many bugfixes related committees

Author: Christoph Ortner

docs/src/tutorials/committee.jl docs/src_outdated/committee.jldocs/src/tutorials/committee.jl renamed to docs/src_outdated/committee.jl

@@ -334,7 +334,9 @@ function ace1_model(; kwargs...)
    end
 
    model_spec = Dict{Symbol, Any}(pairs(kwargs)...)
-   model_spec[:Eref] = ACEpotentials.Models._convert_E0s(kwargs[:Eref])
+   if haskey(model_spec, :Eref)
+      model_spec[:Eref] = ACEpotentials.Models._convert_E0s(kwargs[:Eref])
+   end
    model_spec[:model_name] = "ACE1"
 
    kwargs = _clean_args(kwargs)

docs/src/tutorials/lammps.md docs/src_outdated/lammps.mddocs/src/tutorials/lammps.md renamed to docs/src_outdated/lammps.md

@@ -301,3 +301,23 @@ function energy_forces_virial_basis(
 
    return (energy = E, forces = F, virial = V)
 end
+
+
+function potential_energy_basis(at, calc::ACEPotential{<: ACEModel}, 
+                                ps = calc.ps, st = calc.st; 
+                                domain = 1:length(at), 
+                                nlist    = PairList(at, cutoff_radius(calc)), 
+                                kwargs...)
+   N_basis = length_basis(calc)
+   _e0 = AtomsCalculators.zero_energy(at, calc)
+   T = typeof(ustrip(_e0))
+   E = fill(zero(T) * energy_unit(calc), N_basis)
+
+   for i in domain
+      Js, Rs, Zs, z0 = get_neighbours(at, calc, nlist, i) 
+      v = evaluate_basis(calc.model, Rs, Zs, z0, ps, st)
+      E += v * energy_unit(calc)
+   end
+
+   return E 
+end

docs/src/tutorials/molly.md docs/src_outdated/molly.mddocs/src/tutorials/molly.md renamed to docs/src_outdated/molly.md

@@ -30,12 +30,15 @@ function co_length(model::ACEPotential)
 end
 
 function committee(f, sys::AbstractSystem, model::ACEPotential)
-   E = f(sys, model)
+   if f == potential_energy
+      return co_potential_energy(sys, model)
+   end
+   F = f(sys, model)
    ps0 = model.ps
-   co_E = [ (model.ps = model.co_ps[i]; f(sys, model)) 
+   co_F = [ (model.ps = model.co_ps[i]; f(sys, model)) 
             for i = 1:length(model.co_ps) ]
    model.ps = ps0            
-   return E, co_E
+   return F, co_F
 end
 
 macro committee(ex)
@@ -44,3 +47,22 @@ macro committee(ex)
       committee($(esc_args...))
    end
 end
+
+function co_potential_energy(sys::AbstractSystem, model::ACEPotential)
+   basis_E = potential_energy_basis(sys, model)
+   eref = potential_energy(sys, model.model.Vref) * u"eV"
+   E = dot(basis_E, destructure(model.ps)[1]) + eref
+   co_E = [ dot(basis_E, destructure(model.co_ps[i])[1]) + eref 
+            for i = 1:length(model.co_ps) ]
+   return E, co_E
+end
+
+function co_potential_energy_2(sys::AbstractSystem, model::ACEPotential)
+   f = potential_energy 
+   F = f(sys, model)
+   ps0 = model.ps
+   co_F = [ (model.ps = model.co_ps[i]; f(sys, model)) 
+            for i = 1:length(model.co_ps) ]
+   model.ps = ps0            
+   return F, co_F
+end

docs/src/tutorials/python_ase.md docs/src_outdated/python_ase.mddocs/src/tutorials/python_ase.md renamed to docs/src_outdated/python_ase.md

@@ -35,3 +35,16 @@ maxdiff = maximum(abs(E_per_at[i] - E_per_at[j]) for i = 1:10, j = 1:10 )
 @test ustrip(u"eV", maxdiff) < 1e-9
 
 @info(" ============================================================")
+@info(" ============== Testing for no Eref bug ====================")
+
+# there was never an issue filed for this, but it was an annoying issue 
+# that came up twice by making changes in the model construction heuristics
+
+params1 = (elements = [:Si], rcut = 5.5, order = 3, totaldegree = 12)
+params2 = (; :Eref => [:Si => 0.0], pairs(params1)...)
+model1 = ace1_model(; params1...) 
+model2 = ace1_model(; params2...) 
+sys = bulk(:Si, cubic=true) * 2
+println_slim(@test potential_energy(sys, model1) == potential_energy(sys, model2) == 0.0u"eV")
+
+@info(" ============================================================")

examples/Tutorial/ACEpotentials-Tutorial.ipynb

@@ -104,7 +104,7 @@ println_slim(@test m2.ps == model.ps)
 
 ##
 
-info("Fit a potential with committee")
+@info("Fit a potential with committee")
 
 co_size = 10 
 solver = ACEfit.BLR(factorization = :svd, committee_size = co_size)
@@ -116,6 +116,19 @@ acefit!(data, model;
 
 println_slim(@test length(model.co_ps) == co_size)
 
+E, co_E = @committee potential_energy(data[3], model)
+E
+co_E 
+
+using LinearAlgebra
+M = ACEpotentials.Models
+efv = M.energy_forces_virial_basis(data[3], model)
+e = M.potential_energy_basis(data[3], model)
+println_slim(@test all(efv.energy .≈ e))
+e1, co_e1 = @committee potential_energy(data[3], model)
+e2, co_e2 = M.co_potential_energy_2(data[3], model)
+println_slim(@test e1 ≈ e2)
+println_slim(@test all(co_e1 .≈ co_e2))
 
 ##
 # Add a descriptor test