WorksPhysics and Astronomy
Gamma-Ray Emission from LS 5039: A 30-500 GeV Excess and Phase-Resolved Broadband Spectral Modeling
Abstract
LS~5039 is a well-studied -ray binary exhibiting strong orbital modulation from GeV to TeV energies. We present a phase-resolved analysis of 17.3~yr of Fermi-LAT observations and identify an additional -ray excess in the –~GeV at a significance of . The phase-resolved spectra show pronounced orbital dependence above ~GeV, with the inferior-conjunction more complex than the superior-conjunction. To investigate the broadband spectral energy distribution, we combine the Fermi-LAT spectra with published H.E.S.S.\ and HAWC measurements and compare purely leptonic and lepto-hadronic models using the Bayesian information criterion. In both intervals, relativistic-Maxwellian and magnetospheric components dominate below ~GeV, while shock-accelerated electrons account for much of the GeV–TeV continuum. The purely leptonic fits leave residuals at –~GeV and underpredict the HAWC measurements above ~TeV. In the local one-zone intrabinary shock apex model, synchrotron cooling and Klein-Nishina suppression make the ultra-high-energy emission difficult to reproduce, although more complex leptonic scenarios remain possible. Adding a phenomenological proton component improves the broadband description, with stronger relative support during superior-conjunction than inferior-conjunction. However, the proton parameters remain weakly constrained and the fast-escape hadronic interpretation is energetically challenging, so a hadronic origin is not uniquely established. Future high-sensitivity, phase-resolved observations at multi-TeV and ultra-high energies will be required to test this interpretation.
Introduction
Gamma-ray binaries are compact binary systems composed of a massive star and a compact object, in which most of the nonthermal radiative output is emitted at -ray energies, above MeV. Their high-energy radiation is generally attributed to particles accelerated in relativistic outflows or wind-interaction regions associated with the compact object and its massive companion. The observed emission can involve synchrotron radiation, anisotropic inverse-Compton (IC) scattering, photon–photon pair production () and the resulting absorption and pair cascading, as well as potentially hadronic processes ([25]. Orbital motion periodically changes the viewing geometry, stellar-photon and wind densities, and opacity. These systems therefore provide unique laboratories for studying extreme particle acceleration and radiative processes under periodically varying physical conditions.
The origin of the high-energy emission from -ray binaries remains debated. In the microquasar scenario, accretion powers relativistic jets that can produce leptonic and hadronic emission ([45, 13, 23, 63]). In the pulsar-wind scenario, a rotation-powered neutron-star wind collides with the stellar wind, forming an intrabinary shock (IBS) where particles can be accelerated ([56, 31, 53]). Shock-accelerated electrons can produce synchrotron and IC emission, while relativistic protons may generate -rays through interactions. Magnetospheric outer-gap emission ([64, 19, 55, 15]) and other pulsar-wind processes, including free-wind emission, pair cascades, and striped-wind dissipation ([18 J. Pétri & G.] Dubus 2011), may also contribute. Their different emission sites and phase dependences make phase-resolved observations essential for distinguishing these scenarios.
To date, only a handful of -ray binaries have been firmly identified in the Galaxy ([3, 5, 32, 21]). LS 5039 is among the most prominent and best-characterized members of this class. It consists of an O6.5 V((f)) star with a mass of and a compact companion in an eccentric orbit with a period of d and an eccentricity of ([17, 43]). Although the nature of the compact object remains uncertain, the reported detection of a s periodic signal in hard X-rays provides support for a neutron-star interpretation ([59]). In this work, we adopt the neutron-star and pulsar-wind scenario as a working assumption and investigate its implications for the high-energy emission. With its well-determined orbital ephemeris, extensive multiwavelength coverage, and recently reported ultra-high-energy (UHE) measurements, LS 5039 provides an exceptional laboratory for testing particle-acceleration and radiation models in -ray binaries. In particular, the available GeV–UHE observations span approximately five decades in energy, enabling phase-resolved tests of the underlying particle populations and emission mechanisms.
LS 5039 exhibits nonthermal emission from radio to TeV energies, with clear orbital modulation observed across multiple wavebands ([47, 2, 17, 1, 54, 20]). H.E.S.S. observations above 0.1 TeV revealed a hard TeV spectrum with strong phase-dependent behavior, whereas Fermi-LAT observations revealed a bright GeV component with a spectral cutoff at a few GeV and a distinct modulation pattern ([3, 1]). This GeV–TeV dichotomy suggests that the observed emission cannot be described by a single simple radiative component and may require multiple particle populations, emission zones, or phase-dependent acceleration and cooling processes. The transition region between the GeV and TeV components is therefore particularly important for understanding the radiation mechanisms in LS 5039.
Recent HAWC observations have further extended the spectrum of LS 5039 into the UHE regime, with -rays detected up to TeV ([6]). These measurements provide important additional constraints on broadband emission models. In the IBS environment, radiative cooling and the Klein–Nishina suppression of IC scattering can strongly limit the highest-energy IC emission from relativistic electrons. The UHE observations therefore motivate testing whether the GeV–TeV transition and the UHE spectral extension can be described within a purely leptonic framework, or whether including a hadronic contribution provides a better phenomenological description of the UHE emission.
In this work, we analyze 17.3 yr of Fermi-LAT observations of LS 5039, focusing on the high-energy end of the LAT spectrum, the high-energy excess around 30–500 GeV, and their orbital-phase dependence. We combine the phase-resolved LAT spectra with published H.E.S.S. and HAWC measurements to model the broadband spectral energy distribution (SED) within an IBS framework and compare purely leptonic and lepto-hadronic scenarios. § 2 describes the Fermi-LAT analysis, § 3 and § 4 present the physical model and fitting procedure, respectively, § 5 gives the results and discussion, and § 6 summarizes our conclusions.
Fermi Observation and Data Reduction
Data Reduction
We analyzed -ray events collected by the Fermi-LAT from 2008 August 4 to 2025 November 25 (MJD 54682.687–61004.000). The events were selected within a region of interest (ROI) centered on 4FGL J1826.2−1450 (LS 5039) (R.A. = , decl. = ). Our data analysis based on the standard Pass 8 data-selection criteria with SOURCE-class Front+Back events (evclass=128, evtype=3) in the energy range from 0.1 to 500.0 GeV. To reduce the -ray contamination from the Earth Limb, we applied a maximum zenith-angle cut of . And the good time intervals were selected using the standard filter expression DATA_QUAL>0 && LAT_CONFIG==1. The data reduction and likelihood analysis were carried out with Fermitools version of 2.2.0 with the instrument response functions of P8R3_SOURCE_V3. The Galactic and extragalactic diffuse -ray emissions were modeled using gll_iem_v07.fits and the isotropic diffuse template iso_P8R3_SOURCE_V3_v1.txt.
Spectral Analysis
Based on a model file generated by make4FGLxml.py1 that describes all -ray sources in 4FGL-DR4 within the ROI, we performed a binned maximum likelihood spectral analysis. In the file, the target source LS 5039 **Note—**Best-fit spectral parameters obtained from the 0.1–500.0 GeV spectral fits. The phase-averaged spectrum was fitted with the LP, PLEC4, and BPL models, whereas the SUPC and INFC spectra were fitted with the BPL model. The characteristic energy is model dependent: is the pivot energy for LP and the break energy for BPL, whereas is the equivalent cutoff energy for PLEC4. Numbers in parentheses denote the statistical uncertainties in the last quoted digits, with uncertainties rounded to two significant figures. Parameters without parenthetical uncertainties were fixed during the fit. TS values should be compared only among fits using the same data selection.
**Note—**Parameters 1 and 2 and characteristic energy 1 describe the BPL component, while Parameters 3 and 4 and characteristic energy 2 describe the additional LP component. The LP reference energy was fixed at 100 GeV. Numbers in parentheses denote the statistical uncertainties in the last quoted digits, with uncertainties rounded to two significant figures. and denote the component-wise source-detection TS values reported for the BPL and additional LP components, respectively; they should not be confused with the likelihood-ratio obtained by comparing the complete BPL and BPL+LP models.
has a log-parabola spectral shape (LP). The spectral shapes of other point sources were taken from the 4FGL-DR4, and the spectral parameters for sources located more than from the ROI center were fixed to their catalog values, while others were set free. We tested other two alternative spectral models, a power law with a super-exponential cutoff (PLEC4) and a broken power law (BPL), to describe the -ray emissions from LS 5039. Table 1 summarizes the best-fit parameters and source-detection TS values for the three spectral models. Because LP, PLEC4, and BPL have different numbers of free parameters and are not all mutually nested, we compare them using the Akaike information criterion (AIC; [4]), AIC . The BPL model
| Model | Parameter 1 | Parameter 2 | Characteristic energy (GeV) | TS | AIC | |
| LP | 32956.0 | -44716260.45 | -89432078.90 | |||
| PLEC4 | 33031.2 | -44716259.44 | -89432076.88 | |||
| BPL | 33125.2 | -44716289.64 | -89432135.28 | |||
| BPL | 23611.2 | -19011191.34 | – | |||
| BPL | 10111.2 | -13747005.83 | – |
Table 1. Best-fit spectral parameters for the phase-averaged and phase-resolved spectra
has the lowest AIC and is therefore adopted for the subsequent phase-resolved analysis. To characterize the -ray SED of LS 5039, we divided the 0.1–500.0 GeV energy range into 15 equally logarithmically spaced energy bins and performed a binned maximum-likelihood analysis independently in each bin. In each fit, LS 5039 was described by the best-fit BPL model obtained from the above analysis, with its spectral-shape parameters fixed. The normalizations of LS 5039 and of neighboring sources within of the ROI center were left free, while the parameters of all other sources were fixed at their best-fit values. The resulting SED is shown in Figure 1. At lower energies, our measurements are broadly consistent with those reported in previous studies (e.g., (e.g., H. Yoneda et al. 2021). However, the SED exhibits a more complex spectral structure at higher energies. A high-energy excess is detected in the – GeV relative to the extrapolation of the lower-energy spectral trend. We quantify the significance of the additional component by comparing the BPL and BPL+LP fits over 0.1–500.0 GeV, using identical event selections and background treatments. For the additional LP component, the reference energy is fixed at 100 GeV, while its normalization, spectral index, and curvature are allowed to vary freely. We define .
Figure 1. SED of LS 5039 in the 0.1–500.0 GeV range. Filled pentagons show the measurements obtained in this work, with the green points highlighting the high-energy excess at – GeV. The black dashed and green dashed curves represent the BPL and additional LP components, respectively, while the black solid curve shows their combined BPL+LP model. The blue dashed step curve shows the test statistic (TS) in each energy bin, with values given by the right-hand -axis. All SED data points satisfy TS 4. For comparison, the SED reported by H. Yoneda et al. [58] is shown as open gray circles.
Adopting an asymptotic approximation for the three additional parameters2, we obtain for the phase-averaged spectrum, corresponding to a significance of with . No additional scans or repeated testing were performed, hence the trials factor is taken to be unity.
To investigate whether the detected high-energy excess is associated with LS 5039, we constructed a TS map using events with energies above 30 GeV, as shown in Figure 2. The significant -ray excess in the TS map is spatially coincident with the nominal position

Figure 2. TS map of the GeV -ray emission from LS 5039. The upper panel shows a region with pixels, centered on 4FGL J1826.2−1450, with the target source excluded from the best-fit model. The green star marks the catalog position of 4FGL J1826.2−1450, while the white crosses indicate other -ray sources in 4FGL-DR4. The lower panel shows a zoomed in view of the excess covering a region. The white cross marks the best-fit position obtained with gtfindsrc, while the green star marks the catalog position. The dashed white circle represents the 68% statistical localization region.
of LS 5039, with the high-energy photons concentrated around the source location. At GeV, We further refined the position of the excess using gtfindsrc. The best-fit position is R.A. = and Decl. = (J2000), with a 68% statistical localization radius of . The best-fit position is separated from the position of 4FGL J1826.2−1450 by , placing the catalog position within the 68% error circle. Hence, we conclude that the high-energy -ray excess is very likely associated with LS 5039, rather than arising from contamination by nearby -ray sources or diffuse-background emission.
Phase-Resolved Analysis
To investigate whether the -ray excess is modulated with the orbital phase, we constructed phase-resolved spectra for the superior conjunction (SUPC) and inferior conjunction (INFC) intervals. Our 17.3 yr Fermi-LAT timing analysis gives d, consistent with L. Zeng et al. [62]. We adopt phase zero at periastron, (MJD 51942.59), and define SUPC as and and INFC as , following F. Aharonian et al. [3]; R. Alfaro et al. [6]. As noted by H. Yoneda et al. [58], orbital-phase uncertainties can accumulate over long observational baselines owing to uncertainties in the orbital period and reference epoch. The published H.E.S.S. and HAWC phase-resolved data use the same orbital-phase zero point as adopted here but d, differing from our value of 3.90609 d by only d (). The resulting phase offset, , reaches at most , corresponding to only and of the SUPC and INFC phase widths, respectively. Thus, this period difference does not materially affect the phase-resolved spectral comparison, except possibly near the phase boundaries.
For the SUPC and INFC intervals, we repeated the likelihood analysis using the same event selection, source model, and diffuse backgrounds as in the phase-averaged analysis, selecting only events in the corresponding phase range. Independent broadband fits were performed for the two intervals. The best-fit BPL+LP parameters are listed in Table 2, and the corresponding models are shown in Figure 3. The phase-resolved SEDs were then extracted by fixing the target spectral shape in each energy bin to the best-fit results, while allowing its normalization and those of sources within to vary. Bins with TS are reported as 95% upper limits. The phase-averaged SED is also shown for comparison, and the INFC spectrum exhibits a more complex structure than the SUPC spectrum. Notably, the -ray excess in the 30–500 GeV (colored green) is predominantly detected during the INFC interval. These phase-resolved SEDs are used in the following sections to investigate the orbital modulation of the broadband -ray emission.
| BPL+LP parameter | All phase | SUPC | INFC |
| Parameter 1 | |||
| Parameter 2 | |||
| Characteristic energy 1 (GeV) | |||
| Parameter 3 | |||
| Parameter 4 | |||
| Characteristic energy 2 (GeV) | |||
| 33023.4 | 24398.4 | 9137.6 | |
| 116.9 | 39.6 | 50.5 | |
| -44716325.81 | -19011195.91 | -13747011.55 |
Table 2. Best-fit parameters of the BPL+LP model for the phase-averaged, SUPC, and INFC spectra
Figure 3. SEDs for the SUPC and INFC intervals, shown in blue and red in the left and right panels, respectively. For comparison, the phase-averaged SED from Figure 1 is shown as black points. In both panels, the -ray excess at – GeV is highlighted in green. The blue and red dashed curves represent the BPL components for SUPC and INFC, respectively, the green dashed curves show the additional LP components, and the corresponding solid curves show the combined BPL+LP models. Open green symbols with downward arrows denote 95% upper limits.
Radiative Mechanisms and Particle Cooling
To model the broadband SED of LS 5039, we adopt a hybrid leptonic–hadronic framework in which electrons and protons are accelerated at the IBS formed by the interaction of the pulsar and stellar winds. The model includes IC emission from relativistic electrons [38], emission from accelerated protons [34], and a phase-independent magnetospheric component described by the outer-gap model of J. Takata et al. [55], following Y. Wang et al. [57]. High-energy -rays produced at the IBS are attenuated by absorption in the stellar radiation field, which we account for by multiplying the intrinsic spectrum by ; the energy- and phase-dependent optical depth is calculated following G. Dubus [24] with the full orbital geometry.
We adopt the orbital and stellar parameters of J. Casares et al. [17], with the updated period d [62], K, and . Motivated by the possible hard-X-ray pulsation [59], we assume a neutron star with , , and AU. The corresponding periastron and apastron separations are 0.094 and 0.196 AU, respectively. The adopted geometry and SUPC/INFC intervals are shown in Figure 4. The IBS apex is set by the wind-momentum balance,

Figure 4. Schematic orbital geometry of LS 5039 based on the orbital parameters reported by J. Casares et al. [17]. The gray circle represents the O6.5 V((f)) companion star, while the neutron star follows an eccentric orbit with . The blue and red segments denote the SUPC and INFC intervals, respectively (F. Aharonian et al. [3]). The gray dashed ellipse shows the locus of the IBS apex over the orbit for a wind-momentum ratio of . Periastron, apastron, superior conjunction, and inferior conjunction occur at orbital phases , , , and , respectively. The black dashed lines indicate the periastron and apastron separations, and the arrow denotes the direction toward the observer. The orbital-geometry schematic is adapted from D. Huber et al. [33].
where , , and are the pulsar spin-down power, stellar mass-loss rate, and wind velocity [28, 14]. Adopting erg s, , and cm s [55] gives . For isotropic winds, , , so that and . The apex therefore follows the phase-dependent binary separation; we adopt this simplified local geometry throughout.
The stellar photon energy density at the shock apex is
while the proton number density of the stellar wind is
where is the proton mass. These phase-dependent quantities serve as the target photon field for IC scattering and the target matter field for interactions in the SED modeling.
**Note—**The SUPC interval is defined as and , while the INFC interval is defined as . The quantities , , , , , , and denote the stellar photon-field energy density, binary separation, IC scattering angle, target gas density, magnetic-field strength, maximum electron energy, and maximum proton energy, respectively. The quantities , , , , and denote the orbital inclination, source distance, magnetic-energy fraction, electron acceleration parameter, and proton acceleration parameter, respectively. Angle brackets denote normalized orbital-phase-weighted averages over the corresponding phase interval.
We assume that the IBS magnetic field originates in the pulsar wind and parameterize the post-shock magnetic energy density as , with denoting the magnetic fraction of the local pulsar-wind energy density:
Neglecting shock compression and field geometry, for . We adopt , representative of a weakly magnetized pulsar wind and comparable to values used for shocked pulsar winds and -ray binaries ([36, 25, 55, 9]).
Following the orbital geometry shown in Figure 4, we convert the orbital phase into the true anomaly and compute the instantaneous binary separation from the Keplerian orbit. The SUPC and INFC intervals are defined according to Figure 4. The SUPC and INFC SEDs are integrated over their full phase intervals using normalized Gauss–Legendre quadrature with uniform phase weighting. We use 24 SUPC nodes, split at the occultation boundaries, and 6 INFC nodes. Orbital geometry and local IBS conditions are evaluated at each node; Table 3 lists the corresponding phase-weighted averages.
| Parameter | SUPC interval | INFC interval |
| (erg cm) | ||
| (AU) | 0.141 | 0.170 |
| (deg) | 103.4 | 42.0 |
| (cm) | ||
| (G) | 3.21 | 2.55 |
| (TeV) | 34.10 | 37.72 |
| (deg) | 60 | |
| (kpc) | 2.5 | |
| 1 | ||
| 1 | ||
| (PeV) | 0.346 |
Table 3. Phase-weighted physical parameters for the SUPC and INFC intervals.
As listed in Table 3, the stellar photon energy density at the IBS exceeds the magnetic energy density (), so IC cooling is efficient in the Thomson regime ([8, 51, 24, 11]). For K, the characteristic stellar photon energy is eV. The Thomson–Klein–Nishina transition, , occurs at
([8, 46, 38]). Above this energy, KN suppression reduces the IC cooling rate ([46, 37, 38]), allowing synchrotron losses to determine the maximum electron energy despite ([46, 24, 11, 40]). With parameterizing the acceleration efficiency ( at the Bohm limit), balancing acceleration and synchrotron losses gives ([37, 27])
For and , the node-wise fields are 2.18–4.48 G in SUPC and 2.16–3.56 G in INFC, giving –40.35 and 31.58–40.54 TeV. Since , gives –23 TeV, whereas gives a formal –128 TeV limit but reduces the local confinement limit from to PeV. Thus, lowering alone cannot both accelerate and confine TeV electrons at the IBS apex.
The phase-resolved H.E.S.S. and HAWC spectra show TeV orbital modulation, with HAWC extending beyond TeV ([3, 6]). Since KN-regime IC photons satisfy , the –40 TeV electron limits cannot reproduce the TeV emission in the local IBS model. We therefore test a phenomenological pion-decay component from interactions ([35]).
A possible origin of the TeV emission is pion decay from inelastic interactions of shock-accelerated protons with the stellar wind ([35]). The effective efficiency depends on the proton residence time and mixing across the contact discontinuity, which are not resolved in our one-zone treatment. We therefore include the hadronic component phenomenologically, with the maximum proton energy constrained by acceleration, cooling, escape, and confinement in the local IBS environment.
Equating with gives – eV, showing that cooling does not limit the proton energy under the adopted IBS conditions. A more restrictive local estimate follows from the Hillas confinement condition, , where is the characteristic acceleration and confinement scale. Identifying this scale with the pulsar-to-apex distance, , gives eV. We adopt , with and . For , gives the local limit PeV. Since , gives PeV and gives PeV. Larger acceleration scales are not constrained by the present local model. The maximum particle energies are imposed in NAIMA ([60]) as upper integration limits rather than explicit cutoffs in the analytic particle distributions. The IC and pion-decay calculations are therefore restricted to and , respectively.
Spectral Fitting Procedure
Based on the hybrid leptonic–hadronic radiation model and the binary geometry described in § 3, we perform independent broadband spectral fits to the GeV–UHE SEDs of LS 5039 for the SUPC and INFC intervals using the corresponding phase-resolved data sets. The calculations are carried out with the Python package NAIMA (v0.10; [60]), including IC emission and interactions. Within each conjunction interval, particle-distribution parameters are shared among all phase nodes, while , , , the IC angle, particle-energy limits, and opacity are evaluated locally. The phase-averaged SED is
Here ; is phase independent and unattenuated, and in the purely leptonic model. This treatment ensures consistency with the varying binary environment implied by the eccentric orbital geometry and reduces degeneracies between these environmental quantities and the parameters describing the particle distributions, including their normalizations, characteristic energies, spectral breaks, and cooling-related quantities. We fit the detected GeV–UHE flux points with the Gaussian likelihood implemented in naima,
where denotes the adopted flux uncertainty. The posterior distributions are explored using an ensemble Markov Chain Monte Carlo (MCMC) sampler. We use SUPC and 46 INFC detections; upper limits and X-ray data are excluded from the likelihood and BIC. No additional cross-instrument calibration uncertainties or correlated covariance terms are included.
The model components dominate different, although partially overlapping, parts of the SED. The phase-independent magnetospheric component and IC1 from the relativistic-Maxwellian electron population mainly describe the GeV emission, while IC2 from the broken-power-law electrons accounts for the TeV component. The hadronic component contributes mainly to the UHE tail detected by HAWC. X-ray data are not included in the present fit, as this study focuses on the -ray and higher-energy emission of LS 5039, spanning the GeV to UHE regime. This energy range provides the most direct constraints on the relative contributions of the leptonic and hadronic emission channels.
In the Naima fitting, we consider two electron populations and one proton population. The first electron population, responsible for the thermal electron component (corresponding to IC1), is described by a relativistic Maxwellian distribution,
where is the normalization and is the characteristic electron energy. The second electron population, representing shock-accelerated electrons and producing the IC2 component, is described by a broken power-law distribution,
where is the normalization, is the break energy, and and are the spectral indices below and above the break energy, respectively. We fix the reference energy and high-energy spectral index at GeV and , respectively. The proton population, which produces the hadronic emission through inelastic interactions, is also modeled with a broken power-law distribution,
where is the proton normalization, is the break energy, and and are the spectral indices below and above the break energy, respectively. We fix TeV and , and impose the physically derived as the upper proton integration limit at each phase node. Therefore, for each phase interval, the lepto-hadronic model has eight free parameters: , , , , , , , . SUPC and INFC are fitted independently using the bounded uniform priors in Table 4, with 128 walkers, 1500 burn-in steps, and 10000 production steps. Mean acceptance fractions are 0.460–0.552, maximum –86.0 steps, all chains span at least , and the minimum ESS is –. The posterior parameter estimates are listed in Table 5. The corresponding broadband SED fits are shown in Figure 5, and the posterior distributions are presented as corner plots in Figure 6.
Figure 5. Broadband SED fits for SUPC (left) and INFC (right): purely leptonic models are shown in panels (a,b) and lepto-hadronic models in panels (c,d). Purple dash-dotted, light-blue dashed, and green curves denote the outer-gap, IC1, and IC2 components, respectively; dark-green curves show absorbed IC2. Cyan and violet curves show Syn1 and Syn2. In panels (c,d), pale-orange and orange curves denote intrinsic and absorbed emission. Gray dotted and red solid curves show the intrinsic and absorbed total models. Blue/green pentagons are LAT measurements below/above 30 GeV; open symbols denote 95% upper limits. Gray data are from H.E.S.S. (F. Aharonian et al. [3]), HAWC (R. Alfaro et al. [6]), NuSTAR (H. Yoneda et al. [58]), and COMPTEL (W. Collmar & S. Zhang [20]); X-ray and MeV points are not fitted.
Figure 6. Posterior distributions for the SUPC (left) and INFC (right) fits. The upper and lower rows correspond to the purely leptonic and lepto-hadronic models, respectively.
| Parameter | Prior range |
| (12, 17) | |
| (7, 10) | |
| (15, 30) | |
| (, 4.45) | |
| (1, 3) | |
| (15, 30) | |
| (1.00, 2.54) | |
| (1, 3) |
Note— All priors are uniform in the sampled variables and identical for SUPC and INFC. Electrons span ; because is shared among phase nodes, its upper prior uses . Protons span , with the break-energy upper bound set by . The high-energy indices are fixed at .
Table 4. Uniform prior ranges adopted in the MCMC fits
| Parameter | SUPC | INFC | ||
| Leptonic | Lepto-hadronic | Leptonic | Lepto-hadronic | |
| – | – | |||
| – | – | |||
| – | – | |||
| BIC | ||||
| Mean acceptance fraction | ||||
| Maximum (steps) | ||||
| Minimum |
Table 5. Best-fit parameters and convergence diagnostics from the MCMC analysis. Note—Best-fit parameters and MCMC diagnostics for the SUPC and INFC spectra under the purely leptonic and lepto-hadronic models. Parameter values are posterior medians with uncertainties given by the 16th and 84th percentiles. Here, is the reduced chi-square, BIC is the Bayesian information criterion, is the integrated autocorrelation time, and is the number of retained production steps per walker. The reported maximum and minimum correspond to the least favorable sampled parameter in each fit. All four fits have , indicating adequate posterior sampling. Upper limits and X-ray measurements are excluded from the likelihood and BIC calculations.
We compare the purely leptonic and lepto-hadronic models using the Bayesian information criterion (BIC; [52, 30]),
where , , and are the number of free parameters, number of fitted detections, and maximum likelihood, respectively. Lower BIC indicates relative support for a model; we define
So that favors the lepto-hadronic model. BIC is evaluated independently for SUPC and INFC.
Results and Discussion
Purely Leptonic Fit
Figure 5(a) and (b) show the best-fit purely leptonic SED models for the SUPC and INFC intervals. In both phases, the emission below GeV is dominated by the relativistic-Maxwellian and magnetospheric components, while shock-accelerated electrons connect the GeV and TeV bands. The Maxwellian parameters and are similar in the two intervals, whereas the broken-power-law electron population is phase dependent: for SUPC and for INFC. Thus, the INFC TeV emission
requires a higher break energy and a softer low-energy electron spectrum than in SUPC.
These results are broadly consistent with previous leptonic interpretations of LS 5039, in which anisotropic IC scattering, absorption, and orbital-phase-dependent radiative cooling play important roles in shaping the observed GeV–TeV spectrum ([26, 37, 61]). Compared with simple one-zone IC models, our phase-resolved fits suggest that a single homogeneous leptonic emitter may not fully account for all of the observed spectral features, particularly when the highest-energy TeV measurements are included ([22]). This interpretation is compatible with recent shock-structured and relativistic-flow models, in which the emission region is extended and inhomogeneous, with multiple particle-acceleration and cooling zones contributing at different orbital phases [33].
The leptonic model predicts 3–70 keV synchrotron fluxes from the relativistic-Maxwellian (Syn1) and broken-power-law (Syn2) electron populations of (SUPC) and erg cm s (INFC), compared with NuSTAR values of and erg cm s [58]. SUPC is consistent, whereas INFC is low by .
Although the purely leptonic calculation reproduces the main GeV–TeV spectral structure, it underpredicts the 30–500 GeV emission, particularly during INFC, and the HAWC flux above TeV. These residuals motivate testing an additional high-energy component in the following subsection.
Lepto-hadronic Fit
We extend the purely leptonic calculation by introducing a BPL proton population. Its pion-decay emission is calculated using standard parameterizations of high-energy interactions (G. E. Romero et al. [50]; S. R. Kelner et al. [35]), as described in § 3. The best-fit lepto-hadronic models are shown in Figure 5 (c) and (d), and the corresponding parameters are listed in Table 5. In both phase intervals, the emission below remains dominated by the relativistic-Maxwellian and magnetospheric components, while shock-accelerated electrons produce the higher-energy inverse-Compton emission extending through the TeV band. The additional proton population contributes primarily in the multi-TeV to UHE regime. In SUPC, and , and the BIC decreases from 93.42 to 82.62 (). In INFC, and , with the BIC decreasing from 140.54 to 136.27 (). These values indicate stronger relative support for the lepto-hadronic description in SUPC, but the reduced remains well above unity in both intervals and the proton distributions remain weakly constrained. The BIC differences therefore represent relative model preferences rather than evidence for a uniquely adequate physical model.
For the modeled – proton population, (SUPC) and (INFC). For , the residence times are 12.8 and 15.5 s and and , consistent with rapid proton escape (W. Bednarek & J. Sitarek [7]). The required powers are and , or and . Thus the fast-escape local IBS-apex interpretation is energetically inconsistent with pulsar spin-down alone. As an independent X-ray consistency check, the lepto-hadronic fits predict 3–70 keV synchrotron fluxes of (SUPC) and (INFC). The SUPC value agrees with NuSTAR, whereas INFC is low by (H. Yoneyama et al. 2021). Thus, adding the proton component improves the -ray description but does not resolve the INFC X-ray deficit.
The relative BIC support for the lepto-hadronic description is stronger in SUPC, but the uncertainties do not establish an orbital change in the proton spectral shape. The observed multi-TeV modulation nevertheless indicates that at least part of the VHE emission originates within or near the binary, where anisotropic IC and absorption remain important (V. Bosch-Ramon & D. Khangulyan [12]). Future phase-resolved observations with VLAST (Y. Z. Fan et al. [29]), HAWC, LHAASO, and CTAO will be required to constrain the origin of the highest-energy component.
Physical Implications for the UHE Emission
The inferred phase dependence is broadly compatible with spatially resolved transport calculations of LS 5039 (D. Huber et al. [33]; R. Kissmann et al. [39]), but the present one-zone fits cannot distinguish among variations in acceleration, escape, target density, and shock geometry. Hadronic or hybrid interpretations have been considered in pulsar-powered systems (G. E. Romero et al. [50]; S. R. Kelner et al. [35]; H. Li et al. (2010)), and strong large-scale stellar-wind magnetic fields may in principle support multi-PeV proton acceleration (A. E. Petrov & A. M. Bykov [49]).
The electron-energy constraint applies specifically to the local IBS-apex model: a IC photon requires an electron energy of order 120 TeV (V. Bosch-Ramon & D. Khangulyan [12]), well above the fitted 28–41 TeV limits. Relativistic outflows or larger-scale termination shocks could relax the cooling and confinement constraints (V. Bosch-Ramon & D. Khangulyan [12]; J. R. Martinez & V. Bosch-Ramon [44]), while magnetar-like activity could provide additional nonthermal power (V. Bosch-Ramon & M. V. Barkov [10]). However, a steady large-scale component alone would not explain the observed 2–118 TeV orbital modulation (R. Alfaro et al. [6]). Thus, the present results constrain the local one-zone leptonic model but do not exclude more complex leptonic origins of the UHE emission. UHE emission from LS I +61° 303 and the PWN powered by PSR J1849–0001 shows that -ray binaries and pulsar-powered systems can accelerate particles to TeV or PeV energies [16, 41]. Although these systems differ from LS 5039, they demonstrate that such extreme particle energies are plausible in compact pulsar-powered environments.
The statistical improvement obtained by including the proton component favors a mixed phenomenological description of the broadband SED. However, it does not uniquely identify the physical origin of the UHE emission because the proton distribution remains weakly constrained and extreme leptonic scenarios cannot be excluded. Additional phase-resolved UHE and neutrino observations will be required to distinguish between these possibilities.
Summary
We performed a -ray spectral analysis of LS 5039 using 17.3 yr of Fermi-LAT data and identified an apparent spectral excess at –500 GeV, with a local significance of . We then constructed phase-resolved broadband SEDs by combining the Fermi-LAT results with published H.E.S.S. and HAWC measurements and integrating the model over the full SUPC and INFC intervals. The spectra show phase-dependent structure that motivates a multicomponent interpretation, with the INFC SED exhibiting greater spectral complexity than the SUPC SED.
Within the adopted framework, the GeV–TeV continuum is described predominantly by leptonic components: the relativistic-Maxwellian and magnetospheric components dominate below GeV, while shock-accelerated electrons produce IC emission above GeV and connect the GeV and TeV bands. However, the purely leptonic model leaves residual structure at 30–500 GeV and underpredicts the HAWC flux above TeV. At the highest energies, synchrotron cooling limits the electron energies and IC scattering occurs in the deep Klein–Nishina regime, making the UHE emission difficult to reproduce within the local one-zone IBS-apex model, although more complex leptonic acceleration sites or outflow geometries remain possible.
Adding a proton component improves the broadband spectral description in both phase intervals, reducing the BIC from 93.42 to 82.62 in SUPC and from 140.54 to 136.27 in INFC. Thus, the relative support for the lepto-hadronic model is substantially stronger in SUPC, although the reduced remains well above unity in both intervals. These results support an additional proton-induced contribution to the high-energy emission whose relative strength or spectral properties may vary over the orbit, although the fitted proton spectra remain weakly constrained, particularly during INFC. The local IBSapex hadronic interpretation also remains energetically challenging under the adopted fast-escape conditions.
Overall, our results are consistent with a multicomponent, phase-dependent radiation scenario for LS 5039. Within the adopted framework, an additional component improves the description of the apparent 30–500 GeV excess and the highest-energy emission. However, its hadronic origin is not uniquely established because the proton spectrum remains weakly constrained, the energy budget is challenging, and more complex leptonic scenarios remain possible. Future phase-resolved observations from TeV to beyond 100 TeV with HAWC, LHAASO, and CTAO, complemented by GeV-to-multi-TeV measurements with VLAST [29] and searches for phase-dependent neutrino emission, will help constrain its orbital behavior and physical origin.
Acknowledgments
We would like to thank the anonymous referee for helpful suggestions. This work was supported in part by the National Natural Science Foundation of China under Grant No. 12233006. P.Z. acknowledges support from the Xingdian Talent Support Plan–Youth Project, and by the Scientific Research and Innovation Project of Postgraduate Students in the Academic Degree of Yunnan University Nos. KC-262614856 and KC-262615807.
References
- [1]Abdo, A. A., Ackermann, M., Ajello, M., et al. 2009, ApJL, 706, L56, doi: 10.1088/0004-637X/706/1/L56DOI
- [2]Aharonian, F., Akhperjanian, A. G., Aye, K.-M., et al. 2005, Science, 309, 746, doi: 10.1126/science.1113764DOI
- [3]Aharonian, F., Akhperjanian, A. G., Bazer-Bachi, A. R., et al. 2006, A&A, 460, 743, doi: 10.1051/0004-6361:20065940DOI
- [4]Akaike, H. 1974, IEEE Transactions on Automatic Control, 19, 716, doi: 10.1109/TAC.1974.1100705DOI
- [5]Albert, J., Aliu, E., Anderhub, H., et al. 2006, Science, 312, 1771, doi: 10.1126/science.1128177DOI
- [6]Alfaro, R., Araya, M., Arteaga-Velázquez, J. C., et al. 2025, ApJL, 987, L42, doi: 10.3847/2041-8213/ade760DOI
- [7]Bednarek, W., & Sitarek, J. 2013, MNRAS, 430, 2951, doi: 10.1093/mnras/stt098DOI
- [8]Blumenthal, G. R., & Gould, R. J. 1970, Reviews of Modern Physics, 42, 237, doi: 10.1103/RevModPhys.42.237DOI
- [9]Bosch-Ramon, V. 2021, A&A, 645, A86, doi: 10.1051/0004-6361/202039666DOI
- [10]Bosch-Ramon, V., & Barkov, M. V. 2025, A&A, 703, A299, doi: 10.1051/0004-6361/202555257
- [11]Bosch-Ramon, V., & Khangulyan, D. 2009, International Journal of Modern Physics D, 18, 347, doi: 10.1142/S0218271809014601
- [12]Bosch-Ramon, V., & Khangulyan, D. 2025, A&A, 700, A162, doi: 10.1051/0004-6361/202555514
- [13]Bosch-Ramon, V., & Paredes, J. M. 2004, A&A, 417, 1075, doi: 10.1051/0004-6361:20035615DOI
- [14]Cantó, J., Raga, A. C., & Wilkin, F. P. 1996, ApJ, 469, 729, doi: 10.1086/177820DOI
- [15]Cao, G., Yang, X., & Zhang, L. 2024, Universe, 10, 130, doi: 10.3390/universe10030130DOI
- [16]Cao, Z., Aharonian, F., Bai, Y. X., et al. 2026, PhRvL, 136, 181001, doi: 10.1103/7xhp-tff7DOI
- [17]Casares, J., Ribó, M., Ribas, I., et al. 2005, MNRAS, 364, 899, doi: 10.1111/j.1365-2966.2005.09617.xDOI
- [18]Cerutti, B., Malzac, J., Dubus, G., & Henri, G. 2010, A&A, 519, A81, doi: 10.1051/0004-6361/201014459DOI
- [19]Cheng, K. S., Ruderman, M., & Zhang, L. 2000, ApJ, 537, 964, doi: 10.1086/309051DOI
- [20]Collmar, W., & Zhang, S. 2014, A&A, 565, A38, doi: 10.1051/0004-6361/201323193DOI
- [21]Corbet, R. H. D., Chomiuk, L., Coe, M. J., et al. 2019, ApJ, 884, 93, doi: 10.3847/1538-4357/ab3e32DOI
- [22]del Palacio, S., Bosch-Ramon, V., & Romero, G. E. 2015, A&A, 575, A112, doi: 10.1051/0004-6361/201424713
- [23]Dermer, C. D., & Böttcher, M. 2006, ApJ, 643, 1081, doi: 10.1086/502966DOI
- [24]Dubus, G. 2006, A&A, 451, 9, doi: 10.1051/0004-6361:20054233DOI
- [25]Dubus, G. 2013, A&A Rev, 21, 64, doi: 10.1007/s00159-013-0064-5DOI
- [26]Dubus, G., Cerutti, B., & Henri, G. 2008, A&A, 477, 691, doi: 10.1051/0004-6361:20078261DOI
- [27]Dubus, G., Lamberts, A., & Fromang, S. 2015, A&A, 581, A27, doi: 10.1051/0004-6361/201425394DOI
- [28]Eichler, D., & Usov, V. 1993, ApJ, 402, 271, doi: 10.1086/172130DOI
- [29]Fan, Y. Z., Chang, J., Guo, J. H., et al. 2022, Acta Astronomica Sinica, 63, 27
- [30]Fang, K., Kerr, M., Blandford, R., Fleischhack, H., & Charles, E. 2022, PhRvL, 129, 071101, doi: 10.1103/PhysRevLett.129.071101DOI
- [31]Gaensler, B. M., & Slane, P. O. 2006, ARA&A, 44, 17, doi: 10.1146/annurev.astro.44.051905.092528DOI
- [32]Hinton, J. A., Skilton, J. L., Funk, S., et al. 2009, ApJL, 690, L101, doi: 10.1088/0004-637X/690/2/L101DOI
- [33]Huber, D., Kissmann, R., & Reimer, O. 2021, A&A, 649, A71, doi: 10.1051/0004-6361/202039278DOI
- [34]Kafexhiu, E., Aharonian, F., Taylor, A. M., & Vila, G. S. 2014, PhRvD, 90, 123014, doi: 10.1103/PhysRevD.90.123014DOI
- [35]Kelner, S. R., Aharonian, F. A., & Bugayov, V. V. 2006, PhRvD, 74, 034018, doi: 10.1103/PhysRevD.74.034018DOI
- [36]Kennel, C. F., & Coroniti, F. V. 1984, ApJ, 283, 694, doi: 10.1086/162356DOI
- [37]Khangulyan, D., Aharonian, F., & Bosch-Ramon, V. 2008, MNRAS, 383, 467, doi: 10.1111/j.1365-2966.2007.12572.xDOI
- [38]Khangulyan, D., Aharonian, F. A., & Kelner, S. R. 2014, ApJ, 783, 100, doi: 10.1088/0004-637X/783/2/100DOI
- [39]Kissmann, R., Huber, D., & Gschwandtner, P. 2026, A&A, 706, A211, doi: 10.1051/0004-6361/202556701DOI
- [40]Lange, A., Corbet, R. H. D., Coley, J. B., et al. 2025, ApJ, 986, 181, doi: 10.3847/1538-4357/adda32DOI
- [41]LHAASO Collaboration, Cao, Z., Aharonian, F., et al. 2026, Nature Astronomy, doi: 10.1038/s41550-026-02839-0DOI
- [43]Maíz Apellániz, J., Sota, A., Arias, J. I., et al. 2016, ApJS, 224, 4, doi: 10.3847/0067-0049/224/1/4DOI
- [44]Martinez, J. R., & Bosch-Ramon, V. 2026, A&A, 713, A75, doi: 10.1051/0004-6361/202660821DOI
- [45]Mirabel, I. F., & Rodríguez, L. F. 1999, ARA&A, 37, 409, doi: 10.1146/annurev.astro.37.1.409DOI
- [46]Moderski, R., Sikora, M., Coppi, P. S., & Aharonian, F. 2005, MNRAS, 363, 954, doi: 10.1111/j.1365-2966.2005.09494.xDOI
- [47]Paredes, J. M., Martí, J., Ribó, M., & Massi, M. 2000, Science, 288, 2340, doi: 10.1126/science.288.5475.2340DOI
- [49]Petrov, A. E., & Bykov, A. M. 2026, Advances in Space Research, 77, 8391, doi: 10.1016/j.asr.2026.01.103DOI
- [50]Romero, G. E., Torres, D. F., Kaufman Bernadó, M. M., & Mirabel, I. F. 2003, A&A, 410, L1, doi: 10.1051/0004-6361:20031314-1DOI
- [51]Rybicki, G. B., & Lightman, A. P. 1979, Radiative processes in astrophysics
- [52]Schwarz, G. 1978, Annals of Statistics, 6, 461
- [53]Sierpowska-Bartosik, A., & Torres, D. F. 2008, Astroparticle Physics, 30, 239, doi: 10.1016/j.astropartphys.2008.09.009
- [54]Takahashi, T., Kishishita, T., Uchiyama, Y., et al. 2009, ApJ, 697, 592, doi: 10.1088/0004-637X/697/1/592DOI
- [55]Takata, J., Leung, G. C. K., Tam, P. H. T., et al. 2014, ApJ, 790, 18, doi: 10.1088/0004-637X/790/1/18DOI
- [56]Tavani, M., & Arons, J. 1997, ApJ, 477, 439, doi: 10.1086/303676DOI
- [57]Wang, Y., Takata, J., & Cheng, K. S. 2010, ApJ, 720, 178, doi: 10.1088/0004-637X/720/1/178DOI
- [58]Yoneda, H., Khangulyan, D., Enoto, T., et al. 2021, ApJ, 917, 90, doi: 10.3847/1538-4357/ac0ae1DOI
- [59]Yoneda, H., Makishima, K., Enoto, T., et al. 2020, PhRvL, 125, 111103, doi: 10.1103/PhysRevLett.125.111103DOI
- [60]Zabalza, V. 2015, in International Cosmic Ray Conference, Vol. 34, 34th International Cosmic Ray Conference (ICRC2015), 922, doi: 10.22323/1.236.0922
- [61]Zabalza, V., Bosch-Ramon, V., Aharonian, F., & Khangulyan, D. 2013, A&A, 551, A17, doi: 10.1051/0004-6361/201220589DOI
- [62]Zeng, L., Zhang, M., Ren, C., Zhang, P., & Yan, J. 2024, ApJ, 977, 40, doi: 10.3847/1538-4357/ad7bb6
- [63]Zhang, J.-F., Zhang, L., & Fang, J. 2009, PASJ, 61, 949, doi: 10.1093/pasj/61.5.949DOI
- [64]Zhang, L., & Cheng, K. S. 1997, ApJ, 487, 370, doi: 10.1086/304589DOI



