Exojax

Target
The target of version 1.5 is the radiative transfer. This includes a more sophisticated radiative transfer for pure absorption (i.e. the intensity-based scheme, petitRadtrans like one), with scattering, reflection lights (flux-adding scheme based on [Robinson and Crisp (2018)](https://arxiv.org/abs/1803.02378), LART), more test and improvements of transmission (incl. bug fix), and so on.

Feature List
Radiative Transfer

**Important:** the default RT scheme has been the intensity-based scheme w/ 8 streams (ibased), in ArtEmisPure for instance, since version 1.5 (more accurate and faster than the flux based, fbased, see 475, 457 ). This may change the result slightly because of the difference of the RT scheme.

- [x] Intensity-based radiative transfer for pure absorption 411 412 419
- [x] Toon-type two-stream radiative transfer (with scattering/reflection), flux-adding (and LART) 428 477
- [x] Forward modeling of the reflection spectrum, with an example of Jupiter's reflection spectrum. 477
- [x] transmission improvements (Simpson) and more tests 421 424 463 464

Scattering
- [x] Rayleigh scattering 115 430 434
- [x] Cloud (Mie) scattering using Ackerman and Marley cloud model 477

Other Improvements
- [x] OLA-based convolution of the response (spin and IP) 417
- [x] telluric line fitting 450
- [x] OpaCont includes H-, Rayleigh, Mie, and CIA 434 480

Big Fixes
- bugfix for non-constant gravity 447
- enables to compute gradients for transmission 463 464
- enables OpaDirect for VALD 404
- optelower for HITEMP 454
- slight change (uses nansum) in the f-based transfer 475
- 32 bit issue for mmw 473

Minor Issues
- valid auto T-range became wider than the previous version 345 456 466
- allows 32bit for MultiMol 461

See [here](https://github.com/HajimeKawahara/exojax/milestone/8?closed=1) for more details.

Since version 1.3, ExoJAX contained a critical bug (414) related to the setting of the pressure layer. This release fixes that bug. From v1.3 to v1.4.1, `dParr` calculated with `pressure_layer_logspace`, and consequently the layer optical depth calculated using this, was about half of its intended value. Please update urgently.

This release includes the updates raised in 405 in addition to v1.4.

Target
Version 1.4 targets broadband analysis (e.g., at least the entire H-band) of clear-sky late T dwarf emission; it also adds the capability of a pilot version of transmission spectroscopy.

Update List

- [x] pilot version of the transmission spectra capability 356| [tutorial](https://secondearths.sakura.ne.jp/exojax/tutorials/Transmission_beta.html)
- [x] operators on spectra class 363 (pilot) | [see get started](https://secondearths.sakura.ne.jp/exojax/tutorials/get_started.html)
- [x] multiple molecules handler 388 | [tutorial](https://secondearths.sakura.ne.jp/exojax/tutorials/multimol.html)
- [x] `OpaDirect` for Kurucz 400
- [x] Non air broadening using the fitting formula by Tan et al. (2022) 378 | [doc](https://secondearths.sakura.ne.jp/exojax/userguide/nonair.html)
- [x] required device memory estimate 385 | [doc](https://secondearths.sakura.ne.jp/exojax/userguide/memorysetting.html)
- [x] single broadening parameter for PreMODIT 376 | [doc](https://secondearths.sakura.ne.jp/exojax/userguide/premodit.html#single-broadening-parameter-set)
- [x] uncertainty importing for HITEMP (for HITRAN, we postpone the release until the new release of RADIS) 397
- refactoring more, many small improvements , see [here](https://github.com/HajimeKawahara/exojax/milestone/7?closed=1).

removing deprecated features and renaming

- spec.rtransfer.wavenumber_grid -> use utils.grids.wavenumber_grid instead
- spec.rtransper.dtauM -> use spec.layeropacity.layer_optical_depth

changes
- ExoJAX now ~~strongly~~ recommends using 64-bit as default 385, unless you are confident that 32-bit is enough for computation.
python
from jax import config
config.update("jax_enable_x64", True)

- Users can choose the order of the wavelength grid, but the default is now descending order though v<=1.3 assumes the ascending order. 395

Version 1.3.1 fixes the bug related to HITEMP/H2O opacity computation.
- in set_grid.py the negative value was not allowed for the density grid. So, users could have removed the negative temperature exponent of the HITEMP/H2O. But, this modification could have removed some important H2O lines.

The following is the comparison, ykawashima made: The blue lines removed the lines with n_air < 0.01, yellow clipped to n_air = 0, green no restriction for HITEMP H2O.

Since v1.3, PreMODIT (device-memory friendly calculator) is one of the standard opacity calculators of ExoJAX.

New Features

- [x] more GPU memory saved method in PreMODIT (so called diffmode) 332
- [x] opacity calculator class `opa` 346 [see this tutorial](http://secondearths.sakura.ne.jp/exojax/develop/tutorials/get_started.html)
- [x] atmospheric RT class `art` 346
- [x] optimal Elower max (reducing device memory use) in PreMODIT 331 332 [see this tutorial](http://secondearths.sakura.ne.jp/exojax/develop/tutorials/elower_setting.html)
- [x] optional quantum states 336 338 [see this tutorial](http://secondearths.sakura.ne.jp/exojax/develop/tutorials/select_quantum_states.html)
- [x] faster IP and spin convolutions 350
- [x] molecular mass `mdb.molmass` available 328

schematics of mdb/opa/art classes
See http://secondearths.sakura.ne.jp/exojax/tutorials/get_started.html



Bug Fix

341 MODIT/PreMODIT/DIT
353 pressure reference point in `spec.rtransfer.pressure_layer`

Obsolescence and Replacement

- [x] moldb relationals, instead, fully replaced to `api` (for ExoMol and HiTEMP/HITRAN)
- [x] autospec, instead, new interfaces, `opa` and `art` available