evenmn / vmachine Goto Github PK

In the current version, the interaction energy (inter-particle energy) is restricted to Coulomb interaction, and is located in the Hamiltonian class. At the same time, the kinetic energy is called from the Hamiltonian class, which does not really makes sense.

In order to make VMaChine more flexible and accept other interaction styles, it would be better to create an interaction class where Coulomb is a subclass and None is another subclass (and the default). The local energy should be computed in the sampler.

Add support for the PyTorch C++ API. This can be used for setting up models, autograd, optimization and GPU acceleration

system.cpp contains the write-functions and the parser. These could with advantage be moved to respective files (ex. write.cpp and read.cpp).

Radial onebody density and twobody density does not seem to work as expected

The code can be made more transparent. First, the set*-methods in system are no longer needed after the configuration file was made the way to configure the system. When the set*-methods are removed, the initialize-methods found in all classes can be moved to init, making the code more readable.

There is an issue when trying to compile VMaChine with CMake.

The log should be written in a more readable way

Write up documentation. This is very easy to do using Sphinx

The current VMC algorithm can be improved by allowing for variance minimization and more sophisticated optimization algorithms based on the Hessian

There are a few places where memory is allocated, but never freed up (src/system.cpp and src/initialWeights/automaize.cpp). This is unfortunate and should be fixed asap.

Find a neat way to represent the various objects, preferably similar to the built-in repr()-method in Python.

Include blocking code in VMC code

A Python interface could be implemented in order to ease the usage of the package. Using PyBind11, this should now take more than a few hours