Frank Schäfer
Frank Schäfer
Updated code based on https://github.com/SciML/DifferentialEquations.jl/issues/646 (moving the callback to the problem type instead of the `solve` call). If I choose the number of trajectories `numtraj` >= `Threads.nthreads()`, some trajectories are...
Maybe, it would be a good idea to add also the `pCN` function: https://github.com/SciML/DiffEqNoiseProcess.jl/blob/master/src/pCN.jl (which allows one to generate a correlated noise process given an initial noise realization)
no, we only implemented the one due to Roessler: https://onlinelibrary.wiley.com/doi/abs/10.1002/pamm.200410005 (for his SRK methods in https://github.com/SciML/StochasticDiffEq.jl/issues/182).
yeah we never use the full matrix `Ihat2` but only its entries in `perform_step!`. These (scalar) noise increments are only used in the last stage value in all SRK schemes...
hmm that one looks worse. It's pointing to: ```julia g2 = [zero(noise_rate_prototype) for k=1:m] ``` https://github.com/SciML/StochasticDiffEq.jl/blob/a4497c5fd54f9a7cd895db145e877cc1ef083fae/src/caches/srk_weak_caches.jl#L598 I don't see how one could change that to avoid the `m` stored copies...
Interesting.. and `SOSRI()` and `SOSRA()` do work? The error is pointing to https://github.com/SciML/StochasticDiffEq.jl/blob/a4497c5fd54f9a7cd895db145e877cc1ef083fae/src/caches/srk_weak_caches.jl#L642 which is just the third cached noise vector. `SOSRI()` should have (up to the Butcher table) exactly...
Couldn't we make it a single but longer `dW'` vector and have a ```julia dW = @view dW`[1:m] dZ = @view dW`[m:end] ``` in perform_step() ? (and maybe reserve the...
It happens only in a test environment + it didn't happen before we included the other CPU tests in the queue: https://buildkite.com/julialang/stochasticdiffeq-dot-jl/builds/33#04baae20-0c63-4e81-b71d-db70df5e518d
I had to remove the type fixes (They caused an error: ` MethodError: Cannot `convert` an object of type Float64 to an object of type SArray{Tuple{2},Float64,1,2}` or with `Array{Float64,1}` if...
If we define the functional of the expected value at a time point T as: $$V_{11}^{dt} = E(f(Y(T)^{dt}))$$, the extrapolation methods allow us to get a higher order convergence by...