Open seishiroono opened 5 months ago
The adjoint of a BVP hasn't been added yet. It's actually written down just hasn't been put to the proper repos. We're working this out over the next few months.
Thanks for your reply. I understand the situation. I look forward to having such a wonderful update!
I also need to optimize the initial guess when I solve the differential equation
Even when you have adjoints, the gradient wrt the initial guess would be zero by IFT (unless I am getting something wrong)
I am trying to implement the "shooting method" to solve a simple differential equation (\ddot{x} = -x*(1-x^2) ) as a boundary value problem. I also need to optimize the initial guess when I solve the differential equation.
When I used
DifferentialEquations.jl
andOptimization.jl
for this problem, I encountered the following error. (Please seeoptimization_diff
in my code attached below.)On the other hand, if I solved the differential equation without the optimization (the function
diff_eq
), I did not enter any problem. Also, when I printed the properties of my problem, I found it isProblem type of differential equation: BVProblem...
. So, I wonder if the error comes from the incompatibility betweenOptimization.jl
andDifferentialEquations.jl
.Below is my environment.
MacOS (M2Max) Julia: 1.10.0 Zygote v0.6.68 DifferentialEquations v7.12.0 Optimization v3.20.2 OptimizationOptimJL v0.1.14 SciMLSensitivity v7.51.0