Check how order parameters treat particles with zero neighbors

[bdice]

In #672 and #624, we discussed how the Steinhardt order parameter treats particles with zero neighbors. Steinhardt assigns a value of NaN. We need to check to see how other methods (Hexatic, Translational, SolidLiquid, LocalDescriptors) handle cases with zero neighbors.

The follow-up PR would basically be to review freud for other methods where this problem could arise and copy over the test to make sure the same behavior is consistent, as well as updating docs with the note and snippet added here.

Originally posted by @vyasr in #672 (comment)

[klywang]

For particles without neighbors:

• Steinhardt: Returns nan for wl=True and wl=False
• Hexatic: Returns nan
• Translational: Returns 0
• SolidLiquid: Returns cluster_sizes as array of ones
• LocalDescriptors: num_sphs returns 0 and sph=[]

With the exception of the Translational method, @alacour feel that these are appropriate outputs for particles with zero neighbors.

For the Translational method, we request comments from someone with more expertise on the topic. Specifically, does the value of 0 carry any physical meaning?

[joaander]

For the Translational method, we request comments from someone with more expertise on the topic. Specifically, does the value of 0 carry any physical meaning?

I don't know of anyone with expertise on the topic. Translational in freud is not a standard translational order parameter.

[bdice]

Thanks for checking into that! I discussed with @klywang and @alacour. Our goals for the end of the hackathon are to add documentation and tests for the behavior of each of those classes above. Also we might slip in a LaTeX fix for the docstring of https://freud.readthedocs.io/en/latest/modules/order.html#freud.order.SolidLiquid.particle_harmonics. @klywang also noted that SolidLiquid.compute should return self. It'd be great to have a PR with some or all of those changes.

[vyasr]

@klywang @bdice @alacour Since Translational is deprecated and we're now thinking about plans for 3.0 in light of other open work, I wouldn't worry about the behavior of translational at all.