2FHL J1745.1–3035: A Newly Discovered, Powerful Pulsar Wind Nebula Candidate

The source is in the field of view of three XMM-Newton observations.

• 1.  
  The first observation we studied is a 16 ks one, taken in 2017 and aimed specifically at finding an X-ray counterpart for 2FHL J1745.1–3035 (observation ID: 0782170601; PI: M. Ajello). In the XMM-Newton cameras' field of view, the source was found at a distance from the 2FHL centroid (and therefore at an off-axis angle) ∼35; the J2000 coordinates of the XMM-Newton counterpart centroid are R.A. = 17:45:07.99 and decl. = −30:39:06.17. This source is reported in the fourth release of the XMM-Newton source catalog, ¹³ 4XMM-DR13 (Zolotukhin et al. 2017; Traulsen et al. 2020; Webb et al. 2020), with source ID 4XMM J174507.9-303906 (http://xmm-catalog.irap.omp.eu/source/201032613010001). In Figure 3, we report an image of the XMM-Newton pn 0.3–10 keV observation with the Fermi–LAT 90% confidence positional uncertainty overlaid; it can be seen that the X-rays allow us to reliably identify a bright counterpart of the 2FHL source (magenta circle in Figure 3). As a safety check, we nonetheless verified the properties of the faint source that can be noticed in the upper right part of the image. This a soft X-ray source with no emission above 2 keV, which has a bright (G_AB = 12.5) counterpart in the Gaia DR3 catalog (https://gea.esac.esa.int/archive/; source ID: Gaia DR3 4056681582705849600 Gaia Collaboration et al. 2016, 2023), where the source is flagged as an F-G class star (T_eff ∼ 6000 K) with 100% probability. We thus safely rule it out as a possible counterpart of the 2FHL source. The detection of a bright, compact X-ray counterpart strongly supports a scenario where the 2FHL–detected emission does not have the same origin as the H.E.S.S. one, which would explain why the two objects have such different SEDs. We also note that the following Chandra and NuSTAR observations have been performed targeting the XMM-Newton–detected object. Finally, we used the Aladin web tool ¹⁴ to search for possible counterparts of the X-ray source, but none was identified in the optical images down to magnitudes R_AB ∼ 22. As reported in Table 1, this observation was affected by strong flares, therefore the observation MOS (pn) net exposure time is of ∼9.5 ks (∼6.5 ks).
• 2.  
  The detection of an X-ray source in the XMM-Newton 2017 image allowed us to perform a search in the XMM-Newton archive aimed at finding other observations of the object. We find two such observations: the first one was taken in 2001, is 8 ks long, and targeted the pulsar PSR B1742-30 (also known as PSR J1745-3040; observation ID: 0103261301; PI: F. Jansen). In this observation, our source is found at an off-axis angle ∼10$^{\prime}$. We performed a cross-match between the X-ray coordinates of our target and the Australia Telescope National Facility (ATNF) Pulsar Catalog (Manchester et al. 2005). ¹⁵ The closest source to our X-ray target is indeed PSR B1742-30. Based on the information in the ATNF Pulsar Catalog, PSR B1742-30 is at an ∼200 pc distance from the Earth and has an estimated age of 546 kyr. While a displacement of $10^{\prime}$ (which is 0.6 pc) could be reasonably expected as a consequence of a pulsar kick, a pulsar that old would be expected to be accompanied by a bow-shock nebula at the pulsar position in X-rays, while the relic nebula left behind would not be as compact as the source that we observe either in the X-rays or in the 2FHL catalog. Furthermore, the pulsar E-dot as reported in the ATNF Pulsar Catalog is $\dot{E}$ = 8.5×10³³ erg s⁻¹, a value significantly lower than those commonly measured in pulsars powering PWNe (∼10³⁶ erg s⁻¹; see, e.g., Mattana et al. 2009; Acero et al. 2013). We therefore conclude that the source we detect is likely not associated with the known pulsar PSR B1742-30, and would instead be a newly discovered pulsar.
• 3.  
  More recently, in 2021, our X-ray source was serendipitously observed during a 23 ks observation of the Milky Way plane (observation ID: 0886010401; PI: G. Ponti). In this observation, the object we are studying is located at an off-axis angle ∼9$^{\prime}$.

Zoom In Zoom Out Reset image size

For all of the observations, the XMM-Newton standard data reduction procedure has been performed using the SAS software, version 19.0.0. To extract the source spectra, we used a circular region with radius r_src = 20'', while the background spectra were extracted from a circle having radius r_bkg = 45'' located in proximity to the source and visually inspected to avoid contamination from bright targets. Finally, given the different exposure times of the three observations, we adopted a different spectral binning for each of them: the 2001 and 2017 spectra are binned with 7 counts per bin, while in the 2021 observation, we binned the MOS spectra with 10 counts per bin and the pn spectrum with 15 counts per bin.

2FHL J1745.1–3035 was observed in 2022 May with a ∼25 ks Chandra ACIS-I observation (PI: S. Marchesi). The data set, obtained by the Chandra X-ray Observatory, is contained in the Chandra Data Collection (CDC) cdc.191 doi:10.25574/cdc.191. We performed a standard data reduction using the CIAO software (Fruscione et al. 2006), version 4.14. We then followed the standard astrometric correction procedure reported in the Chandra threads ¹⁶ to correct the astrometry of our Chandra image by using as a reference the USNO-A2.0 astrometric standards catalog (Monet 1998). We find and apply offset corrections ΔR.A. = R.A._Chandra—R.A._USNO = 0471 and Δdecl. = decl._Chandra —decl._USNO = 0278. We detect the Chandra counterpart of 4XMM J174507.9-303906 at the position R.A. = 17:45:08.03 and decl. = −30:39:06.76, just 07 away from the 4XMM source centroid.

In Figure 4, we report a zoom-in of 4XMM J174507.9-303906 in the Chandra 2–7 keV image (top left), in the smoothed XMM-Newton PN 2–10 keV image (top right), in the DSS2-red survey (bottom left; McLean et al. 2000) and in the UKIRT Infrared Deep Sky Survey (UKIDSS) DR11PLUS J, H, and K band survey (bottom left, center, and right; Warren et al. 2007). As can be seen, no clear counterpart of the X-ray source is detected in the optical, where the two closest targets to the X-ray source are two objects both classified as stars with 100% probability in the Gaia DR3 catalog (these GAIA sources are plotted as cyan circles in the figure). There are instead two faint near-infrared sources in the proximity (distance ≲1'') of the Chandra centroid, as reported in the UKIDSS–DR6 Galactic Plane Survey (Lucas et al. 2008). ¹⁷ We flag these sources as U1 (UKIDSS ID: J174507.94+429192822.5, H_AB = 18.1 ± 0.2; K_AB = 17.2 ± 0.2) and U2 (UKIDSS ID: J174508.02+429192824.8; K_AB = 17.2 ± 0.2) in Figure 4. The faintness of these sources with respect to the X-ray brightness of our target, and the high source density of this field in the near-infrared (as a reference, there are 28 UKIDSS sources within a 10'' radius from the Chandra centroid) do not allow us to reliably associate either of the UKIDSS sources to the X-ray object.

Zoom In Zoom Out Reset image size

We used the CIAO tool specextract to extract the Chandra source spectrum from a circle having radius r = 3'' , while the background spectrum was extracted from a r = 25'' circular region located nearby the source. The Chandra spectrum was then binned with 10 counts per bin. In Section 4.3, we will investigate in detail the presence of extended emission in the Chandra image, and the subsequent implications on the nature of 2FHL J1745.1–3035.

2FHL J1745.1–3035 was observed by NuSTAR in 2023 March. Our source is located in close proximity of the Galactic Center (longitude 358481127, latitude −0804527), which means that its NuSTAR observations are strongly affected by stray light (Grefenstette et al. 2021). This is clearly visible in Figure 5, where we also report the region files which we used to extract the source and background spectra, with a radius of 30'' and 45'', respectively. Despite its challenging location, the source is clearly detected in both NuSTAR cameras. Due to the high background, we binned the spectra with 75 counts per bin, to ensure an adequate signal-to-noise ratio.

Zoom In Zoom Out Reset image size

We perform single-epoch fits to search for potential variability in flux and photon index, both commonly observed in the X-ray spectra of PWNe (e.g., Pavlov et al. 2001; Klingler et al. 2018; Guest & Safi-Harb 2020). In particular, X-ray variability in PWNe has already been observed over timescales that vary from months to several years (e.g., the PWN embedded in the SNR Kes 75; Ng et al. 2008; Livingstone et al. 2011; Reynolds et al. 2018), and it has been linked to physical processes in both the pulsar and the surrounding nebula. Such processes include synchrotron cooling of the PWN flow, resulting in a steepening of the X-ray photon index, or Kelvin–Helmoltz instabilities in the shocked winds (Pavlov et al. 2001).

We note that we also searched for short-term variability within the observations, but we did not find any significant evidence of it. Interestingly, the 2017 and 2021 XMM-Newton observations of 4XMM J174507.9-303906 are flagged as variable by the automatic pipeline used in the 4XMM catalog (https://xcatdb.unistra.fr/4xmmdr13/xcatindex.html?detid=107821706010001 for the 2017 observation and https://xcatdb.unistra.fr/4xmmdr13/xcatindex.html?detid=108860104010001 for the 2021 observation). However, we performed a multiband, multi-instrument analysis of the 2017 observation and found no evidence of simultaneous variability in the three XMM-Newton cameras (MOS1, MOS2 and PN) up to 6 keV. At energies greater than 6 keV, a flare is detected in all three cameras, but the same flare is still detected even at energies >10 keV, where the contribution from the source becomes negligible and the emission is dominated by the background. We thus conclude that, in the 2017 observation, the variability reported in the 4XMM catalog can be ascribed to an episode of high background activity that commonly affects XMM-Newton observations. As for the 2021 observation, we compared the MOS1, MOS2, and pn light curves in different bands and once again found no evidence of consistent cross-instrument variability. We report in Appendix B a summary of the light curves we produced for the 2017 and 2021 observations.

For each epoch of the X-ray observations, we fit the available data with an absorbed power-law model, where the column density of the absorber was left free to vary. Since sources nearby the Galactic Center have been shown to be potentially affected by dust scattering (see, e.g., Jin et al. 2018), we performed a consistency check by fitting our data a first time using the XSPEC XScat model (Smith et al. 2016), and then a second time using the standard photoelectric absorption model phabs. Since the results are fully equivalent, we report below those obtained using the standard phabs model.

When fitting XMM-Newton and NuSTAR observations, we also included cross-normalization constants to account for possible differences between different cameras. We also included in the model the absorption due to our own Galaxy, N_H,Gal = 9.7 × 10²¹ cm⁻² (Kalberla et al. 2005). The best-fit results for each epoch are reported in Table 2, while in Figure 6 we plot the best-fit photon index, line-of-sight column density, and 2–10 keV flux we measure in each epoch. The single-epoch spectra and the relative best-fit models are reported in Appendix A.

Zoom In Zoom Out Reset image size

Table 2. Best-fit Results of the Single-epoch X-Ray Spectral Fits of 2FHL J1745.1–3035

Instrument	Observation date	χ2/d.o.f.	Photon Index	NH,los	2–10 keV flux
	(UTC)			1022 cm−2	10−12 erg s−1 cm−2
XMM-Newton 1	2001-03-21	45.1/59	${0.68}_{-0.72}^{+0.87}$	${6.4}_{-3.3}^{+4.8}$	${1.31}_{-1.02}^{+0.15}$
XMM-Newton 2	2017-04-03	68.0/87	${0.67}_{-0.41}^{+0.46}$	${6.1}_{-1.9}^{+2.5}$	${0.95}_{-0.58}^{+0.05}$
XMM-Newton 3	2021-03-16	57.8/67	${0.42}_{-0.36}^{+0.39}$	${4.2}_{-1.4}^{+1.7}$	${0.99}_{-0.31}^{+0.07}$
Chandra	2022-05-20	4.7/10	${0.21}_{-0.57}^{+1.34}$	<6.4	<0.35
NuSTAR	2023-03-27	30.0/46	${1.99}_{-0.32}^{+0.37}$	${18.2}_{-10.7}^{+14.2}$	${0.54}_{-0.45}^{+0.04}$

Note. All spectra are fitted with an absorbed single-power-law model, where N_H,los is the line-of-sight column density of the absorber. χ²/d.o.f. is the reduced χ² of the fit, where d.o.f. are the degrees of freedom. In the XMM-Newton (NuSTAR) data the 2–10 keV flux is computed from the pn (FPMA) data set.

Download table as:  ASCIITypeset image

As can be seen, the photon index and line-of-sight column density do not vary significantly in the three XMM-Newton observations, as well as in the Chandra observation. In all cases, the X-ray photon index is significantly hard (varying in the range Γ_X = [0.2–0.7]), and the X-ray emission is absorbed by material having column density N_H,los∼4–6 ×10²² cm⁻². The NuSTAR best-fit results are different: most importantly, we measure an X-ray photon index that is significantly softer than those measured by the 0.5–10 keV instruments, being Γ_NuS = 1.99${}_{-0.32}^{+0.37}$. The fit to the NuSTAR data also gives a slightly larger line-of-sight column density than the one measured with Chandra and XMM-Newton, which is N_H,los,Nus = 1.8${}_{-1.1}^{+1.4}$ ×10²³ cm⁻². Such a discrepancy can, however, be explained with the fact that NuSTAR lacks the <3 keV coverage that is key to reliably constrain line-of-sight column densities smaller than 10²³ cm⁻².

2FHL J1745.1–3035 shows some marginal evidence for flux variability in the 2–10 keV band. In particular, the 2022 Chandra and the 2023 NuSTAR data are ∼50% fainter than the three XMM-Newton observations taken between 2001 and 2021. However, the relatively large uncertainties on the flux measurements prevent us from studying this tentative trend in further detail. We also note that, in the γ-rays, no information on source flux variability is reported in the 2FHL catalog. There is also no observational evidence for flux variability in the 4FGL source associated with the H.E.S.S. survey and reported in the Fermi–LAT data release 3 catalog (DR3, Abdollahi et al. 2022), which is an incremental version of the 4FGL catalog published in 2020 and is based on the first 12 yr of Fermi–LAT observations between 50 MeV and 1 TeV. Specifically, the source variability index (computed as the sum of the log(likelihood) difference between the flux fitted in each time interval and the average flux over the full catalog interval) is 5.1. As a reference, a value greater than 18.48. ¹⁸ over 12 intervals indicates <1% chance of being a steady source.

Since we did not find any significant line-of-sight column density variability between the four soft X-ray observations, we jointly fit all X-ray spectra by tying this parameter among the different observations. Since we instead found evidence for potential X-ray flux variability, we added to the model a cross-normalization component to take this intrinsic variability into account. We then first fit the spectra with the same absorbed power-law model we used in the single-epoch analysis.

We measure a photon index Γ_X = 1.53${}_{-0.15}^{+0.16}$ and a line-of-sight column density N_H,los = 9.9${}_{-1.1}^{+1.3}$ ×10²² cm⁻². We report in Figure 7, left panel, the single-power-law best-fit model; significant residuals are visible in the ratio between the model and the data above 10 keV. Furthermore, as expected, the best-fit photon index we measure lies between the values we obtain in the soft X-rays (Γ_X ∼ 0.5, as shown in Table 2) and the value measured using NuSTAR, Γ_X = 1.99${}_{-0.37}^{+0.32}$.

Zoom In Zoom Out Reset image size

Given this observational evidence, we performed a new fit, this time using a broken-power-law model: notably, the presence of a break at ∼5–15 keV in the X-ray spectra of PWNe has been reported in several works (see, e.g., An et al. 2014; Nynka et al. 2014; Madsen et al. 2015, 2020; An 2019; Bamba et al. 2022). Such a variation in Γ_X is an intriguing challenge to our understanding of PWN emission mechanisms, since PWNe SED models do not properly fit such an X-ray feature (see, e.g., Tanaka & Takahara 2011). Consequently, additional model components, such as for example different electron injection spectra or radially dependent PWN parameters would be required.

The parameterization of the broken-power-law is

$$\begin{eqnarray}\displaystyle \frac{{dN}}{{dE}}=\left\{\begin{array}{cc}{{KE}}^{-{{\rm{\Gamma }}}_{1}} & \mathrm{if}\ E\leqslant {E}_{{\rm{b}}}\\ {{KE}}_{{\rm{b}}}^{{{\rm{\Gamma }}}_{2}-{{\rm{\Gamma }}}_{1}}{(E/1\,\mathrm{keV})}^{-{{\rm{\Gamma }}}_{2}}) & \mathrm{otherwise}\end{array}\right..\end{eqnarray} \tag{ 1 }$$

Where Γ₁ and Γ₂ are the low- and high-energy photon indices, K is the normalization parameter, and E_b is the energy of the break.

The results of this fit are reported in Figure 7, right panel. As it can be seen, the data are much better constrained across the whole energy range, and in particular at energies E>10 keV. We also measure a significant improvement in the fit, from χ²/d.o.f. = 248.1/264 for the single-power-law model to χ²/d.o.f. = 217.0/262 for the broken-power-law model. The photon index at energies lower than the break is Γ₁ = 0.56${}_{-0.41}^{+0.42}$, in good agreement with the photon indices we measured in the Chandra and XMM-Newton spectra, while the second photon index is Γ₂ = 1.87${}_{-0.24}^{+0.60}$, a value consistent with the one measured in the NuSTAR spectra. Coherently, the best-fit break energy is E_b = 7.1${}_{-0.9}^{+3.0}$ keV, which is where the NuSTAR contribution becomes the dominant one. In Figure 8, we report the confidence contours of the break energy as a function of both photon indices: Γ₁ and Γ₂ are different at the >99% confidence level, a result that further favors the broken-power-law scenario over the single-power-law scenario. We also note that the best-fit column density obtained with the broken-power-law fit, N_H,los = 5.1${}_{-1.4}^{+1.4}$×10²² cm⁻², is significantly larger than the one expected on the line of sight from the neutral Hydrogen measurements by (Kalberla et al. 2005, N_H,Gal = 9.7 × 10²¹ cm⁻²). Such a result supports a scenario where the X-ray emission is taking place in a dense, gas-rich environment.

Zoom In Zoom Out Reset image size

Finally, as a consistency check, and in particular to partially rule out an SNR origin for 2FHL J1745.1–3035, we fitted our data with the nei nonequilibrium ionization collisional plasma model, which is commonly used when modeling the X-ray spectra of SNRs (see, e.g., Harrus et al. 2001; Lazendic & Slane 2006; Katsuda et al. 2009; Prinz & Becker 2013; Leahy et al. 2020; Eagle et al. 2022). We report the best-fit results in Table 3: while the best-fit statistic (χ²/d.o.f. = 225.3/262) is significantly better than the one obtained using the single-power-law model, and fairly consistent with the one we obtain using the broken-power-law model, the best-fit temperature (kT = 34.4${}_{-9.1}^{+24.4}$ keV) is fully unphysical. As a reference, Leahy et al. (2020) analyzed the X-ray spectra of 43 SNR of different ages, explosion energies, and circumstellar medium densities: none of them was found having kT ≥ 4 keV. Such a result reliably rules out a thermal SNR scenario for 2FHL J1745.1–3035.

Table 3. Best-fit Results of the Multi-epoch X-Ray Spectral Fits of 2FHL J1745.1–3035, for the Three Different Models Discussed in the Text

Model	χ2/d.o.f.	NH,los	Γ1	Γ2	Eb	kT	τ	Z
		1022 cm−2			keV	keV	1010 s cm−3	Z⊙
Single Power Law	248.1/264	${9.0}_{-1.1}^{+1.3}$	${1.53}_{-0.15}^{+0.16}$	⋯	⋯	⋯	⋯	⋯
Broken Power Law	217.0/262	${5.1}_{-1.4}^{+1.4}$	${0.56}_{-0.41}^{+0.42}$	${1.87}_{-0.24}^{+0.60}$	${7.1}_{-0.9}^{+3.0}$	⋯	⋯	⋯
nei	225.3/262	${8.9}_{-1.4}^{+0.9}$	⋯	⋯	⋯	${34.4}_{-9.1}^{+24.4}$	<5.95	${0.24}_{-0.11}^{+0.11}$

Note. χ²/d.o.f. is the reduced χ² of the fit, where d.o.f. are the degrees of freedom; N_H,los is the line-of-sight column density of the absorber; Γ₁ is the photon index of the single-power-law model, and the photon index before the break of the broken-power-law model; E_b and Γ₂ are the energy break and the photon index at energies larger than it in the broken-power-law model, respectively; finally, kT, τ, and Z are the best-fit temperature, ionization timescale, and metallicity, derived using the nei thermal model.

Download table as:  ASCIITypeset image

The X-ray counterpart of 2FHL J1745.1–3035 does not show clear evidence for extended emission in the NuSTAR image or in the XMM-Newton images, which implies that the X- and γ-ray emitter must be compact. Extended PWN emission is however commonly observed with Chandra, and indeed a first visual evidence is visible in our Chandra ACIS-I observation as well. As shown in the top panels of Figure 9, both the unbinned Chandra image and the smoothed one present evidence for asymmetric extended emission, with a faint tail of emission detected southward of the source center.

Zoom In Zoom Out Reset image size

To quantitatively search for extended emission in 2FHL J1745.1–3035, we perform a comparison between the 2–7 keV surface brightness radial distribution and the one obtained from a simulated PSF in the same band, computed using the ChaRT and MARX 5.5.1 tools (see e.g., Fabbiano et al. 2017; Jones et al. 2020; Ma et al. 2020; Traina et al. 2021; Silver et al. 2022, for a detailed description of this same approach in nearby active galactic nuclei). We report in Figure 9, bottom left panel, the 2–7 keV surface brightness radial profile of 2FHL J1745.1–3035, which we computed from a set of 11 annuli, where the two radii increase by 05 in each consecutive annulus, and the maximum external radius is 65, and the corresponding expected PSF profile which one would observe for a point-like source. As it can be seen, this quantitative analysis confirms the first qualitative impression, since the source surface brightness exceeds the one predicted for a point-like scenario up to ∼5''. As a caveat, we note that it has been shown (Primini et al. 2011), that that the ray tracing simulations generated by MARX will have a roughly correct total intensity but tend to have PSF wings that are broader than observations and a PSF core narrower than observed, an effect that could in principle affect our simulation as well.

Finally, the low count statistic of the Chandra spectrum prevents us from performing an energy-dependent surface brightness profile, as well as a spatially resolved spectral fit. Nonetheless, a simple tricolor image such as the one we report in Figure 9, bottom right panel, gives a first indication of the presence of two distinct sources of emission: a harder, unresolved one at the center (likely the pulsar) and a fainter, softer diffuse emission in the outskirts (in our scenario, the pulsar wind nebula). We note that Li et al. (2008) analyzed the Chandra and XMM-Newton spectra of a sample of pulsars and PWNe, and found tentative evidence for both direct correlation between the X-ray photon index of the pulsar and its age (i.e., younger pulsars have harder photon indices) and anticorrelation between the PWN photon index and the pulsar age (i.e., younger PWNe have softer photon indices). Both these trends support a young PWN scenario for 2FHL J1745.1–3035, based on the tentative observational evidence we find in the Chandra image.

In the case of two-leptonic particle distributions, the radio–X-ray–2FHL data combination is reasonably reproduced (as shown in the left panel of Figure 10), while no combination of parameters can adequately describe the SED composed by the radio, X-ray, 4FGL, and H.E.S.S. data and, in particular, the 4FGL data (Figure 11, left panel). This is due to the very low radio flux upper limit that forces the electron index to be harder than what is inferred by the 4FGL data (∼2). Introducing another fitting parameter such as the index to the energy cutoff (i.e., an index after the break deviating from the assumed value of 1 here) may be able to reconcile the poor fitting in the Fermi band. An index after the cutoff that is smaller than 1, for instance, would soften the cutoff shape in the synchrotron peak for the low-energy population (Population 1 in left panel of Figure 11). A final possibility would be that the particle population is more complicated than what is explored here.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As mentioned above, the radio, X-ray and 2FHL γ-ray data combination is instead reasonably reproduced by the two-leptonic model (left panel of Figure 10). Assuming the two-leptonic scenario for both data combinations, the predicted properties are consistent with an energetic, young, and/or more evolved PWN where the extended 4FGL and H.E.S.S. γ-ray data may be an extended relic PWN and the point-like 2FHL emission may be a compact, young PWN located close to the pulsar.

In order to explore a hadronic scenario to the broadband data, we fit a pion decay-emission model to characterize the GeV and TeV data together combined with the previously derived leptonic component to characterize the radio and X-ray data. As shown in the right panels of Figures 11 and 10, both data combinations are adequately characterized in the lepto-hadronic scenario, and in particular we are now able to efficiently describe the SED including the 4FGL data, something that could not be done with a purely leptonic model.

We note that the fits to both data sets require very large target densities: more specifically, n_H = 5000 cm⁻³ in the 4FGL+H.E.S.S. data set and n_H = 400 cm⁻³ in the 2FHL data set. The high n_H values are a requirement if the total proton energy W_p is not to exceed the total energy expected to be released from a typical supernova (SN) explosion ${E}_{\mathrm{SN}}={10}^{51}\,$ erg. Even for the highest estimate n_H = 5000 cm⁻³ which characterizes the 4FGL+H.E.S.S. data, W_p ∼ 6 × 10⁵⁰ erg challenges a hadronic γ-ray origin. However, both n_H and W_p may be smaller if there is a significant contribution from nonthermal bremsstrahlung emission (true for both data combinations). A similar representation is presented in Bamba et al. (2009), showing that nonthermal bremsstrahlung radiation can dominate the high-energy emission.

In summary, the SED fitting analysis shows that the two-leptonic model efficiently describes the radio, X-ray, and 2FHL γ-ray data combination, and in particular provides the most accurate modelization of the 2FHL γ-ray spectrum. On the other hand, the lepto-hadronic representation more efficiently describes the radio, X-ray, and 4FGL+H.E.S.S. extended γ-ray emission, as shown in the right panel of Figure 11, under the condition that the ambient particle density is high, n_H = 5000 cm⁻³. We discuss the implications for the best-fit models in more detail in the following section.

In the previous sections, we attempted to determine the most likely origin of the unique and puzzling high-energy source 2FHL J1745.1–3035 through detailed broadband modeling. We tested two data combinations assuming two energetic scenarios: (1) the 4FGL+H.E.S.S. extended γ-ray data is related to the point-like X-ray source or (2) the point-like 2FHL γ-ray data is related to the point-like X-ray source. In the previous sections, we determined that the most favorable models correspond to the lepto-hadronic model for the 4FGL+H.E.S.S. data set, while a two-leptonic model can better characterize the 2FHL data set.

The lepto-hadronic model accurately characterizes the source responsible for the 4FGL+H.E.S.S. observations, requiring a large target proton density n_H = 5000 cm⁻³. If the distance assumed to the source corresponds to the one reported in Bamba et al. (2009), which is d = 8.5 kpc, then 2FHL J1745.1–3035 is likely located within the Galactic Center and thus among dense molecular material. In fact, HESS J1745–303 is spatially coincident with a molecular cloud (Aharonian et al. 2008) consistent with the estimate n_H = 5000 cm⁻³, making this scenario plausible.

The two-leptonic model best characterizes the emission responsible for 2FHL observations and infers a magnetic field strength that is low compared to the average value for the ISM (∼2 μG, Han & Qiao 1994). If we assume that all of the considered data are associated in some way, a depicted scenario is one where the extended γ-ray emission (4FGL+H.E.S.S.) represents an older, diffuse, and softer spectral component whereas the point-like γ-ray emission (2FHL) represents a younger, more compact, and harder spectral component. This type of energy-dependent morphology is often observed in evolved PWNe (e.g., Aharonian et al. 2006; H.E.S.S. Collaboration et al. 2012), but could also be explained by an energetic distant SNR where the highest energy particles have already escaped and are diffusing into the ambient medium (e.g., Eagle et al. 2020). Another possible explanation includes both PWN and SNR contributions, whether they are components of the same system or not. We remark that the models do not consider any electron-to-proton ratio, acknowledging the possibility the emission may be coming from separate, unrelated sources.

A final scenario that would be compatible with the hard X-ray spectrum we measure is a compact accreting binary system where the compact object, either a neutron star or black hole, is accreting material from a stellar companion. SS 433 represents the most well-known example of such a system and is detected from radio to TeV γ-rays with extended emission originating from the outflowing jets of the accreting compact object (HAWC Collaboration et al. 2018). Accreting compact binary systems like SS 433 can be extremely luminous in the X-ray band (≳10³⁹ erg s⁻¹), indicating the the total jet power is much higher (Sudoh et al. 2020). The X-ray spectrum is often modeled as synchrotron emission as in the case for SS 433 (Safi-Harb et al. 2022), and can dominate over the Inverse Compton emission if the magnetic field strength is sufficiently high. Last, the X-ray emission of these systems can be variable on the timescales of days to hours. The lack of both a clear optical counterpart and of significant X-ray variability overt timescales of hours (i.e., within each observation) are however two elements that make this scenario less likely with respect to the compact PWN one.

While the models presented above provide the best representations to the broadband data, there are other limitations to be considered. The most uncertain assumption is the distance used for the modeling, which we assumed to be the same distance reported in Bamba et al. (2009) for the H.E.S.S. source, which is d = 8.5 kpc. It is possible the source generating the observed emission is closer, which would enable lower target densities and lower total particle energies. Second, other particle distributions and hence properties are possible. An alternative broadband representation includes a nonthermal bremsstrahlung component that yields reasonable results (Bamba et al. 2009). Finally, we note that a single population (leptonic or hadronic) cannot explain the broadband data without far exceeding 10⁵¹ erg in total particle energies at a distance of 8.5 kpc.

We remark that 2FHL J1745.1–3035 is unlikely to be of extragalactic origin based on the source selection from the 2FHL catalog, as discussed in Section 1. The spectral index for this source at energies above 50 GeV is much harder than the average spectral index observed for blazars above 50 GeV. Furthermore, 2FHL J1745.1–3035 is located along the Galactic plane and is potentially associated with significant extended GeV and TeV emission (4FGL and H.E.S.S. counterparts). Based on this evidence, we conclude that the most plausible scenario for 2FHL J1745.1–3035 is to be the descendant of a Galactic supernova explosion such as a PWN, SNR, or neutron star or black hole with a stellar companion, but current observations and the presented models prevent us from firmly establishing an origin.