Eigsolve converged to wrong eigenvalue?

[sunnowrun]

I am using eigsolve to compute the eigenvector of a tensor, where the vector has a dimension of 80,000. However, I have observed that eigsolve often converges to an eigenvalue that is not of the largest magnitude.
Is this a common issue? And how can I improve the success rate of converging to the dominant eigenvalue?

[lkdvos]

Sorry for the late reply to this:

Do you think it is possible to provide slightly more information about your case?
Is there any example so we can reproduce this behavior?

In principle, as long as the starting vector has some overlap with the largest magnitude eigenvector, we would expect that you would converge to the correct eigenvector.
Is there any reason to expect that your initial vector is orthogonal to the largest magnitude one?

Does this behavior persist with different settings for the tolerances and Krylov dimension?

There are some caveats with using Krylov methods for degenerate subspaces, but if you are just interested in the largest magnitude eigenvalue that should be fine.

[sunnowrun]

Thank you for your response.

I'm working within a substantial codebase, which makes it challenging to provide a minimal reproducible example.

I have verified that the initial guess vector has a non-zero overlap with the target eigenvector.

Additionally, I encountered occasional convergence issues when using Arpack.jl. Increasing the number of desired eigenvalues (nev) from 1 to 4 appeared to resolve the problem.

I appreciate your insights. Best regards,