Add tests to CI

[lkdvos]

This moves some of the tests that were in the examples folder to a proper testing framework

[lkdvos]

This seems like its still not very stable, and we should probably do something about the overly verbose printing... I might be able to find some more time this afternoon

[pbrehmer]

I will also take a look in a bit, maybe I manage to stabilize the tests.

[pbrehmer]

I like how MPSKit prints out the first and last VUMPS iterations when you set verbosity=1. I think this would be useful to have also in PEPSKit and it would give some useful printing during the tests. Let me quickly add that, while I'm at it!

[lkdvos]

I definitely agree, sadly the machinery for that is still in the untagged master of mpskit, so its not so easy to just import it :p

[pbrehmer]

Maybe, once it's tagged, we can use that machinery to make the output printing and logging a bit more nice :)

[pbrehmer]

Well, the tests ran through on v1.9.3, let's see if I can also get them to run on v1.6.

Update: The gauge-fixing failed (for certain seeds) because the environment dimension was too low - increasing it solved it. I don't get why the success of the gauge fixing should be dependent on the environment dimension? (The error was $\mathcal{O}(1)$ and not just a bit inaccurate, so something failed completely...)

[pbrehmer]

@lkdvos Do you think we could get rid of the Julia 1 tests? I'm not sure if it makes sense to be backwards compatible to v1.0 if PEPSKit anyway still is in a very experimental stage. Also it seems to be difficult to get the tests running consistently on both v1.6 and v1.0. Maybe we could instead test on the latest stable release.

[lkdvos]

I think the Julia 1 tests expand to the latest stable Julia, so 1.10 in this case...

[pbrehmer]

Whoops, that's good to know and definitely makes more sense. I'll try again tomorrow to get the tests running consistently (and maybe to speed them up, so the CI doesn't take ages...)

[lkdvos]

hahahhaha, so now I've fixed v1 but broke 1.6...

[pbrehmer]

It's really annoying that some of these runs are so sensitive (at low bond/environment dimension) to the random initial state. I guess it takes a bit of trying around with the random seed to find one that works for both versions. I can try around tomorrow if you don't have the time/motivation now (which would be perfectly understandable :D)

[lkdvos]

So at this point, I'm basically ready to give up...
Probably it is better to just look for a reasonable physical state somewhere, such as some state which is at the very least approximately close to a groundstate?
I'm fine with merging like this, and maybe we can add an ED thing on a 3x3 lattice later and then just decompose that state and use the center peps tensor?

[pbrehmer]

Thanks for the effort, it really is annoying! Having an ED initial PEPS also sounds like a good idea. I will give it a final go tomorrow but then I'd also be fine with merging.

[lkdvos]

Small FYI, I tried playing around with the Ising PEPS, also does not seem to work (doesnt even want to converge with CTMRG). I think that one might also just be a bit pathological anyways, or I might have implemented it wrong, but I am kind of running out of ideas. I would think the best approach is to just save a reasonable state on disk somewhere, and use that

[pbrehmer]

I would think the best approach is to just save a reasonable state on disk somewhere, and use that

Yes, I agree that this is the most sensible approach in the long run. Maybe we can code up a small state/environment saving and loading interface soon. I guess that's useful to have anyways.

[pbrehmer]

Hope that this runs for now... I think this should be ready to merge!