NTPoly
NTPoly copied to clipboard
william-dawson
A massively parallel library for computing the functions of sparse matrices.
This pull request has dense implementations of each of the core matrix functions in terms of an eigensolver. This might be useful if you want to verify the accuracy of...
I am wondering whether it is possible to perform dense sparse matrix multiplication via NTPoly subroutines. In FHI-aims, we need to perform some matrix multiplication for rotation matrices (sparse) on...
https://pubs.acs.org/doi/full/10.1021/acs.jctc.6b00626 I've tested this out so far and it seems really nice. It will definitely make it easier to integrate the code since you don't need to manage the convergence...
Apparantly SWIG can [automatically generate](http://www.swig.org/Doc4.0/Doxygen.html#Doxygen_translation_overview) documentation in python using the C++ Doxygen.
# Overview # In principle it should be able to follow similar polynomial expansions for the non-symmetric case. In particular, the Taylor series expansions or perhaps using the Pade approximations...
# Overview # NTPoly needs more sophisticated error controls. I think that [this paper](http://aip.scitation.org/doi/abs/10.1063/1.2826343) shows some possible way forward.
One of our builds crashed because `MPI_LONG` is a MPI C datatype. I guess the correct corresponding MPI long integer in Fortran should be `MPI_INTEGER4 `. I fixed this already...
The solver parameters object would be a good place to store things like: * Did the calculation converge? * Did it converge from meeting the threshold or autostop? * What...
These changes appeared to be necessary for tests to pass with OneAPI 2023.2.0. "Fixes" in quotes because the code was correct as written before.
Essentially this would speed up the operation AD = B, where D is a diagonal matrix. My hope is that by putting this in the dense solvers, it can improve...
Metadata
Owner
Metadata
A massively parallel library for computing the functions of sparse matrices.