mrcdr / lambda-lanczos Goto Github PK

Hi! Thank you for the provided package.

Currently, I need to calculate the smallest algebraic eigenvalue of large matrices (>1mln x 1mln). The current serial version is definitely not adapted for this purpose. What do you think, is it possible to parallelize the code using OpenMp (possibly some loops)?
Thank you!
Murod

A small suggestion:
Replace:

#include <lambda_lanczos/lambda_lanczos_util.hpp>

with

#include <lambda_lanczos_util.hpp>

(You will also have to modify the files in the src directory and change the g++ arguments.)

It is not necessary to move lambda_lanczos.hpp or lambda_lanczos_util.hpp. They can remain inside the lambda_lanczos/ subdirectory. But changing the #include statement in lambda_lanczos.hpp will give other developers more freedom to move your files where they prefer.

For example, I use your lambda_lanczos repository as a git submodule inside one of my repositories. (Including your code as a submodule prevents me from modifying your code.)

People who use my repository will probably copy your files to a new directory containing my code. They must move your files to a "lambda_lanczos/" subdirectory, or modify the #include statements. I think C++ programmers are smart enough to figure this out. Anyway, I added a warning message.

This is a very small issue. I hesitate to bother you. (Perhaps I should avoid using git submodules.)

Hi mrcdr,

I very much like your implementation :-) Is there any possibility to figure out the original column / position of the largest Eigenvector? This information could be necessary to me.

Cheers,
Dominik

I think here should be return std::sqrt(r2);

I just wanted to point out that lambda_lanczos.hpp uses std::cout. I suggest adding

#include <iostream>

to the beginning of the file.
Cheers

Andrew