jbcaillau
commented
1 month ago @PierreMartinon in the current code below, compared to the previous tags, I don't see where you would skip the multipliers associated with constraints other than the dynamics (e.g. state constraints) in order to retrieve only the dynamics:
"""
$(TYPEDSIGNATURES)
Recover OCP costate from DOCP multipliers
"""
# +++todo: parse multipliers for the path and boundary constraints (use tuples)
function parse_DOCP_solution_dual(docp, multipliers)
# constraints, costate and constraints multipliers
N = docp.dim_NLP_steps
P = zeros(N, docp.dim_NLP_x)
if !isnothing(multipliers)
index = 1
for i in 1:N
# state equation multiplier for costate
P[i,:] = multipliers[index:index+docp.dim_NLP_x-1]
index = index + docp.dim_NLP_x
end
end
return P
end
PierreMartinon
alesiagr
ocots
@PierreMartinon As discussed: probable shift in
parse_DOCP_solution_dualwhen retrieving the multipliers associated with the dynamics (= adjoints). Bug seem to be activated for problems with nonlinear constraints (= not mere boxes on slices of state / control). See for instance this variant of Goddard that, contrary to this one, has a such a nonlinear constraint: https://github.com/control-toolbox/BVP-DAE/issues/1#issuecomment-2246348545Similar behaviour for these other projects that also have nonlinear (control or mixed) constraints: @alesiagr https://github.com/mctao-inria/sail-therm @PhysicDev https://github.com/PhysicDev/McTao_lunar_solar_sail