WorksPhysics and Astronomy
The Radius of the Neutron Star PSR J0614-3329 from NICER Data
Abstract
Neutron star radius measurements, particularly using X-ray data collected with the Neutron star Interior Composition Explorer (NICER), have provided invaluable information for models of cold, catalyzed matter at densities above that of nuclear saturation. Here we present an analysis of NICER data on the 318-Hz pulsar PSR J0614-3329, which has a mass ~1.4-1.5 solar masses determined from radio observations. The best-fitting model we explore has three uniform-temperature circular hot spots and yields a symmetric 68% credible range for the equatorial circumferential radius of 9.88-12.77 km. We also explore joint fits to the NICER data and the X-ray Multi-Mirror (XMM-Newton) data on this pulsar, but find that the models that best fit both data sets systematically underpredict the XMM-Newton data when, as is standard, it is assumed that the XMM-Newton background is known from observations of surrounding fields. Finally, we discuss the implications of our results, combined with data on other neutron stars, for the properties of the dense matter in neutron star cores.
Corresponding author: M. C. Miller
NASA Einstein Fellow
† Deceased 28 July 2026
INTRODUCTION
The properties of the matter in cores of neutron stars, which has densities above the nuclear saturation density , is much colder than its Fermi temperature, and is highly isospin asymmetric, cannot be explored in terrestrial laboratories or predicted from first principles. Astronomical observations that constrain the properties of this cold, equilibrated, high-density matter are therefore crucial. During the last years, numerous data sets have contributed to our understanding of this matter. Analyses of data on X-ray pulsars acquired using NASA’s Neutron star Interior Composition Explorer (NICER) (see Gendreau et al. 2016) have played a particularly critical role.
NICER’s key capability is that it can determine the arrival time of individual X-ray photons with a precision of nsec or better while also determining their energy to within 10% or better. When folded at a pulsar’s rotational period, the NICER counts in each energy channel form a waveform in pulse phase. The data sets we analyze are therefore two-dimensional, consisting of the number of counts at each phase in each energy channel. When properly analyzed, these data produce constraints on the mass and radius of the star and hence on the equation of state (EOS) of the matter in it. (In this paper “mass” will always mean the star’s gravitational mass and “radius” will always mean the star’s equatorial circumferential radius.)
NICER’s neutron star targets are active radio pulsars and are therefore not accreting. This means that X-ray photons emitted by the stellar surface can propagate to the NICER detectors without interacting with accreting plasma in the system and that each photon emitted by the star can be mapped to a precise position on the star’s surface.
The NICER mission was motivated in part by early studies using synthetic data ((Lamb et al. 2009b,a; Lo et al. 2013; Miller & Lamb 2015; Watts et al. 2016)) which indicated that phase-resolved data, unlike phase-integrated data, could be analyzed in a way that is robust against many types of systematic errors (see (Steiner et al. 2010; Miller 2013; Kajava et al. 2014, and Miller & Lamb 2016) for caveats on radius estimates made using data from thermonuclear X-ray bursts or quiescent low-mass X-ray binaries, and Nättilä et al. 2017 for a possible improvement based on non-uniform surface emission). These studies have since been extended to include joint fits of NICER and X-ray Multi-Mirror Newton data (Holt et al. 2025).
Pulsars with previous NICER-based radius estimates include PSR J0030+0451 ((Miller et al. 2019; Riley et al. 2019; Vinciguerra et al. 2024; Kini et al. 2026)), PSR J0740+6620 ([19] (Miller et al. 2021; Riley et al. 2021; Salmi et al. 2022; Salmi et al. 2024a)), PSR J0437−4715 ([16] (Miller et al. 2026)), PSR 1231−1411 ((Salmi et al. 2024b; Qi et al. 2025)), PSR J0614−3329 (Mauviard et al. 2025), and PSR J2124−3358 (González-Caniulef et al. 2026).
Here we present a new analysis of NICER data on PSR J0614−3329. In § 2 we describe our extraction of the data, in § 3 we provide a brief overview of our methods, and in § 4 we present our results. In § 5 we discuss the implications of our results for the properties of cold, equilibrated matter at densities above the density of nuclear saturation. We summarize our conclusions in § 6.
In Appendix § 8 we present results from joint analyses of NICER and XMM-Newton data, including a comparison with the results of Mauviard et al. (2025) and show that when we make the standard assumption that we know the XMM-Newton background, the joint fit systematically underpredicts the observed XMM-Newton counts. Appendix § 9 provides further details of our results, including plots of the posteriors and corner plots. Posterior samples from our analysis are available on Zenodo via doi:10.5281/zenodo.22131748.
OBSERVATIONS
NICER XTI data
The NICER X-ray Timing Instrument (XTI) data for PSR J0614−3329 used in this analysis were collected between 2018 March 3 and 2023 May 15, corresponding to ObsID range 0030050101–6030050328, with a total unfiltered exposure of 1.6 Ms. The event lists from the individual ObsIDs were subjected to initial screening using the nicerl2 task in the NICERDAS v12 software that is part of HEASoft version 6.33.2, along with the instrument calibration file xti20240206. Additional event filtering was carried out using the psrpipe script from NICERSoft1 with similar criteria to previous analyses. Time intervals were removed when any of the following events occurred: 1) excessive optical loading, as indicated by undershoot rates exceeding 200 counts s; 2) high particle background exceeding 1.5 counts s per Focal Plane Module; 3) elevated solar activity, indicated by a Potsdam planetary geomagnetic index ; 4) space station passages through regions with a low cut-off rigidity, namely, COR_SAX GeV/c. A subsequent cut was made by excising any 16 s intervals with a 2–10 keV countrate above 1.0 counts s, yielding 1.09 Ms of effective exposure time.
Redistribution matrix (RMF) and effective area (ARF) files tailored to the final set of good-time intervals (GTIs) were constructed using the nicerl3 task in NICERDAS. Pulse phases were assigned to the final event list using the photonphase script in PINT (Luo et al. 2021) with a timing ephemeris obtained using Nançay Radio Telescope and Fermi Large Area Telescope data that covered the time span of the NICER observations (M. Kerr, personal communication).
In the analyses we discuss in this paper we used NICER pulse invariant (PI) channels 30–199 inclusive.
XMM-Newton EPIC data
We used an archival XMM-Newton European Photon Imaging Camera (EPIC) observation of PSR J0614−3329 acquired on 2010 October 4. The MOS1/2 and pn instruments were operated in Full Window, which does not offer sufficient time resolution to study the neutron star’s pulsations. Starting with the unprocessed ODF data products, we ran the event files through the emproc and epproc pipelines in SAS version 22.1.0. Next, we screened for instances of elevated background flaring by removing any 100-second segments where the count rate over the entire detector at keV exceeded 0.3 counts s for MOS1/2 and 0.4 counts s for pn. This data processing and filtering procedure resulted in final effective exposure times of 14.4 ks for MOS1/2 and 11.3 ks for pn. The recommended event quality parameter cuts were also made, with FLAG = 0 and PATTERN for MOS1/2 and PATTERN for pn.
The final source spectra were extracted from circular regions of radius 32″ centered on the radio pulsar position. Background estimates were obtained using source-free regions near the target. For MOS1/2 a circular region of radius 125″ on the central CCD at detector coordinates , was chosen whereas for pn a 60″ circular region on the same CCD as the source at detector coordinates , was used.
In the analyses we discuss in Sections A.1 and A.2, we used data from XMM-pn PI channels 57–399 inclusive, from XMM-MOS1 PI channels 60–399 inclusive, and from XMM-MOS2 PI channels 60–399 inclusive.
OVERVIEW OF METHODS
Our approach to modeling NICER and, where relevant, XMM-Newton data has been summarized in several previous papers ((Bogdanov et al. 2019b,a; Miller et al. 2019; Bogdanov et al. 2021; Miller et al. 2021; Salmi et al. 2022; Dittmann et al. 2024; Choudhury et al. 2024b; Miller et al. 2026)), which contain additional details, cross-code comparisons, and justifications. Other groups have used the same basic approach but with different ray-tracing codes, statistical samplers, and assumptions about the background ([16] (Riley et al. 2019; Riley et al. 2021; Salmi et al. 2024a; Vinciguerra et al. 2024; Mauviard et al. 2025; Qi et al. 2025; Kini et al. 2026)).
In brief, our basic assumptions are as follows:
Some regions of the neutron star’s surface are bombarded by highly relativistic particles, which have been accelerated within the star’s magnetosphere toward the star’s surface. This flux of relativistic particles deposits energy deep beneath the star’s surface and this energy heats the surface in those regions where it is being bombarded. As a result, these regions emit X-rays.
The locally measured spectrum and angular dependence of the X-ray emission from each element on the stellar surface depends on the temperature of the element, the chemical composition of the star’s atmosphere, and the local strength of the stellar magnetic field. To model this emission we use the emission from a fully ionized nonmagnetic hydrogen atmosphere computed by Ho & Lai (2001).
We model the pattern of heated regions on the stellar surface phenomenologically, using uniform-temperature circular or oval regions (“hot spots”). These hot spots have arbitrary locations and sizes and, in the case of oval spots, arbitrary aspect ratios and orientations. We number the spots, which are allowed to overlap. If a surface element is at a position that is covered by more than one spot, we assume it emits with the effective temperature of the lowest-numbered spot.
For ray tracing from the star to the observer we use the “oblate Schwarzschild approximation” ([4] (Morsink et al. 2007; Jakab & Morsink 2025)), in which the surface has, to a good approximation, the correct oblate shape produced by the star’s rotation but the actual external spacetime is approximated by the Schwarzschild spacetime.
We assume that counts in the NICER data that do not come from hot spots are not modulated at a frequency commensurate with the rotation of the neutron star and thus treat these counts by including a parameter for each NICER PI channel that represents the background counts in that channel and is independent of the star’s rotational phase and any background counts in other PI channels. This treatment is valid even if the background changes with time; the only assumption is that the background counts populate rotational phases uniformly. Ultimately, we marginalize over these background parameters, independently for each NICER PI channel. Unlike in PSR J0437−4715 (see Miller et al. (2026)), there is no evidence in the PSR J0614−3329 NICER or XMM-Newton data for a nonthermal component modulated at the star’s rotational frequency.
As in our previous analyses of pulsar waveforms, our inference is Bayesian, in that we explicitly specify the models and priors that we assume. As in Miller et al. (2026), our primary sampler is pocomc (Karamamanis et al. 2022). See Sections 3.7 and 3.8 of Miller et al. (2026) for more details about our statistical samplers and Figure 12 of Miller et al. (2026) for a comparison of the radius posterior for PSR J0437−4715 from pocoMC with the very similar radius posterior from the final sampling with emcee.
RESULTS OF OUR ANALYSIS OF NICER PULSE WAVEFORM DATA
We fit three classes of surface-emission models to the NICER data: two circular spots, two oval spots, and three circular spots. All models were run with progressively higher pocoMC resolution until the evidence converged. For our headline three-circle model, we performed additional sampling using emcee to refine the posterior. Table 1 lists our primary model parameters and their priors. The emcee posteriors samples from our three-circle model can be found on Zenodo.
| Parameter | Definition | Assumed Prior |
| Inverse of stellar compactness | ||
| Gravitational mass | ||
| (rad) | Colatitude of spot center | |
| (rad) | Spot half-extension | |
| (keV) | Spot effective temperature | |
| Spot elongation factor | flat from 0 to | |
| (rad) | Spot tilt angle | |
| (cycles) | Spot longitude difference | |
| (rad) | Observer inclination | |
| () | Neutral H column density | |
| (kpc) | Distance | |
| NICER effective area factor |
Table 1. Primary parameters of the pulse waveform models considered in this work.
Model comparison
Table 2 summarizes the Bayesian evidence for each model, together with the symmetric (about the median) 68% credible interval of the inferred radius. We see that the radius ranges overlap among the three models, but that the 3-circle model is clearly preferred, by .
| Model | (km) | |
| 2 circles | ||
| 2 ovals | ||
| 3 circles | 0 | 9.88-12.77 |
Table 2. Summary of the results of the analyses using our pulse waveform models.
Characteristics of the preferred model with three circular spots
Figure 1 shows the mass-radius posterior in our preferred three-circle fit to the NICER data, along with marginalizations of this posterior into mass and radius separately. The dashed lines in the one-dimensional plots show the priors. The mass posterior is essentially the same as the prior, but the radius posterior is strongly informed by the X-ray data.
Figure 1. Mass and radius posteriors (solid lines in the one-dimensional plots; color weighting in the two-dimensional plot), along with their one-dimensional priors (dashed lines), for our featured general three-circular-spot model fit to the NICER data. The mass posterior is clearly determined almost entirely by the prior, but the radius posterior is strongly informed by the X-ray data.
Figure 2 displays the best-fitting parameter combination from our 3-circle model to the NICER data. Note that other spot configurations also provide good fits to the data; in particular, there are some configurations with a single much larger spot ( radian). However, large-spot configurations are strongly disfavored compared with the spot pattern shown in Figure 2.
Figure 2. The hot-spot geometry in our maximum-likelihood 3-circle fit to the NICER data. The black line marks the inferred (dominated by the prior from pulse timing) line-of-sight inclination at which we view the pulsar. Solid lines depict parts of the stellar surface with surface normal vectors pointing out of the page, whereas dashed lines are used to plot parts of the surface with surface normal vectors pointing into the page. The spot colors are simply qualitative labels. Other spot configurations also fit the data well (see Appendix § 9 for some examples), including ones with a large spot ( radian), but large-spot configurations are disfavored by compared with the combination shown here. Spot 1 is colored olive; spot 2, the hottest, is colored blue; and spot 3, the coolest, is colored red.
Non-antipodal emitting-region geometries are favored throughout most of the posterior probability. This is consistent with a surface magnetic field configuration more complex than a centered dipole, possibly including the presence of multipolar components (see, e.g., [8, 15] (Kalapotharakos et al. 2021)). Physically self-consistent multiwavelength modeling, in which X-ray, -ray, and radio observables constrain a common magnetic-field and viewing geometry, would provide complementary information on the geometry.
Goodness of fit checks
The maximum-likelihood 3-circle fit to the NICER data has a phase-channel . The probability of this or larger for a model with this number of degrees of freedom is 83%. The bolometric (probability 50%). Neither probability is low enough to raise concerns, so although this does not prove that it is the right model, it passes these tests. See Appendix § 9 for the full posteriors for our fits with three circular spots to the NICER data.
Figure 3 shows the phase-channel residuals for the maximum-likelihood 3-circle model, defined as
Figure 3. Phase-channel residuals to NICER data from our maximum-likelihood three-circle fit to those data. The color indicates the value of for data and model . NICER PI channel increases upward, whereas the rotational phase bin increases to the right. The lack of any pattern in the residuals, along with the good total indicate that the fit is adequate based on this test.
note that the data and model are both represented in counts, so is dimensionless. Similarly, Figure 4 shows the bolometric (i.e., channel-summed) version of Figure 3. No patterns or notable outliers are evident in either residuals plot.
Figure 4. Top panel: bolometric counts from the NICER data (histogram) and the NICER predictions from the best-fitting three-circle model (orange curve). Bottom panel: for the bolometric data, where is the number of observed NICER counts and is the predicted number for each of the 32 rotational phase bins. Again, no patterns are evident and there are no notable outliers, suggesting that the model passes this test of goodness of fit.
IMPLICATIONS FOR THE EOS
The majority of the properties of neutron stars that we can measure depend only on the equation of state (EOS) of the matter in them. (A notable exception is neutron star cooling rates, which can be used to obtain information about the microscopic properties of the matter in **Note—**Main parameters and priors, where the priors on mass and observer inclination come from Miles et al. (2025) and the distance prior comes from Bassa et al. [5]. The spot parameters describe an oval spot; circular spots have elongation factor 1 and tilt angle 0. The spot parameters are selected independently from each other. By definition the longitude of the center of spot 1 is 0, and the spot longitudes of other spots are with respect to spot 1 (note that after the spots have been constructed, the overall phase of the waveform is marginalized with respect to the data).
**Note—**Log evidence relative to our favored model (in boldface), and posterior radius range, for our three models.
their interiors; see Heinke 2025 for a recent review.) The EOS of a given material is the pressure in the material as a function of intensive quantities, such as the density, temperature, and composition. The thermal temperatures of the matter in the interiors of the neutron stars observed using NICER are thought to be K, much lower than the Fermi temperatures of the matter in their cores, which are K. Hence the thermal pressure in the interiors of these neutron stars does not significantly affect their macroscopic properties.
It is also expected that the matter in the cores of the neutron stars observed using NICER, where the density
is comparable to or greater than the saturation density of nuclear matter, is in chemical equilibrium. Therefore the pressure in this region can be expressed as a function of a single quantity, such as the energy density or the baryon number density , and this equation of state or is a universal function for cold, equilibrated matter at densities above the density of nuclear saturation. Thus, although neutron stars with different masses will have different densities in their cores, it is expected that at a given or the pressure will be the same in any neutron star.
If a neutron star is not rotating, its central density and the EOS of the matter in it determine its mass and structure. Hence, once we know the EOS of neutron-star matter, we can compute the structure of a nonrotating star with a given central density using the relevant structure equations. In particular, we can determine the maximum mass of a nonrotating star composed of matter with this EOS. We can also compute the radius, tidal deformability, moment of inertia, and other macroscopic properties of a star of any mass. Rotation alters the star’s structure, but for the neutron stars that have been observed using NICER, all of which have rotation frequencies Hz, the effect of rotation on these properties is much smaller than the measurement uncertainties (see, e.g., Figure 13 of Miller et al. (2019)). Our procedure for inferring the EOS of neutron-star matter follows that of Miller et al. (2020). We start with a set of EOSs, which we assume all have the same prior probability. For a given EOS, we calculate the likelihood of each of the independent observational data sets, given that EOS, and then multiply their likelihoods to obtain an overall likelihood of the data given that EOS. We then repeat this procedure for each EOS. This gives the posterior probability distribution for the trial EOSs. Using this new, reweighted distribution of EOSs, we derive quantities such as the posterior probability distribution of the maximum mass, the radius at a given mass, or anything else that is predicted by the EOS.
In more detail, we assume that we know the EOS from zero density to some threshold density, and then use a framework to extrapolate to higher densities. We assume that we know the EOS to half nuclear saturation density, because this is approximately where the crust transitions to the core (Hebeler et al. (2013)). In previous work we used the QHC19 EOS (Baym et al. [6]) up to half of saturation density. In contrast, here our EOS below half of saturation density is the more microscopically realistic QHC21 EOS (Kojo et al. (2022)), which was inspired in part by NICER analyses of the pulsar PSR J0740+6620 and which uses the chiral effective field theory EOS (Lonardoni et al. (2020); ); Drischler et al. [20]) below 1.5 times saturation density. The QHC19 and QHC21 low-density priors give nearly indistinguishable EOS posteriors after neutron star observations are included.
We then need to decide on our EOS extrapolation framework. Because of its flexibility, and for continuity with previous papers, we use the Gaussian processes framework described in Miller et al. (2021), and in fact use the same equations of state above as in previous papers (Miller et al. 2021; Dittmann et al. [19]; Miller et al. 2026) so that any differences come from added data (and as a minor contribution, the assumptions about the EOS below ) rather than from the EOS framework.
The final decision to make is the data sets to include in our analyses. As in Miller et al. (2026) we use the updated masses of PSR J1614−2230 (; see Agazie et al. [3]) and PSR J0348+0432 (; see Saffer et al. 2025); the tidal deformability from the double neutron star merger GW170817 (Abbott et al. [1]) and the probable double neutron star merger GW190425 (Abbott et al. 2020); and the NICER-based (sometimes including XMM-Newton data) mass and radius estimates of PSR J0030+0451 (Miller et al. 2019), PSR J0740+6620 (Dittmann et al. [19]), and PSR J0437−4715 (Miller et al. 2026), to which we add the measurement of PSR J0614−3329 in the current paper.
For the mass-radius and mass-deformability data sets we used kernel density estimation with a Gaussian kernel that has 0.1 times the bandwidth recommended by Silverman (1986); see Section 5.1 of Miller et al. (2021) for details. We also included a constraint on the nuclear symmetry energy at saturation density (defined as the energy per nucleon of pure neutron matter minus the energy per nucleon of matter with an equal number of neutrons and protons) of MeV (Tsang et al. 2012), where we imposed beta equilibrium including muons using the approach in Section II of Blaschke et al. [9]. Finally, we explored the consequences of including the high estimated mass of from PSR J0952−0607 (Romani et al. 2026; note that this pulsar has a high rotation rate of 707 Hz and that Romani et al. 2026 find that the zero-rotation mass equivalent is , which is what we use).
As emphasized by Romani et al. (2026), although the best-estimate mass for this pulsar is high and therefore constraining, concerns about systematics in the modeling of PSR J0952−0607 and similar “spider” pulsars suggests caution in using them in EOS constraints.
Given these assumptions and data sets, the main results are in Table 3 (which shows the symmetric 68% credible ranges for the maximum stable mass of a non-rotating neutron star and the radius of a fiducial 1.4 neutron star) and Figure 5, which shows the symmetric 90% credible range of the pressure as a function of baryon number density, in both cases emphasizing the effect of the new measurement.
| Quantity | Data set | median | ||
| Miller et al. (2026) | 2.07 | 2.20 | 2.44 | |
| This work | 2.02 | 2.15 | 2.38 | |
| PSR J0952−0607 | 2.26 | 2.41 | 2.59 | |
| (km) | Miller et al. (2026) | 12.09 | 12.56 | 13.03 |
| (km) | This work | 11.91 | 12.44 | 12.88 |
| (km) | PSR J0952−0607 | 11.88 | 12.38 | 12.82 |
Table 3. Maximum Masses and Fiducial Radii Including the PSR J0614−3329 Measurements Reported Here
Figure 5. Posterior range in log pressure as a function of log baryon number density in units of the saturation density , for our QHC21 prior (black dotted line; the grey dotted line shows the QHC19 prior), our posterior using the QHC21 prior with all information other than our PSR J0614−3329 measurement (blue dashed line), and our posterior with all information including our PSR J0614−3329 measurement (red solid line). For each line type, at a given number density the lower curve shows the 5th percentile of the pressure, and the upper curve shows the 95th percentile of the pressure. We do not include constraints based on the mass of PSR J0952−0607 from Romani et al. (2026). The posteriors (not shown) that use the QHC19 low-density prior are essentially indistinguishable from the posteriors that we display here. We see that our results for PSR J0614−3329 are broadly consistent with our previous analyses, as indicated by the small change that inclusion of this measurement makes to the pressure posterior. See the main text for additional details.
Note— Median and values of the maximum mass of a non-rotating neutron star (where TOV refers to the papers Tolman (1939) and Oppenheimer & Volkoff (1939)) and the radius of a neutron star with a fiducial mass 1.4 , from our analysis prior to our PSR J0614−3329 measurement, then including that measurement, and then including the mass measurement of PSR J0952−0607 by Romani et al. (2026), spin-corrected to (see text for details). We see that the addition of our comparatively small measured radius for PSR J0614−3329 decreases the estimated maximum mass and fiducial radius slightly, and that further inclusion of the estimated mass of PSR J0952−0607 increases the quantiles for the maximum mass substantially but only lowers the quantiles for the fiducial radius by a small amount.
CONCLUSIONS
Our analysis of NICER data on the pulsar PSR J0614−3329 yields an estimate of its radius consistent with previous NICER-based radius measurements of the pulsars PSR J0030 and PSR J0437, though centered at a smaller radius. The constraints we have obtained here on the EOS of matter at a few times nuclear saturation density, when combined with the constraints obtained from other analyses of astronomical data, including previous analyses of NICER data, continue to refine our understanding of the EOS of neutron-star matter, thereby further improving our understanding of the properties of cold, equilibrated matter at very high densities.
ACKNOWLEDGMENTS
The authors thank Gordon Baym and Toru Kojo for illuminating discussions regarding the equation of state of dense matter. A.J.D. and M.C.M. were supported in part by NASA ADAP grant No. 80NSSC21K0649. M.C.M. and C.C. were also supported in part by NASA grants 80NSSC25K7108, 80NSSC25K0511, and
80NSSC26K0667. Part of this work was performed at the Aspen Center for Physics, which is supported by National Science Foundation grant PHY-2210452. We are grateful to the Institute for Nuclear Theory at the University of Washington for its kind hospitality and stimulating research environment; this research was therefore supported in part by the INT’s U.S. Department of Energy grant No. DE-FG02-00ER41132. A.J.D. was supported by NASA through the Hubble Fellowship Program grant No. HST-HF2-51553.001, awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA, under contract NAS5-26555. J.B. was supported by UKRI FLF grant number MR/Y018257/1 (PI: Graber). S.B. was supported in part by NASA grants 80NSSC20K0275, 80NSSC22K0728, and 80NSSC25K7544. The research of S.M.M. is supported by NSERC Discovery Grant RGPIN-2026-05707. C.C. and Z.W. were supported by NASA under award number 80GSFC24M0006. C.K. was partly supported by NASA grants 22-ADAP22-0142 and 22-TCAN22-0027. Some of the resources used in this work were provided by the NASA High-End Computing (HEC) Program through the NASA Center for Climate Simulation (NCCS) at Goddard Space Flight Center. The authors acknowledge the University of Maryland supercomputing resources (http://hpcc.umd.edu) that were made available for conducting the research reported in this paper. Portions of this work performed at NRL were supported by NASA. This research has made use of data products and software provided by the High Energy Astrophysics Science Archive Research Center (HEASARC), which is a service of the Astrophysics Science Division at NASA/GSFC and the High Energy Astrophysics Division of the Smithsonian Astrophysical Observatory. We acknowledge extensive use of NASA’s Astrophysics Data System (ADS) bibliographic services and arXiv.
Facility: NICER ((Gendreau et al. 2016))
Software: emcee ((Foreman-Mackey et al. 2013)), Python ((Oliphant 2007)), NumPy ((Harris et al. 2020)), Matplotlib ((Hunter 2007)), PGF/TikZ ((Tantau 2013)), Cython ([7]), schwimmbad ((Price-Whelan & Foreman-Mackey 2017)), HEASoft ((Nasa High Energy Astrophysics Science Archive Research Center (Heasarc) 2014)), PINT ((Luo et al. 2021)), and pocoMC ((Karamian et al. 2022)).
Joint Fits with XMM-Newton data
Joint analyses
Our approach to joint fits of NICER and XMM-Newton data is described in detail in Miller et al. (2021) and Dittmann et al. [19]. To the parameters listed in Table 1 we add effective area factors for the XMM-pn, XMM-MOS1, and XMM-MOS2 data. For comparison with Mauviard et al. (2025), our priors for these parameters are all independent normal distributions with means of 1 and standard deviations of 0.104 (i.e., the same as our prior on the NICER effective area factor), although in reality there are likely to be much tighter correlations between the effective area parameters of the XMM-Newton instruments.
Because the time resolution and effective area of XMM-Newton are insufficient to yield useful phase-resolved data for PSR J0614−3329, we need to treat the XMM-Newton background in a different way from the NICER background. Our standard assumption, which is consistent with the assumption used by all groups in published joint NICER-XMM analyses, is that the only non-spot contribution to the XMM-Newton data is the measured background around PSR J0614−3329 (see § 2.2). We further assume that the counts measured in the background represent Poisson samples from the expected background. Thus if in some XMM-Newton PI channel there are measured background counts (where is an integer), the prior probability that the average background during that observation is is given by . Using this prior probability distribution for the background, and adjusting for the relative extraction areas on the sky and instrument responses for the source and background regions, we marginalize over the background after multiplying by the likelihood of the XMM-Newton data given the spot model and the background.
However, when we use this assumption there is tension between the NICER and the XMM-Newton data, in the sense that the best joint fit underestimates the XMM-Newton counts. This is evident in Figure 6, which shows the cumulative difference between the best joint-fit model and the data in each of the XMM-Newton instruments. The XMM-MOS2, XMM-MOS1, and XMM-pn models are, respectively, , , and low compared with the data. The NICER-only fits have an even greater deficit relative to the XMM-Newton data: for example, our maximum-likelihood NICER-only three-circle model has an XMM-pn deficit
Figure 6. Comparison between the counts predicted by the best joint-fit model and the data for the three XMM-Newton instruments, as function of PI channel. The top panel displays the normalized residual in each PI channel. The lower panels display the cumulative sum, starting at the minimum PI channel for each instrument ( for XMM-pn and for XMM-MOS1 and XMM-MOS2), of the difference between the model and data. The second panel from the top shows the cumulative differences for XMM-pn, XMM-MOS1, and XMM-MOS2. The remaining three panels plot the cumulative differences for each instrument separately, combined with the (darkest region), , and (lightest region) ranges expected from Poisson statistics if the model were correct. We see that in all three instruments, most notably for XMM-pn, the predictions in the best joint-fit model fall short of the data.
of 208.934 counts relative to the data, an XMM-MOS1 deficit of 72.059 counts relative to the data, and an XMM-MOS2 deficit of 53.263 counts relative to the data.
Although Figure 6 only shows the results for the best fit to the joint NICER+XMM data sets, we have performed the same exercise for a set of 100 samples drawn from the posterior. Out of these, no model overpredicted the counts in any XMM-Newton instrument: the total model XMM-pn counts ranged from 444 to 500, versus 553 observed counts; the total model XMM-MOS1 counts ranged from 129 to 160, versus 165 observed counts; and the total model XMM-MOS2 counts ranged from 120 to 147, versus 148 observed counts. We caution that these posterior samples were chosen by finding combinations with high NICER+XMM likelihoods rather than reproducing the total counts in a given instrument. Nonetheless, the systematic underestimation of the XMM-Newton data could suggest that there is additional background in the XMM-Newton data beyond what is captured in fields near PSR J0614−3329.
The consequence is that in NICER-XMM joint fits without additional background, the fit assigns more of the unpulsed emission to the spots, which requires a more compact star to increase gravitational lensing and thus (given the well-constrained mass) requires a smaller radius. This is why, using standard assumptions about XMM-Newton backgrounds, the inferred radius drops from 12.765 km in our NICER-only analysis to 11.88 km when XMM-Newton data are incorporated under the assumption of known background.
We could instead use the approach of Holt et al. (2025), in which the XMM-Newton background is allowed to include an additional component. In Table 4 we compare the resulting radius ranges between our standard background assumption (i.e., no additional background) and the case treated by Holt et al. (2025), which allows for additional background up to the known background. We see that the assumption about the XMM-Newton background makes a large difference to the inferred radius.
| Background | (km) |
| Standard | 9.52 – 11.88 |
| Additional | 11.60 – 14.76 |
| NICER-only | 9.88 – 12.77 |
Table 4. Comparison of three-circle NICER+XMM joint fit results using only the known XMM-Newton background versus allowing for additional background as in Holt et al. (2025). We also show our headline NICER-only result for comparison.
Thus our standard background assumption leads to a discrepancy with the data, and the addition of unknown background changes the radius posterior significantly. For these reasons, our headline result comes from our analysis of just the NICER data.
Comparison with 2025ApJ...995...60M
The analysis of PSR J0614−3329 by Mauviard et al. (2025) was a joint analysis of NICER and XMM-Newton observations. Their headline result was a symmetric 68% credible range in the radius of km. This result used the “ST+PDT” model in the publicly available X-PSI package (Riley et al. 2023). In this model, one spot (ST, for single temperature) is a uniform-temperature circle, whereas the other (PDT, for protruding double temperature) is two uniform-temperature circles on top of each other with at least a point of overlap, and pixels in the overlap region emit with the effective temperature of the “supersized” circle. No parts of the ST and PDT spots can overlap each other, and all three circles must be less than or equal to in angular radius. Thus ST+PDT is a subset of our three-circle model.
Mauviard et al. (2025) use the same NICER data that we do, but an older extraction of the XMM-Newton data. Thus for our comparison tests, we use the same XMM-Newton source and background data as in Mauviard et al. (2025).
Mauviard et al. (2025) use the same approach to the NICER background that we do, but treat the XMM-Newton background differently. In our analysis as well as that of Mauviard et al. (2025), the background data are counts in the region around the PSR J0614−3329 pointing, and take the form of counts in the PI channels of each of the three XMM-Newton instruments. Both groups estimate the probability distribution of the long-term average number of counts based on the observed counts, and both groups then apply a scaling factor between the background data and the PSR J0614−3329 pointing. The difference lies in the translation between observed background counts and the probability distribution of the long-term average number of background counts.
Suppose that in a given XMM-Newton PI channel , there are counts in the background observation. The task is to determine the prior distribution of the expected number , that is, to determine . Then the two approaches are:
As we described in Section A.1, we assume that the counts are drawn from a Poisson distribution, so that .
Mauviard et al. (2025) follow Riley et al. (2021) in their approach. The prior on is flat between and . If but then the prior is flat between 0 and . If then is set to the next nonzero background count number going upward in PI channel and the prior on is determined using this new .
The prior from either approach is then multiplied by the likelihood of the data, in channel , in the pointed observation given the assumed background and the counts contributed by the spot model under consideration. This posterior is then marginalized over the background to get the factor contributed by that channel in that data set for the given spot model.
To compare our results with those of Mauviard et al. (2025) we fit the data using the ST+PDT model with the same priors as in Mauviard et al. (2025) and use their treatment of the XMM-Newton background. Using pocoMC, with repeated runs with higher precision until apparent convergence, we find km, compared with km from Mauviard et al. (2025). Thus for the same data, model, and background treatment our symmetric 68% credible region is 28% broader than that of Mauviard et al. (2025). Based on previous comparisons [13, 18] (Nelson et al. 2020; Buchner 2023; Lemos et al. 2023), we suspect that this is because the MultiNest precision parameters used by Mauviard et al. (2025) (in this case, live points and a sampling efficiency of 0.05) are insufficient for convergence.
Posterior Distributions
Table 5 lists the median, , and points in the posterior distributions obtained by fitting our models to only the NICER data, assuming a fully ionized hydrogen atmosphere, and the resulting maximum likelihood values for each of the parameters in these models. We display the complete corner plot of the posteriors from these same analyses in Figure 7. Although the radius posterior is unimodal, many of the spot geometry posteriors are multimodal. We illustrate in Figure 8 the best-fitting spot configurations from three other modes, to complement that displayed in Figure 2.
Figure 7. Posterior probability density distributions from our analysis of the NICER data on PSR J0614−3329, where the units are the same as in Table 5. This corner plot displays results for our featured model, in which we have three uniform-temperature circular spots with arbitrary overlap. Dotted traces in the one-dimensional distributions depict prior information.
| Parameter | median | Maximum Likelihood | ||||
| 11.157 | 9.883 | 12.765 | 8.869 | 14.809 | 11.910 | |
| 0.190 | 0.166 | 0.214 | 0.144 | 0.236 | 0.185 | |
| 1.436 | 1.366 | 1.508 | 1.295 | 1.578 | 1.494 | |
| 1.541 | 1.537 | 1.545 | 1.533 | 1.550 | 1.542 | |
| 0.017 | -0.484 | 0.511 | -0.917 | 0.920 | -0.057 | |
| 0.098 | 0.035 | 0.136 | 0.017 | 0.182 | 0.095 | |
| 0.087 | 0.078 | 0.122 | 0.073 | 0.149 | 0.083 | |
| -0.007 | -0.576 | 0.564 | -0.950 | 0.951 | -0.058 | |
| 0.099 | 0.033 | 0.155 | 0.016 | 0.383 | 0.021 | |
| 0.085 | 0.060 | 0.125 | 0.038 | 0.150 | 0.132 | |
| 0.466 | 0.301 | 0.548 | 0.215 | 0.582 | 0.553 | |
| -0.013 | -0.615 | 0.595 | -0.937 | 0.933 | 0.209 | |
| 0.123 | 0.034 | 1.629 | 0.016 | 2.427 | 0.098 | |
| 0.063 | 0.035 | 0.120 | 0.026 | 0.146 | 0.062 | |
| 0.547 | 0.286 | 0.743 | 0.129 | 0.785 | 0.299 | |
| 0.597 | 0.167 | 1.441 | 0.023 | 2.639 | 0.034 | |
| 0.605 | 0.560 | 0.650 | 0.515 | 0.696 | 0.620 | |
| 0.981 | 0.875 | 1.085 | 0.772 | 1.189 | 1.029 |
Table 5. Three-circle fit to NICER Data on PSR J0614−3329. Note—A comparison of the , , median, , and , and maximum likelihood values inferred from our analysis of NICER data for PSR J0614−3329. Here we use our featured model, which has three uniform-temperature circular hot spots plus a phase-independent background.
Figure 8. Best-fitting spot configurations within different modes (see Figure 7), where we have excluded the mode in which the global best fit resides (see Figure 2) and ignored the subdominant modes that include very large spots. In each case the first, second, and third spots are colored olive, blue, and red respectively. As before, solid lines are used to plot parts of the stellar surface with surface normal vectors pointing out of the page, while dashed lines are used to plot parts of the surface with surface normal vectors pointing into the page.
References
- [1]Abbott, B. P., Abbott, R., Abbott, T. D., et al. 2017, Physical Review Letters, 119, 161101, doi: 10.1103/PhysRevLett.119.161101DOI
- [3]Agazie, G., Alam, M. F., Anumarlapudi, A., et al. 2023, ApJL, 951, L9, doi: 10.3847/2041-8213/acda9aDOI
- [4]AlGendy, M., & Morsink, S. M. 2014, ApJ, 791, 78, doi: 10.1088/0004-637X/791/2/78DOI
- [5]Bassa, C. G., Antoniadis, J., Camilo, F., et al. 2016, MNRAS, 455, 3806, doi: 10.1093/mnras/stv2607DOI
- [6]Baym, G., Furusawa, S., Hatsuda, T., Kojo, T., & Togashi, H. 2019, ApJ, 885, 42, doi: 10.3847/1538-4357/ab441eDOI
- [7]Behnel, S., Bradshaw, R., Citro, C., et al. 2011, Computing in Science & Engineering, 13, 31, doi: 10.1109/MCSE.2010.118DOI
- [8]Bilous, A. V., Watts, A. L., Harding, A. K., et al. 2019, ApJL, 887, L23, doi: 10.3847/2041-8213/ab53e7DOI
- [9]Blaschke, D., Alvarez-Castillo, D. E., & Klähn, T. 2016, arXiv e-prints, arXiv:1604.08575, doi: 10.48550/arXiv.1604.08575DOI
- [13]Buchner, J. 2016, Statistics and Computing, 26, 383, doi: 10.1007/s11222-014-9512-y
- [15]Chen, A. Y., Yuan, Y., & Vasilopoulos, G. 2020, ApJL, 893, L38, doi: 10.3847/2041-8213/ab85c5DOI
- [16]Choudhury, D., Salmi, T., Vinciguerra, S., et al. 2024a, ApJL, 971, L20, doi: 10.3847/2041-8213/ad5a6fDOI
- [18]Dittmann, A. 2024, The Open Journal of Astrophysics, 7, 79, doi: 10.3233/001c.123872
- [19]Dittmann, A. J., Miller, M. C., Lamb, F. K., et al. 2024, ApJ, 974, 295, doi: 10.3847/1538-4357/ad5f1eDOI
- [20]Drischler, C., Han, S., Lattimer, J. M., et al. 2021a, PhRvC, 103, 045808, doi: 10.1103/PhysRevC.103.045808DOI







