NICER constraints#
NICER (Neutron star Interior Composition Explorer) is a NASA X-ray telescope on the International Space Station that measures X-ray pulse profiles from millisecond pulsars. By modelling the X-ray emission from hot spots on the neutron star surface, NICER constrains the stellar mass and radius simultaneously, providing direct input for equation-of-state inference.
JESTER currently supports five pulsars observed by NICER:
PSR J0030+0451
PSR J0437-4715
PSR J0614-3329
PSR J0740+6620
PSR J1614-2230
The figure below shows the mass-radius posteriors for the five supported pulsars, with filled contours at the 68% and 90% credible intervals. Where both the Amsterdam and Maryland groups have published an independent analysis for a pulsar, both are overlaid using each group’s preferred/headline dataset.
Below, we provide more details on the datasets that are available for each pulsar, and how to use them in jester.
(Source code, png, hires.png, pdf)
Mass-radius posteriors from NICER for the five supported pulsars. Filled contours show the 68% (darker) and 90% (lighter) credible intervals from kernel density estimation of the published posterior samples. Each pulsar is shown in a single colour (filled squares in the legend); where both the Amsterdam and Maryland groups have published an independent analysis, both are overlaid using the preferred/headline dataset from each group, with solid contours for Amsterdam and dashed contours for Maryland.#
Likelihood#
For each pulsar, JESTER uses a normalizing flow trained on the published M-R posterior samples.
The flow learns the joint density \(p(M, R \mid \text{data})\) for a given analysis group and hotspot model.
Pretrained flows are shipped with the jester repository along with the posterior samples downloaded from Zenodo releases.
The flows are stored in the directory jesterTOV/inference/flows/models/nicer_maf/.
When evaluating the likelihood for a candidate EOS, the computation proceeds as follows. At initialization, a fixed set of \(N\) masses \(M^{(i)}\) is drawn once from the marginal mass distribution of the flow. For each EOS evaluation, the radius \(R_\mathrm{EOS}(M^{(i)})\) is obtained by linear interpolation along the EOS M-R curve at each pre-sampled mass. The per-group log-likelihood is
evaluated numerically as \(\mathrm{logsumexp}_i\!\left(\log p_\mathrm{flow}^{(i)}\right) - \log N\). If multiple analysis groups are used (e.g. Amsterdam and Maryland for PSR J0030+0451), the final log-likelihood averages over groups (with equal weights) in the same way as the above equation.
Mass samples that exceed \(M_\mathrm{TOV}\) receive a large negative penalty, if enabled by the user (switched off by default).
The pre-sampling at initialization fixes the mass grid for the entire run, which ensures deterministic and smooth likelihood evaluations — important for sampler convergence.
The default implementation is NICERLikelihood (flow-based).
A legacy NICERKDELikelihood based on kernel density estimation of the raw posterior samples is also available for comparison.
Data and flows#
Mass-radius posterior samples for all supported pulsars are stored as compressed NumPy archives (.npz) under jesterTOV/inference/data/NICER/.
Each file contains radius (km), mass (solar masses), and a metadata dictionary with the pulsar name, analysis group, hotspot model, Zenodo record URL, and paper reference.
The raw Zenodo archives (which can be several gigabytes) are downloaded into jesterTOV/inference/data/NICER/zenodo_data/ (gitignored).
Only the lightweight extracted .npz files are version-controlled.
Download script: jesterTOV/inference/data/NICER/download_nicer.py
This single script handles the full pipeline — downloading Zenodo archives, extracting mass-radius samples from raw text files and tar archives, and downsampling large posteriors to at most 100,000 samples. It requires the zenodo-get package:
uv pip install zenodo-get
uv run python jesterTOV/inference/data/NICER/download_nicer.py
Warning
Downloading the raw Zenodo archives can take a significant amount of time. Several
records contain full posterior chains that are multiple gigabytes in size. Make sure
you have sufficient disk space and a stable internet connection before running the
script. The processed .npz files (the only outputs you need for inference) are
already version-controlled in the repository, so you only need to run this script
if you want to reproduce the extraction from scratch.
Zenodo record identifiers and paper metadata are maintained in
jesterTOV/inference/data/NICER/zenodo_downloader.py.
PSR J0030+0451#
First millisecond pulsar observed by NICER with sufficient quality for mass-radius inference. It has since been analyzed independently by the Amsterdam (X-PSI) and Maryland groups multiple times as more NICER exposure accumulated.
Amsterdam group — Kini, Mauviard, Salmi, et al. 2026 (arXiv:2602.23743; Zenodo 18741942)
Most recent Amsterdam analysis, using six years of NICER data:
J00300451_amsterdam_PDTU_NICERXMM_Kini2026.npz— PDT-U, NICER+XMM (recommended as default Amsterdam flow for J0030+0451)
Amsterdam group — Riley et al. 2019 (ApJL 887, L21; Zenodo 7096789)
Original Amsterdam analysis, using the 2017-2018 NICER-only dataset. Five hotspot geometries are available:
J00300451_amsterdam_ST_S_NICER_only_Riley2019.npz— ST symmetricJ00300451_amsterdam_ST_U_NICER_only_Riley2019.npz— ST unrestrictedJ00300451_amsterdam_CDT_U_NICER_only_Riley2019.npz— CDT unrestrictedJ00300451_amsterdam_ST_EST_NICER_only_Riley2019.npz— ST+ESTJ00300451_amsterdam_ST_PST_NICER_only_Riley2019.npz— ST+PST
Maryland group — Miller et al. 2019 (ApJL 887, L24; Zenodo 3473466)
Two hotspot geometries x two prior choices:
J00300451_maryland_2spot_NICER_only_RM.npz— 2-spot, restricted-model priorJ00300451_maryland_2spot_NICER_only_full.npz— 2-spot, full priorJ00300451_maryland_3spot_NICER_only_RM.npz— 3-spot, restricted-model priorJ00300451_maryland_3spot_NICER_only_full.npz— 3-spot, full prior
Amsterdam group — Vinciguerra et al. 2023 (ApJ 961, 62; Zenodo 8239000)
Intermediate Amsterdam analysis (between Riley et al. 2019 and Kini et al. 2026):
J00300451_amsterdam_ST_U_NICER_only_Vinciguerra2023.npz— ST-UJ00300451_amsterdam_ST_PST_NICER_only_Vinciguerra2023.npz— ST+PSTJ00300451_amsterdam_ST_PDT_NICER_only_Vinciguerra2023.npz— ST+PDTJ00300451_amsterdam_PDT_U_NICER_only_Vinciguerra2023.npz— PDT-U
PSR J0437-4715#
The nearest and brightest millisecond pulsar, observed with NICER only. Two independent analyses are available and both are shown in the overview figure above.
Amsterdam group — Choudhury et al. 2024 (ApJL 971, L20; Zenodo 13766753)
Headline result using CST+PDT hotspot model with 3C50 background:
J04374715_amsterdam_CST_PDT_NICER_only_Choudhury2024.npz
Maryland group — Miller, Dittmann, Holt et al. 2026 (ApJL 1000, L48; Zenodo 17833896)
Headline result. The raw file (J0437_NICER_RM.txt) stores importance weights alongside mass and radius rather than equal-weight samples, so the extraction script importance-resamples it to
an equal-weight posterior before saving:
J04374715_maryland_3spotGPL_NICER_only_RM.npz
PSR J0614-3329#
A 1.4 solar-mass edge-on pulsar observed with NICER only. Two independent analyses are available and both are shown in the overview figure above.
Amsterdam group — Mauviard et al. 2025 (ApJ 995, 60; Zenodo 17380576)
Headline result using ST+PDT hotspot model:
J06143329_amsterdam_ST_PDT_NICER_only_Mauviard2025.npz
Maryland group — Miller, Dittmann, Holt, et al. 2026 (arXiv:2609.00965; Zenodo 22131748)
Headline result. The favored model uses three uniform-temperature circular
hot spots fit to NICER-only data. The raw file (J0614_NICER_rm.txt)
stores importance weights alongside mass and radius rather than equal-weight
samples, so the extraction script importance-resamples it to an equal-weight
posterior before saving:
J06143329_maryland_3circle_NICER_only_RM.npz
PSR J0740+6620#
The most massive millisecond pulsar with a precise radio timing mass, providing high-density EOS constraints. Analyzed jointly with XMM-Newton data.
Amsterdam group — Salmi et al. 2024 (ApJ 974, 294; Zenodo 10519473)
Most recent analysis using gamma hotspot model with NICER+XMM-Newton data:
J07406620_amsterdam_gamma_NICERXMM_equal_weights_recent.npz
Amsterdam group — Riley et al. 2021 (ApJL 918, L27; Zenodo 7096886)
Original Amsterdam analysis using the ST-U hotspot model with NICER+XMM-Newton data:
J07406620_amsterdam_STU_NICERXMM_Riley2021.npz
Note: superseded by the Salmi et al. 2024 dataset above.
Maryland group — Dittmann et al. 2024 (ApJ 974, 295; Zenodo 10215109)
Headline result (NICER+XMM, fully ionized hydrogen atmosphere). The raw file stores importance weights alongside mass and radius, so the extraction script importance-resamples it to an equal-weight posterior before saving:
J07406620_maryland_NICERXMM_full_Dittmann2024.npz
Maryland group — Miller et al. 2021 (ApJL 918, L28; Zenodo 4670689)
Three dataset combinations x two prior choices:
J07406620_maryland_unknown_NICER_only_RM.npzJ07406620_maryland_unknown_NICER_only_full.npzJ07406620_maryland_unknown_NICERXMM_RM.npzJ07406620_maryland_unknown_NICERXMM_full.npzJ07406620_maryland_unknown_NICERXMM_relative_RM.npzJ07406620_maryland_unknown_NICERXMM_relative_full.npz
“NICER+XMM” indicates joint analysis; “relative” includes relative calibration between instruments; “RM” uses a restricted-model prior.
PSR J1614-2230#
A massive (\(\sim 1.94\,M_\odot\)) and compact millisecond pulsar, observed with NICER, XMM-Newton and Chandra.
Amsterdam group — Mauviard et al. 2026 (arXiv:2609.00172; Zenodo 22163155)
Headline result using the ST-U hotspot model (two circular hot spots), with \(M = 1.937^{+0.012}_{-0.013}\,M_\odot\) and \(R_\mathrm{eq} = 10.06^{+1.25}_{-0.87}\) km:
J16142230_amsterdam_STU_NICER_only_Mauviard2026.npz
Pulsars without samples#
The following NICER analyses are deliberately not included:
PSR J2124-3358 (González-Caniulef et al. 2026, arXiv:2607.03721; Zenodo 20640366): at the time of writing, the posterior samples for this pulsar are not yet public.
PSR J1231-1411 (Salmi et al. 2024, ApJ 976, 58; Zenodo 13358349): samples are public, but we know from private communication that they are probably not suitable for equation-of-state inference, so they are not used here.
Loading the data#
Each .npz file can be loaded directly:
import numpy as np
data = np.load("NICER/J07406620_amsterdam_gamma_NICERXMM_equal_weights_recent.npz",
allow_pickle=True)
radius = data["radius"] # km, shape: (n_samples,)
mass = data["mass"] # solar masses
meta = data["metadata"].item() # dict: psr, group, zenodo_record, paper, …
In normal usage the likelihood classes load the data automatically based on your YAML configuration.