About

Pylada is a modular python framework to control physics simulations, from DFT to empirical pseudo-potentials to point ion electrostatics. It’s goal is to provide the basic building blocks from which methods incorporating different Hamiltonians can be constructed. It is designed around three main concepts:

• constructing and manipulating periodic crystal structures. For instance, it is possible, starting from the unit cell of spinel to automatically create a supercell with vacancies or substitutions.
• manipulating functionals, such as VASP, and extracting their output. For instance, one could throw diamond at VASP, expect it to relax the structure, then perform a static calculation for maximum accuracy. Or possibly perform an epitaxial relaxation, something VASP does not do per say.
• Manipulate, launch, and check the results for thousands of calculations simultaneously. Think of calculating different of carbone allotropes. It would be impractical to qsub each and every job and then check that each ran correctly. Hence, Pylada provides an interface for that.

With these building blocks in hand, it is possible, fairly easy, and - I’m told - even fun, to construct complex computational routines. Computing phonons or formation enthalpy requires some know-how, certainly. But it is also quite repetitive. Pylada takes the gruelling out of it.

Here are a few of the projects Pylada helped us with.

• When brute force suffices: A₂BX₄ (A, B cations, X=O, S, Se, Te) form a vast family of compounds crystallizing in forty different structures. However, hundreds of possible combinations of A, B, X have never been identified in nature before. Do these exist and have simply never been reported? Or are they thermodynamically unstable and could never have been found? To answer this question, we systematically computed the formation enthalpy of hundreds of unknown A₂BX₄ as well as that of their (reported) competing binaries and ternaries. We found indeed quite a few compounds, some which may be fairly easily grown, which simply passed through the sieve. Pylada made it a sinch to perform these individual calculations. It allowed us to shove them in a database and then scour the data to automatically create phase diagrams for each system.
• Stay abstract and leave off mano a mano calculations: One of the goals of the project described above was to explore stable A₂BX₄ for semi-conducting properties. More specifically, we were interested in predicting whether these compounds can be doped and how much. Or whether some point defects intrinsic to the material would form spontaneously when external dopants are introduced and limit the material’s carrier concentration. To do this, we created a simple script to automatically determine possible vacancies and substitutions, as well as their charge state. These defects are all the same, so no need to worry and create each, one at a time. Pylada made it possible to create the systems, launch the appropriate chain of calculations, and gather the results, and let physicists do physics rather than word processing input files.
• Manipulating DFT functionals: This DFT functional was introduced to reproduce GW band-structures. It works by fitting a non-local potential. Originally, each fit would be performed by hand for each element. It turned out to be tedious. Especially since for more complicated compounds one would rather fit all the elements involved simultaneously. But Pylada offers python wrappers around DFT codes, and python offers a number of optimization functions. The two together make it possible to carry out painless optimization of a empirical DFT functional.
• Searching for and finding a needle: Silicon is a possibly God’s greatest gift for nerds everywhere. It is n-type. It is p-type. It naturally forms a protective insulating oxide layer. However, it is an indirect gap material and absorbs visible light only through the mediation of phonons. It has been known for a while that it is possible to create Si/Ge superlattices which are nominally direct gap. Indeed, Si₆Ge₄ on a strained (001) substrate absorb light at the band edges. But only academically so. It is possible that a superlattice with specific motif and Si and Ge layer with meaningful photon absorption exists somewhere, but there a literally billions upon trillions of possibilities. So which one to start off with? At that point, research tapered off. We combined an empirical pseudo-potential method with a genetic algorithm to perform the search for us. Eventually, the search payed off.
• screwing up: Speaking from painful and embarrassing personal experience, it is extremely easy to launch tens and hundreds of meaningless jobs. The physics is still for the user to figure out.