AMGX Fails to Converge to the Correct Solution

[fxuecnu]

Hello everyone,

I'm currently trying to solve a large linear system using AMGX and I'm a bit stuck. My matrix data is in CSR (Compressed Sparse Row) format. I must confess that my knowledge of linear algebra isn't very strong, which might be part of the problem.

I've attempted to use almost all the default configuration files provided by AMGX, but I can't seem to get the correct results. Interestingly, I can solve the system correctly in MATLAB, but it's too slow, which is why I turned to AMGX.

I'm not sure how to tweak the configuration files to get the correct solution in AMGX. Could anyone offer some advice or share experiences that might point me in the right direction? Any suggestions on specific parameters or configurations that could be crucial for improving the solver's accuracy or performance would be greatly appreciated.

Thank you!
matrix_info.zip

[marsaev]

Can I ask you how you constructed this matrix? There are a lot of zeros on the diagonal, which doesn't play well with typical smoothers which scale by inverted diagonal value.

[fxuecnu]

Hello Marsaev, Thank you for your response. I generated the matrix in the same manner as in issue 305. 徐凡 副研究员 华东师范大学 ***@***.*** 18851122579 上海市东川路500号河口海岸大楼 签名由 网易灵犀办公 定制 Original: From：marsaev ***@***.***>Date：2024-07-04 09:01:06(中国 (GMT+08:00))To：NVIDIA/AMGX ***@***.***>Cc：fxuecnu ***@***.***> , Author ***@***.***>Subject：Re: [NVIDIA/AMGX] AMGX Fails to Converge to the Correct Solution (Issue #302) Can I ask you how you constructed this matrix? There are a lot of zeros on the diagonal, which doesn't play well with typical smoothers which scale by inverted diagonal value. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>