VUMPS, Lapack running error at final outer_iters

DMRG and Numerical Methods
1

I got Lapack error. I set outer_iters = 5.
When running calculations on some machines, it gives a Lapack error.

ERROR: LoadError: LAPACKException(22)

2

I’m not sure we’ll be able to say much, I suspect its related to this issue: Occasional `LAPACKException(22)` in `eigsolve` · Issue #57 · Jutho/KrylovKit.jl · GitHub

If you have a degeneracy, I would suggest adding a symmetry breaking term and slowly turning it off if possible

1 Like
3

Thanks
but It doesn’t seem to be that kind of issue. When running the calculation on a different computer, it works without any problems.
There’s no error when setting max_vumps_iters=2 and running the calculation.

4

Different computers can have different floating point roundoffs, which could change the eigenvalue degeneracy enough to cause the eigensolver to either work or not work.

5

Thanks!
How can I solve this problem by working around it?
Now solver tolerance is set as follows.
‘solver_tol = (x → x / 100)’

Would it work if I changed this as shown below?
‘solver_tol = (x → x / 10)’

6

I’m not sure, it seems like the source of the error is a bit mysterious, it might help to change the KrylovKit eigensolver parameters or add symmetry breaking terms as Ryan suggested. I think you’ll just have to experiment with things, feel free to update us if you find a robust fix.

An alternative could be to try a different eigensolver implementation, like GitHub - JuliaLinearAlgebra/Arpack.jl: Julia Wrappers for the arpack-ng Fortran library, but that would require changing the source code of ITensorInfiniteMPS.jl (and it may have the same issue as KrylovKit.jl since I think they both call LAPACK anyway for the core diagonalization).

7

Thanks!
I will test it

8

After updating the Julia version to 1.11.1, the error disappeared.
Previously, I was using version 1.8.5.

9

Interesting, I’m not sure why the Julia version would affect this problem but that’s good to know, thanks for the update.