How do I speed up ODE15s?
19 次查看(过去 30 天)
显示 更早的评论
Hello everybody I am trying to speed up a MATLAB code which numerically solves for deactivation process in a catalyst. The algorithm of the code which simply solves mass and momentum conservation equations plus an equation for chemical reaction over time and space(1D) is like below:
main()
statements
Initial values and boundary conditions
Timespan=[0 3600] // By discretizing equations in Z direction we get ODEs and then implement stiff ode solver to sweep in time
[t,w]=ODE15s(@transientfunction,timespan,w0)
outputs+graphs
end
transientfunction()
"FOR" loop // "for" loop is implemented to allocate the differential values at each cell which is around 100 cells.
end
***************************************************************Problem******************************************************
My problem here is that simulation of one hour (3600s) of deactivation process takes two days!!!(at the end I want to simulate it for 5years which I guess my grandson can check the result and be proud of his grandpa) I have tried to use "parfor" in transientfunction with 12 workers but not only the speed did not increase but also decreased! Time step in ODE15s solver is,specially at the beginning, gravely small. Is there any suggestion or advice out there that might help me or help this nice and lovely ode solver to speed up I look forward to your suggestions Thank you all Rasoul
4 个评论
Cam Salzberger
2017-10-5
Hey Rasoul,
Maybe just post a couple of snippets, showing the for loop limits and a couple of assignments if possible. I can try to have a look on here.
I did just notice you mentioning assigning about 100 cells. I'm assuming that means you're returning a vector of length ~100 from your transientfunction? That's a lot of variables to solve over, and fairly uncommon for these types of problems. I'm not sure if it's possible to narrow that down (very problem dependent) or maybe use a different framework (optimization to get close first or something)?
Definitely don't go the way of the parfor in the transient function. Way too much overhead for something that gets called that often.
-Cam
采纳的回答
Josh Meyer
2017-9-21
Some things to try...
- For stiff problems you should always specify the Jacobian or its sparsity pattern via the options structure. Sometimes this can greatly improve performance since the solver estimates it numerically when you don't specify it. See odeset for more info.
- Try some of the other stiff solvers since they use a slightly cruder tolerance to calculate steps ( ode23s, ode23t, ode23tb). Since your ODE function contains a loop the fastest solver is likely to be the one with the fewest function evaluations. You can check that by setting 'Stats','on' with odeset and comparing the stats of each solver.
- If your equations include some conservation equations then you might have a system of Differential Algebraic Equations. If that's the case then you can continue using ode15s but might need to set some options as outlined on the linked page.
更多回答(0 个)
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Ordinary Differential Equations 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!