search

atmtools / arts

The Atmospheric Radiative Transfer Simulator
https://www.radiativetransfer.org/
Other
62 stars 28 forks source link

Cpp - disort #760

Closed riclarsson closed 3 months ago

riclarsson commented 4 months ago

This is a reimplementation of the PythonicDISORT solver.

It still lacks clear citation of the original software. This must be done as part of the user-facing interface, which is not yet part of this draft.

Remaining work: 1) Make user-facing test. Method and all. 2) Run speed comparison versus cdisort in the 2.6.x branch. 3) Remove cdisort from ARTS3. 4) Remove the GSL Legendre Gauss Quadrature as a new open-source implementation that is not GPL is used. 5) Add user-facing citation for the new Legendre Gauss Quadrature scheme. 6) Get approval that the inclusion of Boost in ARTS3 is OK.

riclarsson commented 4 months ago

As a note for testing. In this iteration the following things limit what is tested:

1) b_pos and b_neg are always single-values 2) The Fourier error is not tested and will throw if used 3) f_arr is never empty in TMS-corrections. (Honestly, TMS correction should perhaps not be called then.) 4) There are no BDRFs when there's a source term.

The test follows what PythonDISORT tests as of April 2nd 2024 and should be updated before merge to be with the unreleased version.

riclarsson commented 3 months ago

This now contains enough improvements that it can be merged in my opinion.

I cannot continue with this in an effectively. It requires integration with the overall ARTS still, which is much more your expertise for DISORT-y stuff, @m-brath . Of course, I am around for helping. But I cannot effectively do the integration myself. It's a little bit of a chicken-and-egg problem with the scattering update branch since so much code might have to be switched later. As the test-file suggests, this at least produces the same results as the original algorithm and the same results as cdisort.

@olemke , I also removed Eigen with this update and made us use the distributed version instead (same version). This removes quite a few errors as we don't report on bad code warnings in external libraries.