@Szabolcs Seems like the use case for Language`ExpressionStore?

I suppose the issue is that that only works for truly identical objects... on the other hand you can use an Association to keep track of the ExpressionStore keys since the value will be the unique instance and the key will be matched via SameQ or whatever Hash mechanism Mathematica uses

@Szabolcs Ah wait I misunderstood what you wanted. You don't need to associate values to the object. You just want the object. In that case Association will work. If you want to associate values to the object then you want something like this:

objectCache[store_Language`ExpressionStore[], keys_Association][key_] :=
  store@"get"[keys[key]];

And then you'll define an OOP interface on that

@JasonB. Any thoughts on why MoleculePlot3D@Molecule["ClP(Cl)(Cl)(Cl)Cl"] works but MoleculePlot3D@Molecule["FS(F)(F)(F)(F)F"] fails?

@bobthechemist The internal message i get is "UFFTYPER: Unrecognized atom type: S_6+6 (1)"

We use the method described here, pubs.acs.org/doi/abs/10.1021/acs.jcim.5b00654, to generate coordinates. That relies on atom typing, and hexavalent sulfur seems to be missing.

databases like pubchem, chemspider, etc, also seem to be missing 3D coordinates for SF6, probably they rely on similar code

This gives a decent result - Import["https://raw.githubusercontent.com/aoterodelaroza/refdata/master/20_g3/sf6.xyz"]

The bond lengths look right, haven't checked the angles though

Would it be unethical to ask someone for a formula then to try and do it myself. Terrible at math

Hey @LukasLang, I am enjoying the ease of PlotGrid, but it is resizing the ticks on the bottom of the plots? I have customized the sizes with a function, and the left sides do not get resized as the bottom sides do (they get bigger). Any advice?

@CATrevillian That should not happen (in theory at least ;) ) - can you show a specific example that reproduces the issue?

@JasonB. The reference and xyz file are helpful, thank you. Ultimately, I just want something that's hexavalent and has a oh point group. Out of the 12 "common" point groups typically taught in undergrad chem, there are 4 I cannot seem to find in ChemicalData.

fktick[min_, max_, dn_] :=
 Table[If[EvenQ[i], {1 + i*dn/Floor[max],
    i, {.01, 0.}}, {1 + i*dn/Floor[max], i, {.005, 0.}}], {i,
   Ceiling[min], Floor[max], 1}]
fntick[ndiv_][min_, max_] :=
 Table[If[Divisible[i, ndiv], {i, i, {.01, 0.}},
   If[EvenQ[i], {i, "", {0., 0.}}, {i, "", {0., 0.}}]], {i,
   Ceiling[min], Floor[max], 1}]
randset[{min_, max_, dn_}, ri_, rf_,
   rn_] := {RandomReal[{ri, rf}, {rn, dn + 1}]};
mweSEPlot[set_, vdiv_ : 4] := Module[{y = (Evaluate @@ set)[[1]]},
   ListLinePlot[y, PlotRange -> {Full, All}, Frame -> True,

@LukasLang for sure! ^^There's a decently stripped-down MWE that creates the problem^^

^^Show^^ produces that, but PlotGrid produces the following:

@CATrevillian Thanks for the example! Look like this is caused by this bug that has existed since version 7.0.1. Since I am using AspectRatio->Full to make the individual plots fill their grid space, custom tick mark lengths are broken.

I'll look into implementing the suggested workaround from the linked post into the next version, but I am not very hopeful that it will work out (since that fix relies on knowing the final ImageSize, which PlotGrid can not in general). As a manual workaround, I'd suggest you rescale the ticks manually for now...

@GWarner you mean here in this chat room or in general?

@ChrisK either or both

@GWarner give it a try!

@User21 A colleague has developed an indexing method which allows assembling the system of equations for FEM in CSR format directly from element matrices. It works well with the MKL Pardiso sparse solver. Is this something that you or WRI might be interested in? shearmode@gmail.com

@LeonidShifrin One of the problems with classical caching is that the keys of the cache make take up a lot of storage. Imagine dealing with a million-vertex graph. I don't want to store that in the cache. What do you think about @kirkus's suggestion to cache based on the Hash instead? I assume the chance of collisions is negligible?

Hash[] itself will cache its result, so using Hash on a large input is not a problem.

@ChrisK I was looking for a formula to describe traveling outward along a spiral (not Fibonacci) staying the same distance from the previous pass. I was hoping to adapt it to a grid search but decided there would be too much overlap in areas already checked.

@Szabolcs Yeah, I guess, for large enough expressions, Hash can provide some memory savings. You will still probably need to use buckets and resolve the collisions, but that should not be hard to implement. For buckets, you could just use lists, and perform linear search, since there won't be too many collisions in general. Overall, this doesn't sound like something hard to implement.

@Chrisk Closest example I could find. And it wont work well for a square grid.

Thanks !

It seems to me defining the coordinate system corresponding to slicing through a given object would be a useful thing to do?

@LukasLang aha! That makes a bit of sense. I wonder if one could make a generalized wrapper for the defined tick labels? Hmm, I'll have to think on it--for now I'll just manually adjust them.

@LukasLang how can you just use a PlotLegend for the whole kit and caboodle, rather than having them be automatically combined? I can't do it by simply using PlotLegend in PlotGrid