeiin.yaml
.
system:
phase_models: myphases.json
datasets: myinputdatasets
generate_parameters:
excess_model: linear
ref_state: SGTE91
Then ESPEI can be run by running
espei input espeiin.yaml
Multiphase only¶
If you have a database already and just want to do a multiphase fitting, you can specify a starting TDB file (named mytdb.tdb
) with
system:
phase_models: myphases.json
datasets: myinputdata
mcmc:
iterations: 1000
input_db: mytdb.tdb
The TDB file you input must have all of the degrees of freedom you want as FUNCTIONs with names beginning with VV
.
Restart from previous runphase only¶
If you’ve run an MCMC fitting already in ESPEI and have a trace file called myprevioustrace.npy
, then you can resume the calculation with the following input file
system:
phase_models: myphases.json
datasets: myinputdata
mcmc:
iterations: 1000
input_db: mytdb.tdb
restart_trace: myprevioustrace.npy
Full run¶
A minimal full run of ESPEI with single phase fitting and MCMC fitting is done by the following
system:
phase_models: myphases.json
datasets: myinputdata
generate_parameters:
excess_model: linear
ref_state: SGTE91
mcmc:
iterations: 1000
Input Customization¶
ESPEI lets you control many aspects of your calculations with the input files shown above. See ESPEI YAML input files for a full description of all possible inputs.
FAQ¶
Q: There is an error in my JSON files¶
A: Common mistakes are using single quotes instead of the double quotes required by JSON files. Another common source of errors is misaligned open/closing brackets.
Many mistakes are found with ESPEI’s checkdatasets
utility.
Run espei checkdatasets myinputdatasets
on your directory myinputdatasets
.
Q: How do I analyze my results?¶
A: By default, ESPEI will create trace.npy
and lnprob.npy
for the MCMC chain at the specified save interval and according to the save interval (defaults to ever iteration).
These are created from arrays via numpy.save()
and can thus be loaded with numpy.load()
.
Note that the arrays are preallocated with zeros.
These filenames and settings can be changed using in the input file.
You can then use these chains and corresponding logprobabilities to make corner plots, calculate autocorrelations, find optimal parameters for databases, etc..
Finally, you can use py:mod:espei.plot functions such as multiplot
to plot phase diagrams with your input equilibria data and plot_parameters
to compare singlephase data (e.g. formation and mixing data) with the properties calculated with your database.
Q: Can I run ESPEI on a supercomputer supporting MPI?¶
A: Yes! ESPEI has MPI support. See the Advanced Schedulers page for more details.
Q: How is the log probability reported by ESPEI calculated?¶
MCMC simulation requires determining the probability of the data given a set of parameters, \(p(D\theta)\). In MCMC, the log probability is often used to avoid floating point errors that arise from multiplying many small floating point numbers. For each type of data the error, often interpreted as the absolute difference between the expected and calculated value, is determined. For the types of data and how the error is calculated, refer to the ESPEI paper [1].
The error is assumed to be normally distributed around the experimental data point that the prediction of a set of parameters is being compared against.
The log probability of each data type is calculated by the log probability density function of the error in this normal distribution with a mean of zero and the standard deviation as given by the data type and the adjustable weights (see data_weights
in ESPEI YAML input files).
The total log probability is the sum of all log probabilities.
Note that any probability density function always returns a positive value between 0 and 1, so the log probability density function should return negative numbers and the log probability reported by ESPEI should be negative.
Q: Why is the version of ESPEI ‘0+unknown’?¶
A: A version number of '0+unknown'
indicates that you do not have git installed.
This can occur on Windows where git is not in the PATH (and the Python interpreter cannot see it).
You can install git using conda install git
on Windows.
References¶
[1] 
