betaFeedback

WorksPhysics and Astronomy

Systematic Study of Cosmic Rays in Star-Forming and Starburst Galaxies using Fermi-LAT and Multiwavelength Data

Abstract

The majority of γγ rays from star-forming galaxies (SFGs) and starburst galaxies (SBGs) are expected to originate from interactions between high-energy cosmic rays (CRs) and the interstellar medium, making γγ-ray observations a powerful tool for probing CR populations and interstellar environments. Using 15 years of Fermi-LAT data, we conducted a systematic study of seven sources. The γγ-ray spectra of the bright SBGs M82 and NGC~253 exhibit a decrease in energy flux below a few hundred MeV, suggesting a substantially smaller contribution from CR electrons than previously inferred. Detailed fits of the spectral energy distribution, which combine our GeV γγ-ray measurements with reported radio and TeV γγ-ray results, indicate that the electron-to-proton ratios in M82 and NGC~253 are comparable to that of the Milky Way. In contrast, energy densities of CRs in these two SBGs are 2.5–8 times lower than that of the magnetic field, unlike the Milky Way. For the nearby Local Group galaxy M31, we decomposed the γγ-ray emission into two components, revealing a central source with a soft spectrum and a compact size of less than 0.33 degree, which indicates a considerable contribution from unresolved sources. Our new analysis also reduces the γγ-ray fluxes of several SBGs, bringing all SBG sample members to levels comparable to or below the calorimetric limit, with calorimetric efficiencies ranging from 0.33 to 1.20. In contrast, all SFG sample members exhibit low efficiencies at the 10% level. We developed a theoretical model of interstellar γγ-ray emission that successfully reproduces the observed calorimetric efficiencies, providing a robust framework for interpreting and predicting γγ-ray emission from SFGs and SBGs.

Introduction

Cosmic rays (CRs) in the Milky Way Galaxy are believed to originate from supernova remnants and other star-formation processes, such as the winds of massive stars and pulsars (e.g., V. L. Ginzburg & S. I. Syrovatskii [30]; Y. Butt [20]). Similarly, galaxies with high star-formation rates are expected to exhibit CR densities comparable to or greater than those of the Milky Way. In these external galaxies, CRs generate electromagnetic radiation through mechanisms analogous to those in the Milky Way. Specifically, CR electrons produce synchrotron radiation via interactions with interstellar magnetic fields and non-thermal Bremsstrahlung radiation via interactions with interstellar gas. They also upscatter the interstellar radiation field (ISRF) via inverse Compton (IC) scattering. CR protons produce γ\gamma rays through nuclear–nuclear collisions with interstellar gas. Multiwavelength observations of external galaxies therefore provides insights into both CR populations and the properties of their interstellar environments, including magnetic fields, the ISRF, and gas content, thereby advancing understanding of the underlying astrophysical processes.

The γ\gamma-ray band is particularly important because it is sensitive to CR protons within galaxies. CR protons produce intense GeV and TeV γ\gamma rays through the decay of neutral pions, which are generated by nuclear–nuclear interactions with ambient interstellar gas. This process typically results in γ\gamma-ray emission that dominates over that from CR electrons in typical interstellar environments. Fermi-LAT (W. B. Atwood et al. [17]) detected GeV γ\gamma-rays from the nearest star-forming galaxies (SFGs), specifically the LMC, SMC, and M31 (A. A. Abdo et al. [1]; A. A. Abdo et al. [2]; A. A. Abdo et al. [3]), shortly after its launch. The instrument also identified significant GeV γ\gamma-ray emission from the nearby starburst galaxies (SBGs) M82 and NGC 253 (A. A. Abdo et al. [4]). Concurrently, ground-based Cherenkov telescopes observed significant TeV γ\gamma-rays from these two galaxies [58, 6], thereby establishing SBGs as γ\gamma-ray sources. A strong correlation between observed GeV γ\gamma-ray luminosities and total infrared luminosities [9] suggests that the majority of γ\gamma rays originate from interstellar CRs produced by star-formation activity, rather than from the nuclear supermassive black hole. The hard spectrum observed in SBGs (photon index Γ∼2.2\Gamma\sim2.2) indicates that these galaxies act as CR calorimeters, in contrast to the Milky Way, which is escape-dominated [10]. These observations have prompted theoretical investigations into CR transport in both SBGs and the Milky Way. For example, M. R. Krumholz et al. [36] proposed that CR diffusion in SBGs is nearly energy-independent up to about 1 TeV. Furthermore, the absence of a decrease in γ\gamma-ray flux below a few hundred MeV in M82, which would be expected in a proton-dominated scenario, suggests a high CR electron-to-proton ratio (KepK_{\rm ep}), as indicated by VERITAS 2025 [8].

Understanding γ\gamma rays produced by star-formation activity is also essential for investigating the activity of an active galactic nucleus (AGN). When galactic nucleus emission does not dominate the observed γ\gamma rays, as is typical for non-jetted AGNs, the γ\gamma-ray contributions from the nucleus and from star-formation activity are expected to be comparable. Current and near-future GeV/TeV γ\gamma-ray instruments are not able to resolve nuclear emission from large-scale emission except in local group galaxies. However, differences in spectral shape can enable their distinction. For example, B. Eichmann et al. [26] presented a detailed model of multi-wavelength data for the Seyfert galaxy NGC 1068, noting that coronal emission from the Seyfert nucleus would appear below 100 MeV. Indeed, M. Ajello et al. [15] reported significant γ\gamma-ray emission below a few hundred MeV from the galaxy using the Fermi data. The observed flux exceeds the expected starburst component in a proton-dominated scenario, leading the authors to attribute the emission to MeV radiation from the corona surrounding the Seyfert nucleus.

In short, accurate measurements of γ\gamma rays from SFGs and SBGs are essential for understanding CRs and the interstellar environments of these systems. Furthermore, improved knowledge of SFGs and SBGs enables robust identification and characterization of coronal emission in nearby Seyfert galaxies — many of which exhibit starburst activity — relative to the large-scale starburst components. Recent improvements in the Fermi-LAT instrument response (Pass 8; W. Atwood et al. [16]; P. Bruel et al. [19]), together with more than 15 years of accumulated observations, now make it possible to study γ\gamma-ray emission from SFGs, SBGs, and Seyferts with unprecedented accuracy and uniformity. Motivated by these scientific and technical advances, we have carried out a systematic analysis of individually detected SFGs and SBGs with the Fermi-LAT. The methodology and results are detailed in this paper. A dedicated analysis of Seyfert galaxies, including a comprehensive comparison with SFGs and SBGs, will be reported separately (Barnier et al. 2026, in preparation). This paper is organized as follows. Section 2 describes the target selection, γ\gamma-ray observations, and analysis procedures. Section 3 presents the results for each galaxy. Section 4 discusses the CR properties inferred from detailed modeling of the multi-wavelength spectra from radio to TeV γ\gamma rays, as well as comparisons with theoretical calculations of the GeV γ\gamma rays and among our targets. Section 5 summarizes the findings and outlines future prospects.

Target Selection and Observations

Target Selection

We aim to accurately characterize the γ\gamma-ray spectra of SFGs and SBGs and therefore focus on sources that are individually detected by Fermi-LAT. For source selection, we used the 12-year Fermi-LAT source catalog (4FGL-DR31; S. Abdollahi et al. [5]), which contains more than 6,000 γ\gamma-ray sources. The catalog provides source classifications: six objects are listed as normal galaxies, eight as starburst galaxies, and four as Seyfert galaxies. In general, galaxy types can be distinguished using their broadband spectral properties from radio to X-ray energies. Systems with high radio to optical fluxes are typically jetted AGN (radio galaxies, blazars, etc.). Among galaxies with relatively weak radio emission, those with strong infrared emission are classified as SFGs or SBGs, while those with strong hard X-ray emission are identified as Seyfert galaxies. From the 14 objects labeled as “normal galaxies” or “starburst galaxies” in the 4FGL, we selected six as our SFG/SBG samples and analyzed them in this work. The other eight galaxies are excluded for the following reasons:

  • Four relatively faint systems with photon flux in 1 GeV to 100 GeV fph,1–100≤10−10f_{\mathrm{ph},1\text{--}100} \le10^{-10} photons s−1 cm−2\mathrm{s}^{-1}\,\mathrm{cm}^{-2}: we excluded them because they are not suitable for a detailed analysis of γ\gamma-ray emission.

  • NGC 1068 and NGC 4945: although they are labeled as starbursts in 4FGL due to their high infrared luminosity (reflecting high star-formation rates), both exhibit strong hard X-ray emission indicative of significant nuclear activity, and have been classified as Seyfert galaxies in several previous Fermi-LAT studies (e.g., M. Ajello et al. [15]). We follow this convention and defer their analysis to the Seyfert-focused companion paper (Barnier et al. 2026, in preparation).

  • The LMC and SMC: they are nearby γ\gamma-ray bright galaxies, but their large angular extent complicates the analysis. For simplicity, we excluded them from our target list for the analysis, and instead use their published γ\gamma-ray fluxes in the discussion (§ 4.5).

Instead, we additionally included M33, previously studied as an SFG by M. Ajello et al. [14]. Our final target list is summarized in Table 1. All our sources were already analyzed in M. Ajello et al. [14], but our dedicated spectral and spatial modeling (for M82, NGC 253, and M31; see § 3.1–3.3) and use of more data (10 yr vs. 15 yr) and newer Galactic diffuse and point-source models enable improved characterization of the γ\gamma-ray emission. We adopted the M. Ajello et al. [14] distances2 for easy comparison.

Common name4FGL nameDistance* Mpc(RA, DEC)J2000.0†_{\mathrm{J2000.0}}^\dagger
M824FGL J0955.7+69403.7(148∘.95,+69∘.67)(148^\circ.95, +69^\circ.67)
NGC 2534FGL J0047.5-25173.3(11∘.90,−25∘.29)(11^\circ.90, -25^\circ.29)
M314FGL J0043.2+41140.78(10∘.82,+41∘.24)(10^\circ.82, +41^\circ.24)
Arp 2204FGL J1534.7+233179.9(233∘.70,+23∘.53)(233^\circ.70, +23^\circ.53)
NGC 21464FGL J0618.1+781918.0(94∘.53,+78∘.33)(94^\circ.53, +78^\circ.33)
Arp 2994FGL J1128.2+583147.7(172∘.14,+58∘.56)(172^\circ.14, +58^\circ.56)
M33–0.85(23∘.61,+30∘.78)(23^\circ.61, +30^\circ.78)

Table 1. List of SFGs/SBGs analyzed in this paper

γ-ray Observation and Analysis Procedure

Fermi-LAT (W. B. Atwood et al. [17]) is a pair-tracking γ\gamma-ray telescope sensitive to photons from ∼20\sim20 MeV to beyond 300 GeV. Its wide field of view (∼2.4\sim2.4 sr) and continuous sky-survey operation — covering nearly the entire sky every three hours — make it an ideal instrument for systematic studies of SFGs/SBGs (as well as other source classes) in the sub-GeV to GeV band. Routine LAT science operations began on August 4, 2008. For this work, we analyzed 15 years of data collected from August 4, 2008 to August 2, 2023. The analysis was performed using the standard LAT analysis software package Fermitools3 v2.2.0, together with Fermipy4 v1.3.1, a Python-based pipeline that automates many LAT analysis tasks (M. Wood et al. [62]).

We defined a 15∘×15∘15^\circ\times15^\circ region of interest (ROI) and selected events belonging to the SOURCE class to maximize the photon statistics. We set the ROI aligned in Galactic coordinates to reduce potential contamination from Galactic diffuse emission, but will show the resulting maps in equatorial coordinates for easy comparison with past studies. Data from 50 MeV to 102.4 GeV were binned spatially with a pixel size of 0\fdg1. As described below, we considered logarithmically spaced energy intervals, each increasing by a factor of two (e.g., 50–100 MeV, 100–200 MeV, and so forth until 5120–10240 MeV) in the analysis.

In order to reduce contamination from γ\gamma rays produced in Earth’s atmosphere, we applied zenith-angle cuts of 80∘80^\circ, 90∘90^\circ, and 105∘105^\circ for events in the 50–100 MeV, 100–400 MeV, and ≥400\geq400 MeV bands, respectively. These thresholds account for the LAT’s relatively poor angular resolution at low energies (68% containment radius, r68r_{68}, is ∼5∘\sim5^\circ at 100 MeV). Similarly, we restricted the event selection to PSF types 2 and 3 (the second-best and best angular resolution) for 50–100 MeV, and those of PSF types 1, 2, and 3 for 100–400 MeV. Above 400 MeV, where the angular resolution is significantly better (r68≤2∘r_{68} \leq2^\circ), we included all PSF event types.

For spectral analysis, we divided each energy interval into three bins (i.e., about 10 bins per decade) to capture the effect of finite energy dispersion. We computed the binned maximum likelihood for our ROI using the P8R3_SOURCE_V3 instrument response functions, which match our event selections. Below 400 MeV, we employed the “summed likelihood” method in Fermitools, treating each PSF type independently to best use the angular resolution and better separate the target galaxy from nearby sources. Above 400 MeV, where confusion is less severe, we combined all PSF types for simplicity. Our model includes the standard Galactic diffuse emission model (gll_iem_v07.fits), the isotropic spectral template (iso_P8R3_SOURCE_V3_v1.txt), and all point-source models from the 4FGL-DR3 catalog (S. Abdollahi et al. [5]). For point sources within 7\fdg5 of the ROI center, we allowed both the normalization and spectral index to vary freely. We also left free the normalization and spectral index of the Galactic diffuse background model, as well as the normalization of the isotropic spectral template. The Sun and Moon are known γ\gamma-ray emitters whose motion across the sky can affect LAT analyses if they pass near the ROI. For all targets in our sample, the ecliptic plane lies outside the ROI, and the solar and lunar emissions in the ROI are both weak and very extended in these configurations. Therefore their contribution will be absorbed by the isotropic component, and we did not include templates for the Sun or Moon.

The spectrum of each galaxy below a few hundred MeV is essential for constraining KepK_{\mathrm{ep}}, where the LAT’s angular resolution is relatively poor. An inadequate global spectral model can distort the modeling of nearby sources and bias the derived spectral energy distribution (SED) of the target source. To mitigate this, we adopted two alternative spectral models — the power-law (PL) and log-parabola (logP) — for bright systems (M82, NGC 253, and M31).

The differential photon flux (dN/dEdN/dE) as a function of energy (EE) of the PL model is given by

dNdE=K(EE0)−Γ,(1)\frac{dN}{dE} = K \left(\frac{E}{E_0}\right)^{-\Gamma}, \tag*{(1)}

where KK is the normalization, E0E_0 is the reference energy, and Γ\Gamma is the photon index. The logP model is expressed as

dNdE=K(EEb)−(α+βlog⁡(E/Eb)),(2)\frac{dN}{dE} = K \left(\frac{E}{E_b}\right)^{-(\alpha+\beta\log(E/E_b))}, \tag*{(2)}

where EbE_b is the reference energy, and α\alpha and β\beta describe the spectral slope and curvature, respectively. The logP model is widely used to represent curved γ\gamma-ray spectra. When applying the logP model, we fixed EbE_b to the value listed in the 4FGL catalog. If the 4FGL catalog uses the PL model, we adopted Eb=1000E_b = 1000 MeV, typical of a Milky-Way-like emission. After obtaining the best-fit global spectral model, we derived the SED by fixing all parameters except the normalization of the target source and refitting the normalization independently in each energy interval (50–100 MeV, 100–200 MeV, etc.). We then compared the resulting SED with the best-fit global spectral model and selected the model function (PL or logP) that provides the better overall agreement.

To evaluate the detection significance in each energy interval, we used the test statistic (TS), defined as TS = 2(ln⁡L1−ln⁡L0)2(\ln L_1 - \ln L_0), where L1L_1 and L0L_0 are the maximum likelihood values with and without the source, respectively. The TS approximately follows a χ2\chi^2 distribution with degrees of freedom equal to the number of free parameters (J. R. Mattox et al. [39]). For intervals in which TS <4< 4 (corresponding to ≤2σ\leq2\sigma significance), we report 95% upper limits.

Data Analysis of Each Object

M82

Figure 1 (panels (a) and (c)) presents the count maps of M82, constructed separately for energies above and below 400 MeV. The source is distinctly detected in the high-energy band and remains identifiable in the low-energy band. As outlined in § 2, two spectral models were employed: the logP model (as adopted in 4FGL) and the PL model. The PL model yields Γ=2.145±0.027\Gamma= 2.145 \pm0.027, while the logP model provides α=2.021±0.046\alpha= 2.021 \pm0.046 and β=0.110±0.024\beta= 0.110 \pm0.024, with EbE_{\mathrm{b}} fixed at 956 MeV. Our best-fit PL index is almost identical to that of M. Ajello et al. [14] (Γ=2.14\Gamma= 2.14), who employed only the PL model because the curvature was not significant due to low statistics. Our logP model parameters are similar to those of 4FGL (α=2.115\alpha= 2.115 and β=0.074\beta= 0.074) and require a moderately curved spectrum. The SEDs were evaluated as described in § 2 and compared with the best-fit global models in Figure 2. The best-fit PL model does not adequately reproduce the SED, whereas the logP model shows good agreement. Comparison of SEDs based on the PL and logP models indicates that use of the PL model artificially increases the low-energy flux of M82, particularly below a few hundred MeV, by reducing the low-energy fluxes in models for diffuse emission and/or nearby sources.

Figure 1. (a) Data count map and (b) model count map in the vicinity of M82 in ≥400\ge400 MeV. (c) Data count map and (d) model count map in a 4∘×4∘4^\circ\times4^\circ field of M82 in ≤400\le400 MeV. In all panels, the position of M82 is indicated by a cyan circle, and those of 4FGL sources are indicated by green crosses. Alt text: Two data count maps and two model count maps for the M82 region.

(a)

SED and a best-fit spectral model of M82 obtained by the logP model

(b)

Figure 2. SED and a best-fit spectral model of M82, obtained by (a) the PL model and (b) the logP model. The choice of the spectral model formula affects the SED data points (see text for details), and we adopt the logP model results in the discussion section. Alt text: Two SED plots for M82.

These results indicate that the GeV and sub-GeV emission from M82 exhibits mild curvature5, and use of the PL model introduces bias in the low-energy flux. This bias will affect the SED fitting described in § 4.2. Therefore, the logP model was adopted for subsequent analysis (this Section) and discussion (§ 4). The source is detected at a TS of 1702, corresponding to a significance greater than 40σ40\sigma. The resulting model count maps are shown in Figure 1 (panels (b) and (d)), and the spectral parameters are summarized in Table 2. Numerical values of the SED are provided in Table 3.

NameTSfe,0.1−100f_{\mathrm{e},0.1-100}Γ\Gamma (PL)α\alpha (logP)β\beta (logP)Eb (logP)
M8217028.19e-12–2.021±0.0462.021 \pm0.0460.110±0.0240.110 \pm0.024956
NGC 2538996.31e-12–1.991±0.0641.991 \pm0.0640.130±0.0350.130 \pm0.0351048
M31(center)75.11.87e-12–2.29±0.192.29 \pm0.190.33±0.130.33 \pm0.13913.1
M31(SE)44.21.50e-122.05±0.182.05 \pm0.18–––
Arp 22053.91.72e-122.19±0.162.19 \pm0.16–––
NGC 214653.11.33e-122.00±0.132.00 \pm0.13–––
Arp 29955.81.31e-122.16±0.162.16 \pm0.16–––
M3344.61.58e-122.20±0.122.20 \pm0.12–––

Table 2. Best-fit spectral parameters of our SFG/SBG samples. fe,0.1−100f_{\mathrm{e},0.1-100} gives the integrated energy flux from 100 MeV to 100 GeV. Γ\Gamma is the index of the PL model, and α\alpha, β\beta, and EbE_{\mathrm{b}} are parameters of the logP model.

EnergyM82 (logP)NGC 253 (logP)M31 (center; logP)M31 (SE; PL)
GeV
0.05–0.1≤0.96\le0.96≤0.63\le0.63≤0.37\le0.37≤0.48\le0.48
0.1–0.20.83±0.150.83 \pm0.150.52±0.160.52 \pm0.16≤0.38\le0.38≤0.39\le0.39
0.2–0.40.96±0.110.96 \pm0.110.71±0.120.71 \pm0.120.31±0.120.31 \pm0.12≤0.43\le0.43
0.4–0.81.120±0.0871.120 \pm0.0870.872±0.0940.872 \pm0.0940.49±0.100.49 \pm0.100.22±0.100.22 \pm0.10
0.8–1.61.098±0.0761.098 \pm0.0760.905±0.0840.905 \pm0.0840.331±0.0800.331 \pm0.0800.199±0.0730.199 \pm0.073
1.6–3.21.198±0.0851.198 \pm0.0850.824±0.0890.824 \pm0.0890.216±0.0670.216 \pm0.067≤0.20\le0.20
3.2–6.40.748±0.0860.748 \pm0.0860.80±0.110.80 \pm0.110.100±0.0520.100 \pm0.0520.125±0.0570.125 \pm0.057
6.4–12.80.525±0.0980.525 \pm0.0980.42±0.110.42 \pm0.110.087±0.0590.087 \pm0.0590.111±0.0650.111 \pm0.065
12.8–25.60.57±0.140.57 \pm0.140.27±0.130.27 \pm0.13≤0.079\le0.0790.162±0.0890.162 \pm0.089
25.6–51.20.22±0.120.22 \pm0.120.36±0.190.36 \pm0.19≤0.12\le0.12≤0.25\le0.25
51.2–102.40.15±0.150.15 \pm0.150.17±0.170.17 \pm0.17≤0.20\le0.200.42±0.290.42 \pm0.29

Table 3. SED of bright objects.

NGC 253

Figure 3 represents the data and model count maps of NGC 253: the source is clearly detected above 400 MeV and is identifiable below 400 MeV. Similar to the analysis of M82, both the logP model (adopted in 4FGL) and the PL model were tested: the PL model yields Γ=2.105±0.038\Gamma= 2.105 \pm0.038, while the logP model provides α=1.991±0.064\alpha= 1.991 \pm0.064 and β=0.130±0.035\beta= 0.130 \pm0.035, with EbE_{\mathrm{b}} fixed at 1048 MeV. Again, our best-fit PL index is almost identical to that of M. Ajello et al. [14] (Γ=2.10\Gamma= 2.10), and our logP model parameters are similar to those of 4FGL (α=2.066\alpha= 2.066 and β=0.089\beta= 0.089). As shown by Figure 4, the logP model demonstrated better agreement with the SED data points, while the use of the PL model artificially increased the low-energy flux. Like M82, we adopted the logP model for NGC 253 to prevent the bias in low-energy flux from affecting the SED fitting (§ 4.3). Table 2 summarizes the spectral parameters, including TS of 899 (detection significance of about 30σ30\sigma). The numerical values of the SED are provided in Table 3.

Figure 3. (a) Data count map and (b) model count map in a 4∘×4∘4^\circ\times4^\circ field of NGC 253 in ≥400\geq400 MeV. (c) Data count map and (d) model count map in the vicinity of NGC 253 in ≤400\leq400 MeV. In all panels, the position of NGC 253 is indicated by a cyan circle, and those of 4FGL sources are indicated by green crosses. Alt text: Two data count maps and two model count maps for the NGC 253 region.

Figure 4. SED and a best-fit spectral model of NGC 253, obtained by (a) the PL model and (b) the logP model. Alt text: Two SED plots for NGC 253.

M31

M31 is a Local Group galaxy located at a distance of ∼0.78\sim0.78 kpc and therefore spatially resolved even with the moderate angular resolution of the Fermi-LAT (68% containment radii of ∼0.9∘\sim0.9^\circ and 0.2∘0.2^\circ at 1 GeV and 10 GeV, respectively). Using seven years of Fermi-LAT data, M. Ackermann et al. [12] reported that the GeV γ\gamma-ray emission is dominated by an extended nuclear component with a radius of ∼0.4∘\sim0.4^\circ. More recently, Y. Xing et al. [64] showed that the emission is decomposed into two point-like sources: one at the M31 center and another located ∼0.4∘\sim0.4^\circ to the southeast (SE). We confirmed this result by examining TS maps (constructed with the 4FGL M31 source removed) above 1 GeV and 3 GeV (Figure 5). The two sources are clearly separated in the ≥3\ge3 GeV map, with the central source exhibiting a softer spectrum.

Figure 5. TS maps in the vicinity of M31, (a) above 1 GeV and (b) above 3 GeV. Both maps are constructed with the 4FGL source for M31 removed. In both panels, the central M31 source is indicated by a cyan circle, and the SE source is indicated by a cyan cross. Alt text: Two TS maps for M31.

Accordingly, we modeled M31 emission using two point sources at the positions reported in Y. Xing et al. [64]. The PL model adequately describes the SE source spectrum6, whereas the central source shows clear spectral curvature and is better represented by the logP model, as shown in Figure 6. The comparison between the central and SE source spectra, obtained using the logP and PL models, respectively, is shown in Figure 7. The spectral parameters and the numerical values of the SED data points are listed in Table 2 and Table 3. The curvature parameter (β\beta) of the central source is about three times larger than those of M82 and NGC 253. The central and SE sources are detected with TS values of ∼75\sim75 (>8σ> 8\sigma) and ∼44\sim44 (>6σ> 6\sigma), respectively. The data and model count maps, generated using the final model configuration, are presented in Figure 8.

Figure 6. SEDs and best-fit spectral models of M31’s central source (panels a and b) and the SE source (panels c and d). The results from the PL model are shown in panels a and c, and those from the logP model are shown in panels b and d. Alt text: Four SED plots for M31.

SED plot for M31

Figure 7. SED and best-fit spectral models for the M31’s central source (in red, logP model is used) and the SE source (in black, PL model is used). For clarity, SED data points for the SE source are shifted horizontally. Alt text: An SED plot for M31.

Figure 8. (a) Data cont map and (b) model count map in the vicinity of M31 in ≥400\geq400 MeV. In both panels, the position of the M31’s central source is indicated by a cyan circle, that of the SE source is indicated by a cyan cross, and those of 4FGL sources are indicated by green crosses. Alt text: Two data count maps and two model count maps for the M31 region.

We also tested spatial extension of the central source by modeling it as a uniform disk and scanning the radius in steps of 0.01∘0.01^\circ. To take advantage of the improved angular resolution, we used data above 800 MeV. The profile of the TS value indicates that the source is consistent with being point-like, and we derived a 95% upper limit on the radius of 0.33∘0.33^\circ.

Other Objects

For other weak sources where photon statistics are limited (Arp 220, NGC 2146, Arp 299, and M33; each has TS of ∼50\sim50), we adopted the PL model for simplicity. Use of the logP model reduces the flux by 15% for M33 and ≤7%\le7\% for other sources and will not affect the discussion in § 4. Regarding M33, S.-Q. Xi et al. (2020) reported that the γ\gamma-ray emission is located at NGC 604, an H II region situated ∼0.14\sim0.14 northeast of the galaxy’s center. We confirmed their finding and modeled the γ\gamma-ray emission as a point source at NGC 604. The resulting spectral parameters and numerical values of the SEDs of four galaxies are listed in Table 2 and Table 4, and the spectra are shown in Figure 9; a wider energy range was used to accommodate their low fluxes. The data and model count maps are presented in Figure 10. For reference, the spectra obtained using the logP model are shown in Appendix § 1.

EnergyArp 220Arp 299NGC 2146M33
GeV
0.05–0.1≤0.752\le0.752≤0.584\le0.584≤0.303\le0.303≤0.485\le0.485
0.1–0.4≤0.397\le0.3970.174±0.0870.174 \pm0.087≤0.252\le0.252≤0.360\le0.360
0.4–1.60.268±0.0620.268 \pm0.0620.143±0.0420.143 \pm0.0420.132±0.0460.132 \pm0.0460.288±0.0620.288 \pm0.062
1.6–6.40.112±0.0400.112 \pm0.0400.125±0.0320.125 \pm0.0320.132±0.0340.132 \pm0.0340.105±0.0410.105 \pm0.041
6.4–25.60.121±0.0510.121 \pm0.0510.083±0.0400.083 \pm0.0400.086±0.0400.086 \pm0.0400.112±0.0500.112 \pm0.050
25.6–102.40.097±0.0870.097 \pm0.087≤0.206\le0.2060.144±0.0840.144 \pm0.084≤0.197\le0.197

Table 4. SED of weak objects, obtained by the PL model.

Figure 9. SED and a best-fit spectral model of (a) Arp 220, (b) NGC 2146, (c) Arp 299, and (d) M33, obtained by the PL model. Alt text: Four SED plots for other objects.

Figure 10. Data count map (left) and model count map (right) in ≥400\geq400 MeV in the vicinity of Arp 220 (panels a and b), NGC 2146 (c and d), Arp 299 (e and f), and M33 (g and h). In all panels, the position of the target galaxy is indicated by a cyan circle, and those of 4FGL sources are indicated by green crosses. Alt text: Four data count maps and Four model count maps for other object regions.

Compared to the previous analysis by M. Ajello et al. [14], we obtained softer spectra (ΔΓ≥0.25\Delta\Gamma\ge0.25) for Arp 220 and NGC 2146. This difference is attributed to the improved background models and photon statistics. Contamination from diffuse emission and nearby sources is considerable for Arp 220, while diffuse emission contamination is severe for NGC 2146 (see Figure 10). Consequently, significantly reduced γ\gamma-ray fluxes (by a factor of about 2) were obtained for these galaxies.

The photon index (Γ\Gamma) for M33 is also smaller (by about 0.20) than that reported by M. Ajello et al. [14], partially due to the slight offset of the source from the galaxy center. When the source position was fixed at the galaxy center, the resulting index was 2.30±0.122.30 \pm0.12, consistent with that (Γ=2.41\Gamma= 2.41) of M. Ajello et al. [14].

Discussion

In the following, we first model the SEDs of two bright SBGs (M82 and NGC 253) and one moderately bright SFG (M31) in Sections 4.1 and 4.2, respectively. Then we discuss all our samples (including other weak sources) collectively in § 4.5.

SED Modeling of γ\gamma-ray bright SBGs

M82 and NGC 253 are the two nearest starburst galaxies, and there are several past studies of non-thermal emission from them [42, 43 and references therein] (E. de Cea del Pozo et al. 2009); each of these works has its own approach to achieve their objective. In order to model the SEDs of M82 and NGC 253, we employed Naima [66], a Python-based package for computing non-thermal radiation, including radio, X-ray, GeV and TeV γ\gamma rays, from relativistic protons and electrons. Naima has been widely used in GeV and TeV γ\gamma-ray data analysis. It allows deriving CR proton and electron spectra by fitting multiwavelength data under a single-emission-region approximation. We fixed interstellar environmental parameters using values reported in the literature for each galaxy, as detailed in § 4.2 and § 4.3. When literature values depend on distance, such as luminosities, we scaled them to match the adopted distances (hereafter DD) summarized in Table 1. We then derived the CR proton and electron spectra by fitting our GeV data together with radio and TeV data compiled from previous studies. Diffuse X-ray emission in SBGs may also originate from interstellar electrons through synchrotron radiation or IC scattering [e.g. 37]. However, accurately removing point-source contributions and thermal components is challenging. Consequently, we excluded X-ray data from the SED fitting, but confirmed that the apparently diffuse X-ray fluxes reported in the literature are consistent with our SED models.

In our SED fitting, we specified three interstellar quantities: the target gas density for computing pion-decay and electron non-thermal Bremsstrahlung emission; the ISRF (temperature and energy density) for IC scattering; and the magnetic field strength for synchrotron emission. The CR proton spectrum was modeled as a power-law in momentum. The injection spectrum of accelerated CR electrons is also expected to follow a power-law, but electrons experience substantial energy losses due to ionization, Bremsstrahlung, synchrotron radiation, and IC scattering. The total energy-loss rate depends on electron energy EeE_{\mathrm{e}} and can be expressed as b(Ee)=C1+C2Ee+C3Ee2b(E_{\mathrm{e}})=C_{1}+C_{2}E_{\mathrm{e}}+C_{3}E_{\mathrm{e}}^{2}. The ratio C1/C2C_{1}/C_{2} is set by the balance between ionization and Bremsstrahlung losses, while C2/C3C_{2}/C_{3} is determined by the relative importance of Bremsstrahlung, synchrotron, and IC losses. Consequently, the steady-state electron spectrum is proportional to the injected spectrum divided by b(Ee)/Eeb(E_{\mathrm{e}})/E_{\mathrm{e}}, resulting in a smoothly-broken double-broken power-law shape [e.g. 38] (A. Acharya et al. 2025). The first break energy Ebk1=C1/C2E_{\mathrm{bk1}}=C_{1}/C_{2} is universal, whereas the second break Ebk2=C2/C3E_{\mathrm{bk2}}=C_{2}/C_{3} depends on the gas density, ISRF energy density, and magnetic-field energy density. Thus, our CR spectral model contains four free parameters: the normalization and index for CR protons and for CR electrons.

In addition to the synchrotron, free-free emission (thermal Bremsstrahlung) contributes to the radio spectrum. According to P. K. G. Williams & G. C. Bower [61], the radio emission of M82 and NGC 253 is severely affected by free-free absorption below ∼1\sim1 GHz and by dust emission above ∼100\sim100 GHz. We therefore fitted the radio data in the 1–100 GHz range using a model consisting of synchrotron plus free-free emission, adopting a fixed spectral index of 0.1 for the free-free component. Its normalization was allowed to vary freely, giving a total of five free parameters in the modeling.

Modeling of M82

To model M82 SED, we followed B. C. Lacki & T. A. Thompson [37] and related references therein for interstellar quantities. A radius (RR) of 250 pc and a half-height (hh) of 100 pc were adopted for the starburst core, based on M. Goetz et al. [31]. The ISRF was characterized using the SED compilation by L. Silva et al. [52]. The peak wavelength λp\lambda_{\mathrm{p}} in the infrared band, related to temperature as λp=3.92kT\lambda_{\mathrm{p}} = 3.92 kT (where kk is the Boltzmann constant and TT is temperature), is ∼\sim60 μ\mum, corresponding to a temperature TIRT_{\mathrm{IR}} = 60 K.7 The infrared luminosity LIRL_{\mathrm{IR}} in the wavelength range 8–1000 μ\mum was taken from D. B. Sanders et al. [50].8 Considering the starburst core size (2πR2c2\pi R^{2}c, where cc is the speed of light), the energy density in the infrared was estimated as Uph,IRU_{\mathrm{ph,IR}} = 1280 eV cm−3^{-3} for DD = 3.7 Mpc. The peak wavelength and luminosity in the optical band are about 1/50 and 1/4 of those in the infrared, respectively (L. Silva et al. [52]). Accordingly, ToptT_{\mathrm{opt}} = 3000 K and Uph,optU_{\mathrm{ph,opt}} = 320 eV cm−3^{-3} were obtained. In accordance with B. C. Lacki & T. A. Thompson [37], energy equipartition was assumed, resulting in an estimated magnetic-field strength BB of 250 μ\muG. Given the significant uncertainty of BB, its influence on the conclusions is addressed in § 4.5. For gas number density, we referred to A. Weiß et al. [59] who measured the molecular mass in the central region of M829, and derived a proton number density of npn_{\mathrm{p}} = 220 cm−3^{-3} for DD = 3.7 Mpc. The interstellar quantities used in the SED fitting are summarized in Table 5. Using these interstellar parameters and the formulae from A. Acharyya et al. [8], Ebk1∼0.4E_{\mathrm{bk1}} \sim0.4 GeV and Ebk2∼0.7E_{\mathrm{bk2}} \sim0.7 GeV were obtained and are used as fixed values (Table 6).

NameRR pchh pcUph,IRU_{\mathrm{ph,IR}} eV cm−3^{-3}TIRT_{\mathrm{IR}} KUph,optU_{\mathrm{ph,opt}} eV cm−3^{-3}ToptT_{\mathrm{opt}} KBB μ\muGnpn_{\mathrm{p}} cm−3^{-3}
M822501001280603203000250220
NGC 25315050900602253000210320

Table 5. Interstellar quantities assumed for the SED fit

NameWp(≥1W_{\mathrm{p}}(\geq1 GeV) 105310^{53} ergαp\alpha_{\mathrm{p}}We(≥1W_{\mathrm{e}}(\geq1 GeV) 105110^{51} ergαe\alpha_{\mathrm{e}}Kep,1GeVK_{\mathrm{ep,1GeV}}Kep,10GeVK_{\mathrm{ep,10GeV}}Ebk1E_{\mathrm{bk1}} GeVEbk2E_{\mathrm{bk2}} GeV
M823.40−0.27+0.213.40^{+0.21}_{-0.27}2.52−0.05+0.062.52^{+0.06}_{-0.05}3.69−0.23+0.113.69^{+0.11}_{-0.23}2.02−0.15+0.052.02^{+0.05}_{-0.15}1/281/1470.40.7
NGC 2531.54−0.11+0.111.54^{+0.11}_{-0.11}2.41−0.04+0.042.41^{+0.04}_{-0.04}1.66−0.23+0.071.66^{+0.07}_{-0.23}2.03−0.02+0.142.03^{+0.14}_{-0.02}1/261/1200.41.5

Table 6. CR spectral parameters obtained by the SED fit

We used radio data from P. K. G. Williams & G. C. Bower [61], with the measured radio extent of ∼35′′×∼10′′\sim35^{\prime\prime}\times\sim10^{\prime\prime} closely matching the starburst core size. The SEDs in the radio, GeV γ\gamma rays (§ 3.1), and TeV γ\gamma rays (A. Acharyya et al. [8]) were fitted, and the results are summarized in Figure 11. The best-fit indices are 2.52 for CR protons and 2.02 for CR electrons, with KepK_{\mathrm{ep}} estimated at about 1/30 at 1 GeV and about 1/150 at 10 GeV, as detailed in Table 6. In the X-ray regime, the M82 core was examined in detail by D. K. Strickland & T. M. Heckman [54] and Iwasawa, K. et al. [33] using Chandra data. Iwasawa, K. et al. [33] analyzed the starburst core within a region of ∼20′′×∼10′′\sim20^{\prime\prime}\times\sim10^{\prime\prime}, removing point sources and the thermal emission contributions, and identified diffuse hard X-ray emission with a photon index Γ∼1.3\Gamma\sim1.3. By converting their reported luminosity10 into that for DD = 3.7 Mpc, we obtained an SED flux (E2⋅dN/dEE^{2}\cdot dN/dE) of ∼0.70×10−12\sim0.70\times10^{-12} erg s−1^{-1} cm−2^{-2} in 4–8 keV. We remind that one can resolve only very bright sources, and modeling the unresolved source contribution is uncertain. For example, the Galactic Center’s apparently diffuse X-ray emission, once widely considered truly diffuse, was found to be mostly point-source in origin (e.g., M. Revnivtsev et al. [48]). The situation is likely more difficult for an external galaxy like M82. Accordingly, we treat this value as an upper limit. We confirm that it is consistent with our SED model, as illustrated in Figure 11.

(a) (a) proton model spectrum

(b) (b) electron model spectrum

(c) (c) SED of M82

Figure 11. (a) Best-fit proton spectrum and (b) electron spectrum for M82. (c) SED and best-fit model components for M82. Alt text: Three plots to show CR spectral models and the SED fit for M82.

Note: Ebk1E_{\mathrm{bk1}} and Ebk2E_{\mathrm{bk2}} are pre-computed from the assumed interstellar quantities and fixed. WpW_{\mathrm{p}} and WeW_{\mathrm{e}} are best-fit total energies of CR protons and electrons, respectively. αp\alpha_{\mathrm{p}} and αe\alpha_{\mathrm{e}} are best-fit indices of CR protons and electrons, respectively. They give KepK_{\mathrm{ep}} and values at 1 GeV and 10 GeV are also presented. Best-fit normalizations of the free-free emission are 0.68 ±\pm 0.05 and 0.67 ±\pm 0.06 times the measured flux for M82 (at 92 GHz) and NGC 253 (at 94 GHz), respectively.

Modeling of NGC 253

We also followed B. C. Lacki & T. A. Thompson [37] and references therein for the interstellar parameters of NGC 253. For the starburst core, we adopted R=150R = 150 pc and h=50h = 50 pc. Using the infrared luminosity reported by D. B. Sanders et al. [50]11, we obtained LIRL_{\mathrm{IR}} as 1010.49 L⊙10^{10.49}\ L_\odot for D=3.3D = 3.3 Mpc. Approximately half of this luminosity originates from the nuclear starburst (V. P. Melo et al. [40]), and we halved LIRL_{\mathrm{IR}} above. Following B. C. Lacki & T. A. Thompson [37], we assumed Uph,IR/Uph,opt=4U_{\mathrm{ph,IR}}/U_{\mathrm{ph,opt}} = 4 and energy equipartition as in M82, giving the ISRF and magnetic-field energy densities to be ∼70%\sim70\% of those for M82. Specifically, we adopted Uph,IR=900U_{\mathrm{ph,IR}} = 900 eV cm−3^{-3}, Uph,opt=225U_{\mathrm{ph,opt}} = 225 eV cm−3^{-3}, and B=210 μGB = 210\ \mu\mathrm{G}. For the ISM gas mass, we referred to K. Sakamoto et al. [49], who identified several molecular clumps in the central molecular zone12. A proton number density was estimated as np=320n_p = 320 cm−3^{-3} for D=3.3D = 3.3 Mpc. Consequently, Ebk2E_{\mathrm{bk2}} was set to 1.5 GeV, about twice the value adopted for M82. These parameters are summarized in Table 5 and Table 6.

P. K. G. Williams & G. C. Bower [61] measured the radio emission at 1–7 GHz and compiled additional data at 20–100 GHz from the literature. They found that the emission is spatially extended and modeled it as the sum of core and extended components. Their adopted core size of 16′′×10′′16^{\prime\prime} \times10^{\prime\prime} is consistent with the starburst-core dimensions used in our modeling. We fitted the SED using the core-only radio data from P. K. G. Williams & G. C. Bower [61], our GeV γ\gamma-ray data (§ 3.2), and TeV γ\gamma-ray data from H. E. S. S. Collaboration et al. (2018). The resulting fit is shown in Figure 12. The best-fit spectral indices are 2.41 for CR protons and 2.03 for CR electrons, with KepK_{\mathrm{ep}} at 1 GeV and 10 GeV similar to those for M82. These results are summarized in Table 6. For the X-ray band, we referred to D. R. Wik et al. [60], who analyzed NuSTAR and Chandra data and estimated the IC flux (with Γ∼1.6\Gamma\sim1.6) from the core using several emission-region models. Their upper limit for a circular region of radius 15′′15^{\prime\prime} is 2.0×10−132.0 \times10^{-13} erg s−1^{-1} cm−2^{-2} at 20 keV. This limit is consistent with our model prediction, as shown in Figure 12.

Figure 12. SED and best-fit model components for NGC 253. Alt text: A plot to show the SED fit for NGC 253.

SED Modeling of M31

As described in § 3.1, we found that the γ\gamma-ray emission from M31 consists of two components: one located at the galaxy’s center and another ∼0.4∘\sim0.4^\circ to the SE, confirming the results of Y. Xing et al. [64]. Given the 4FGL source density of ∼0.17\sim0.17 deg−2^{-2}, the probability of a background source lying within 0.4∘0.4^\circ is non-negligible, at roughly the 10% level. In the following, we therefore focus on the central source.

The central source dominates the γ\gamma-ray emission around 1 GeV (see Figure 7), and its physical origin has been debated (e.g., M. Ackermann et al. 2017) due to its apparently compact morphology. In our analysis, the 95% upper limit on its radius is only 0.33∘0.33^\circ, corresponding to 4.5 kpc — well below the extent of the 10 kpc ring (e.g., M. W. L. Smith et al. [53]). Although millisecond pulsars (MSPs) may contribute (e.g., M. Persic et al. [46]), and previous studies have discussed a significant contribution (e.g., M. Ackermann et al. [12]) to explain the location and size of the γ\gamma-ray emission, the spectrum we obtained appears to contradict this interpretation. F. Calore et al. [21] showed that the spectrum of Galactic MSPs is well described by an exponential-cutoff power law with a mean photon index of ∼1.3\sim1.3 and a cutoff energy of ∼2.4\sim2.4 GeV, producing a characteristic SED peak at 2–3 GeV (see their Figure 11). This spectral shape is inconsistent with the γ\gamma-ray spectrum from M31, where we observe the cutoff at ≤1\le1 GeV (see Figure 7), requiring a much softer spectrum for MSPs in M31.

Setting aside the small source extension, let us first investigate the scenario of interstellar γ\gamma-ray emission originating from star-formation activity. Due to the limited photon statistics, with the flux less than half those of M82 and NGC 253, we do not fit the SED. Instead, we employed the theoretical modeling of γ\gamma-ray emission from SFGs and SBGs developed by J. Chen & T. Totani [22]. This model estimates γ\gamma-ray emission from CR protons and leptons based on several key galactic parameters: star formation rate ψ\psi, gas mass MgasM_{\mathrm{gas}}, stellar mass MstarM_{\mathrm{star}}, effective radius ReffR_{\mathrm{eff}}, and luminosities in infrared and optical bands (LIRL_{\mathrm{IR}} and LoptL_{\mathrm{opt}}), under the assumption of a one-zone disk-like geometry. In the original formulation, CR particles are assumed to be confined within the gas scale height HgasH_{\mathrm{gas}} from the galactic plane. To improve the accuracy of γ\gamma-ray emission predictions, we have newly introduced a CR scale height HCRH_{\mathrm{CR}}, which is significantly larger than HgasH_{\mathrm{gas}} in non-calorimetric cases. We have also updated the key model parameters to reproduce the canonical values of mean path length (grammage) and CR escape time for the Milky Way. Further methodological details are provided in Appendix 2. We remind readers that this modeling does not fit the observed SED; instead, it determines the γ\gamma-ray spectral shape and flux normalization a priori. We then compared the model prediction of the interstellar γ\gamma-ray emission with the measured spectrum to interpret the observed γ\gamma-ray emission.

Specifically, we adopted the following parameters and summarized them in Table 7: ψ=0.53 M⊙ yr−1\psi= 0.53\ M_\odot\ \mathrm{yr}^{-1}, Mgas=7.6×109 M⊙M_{\mathrm{gas}} = 7.6 \times10^9\ M_\odot, Mstar=86×109 M⊙M_{\mathrm{star}} = 86 \times10^9\ M_\odot and Reff=10R_{\mathrm{eff}} = 10 kpc. These values are adopted from N. Shimono et al. (2021), with ReffR_{\mathrm{eff}} updated to correspond to the gas disk size (10 kpc ring). LIRL_{\mathrm{IR}} and LoptL_{\mathrm{opt}}, which determine the seed photon distribution for IC scattering, are also sourced from the literature and are presented in Table 7. Three characteristic timescales (proton-proton collision time tppt_{\mathrm{pp}}, diffusion loss time tdifft_{\mathrm{diff}}, and advection loss time tadvt_{\mathrm{adv}}) calculated by the model are shown in Figure 13(a) as a function of CR proton energy EpE_p. For comparison, we also give time scales of the Milky Way. In the GeV (and TeV) energy band, tppt_{\mathrm{pp}} is more than one order of magnitude larger than tdifft_{\mathrm{diff}}, resulting in a calorimetric efficiency as a function of EpE_p, fcal(Ep)f_{\mathrm{cal}}(E_p), less than 10% (See Figure 13(b)). Accordingly, the predicted flux of the interstellar γ\gamma-ray emission is low despite a high star-formation rate, and the calculated spectrum is comparable to or lower than the measurement, as summarized in Figure 13(c). This explains why we do not detect significant γ\gamma-ray emission from the 10 kpc ring in M31. Interestingly, predicted π0\pi^0-decay γ\gamma-ray emission peaks at ∼1\sim1 GeV and decays steeply below 200 MeV, being compatible with the data.

Nameψ\psiMgasM_{\mathrm{gas}}MstarM_{\mathrm{star}}ReffR_{\mathrm{eff}}DDLIRL_{\mathrm{IR}}LoptL_{\mathrm{opt}}
MmathrmO_{\\mathrm{O}} yr−1^{-1}109 M⊙10^9\ M_\odot109 M⊙10^9\ M_\odotkpcMpc104310^{43} erg s−1^{-1}104310^{43} erg s−1^{-1}
M31(center)0.537.686.0100.780.9412.2
NGC 214611.410.487.11.717.245.07.27
Milky Way2.64.964.36–5.219.6

Table 7. Parameters to model γ\gamma-ray emission from M31, NGC 2146, and Milky Way

Figure 13. (a) Timescales of CR proton interaction and propagation of M31. For comparison, those of the Milky Way are overlaid. The net propagation time tprt_{\mathrm{pr}}, defined as (tdiff−1+tadv−1)−1(t_{\mathrm{diff}}^{-1}+t_{\mathrm{adv}}^{-1})^{-1}, is also presented. (b) Calorimetric efficiency of M31, Milky Way, and NGC 2146. (c) SED of the M31 central source and the model prediction of interstellar γ\gamma rays. Alt text: Three plots to show CR proton timescales, calorimetric efficiencies, and the SED. All include model curves for M31.

As discussed above, the scenario in which the central source originates from the interstellar γ\gamma-ray emission is problematic given the observed small source extent; there must be a considerable contribution from unresolved sources located in the vicinity (≤4.5\le4.5 kpc) of the M31 center. Our analysis and modeling indicate that the contribution from unresolved sources is considerable, with the spectral shape peaking at ∼1\sim1 GeV. One possible solution is that the average spectrum of MSPs in M31 is much softer than that in the Milky Way. Indeed, Galactic MSPs exhibit softer spectra when the rotational power is lower (i.e., the PSR is older), as discussed by M. Kerr et al. [35]. Hence, if the average age of MSPs in M31 is much older, MSPs may explain the observed spectrum of M31’s central source. Although this is an interesting scenario, it requires detailed modeling of the MSP population in M31 and is outside the scope of this work.

The magnetic field strength (BB) in the model, determined from ψ\psi and the galaxy size, is ∼1.7 μ\sim1.7\ \muG. This value is lower than several estimates derived from energy equipartition between CRs and the magnetic field (e.g., R. Beck & E. M. Berkhuijsen [18]; A. Fletcher et al. [27]). Nevertheless, using B=6 μB = 6\ \muG from these estimates yields similar γ\gamma-ray emission, as shown in Figure 13(c). Therefore, the discussion on the comparison between the γ\gamma-ray spectral data and the model is not affected.

Notes: Parameters for Milky Way and NGC 2146 are taken from J. Chen & T. Totani [22]. For M31, LIRL_{\mathrm{IR}} and LoptL_{\mathrm{opt}} are taken from M. Ajello et al. [14] and G. A. C. W. (1991), respectively. Other M31’s parameters are taken from N. Shimono et al. (2021), but ReffR_{\mathrm{eff}} is updated to correspond to the gas disk size (see text).

CR Production and Propagation of SFGs/SBGs

In SFGs and SBGs, interstellar γ\gamma rays contribute significantly to the total γ\gamma-ray emission. This contribution is substantial in SFGs and predominant in SBGs. Figure 14 presents the relationship between infrared luminosity (LIRL_{\mathrm{IR}}) and γ\gamma-ray luminosity (LγL_{\gamma}) for the selected targets. Specifically, γ\gamma-ray luminosity above 100 MeV is plotted as a function of infrared luminosity in the 8–1000 μ\mum band, as reported by M. Ajello et al. [14]. As discussed in § 4.4, the contribution from unresolved sources is considerable due to the limited spatial extent of M31’s central source. In contrast, most of the potential interstellar γ\gamma-ray emission, which is expected to be symmetric about the galaxy’s center, will be absorbed by the central source model below 1 GeV, where angular resolution is moderate. Therefore, the observed LγL_{\gamma} of M31’s central source is treated as an upper limit for interstellar γ\gamma rays. As detailed in § 3.4, the γ\gamma-ray emission from M33 is offset by ∼\sim0°.14 from the galaxy’s center and is associated with the star-forming region NGC 604. Since NGC 604 is 3–6 Myr old (e.g., M. Relaño & R. C. Kennicutt [47]), which is considerably younger than the typical interaction time of CR protons in SFGs in the GeV range (>10 Myr; see Figure 13(a)), point source contributions are likely to be significant. Consequently, the observed LγL_{\gamma} is also considered an upper limit. For comparison, data points from M. Ajello et al. [14] for the selected targets, as well as for the LMC,13 SMC, and Milky Way, are included. A model curve for the CR calorimetric limit from M. Ackermann et al. [9] is also presented.

Figure 14. γ\gamma-ray luminosity vs. infrared luminosity of our targets and the SMC, the LMC, and the Milky Way. Alt text: A plot to compare γ\gamma-ray luminosity and infrared luminosity.

Our new analyses substantially reduce the γ\gamma-ray luminosities for NGC 2146 and Arp 220 by about a factor of 2, aligning all SBGs with or below the calorimetric limit, with observed fcalf_{\mathrm{cal}} values (ratio of the measured γ\gamma-ray luminosity to the calorimetric limit) ranging from 0.33 to 1.20. To verify this, we computed the theoretical fcal(Ep)f_{\mathrm{cal}}(E_{\mathrm{p}}) for NGC 2146 using the same procedure adopted for M31 and Milky Way (see Table 7 for model parameters). The resulting efficiency exceeds 0.8 at a few GeV (Figure 13(b)), which is consistent with the observed value of 0.65. These findings demonstrate that interstellar γ\gamma-ray emission dominates all SBGs in our sample. The modest variations in fcalf_{\mathrm{cal}} among our sources will enable accurate predictions of the starburst contribution to both the extragalactic γ\gamma-ray background and the γ\gamma-ray emission of individual galaxies; the latter allows distinguishing nuclear emission from the starburst component in Seyfert galaxies. In addition, the SEDs obtained for M82 and NGC 253 will provide observation-based templates for future studies.

In contrast, all SFGs presented in Figure 14 exhibit low fcalf_{\mathrm{cal}} values at the 10% level, except for the SMC, which has fcal=0.28f_{\mathrm{cal}}=0.28. As a low-mass dwarf galaxy, the SMC is expected to have significant contribution from individual point sources. Among the SFGs analyzed, the Milky Way displays unique γ\gamma-ray characteristics. Although its fcalf_{\mathrm{cal}} is only ∼\sim0.1, it remains dominated by interstellar γ\gamma-ray emission (e.g., F. Acero et al. [7]), due to its relatively high star-formation rate, which is within a factor of 2 to 4 of those observed in the SBGs NGC 253 and M82.

Regarding the CR property of individual galaxies, our detailed SED fits (§ 4.1) provide constraints on KepK_{\mathrm{ep}}. The inferred KepK_{\mathrm{ep}} values of M82 and NGC 253 (1/25–1/30 at 1 GeV and 1/120–1/150 at 10 GeV) are comparable to those observed in the Milky Way (e.g., S. Gabici [29]). The higher KepK_{\mathrm{ep}} for M82, previously inferred by A. Acharyya et al. [8], is no longer necessary given the revised SED below a few hundred MeV. In scenarios invoking the diffusive shock acceleration at supernova remnants (SNRs), KepK_{\mathrm{ep}} is generally expected to be universal (e.g., L. Merten et al. [41]), and a value about 1/100 is frequently adopted to predict γ\gamma-ray emission from SFGs and SBGs (e.g., T. M. Yoast-Hull et al. [65]; J. Chen & T. Totani [22]). Despite uncertainties arising from potential energy losses of CR electrons during acceleration or escape, as well as contributions from CR accelerators other than SNRs, the current results offer observational support for the canonical KepK_{\mathrm{ep}} value.

Turning to the spectral indices, we obtained 2.4–2.5 and ∼\sim2.0 for αp\alpha_{\mathrm{p}} and αe\alpha_{\mathrm{e}}, respectively (see Table 6). The latter gives the spectral index of CR electrons of about 3.0 above Ebk2E_{\mathrm{bk2}}. Consequently, both CR protons and electrons show harder spectral indices than those observed in the Milky Way, consistent with expectations for a system characterized by high CR calorimetric efficiency rather than significant CR escape.

Lastly, the CR proton energy densities (UCRU_{\mathrm{CR}}) of M82 and NGC 253 are calculated from WpW_{\mathrm{p}} and the starburst core size, yielding values of ∼\sim190 eV cm−3^{-3} and ∼\sim470 eV cm−3^{-3}, respectively. We remind that an independent analysis by M. Ajello et al. (2025) gave similar UCRU_{\mathrm{CR}} for M82. Our values are ∼\sim8 and ∼\sim2.5 times smaller than the magnetic-field energy densities (UmagU_{\mathrm{mag}}). Although the magnetic field strength BB is subject to significant uncertainty, reducing that of M82 by a factor of two (which would decrease UmagU_{\mathrm{mag}} by a factor of four) prevents a satisfactory fit to the multiwavelength SED. In this scenario, the inferred CR electron spectrum required to reproduce the radio data becomes too intense and leads to an overprediction of γ\gamma-ray emission below a few hundred MeV. Therefore, we conclude that the UCRU_{\mathrm{CR}} of SBGs, at least for M82 and NGC 253, is well below energy equipartition, in contrast to the Milky Way. This could be due to the strong loss of CR protons (via interactions with interstellar gas) and CR electrons (via interactions with ISRF and/or magnetic field) in dense environments, as discussed by, e.g., R. M. Crocker et al. [23]. We note that the assumption of energy equipartition between CRs and the magnetic field is frequently used to estimate BB of SBGs from synchrotron emission (e.g., B. Adebahr et al. [13]; M. Persic & Y. Rephaeli [44]), and these estimates may be subject to bias.

Summary and Prospect

We analyzed 15 years of Fermi-LAT data to characterize GeV and sub-GeV γ\gamma-ray emission from seven SFGs and SBGs, aiming to investigate their CR properties and interstellar environments. The galaxies studied include M82, NGC 253, M31, Arp 220, NGC 2146, Arp 299, and M33. While these galaxies were previously detected in GeV γ\gamma-rays, our improved data analysis, incorporating additional data and updated Galactic diffuse background and source models, enables a more detailed investigation of their γ\gamma-ray emission. We found that the two bright SBGs, M82 and NGC 253, exhibit mild spectral curvature. Our updated SEDs show a clear decrease in energy flux below a few hundred MeV, indicating a substantially smaller contribution from CR electrons than previously inferred. Through detailed multiwavelength SED fitting, spanning radio to TeV γ\gamma-ray energies, we determined that CR electron-to-proton ratios are comparable to that of the Milky Way. However, the CR energy densities in these two SBGs are significantly lower than the energy equipartition with the magnetic field energy density, in contrast to the Milky Way. For M31, the γ\gamma-ray emission consists of two point-like sources, with the central source displaying a soft spectrum and a compact size of less than 0∘.330^{\circ}.33, suggesting a considerable contribution from unresolved sources in addition to interstellar γ\gamma-ray emission. Our revised analysis also yields lower γ\gamma-ray flux estimates for several SBGs. The updated γ\gamma-ray luminosities of our SBG sample members are comparable to or below the calorimetric limit, with calorimetric efficiencies ranging from 0.33 to 1.20. In contrast, all our SFG sample members exhibit low efficiencies at the 10% level, except for the SMC, where point source contributions are likely significant. We further developed a theoretical model for interstellar γ\gamma-ray emission in SFGs and SBGs, which successfully reproduces the observed calorimetric efficiencies of NGC 2146 (SBG) and M31 (SFG). Collectively, these results provide a robust framework for interpreting and predicting interstellar γ\gamma-ray emission from SFGs and SBGs.

Acknowledgments

The Fermi LAT Collaboration acknowledges generous ongoing support from a number of agencies and institutes that have supported both the development and the operation of the LAT as well as scientific data analysis. These include the National Aeronautics and Space Administration and the Department of Energy in the United States, the Commissariat à l’Energie Atomique and the Centre National de la Recherche Scientifique / Institut National de Physique Nucléaire et de Physique des Particules in France, the Agenzia Spaziale Italiana and the Istituto Nazionale di Fisica Nucleare in Italy, the Ministry of Education, Culture, Sports, Science and Technology (MEXT), High Energy Accelerator Research Organization (KEK) and Japan Aerospace Exploration Agency (JAXA) in Japan, and the K. A. Wallenberg Foundation, the Swedish Research Council and the Swedish National Space Board in Sweden.

Additional support for science analysis during the operations phase is gratefully acknowledged from the Istituto Nazionale di Astrofisica in Italy and the Centre National d’Études Spatiales in France. This work performed in part under DOE Contract DE-AC02-76SF00515.

Funding

Part of this work was supported by JSPS KAKENHI Grant Numbers 23K25882 and 23H04895 (T.M.), 26H00822 and 26H02063 (T.T.), and by JST SPRING Grant Number JPMJSP2108 (J.C.).

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

SEDs of Weak objects

In Figure 15, we give SEDs of weak objects obtained using logP model.

(a) Arp 220

SED of NGC 2146

(b) NGC 2146

(c) Arp 299

SED of M33

(d) M33

Figure 15. SED and a best-fit spectral model of (a) Arp 220, (b) NGC 2146, (c) Arp 299, and (d) M33, obtained by the logP model. Alt text: Four SED plots for other weak objects.

Theoretical Modeling of Gamma-ray SED

In order to model the SED of the M31’s central source (the flux is approximately half that of M82 and NGC 253), we employ the model of γ\gamma-ray emission from SFGs and SBGs developed by J. Chen & T. Totani [22] (hereafter C25) based on T. Sudoh et al. [57] and N. Shimono et al. (2021). This model calculates γ\gamma-ray emission from CR protons and electrons based on several key galactic parameters: star formation rate ψ\psi, gas mass MgasM_{\mathrm{gas}}, stellar mass MstarM_{\mathrm{star}}, effective radius ReffR_{\mathrm{eff}}, and luminosities in optical and infrared bands (LoptL_{\mathrm{opt}} and LIRL_{\mathrm{IR}}), assuming a one-zone disk-like geometry. However, the model used here has undergone several significant improvements compared to C25, which are described below.

In the models up to C25, it was assumed that CR particles are confined within the gas scale height HgasH_{\mathrm{gas}} from the Galactic plane; however, in more realistic CR transport models (e.g., GALPROP; A. W. Strong & I. V. Moskalenko [55]; A. W. Strong et al. [56]), the CR scale height HCRH_{\mathrm{CR}} is considerably larger than HgasH_{\mathrm{gas}}. This is particularly important to CR electrons interacting with the radiation field extending from the gas disk. Therefore, in the new model, the scale height of the cosmic ray distribution is determined based on physical considerations as follows. Let HhaloH_{\mathrm{halo}} denote the scale height of the geometric boundary at which cosmic rays escape beyond the galaxy. In the Milky Way, which serves as the model reference, 1 kpc is adopted for HhaloH_{\mathrm{halo}} from a typical scale height for hot gas, and HhaloH_{\mathrm{halo}} in other galaxies is determined on the assumption that the ratio of Hhalo/Reff=1/6H_{\mathrm{halo}}/R_{\mathrm{eff}} = 1/6 is constant across all galaxies.

For a given energy of CR protons, we define HCR,pH_{\mathrm{CR,p}} as the scale height at which CR protons propagate without losing a significant amount of their energy. The time tpr,pt_{\mathrm{pr,p}} to propagate the distance HCR,pH_{\mathrm{CR,p}} is determined by the shorter of the two time scales associated with diffusion (tdifft_{\mathrm{diff}}) and advection (tadvt_{\mathrm{adv}}), i.e., tpr,p=(tdiff−1+tadv−1)−1t_{\mathrm{pr,p}} = (t_{\mathrm{diff}}^{-1} + t_{\mathrm{adv}}^{-1})^{-1}, where two time scales are calculated by the formulation of C25. Here, it is assumed that the same formulation for diffusion in C25 can be applied to the height HCR,pH_{\mathrm{CR,p}}, beyond HgasH_{\mathrm{gas}}. This is not unreasonable, provided that energy is roughly equally distributed between CRs and magnetic fields by strong coupling. If the spread of CR protons is limited by energy loss due to the pp reaction, the scale height will be H~CR,p\widetilde{H}_{\mathrm{CR,p}}, which is determined by tpr,p(HCR,p)=tpp(HCR,p)t_{\mathrm{pr,p}}(H_{\mathrm{CR,p}}) = t_{\mathrm{pp}}(H_{\mathrm{CR,p}}), where tpp=(HCR,p/Hgas)(ngasσppvp)−1t_{\mathrm{pp}} = (H_{\mathrm{CR,p}}/H_{\mathrm{gas}})(n_{\mathrm{gas}}\sigma_{\mathrm{pp}}v_{\mathrm{p}})^{-1}, and vpv_{\mathrm{p}} is the velocity of protons.14 The (HCR,p/Hgas)(H_{\mathrm{CR,p}}/H_{\mathrm{gas}}) factor takes into account the proportion of time that protons spend in the gas disk. The final HCR,pH_{\mathrm{CR,p}} is determined by the following equation:

HCR,p={Hgas(if H~CR,p<Hgas)H~CR,p(if Hgas≤H~CR,p≤Hhalo)Hhalo(if Hhalo<H~CR,p)(3)H_{\mathrm{CR,p}} = \begin{cases} H_{\mathrm{gas}} & \text{(if }\widetilde{H}_{\mathrm{CR,p}} < H_{\mathrm{gas}}\text{)} \\ \widetilde{H}_{\mathrm{CR,p}} & \text{(if }H_{\mathrm{gas}} \le\widetilde{H}_{\mathrm{CR,p}} \le H_{\mathrm{halo}}\text{)} \\ H_{\mathrm{halo}} & \text{(if }H_{\mathrm{halo}} < \widetilde{H}_{\mathrm{CR,p}}\text{)} \end{cases} \tag*{(3)}

Due to significant uncertainties regarding the diffusion coefficients of CR protons and electrons, we will determine it so as to reproduce the mean path length (grammage) λ=Mgas/(2πReff2HCR,p)vptpr,p=10 g cm−2\lambda= M_{\mathrm{gas}}/(2\pi R_{\mathrm{eff}}^{2} H_{\mathrm{CR,p}}) v_{\mathrm{p}}t_{\mathrm{pr,p}} = 10\,\mathrm{g\,cm^{-2}} for CR protons at E=1 GeVE = 1\,\mathrm{GeV}, which is a typical value for the Milky Way. We find that we can reproduce this λ\lambda value by applying a correction factor of 2.9 to the formula of the C25 diffusion coefficient; therefore, we apply this correction factor to all galaxies and cosmic ray energies. In this case, 1 GeV CR protons in the Milky Way are now non-calorimetric (HCR,p=HhaloH_{\mathrm{CR,p}} = H_{\mathrm{halo}}), and the escape time from the CR halo is 1.4×1071.4 \times10^{7} years (see Figure 13(a)).

It should be noted that the galaxy becomes non-calorimetric to CR protons only in the final case of eq. (A1), i.e., Hhalo<H~CR,pH_{\mathrm{halo}} < \widetilde{H}_{\mathrm{CR,p}}. Then the gamma-ray luminosity is calculated using the calorimetric efficiency fcal=1−exp⁡[−tpr,p/tpp]f_{\mathrm{cal}} = 1 - \exp[-t_{\mathrm{pr,p}}/t_{\mathrm{pp}}]. Conversely, when HCR,p≤HhaloH_{\mathrm{CR,p}} \le H_{\mathrm{halo}}, CR protons experience substantial energy loss before reaching HhaloH_{\mathrm{halo}}, resulting in a calorimetry of the galaxy. In this case, since tpr,p=tppt_{\mathrm{pr,p}} = t_{\mathrm{pp}} always, fcalf_{\mathrm{cal}} is calculated as fcal=1−exp⁡[−Hhalo/(Hhalo+HCR,p)]f_{\mathrm{cal}} = 1 - \exp[-H_{\mathrm{halo}}/(H_{\mathrm{halo}} + H_{\mathrm{CR,p}})], such that it connects continuously to the non-calorimetric fcalf_{\mathrm{cal}} when HCR,p=HhaloH_{\mathrm{CR,p}} = H_{\mathrm{halo}}, and fcalf_{\mathrm{cal}} asymptotically approaches 1 when HCR,p≪HhaloH_{\mathrm{CR,p}} \ll H_{\mathrm{halo}}.

MW and M31 are non-calorimetric, but NGC 2146 is an example of a calorimetric galaxy, as illustrated in Figure 16. Here, various time scales for this galaxy are shown in Figure 16a, in comparison to those of MW and M31 in Figure 13(a). Figure 16b summarizes HCR,pH_{\mathrm{CR,p}} of these three galaxies. From a few GeV to a few tens GeV, HCR,pH_{\mathrm{CR,p}} of NGC 2146 is lower than HhaloH_{\mathrm{halo}}, meaning that this galaxy is calorimetric in this energy range.

Timescales of CR proton interaction and propagation of NGC 2146

(a)

(b)

Figure 16. (a) Timescales of CR proton interaction and propagation of NGC 2146. (b) HCR,pH_{\mathrm{CR,p}} of three galaxies. They are equal to HhaloH_{\mathrm{halo}} except EpE_p from a few GeV to a few tens GeV for NGC 2146. Alt text: Two plots to show CR proton timescales and calorimetric efficiencies. All include model curves for NGC 2146.

The scale height of CR electrons, HCR,eH_{\mathrm{CR,e}}, is determined in a manner similar to that for protons. Propagation by diffusion or advection is the same as that of protons. The energy loss timescale for electrons is the cooling time tcl,et_{\mathrm{cl,e}} due to ionization, bremsstrahlung, and synchrotron and IC radiation. Here, regarding the energy loss rates due to ionization and bremsstrahlung, the factor Hgas/HCR,eH_{\mathrm{gas}}/H_{\mathrm{CR,e}} is applied to that of C25 to account for the residence time within the gas disk. On the other hand, the energy losses due to synchrotron and IC radiation are taken from C25 without modification. This is a reasonable assumption, given that the radiation field from the disk persists without significant attenuation up to the halo scale height, Hhalo=(1/6)ReffH_{\mathrm{halo}} = (1/6)R_{\mathrm{eff}}, and that cosmic rays and magnetic fields are strongly coupled. See P. M. W. Kalberla et al. [34] and Y. Fukui et al. [28] for details.

References

  1. [1]Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010a A&A, 512, A7
  2. [2]Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010b A&A, 523, A46
  3. [3]Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010c A&A, 523, L2
  4. [4]Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010d ApJL, 709, L152
  5. [5]Abdollahi, S., Acero, F., Baldini, L., et al. 2022 ApJS, 260, 53
  6. [6]Acero, F., Aharonian, F., Akhperjanian, A. G., et al. 2009 Science, 326, 1080
  7. [7]Acero, F., Ackermann, M., Ajello, M., et al. 2016 ApJS, 223, 26
  8. [8]Acharyya, A., Adams, C. B., Bangale, P., et al. 2025 ApJ, 981, 189
  9. [9]Ackermann, M., Ajello, M., Allafort, A., et al. 2012a ApJ, 755, 164
  10. [10]Ackermann, M., Ajello, M., Atwood, W. B., et al. 2012b ApJ, 750, 3
  11. [11]Ackermann, M., Albert, A., Atwood, W. B., et al. 2016 A&A, 586, A71
  12. [12]Ackermann, M., Ajello, M., Albert, A., et al. 2017 ApJS, 836, 208
  13. [13]Adebahr, B., Krause, M., Klein, U., Heald, G., & Dettmar, R.-J. 2017 A&A, 608, A29
  14. [14]Ajello, M., Di Mauro, M., Paliya, V. S., & Garrappa, S. 2020 ApJ, 894, 88
  15. [15]Ajello, M., Murase, K., & McDaniel, A. 2023 ApJL, 954, L49
  16. [16]Atwood, W., Albert, A., Baldini, L., et al. 2013 arXiv e-prints, arXiv:1303.3514arxiv.org/abs/1303.3514
  17. [17]Atwood, W. B., Abdo, A. A., Ackermann, M., et al. 2009 ApJ, 697, 1071
  18. [18]Beck, R., & Berkhuijsen, E. M. 2025 A&A, 700, A198
  19. [19]Bruel, P., Burnett, T. H., Digel, S. W., et al. 2018 arXiv e-prints, arXiv:1810.11394arxiv.org/abs/1810.11394
  20. [20]Butt, Y. 2009 Nature, 460, 701
  21. [21]Calore, F., Di Mauro, M., & Donato, F. 2014 ApJ, 796, 14
  22. [22]Chen, J., & Totani, T. 2025 MNRAS, 540, 3221
  23. [23]Crocker, R. M., Krumholz, M. R., & Thompson, T. A. 2021 MNRAS, 502, 1312
  24. [26]Eichmann, B., Oikon­omou, F., Salvatore, S., Dettmar, R.-J., & Tjus, J. B. 2022 ApJ, 939, 43
  25. [27]Fletcher, A., Berkhuijsen, E. M., Beck, R., & Shukurov, A. 2004 A&A, 414, 53
  26. [28]Fukui, Y., Sano, H., Sato, J., et al. 2017 ApJ, 850, 71
  27. [29]Gabici, S. 2022 A&AR, 30, 4
  28. [30]Ginzburg, V. L., & Syrovatskii, S. I. 1964, The Origin of Cosmic Rays
  29. [31]Goetz, M., Downes, D., Greve, A., & McKeith, C. D. 1990 A&A, 240, 52
  30. [33]Iwasawa, K., Norman, C., Gilli, R., Gandhi, P., & Pérez-Torres, M. A. 2023 A&A, 674, A77
  31. [34]Kalberla, P. M. W., Burton, W. B., Hartmann, D., et al. 2005 A&A, 440, 775
  32. [35]Kerr, M., Wadiasingh, Z., Lavrov, A., et al. 2026 The Astrophysical Journal Letters, 1001, L36
  33. [36]Krumholz, M. R., Crocker, R. M., Xu, S., et al. 2020 MNRAS, 493, 2817
  34. [37]Lacki, B. C., & Thompson, T. A. 2013 ApJ, 762, 29
  35. [38]Longair, M. S. 2011, High Energy Astrophysics
  36. [39]Mattox, J. R., Bertsch, D. L., Chiang, J., et al. 1996 ApJ, 461, 396
  37. [40]Melo, V. P., Pérez García, A. M., Acosta-Pulido, J. A., Muñoz-Tuñón, C., & Rodríguez Espinosa, J. M. 2002 ApJ, 574, 709
  38. [41]Merten, L., Becker Tjus, J., Eichmann, B., & Dettmar, R.-J. 2017 Astroparticle Physics, 90, 75
  39. [42]Peretti, E., Blasi, P., Aharonian, F., & Morlino, G. 2019 MNRAS, 487, 168
  40. [43]Persic, M., Rando, R., & Rephaeli, Y. 2025 A&A, 701, A154
  41. [44]Persic, M., & Rephaeli, Y. 2010 MNRAS, 403, 1569
  42. [45]Persic, M., & Rephaeli, Y. 2022 A&A, 666, A167
  43. [46]Persic, M., Rephaeli, Y., & Rando, R. 2024 A&A, 685, A47
  44. [47]Relaño, M., & Kennicutt, Jr., R. C. 2009 ApJ, 699, 1125
  45. [48]Revnivtsev, M., Sazonov, S., Churazov, E., et al. 2009 Nature, 458, 1142
  46. [49]Sakamoto, K., Mao, R.-Q., Matsushita, S., et al. 2011 ApJ, 735, 19
  47. [50]Sanders, D. B., Mazzarella, J. M., Kim, D.-C., Surace, J. A., & Soifer, B. T. 2003 AJ, 126, 1607
  48. [52]Silva, L., Granato, G. L., Bressan, A., & Danese, L. 1998 ApJ, 509, 103
  49. [53]Smith, M. W. L., Eales, S. A., Gomez, H. L., et al. 2012 ApJ, 756, 40
  50. [54]Strickland, D. K., & Heckman, T. M. 2007 ApJ, 658, 258
  51. [55]Strong, A. W., & Moskalenko, I. V. 1998 ApJ, 509, 212
  52. [56]Strong, A. W., Moskalenko, I. V., & Ptuskin, V. S. 2007 Annual Review of Nuclear and Particle Science, 57, 285
  53. [57]Sudoh, T., Totani, T., & Kawanaka, N. 2018 PASJ, 70, 49
  54. [58]VERITAS Collaboration, Acciari, V. A., Aliu, E., et al. 2009 Nature, 462, 770
  55. [59]Weiß, A., Neininger, N., Hüttemeister, S., & Klein, U. 2001 A&A, 365, 571
  56. [60]Wik, D. R., Lehmer, B. D., Hornschemeier, A. E., et al. 2014 ApJ, 797, 79
  57. [61]Williams, P. K. G., & Bower, G. C. 2010 ApJ, 710, 1462
  58. [62]Wood, M., Caputo, R., Charles, E., et al. 2017 in International Cosmic Ray Conference, Vol. 301, 35th International Cosmic Ray Conference (ICRC2017), 824
  59. [64]Xing, Y., Wang, Z., Zheng, D., & Li, J. 2023 ApJL, 945, L22
  60. [65]Yoast-Hull, T. M., Everett, J. E., Gallagher, III, J. S., & Zweibel, E. G. 2013 ApJ, 768, 53
  61. [66]Zabalza, V. 2015 in International Cosmic Ray Conference, Vol. 34, 34th International Cosmic Ray Conference (ICRC2015), 922

Paper details

vlnc:W_kdcye92j
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 1

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents