Add superacids category: HF/SbF5 mixture densities and HF structure factor (Ref #379)

.gitignore

@@ -15,6 +15,8 @@
 !dvc.yaml
 *.pdf
 *.hdf5
+*.mda
+*.mdt
 *.traj
 *.ipynb
 *.jpg

docs/source/user_guide/benchmarks/index.rst

@@ -19,3 +19,4 @@ Benchmarks
     conformers
     molecular_dynamics
     defect
+    superacids

docs/source/user_guide/benchmarks/superacids.rst

@@ -0,0 +1,98 @@
+==========
+Superacids
+==========
+
+HF/SbF5 mixture densities
+=========================
+
+Summary
+-------
+
+Performance in predicting the liquid density of HF/SbF5 mixtures, the archetypal
+superacid system, at three compositions: pure HF, a 10 mol % SbF5 mixture, and pure
+SbF5. Each system consists of about 200 atoms, simulated at 288.65 K and 1 atm.
+
+Metrics
+-------
+
+1. MAPE
+
+For each composition, the density is calculated from the average volume of an NPT
+molecular dynamics run of 100 ps. The first half of the simulation is discarded as
+equilibration. The mean absolute percentage error over the three compositions is
+compared to the reference densities, obtained from experiment.
+
+Computational cost
+------------------
+
+High: about 8 GPU hours per model for all three compositions, for a model that
+includes a dispersion correction.
+
+Data availability
+-----------------
+
+Input structures:
+
+* Generated with Packmol, available from the ML-PEG S3 bucket
+
+Reference data:
+
+* Shair and Schurig, Vapor-Liquid Equilibrium of Antimony Pentafluoride-Hydrogen
+  Fluoride. Ind. Eng. Chem. 43, 1624 (1951). https://doi.org/10.1021/ie50499a042
+* Experimental
+
+
+HF structure factor
+===================
+
+Summary
+-------
+
+Performance in predicting the total neutron structure factor of liquid HF, which probes
+the hydrogen-bonded chain structure of the liquid. A system of 100 HF molecules is
+simulated at 296 K and 1.2 bar for 50 ps of NPT molecular dynamics.
+
+Since the experimental reference is measured on the deuterated liquid, all H are
+transmuted to D before the structure factor is computed. S(q) is obtained with MDANSE,
+as the Fourier transform of the pair distribution function weighted by coherent
+scattering lengths, using the second half of the trajectory sampled every other frame.
+The real-space cutoff is kept the same for every model, so that the transform is
+truncated identically.
+
+Metrics
+-------
+
+1. S(q) R-factor
+
+The relative deviation of the calculated structure factor from experiment,
+``sum|S_exp - S_calc| / sum|S_exp|``, evaluated from the first experimental point to
+4 1/A. Calculated and experimental structure factors are computed on the same grid of
+scattering vectors.
+
+2. First peak position error
+
+The absolute error in the position of the maximum of S(q), located over the same range
+as the R-factor. The position is read directly off the grid of scattering vectors,
+whose spacing is 0.05 1/A.
+
+The "good" threshold of the R-factor is the statistical noise of the protocol itself,
+measured by splitting the production window of one model into two halves.
+
+Computational cost
+------------------
+
+High: about 6 GPU hours per model, for a model that includes a dispersion correction.
+
+Data availability
+-----------------
+
+Input structures:
+
+* Generated with Packmol, available from the ML-PEG S3 bucket
+
+Reference data:
+
+* McLain, Benmore, Siewenie, Urquidi and Turner, On the Structure of Liquid Hydrogen
+  Fluoride. Angew. Chem. Int. Ed. 43, 1952 (2004).
+  https://doi.org/10.1002/anie.200353289
+* Experimental