RDF calculation results are not normalized when using "around" in atom group selection

[Chengeng-Yang]

Dear Developers,

I'm posting this because I have a question regarding RDF calculation inconsistency. I realized that the calculation results are not normalized when I used "around" in atom group selection of one group.

I have a quick example below:

Case 1
from MDAnalysis.tests.datafiles import TPR,XTC
from MDAnalysis.analysis import rdf

u1 = mda.Universe(TPR,XTC)
res60 = u1.select_atoms('resid 60')
water60 = u1.select_atoms('resname SOL')
rdf1 = rdf.InterRDF(res60, water60,
nbins=75, # default
range=(0.0, 15.0), # distance in angstroms
)
rdf1.run()

plt.plot(rdf1.results.bins,rdf1.results.rdf)

Case 2
from MDAnalysis.tests.datafiles import TPR,XTC
from MDAnalysis.analysis import rdf

u1 = mda.Universe(TPR,XTC)
res60 = u1.select_atoms('resid 60')
water60 = u1.select_atoms('byres resname SOL and around 15 group res60',res60 = res60, updating = True)
rdf1 = rdf.InterRDF(res60, water60,
nbins=75, # default
range=(0.0, 15.0), # distance in angstroms
)
rdf1.run()

plt.plot(rdf1.results.bins,rdf1.results.rdf)

Question

Could anyone help explain why there's such a difference in the rdf calculation and the normalization doesn't work? Usually, norm is set to "rdf" as default. So I was confused how the calculation works. I'm using MDAnalysis 2.4.3 and python 3.8.18. Thanks so much and I appreciate any answers.

[orbeckst]

Looking at the source code

mdanalysis/package/MDAnalysis/analysis/rdf.py

Lines 314 to 318 in 263bbe6

	if self.norm == "rdf":
	# Number of each selection
	nA = len(self.g1)
	nB = len(self.g2)
	N = nA * nB

I see that the normalization takes the size of the AtomGroups g1 and g2 at the last frame. With an updating AtomGroup (as in your second example), you will get a different number than if you had the same number of waters in the group at every step (first example). Thus, I'd say, the norm="rdf" is not guaranteed to work with updating groups.

I'd think, norm=None will give similar numbers — can you check?

You can raise an issue for this problem (if you have the "Create issue from this discussion" link then you can use it directly). At a minimum, RDF should fail if g1 or g2 are updating and norm is not None. Ideally, we calculate the normalization correctly.