With the addition of
SROCKC2 all the methods mentioned in StochasticDiffEq.jl Issue #73 are complete.
KomBurSROCK2 is similar to
SROCK2 for Stratonovich Problems.
I completed implementing
TangXaioSROCK2W2Ito. The last of these is not currently used. One of the main advantages of SROCK methods is the varying number of stages which can be chosen from depending on the spectral radius of the problem. But the paper describing
TangXaioSROCK2W2Ito only has constants for 5 stages and it is not trivial to calculate these constants so this method is not suggested. Along with these I also tried improving
RKMilCommute GPU Compatibility and Memory utilization. I also improved the General Noise handling in
SROCK1, adding noise rate type stability along with some memory optimisation. I added more robust tests for Commutative Noise Solvers.
I have completed the implementation of
ESERK4 method. I also worked towards making the ROCK and SERK solvers GPU compatible. This allows using DifferentialEquations.jl problem type mixed with neural networks. These methods are a very promising option in training data generation, for example
ROCK2 taking 3 minutes against 40 minutes with
The Julia Community is simply amazing. Along the way I had my share of genuine but sometimes stupid and irritating doubts, everybody was really helpful. Slack is very active and very well organised. I’m really grateful to be a part of this community.