Incorrect eigvector in case of ex4_4k input

[Hegden]

Matlab (test3.m) uses: mat2hsssym, diagonal block size=128, A=load('hsseigen/MatrixData/ex4_4k.txt');
CPP: ./Test ex4_4k.txt 4096 128 1 1 1
both Matlab and cpp software try to obtain the eigenvector corresponding to the largest eigenvalue (4096th in this case).
As can be seen from the picture uploaded along with this bug, the first 256 components of the eigenvector differ. The .txt files uploaded along with this bug, show the exact values of vector components.

matlabvec.txt
cppvec.txt

[priteshiitdh]

Hi Sir,

Regarding the issue raised above, I would suggest performing the following quick checks:

• Verify whether the vectors obtained from CPP and MATLAB have the same magnitude and direction.
• Verify whether the orthogonal property holds true for the vectors.

Performing these checks should help us identify any potential discrepancies between the vectors obtained from the two platforms.

[Hegden]

as can be seen from the data files attached, vectors obtained from CPP and MATLAB have different direction. The magnitude is almost same except for the first 256 components of the vector.

[priteshiitdh]

Hi @Hegden Sir,
I have calculated the eigenvalues using Hsseigen and the LAPACKE routine, and have computed the relative error. I then plotted the results as mentioned in the previous comment. As you can see, the eigenvalues appear to be within a tolerable limit.

[priteshiitdh]

I was able to execute the eigenvector test for LAPACK and HSSEigen, and it seems to have a similar error to MATLAB's.