Hi Lufter,
One limitation of the rate equation governing equation is that it needs to re-diagonalize the Hamiltonian when the direction of the magnetic field changes (since that defines the rate equation quantization axis). In a MOT simulation, this happens at every timestep in evolve motion, making the simulation very slow (even when random recoil is false).
My first suggestion is to remove max_step=1 from the call to evolve_motion and use max_scatter_probability (=0.1 is default) to control the step size instead. That might give you a marginal improvement in speed (particularly if you investigate higher values of max_scatter_probability). You could also reduce the solver precision using the rtol and atol arguments of the underlying call to scipy’s solve_ivp (though this may lead to nonsense results, I typically need to increase the precision from the pylcp defaults).
Another possibility is to switch to the OBE governing equation. The OBE always uses the z axis as the quantization axis, so there’s no repetitive Hamiltonian diagonalization. Because bosonic alkaline-earths only have 4 state Hamiltonians, the simulations might be faster than the rate equations. But, I have not tested this myself. (Please let me know how it goes if you try it).
Finally, my intuition is that the broad line dynamics may be faster than the narrow line dynamics. You might be able to run shorter simulations once you figure out when the MOT reaches steady state.
Let us know if you have any more questions,
Daniel
--
You received this message because you are subscribed to the Google Groups "pylcp" group.
To unsubscribe from this group and stop receiving emails from it, send an email to
pylcp+un...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/pylcp/2a000de5-5723-4725-8b43-bba8720c24e0n%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.