History

Version 2.2.2

What’s Changed

Full Changelog: https://github.com/HajimeKawahara/exojax/compare/v2.2.1…v2.2.2

Version 2.2.1

  • bug fix for when loading OpaPremodit with MdbHitemp #639

  • added save/load mode for OpaCKD #635 #636 #637

What’s Changed

Full Changelog: https://github.com/HajimeKawahara/exojax/compare/v2.2.0…v2.2.1

Version 2.2.0

The main target for version 2.2.0 is the capability of differentiable TCE (thermal chemical equilibrium). Note that we are developing the differentiable TCE package ExoGibbs.

Equilibrium Retrieval

Overview

Release 2.2.0 (branch release_v2_2_0) is a substantial update since tag v2.1, focusing on PreMODIT performance/memory, database API refactors, packaging modernization, and new snapshot/provider abstractions to decouple opacity from DB classes. Summary below is based on local Git history; I can cross-check the milestone if you paste it.

pyproject

We migrate ExoJAX from setup.py to a modern PEP 517/518 and PEP 621 compliant pyproject.toml build. (#626)

restructure

  • MdbExomol, MdbHitemp, MdbHitran, MdbHargreaves, AdbVald, AdbSepVald, AdbKurucz, XdbExomolHR, PdbCloud, CdbCIA moved to subdirectories under database (#622, #623, #624)

These classes can be called as:

from exojax.database import MdbExomol

Connect MDB to OpaPremodit

What’s new: OpaPremodit now accepts databases via high-level helpers and data-only snapshots. This decouples opacity from DB classes, improves memory use, and keeps APIs clean.

Snapshots (advanced/portable)

  • Export once, construct anywhere:

    snap = mdb.to_snapshot()
opa = OpaPremodit.from_snapshot(snap, nu_grid)
  • Optional providers override:

    • Partition: ExoMol/HITRAN providers are auto-selected; pass a custom pf_provider if needed.

    • Broadening: ExoMol/HITRAN strategies are auto-selected; pass a custom broadening_strategy to override.

Memory and Runtime Controls

  • memory_policy: central knob for allow_32bit, nstitch (ν-stitching), and cutwing. Policy values override constructor args.

  • broadening_resolution:

    • manual: fixed resolution (highest memory)

    • minmax: DB-driven min/max grids (medium memory)

    • single: single broadening set (lowest memory)

Notes

  • from_mdb calls .to_snapshot() internally and frees the DB by default (delete_mdb_after_init=True). Keep it with delete_mdb_after_init=False.

  • Ensure MDB lines intersect nu_grid; otherwise a ValueError is raised.

  • Direct constructor OpaPremodit(mdb, nu_grid, ...) remains supported; from_mdb is the preferred path for clarity and memory efficiency.

Highlights

  • PreMODIT: adds MemoryPolicy, provider-based partition/broadening, from_mdb / from_snapshot constructors, and precomputed line_strength_Tref for speed.

  • Snapshots: new MDBSnapshot / MDBMeta / Lines DTOs; MdbExomol, MdbHitran, MdbHitemp implement .to_snapshot().

  • Database refactor: split ExoMol vs HITRAN/HITEMP broadening paths; reorganized modules; added XdbExomolHR for high-res empirical lines.

  • Atomic DB (VALD/Kurucz): AdbVald, AdbSepVald, AdbKurucz exposed via exojax.database.

  • CIA: dedicated exojax.database.cia.api.CdbCIA with contdb kept as a deprecation shim.

  • Float32: optional 32-bit mode for PreMODIT (faster + lower memory).

  • Chemistry: groundwork to connect equilibrium chemistry (exogibbs), new tutorial/docs, and H2 CIA usage in examples.

  • Packaging: switch to pyproject.toml + setuptools-scm; Python ≥ 3.9; CI config updates.

  • Tests/Docs: snapshot tests for MDBs, additional opacity tests, relaxed tolerances where needed; “Get Started” refresh.

Breaking/Deprecations

  • exojax.database.moldb removed; use exojax.database.exomol / hitran / hitemp classes (MdbExomol, MdbHitran, MdbHitemp) or atomic Adb*.

  • Prefer exojax.database.cia.api.CdbCIA; contdb.CdbCIA emits DeprecationWarning.

  • Minimum Python now 3.9 (declared in pyproject.toml).

Included PRs (since v2.1)

  • #631 di_mdb_opa: DB–opacity integration refactor, memory/policies/providers, tests/docs.

  • #630 premodit_mem: PreMODIT memory improvements + snapshot path and tests.

  • #629 connect_exogibbs: equilibrium chemistry integration + tutorial.

  • #626 pyproject: packaging migration, Python version floor, CI.

  • #624 mmrefacmdb / #623 morerefactmdb / #622 refacmdb: DB refactors (molinfo/qstate moves, ExoMolHR restructure, HITRAN/HITEMP splits).

  • #620 32bitxs: float32 cross-section support + tests.

bug fixes

  • #619 #620 32-bit overflow

What’s Changed

Full Changelog: https://github.com/HajimeKawahara/exojax/compare/v2.1…v2.2.0

Version 2.1

The main target of v2.1 is major refactoring for the new development of ExoJAX3 (#606 #607 #608) and to implement the differentiable CKD (#484). This capability will be used to computes magnitude and/or low/mid res spectrum to calibrate high-res spectra.

New Features

  • Correlated-k distribution (CKD; beta-version), #484 #610 #614 - tutorial for emission - tutorial for transmission - what is CKD?

  • extra database for ExoMolHR #595 #600 , see this tutorial

  • FeH line list by Hargreaves et al. (2010), #603 by @YuiKasagi

  • checkpoint (memory release) for transmission reverse-mode diff #604 by @sh-tada

See #597 as direction toward ExoJAX3

Refactoring

Separates database, opacity, rt, postproc from spec, and a script for fixing #606 #607 #611

The main change in this update is the redistribution of modules previously located in exojax.spec into four function-specific subpackages: exojax.database (db), exojax.opacity (opa), exojax.rt (art), and exojax.postproc. To assist with this transition, we have included a script that automatically updates your import statements from exojax.spec to their appropriate new locations. The script is located at exojax/scripts/fix_v2_1.py. You will need to install libcst to run it.

… code-block:: sh

pip install libcst

Please make sure to back up your original files before using the script. Usage instructions are as follows:

… code-block:: sh

python -m fix_v2_1 your_project/

  • OpaPremodit, OpaDirect, OpaModit can be called such as

… code-block:: python

from exojax.opacity import OpaPremodit

Bug Fixes

  • Reflect updates to astropy (unit “Angstrom”) by @YuiKasagi #615

Removed methods

  • removed dtau_mmwl

  • merged nu2wav and wav2nu from unitconvert to utils.grids

Version 2.0

In this update, we include further refactoring, additional documentation, and new features for device-memory saving algorithms, opart and The nu-stitching , which were not described in the initial submission ( version 1 on the ArXiv manuscript ).

Opart: layer-by-layer opacity and radiative transfer

The opart method has recently demonstrated at least comparable computation times to conventional methods for cases like pure absorption with emission (EmisPure). By calculating without separating opa and art, it eliminates dependency on nlayer for memory requirements. This means that calculations can likely be performed even on GPUs with limited device memory, such as those without the large 80GB memory capacity of an A100 GPU. We consider this a valuable addition for version 2. See here as a tutorial.

Available Opart classes

  • OpartEmisPure

  • OpartReflectPure

  • OpartEmisScat

  • OpartReflectEmis

#542 #546 #547 #558 #565 #559 #556 #557 #568

Wavenumber Stitching ($nu$-stitching)

We implemented the nu-stitching , the method to divide the wavenumber grid, using OLA , in PreMODIT as an option (with nstitch and cutwing arguments).

from exojax.opacity import OpaPremodit
opa = OpaPremodit(mdb, nus, nstitch=10, auto_trange=[500,1300], cutwing = 0.015)

See here for the details.

Note that $nu$-stitching can be used even in opart.

#566 #581

Improvements

  • (OpaPremodit, OpaModit) Reduction in device memory usage by performing zero-padding within the scan #577 #578

  • (MdbExomol) Exomol broadener_species option, the capability of air and/or other species #580

Documentation Update

  • visualization of LBD #548 #553

  • GP modeling in getting started #554

  • How to use SVI with ExoJAX #572

  • How to use Nested Sampling with ExoJAX #107 #570

  • others #560 #561

Removed

  • spec.modit.calc_xsection_from_lsd #576

  • spec.modit.xsvector #576

  • spec.modit.xsmatrix #576

Version 1.6

The target for version 1.6 is to introduce features that facilitate the application to more practical data. Also, v 1.6 will be the fixed version of ExoJAX2 prior to the paper II submission. It is planned to be released as a pre-release of ExoJAX (ver 2).

  • practical use of a cloud model. See documents/tutorials/jupiters

  • comprehensive review

  • eq methods (and neq ) in opa and mdb to enable to use == (and !=) operators. #509

  • compatible to radis 0.15.2, in particular, “vaex as an option” strategy #500 #506

  • as a result, ExoJAX supports python 3.9 3.10, 3.11, 3.12,

  • new art, ArtAbsPure, computes a pure absorption spectrum w/ and w/o a single reflection at Psurf.

Comparison

  • Comparison with a tabulated cross section #491

  • with pRT #427

Bug Fixes

  • side effect on a multiple call of opa #510

  • Stark broadening #489

  • solar abundance definition (utils.zsol.nsol) #535

Changes

removed

  • recexomol #501

  • lpf.auto_crosssection #498

  • utils.recexomol #501

  • Sij0 attribute from api.py (MdbExomol, MdbHitemp, MdbHitran) #515

  • atm.mixratio.vmr2mmr, atm.mixratio.mmr2vmr (use atm.atmconvert.vmr_to_mmr, atm.atmconvert.mmr_to_vmr instead)

  • spec.dit.dtauM_vald_old

all methods in dynamics package - getE.getE, rvfunc.rvcoref, rvfunc.rvf,``rvfunc.rvf2``, rvfunc.rv_semi_amplitude, rvfunc.rvf1,

several methods in spec.rtransfer (deprecated since v1.5) - dtauM,``dtauCIA``, dtauHminus, dtauVALD, pressure_layer

renamed

  • line_strength_ref_original was introduced for line strength at T=Tref_original, instead of line_strength_ref #515

  • self.line_strength_ref (which depends on self.Tref) → self.line_strength(Tref) #515

  • instances_from_dataframe → attributes_from_dataframe

changes arguments

  • qr_interp(T) → qr_interp(T,Tref) in MdbExomol #515

  • qr_interp(iso,T) → qr_interp(iso,T,Tref) in MdbHitemp/Hitran #515

  • qr_interp_line(T) → qr_interp_line(T,Tref) in MdbHitemp/Hitran #515

new methods

  • computes X, Y, Z, utils.zsol.mass_fraction_XYZ(number_ratio_elements) #535

Version 1.5

  • Intensity-based radiative transfer for pure absorption #411 #412 #419

  • Toon-type two-stream radiative transfer (with scattering/reflection), flux-adding (and LART) #428 #477

  • Forward modeling of the reflection spectrum, with an example of Jupiter’s reflection spectrum. #477

  • Transmission code improvement (Simpson) and more tests #421 #424 #463 #464

  • Rayleigh scattering #115 #430 #434

  • Cloud (Mie) scattering using Ackerman and Marley cloud model #477

Version 1.4

  • transmission spectra #356

  • operators on spectra class #363

  • multiple molecules handler #388

ExoJAX now recommends using 64-bit as default, but can use 32 bit if you are confident (and for real science use).

Version 1.3

  • more GPU memory saved method in PreMODIT (so called diffmode) #332

  • opacity calculator class opa #346 see this tutorial

  • atmospheric RT class art #346

  • optimal Elower max (reducing device memory use) in PreMODIT #331 #332 see this tutorial

  • optional quantum states #336 #338 see this tutorial

  • faster IP and spin convolutions #350

  • molecular mass mdb.molmass available #328

Version 1.2

  • Moved on a common I/O of molecular/atomic database with radis.api for ExoMol/HITRAN/HITEMP #272

  • Removed old moldb for ExoMol/HITRAN/HITEMP

  • PreMODIT algorithm, applicable to a wide wavenumber range #265 #287 #288 #307

  • Memory saved version of spin rotation and instrumental response #295

Version 1.1

  • VALD3

  • reverse mode available

  • compatibility with JAXopt #212

Version 1.0

  • Auto-differentiable Spectrum Model of exoplanets/brown dwarfs built on JAX using the molecular/atomic database, ExoMol, HITRAN/HITEMP and VALD3.

  • Bayesian inference using HMC-NUTS/NumPyro

  • Two opacity calculators available: Direct computation of the Voigt line profile (lpf) and the modified discrete integral transform (MODIT). The latter is a fast opacity calculator for the number of lines N >~ 1000.

  • Transparent open-source code with documentations, including a peer-reviewed paper, API, user guide, and tutorials using real data

  • HITRAN/CIA and H- as continuous opacity

  • Cloud modeling based on Ackerman and Marley

  • Quick computation of the opacity and emission spectra for observers (autospec)

Before Version 1