pith. sign in

arxiv: 2606.30886 · v1 · pith:JIBTZ4QMnew · submitted 2026-06-29 · 🌌 astro-ph.HE · gr-qc· physics.data-an

Multipolar Magnetic-Field Inference for PSR J0740+6620 with Neural-Network-Accelerated NICER Pulse-Profile Modeling

Farhana Taiyebah , Constantinos Kalapotharakos , Greg Olmschenk , Wendy F. Wallace , Soumi De , Abu Bucker Siddik , Diane Oyen , Thibault Lechien
show 1 more author
Zorawar Wadiasingh
This is my paper

Pith reviewed 2026-07-01 01:31 UTC · model grok-4.3

classification 🌌 astro-ph.HE gr-qcphysics.data-an
keywords PSR J0740+6620NICER pulse profilemultipolar magnetic fieldneural network surrogatemillisecond pulsarmagnetic field inferenceoffset dipolebolometric light curve
0
0 comments X

The pith

An offset dipole-plus-quadrupole magnetic field reproduces the double-peaked NICER pulse profile of PSR J0740+6620 while a zero-offset version is disfavored.

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

The paper models the emitting regions on PSR J0740+6620 as the footpoints of open field lines belonging to an offset dipole combined with an axisymmetric quadrupole, rather than as fixed geometric hotspots. Neural-network surrogates trained on 51.2 million synthetic bolometric light curves accelerate the likelihood calculations enough to permit an 11-dimensional Markov Chain Monte Carlo exploration with stellar mass, radius, inclination, and temperature ratio held fixed. The resulting posteriors show that the offset configuration can match the observed 32-bin profile, while the restricted zero-offset case is disfavored inside the chosen field basis. Pulse phase data tightly constrain azimuthal placement of the emission regions, but latitude, surface extent, and detailed morphology remain only weakly determined. The two temperature-weight prescriptions produce nearly identical results, indicating limited sensitivity to that modeling detail.

Core claim

By representing the emitting regions as the open-field-line footpoints of a static vacuum multipolar field consisting of an offset dipole plus axisymmetric quadrupole, and by using convolutional neural-network surrogates to evaluate 5.12 times 10 to the 7 synthetic light curves, the analysis shows that this model reproduces the observed double-peaked bolometric NICER pulse profile. The posteriors are broad and multimodal, the maximum-likelihood values are similar for the two calibrated temperature weights, and a restricted zero-offset run is disfavored within the adopted parameterization.

What carries the argument

The offset dipole plus axisymmetric quadrupole static vacuum multipolar field, whose open-field-line footpoints define the locations and shapes of the emitting regions.

If this is right

  • The neural-network surrogates accelerate likelihood evaluations by a factor of at least 400, making an 11-dimensional magnetic-field inference computationally feasible on parallel CPU resources.
  • The similarity of posteriors for the two temperature-weight prescriptions demonstrates weak sensitivity to that modeling choice.
  • Pulse phases constrain the approximate azimuthal placement of emission regions while latitude, surface extent, and morphology remain weakly constrained.
  • The method extends neural-network-accelerated multipolar inference to PSR J0740+6620 and supports future energy-dependent and force-free extensions.

Where Pith is reading between the lines

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

  • Jointly varying the fixed stellar and geometric parameters together with the magnetic-field parameters would likely expose additional degeneracies or tighten the constraints.
  • Adding gamma-ray pulse-profile data to the X-ray likelihood could help resolve the broad multimodal posteriors in the multipolar coefficients.
  • The same surrogate approach could be applied to other NICER-observed millisecond pulsars to test whether offset multipolar geometries are common.

Load-bearing premise

Stellar mass, radius, observer inclination, and hotspot temperature ratio are fixed to the Dittmann et al. (2024) maximum-likelihood values, and emitting regions coincide exactly with open-field-line footpoints of the static vacuum multipolar field.

What would settle it

An MCMC exploration restricted to zero offset that produces a comparable or higher likelihood value for the 32-bin NICER profile would show that the offset is not required within the adopted field basis.

Figures

Figures reproduced from arXiv: 2606.30886 by Abu Bucker Siddik, Constantinos Kalapotharakos, Diane Oyen, Farhana Taiyebah, Greg Olmschenk, Soumi De, Thibault Lechien, Wendy F. Wallace, Zorawar Wadiasingh.

Figure 1
Figure 1. Figure 1: Calibration of the simplified bolometric light-curve model. Red points show the background-sub￾tracted NICER pulse profile of PSR J0740+6620 with 1σ uncertainties. The black curve is the A. J. Dittmann et al. (2024) maximum-likelihood bolometric reference model. The blue and orange curves show our calibrated forward-model light curves for Tw = 1.31 and Tw = 1.41, respectively. Both calibrations reproduce t… view at source ↗
Figure 2
Figure 2. Figure 2: Test-set accuracy of the two neural-network surrogate models. The curves show the distribution of log10(MdNSE) for 105 held-out light curves, evaluated sep￾arately for the networks trained with Tw = 1.31 (blue) and Tw = 1.41 (orange). The two distributions nearly overlap, with median log10(MdNSE) ≈ −2, indicating similar surro￾gate accuracy for the two emission prescriptions. emission-calibration quantitie… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of physical-model light curves (dark red) and neural-network surrogate predictions for Tw = 1.31 (left, blue) and Tw = 1.41 (right, orange), evaluated on the held-out test set. Each panel shows five representative cases corresponding to the median and the ±1σ and ±2σ quantiles of the log10(MdNSE) distribution, ordered from best to worst (top to bottom) surrogate fidelity. The green curve beneath… view at source ↗
Figure 4
Figure 4. Figure 4: Log-likelihood convergence of the neural-network-accelerated-MCMC chains for PSR J0740+6620 under Tw = 1.31 (blue) and Tw = 1.41 (orange). Left: Full 2 × 106 -iteration run. Solid lines show the binned median ln L across all chains; shaded bands indicate the inter-quartile range (IQR, 25th–75th percentile); dotted lines show the running maximum. Right: A comparison of distributions of the log-likelihood ov… view at source ↗
Figure 5
Figure 5. Figure 5: Joint posterior distribution of the 11 magnetic-field parameters for PSR J0740+6620, overlaying results from Tw = 1.31 (blue) and Tw = 1.41 (orange). Each posterior is constructed from the last 10% of the respective post-burn-in chain (∼9.51 × 105 samples, each subsampled to 1 × 105 for display). Contours in the off-diagonal panels enclose the 68% and 95% credible regions. Diagonal panels show the marginal… view at source ↗
Figure 6
Figure 6. Figure 6: Posterior predictive light-curve ensemble for PSR J0740+6620 using 104 parameter samples drawn from it￾erations 1.5×106 –2×106 of the post-burn-in chain for the Tw = 1.41 run. Top: background-subtracted NICER ob￾served counts (red points with 1σ error bars), overlaid with the ±1σ (dark), ±2σ (medium), and ±3σ (light) contours of the predicted distribution and the median model (black line). Bottom: normaliz… view at source ↗
Figure 7
Figure 7. Figure 7: Surface emission geometry of PSR J0740+6620 inferred from the neural-network-accelerated MCMC posterior, shown on Mollweide projections of the stellar surface. (a) Accumulated hotspot density map constructed from 104 post-burn-in posterior samples. The color scale indicates the fraction of posterior samples for which each surface element belongs to the open-field-line emission region. Solid red and green c… view at source ↗
Figure 8
Figure 8. Figure 8: Posterior distribution for the centered, five-parameter magnetic-field model. In this run, both magnetic moments are fixed at the stellar center, so xD = yD = zD = xQ = yQ = zQ = 0. The remaining parameters are the dipole and quadrupole orientations and the strength ratio BQ/BD. The posterior is constructed from the final 10% of the post-burn-in samples for the Tw = 1.41 run. This model tests whether a cen… view at source ↗
Figure 9
Figure 9. Figure 9: Fit quality for the offset and zero-offset mag￾netic-field models. The curves show the post-burn-in dis￾tributions of reduced chi-squared, χ 2 r = (−2 ln L − C)/dof, where C = P i ln(2πσ2 i ). Blue and orange show the full offset dipole-plus-quadrupole model for Tw = 1.31 and Tw = 1.41, respectively; green shows the centered five-parameter model for Tw = 1.41. Dashed lines mark the medians, and the dot￾ted… view at source ↗
Figure 10
Figure 10. Figure 10: Posterior distribution of magnetic moment directions for PSR J0740+6620, constructed from 104 post-burn-in samples of the Tw = 1.41 run. (a) Dipole magnetic moment direction, determined by (αD, ϕD). (b) Quadrupole magnetic moment direction, determined by (αQ, ϕQ). present model includes only the rotated axisymmetric quadrupole component. A more complete multipolar ex￾pansion, combined with force-free magn… view at source ↗
read the original abstract

We investigate the multipolar surface magnetic-field structure of the high-mass millisecond pulsar PSR J0740+6620 using the 32-bin bolometric NICER pulse profile of Dittmann et al. (2024). Building on the neural-network surrogate framework of Olmschenk et al. (2025), we model the emitting regions as open-field-line footpoints of an offset dipole plus axisymmetric quadrupole static vacuum field, rather than as prescribed geometric hotspots. We fix the stellar mass, radius, observer inclination, and hotspot temperature ratio to the Dittmann et al. (2024) maximum-likelihood values and explore the resulting 11-dimensional magnetic-field space. To make this feasible, we train convolutional neural-network surrogates on $5.12\times10^7$ synthetic bolometric light curves and use them in a parallel ensemble Markov Chain Monte Carlo calculation on 4000 CPU cores, accelerating likelihood evaluations by a factor of $\gtrsim 400$. We perform independent inferences for two calibrated temperature-weight prescriptions, Tw=1.31 and Tw=1.41, encoding the relative bolometric weight associated with the hotspot temperature difference. The posteriors, posterior-predictive light curves, and maximum-likelihood values are very similar, indicating weak sensitivity to this choice. The offset model reproduces the observed double-peaked profile and yields broad, multimodal posteriors, reflecting both the background-dominated data and degeneracies of the multipolar parameterization. The hotspot-density map shows that pulse phases constrain the approximate azimuthal placement of the emission, while latitude, surface extent, and morphology remain weakly constrained. A restricted zero offset run is disfavored within the adopted field basis. This work extends neural-network-accelerated magnetic-field inference to PSR J0740+6620 and motivates future energy-dependent, force-free, and joint X-ray/$\gamma$-ray extensions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript performs an 11-dimensional Bayesian inference of the surface magnetic field of PSR J0740+6620 from its 32-bin NICER bolometric pulse profile, modeling emission regions as open-field-line footpoints of an offset dipole plus axisymmetric quadrupole vacuum field. Stellar mass, radius, observer inclination, and hotspot temperature ratio are fixed to the Dittmann et al. (2024) maximum-likelihood values; convolutional neural-network surrogates trained on 5.12×10^7 synthetic light curves accelerate likelihood evaluations by ≳400× to enable parallel ensemble MCMC. The offset model is reported to reproduce the observed double-peaked profile, with broad multimodal posteriors, while a restricted zero-offset run is disfavored; results are insensitive to the choice of two calibrated temperature-weight prescriptions.

Significance. If the results hold, the work demonstrates computational feasibility of neural-network-accelerated multipolar field inference on real high-mass MSP data, extending the Olmschenk et al. (2025) surrogate framework. The scale of the training set and parallel implementation on 4000 cores are notable strengths that enable exploration of an 11-dimensional parameter space.

major comments (2)
  1. [Abstract] Abstract: The claim that a restricted zero-offset run is disfavored is obtained after fixing stellar mass, radius, observer inclination, and hotspot temperature ratio to the Dittmann et al. (2024) maximum-likelihood point derived under a prescribed geometric hotspot model. Because these values are not re-optimized or marginalized under the open-field-line footpoint geometry, the apparent preference for nonzero offset may be an artifact of the conditioning rather than a robust comparison within the adopted field basis.
  2. [Abstract] Abstract / surrogate training description: No quantitative validation metrics (e.g., test-set mean absolute error, maximum surrogate prediction error, or posterior sensitivity to surrogate uncertainty) are reported for the convolutional neural-network surrogates, which are load-bearing for the reliability of the reported posteriors and the disfavoring statement.
minor comments (1)
  1. [Abstract] The abstract states that the posteriors are 'very similar' for Tw=1.31 and Tw=1.41 but does not quantify the level of agreement (e.g., via overlap integrals or parameter shifts) or provide the physical motivation for these specific calibrated values.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive comments on our manuscript. We address each major comment below and indicate where revisions will be made to improve clarity and completeness.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that a restricted zero-offset run is disfavored is obtained after fixing stellar mass, radius, observer inclination, and hotspot temperature ratio to the Dittmann et al. (2024) maximum-likelihood point derived under a prescribed geometric hotspot model. Because these values are not re-optimized or marginalized under the open-field-line footpoint geometry, the apparent preference for nonzero offset may be an artifact of the conditioning rather than a robust comparison within the adopted field basis.

    Authors: We agree that the comparison is conditional on the fixed stellar mass, radius, observer inclination, and temperature ratio taken from the Dittmann et al. (2024) maximum-likelihood point, which was obtained under a different geometric hotspot prescription. Our statement that the zero-offset model is disfavored therefore holds only within this fixed-parameter setup and the adopted multipolar field basis; it is not a fully marginalized result. We will revise the abstract to explicitly note this conditioning and the limited scope of the model comparison, while retaining the reported finding that, under these choices, the zero-offset run is disfavored. revision: yes

  2. Referee: [Abstract] Abstract / surrogate training description: No quantitative validation metrics (e.g., test-set mean absolute error, maximum surrogate prediction error, or posterior sensitivity to surrogate uncertainty) are reported for the convolutional neural-network surrogates, which are load-bearing for the reliability of the reported posteriors and the disfavoring statement.

    Authors: We acknowledge that the current manuscript does not report quantitative validation metrics such as test-set mean absolute error, maximum prediction error, or an assessment of how surrogate uncertainty affects the posteriors. Although the training-set size and the resulting speed-up factor are stated, these additional diagnostics are needed to substantiate the reliability of the MCMC results. In the revised manuscript we will add an appendix or dedicated subsection presenting the held-out test-set errors, maximum deviations, and a brief sensitivity test of the posteriors to surrogate noise, thereby addressing the referee's concern directly. revision: yes

Circularity Check

0 steps flagged

Minor self-citation of surrogate framework; central inference remains independent of fitted inputs

full rationale

The paper performs Bayesian inference on an 11-dimensional magnetic field parameter space using neural network surrogates trained on 5.12e7 synthetic light curves generated from the adopted multipolar field model. Stellar parameters are fixed to values from an external reference (Dittmann et al. 2024), and the data are from NICER observations reported therein. The comparison between offset and zero-offset models is conducted within this fixed setup, but the posteriors and model preference do not reduce by construction to the input parameters or prior citations. The surrogate training and MCMC sampling provide independent content, making the analysis self-contained against external benchmarks with only minor reliance on prior framework papers.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract-only review; the model introduces an 11-dimensional parameter space for the offset dipole and quadrupole components whose values are constrained by the data, plus the assumption that emission occurs at open-field-line footpoints.

free parameters (1)
  • 11-dimensional magnetic-field parameters
    Parameters describing the offset dipole plus axisymmetric quadrupole components are explored via MCMC.
axioms (1)
  • domain assumption Magnetic field is a static vacuum offset dipole plus axisymmetric quadrupole
    Used to define the open-field-line footpoints that determine the emitting regions.

pith-pipeline@v0.9.1-grok · 5932 in / 1269 out tokens · 43584 ms · 2026-07-01T01:31:47.249598+00:00 · methodology

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

29 extracted references · 27 canonical work pages · 1 internal anchor

  1. [1]

    A., Cheng, A

    Alpar, M. A., Cheng, A. F., Ruderman, M. A., & Shaham, J. 1982, Nature, 300, 728, doi: 10.1038/300728a0

    work page doi:10.1038/300728a0 1982

  2. [2]

    2010, ApJ, 715, 1282, doi: 10.1088/0004-637X/715/2/1282

    Bai, X.-N., & Spitkovsky, A. 2010, ApJ, 715, 1282, doi: 10.1088/0004-637X/715/2/1282

    work page doi:10.1088/0004-637x/715/2/1282 2010

  3. [3]

    Bhattacharya, D., & van den Heuvel, E. P. J. 1991, Phys. Rep., 203, 1, doi: 10.1016/0370-1573(91)90064-S

    work page doi:10.1016/0370-1573(91)90064-s 1991

  4. [4]

    V., Watts, A

    Bilous, A. V., Watts, A. L., Harding, A. K., et al. 2019, ApJL, 887, L23, doi: 10.3847/2041-8213/ab53e7

    work page doi:10.3847/2041-8213/ab53e7 2019

  5. [5]

    S., et al

    Bogdanov, S., Guillot, S., Ray, P. S., et al. 2019, ApJL, 887, L25, doi: 10.3847/2041-8213/ab53eb

    work page doi:10.3847/2041-8213/ab53eb 2019

  6. [6]

    Y., Yuan, Y., & Vasilopoulos, G

    Chen, A. Y., Yuan, Y., & Vasilopoulos, G. 2020, ApJL, 893, L38, doi: 10.3847/2041-8213/ab85c5

    work page doi:10.3847/2041-8213/ab85c5 2020

  7. [7]

    1999, ApJ, 511, 351, doi: 10.1086/306652

    Contopoulos, I., Kazanas, D., & Fendt, C. 1999, ApJ, 511, 351, doi: 10.1086/306652

    work page doi:10.1086/306652 1999

  8. [8]

    T., Fonseca, E., Ransom, S

    Cromartie, H. T., Fonseca, E., Ransom, S. M., et al. 2020, Nat. Astron., 4, 72, doi: 10.1038/s41550-019-0880-2

    work page doi:10.1038/s41550-019-0880-2 2020

  9. [9]

    J., Miller, M

    Dittmann, A. J., Miller, M. C., Lamb, F. K., et al. 2024, ApJ, 974, 295, doi: 10.3847/1538-4357/ad5f1e

    work page doi:10.3847/1538-4357/ad5f1e 2024

  10. [10]

    T., Pennucci, T

    Fonseca, E., Cromartie, H. T., Pennucci, T. T., et al. 2021, ApJL, 915, L12, doi: 10.3847/2041-8213/ac03b8

    work page doi:10.3847/2041-8213/ac03b8 2021

  11. [11]

    C., Arzoumanian, Z., Adkins, P

    Gendreau, K. C., Arzoumanian, Z., Adkins, P. W., et al. 2016, in Proc. SPIE, Vol. 9905, 99051H, doi: 10.1117/12.2231304

    work page doi:10.1117/12.2231304 2016

  12. [12]

    2010, Communications in Applied Mathematics and Computational Science, 5, 65, doi: 10.2140/camcos.2010.5.65

    Goodman, J., & Weare, J. 2010, Communications in Applied Mathematics and Computational Science, 5, 65, doi: 10.2140/camcos.2010.5.65

    work page doi:10.2140/camcos.2010.5.65 2010

  13. [13]

    K., & Muslimov, A

    Harding, A. K., & Muslimov, A. G. 2001, ApJ, 556, 987, doi: 10.1086/321589

    work page doi:10.1086/321589 2001

  14. [14]

    K., & Muslimov, A

    Harding, A. K., & Muslimov, A. G. 2002, ApJ, 568, 862, doi: 10.1086/338985

    work page doi:10.1086/338985 2002

  15. [15]

    2015, in Proceedings of the IEEE International Conference on Computer Vision (ICCV), 1026–1034, doi: 10.1109/ICCV.2015.123

    He, K., Zhang, X., Ren, S., & Sun, J. 2015, in Proceedings of the IEEE International Conference on Computer Vision (ICCV), 1026–1034, doi: 10.1109/ICCV.2015.123

    work page doi:10.1109/iccv.2015.123 2015

  16. [16]

    Deep Residual Learning for Image Recognition,

    He, K., Zhang, X., Ren, S., & Sun, J. 2016, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 770–778, doi: 10.1109/CVPR.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  17. [17]

    K., & Kazanas, D

    Kalapotharakos, C., Harding, A. K., & Kazanas, D. 2014, ApJ, 793, 97, doi: 10.1088/0004-637X/793/2/97

    work page doi:10.1088/0004-637x/793/2/97 2014

  18. [18]

    2021, ApJ, 907, 63, doi: 10.3847/1538-4357/abcec0

    Kazanas, D. 2021, ApJ, 907, 63, doi: 10.3847/1538-4357/abcec0

    work page doi:10.3847/1538-4357/abcec0 2021

  19. [19]

    The swept-back multipolar magnetic field of neutron stars: Application to NICER MSP J0030+0451

    Kundu, A., Kalapotharakos, C., Wadiasingh, Z., et al. 2026, arXiv e-prints. https://arxiv.org/abs/2604.19534

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  20. [20]

    2026, in preparation

    Lechien, T., Olmschenk, G., Kalapotharakos, C., et al. 2026, in preparation

    2026

  21. [21]

    2019, in International Conference on Learning Representations (ICLR)

    Loshchilov, I., & Hutter, F. 2019, in International Conference on Learning Representations (ICLR). https://openreview.net/forum?id=Bkg6RiCqY7

    2019

  22. [22]

    C., Lamb, F

    Miller, M. C., Lamb, F. K., Dittmann, A. J., et al. 2019, ApJL, 887, L24, doi: 10.3847/2041-8213/ab50c5

    work page doi:10.3847/2041-8213/ab50c5 2019

  23. [23]

    C., Lamb, F

    Miller, M. C., Lamb, F. K., Dittmann, A. J., et al. 2021, ApJL, 918, L28, doi: 10.3847/2041-8213/ac089b 18

    work page doi:10.3847/2041-8213/ac089b 2021

  24. [24]

    2025, ApJ, 991, 169, doi: 10.3847/1538-4357/ae03c0

    Olmschenk, G., Broadbent, E., Kalapotharakos, C., et al. 2025, ApJ, 991, 169, doi: 10.3847/1538-4357/ae03c0

    work page doi:10.3847/1538-4357/ae03c0 2025

  25. [25]

    Broadbent, E., & Wallace, W. F. 2025, haplo, Zenodo, doi: 10.5281/zenodo.14814679 P´ etri, J., Guillot, S., Guillemot, L., et al. 2023, A&A, 680, A93, doi: 10.1051/0004-6361/202346913

    work page doi:10.5281/zenodo.14814679 2025

  26. [26]

    K., Hebeler, K., et al

    Raaijmakers, G., Greif, S. K., Hebeler, K., et al. 2021, ApJL, 918, L29, doi: 10.3847/2041-8213/ac089a

    work page doi:10.3847/2041-8213/ac089a 2021

  27. [27]

    E., Watts, A

    Riley, T. E., Watts, A. L., Bogdanov, S., et al. 2019, ApJL, 887, L21, doi: 10.3847/2041-8213/ab481c

    work page doi:10.3847/2041-8213/ab481c 2019

  28. [28]

    E., Watts, A

    Riley, T. E., Watts, A. L., Ray, P. S., et al. 2021, ApJL, 918, L27, doi: 10.3847/2041-8213/ac0a81

    work page doi:10.3847/2041-8213/ac0a81 2021

  29. [29]

    2024, ApJ, 974, 294, doi: 10.3847/1538-4357/ad5f1f

    Salmi, T., Choudhury, D., Kini, Y., et al. 2024, ApJ, 974, 294, doi: 10.3847/1538-4357/ad5f1f

    work page doi:10.3847/1538-4357/ad5f1f 2024