How to autodifferentiate the results of NLsolve?

[Tamas_Papp]

Also, this univariate worked example may help to get you started:

ChainRulesCore and ForwardDiff

Thanks. I managed to adapt the docs example to use the frule with ForwardDiff. This MWE uses bisection (to keep it simple) to solve x^{\varepsilon_1} + x^{\varepsilon_2} = \theta \qquad \varepsilon_1, \varepsilon_2, \theta > 0 If it sees ForwardDiff.Dual, it just invokes the frule. This works and I don’t see any obvious problems with inference, but suggestions to improve how I hook into ForwardDiff are welcome (the rest is really just an MWE). using ForwardDiff, ChainRulesCore, FiniteDiffer…