Discovery of A Compact X-ray Object with A 614s Periodicity in the Direction of the Galactic center

We report on analysis of X-ray, optical and radio observations of the previously overlooked X-ray source 2CXO\,J174517.0$-$321356 located just 3.2$^{\circ}$ away from the Galactic center. Timing analysis of X-ray observations of the source with \textit{XMM-Newton} reveals periodic pulsations with periods of 1228\,s and 614\,s, with the latter being tentatively considered fundamental. On the other hand, an observation of the object with \textit{NuSTAR} reveals hard thermal-bremsstrahlung spectrum. Inspection of the archival VLT image reveals, however, no obvious optical counterpart down to $\rm{R}>25\,$mag. Observations made with ATCA showed a possible faint radio counterpart with a positive spectral index ($\alpha>0.51$) between 1--3\,GHz, but follow-up ATCA and VLA observations at frequencies between 4.5--10\,GHz and 3--22\,GHz, respectively, could not detect it. Given the properties in these three bands, we argue that the most likely origin of the X-ray source is emission from a new intermediate polar close to the Galactic center. Alternatively, and less likely, it is an ultra-compact X-ray binary, which is one of the most compact X-ray binaries.

Identification of multiple serendipitously discovered X-ray sources in crowded regions of the Galactic plane, and especially the center is notoriously difficult (Hong et al. 2016). This is due to the fact that most of the sources are faint in X-rays which precludes sensitive searches for possible pulsed signal and detailed broadband spectral studies. Therefore, X-ray data alone is most often not sufficient to establish the origin of the source. Additionally, density of optical and infrared sources together with high absorption in the Galactic plane make it complicated to identify and study properties of potential counterpart sources. In fact, even detection of periodic signals in the X-ray band is often insufficient to establish the origin of an X-ray source as characteristic spin and orbital periods of NS (neutron star) and WD (white dwarf) X-ray binaries partly overlap (hundreds of seconds to thousands of seconds).
For example, 4U 1820−30 is the most compact X-ray binary so far, but the orbital nature of 4U 1820−30's 685 s signal discovered in the EXOSAT data was not obvious. It was deduced based on its high L X , period stability within one year and binary evolution models (Stella et al. 1987). Another example is the most compact binary known to date RX J0806.3+1527. Its 321 s signal detected in the ROSAT data was mistakenly treated as a rotational period (Israel et al. 1999) before its orbital nature was deduced from a series of spectroscopic observations (Israel et al. 2002;Roelofs et al. 2010). Misidentifications of rotational periodic signals occurred too. For example, although initially different scenarios (Farrell et al. 2010;Malizia et al. 2010) were suggested, AX J1740.2−2903 (or 2XMM J174016.0−290337) was finally found to be an intermediate polar (IP) by optical spectroscopic observations (Masetti et al. 2012;Halpern & Gotthelf 2010). It is located 1.2 • away from the Galactic center and has a 626 s modulation in its XMM-Newton light curve. Another example is OGLE-UCXB-01 (Pietrukowicz et al. 2019), which is also suspected to be an IP (Peng & Shen 2021), despite its name.
Here, we report a newly discovered periodic variability with period 614 s from a previously overlooked X-ray source 2CXO J174517.0−321356 (R.A., Dec=266.32087 • , −32.23237 • ; J1745−3213 hereafter, Gong 2019; Evans et al. 2019) which is 3.2 • away from the Galactic center. By analysing its X-ray, optical and radio data, we argue the signal is probably due to a rotational period making it one of the IPs with confirmed periods in the direction of the Galactic center, or less likely, an orbital period making it one of the most compact X-ray binaries 1 .

Spectral Analysis
For XMM, the MOS data were reduced by SAS 19 (Gabriel et al. 2004) with standard procedures. The two event files of MOS1 and MOS2 were generated by emchain, filtered by gti files, which were generated by espfilt automatically, and then barycentric corrected by barycen and DE405 ephemeris. Source photons were extracted between 0.3-10 keV The red ellipse (30"x40") marks the source extraction region between 0.3-10 keV for EPIC-MOS. We selected the region between the green dashed lines for the background extraction, which also avoids the source marked with an X. Right panel: The VIMOS R-band image of J1745−3213. Bias, flat-field and astrometric corrections have been applied. Color scale is fine-tuned in DS9 to highlight faint features. The red circle indicates the Chandra position.
using evselect around the position given by Chandra. Due to the large off-axis angle 2 of J1745−3213, which causes elongated images of point sources on the MOS detectors, and the existence of a source nearby (marked by the red X in the left panel of Fig. 1), we tried different regions and finally adopted an elliptical region with semi-major and semiminor axes of r a × r b =30 ′′ × 40 ′′ (MOS1+MOS2≈3,500 photons) as the source extraction region, and a non-concentric ring as the background region ( Fig. 1, left ).
We used Sherpa of CIAO 4.14 (Fruscione et al. 2006) to perform the spectral fitting. Every spectral bin had 25 counts at least. Hence, we adopted chi2datavar instead of the default chi2gehrels as the fit statistic 3 . The background was subtracted channel by channel using subtract. The two MOS spectra were fit simultaneously. Among the commonly used models 4 , we found a simple apec, thermal-bremsstrahlung, thermal-bremsstrahlung with a partial covering absorber (e.g., Haberl et al. 2002) and diskbb led to poor or unreasonable fits to the data. Among most of the models used, residuals around 6.7 keV suggested presence of an iron feature which was thus included in the fits (Fig. 2, left ). Due to its broadness (≈200 eV), the feature is likely to be a mixture of three kinds of iron lines (6.4, 6.7 and 7.0 keV). Nevertheless, adding three Gaussian components made the fitting complicated and difficult to find a global minimum. We also tried to separate the three possible lines like Tomsick et al. (2016) and Coughenour et al. (2022). However, no matter the three line energies and line widths were fixed or not, our fits could not improve significantly or be reasonable. Thus, we only adopted one single Gaussian component, but left the line width free or fixed it at 0.05 keV. Hereafter, all the models we mention include an iron feature.

Timing Analysis
We used the Stingray (Huppenkothen et al. 2019) code, which performs Z 2 n test originally introduced in Buccheri et al. (1983) to detect signals in MOS1, MOS2 and the combined data sets, respectively. We searched in the frequency range 0.0001-1.6 Hz (MOS sampling time 0.3 s) with a frequency step size of 1.2×10 −6 Hz (1/(10×ExpTime)). Assuming the folded profile can be described by a superposition of three sinusoidal components (n=3), we found the corresponding periods of the most significant signals are around 614.1 s (7.8σ) and 1228.3 s (7.3σ) in the combined data set (Fig. 3, left ). Nevertheless, the most significant signal in MOS2 data is 613.7 s (1.5σ). We then folded the X-ray light curve. According to the MOS1 data set, which has the more significant signal (614.2 s (5.5σ)), it seems there are two different peaks within one cycle (Fig. 3, right ).
Seven successful Neutron Star Interior Composition Explorer (NICER; Gendreau et al. 2016) observations were made between 2019-09-19 and 2019-11-05. We analysed the data using HEASoft 6.28 and CALDB version XTI (20200722). Photons were extracted by nicerl2 and barycentric corrected by barycorr. No significant signals around   614 s or 1228 s were detected. Since only about 15 ks raw data were generated in total and J1745−3213 was much fainter than the detection limit of NICER (Remillard et al. 2022), the data are dominated by background noise and the non-detection was not surprising. Based on the longest observation (ID=2200830102, 7ks), the upper limit of the pulsed fraction (Vaughan et al. 1994;Bachetti et al. 2021) is about 19% with a confidence level of 95%.

X-ray Observations and Data Analysis
In order to verify its periodicity and check its spectral shape in the hard X-ray band, we made a quasi-simultaneous XMM-Newton (ObsID=0865510101; PI: Gong) and NuSTAR (ObsID=30601021002; PI: Mori) observation on 2021-03-02. This time, the nearby black hole binary H1743−322, the target of the old XMM observation, was in the quiescent state.
The new XMM observation was executed in the full window mode. Its data were reduced by the same procedures in Section 2.1. J1745−3213 was on-axis and covered by MOS and PN simultaneously. We adopted a 20 ′′ source extraction region and obtained 3,600 photons (1,770 PN photons and 1,820 MOS photons, respectively) between 0.3-10 keV after the GTI correction. Note that the counts had no obvious advantage compared with the old XMM observation, but the source extraction regions were different. The background region was simply set to be a homocentric annulus with an inner radius of 20 ′′ and an outer radius of 40 ′′ because this time the X source in Fig. 1 was easily excluded.
The NuSTAR data were reduced using the two commands nupipeline and nuproducts in HEASoft 6.28 with the newest calibration files. The source spectra and photons between 3-78 keV were extracted from r=40 ′′ circular regions centered at the centroids of J1745−3213 ′ s photons, which had ≈ 6 ′′ offsets compared with the Chandra position. The  . Left panel: Z 2 n statistics obtained from the combined MOS1 and MOS2 data with 3,500 photons. The 614.1 s (7.8σ) signal is slightly stronger than the 1228.3 s signal (7.3σ) and much stronger than the 1841 s harmonic signal. Right panels: Comparison of background-subtracted pulse profiles. The MOS1 data set, which has more significant signals, shows two different peaks within 1228 s. spectra were grouped by grppha and every spectral bin had a minimum of 25 counts. The background was selected as a 4 ′ × 2.5 ′ rectangular region ≈ 2 ′ away. The barycentric correction was applied by setting the input parameter 'barycorr' of nupipeline to 'yes'.

Spectral Analysis
The MOS and PN spectra were fit simultaneously. If no parameters were fixed, BB yielded the best reduced χ 2 and a consistent fit with the previous XMM observation (the 1st part of Table.3). Both fits had a BB temperature of ≈ 2 keV. If line width was fixed, an apec model with a partial covering absorber led to a comparable reduced χ 2 , with which F-test could not distinguish which model is better. The XMM and NuSTAR spectra were fit between 0.3-20 keV simultaneously. As the photons gets harder, the continuum can not be described by BB anymore. Instead, it is characterised by a partial covering thermal-bremsstrahlung (Fig. 2, middle) or PL (the 2nd part of Table.3). We selected 3-50 keV as the hardest band in case of the significant detector background ). The NuSTAR spectrum (the 3rd part of Table.3) is described by an absorbed thermal-bremsstrahlung (Fig. 2, right ) with an unabsorbed flux and luminosity of 3.8 +0.4 −0.4 × 10 −12 erg cm −2 s −1 and 4.6 +0.5 −0.5 × 10 32 ×(D/1 kpc) 2 erg s −1 , respectively. A slightly inferior model is absorbed apec. Adding a partial covering component to either of the two models would yield a worse reduced χ 2 and let the partial covering factor hit its maximum boundary 1.0.

Timing Analysis
The MOS and PN sampling times of the new observation were 2.6 s and 73.4 ms, respectively. For the PN data set, we searched in the frequency range 0.0001-6.8 Hz with a frequency step size of 2.0 × 10 −6 Hz (1/(10×ExpTime)) via Z 2 n test (n=3). The two signals were confirmed (Fig. 4, left ). Nevertheless, we found the corresponding period of the most significant signal is around 613.5 s (4.7σ), which is weakened by strong red noise. No shorter significant signals were discovered. For the combined MOS data set, we searched in the frequency range 0.0001-0.19 Hz and found the corresponding period of the most significant signal is around 613.8 s (7.0σ). The 1,228 s signal was both weaker in the two cases (Fig. 4, left ). Considering in this observation J1745−3213 was on-axis and had a small source extraction region, and especially the nearly identical peaks we obtained in the MOS data (Fig. 4, right ), we take the 613.8 +1.0 −0.8 s (Huppenkothen et al. 2019) signal as the fundamental period tentatively.
Surprisingly, no signals were detected in the NuSTAR photons no matter a r=40 ′′ or a r=20 ′′ extraction region was adopted. The non-detection was not the result of the choice of energy band because 3-10 keV and 3-78 keV were both tried. We attribute the non-detection to the high background of J1745−3213 in the NuSTAR observation. J1745−3213 is fortuitously located in a field with archival VLT observations. Its VLT R-band image (Fig. 1, right ) displays a patchy environment and a nonuniform extinction near J1745-3213. We aligned the astrometry of the VLT-VIMOS image onto the Chandra image using four Tycho stars in the field of view. No optical point sources brighter than the limit of R = 25 mag 5 were detectable within 1 ′′ of the X-ray position.  However, the extinction towards the Galactic center is very high. Assuming R V = 3.1 and adopting a moderate N H (2.0×10 22 cm −2 ) based on Table.2 and Table.3, we obtained A(V) = 11.2 mag and A(R) ≈ 9.6 mag (λ ef f =6,312Å) 6 using the extinction model from Cardelli et al. (1989). Therefore, we can exclude a G5-type giant star within 11 kpc, an M5-type giant star within 37 kpc, an A0-type main sequence star within 9 kpc, an A5-type main sequence star within 5.3 kpc, and a B8-type main sequence star within 13.4 kpc (Table 15.7, Allen!'s Astrophysical Quantities, 4th Ed.). But, if we adopted a smaller N H (1.0×10 22 cm −2 ), an M0-type main sequence star within 3.6 kpc, a G2-type main sequence star within 16.8 kpc and an A5-type main sequence star within 50 kpc can be excluded.
We note there is a reported cold clump PGCC G357.16−1.66 (Planck Collaboration et al. 2016) in the same direction. With semi-major and semi-minor axes of 4.5 ′ × 3.4 ′ , and an angular separation of just 40 ′′ , the extinction may be enhanced by the clump.

ATCA
Since the optical data showed no source at the location of the Chandra position, we requested and received radio observations on the ATCA. J1745−3213 was observed over a period of four epochs: 2018-12-22, 2018-12-23, 2018-12-27 and 2019-01-03 for a total integration of ≃11.5 hours on source. For each observing epoch, the array was in the 1.5D configuration with a minimum baseline length of 107 m and a maximum baseline length of nearly 4,438 m. We utilise the Compact Array Broadband Backend (CABB, Wilson et al. 2011) with our observations being taken at a centre frequency of 2,100 MHz with 2 GHz of instantaneous bandwidth. Continuum data was recorded in 2,048 independent channels of 1 MHz bandwidth each. The flux and bandpass calibrator used for these observations was PKS 1934−638 while PKS 1759−39 was used as a phase calibrator.
Data were reduced, calibrated and imaged using the miriad software package (Sault et al. 1995) with standard routines. Flagging of the data was largely done with the automated task pgflag. We made naturally-weighted total intensity (Stokes I) mutifrequency images of 900 MHz of bandwidth, centred on 1,730 and 2,550 MHz (Fig. 5). The resultant images have a measured root-mean-squared noise (σ) of 24 and 20 µJy beam −1 for the 1,730 MHz and 2,550 MHz images, respectively.
Interestingly, the source was detected in the upper band, and not the lower band (Fig. 5). The upper-band image centred at 2,550 MHz shows a source at the centre of the imaged field with a peak intensity of 128 µJy beam −1 , equivalent to 6.4σ, and suggests a strong positive spectral index of α > 0.51 where S ν ∝ ν α . Imaging the entire 2 GHz of bandwidth results in a detection of lesser significance (5.1σ) when compared to the upper-band image.
The existence of a signal in only the higher-frequency band is unlikely due to a difference in sensitivity between the two bands, because the number of flagged visibilities due to radio frequency interference (RFI) between the two bands results in the higher-frequency band being only about 10% more sensitive than the lower-frequency band. To test whether the radio emission at the source position was an artefact related to either RFI or some other effect, we systematically removed antennae and observational epochs. In all cases the radio source remained, although we note that removing single epochs provided different source intensity, suggesting J1745−3213 is mildly radio variable. Despite radio emission remaining for all tests, we do remain somewhat cautious of the source detection.
Out of caution, we followed up our initial set of radio observations with a further ATCA radio observation taken at central frequencies of 5.5 and 9 GHz. This later radio observation was recorded simultaneously with a bandwidth of 2 GHz at each frequency. It was made on 2020-06-18 when the array was in the 1.5C configuration. Calibration, data reduction and analysis followed the same steps as reported above.
J1745−3213 was not detected in either the 5.5 or 9 GHz band. We measured an RMS noise of 17 and 15 µJy beam −1 for 5.5 GHz and 9 GHz, respectively, over the source position. While tests suggest that the detected radio emission from our 2018/2019 data was likely real, the non-detection in the lower-half of the band, and the non-detection at later times (at a higher frequency) means that the detection at 2,550 MHz should be regarded as preliminary.

VLA
We also performed follow-up radio observations using the VLA. J1745−3213 was observed for six hours on four separate dates (Project Code=21A-156; PI: Gong). The first two observations measured the spectrum of J1745−3213 across five radio bands from 2 to 26 GHz during two separate one hour scheduling blocks. One for the three lowest frequencies and the other for the two remaining high frequencies. The exposures in each band range between 705 s to 948 s duration. The third and fourth observations were two hours each in the C and X bands (4-8 and 8-12 GHz, respectively) with an on-source time of about 1.6 hours each. During these observations, the VLA cycled between the target J1745−3213 and the phase calibrator J1744−3116 about every 570 s with about 525 s being on target. Each observation used the 3-bit samplers for wideband coverage. The radio sources J1331+305 (=3C286) and J1744−3116 were used as the flux and polarization, and the phase and gain calibrators for each observation, respectively. No polarization leakage calibrators were observed, because the cross-polarization is <1% and varies slowly over several months. This accuracy is sufficient for our needs. Table.4 gives a log of the observations and the measured flux densities in each band. No radio source was detected at the location of J1745−3213 in any of the observations (e.g., Fig. 6).  The data were calibrated using version 5.6.3 of the CASA (Common Astronomy Applications Software) calibration pipeline (McMullin et al. 2007). The imaging application tclean is used to generate IQUV and RRLL images of each target scan to check for any source confusion and radio frequency interference. None were found. The flux densities are measured by fitting a point source to the UV data using the Julia programming language package Visfit (Barrett 2022). The package uses a box-constrained Levenberg-Marquardt algorithm to minimize the model residuals. The position of the point source is constrained to be within a 4 ′′ × 4 ′′ region centered on the X-ray source position.

THE NATURE OF J1745−3213
Based on the 614 s signal, the 6.7-keV iron feature, non-detection of an optical counterpart and its small Galactic latitude (l, b = 357.14866 • , −1.65597 • ), it is safe to conclude that J1745−3213 is a Galactic compact object. Nevertheless, as we point out in the first section, it is not necessarily easy to distinguish between a rotational period and an orbital period. Being devoid of an optical counterpart makes a secure identification difficult. Below, we mainly discuss its possibility of being an X-ray binary or a white dwarf binary.
The signal may be interpreted as the rotational period of an accreting pulsar. Pulsars in LMXBs typically spin faster (P spin < 200 s) than J1745−3213 due to high accretion torques. Even the exceptions to the LMXB fast spin rule, symbiotic LMXBs (P spin > 100 s, e.g., Masetti et al. 2007) are unlikely too due to their rarity and the presence of M-type giants as companions. The parameter space of a NS-HMXB scenario is quite restricted due to the optical and X-ray faintness of J1745−3213. Being a HMXB would require an L X much higher than 1.9 × 10 32 ×(8/1 kpc) 2 erg s −1 (Grimm et al. 2002).
The signal may be the orbital period of a LMXB, or further, an ultra-compact X-ray binary (UCXB), which belongs to LMXBs with extremely short orbital periods (<1 hour, e.g., in 't Zand et al. 2007a). Due to their short orbital periods, which lead to small Roche lobes, the donors of UCXBs are supposed to be hydrogen-stripped stars or white dwarfs. They tend to be located in globular clusters and the Galactic bulge, while about 40% (Heinke et al. 2013) are just Galactic field UCXBs. A small donor would make J1745−3213 optically faint (in 't Zand et al. 2007bZand et al. , 2009. The existence of an iron feature suggests a He-rich donor (Koliopanos et al. 2014). One may argue that a persistently and extremely low Eddington ratio (≈0.01% even if D=10 kpc and M=1.4 M ⊙ ) for J1745−3213 is incompatible with a supposedly high L X driven by gravitational radiation (Bildsten & Deloye 2004). But, it may be a transient UCXB (Heinke et al. 2013) and a very faint X-ray binary (Heinke et al. 2015), which both have high concentrations near the Galactic center (Bahramian et al. 2021). If the BB choice in the XMM band is correct, then J1745−3213 is a NS-UCXB because the X-ray spectrum would be much harder for a black hole X-ray binary with such a low L X . The low L X can be explained by a small accretion flow to a neutron star limited by the propeller effect and a small disk (Heinke et al. 2015;van den Eijnden et al. 2018b). The BB emission may originate from a hot spot with a radius about 11.3×(D/1 kpc) m (Lamb et al. 1973), or more likely, from two antipodal spots on the surface of J1745−3213. Similar spots appeared on the surfaces of the NS X-ray binaries 1E1743.1−2843 (kT =1.8 keV, Lotti et al. 2016) and Cir X-1 (kT =1.6 keV, Schulz et al. 2020). The hard band X-ray spectrum of J1745−3213 is consistent with an IP scenario, but not exclusive to it (e.g., Chakrabarty et al. 2014). If the fundamental period is 1228 s, one piece of evidence against the UCXB scenario is its possible pulse profile. A 180 • -separated double-peaked pulse profile usually corresponds to a rotational signal from compact objects like IPs and accreting pulsars (Annala & Poutanen 2010). A standard example of sources with this sort of pulse profile is black widow pulsars, but a 180 • -separated double-peaked pulse profile would require a fine-tuning of the geometries of intra-binary shocks (Romani & Sanchez 2016). Although orbital modulations of X-ray binaries rarely generate this kind of profile, positive examples do exist, e.g., the HMXB Vela X-1 (Malacaria et al. 2016). If the radio observations in the 1-3 GHz range are accurate, the emission contrast between the lower and higher band of 1-3 GHz could be explained by self-absorbed jets (Kaiser 2006) naturally.
The spin of polars is usually synchronized or slightly asynchronous to their orbital motion. Except AM CVn systems, most of the CVs have orbital periods longer than 80 mins (Mukai 2017). For the white dwarf binary scenario, considering the length of the 614 s signal, we interpret it as the rotational period of an IP or the orbital period of an AM CVn system. IPs (Warner 2003;Mukai 2017) are a subclass of CVs with moderate magnetic fields strong enough to truncate accretion discs, but weaker than the magnetic fields of polars, which prevent the formation of discs. The accretion stream would follow magnetic filed lines and form a shock when it collides with the magnetic poles of white dwarfs. IPs can show Fe K α complex (Hellier, & Mukai 2004), but this is almost ubiquitous in X-ray sources. A secure identification requires a spectroscopic observation in the optical band (e.g., Hare et al. 2021), which is impossible for J1745−3213. Arguments in favor of an IP identity include: 1) the hard band X-ray spectral shape is consistent with IPs (Mukai 2017). Sources with thermal bremsstrahlung temperatures in the range of dozens of keV are often identified to be IPs Clavel et al. 2016;Hare et al. 2021;Coughenour et al. 2022), 2) at least eight IPs show 180 • -separated double-peaked pulse profiles (Norton et al. 1999(Norton et al. , 2002Nasiroglu et al. 2012) generated by two-pole accretion, which is good news for an IP scenario if 1228 s is fundamental, and 3) IPs have late-type companions. The strongest counter-argument to the IP nature is the tentative radio detection by ATCA. IPs are not strong radio emitters (Barrett et al. 2017(Barrett et al. , 2020. J1745−3213 even showed evidence of radio emission with a positive spectral index. But follow-up radio observations could not confirm it. Therefore, either that detection is an artefact or J1745−3213 is transient in the radio band. If J1745−3213 is indeed an IP, using the PSR model derived by Suleimanov et al. (2019), we can estimate a 1.0 +0.2 −0.2 M ⊙ mass (last line of Table.3) for J1745−3213, which is compatible with the typical mass of an IP (Shaw et al. 2020). For AM CVn systems in particular, they can have Fe K α emission (Ramsay et al. 2005) and usually can be modelled by optically thin thermal plasma emission (Ramsay et al. 2007). Nevertheless, for those short-period AM CVn systems, they are modelled by BB emission with temperatures of dozens of eV (Nelemans et al. 2004;Solheim 2010).
In summary, properties like the hard thermal bremsstrahlung spectrum, double-peaked X-ray pulse profile (if 1228 s is fundamental), faint X-ray luminosity, and an invisible companion in the optical band are naturally compatible with an IP origin. Considering either of these properties can be cracked, the nonexistence of an optical counterpart, and the possibility that the J1745−3213 is transient in the radio band, based on the current data, we argue the second choice for J1745−3213 is an ultra-compact X-ray binary. A deep search for an infrared counterpart may help reveal its nature.