Pith. sign in

REVIEW 3 major objections 7 minor 242 references

This paper shows that a fully differentiable pipeline from hydrodynamics through chemistry to radiative transfer can recover physical and chemical parameters of a shock directly from its molecular line spectrum using gradient-based optimiza

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 00:33 UTC pith:MURVS4LT

load-bearing objection A legitimate proof-of-concept for end-to-end differentiable astrochemistry, with a real inverse-crime caveat in the retrieval demo. the 3 major comments →

arxiv 2607.28727 v1 pith:MURVS4LT submitted 2026-07-30 astro-ph.IM

End-to-end differentiable retrieval of molecular spectra using hydrodynamics, chemistry, and radiative transfer

T. Grassi , G. Vermari\"en , J.E. Pineda , S. Spezzano , S. Bovino , P. Caselli This is my paper
classification astro-ph.IM
keywords differentiable programmingautomatic differentiationradiative transferastrochemistrymolecular line spectrashock modelsinverse problemsgradient-based optimization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors set out to prove that the full chain of physical processes connecting a shock to the molecular line radiation we observe—gas dynamics, time-dependent chemistry, and photon emission and absorption—can be treated as one differentiable function of the underlying parameters. They build such a pipeline and show that by minimizing the difference between a computed HCO+ spectrum and a target spectrum, gradient descent retrieves the upstream density, velocity, and temperature of the shock, plus a chemical integration time, two parameters of the dissociative recombination rate of HCO+, and the microturbulence. The key result is that those parameters are recoverable from the spectrum alone, with no surrogate model or hand-tuning. Because every module is differentiable, the pipeline also yields the sensitivity of each spectral channel to every parameter. If extended to more realistic physics, this opens a route to inferring both physical conditions and chemical reaction rates directly from observations.

Core claim

On the paper's own terms, the central claim is that the inverse problem of molecular-line spectrum formation—from shock hydrodynamics to level populations to emergent line profiles—can be inverted by gradient-based optimization because every step is differentiable end to end. In the demonstration, the pipeline's forward model generates the target HCO+ (1-0) spectrum of a 1D shock with a shock-capturing Riemann solver, post-processed 25-reaction chemistry, and a radiative transfer module that solves for level populations under statistical equilibrium. Seven unknown parameters are optimized simultaneously. The paper reports that six parameters—upstream number density, velocity, temperature, mi

What carries the argument

The load-bearing object is a sequence of three differentiable operators written with automatic differentiation: a 1D conservative hydrodynamic solver using a Riemann solver and a Courant-limited adaptive timestep; a modified differentiable chemical integrator for stiff ordinary differential equations that supports backpropagation via a recursive checkpointing adjoint; and a radiative transfer code that computes level populations by solving a linear system at statistical equilibrium, then integrates emission with line-of-sight absorption, Doppler shifts, and microturbulence, all vectorized over spatial cells and frequency channels. The crucial technical enabler is checkpoint-and-recompute adj

Load-bearing premise

The load-bearing premise is that the forward model—1D ideal hydrodynamics, post-processed 25-reaction chemistry, and collision-only level populations—represents real shock emission closely enough that parameters retrieved from synthetic spectra will also be correct for real observations; if that physics is too simplified, the successful retrieval is an artifact of the model.

What would settle it

Run the same gradient-descent retrieval on an observed HCO+ (1-0) spectrum of a shock with independently known physical conditions (e.g., from multi-line observations). If the retrieved density, velocity, and temperature disagree with the independent values beyond the few-tens-of-percent level seen in the synthetic test, the pipeline's claim to invert spectra is refuted. Alternatively, within the synthetic framework, add a magnetic field or couple the chemistry to the energy equation; if the recovery of injected parameters then fails, the current success depends on the omitted physics.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the retrieval works on real spectra, observers could fit full line profiles to recover shock velocity, density, and temperature without iterative manual adjustment.
  • Chemical reaction rate coefficients, at least those that strongly influence the tracer abundance, can be constrained directly from emission data, potentially revising astrochemical networks.
  • The same gradient machinery provides per-channel sensitivity maps, identifying which physical parameters control which velocity ranges of a line.
  • Because the chemistry is time-dependent, the method applies to non-equilibrium chemistry, unlike steady-state analyses.
  • The approach is a working template for building differentiable multi-physics models for other diagnostics, limited only by the ability to differentiate the forward physics.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A natural next test is to apply the pipeline to real observations of a well-characterized shock and compare retrieved parameters with independent estimates; the paper's synthetic-only demonstration leaves open how model error propagates.
  • The biggest barrier to practical use is likely computational: differentiating stiff chemical networks with hundreds of species, or coupling chemistry to the energy equation, will require solver development beyond what is demonstrated.
  • One could extend the method to simultaneously fit multiple lines and species, which would lift some of the degeneracies the authors note (for instance, the poorly-constrained chemical integration time).
  • The failure of the retrieval to constrain the chemical integration time suggests a general caution: gradient fits recover only parameters the spectrum is truly sensitive to; sensitivity analysis should accompany every retrieval.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The manuscript presents a JAX-based pipeline that couples a 1D HLL hydrodynamics code, a differentiable chemistry code (a modified Carbox), and a line radiative transfer module, with checkpointed reverse-mode automatic differentiation through adaptive time stepping. The authors benchmark the hydrodynamics with Sod's shock tube, the chemistry by fitting 63 rate parameters to KIDA-based synthetic abundance trajectories, and the radiative transfer against PDR benchmark/Cloudy level populations and LOC spectra. They then perform an end-to-end optimization of seven shock/chemical/line parameters against a synthetic HCO+ J=1-0 spectrum and present a local sensitivity analysis. The central claim is that this is the first end-to-end differentiable pipeline of hydrodynamics, chemistry, and radiative transfer, enabling gradient-based retrieval of physical and chemical parameters from molecular spectra.

Significance. If the framework is validated beyond synthetic self-consistency, it would provide a novel and potentially powerful tool for molecular-line inference, avoiding surrogate models and finite-difference gradients. The modular benchmarks (Sod, PDR/Cloudy, LOC) and the use of KIDA rate values as ground truth are concrete strengths, as is the vectorized implementation and the use of checkpointing for adaptive ODE/hydro time-stepping. The paper also offers a useful local sensitivity analysis. However, the current end-to-end demonstration is an inverse-crime test against the same forward model, so the physical retrieval claim is not yet established.

major comments (3)
  1. [Sect. 3; Fig. 2; Table 1] The end-to-end retrieval optimizes against a target spectrum generated by the same discretized forward model (same 128-cell HLL hydrodynamics, same 25-reaction network, same optically thin collision-only RT) without noise. Success therefore shows that the optimizer can invert the code, not that the recovered parameters are the physical conditions of a real shock. Because the abstract concludes the framework 'can recover parameters ... directly from molecular line spectra', please add an out-of-sample test—e.g., target spectra produced by an independent code (Cloudy/LOC or a different shock model), or at least noise-added and resolution-degraded spectra—or explicitly reframe the demonstration as a numerical proof-of-concept. This is a load-bearing distinction for the central claim.
  2. [Table 1; Sect. 3] The statement that 'the expected parameter values are all reproduced' is contradicted by t_end: target = 1e5 yr, optimized = 5.5e3 yr, a factor of ~18 below target. The text's explanation (the spectrum is insensitive) means t_end is not identifiable from this dataset. In addition, the paper states that not all randomized initializations converge and that the learning rate must be tuned. Please quantify the convergence rate over seeds, report how the displayed solution was selected, and treat t_end as a non-recovered parameter in the abstract and conclusions.
  3. [Sect. 2.3, 2.3.1, 4] The radiative transfer used in the retrieval omits radiation-induced transitions in the level populations, i.e., it assumes optically thin/collision-only excitation. Since HCO+ 1-0 can be optically thick in the density/temperature regime of the shock example, the demonstration does not exercise the regime claimed by the abstract ('reproduce observed molecular line emission'). Either repeat the retrieval with the full RT version (which the authors say is already implemented and tested) or explicitly limit the claim to optically thin, collisionally excited lines.
minor comments (7)
  1. [Sect. 6] The source code is not public. For a methods paper whose contribution is a pipeline, please provide a repository or a clear release plan; this is important for reproducibility.
  2. [References] The two Kawahara et al. 2025 entries (ApJ 985, 263) appear to be duplicates; please check.
  3. [Sect. 2.3, Eq. (10)] The sentence 'units of 10^23 Jy' is confusing; 1 erg s^-1 cm^-2 Hz^-1 equals 10^23 Jy, so the spectrum is in cgs units. Please rephrase.
  4. [Eq. (7)] The term '32 turb(z)' appears to be a rendering error for the squared microturbulence velocity; please correct.
  5. [Sect. 2.3] The text mentions 'T(v)' where the spatial coordinate is intended; should read 'T(z)'.
  6. [Appendix E, Table E.1] Reactions 15 and 16 both list C+ + OH → CO+ + H; reaction 16 likely should have a different product. Please verify.
  7. [Fig. D.1] The top panel y-axis would benefit from an explicit label '(log)' or a note that the loss is shown on a logarithmic scale.

Circularity Check

2 steps flagged

Retrieval demonstrations are self-consistency fits to synthetic targets from the same differentiable forward model (inverse-crime setup), so the headline 'recovery' is partially circular; component benchmarks provide some independent grounding.

specific steps
  1. fitted input called prediction [Sect. 3, Eqs. (5)-(10), Table 1, Appendix D]
    "As an example, we model a 1D shock in a box of size L=10^4 au with 128 grid elements ... The output quantity n_HCO+(z) is provided to the radiative transfer code, alongside T(z) and v(z) from the hydrodynamical code. ... We want to minimize the L2-norm of the difference of the target and optimized I(v), normalized to the maximum of the target. ... We do not include any mock observational noise. ... In Table 1, the expected parameter values are all reproduced (n_L, T_L, v_L, v_turb, alpha_18 and beta_18)."

    The target spectrum is generated by the same discretized operators that are being inverted: the 128-cell HLL hydrodynamics, the 25-reaction Carbox network, and the collision-only radiative transfer of Eqs. (5)-(10). With no noise, the loss compares I(theta) with I(theta_target), both obtained from identical R(C(H(...))) code, so the optimizer can in principle recover exactly the generating parameter set. Calling these 'expected parameter values' and saying they are 'reproduced' presents a self-consistency fit as a retrieval; it does not establish physical identifiability or forward-model adequacy. It is partial circularity, not a tautology, because the optimization is nontrivial and the modules have external benchmarks.

  2. fitted input called prediction [Sect. 2.2 and Appendix B]
    "We tested the capability of our code of leveraging automatic differentation by predicting the rate coefficient of a chemical network using a given set of time-dependent abundances n(t;T,zeta) at different temperatures and cosmic ray ionization rates. ... The optimization target N_t x N_T x N_zeta matrix is constructed by running Carbox on a grid of T and zeta values for a specific set of alpha, beta, and gamma (i.e., the values found in the KIDA database, Wakelam et al. 2024)."

    The target abundance matrix used as ground truth is produced by running the same Carbox solver whose rate parameters are then optimized. Recovering the injected KIDA alpha, beta, gamma therefore shows that gradient descent can invert Carbox's own ODE (Eq. 4), not that the rate coefficients are constrained by independent data. The paper labels this 'predicting the rate coefficient', but it is a fit to synthetic output generated by the same differentiable solver. The use of external KIDA values and the explicit framing as a numerical capability test keep this from being fully circular.

full rationale

The paper's strongest numerical claims reduce to self-consistency tests against synthetic data from the same forward model. In Sect. 3, both the target and optimized HCO+ spectra are produced by the same 1D HLL hydrodynamics + Carbox chemistry + collision-only radiative-transfer chain, with no added noise; matching them therefore demonstrates that the optimizer can invert the code, not that the recovered n_L, v_L, T_L, v_turb, alpha_18, beta_18 are uniquely determined physical shock parameters. The paper itself notes some initial conditions do not converge and t_end is recovered far below target, so the loss landscape is not tightly constrained. Similarly, Appendix B constructs the rate-coefficient target matrix by running Carbox with KIDA rates and then optimizes those same rates, which validates gradients and optimization, not the chemistry. The paper is honest that these are 'controlled synthetic data' and provides independent module-level benchmarks (Sod shock, Cloudy PDR level populations, LOC spectra), and the KIDA values are external. There is no load-bearing self-citation chain: Carbox is public and the modifications are described. On balance, the central 'recovery' demonstrations are partially circular because the predictions are generated by and fitted to the same model, but the framework itself has independent content and external checks, so the score is 6 rather than higher.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 0 invented entities

The pipeline itself introduces no fitted constants: the parameters optimized in the demonstrations (shock left-state, t_end, alpha18, beta18, v_turb, 63 rate coefficients) are the targets of recovery with known ground truths, not ad hoc numbers hiding derivation gaps. The fixed physical inputs (gamma, mu, zeta, A_V, initial abundances) are domain assumptions listed below.

axioms (5)
  • domain assumption 1D single-velocity ideal-gas hydrodynamics with fixed gamma=1.4 and mean molecular weight mu=2.33 m_p
    Sect. 3: shock initial conditions and equation of state; ignores magnetic fields, obliquity, and varying mu.
  • domain assumption Chemistry is post-processed after the complete hydrodynamic integration
    Sect. 3: 'chemistry is calculated after the complete hydrodynamic integration rather than during each hydrodynamical step'; no chemical feedback on temperature.
  • domain assumption Level populations are computed without radiation-induced transitions in the main benchmark
    Sect. 2.3.1 and 4: optically thick radiation coupling implemented but not used for the demonstrated retrieval.
  • domain assumption The 25-reaction gas-phase network and fixed initial C/O abundances capture the needed shock chemistry
    Sect. 2.2 and Appendix E: network is 'relatively simple' and 'only for illustrative purposes'; C and O initialized at 1e-4 and 3e-4 of n(z).
  • domain assumption Synthetic target spectra are noiseless and generated by the same forward pipeline
    Sect. 3: 'We do not include any mock observational noise'; the retrieval is an inverse-crime test.

pith-pipeline@v1.3.0-alltime-deepseek · 17545 in / 13027 out tokens · 132333 ms · 2026-08-03T00:33:02.634983+00:00 · methodology

0 comments
read the original abstract

We aim to reproduce observed molecular line emission using a pipeline that couples hydrodynamics, chemistry, and radiative transfer, capable of simultaneously optimizing all relevant physical and chemical parameters. We developed an end-to-end differentiable Jax pipeline consisting of a custom hydrodynamical code, a modified version of the differentiable chemical code Carbox, and a custom radiative transfer code. We tested the framework using controlled synthetic data. We demonstrate that the framework can recover the parameters of hydrodynamical shock models directly from molecular line spectra and optimize selected chemical reaction rate coefficients through gradient-based optimization. The differentiable formulation enables efficient optimization of the coupled physical and chemical model while preserving the full time-dependent evolution.

Figures

Figures reproduced from arXiv: 2607.28727 by G. Vermari\"en, J.E. Pineda, P. Caselli, S. Bovino, S. Spezzano, T. Grassi.

Figure 1
Figure 1. Figure 1: Sketch of the pipeline. The first step is to run a hydrodynamical code that evolves the density ρ(z), velocity v(z), and temperature T(z). Then the chemical evolution produces the chemical species abundances n(z). Finally, the radiative transfer produces the emission lines I(ν), or equivalently I(3). Our code allows us to backpropagate the loss gradients with respect to the target spectra. pass requires ac… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of the target quantities (solid blue), evaluated at epoch 0 (dotted green), and at epoch 270 (solid orange). We show the HCO+ (1 → 0) emission normalized by the maximum of the target emission profile (top left), the density profile (top center), the velocity profile (top right, left black y-scale), the n(HCO+ ) · n(H2) product using the densities of the target (top right, right red y-scale), the… view at source ↗
Figure 3
Figure 3. Figure 3: Top panel: the HCO+ (1 → 0) emission normalized by the max￾imum of the target emission profile (same as top left panel of [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

242 extracted references · 44 canonical work pages · 10 internal anchors

  1. [1]

    1997 , month = aug, volume =

    Modeling primordial gas in numerical cosmology , author =. 1997 , month = aug, volume =. doi:10.1016/s1384-1076(97)00010-9 , keywords =

    doi:10.1016/s1384-1076(97)00010-9 1997

  2. [2]

    , keywords =

    A photon dominated region code comparison study. , keywords =. doi:10.1051/0004-6361:20065918 , archivePrefix =. astro-ph/0702231 , primaryClass =

    Pith/arXiv arXiv

  3. [3]

    and Leer, Bram van , title =

    Harten, Amiram and Lax, Peter D. and Leer, Bram van , title =. SIAM Review , volume =. 1983 , doi =

    1983

  4. [4]

    arXiv e-prints , keywords =

    On Neural Differential Equations. arXiv e-prints , keywords =. doi:10.48550/arXiv.2202.02435 , archivePrefix =. 2202.02435 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2202.02435

  5. [5]

    Journal of Computational Physics , keywords =

    Kinematic dynamos using constrained transport with high order Godunov schemes and adaptive mesh refinement. Journal of Computational Physics , keywords =. doi:10.1016/j.jcp.2006.01.042 , archivePrefix =. astro-ph/0601715 , primaryClass =

    Pith/arXiv arXiv 2006

  6. [6]

    BIT Numerical Mathematics , volume=

    Singly diagonally implicit Runge--Kutta methods with an explicit first stage , author=. BIT Numerical Mathematics , volume=. 2004 , publisher=

    2004

  7. [7]

    and Brown, Peter N

    Hindmarsh, Alan C. and Brown, Peter N. and Grant, Keith E. and Lee, Steven L. and Serban, Radu and Shumaker, Dan E. and Woodward, Carol S. , title =. 2005 , issue_date =. doi:10.1145/1089014.1089020 , journal =

    arXiv 2005

  8. [8]

    A Survey of Several Finite Difference Methods for Systems of Nonlinear Hyperbolic Conservation Laws

    Review. A Survey of Several Finite Difference Methods for Systems of Nonlinear Hyperbolic Conservation Laws. Journal of Computational Physics , keywords =. doi:10.1016/0021-9991(78)90023-2 , adsurl =

    doi:10.1016/0021-9991(78)90023-2

  9. [9]

    A new high resolution code called RAMSES

    Cosmological hydrodynamics with adaptive mesh refinement. A new high resolution code called RAMSES. , keywords =. doi:10.1051/0004-6361:20011817 , archivePrefix =. astro-ph/0111367 , primaryClass =

    Pith/arXiv arXiv

  10. [10]

    , keywords =

    Incorporating Ambipolar and Ohmic Diffusion in the AMR MHD Code RAMSES. , keywords =. doi:10.1088/0067-0049/201/2/24 , archivePrefix =. 1206.2476 , primaryClass =

    Pith/arXiv arXiv

  11. [11]

    , keywords =

    A high order Godunov scheme with constrained transport and adaptive mesh refinement for astrophysical magnetohydrodynamics. , keywords =. doi:10.1051/0004-6361:20065371 , archivePrefix =. astro-ph/0607230 , primaryClass =

    Pith/arXiv arXiv

  12. [12]

    , keywords =

    The challenges of modelling microphysics: ambipolar diffusion, chemistry, and cosmic rays in MHD shocks. , keywords =. doi:10.1093/mnras/sty3519 , archivePrefix =. 1901.00504 , primaryClass =

    Pith/arXiv arXiv 1901

  13. [13]

    , keywords =

    Numerical methods for non-LTE line radiative transfer: Performance and convergence characteristics. , keywords =. doi:10.1051/0004-6361:20021226 , archivePrefix =. astro-ph/0208503 , primaryClass =

    Pith/arXiv arXiv

  14. [14]

    An accelerated lambda iteration method for multilevel radiative transfer. I. Non-overlapping lines with background continuum. , keywords =

  15. [15]

    An accelerated lambda iteration method for multilevel radiative transfer. II. Overlapping transitions with full continuum. , keywords =

  16. [16]

    1973 , month = may, volume =

    Radiative and Dielectronic Recombination Coefficients for Complex Ions , author =. 1973 , month = may, volume =

    1973

  17. [17]

    2026 , journal =

    Probing the ion-neutral drift velocity towards the L1544 prestellar core: Detection of ambipolar diffusion using N2d+ and para-NH2D , author =. 2026 , journal =

    2026

  18. [18]

    , keywords =

    A fast neural emulator for interstellar chemistry. , keywords =. doi:10.1093/mnras/stae1432 , archivePrefix =. 2406.02387 , primaryClass =

    Pith/arXiv arXiv

  19. [19]

    1994 , month = jun, volume =

    The Photoelectric Heating Mechanism for Very Small Graphitic Grains and Polycyclic Aromatic Hydrocarbons , author =. 1994 , month = jun, volume =. doi:10.1086/174188 , adsurl =

    doi:10.1086/174188 1994

  20. [20]

    1987 , booktitle =

    Modular learning in neural networks , author =. 1987 , booktitle =

    1987

  21. [21]

    2022 , month = nov, volume =

    Tracing Interstellar Heating: An ALCHEMI Measurement of the HCN Isomers in NGC 253 , author =. 2022 , month = nov, volume =. doi:10.3847/1538-4357/ac91ce , keywords =

    doi:10.3847/1538-4357/ac91ce 2022

  22. [22]

    , keywords =

    Neural Network Constraints on the Cosmic-Ray Ionization Rate and Other Physical Conditions in NGC 253 with ALCHEMI Measurements of HCN and HNC. , keywords =. doi:10.3847/1538-4357/ad85db , archivePrefix =. 2409.13821 , primaryClass =

    Pith/arXiv arXiv

  23. [23]

    2005 , month = jan, booktitle =

    Chemistry in pre-stellar cores , author =. 2005 , month = jan, booktitle =

    2005

  24. [24]

    doi:10.1051/0004-6361/201321276 , eid =

    2013 , month = jul, volume =. doi:10.1051/0004-6361/201321276 , eid =

    work page doi:10.1051/0004-6361/201321276 2013

  25. [25]

    2019 , month = dec, volume =

    The 3D Structure of CO Depletion in High-mass Prestellar Regions , author =. 2019 , month = dec, volume =. doi:10.3847/1538-4357/ab53e4 , eid =

    work page doi:10.3847/1538-4357/ab53e4 2019

  26. [26]

    2020 , month = jun, volume =

    A new proxy to estimate the cosmic ray ionization rate in dense cores , author =. 2020 , month = jun, volume =. doi:10.1093/mnrasl/slaa048 , archiveprefix =. 2003.05416 , primaryclass =

    Pith/arXiv arXiv 2020

  27. [27]

    2021 , month = oct, volume =

    Chemical analysis of prestellar cores in Ophiuchus yields short timescales and rapid collapse , author =. 2021 , month = oct, volume =. doi:10.1051/0004-6361/202141252 , keywords =

    doi:10.1051/0004-6361/202141252 2021

  28. [28]

    2024 , month = apr, volume =

    Emulating the interstellar medium chemistry with neural operators , author =. 2024 , month = apr, volume =. doi:10.1051/0004-6361/202449193 , eid =

    doi:10.1051/0004-6361/202449193 2024

  29. [29]

    arXiv e-prints , keywords =

    Ray-trax: Fast, Time-Dependent, and Differentiable Ray Tracing for On-the-fly Radiative Transfer in Turbulent Astrophysical Flows. arXiv e-prints , keywords =. doi:10.48550/arXiv.2511.09389 , archivePrefix =. 2511.09389 , primaryClass =

    doi:10.48550/arxiv.2511.09389

  30. [30]

    , keywords =

    LIME - a flexible, non-LTE line excitation and radiation transfer method for millimeter and far-infrared wavelengths. , keywords =. doi:10.1051/0004-6361/201015333 , archivePrefix =. 1008.1492 , primaryClass =

    Pith/arXiv arXiv

  31. [31]

    LAsMA observations of H-MM1 and Oph D

    Nuclear spin ratios of deuterated ammonia in prestellar cores. LAsMA observations of H-MM1 and Oph D. , keywords =. doi:10.1051/0004-6361/202346578 , archivePrefix =. 2311.08006 , primaryClass =

    Pith/arXiv arXiv

  32. [32]

    , keywords =

    On the Timescale for the Formation of Protostellar Cores in Magnetic Interstellar Clouds. , keywords =. doi:10.1086/318348 , archivePrefix =. astro-ph/0009262 , primaryClass =

    Pith/arXiv arXiv

  33. [33]

    , keywords =

    The Green Bank Ammonia Survey: Data Release 2. , keywords =. doi:10.3847/1538-4365/ae11b1 , archivePrefix =. 2510.10607 , primaryClass =

    arXiv

  34. [34]

    , keywords =

    Chemical segregation analysed with unsupervised clustering. , keywords =. doi:10.1051/0004-6361/202553843 , archivePrefix =. 2507.22482 , primaryClass =

    Pith/arXiv arXiv

  35. [35]

    , keywords =

    Deuterated methanol in the pre-stellar core L1544. , keywords =. doi:10.1051/0004-6361/201423858 , archivePrefix =. 1408.2491 , primaryClass =

    Pith/arXiv arXiv

  36. [36]

    , keywords =

    The Ionization Fraction in Dense Cloud Cores. , keywords =. doi:10.1086/305624 , adsurl =

    doi:10.1086/305624

  37. [37]

    Molecular Ions in L1544. I. Kinematics. , keywords =. doi:10.1086/324301 , archivePrefix =. astro-ph/0109021 , primaryClass =

    Pith/arXiv arXiv

  38. [38]

    doi:10.1051/0004-6361:20030526 , keywords =

    2003 , month = may, volume =. doi:10.1051/0004-6361:20030526 , keywords =

    doi:10.1051/0004-6361:20030526 2003

  39. [39]

    2011 , month = dec, booktitle =

    Observational Studies of Pre-Stellar Cores and Infrared Dark Clouds , author =. 2011 , month = dec, booktitle =. doi:10.1017/s1743921311024835 , editor =

    work page doi:10.1017/s1743921311024835 2011

  40. [40]

    1992 , month = feb, volume =

    A Hydrodynamic Approach to Cosmology: Methodology , author =. 1992 , month = feb, volume =. doi:10.1086/191630 , adsurl =

    doi:10.1086/191630 1992

  41. [41]

    2019 , journal =

    Data-driven discovery of coordinates and governing equations , author =. 2019 , journal =. doi:10.1073/pnas.1906995116 , url =

    doi:10.1073/pnas.1906995116 2019

  42. [42]

    2013 , month = jul, volume =

    Insights into the content and spatial distribution of dust from the integrated spectral properties of galaxies , author =. 2013 , month = jul, volume =. doi:10.1093/mnras/stt523 , keywords =

    doi:10.1093/mnras/stt523 2013

  43. [43]

    2016 , month = oct, volume =

    Modelling and interpreting spectral energy distributions of galaxies with BEAGLE , author =. 2016 , month = oct, volume =. doi:10.1093/mnras/stw1756 , keywords =

    doi:10.1093/mnras/stw1756 2016

  44. [44]

    2020 , month = dec, journal =

    The frontier of simulation-based inference , author =. 2020 , month = dec, journal =. doi:10.1073/pnas.1912789117 , keywords =

    doi:10.1073/pnas.1912789117 2020

  45. [45]

    , keywords =

    Observing the gas temperature drop in the high-density nucleus of L 1544. , keywords =. doi:10.1051/0004-6361:20077613 , archivePrefix =. 0705.0471 , primaryClass =

    Pith/arXiv arXiv

  46. [46]

    , keywords =

    The NH _ 2 D hyperfine structure revealed by astrophysical observations. , keywords =. doi:10.1051/0004-6361/201527768 , archivePrefix =. 1601.00162 , primaryClass =

    Pith/arXiv arXiv

  47. [47]

    , keywords =

    Radiative transfer as a Bayesian linear regression problem. , keywords =. doi:10.1093/mnras/stac3461 , archivePrefix =. 2211.12547 , primaryClass =

    Pith/arXiv arXiv

  48. [48]

    1972 , month = aug, volume =

    The Density of H2 Molecules in Dark Interstellar Clouds , author =. 1972 , month = aug, volume =

    1972

  49. [49]

    Monthly Notices of the Royal Astronomical Society , volume =

    Rotational relaxation of HCO+ and DCO+ by collision with H2 , author =. Monthly Notices of the Royal Astronomical Society , volume =. 2020 , month =. doi:10.1093/mnras/staa2308 , issn =

    work page doi:10.1093/mnras/staa2308 2020

  50. [50]

    , keywords =

    Disentangling CO Chemistry in a Protoplanetary Disk Using Explanatory Machine-learning Techniques. , keywords =. doi:10.3847/1538-4357/ad11ed , archivePrefix =. 2312.05254 , primaryClass =

    Pith/arXiv arXiv

  51. [51]

    , keywords =

    Laboratory and radio-astronomical spectroscopy of the hyperfine structure of N _ 2 D ^ +. , keywords =. doi:10.1051/0004-6361:20034025 , adsurl =

    work page doi:10.1051/0004-6361:20034025

  52. [52]

    , author =

    Photoelectric heating of interstellar gas. , author =. 1978 , month = apr, volume =. doi:10.1086/190513 , adsurl =

    doi:10.1086/190513 1978

  53. [53]

    1996 , month = sep, volume =

    Structure of Stationary Photodissociation Fronts , author =. 1996 , month = sep, volume =. doi:10.1086/177689 , archiveprefix =. astro-ph/9603032 , primaryclass =

    Pith/arXiv arXiv 1996

  54. [54]

    2003 , month = jan, volume =

    Interstellar Dust Grains , author =. 2003 , month = jan, volume =. doi:10.1146/annurev.astro.41.011802.094840 , archiveprefix =. astro-ph/0304489 , primaryclass =

    Pith/arXiv arXiv 2003

  55. [55]

    2010 , month = aug, volume =

    The rotational excitation of HCN and HNC by He: temperature dependence of the collisional rate coefficients , author =. 2010 , month = aug, volume =. doi:10.1111/j.1365-2966.2010.16826.x , adsurl =

    arXiv 2010

  56. [56]

    Application to model predictions , author =

    Quantifying the informativity of emission lines to infer physical conditions in giant molecular clouds: I. Application to model predictions , author =. 2024 , month = nov, volume =. doi:10.1051/0004-6361/202451588 , keywords =

    doi:10.1051/0004-6361/202451588 2024

  57. [58]

    2012 , month = sep, volume =

    The impact of collisional rate coefficients on molecular hyperfine selective excitation , author =. 2012 , month = sep, volume =. doi:10.1111/j.1365-2966.2012.21601.x , adsurl =

    arXiv 2012

  58. [59]

    1992 , month = mar, volume =

    Anisotropic Line Emission and the Geometry of the Broad-Line Region in Active Galactic Nuclei , author =. 1992 , month = mar, volume =. doi:10.1086/171063 , keywords =

    doi:10.1086/171063 1992

  59. [60]

    , keywords =

    Chemical post-processing of magneto-hydrodynamical simulations of star-forming regions: robustness and pitfalls. , keywords =. doi:10.1093/mnras/stab1525 , archivePrefix =. 2105.01649 , primaryClass =

    Pith/arXiv arXiv

  60. [61]

    2013 , month = aug, volume =

    Rate of Formation of Hydrogen Molecules by Three-body Recombination during Primordial Star Formation , author =. 2013 , month = aug, volume =. doi:10.1088/2041-8205/773/2/l25 , keywords =

    doi:10.1088/2041-8205/773/2/l25 2013

  61. [62]

    2015 , booktitle =

    Cosmic-ray heating of molecular cloud cores , author =. 2015 , booktitle =

    2015

  62. [63]

    2019 , month = oct, volume =

    Structure and kinematics of the Taurus star-forming region from Gaia-DR2 and VLBI astrometry , author =. 2019 , month = oct, volume =. doi:10.1051/0004-6361/201935928 , eid =

    doi:10.1051/0004-6361/201935928 2019

  63. [64]

    ISM properties in low-metallicity environments. II. The dust spectral energy distribution of NGC 1569 , author =. 2003 , month = aug, volume =. doi:10.1051/0004-6361:20030814 , keywords =

    doi:10.1051/0004-6361:20030814 2003

  64. [65]

    Formalism and benchmarking , author =

    A dust spectral energy distribution model with hierarchical Bayesian inference - I. Formalism and benchmarking , author =. 2018 , month = may, volume =. doi:10.1093/mnras/sty189 , keywords =

    work page doi:10.1093/mnras/sty189 2018

  65. [66]

    Scaling relations and constraints on the dust build-up in galaxies with the DustPedia and DGS samples , author =

    A nearby galaxy perspective on dust evolution. Scaling relations and constraints on the dust build-up in galaxies with the DustPedia and DGS samples , author =. 2021 , month = may, volume =. doi:10.1051/0004-6361/202039701 , keywords =

    doi:10.1051/0004-6361/202039701 2021

  66. [67]

    2011 , month = sep, howpublished =

    PySpecKit: Python Spectroscopic Toolkit , author =. 2011 , month = sep, howpublished =

    2011

  67. [68]

    , keywords =

    Pyspeckit: A Spectroscopic Analysis and Plotting Package. , keywords =. doi:10.3847/1538-3881/ac695a , archivePrefix =. 2205.04987 , primaryClass =

    Pith/arXiv arXiv

  68. [69]

    2012 , month = sep, volume =

    Cosmic-Ray and X-Ray Heating of Interstellar Clouds and Protoplanetary Disks , author =. 2012 , month = sep, volume =. doi:10.1088/0004-637x/756/2/157 , eid =

    doi:10.1088/0004-637x/756/2/157 2012

  69. [70]

    2010 , month = may, volume =

    Modelling CO formation in the turbulent interstellar medium , author =. 2010 , month = may, volume =. doi:10.1111/j.1365-2966.2009.15718.x , keywords =

    arXiv 2010

  70. [71]

    2012 , month = mar, volume =

    Approximations for modelling CO chemistry in giant molecular clouds: a comparison of approaches , author =. 2012 , month = mar, volume =. doi:10.1111/j.1365-2966.2011.20260.x , archiveprefix =. 1102.0670 , primaryclass =

    arXiv 2012

  71. [72]

    Impact of rate coefficient uncertainties , author =

    Simulating the formation of massive seed black holes in the early Universe - II. Impact of rate coefficient uncertainties , author =. 2015 , month = nov, volume =. doi:10.1093/mnras/stv1781 , archiveprefix =. 1504.00514 , primaryclass =

    Pith/arXiv arXiv 2015

  72. [73]

    2017 , month = jul, volume =

    A Simple and Accurate Network for Hydrogen and Carbon Chemistry in the Interstellar Medium , author =. 2017 , month = jul, volume =. doi:10.3847/1538-4357/aa7561 , eid =

    doi:10.3847/1538-4357/aa7561 2017

  73. [74]

    MaNN: Multiple Artificial Neural Networks for modelling the Interstellar Medium

    MaNN: Multiple Artificial Neural Networks for modelling the Interstellar Medium , author =. 2011 , month = mar, journal =. doi:10.48550/arXiv.1103.0509 , eid =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1103.0509 2011

  74. [75]

    2011 , month =

    ROBO: A model and a code for studying the interstellar medium , author =. 2011 , month =

    2011

  75. [76]

    2014 , month = apr, journal =

    KROME - a package to embed chemistry in astrophysical simulations , author =. 2014 , month = apr, journal =. doi:10.1093/mnras/stu114 , archiveprefix =. 1311.1070 , primaryclass =

    Pith/arXiv arXiv 2014

  76. [77]

    2017 , month = apr, volume =

    A detailed framework to incorporate dust in hydrodynamical simulations , author =. 2017 , month = apr, volume =. doi:10.1093/mnras/stw2871 , archiveprefix =. 1606.01229 , primaryclass =

    Pith/arXiv arXiv 2017

  77. [78]

    2022 , month = dec, volume =

    Reducing the complexity of chemical networks via interpretable autoencoders , author =. 2022 , month = dec, volume =. doi:10.1051/0004-6361/202039956 , eid =

    doi:10.1051/0004-6361/202039956 2022

  78. [79]

    , keywords =

    Mapping synthetic observations to pre-stellar core models: An interpretable machine learning approach. , keywords =. doi:10.1051/0004-6361/202453266 , archivePrefix =. 2502.07874 , primaryClass =

    arXiv

  79. [80]

    2016 , month = aug, volume =

    A New Reference Chemical Composition for TMC-1 , author =. 2016 , month = aug, volume =. doi:10.3847/0067-0049/225/2/25 , keywords =

    doi:10.3847/0067-0049/225/2/25 2016

  80. [81]

    doi:10.1051/0004-6361/202037871 , eid =

    2021 , month = jan, volume =. doi:10.1051/0004-6361/202037871 , eid =

    doi:10.1051/0004-6361/202037871 2021

Showing first 80 references.