Lightspeed (LS) is a C++/Python library to facilitate the rapid development and deployment of new ideas in Gaussian basis electronic structure theory (EST). The main precept of LS is that most EST methods consist of two fundamentally different operations:
-
A few common numerical kernels, which are encountered in many places. Examples include Gaussian integrals, evaluation of basis function properties on grid points, construction of J/K matrices, construction of low-rank factorizations of interaction integral tensors, contractions of density matrices with integrals and integral derivatives, etc. These operations are often exceedingly rate-limiting in terms of FLOPs, memory, and disk storage. Moreover, there may exist several different algorithms to do any one of these operations.
-
An enormous number of combinations of these numerical kernels to produce the myriad methods of EST. One notable example is how HF/DFT, CIS/TD-HF/TDA/TD-DFT, CPHF, and limited active-space CASCI/CASSCF methods can all be formulated in terms of rate-limiting J/K builds, plus large amounts of wrapper code that is not rate limiting (eigensolves, guesses, iteration control and stabilization, etc).
Lightspeed does not do electronic structure theory -- Lightspeed helps you do electronic structure theory. LS accomplishes this by providing a clean interface to fast algorithms for evaluation of the heavy numerical kernels in Point 1, leaving more time and flexibility for the user to focus on the layering of these kernels to produce useful EST methods in Point 2.
The part of this repository which constitutes LS is in src/*
EST (Electronic Structure Theory) is a simple Python library which implements production-scale electronic structure methods in a flexible, extensible manner. One of the key design goals with EST is that there is no "main" method and that EST methods should be built as objects for each problem encountered. These objects can solve for the electronic wavefunction, compute extra desired properties on demand (gradients, couplings, overlaps, observables), and allow fine-grained access to the state variables computed during wavefunction optimization.
The part of this repository which constitutes EST in in est/*
The major LS source directories are:
- src - the main C++/Python source tree constituting the heart of LS
- est - a simple python-based electronic structure code stack
- data - ancillary data needed to use LS, such as basis sets
- tests - unit/production tests for the LS library
More minor LS source directories are:
- md - a simple adiabatic / overlap following molecular dynamics code.
- apps - small plugin applications using LS and EST to accomplish custom theoretical chemistry workflows.
- src/lightspeed - C++ library API (wrapped ~100% to Python)
- src/python - C++ -> Python wrapping and Python extensions
- est - Python EST code stack
More detailed documentation, installation instructions, tutorials, and unit tests are forthcoming.
- Robert Parrish ([email protected])
- Xin Li - Integrals/Gradients/Infrastructure
- Jason Ford - BlurBox Routines/Infrastructure
- Ruben Guerrero - Davidson/Potential Gradients
RMP is pleased to acknowledge many ideas from Justin Turney, Francesco Evangelista, Daniel Smith, Lori Burns, David Sherrill, Nathan Luehr, Ed Hohenstein, and Todd Martinez