Launching a calculation is not everything. One wants the results. Pylada makes it easy to interface with those results directly in the python language. Every functional in Pylada returns an extraction object capable of grepping the relevant output.
>>> result = vasp(structure)
>>> print result.success
True
>>> print result.eigenvalues * 2
[5, 6, 7] * eV
The above checks that the calculation ran to completion, and then multiplies the eigenvalues by two. At this point, one could perform any sort of post-processing, and then automatically launch a subsequent calculation.
Warning
Success means the calculation ran to completion, specifically that the lines giving the elapsed time exist in the OUTCAR. It does not mean that the results are meaningful.
The extraction object can be obtained without rerunning VASP. There are to ways to go about this. One is, well, to rerun (but without rerunning, if that makes sense):
>>> # first time, vasp is executed, presumably.
>>> result = vasp(structure, outdir='mycalc', ispin=1)
>>>
>>> # abort if not successful
>>> assert result.success
>>>
>>> ... do something ...
>>>
>>> # second time, OUTCAR exists. It is NOT overwritten.
>>> # The extraction object is returned immediately.
>>> result = vasp(structure, outdir='mycalc', ispin=2)
In the example above, VASP is actually launched the first time. However, on the second pass, an OUTCAR is found. If it is a successful run, then Pylada will not overwrite it. It does not matter whether the structure has changed, or whether the VASP parameters are different. Pylada will never overwrite a successful run. Not unless specifically requested to. The returned extraction object corresponds to the OUTCAR. Hence, on the second pass, it is the results of the first call which are returned. Unless, of course, a successful calculation already existed there prior to the first run, in which case Pylada would never ever have been so crass as to overwrite it.
The second method is to create an extraction object directly:
>>> from pylada.vasp import Extract
>>> result = Extract('/path/to/directory')
In the above, it is expected that the OUTCAR is called OUTCAR. The path can also be made to a file with a name other than OUTCAR.
Note
An extraction object can be created for any OUTCAR, whether obtained via Pylada or not. Some information that Pylada automatically appends to an OUTCAR may not be obtainable, however.
To find out what Pylada can extract, do result.[TAB] in the ipython environment. Or checkout Extract.
If you know how to use regular expressions, creating a property like those above is generally fairly simple. Edit the file “vasp/extract/base.py”, reinstall, and you’re golden. And send your snippet back this way.
Pylada arranges all calculations within directories, with a single VASP calculation per sub-directory. It can be expedient to extract simultaneously all the results contained within a directory and its subdirectories. One approach is to use jobfolders and the ipython interface. Failing that, however, it still possible to extract all the results within a tree of directories simultaneously. When used in conjunction with plotting software such as matplotlib, it makes it really easy to synthesize and understand the results from a set of calculations. It all comes down to a few simple lines:
>>> from pylada.vasp import MassExtract
>>> a = MassExtract('some/path')
>>> a.total_energies
{
'/this/path/': array(-666.667) * eV,
'/that/path/': array(-999.998) * eV
}
“this/path” and “that/path” are directories in “some/path” where OUTCAR files exist. The return is a ForwardingDict instance. It is possible to string together attributes to get to those of interest:
>>> a.structure.scale
{
'/this/path/': 5.45,
'/that/path/': 5.65
}
From there, it is one simple step to plot, say, energies with respect to the scale (first, run ipython with the -pylab flag to import matplotlib related stuff, or run the ipython notebook app):
>>> x = array(a.structure.scale.values())
>>> y = array(a.total_energies.values())
>>> plot x, y
MassExtract behaves exactly like the collect object.