Hi,
I was wondering if the modified Pennes bioheat equation (kWaveDiffusion) could be optimized in theory with compiled binaries in the same fashion as kspaceFirstOrder3D? If so, would it be feasible to use the C++ code for kspaceFirstOrder3D (e.g. KSpaceFirstOrder3DSolver.cpp) as a starting point for a C++-based kWaveDiffusion, or is the problem much more complicated than that?
Thank you!
Hi Vich,
Yes in principle these codes are closely related (e.g., both use an explicit time stepping scheme and derivatives calculated using FFTs). We have it in the development pipeline to write a C++ code, but haven't started on it yet, so it might be some time away!
Brad
I realize it must be really time-consuming to maintain and add new features to a project as big as k-Wave has become... I appreciate your hard work and that of everyone who has contributed to k-Wave: theoretical work, programming, documentation, physical validation, all of that is extremely valuable to us. Thank you!
Hi,
I was wondering if nowadays there is an available C++_based KWaveDiffusion.
Congratulations for this great toolbox! Thank you!
Hi Davide,
It's in the development pipeline, but we won't have anything ready for release for a while I'm afraid. In general, the stability limit for the diffusion code is large (due to the k-space correction), so if you're trying to speed things up, you could conduct a convergence test to see how big you can make your time steps and still get accurate results.
Brad
Hi -- I have three questions about KWaveDiffusion.
1) The last post on this thread is 3+ years old. Has there been progress in making KWaveDiffusion faster since then, for example using a C++ implementation?
2) I have been using the following calculation to set the time step:
>> DD = medium.thermal_conductivity ./ (medium.density .* medium.specific_heat); % diffusivity map in m^2/s
>> diffTimeVoxel = 0.001^2 / ( 2 * max(DD(SIM>1)) ); % time it takes for heat to diffuse of 1 voxel (calculated using the fastest diffusivity value in the domain)
>> dt = diffTimeVoxel / CFL; % I seem to obtain decent results for CFL=10 and CFL=5
This works OK, I wonder what the developers think of this approach and what value they would recommend for CFL, based on theoretical considerations or experience.
3) There are numerical difficulties when using medium.density=1 for air voxels, which is the correct physical value. Setting the density of air voxels to 1000 improves convergence dramatically. It does not seem to affect the results too much because the thermal conductivity of air is small, therefore heat simply does not propagate into that compartment anyway. What do the developers think of this fix?
Thanks! Bastien
You must log in to post.