docs for loss throughout simulation

docs/src/differentiable.md

@@ -60,7 +60,7 @@ neighbor_finder = DistanceVecNeighborFinder(
     dist_cutoff=1.5,
 )
 lj = LennardJones(nl_only=true, force_units=NoUnits, energy_units=NoUnits)
-# Currently this is required too though here it does not affect the simulation
+# Currently required for speed though here it does not affect the simulation
 crf = CoulombReactionField(dist_cutoff=1.5, nl_only=true, coulomb_const=0.0,
                            force_units=NoUnits, energy_units=NoUnits)
 general_inters = (lj, crf)
@@ -101,11 +101,15 @@ function loss(σ)
     return loss_val
 end
 ```
-Now we can obtain the gradient of `loss` with respect to the atom property `σ`.
+Currently only the combination of `LennardJones` and `CoulombReactionField` works on the GPU or fast on the CPU.
+Other types work but run slower on the CPU.
+This will change in future.
+
+We can obtain the gradient of `loss` with respect to the atom property `σ`.
 ```julia
 grad = gradient(loss, σtrue)[1]
 ```
-We can use this gradient in a training loop to optimise `σ`, starting from an arbitrary value.
+This gradient can be used in a training loop to optimise `σ`, starting from an arbitrary value.
 ```julia
 function train()
     σlearn = 0.60 / scale_σ_to_dist
@@ -144,6 +148,12 @@ The final value we get is 0.449, close to the theoretical value of 0.445 if all
 The RDF looks as follows, with the purple line corresponding to the desired distance to the closest neighbor.
 ![LJ RDF](images/rdf_lj.png)
 
+It is common to require a loss function formed from values throughout a simulation.
+In this case it is recommended to split up the simulation into a set of short simulations, each starting from the previous final coordinates and velocities.
+This runs the same simulation but makes the intermediate coordinates and velocities available for use in the loss function.
+It is recommended to have a single value to accumulate values of the loss function.
+For example, the RMSD could be calculated from the coordinates every 100 steps and added to a variable that is then divided by the number of chunks to get a loss value corresponding to the mean RMSD over the simulation.
+
 ## Specific interactions
 
 Next we look at obtaining gradients through simulations with specific interactions, e.g. bonds or angles between specified atoms.