What is the highest model order (No. of differential equations) with which somebody has succesfully used PyDSTool?
I have a model with 10000+ nonlinear ODEs, is it possible to use PyDSTool?
Equations are Stiff & Nonlinear. I am using fortran90, for transient simulation purpose with DVODE as ODE solver. Is it possible to use such large ordered system with PyDSTool? Any experiences, please share.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
In lieu of any other responses, I'm not familiar with examples over 1000 personally. I don't see that 10000 would be a problem if you have plenty of RAM. You might need to run shorter time epochs if it fills up quickly, maybe saving and restarting then stitching it all together afterwards. That's all quite easy to do. Stiff systems with Radau will of course use small steps for all variables whenever any one variable is stiff. Radau is also not a sparse solver, so there will be very large memory footprints if you have very stiff systems. In any case, don't use the built in VODE!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Exactly, "Shorter Time Epochs" of saving and restarting.
This is just what I am doing, as of now to solve for the transient simulation of the large ordered system.
Good to hear, for large order system it shall work.
Can you please provide any references of 1000 ordered system? I would have a look for my betterment.
Windows64 and linux64 bit, can PyDSTool work without any hiccups?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I don't have any references, sorry. I don't recall who has told me about that and I don't think it has been recorded anywhere. No hiccups at all for 64bit? Maybe not, especially on Windows -- mostly because the scipy stack including cygwin/mingw and gcc/gfortran can be tricky to install in a way that allows users to link their own gcc-made DLLs (which is how Dopri/Radau work). Bundled binary environments like Anaconda are usually compiled with an MS C compiler that can't be linked to unless you own that same compiler. You might have to install the entire python stack from source, but then BLAS etc. become the stumbling points. I'm not sure what the current status of that situation is as I haven't done any Windows setups for more than a year. But I know of many folk who are using it on 64 bit linux/OS X.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
What is the highest model order (No. of differential equations) with which somebody has succesfully used PyDSTool?
I have a model with 10000+ nonlinear ODEs, is it possible to use PyDSTool?
Equations are Stiff & Nonlinear. I am using fortran90, for transient simulation purpose with DVODE as ODE solver. Is it possible to use such large ordered system with PyDSTool? Any experiences, please share.
In lieu of any other responses, I'm not familiar with examples over 1000 personally. I don't see that 10000 would be a problem if you have plenty of RAM. You might need to run shorter time epochs if it fills up quickly, maybe saving and restarting then stitching it all together afterwards. That's all quite easy to do. Stiff systems with Radau will of course use small steps for all variables whenever any one variable is stiff. Radau is also not a sparse solver, so there will be very large memory footprints if you have very stiff systems. In any case, don't use the built in VODE!
Exactly, "Shorter Time Epochs" of saving and restarting.
This is just what I am doing, as of now to solve for the transient simulation of the large ordered system.
Good to hear, for large order system it shall work.
Can you please provide any references of 1000 ordered system? I would have a look for my betterment.
Windows64 and linux64 bit, can PyDSTool work without any hiccups?
I don't have any references, sorry. I don't recall who has told me about that and I don't think it has been recorded anywhere. No hiccups at all for 64bit? Maybe not, especially on Windows -- mostly because the scipy stack including cygwin/mingw and gcc/gfortran can be tricky to install in a way that allows users to link their own gcc-made DLLs (which is how Dopri/Radau work). Bundled binary environments like Anaconda are usually compiled with an MS C compiler that can't be linked to unless you own that same compiler. You might have to install the entire python stack from source, but then BLAS etc. become the stumbling points. I'm not sure what the current status of that situation is as I haven't done any Windows setups for more than a year. But I know of many folk who are using it on 64 bit linux/OS X.