agerlach

> The main problem here is that algorithms that use vex::symbolic are not allowed to use any data-dependent branching. This is because C++ does not allow overloading conditional operators. That...

> I don't see any right away, but with the logic you showed the update to dt will be wasted anyway (the kernel will finish right after the update). Why...

> I meant a loop inside the kernel Would you implement the loop in the kernel by manually updating the output stream, e.g. ``` cpp body

I don't recall what exactly I meant by "updating". Here is a MWE I just put together, I'm not sure if it captures the same problem I had in the...

> It's maxiters: 🤦 Thanks I knew that some slack was required for maxiters for the various algs. It was the crashes that are the main issue. Its an easy...

I'll take a look at it. May not be until tomorrow though.

> But it's having an issue that the domain transformation is increasing nout? @agerlach is that supposed to happen? Yes. B/c the gradient is computed as the integral of the...

`CubatureJLh()` and `CubatureJLp()` have the same issues ```julia using Quadrature, Cubature qprob = QuadratureProblem((x,p)->sin.(x),0,π; nout=1, batch=0) sol = solve(qprob, CubatureJLh(), reltol=1e-3,abstol=1e-3) qprob = QuadratureProblem((x,p)->sin.(x),[0],[π]; nout=1, batch=0) sol = solve(qprob, CubatureJLh(),...

I understand that you may want to treat them different, however, both should work.

A couple of thoughts off the top of my head: 1. The Sherman–Morrison–Woodbury formula can be used to perform rank 1 updates on the inverse of a matrix. I have...