[rhh006]

This pull requests suggests adding kWaveDiffusion to k-wave-python, ported from the MATLAB k-Wave toolbox. Please provide any feedback and suggestions.

Confidence Score: 4/5

Fix the heterogeneous dt_limit calculation before merging.

Findings

1.  Unsafe steps look safe ▶
2.  Test skips changed derivatives ▶

Summary

Adds kWaveDiffusion for periodic thermal simulations, with optional CuPy execution, heat sources, perfusion, sensor readings, and thermal dose.

• Temperature simulations now run from material and heat inputs on a grid.
• The solver tracks thermal dose and marks affected grid cells.
• The solver reports a time-step limit for its diffusion setup.

_{Reviews (3) · Last reviewed commit: "Implement uniformity checks for diffusio..." · Reviewed by Greptile}

[greptile-apps]

Test skips changed derivatives

test_nyquist_mode_decay never reaches the heterogeneous code changed to fix alternating temperature patterns. Its np.full(N, D) coefficient now passes _is_uniform, so the test uses _homogeneous_diffusion_update. Restoring the old, broken first derivatives would still pass this test.

Add an expected-value test that reaches _heterogeneous_diffusion_update and includes an alternating temperature component.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[greptile-apps]

Unsafe steps look safe

With use_kspace=False, the new dt_limit calculation multiplies spatial coefficients by Fourier frequencies at matching array positions. Those positions do not describe the same thing.

For example, with no perfusion and diffusion_coeff positive only at index zero, the product is zero everywhere because k_squared[0] is zero. The property returns infinity, but _heterogeneous_diffusion_update still applies diffusion and needs a finite step limit. Users relying on this value can choose steps that make temperature errors grow. Bound the actual update instead of multiplying these arrays elementwise.