You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I ran into an interesting numerical issue in Julia that seems to be related to a bug in LinearAlgebra.
Specifically, I have a vector with norm $\approx 1$ that behaves as expected when using a Linux machine but has various Nan issues on a Mac (Apple Silicon).
I will include a minimal working example below, but there is also more discussion at ITensor/ITensorMPS.jl#110. (I first experienced the issue when using this package but it seems independent of it.)
To verify it's a problem with a buggy OpenBLAS kernel you can fallback to the generic one by exporting the environment variable OPENBLAS_CORETYPE=armv8 before starting Julia. Conversely, you should be able to reproduce the issue in Julia v1.10 by exporting OPENBLAS_CORETYPE=neoversen1, because between Julia v1.10 and Julia v1.11 OpenBLAS changed the autodetection of the CPU on Apple Silicon: JuliaLang/julia#52762.
I ran into an interesting numerical issue in Julia that seems to be related to a bug in LinearAlgebra.
Specifically, I have a vector with norm$\approx 1$ that behaves as expected when using a Linux machine but has various Nan issues on a Mac (Apple Silicon).
I will include a minimal working example below, but there is also more discussion at ITensor/ITensorMPS.jl#110. (I first experienced the issue when using this package but it seems independent of it.)
Here is the vector itself:
Here are some examples of the kinds of things that go wrong on a Mac:
The text was updated successfully, but these errors were encountered: