Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

A physics-constrained neural field, trained through a differentiable ray tracer, recovers fine 3D structure in the HD 163296 disk from ALMA data — a CO layer that narrows and flattens past ~400 au, unseen by other models.

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

2026-08-05 10:47 UTC pith:BGFP7LPV

load-bearing objection A genuinely useful differentiable RT tool and a plausible but under-tested morphological claim; the plateau rests on an unexamined abundance threshold. the 3 major comments →

arxiv 2509.03623 v1 pith:BGFP7LPV submitted 2025-09-03 astro-ph.EP cs.CV

Revealing Fine Structure in Protoplanetary Disks with Physics Constrained Neural Fields

Aviad Levis , Nhan Luong , Richard Teague , Katherine. L. Bouman , Marcelo Barraza-Alfaro , Kevin Flaherty This is my paper
classification astro-ph.EP cs.CV
keywords protoplanetary disksneural fieldsdifferentiable line radiative transferALMA observationsHD 163296CO emission surfaceBayesian inferencescientific machine learning
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 paper sets out to show that the fine three-dimensional structure of a protoplanetary disk can be recovered directly from ALMA spectral cubes by representing gas temperature as a neural field — a continuous, coordinate-based function — instead of a low-dimensional analytic profile. To make that practical, the authors build RadJAX, a fully differentiable, GPU-vectorized line radiative transfer solver that renders velocity-resolved cubes in milliseconds rather than minutes and cuts MCMC inference from months of CPU time to hours on one GPU. Applied to the 12CO emission of HD 163296, the neural reconstruction fits the data about 19% better in mean χ² than the standard parametric model and removes the 'stair-case' separation between the upper and lower CO layers that the analytic model produces. Its central scientific result is a morphological transition: the CO-rich layer narrows and flattens, holding a nearly constant height beyond roughly 400 au — structure the parametric model cannot represent and the purely geometric surface estimate misreads as a dip. The paper argues this demonstrates a new regime of data-driven discovery, in which flexible but physics-constrained models turn high-resolution observations into structure that current approaches miss.

Core claim

On the paper's own terms, the discovery is that reproducing the ALMA 12CO(2–1) emission of HD 163296 forces the CO-emitting layer to stop flaring: the recovered temperature field drives a transition from a geometrically thick inner layer to a thin outer one, with the layer height plateauing at a nearly constant value beyond about 400 au. The paper reads this plateau as the feature behind the observed merging of the upper and lower CO layers in the image plane — the place where the parametric model fails and produces stair-case artifacts. Two pieces of evidence are offered for its reality: the plateau persists when the model is trained on only the red- or blue-shifted half of the data cube (c

What carries the argument

The load-bearing object is a physics-constrained neural field: a small multilayer perceptron that maps disk coordinates (r, |z|) to gas temperature with azimuthal and midplane mirror symmetry, using sinusoidal positional encoding of degree L=4 to control its spatial bandwidth. Temperature is the only free field; everything else is derived — H2 density by integrating hydrostatic equilibrium, the CO-emitting layer by Eq. 9 (a hard step function that keeps CO only where T > 19 K and the vertical column exceeds a fixed photodissociation threshold), and velocity by height-dependent Keplerian rotation with a pressure-gradient correction. These fields feed RadJAX, a GPU-vectorized, fully differenti

Load-bearing premise

The recovered CO-layer shape, including the plateau beyond 400 au, is produced by a hard step function on temperature and H2 column density (Eq. 9), with freeze-out fixed at 19 K and photodissociation above 1.256×10^21 cm^-2, applied to the fitted temperature field; if real CO abundance varies smoothly or obeys different chemistry, the inferred flattening may not be physical.

What would settle it

Replace Eq. 9's step-function CO abundance with a smoothly varying abundance from a thermochemical model in the same neural-field reconstruction: if the plateau beyond 400 au disappears or shifts, the claimed morphology is an artifact of the cutoff rather than a disk property. Observationally, map the same outer-disk layer with an optically thin tracer such as 13CO or C18O at high signal-to-noise — a continuously flaring emission surface past 400 au would directly contradict the flat layer.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If the plateau is real, the outer disk of HD 163296 is vertically far thinner than standard flared-disk models assume, which would mean current prescriptions for where CO resides in outer disks omit real physics.
  • RadJAX makes full Bayesian inference on high-resolution ALMA cubes feasible in hours on a single GPU, turning population-scale structural modeling of many disks from prohibitive into routine.
  • The neural model fits better in mean χ² while generalizing to held-out red- and blue-shifted channels, so its recovered temperature field is a candidate for the true vertical thermal structure rather than an interpolation artifact.
  • On the higher-resolution MAPS data the same framework recovers consistent radial structure but colder atmospheric temperatures and stronger turbulence, implying that resolution materially changes what can be inferred about these parameters.
  • Because the geometric dip radius and the neural plateau radius coincide, geometric surface extraction is biased low where the outer disk turns optically thin, and radiative-transfer-based reconstruction is the safer tool in that regime.

Where Pith is reading between the lines

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

  • My inference: if freeze-out and photodissociation set the plateau radius, that radius should track the local UV field and dust surface area; a testable extension is to cross-correlate the recovered layer height with dust-continuum substructure to see whether the plateau marks a change in grain properties.
  • My inference: the same end-to-end differentiable pipeline should transfer to other molecules (13CO, C18O, CS) and to other disks in the MAPS and exoALMA samples; a multi-line, multi-disk run would show whether the thick-to-thin transition is generic to protoplanetary disks or peculiar to HD 163296.
  • My inference: because gas above the CO layer emits nothing, the temperature gradient there is unconstrained by the data, so the plateau could be a temperature effect, a CO-abundance effect, or both; an independent scale-height measurement from dust continuum or scattering would break that degeneracy.
  • My inference: the step-function CO threshold of Eq. 9 does more work than the paper emphasizes — re-running the same fitted temperature field with a smooth CO abundance curve would reveal how much of the plateau is data-driven versus imposed by the chemistry prescription.

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 / 5 minor

Summary. This paper introduces RadJAX, a GPU-accelerated, differentiable line radiative transfer solver, and combines it with a physics-constrained neural field representation of the gas temperature to reconstruct the CO-emitting layer of the HD 163296 protoplanetary disk from ALMA CO(2-1) data cubes. The temperature field is the primary free function; H2 density, CO abundance, and velocity are derived from hydrostatic equilibrium, a freeze-out/photodissociation threshold prescription, and Keplerian rotation. The authors report a ~19% reduction in mean chi-squared relative to an 8-parameter parametric model, reproduce published MCMC posteriors for the parametric model at much lower cost, validate RadJAX against RADMC-3D on a simulation, and cross-validate the neural reconstruction on red- and blue-shifted channel subsets. The principal new scientific claim is that the CO-rich layer narrows and flattens beyond ~400 au, a morphology missed by the parametric model and not captured by the geometric disksurf emission-surface estimate.

Significance. If the recovered morphology is robust, the paper reports an interesting and potentially important transition in the vertical structure of HD 163296's outer disk, while the methodological contribution — a differentiable GPU radiative transfer solver enabling gradient-based optimization of high-dimensional neural fields — is timely and would be of broad use in protoplanetary disk analysis. The paper has real strengths: RadJAX is validated against an established solver on an independent simulation; the neural model's fit improvement is quantified; withheld-channel cross-validation is a sensible generalization check; and the MCMC validation against prior parametric results is reassuring. However, the central morphological claim rests on an untested hard-threshold CO abundance prescription, and no code or data are released, which limits reproducibility. The significance of the scientific claim is therefore conditional on additional sensitivity analysis.

major comments (3)
  1. [Methods, Eq. (9); Results, Figs. 5 and 9] The plateau beyond 400 au is not a direct observable. The CO layer is defined by Eq. (9) with fixed thresholds Tfreeze = 19 K and Ndissoc = 1.256e21 cm^-2, and the neural temperature field is explicitly stated in the Fig. 5 caption to be unconstrained above the CO layer. The recovered plateau is therefore a contour of the fitted temperature field under a hard abundance step, not a measured quantity. The red/blue channel withholding tests in Fig. 9 test spectral generalization but do not perturb Eq. (9) or its parameters. I ask for sensitivity tests: vary Tfreeze and Ndissoc over physically plausible ranges, and replace the step function with a smooth CO abundance transition. If the plateau persists, the claim is substantially strengthened; if not, the conclusion should be softened or reframed as a property of the adopted prescription.
  2. [Results, Fig. 6 and surrounding text] The geometric disksurf surface is dismissed as unreliable beyond 400 au because the outer disk may be optically thin. But the same optically thin regime weakens the direct observational constraint on the neural radiative transfer solution: if the emission is faint and the CO layer is set by thresholds rather than by a bright tau~1 surface, the outer plateau is largely determined by the neural prior and the abundance prescription. The agreement between the geometric dip and the neural plateau is suggestive, but it is not an independent validation. A synthetic recovery test — inject a known plateau (or known non-plateau) into a model cube, add noise, and ask whether the neural pipeline recovers it — would directly address whether the observed data actually constrain the outer surface height. Without such a test, the claim that the outer disk is 'vertically thinner' overreaches the evidence
  3. [Results, Fig. 10; Discussion; no code/data release] The claimed 'up to 10,000x speedups' over RADMC-3D is a central advertised contribution, but Figure 10 reports a speedup of 'over 3000x' at high spectral resolution, and no configuration achieving 10,000x is shown. The manuscript also does not provide a link to RadJAX code or the trained neural fields, making the inverse-modeling result unreproducible. Please release the solver and the reconstruction code (or at minimum the trained model and data-processing scripts), and clarify the exact benchmark conditions under which 10,000x is obtained versus the 3000x shown. This is important both for the reader to trust the performance claim and for the community to build on the method.
minor comments (5)
  1. [Abstract] Typo: 'Our work establish a new paradigm' should be 'Our work establishes...'.
  2. [Results, Fig. 4] The sentence 'This localized improvement could indicate that the parametric model is biased toward an asymmetric feature' is speculative; consider clarifying that this is one possible interpretation, or rephrasing to avoid implying a conclusion.
  3. [Methods, MCMC Inference] The description of the two-stage MCMC re-centering is clear, but the choice to discard 75% of samples as burn-in based on 'walker plots' would be better supported by a quantitative convergence diagnostic (e.g., integrated autocorrelation time or Gelman-Rubin).
  4. [References] Reference numbering in the author list is inconsistent: the affiliation footnotes skip department 6 and list seven departments for five authors. Also, some references in the text (e.g., the Isella et al. dust asymmetry) are cited only parenthetically without context; expand where needed.
  5. [Methods, Eq. (9)] The factor 0.706 is said to convert H2 column to total gas column, but it would help to state the assumed mean molecular weight explicitly in the text rather than only citing [32].

Circularity Check

0 steps flagged

No significant circularity: the recovered CO-layer morphology is a fitted model output, not an assumed input; the paper's red/blue cross-validation provides independent support, and the acknowledged unconstrained region above the layer is an identifiability limitation rather than a circular step.

full rationale

The derivation chain is self-contained and non-circular. The neural temperature field is optimized directly against the ALMA CO data cube using the differentiable RadJAX radiative transfer solver. The CO-rich layer is then derived by applying the chemical prescription in Methods Eq. 9 (XCO = 10^-4 only when T > Tfreeze and 0.706 N_H2 > Ndissoc, with Tfreeze=19 K and Ndissoc=1.256e21 cm^-2, fixed from external literature, Visser et al. 2009). The plateau beyond ~400 au is not an input or a fitted parameter; it emerges from the fitted neural temperature field through this threshold. The paper tests robustness by fitting exclusively to red- or blue-shifted channel subsets and shows the plateau persists (Fig. 9), which is genuine cross-validation rather than a forced identity. The paper also compares against an independent geometric extraction (disksurf), further grounding the result. The main caveat is explicitly stated in the Fig. 5 caption: 'Steep temperature gradients above the CO layer remain unconstrained, as they produce no observable emission.' This means the height of the upper CO surface is weakly constrained by the data, and the plateau location could be influenced by the network's smoothness prior and by the fixed Eq. 9 thresholds. That is a real robustness/identifiability concern, but it does not make the claim circular: the layer shape is not defined to equal the observed emission surface by construction, and no fitted parameter is renamed as a prediction. Self-citations (Levis et al. for neural fields; Teague for disksurf) are method provenance, not load-bearing justifications of the central result, and no uniqueness theorem or ansatz is imported solely through self-citation. The paper's central claim is therefore a legitimate, if imperfectly constrained, reconstruction result.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 0 invented entities

The central morphological claim depends on the free neural temperature field, the hydrostatic density, and the hard-threshold CO abundance model. No new physical entities are introduced.

free parameters (4)
  • Neural network weights of temperature MLP = tens of thousands (4x64 ReLU MLP, positional encoding L=4)
    The temperature field T(r,|z|) is the primary free function; its weights are optimized against the ALMA data with ADAM.
  • Parametric model parameters (Tmid, Tatm, q, qin, rbreak, rin, log10(rc), vturb) = posterior medians in Table 1
    Used in the MCMC baseline (Figure 7, Table 1) to validate RadJAX and to compare with the neural model.
  • Noise levels per dataset = 7 mJy/beam (Flaherty), 2 mJy/beam (MAPS)
    Estimated from background pixels; used to define chi2 loss and likelihood. Not central to the morphological claim.
  • Pixel stride for MAPS data = 3 pixels
    Chosen to decorrelate beam-smeared pixels; changes the effective weighting of the loss and affects small-scale parameter recovery.
axioms (6)
  • domain assumption Hydrostatic equilibrium relates density and temperature (Eq. 1)
    Used to derive nH2 from the neural T(r,z); assumes no bulk vertical motion or magnetic support.
  • domain assumption CO abundance is a step function of temperature and H2 column (Eq. 9)
    Sets the CO layer geometry; directly shapes the recovered emission surface and the plateau.
  • domain assumption Azimuthal and midplane mirror symmetry of temperature
    Reduces the neural field to T(r,|z|); excludes warps or lopsided structures that could mimic the plateau.
  • domain assumption LTE and Gaussian line profile (Eq. 15) with thermal and turbulent broadening
    Standard molecular line radiative transfer; ignores non-LTE excitation effects.
  • domain assumption Keplerian rotation with pressure-gradient correction (Eq. 6)
    Maps velocity to spatial coordinate through Doppler shift; assumes no radial flows or embedded planets in the fitted region.
  • domain assumption Fixed stellar and disk parameters (M*=2.3 Msun, Mgas=0.09 Msun, gamma=1, rscale=150 AU, inclination 47.5 deg, etc.)
    Taken from prior studies; errors in these values propagate into the vertical scale height and the radius of the plateau.

pith-pipeline@v1.4.0-alltime-deepseek-medium · 13221 in / 10354 out tokens · 100870 ms · 2026-08-05T10:47:06.705791+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Revealing Fine Structure in Protoplanetary Disks with Physics Constrained Neural Fields." pith.science (2026). https://pith.science/paper/BGFP7LPV

@misc{pith2026250903623,
  author       = {Pith},
  title        = {Pith review of: Revealing Fine Structure in Protoplanetary Disks with Physics Constrained Neural Fields},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BGFP7LPV}},
  note         = {Machine review of arXiv:2509.03623}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Protoplanetary disks are the birthplaces of planets, and resolving their three-dimensional structure is key to understanding disk evolution. The unprecedented resolution of ALMA demands modeling approaches that capture features beyond the reach of traditional methods. We introduce a computational framework that integrates physics-constrained neural fields with differentiable rendering and present RadJAX, a GPU-accelerated, fully differentiable line radiative transfer solver achieving up to 10,000x speedups over conventional ray tracers, enabling previously intractable, high-dimensional neural reconstructions. Applied to ALMA CO observations of HD 163296, this framework recovers the vertical morphology of the CO-rich layer, revealing a pronounced narrowing and flattening of the emission surface beyond 400 au - a feature missed by existing approaches. Our work establish a new paradigm for extracting complex disk structure and advancing our understanding of protoplanetary evolution.

discussion (0)

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

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

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

    astro-ph.IM 2026-07 conditional novelty 6.0

    An end-to-end differentiable JAX pipeline couples 1D hydrodynamics, time-dependent chemistry, and radiative transfer, and recovers shock and rate parameters from synthetic HCO+ spectra.

  2. IRIS: Deciphering Spectral-Line Imagery of the Galactic Center by Machine-Learning on Simulations

    astro-ph.GA 2026-07 unverdicted novelty 6.0

    A CNN trained on AREPO simulations and synthetic observations reverts edge-on 13CO spectral data to top-down views of the CMZ as a proof-of-concept for supervised reversion.

Reference graph

Works this paper leans on

34 extracted references · 20 canonical work pages · cited by 2 Pith papers · 5 internal anchors

  1. [1]

    The Disk Substructures at High Angular Resolution Project (DSHARP). I. Motivation, Sample, Calibration, and Overview

    Sean M. Andrews et al. “The Disk Substructures at High Angular Resolution Project (DSHARP). I. Motivation, Sample, Calibration, and Overview”. In: ApJ 869.2, L41 (Dec. 2018), p. L41. doi: 10.3847/2041- 8213/aaf741. arXiv: 1812. 04040 [astro-ph.SR]

    doi:10.3847/2041- 2018

  2. [2]

    Physical Processes in Protoplanetary Disks

    Philip J. Armitage. “Physical Processes in Protoplanetary Disks”. In: From Pro- toplanetary Disks to Planet Formation: Saas-Fee Advanced Course 45. Swiss Society for Astrophysics and Astronomy . Ed. by Marc Audard, Michael R. Meyer, and Yann Alibert. Berlin, Heidelberg: Springer Berlin Heidelberg, 2019, pp. 1–150. isbn: 978-3-662-58687-7. doi: 10.1007/978...

    doi:10.1007/978- 2019

  3. [3]

    Kinematic signatures of planet-disk interactions in VSI-turbulent protoplanetary disks

    Marcelo Barraza-Alfaro, Mario Flock, and Thomas Henning. “Kinematic sig- natures of planet-disk interactions in vertical shear instability-turbulent proto- planetary disks”. In: A&A 683, A16 (Mar. 2024), A16. doi: 10 . 1051 / 0004 - 6361/202347726. arXiv: 2310.18484 [astro-ph.EP]

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  4. [4]

    exoALMA. XVI. Predicting Signatures of Large-scale Turbulence in Protoplanetary Disks

    Marcelo Barraza-Alfaro et al. “exoALMA. XVI. Predicting Signatures of Large- scale Turbulence in Protoplanetary Disks”. In: ApJ 984.1, L21 (May 2025), p. L21. doi: 10.3847/2041-8213/adc42d. arXiv: 2504.19853 [astro-ph.EP]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    JAX: composable transformations of Python+NumPy programs

    James Bradbury et al. JAX: composable transformations of Python+NumPy programs. Version 0.3.13. 2018. url: http://github.com/jax-ml/jax

    work page 2018

  6. [6]

    A dusty filament and turbulent CO spirals in HD135344B-SAO206462

    Simon Casassus et al. “A dusty filament and turbulent CO spirals in HD 135344B - SAO 206462”. In: MNRAS 507.3 (Nov. 2021), pp. 3789–3809. doi: 10.1093/ mnras/stab2359. arXiv: 2104.08379 [astro-ph.EP]

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  7. [7]

    Structure of the DM Tau Outer Disk: Probing the vertical kinetic temperature gradient

    E. Dartois, A. Dutrey, and S. Guilloteau. “Structure of the DM Tau Outer Disk: Probing the vertical kinetic temperature gradient”. In: A&A 399 (Feb. 2003), pp. 773–787. doi: 10.1051/0004-6361:20021638

    doi:10.1051/0004-6361:20021638 2003

  8. [8]

    RADMC-3D: A multi-purpose radiative transfer tool

    CP Dullemond et al. “RADMC-3D: A multi-purpose radiative transfer tool”. In: Astrophysics Source Code Library (2012), ascl–1202

    work page 2012

  9. [9]

    Measuring turbulent motion in planet-forming disks with ALMA: A detection around DM Tau and non-detections around MWC 480 and V4046 Sgr

    Kevin Flaherty et al. “Measuring Turbulent Motion in Planet-forming Disks with ALMA: A Detection around DM Tau and Nondetections around MWC 480 and V4046 Sgr”. In: ApJ 895.2, 109 (June 2020), p. 109. doi: 10.3847/1538- 4357/ab8cc5. arXiv: 2004.12176 [astro-ph.SR]

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  10. [10]

    A Three-dimensional View of Turbulence: Constraints on Turbulent Motions in the HD 163296 Protoplanetary Disk Using DCO +

    Kevin M. Flaherty et al. “A Three-dimensional View of Turbulence: Constraints on Turbulent Motions in the HD 163296 Protoplanetary Disk Using DCO +”. In: ApJ 843.2, 150 (July 2017), p. 150. doi: 10.3847/1538-4357/aa79f9. arXiv: 1706.04504 [astro-ph.EP]

    Pith/arXiv arXiv 2017

  11. [11]

    Weak Turbulence in the HD 163296 Protoplanetary Disk Revealed by ALMA CO Observations

    Kevin M. Flaherty et al. “Weak Turbulence in the HD 163296 Protoplanetary Disk Revealed by ALMA CO Observations”. In: ApJ 813.2, 99 (Nov. 2015), p. 99. doi: 10.1088/0004-637X/813/2/99. arXiv: 1510.01375 [astro-ph.SR]

    Pith/arXiv arXiv 2015

  12. [12]

    emcee: the MCMC hammer

    Daniel Foreman-Mackey et al. “emcee: the MCMC hammer”. In: Publications of the Astronomical Society of the Pacific 125.925 (2013), p. 306

    work page 2013

  13. [13]

    Ensemble samplers with affine invari- ance

    Jonathan Goodman and Jonathan Weare. “Ensemble samplers with affine invari- ance”. In: Communications in applied mathematics and computational science 5.1 (2010), pp. 65–80

    work page 2010

  14. [14]

    Predicting the kinematic evidence of gravitational instability

    C. Hall et al. “Predicting the Kinematic Evidence of Gravitational Instability”. In: ApJ 904.2, 148 (Dec. 2020), p. 148. doi: 10.3847/1538-4357/abac17. arXiv: 2007.15686 [astro-ph.SR]

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  15. [15]

    Aperture synthesis with a non-regular distribution of interferom- eter baselines

    JA H¨ ogbom. “Aperture synthesis with a non-regular distribution of interferom- eter baselines”. In: Astronomy and Astrophysics Supplement, Vol. 15, p. 417 15 (1974), p. 417

    work page 1974

  16. [16]

    The Disk Substructures at High Angular Resolution Project (DSHARP). IX. A high-definition study of the HD 163296 planet-forming disk

    Andrea Isella et al. “The Disk Substructures at High Angular Resolution Project (DSHARP). IX. A high-definition study of the HD 163296 planet-forming disk”. In: The Astrophysical Journal Letters 869.2 (2018), p. L49

    work page 2018

  17. [17]

    Adam: A method for stochastic optimiza- tion

    Diederik P Kingma and Jimmy Ba. “Adam: A method for stochastic optimiza- tion”. In: ICLR (2014). 20

    work page 2014

  18. [18]

    Gravitationally Lensed Black Hole Emission Tomography

    Aviad Levis et al. “Gravitationally Lensed Black Hole Emission Tomography”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022, pp. 19841–19850

    work page 2022

  19. [19]

    Orbital polarimetric tomography of a flare near the Sagit- tarius A* supermassive black hole

    Aviad Levis et al. “Orbital polarimetric tomography of a flare near the Sagit- tarius A* supermassive black hole”. In: Nature Astronomy 8.6 (2024), pp. 765– 773

    work page 2024

  20. [20]

    NeRF: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall et al. “NeRF: Representing scenes as neural radiance fields for view synthesis”. In: ECCV (2020)

    work page 2020

  21. [21]

    Molecules with ALMA at Planet-forming Scales (MAPS). I. Program Overview and Highlights

    Karin I. ¨Oberg et al. “Molecules with ALMA at Planet-forming Scales (MAPS). I. Program Overview and Highlights”. In: ApJS 257.1, 1 (Nov. 2021), p. 1. doi: 10.3847/1538-4365/ac1432. arXiv: 2109.06268 [astro-ph.EP]

    Pith/arXiv arXiv 2021

  22. [22]

    Planet Formation Signposts: Observability of Circum- planetary Disks via Gas Kinematics

    Sebastian Perez et al. “Planet Formation Signposts: Observability of Circum- planetary Disks via Gas Kinematics”. In: ApJ 811.1, L5 (Sept. 2015), p. L5. doi: 10.1088/2041-8205/811/1/L5. arXiv: 1505.06808 [astro-ph.EP]

    Pith/arXiv arXiv 2015

  23. [23]

    Direct mapping of the temperature and velocity gradients in discs. Imaging the vertical CO snow line around IM Lupi

    C. Pinte et al. “Direct mapping of the temperature and velocity gradients in discs. Imaging the vertical CO snow line around IM Lupi”. In: A&A 609, A47 (Jan. 2018), A47. doi: 10 . 1051 / 0004 - 6361 / 201731377. arXiv: 1710 . 06450 [astro-ph.SR]

    work page 2018

  24. [24]

    Kinematic Structures in Planet-Forming Disks

    C. Pinte et al. “Kinematic Structures in Planet-Forming Disks”. In: Protostars and Planets VII . Ed. by S. Inutsuka et al. Vol. 534. Astronomical Society of the Pacific Conference Series. July 2023, p. 645. doi: 10.48550/arXiv.2203.09528. arXiv: 2203.09528 [astro-ph.EP]

    Pith/arXiv arXiv doi:10.48550/arxiv.2203.09528 2023

  25. [25]

    A spatially resolved vertical temperature gradient in the HD 163296 disk

    Katherine A Rosenfeld et al. “A spatially resolved vertical temperature gradient in the HD 163296 disk”. In: The Astrophysical Journal 774.1 (2013), p. 16

    work page 2013

  26. [26]

    An atomic and molecular database for analysis of submillimetre line observations

    Fredrik L Sch¨ oier et al. “An atomic and molecular database for analysis of submillimetre line observations”. In: Astronomy & Astrophysics 432.1 (2005), pp. 369–379

    work page 2005

  27. [27]

    Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains

    Matthew Tancik et al. “Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains”. In: NeurIPS (2020)

    work page 2020

  28. [28]

    disksurf: Extracting the 3D Structure of Protoplanetary Disks

    Richard Teague et al. “disksurf: Extracting the 3D Structure of Protoplanetary Disks”. In: Journal of Open Source Software 6.67 (2021), p. 3827. doi: 10.21105/ joss.03827. url: https://doi.org/10.21105/joss.03827

    doi:10.21105/joss.03827 2021

  29. [29]

    exoALMA. I. Science Goals, Project Design, and Data Products

    Richard Teague et al. “exoALMA. I. Science Goals, Project Design, and Data Products”. In: ApJ 984.1, L6 (May 2025), p. L6. doi: 10 . 3847 / 2041 - 8213 / adc43b. arXiv: 2504.18688 [astro-ph.EP]

    Pith/arXiv arXiv 2025

  30. [30]

    Mapping the Complex Kinematic Substructure in the TW Hya Disk

    Richard Teague et al. “Mapping the Complex Kinematic Substructure in the TW Hya Disk”. In: ApJ 936.2, 163 (Sept. 2022), p. 163. doi: 10.3847/1538- 4357/ac88ca. arXiv: 2208.04837 [astro-ph.EP]

    Pith/arXiv arXiv 2022

  31. [31]

    Molecules with ALMA at Planet-forming Scales (MAPS). XVIII. Kinematic Substructures in the Disks of HD 163296 and MWC 480

    Richard Teague et al. “Molecules with ALMA at Planet-forming Scales (MAPS). XVIII. Kinematic Substructures in the Disks of HD 163296 and MWC 480”. In: ApJS 257.1, 18 (Nov. 2021), p. 18. doi: 10.3847/1538-4365/ac1438. arXiv: 2109.06218 [astro-ph.EP]. 21

    Pith/arXiv arXiv 2021

  32. [32]

    The photodissociation and chemistry of CO isotopologues: applications to interstellar clouds and circum- stellar disks

    R. Visser, E. F. van Dishoeck, and J. H. Black. “The photodissociation and chemistry of CO isotopologues: applications to interstellar clouds and circum- stellar disks”. In: A&A 503.2 (Aug. 2009), pp. 323–343. doi: 10 . 1051 / 0004 - 6361/200912129. arXiv: 0906.3699 [astro-ph.GA]

    Pith/arXiv arXiv 2009

  33. [33]

    A Parametric Modeling Approach to Measuring the Gas Masses of Circumstellar Disks

    Jonathan P. Williams and William M. J. Best. “A Parametric Modeling Approach to Measuring the Gas Masses of Circumstellar Disks”. In: ApJ 788.1, 59 (June 2014), p. 59. doi: 10.1088/0004- 637X/788/1/59. arXiv: 1312.0151 [astro-ph.EP]

    Pith/arXiv arXiv 2014

  34. [34]

    Single View Refractive Index Tomography with Neural Fields

    Brandon Zhao et al. “Single View Refractive Index Tomography with Neural Fields”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 25358–25367. 22

    work page 2024