QED tests

[niclaurenti]

PR made to address issue #1735 , i.e. to create a low precision QED theory (similar to theory 399) that will be used to test the code and in particualr the photon module of validphys.

[scarlehoff]

Let's sit on this until @andreab1997 updates evolven3fit_new to utilize eko 0.13

[scarlehoff]

For the small eko for Theory 399 I've used evolven3fit_new -q 500 --q-points 10 produce_eko -i 0.00001 --x-grid-points 10 399 eko.tar

This produces an eko that weights only ~70 MB which I think it is ok for the tests. I'm going to open a PR with tests for evolven3fit_new, you can rebase on top of that branch and use the same test.

[RoyStegeman]

@niclaurenti going forward, could you try rebasing a bit more often to avoid the merge commit?

[niclaurenti]

@scarlehoff @RoyStegeman Since the call to fiatlux takes a lot should I make that if debug: true the photon is computed on a smaller number of points and reinterpolated?

[scarlehoff]

Is the number of points not exposed in the runcard? One could want to run with debug: true while having the correct photon.

[niclaurenti]

Is the number of points not exposed in the runcard? One could want to run with debug: true while having the correct photon.

No it is computed for every point in the XGRID. That's why it takes so much.
In the runcard there is

fiatlux:
  luxset: NNPDF40_nnlo_as_01180
  additional_errors: true # should be set to true only for the last iteration
  luxseed: 1234567890

Maybe I could add a key debug under fiatlux. Or we can just accept the slowness of generating the photon

[scarlehoff]

I think a better solution is to expose the integration precision to the runcard.

nnpdf/validphys2/src/validphys/photon/compute.py

Line 22 in 6bfc220

"eps_base": 1e-5, # precision on final integration of double integral.

Then it becomes fast and you can test that things are working (we don't care about correctness in the test after all).

Edit: I just tried with eps_base: 1e-2 and the Basic_runcard_qed.yml ran quite fast.

[niclaurenti]

@scarlehoff @RoyStegeman where does the test_evolven3fit.py is supposed to download the theories 399 and 398? Or they are downloaded somewhere else?

[scarlehoff]

I haven't looked at the error, but the easiest solution is to add API.theoryid(theoryid=THEORYID_399) at some point at the beginning of the test so that vp downloads it.

[niclaurenti]

Should I add some other tests?

[RoyStegeman]

I think this is done now?

[niclaurenti]

I think this is done now?

yes let's wait for the test to pass and then we can merge

[RoyStegeman]

Thanks - looks good to me and tests seem to pass.