Identification of 1RXS J165424.6-433758 as a Polar Cataclysmic Variable

We present the results of our X-ray, ultraviolet, and optical follow-up campaigns of 1RXS J165424.6-433758, an X-ray source detected with the Swift Deep Galactic Plane Survey. The source X-ray spectrum (Swift and NuSTAR) is described by thermal bremsstrahlung radiation with a temperature of kT = 10.1 ± 1.2 keV, yielding an X-ray (0.3–10 keV8) luminosity L X = (6.5 ± 0.8) × 1031 erg s−1 at a Gaia distance of 460 pc. Spectroscopy with the Southern African Large Telescope revealed a flat continuum dominated by emission features, demonstrating an inverse Balmer decrement, the λ4640 Bowen blend, almost a dozen He i lines, and He ii λ4541, λ4686, and λ5411. Our high-speed photometry demonstrates a preponderance of flickering and flaring episodes, and revealed the orbital period of the system, P orb = 2.87 hr, which fell well within the cataclysmic variable (CV) period gap between 2 and 3 hr. These features classify 1RXS J165424.6-433758 as a nearby polar magnetic CV.

1. INTRODUCTION Cataclysmic Variables (CVs) are binary systems composed of a white dwarf (WD) and a late-type main sequence star.An accretion disk will also form if the WD magnetic field is < 10 6 G (Warner 1995).In systems with a highly magnetic WD (∼ 10 6−9 G) the accretion disk is either fully (Cropper 1990, e.g., AM Herculis; a polar), or partially disrupted (Patterson 1994, e.g., DQ Herculis; an intermediate polar (IP)).
In polar magnetic CVs (mCVs), matter from the companion star accretes onto the WD poles following the magnetic field lines.As the matter in the accretion column collides with the WD atmosphere a shock is formed, which emits cyclotron (optical/infrared) and thermal bremsstrahlung (X-rays) radiation.Polars have typical soft X-ray (0.3 − 10 keV) luminosities of ∼ 10 30−32 erg s −1 .
In these systems, the large magnetic field (> 10 7 G) causes synchronicity with the companion star, such that the orbital period and WD rotational period are the same.Angular momentum losses lead to an evolution of the orbital period over time.In CVs with large orbital periods (> 3 hr), the orbit is dominated by magnetic breaking, while in short period binaries (< 2 hr) gravitational radiation losses are the dominant factor.However, magnetic breaking is thought to be ineffective when the companion star becomes fully convective, leading to a so called "period gap" between 2 − 3 hr (Knigge 2006).Despite this, many polars are discovered with orbital periods lying in this gap, requiring new theories of mCV formation and evolution (e.g., magnetic field evolution or reduced magnetic braking; Li et al. 1994;Belloni et al. 2020;Webbink & Wickramasinghe 2002).
The disruption of the accretion disk reduces the thermal instabilities in these systems, making novae and outbursts less common than in non-magnetic CVs.As a result, magnetic CVs are rare in optical surveys (Oliveira et al. 2017(Oliveira et al. , 2020)).For example, Szkody et al. (2020) found that only ∼ 1/100 CVs discovered in optical surveys are magnetic.However, using a volume limited (< 150 pc) sample, Pala et al. (2020) showed that 36% of CVs, selected using Gaia parallaxes, are magnetic.This emphasizes the need for X-ray surveys (e.g., e-ROSITA) to improve the discovery rate of mCVs (see also Bernardini et al. 2012).
In this work, we introduce the discovery of a new polar through the Swift Deep Galactic Plane Survey (DGPS; O' Connor et al. 2023a), a Swift Key Project and NuS-TAR Legacy Project.The DGPS covers 40 deg 2 of the Galactic Plane between 10 < |l| < 30 deg in longitude and |b| < 0.5 deg in latitude, to luminosity L X > 10 34 erg s −1 , out to 3-6 kpc.The DGPS has also revealed the discovery of a distant IP CV (Gorgone et al. 2021) and a candidate IP (O'Connor et al. 2023b), totaling three mCVs serendipitously discovered by the survey.
1RXS J165424.6-43375(hereafter J1654) was discovered with ROSAT in 1990 August, and later with ASCA as AX J165420-4337.The Chasing the Identification of ASCA Galactic Objects (ChIcAGO) survey identified the X-ray counterpart to AX J165420-4337 using Chandra (Anderson et al. 2014).The source was later observed with Swift (Reynolds et al. 2013) as part of a shallow Galactic plane survey.Takata et al. (2022) analyzed a sample of unclassified X-ray sources with Gaia counterparts, including J1654 (G596 in that paper).Using a combination of Swift, NuSTAR, and the Transiting Exoplanet Survey Satellite (TESS), they classify J1654 as a candidate polar.In particular, their result is due to the periodicity of 2.88 hr uncovered by TESS, and the fact that no other periodicity is identified, suggesting the spin and orbital periods may be locked.However, they do not exclude the possibility that J1654 is a IP.
Here, we present DGPS observations and follow-up of J1654 obtained using our approved Target of Opportunity (ToO) programs.We obtained observations with Swift, NuSTAR, the Southern African Large Telescope (SALT), and the South African Astronomical Observatory (SAAO) 1-m telescope.We also analyze archival optical and infrared observations, and archival XMM-Newton observations.We uncover clear low-and highstate transitions in the ultraviolet, optical, infrared, and X-rays.The low-state photometry allows us to analyze the secondary star, determining that it is likely of late spectral type.High-speed photometry carried out over multiple nights yields an orbital period of ∼2.87 hr, in agreement with Takata et al. (2022).Our optical spectroscopy reveals numerous lines of hydrogen and helium, demonstrating an inverse Balmer decrement.In particular, our spectroscopic observations point towards the classification of the source as a typical polar at a Gaia distance of ∼460 pc, solidifying its previous classification (Takata et al. 2022).
We present the observations and analysis in §2.The results of our timing and spectroscopic analyses are presented in §3 with a brief discussion in §4, and conclusions in §5.Throughout the manuscript errors are quoted at the 1σ level and upper limits at the 3σ level, unless otherwise stated.

OBSERVATIONS AND DATA ANALYSIS
We begin by presenting our ultraviolet, optical and infrared data analysis in §2.1, including optical spectroscopy ( §2.1.3)and high-speed photometry ( §2.1.4).This is followed by the description of our X-ray data analysis in §2.2.
The Gaia counterpart yields a nearby distance d = 463 ± 24 pc (Bailer-Jones et al. 2021) for J1654.Moreover, the counterpart has a high proper motion of µ RA = 22.2 ± 0.3 mas yr −1 and µ DEC = −7.3± 0.3 mas yr −1 .However, this proper motion is not high enough to have impacted the source's X-ray localization over the past 20 years by more than ∼0.5 ′′ .
We also performed an analysis of public archival data obtained in multiple epochs.We retrieved calibrated VPHAS images from the ESO Archive Science Portal 1 .VPHAS photometry was performed in two epochs: 2013 July 27 in ri(Hα) and 2015 April 23 in the ugr filters.We performed aperture photometry using SExtractor with an aperture radius of 1.5× the image full width at half maximum (FWHM).The photometry was calibrated to VPHAS DR2 for the ri filters and ATLAS for the g filter.The photometry derived on 2013 July 27 agrees with that presented in VPHAS DR2, but on 2015 April 23 the source is brighter by ∼2 magnitudes, which is indicative of a high-state vs low-state transition (Warner 1999;Hessman et al. 2000;Wu & Kiss 2008).The photometry is reported in Table 1.
We performed a similar analysis of the archival VVV data to search for state changes.The calibrated images were retrieved from the ESO Archive Science Portal.Observations covering the field of J1654 were obtained in the ZY JHK filters.The Z-band photometry covers five epochs, whereas Y JH images were obtained only in two epochs on 2014 November 12 and 2017 January 18.We performed photometry using SExtractor, and determined the photometric zeropoint by calibrating to point sources in the VIRAC catalog (Smith et al. 2020) for the ZY JH bands and VVV DR2 (Minniti et al. 2017) for the K-band.Based on the Z-band variability (∆m λ ∼ 1.1 mag), we find that the observations obtained on 2017 January 18 are in a high-state, and those from 2014 November 12 are in a low-state (see Table 1).
We further searched images obtained by the Spitzer Space Telescope (hereafter Spitzer ) through GLIMPSE.PSF photometry was performed with SExtractor, calibrated to GLIMPSE standards in the field.The results are shown in Table 1.

Ultraviolet Observations
Data obtained with the Swift UltraViolet and Optical Telescope (UVOT; Roming et al. 2005;Breeveld et al. 2011) were analyzed using HEASoft v6.29c with CALDB version 20211108.The counterpart was detected in all single UVOT exposures in the uvw2, uvm2, uvw1, and u filters.A finding chart is shown in Figure 1.We performed aperture photometry using the uvotsource task with a circular source aperture of radius 5 ′′ and a nearby circular background region of radius 15 ′′ .The photometry is reported in Table A2 in the AB magnitude system.
We likewise analyzed archival data obtained with the XMM-Newton Optical/UV Monitor (OM) Telescope (Mason et al. 2001).Observations were carried out in the U V M 2, U V W 1, B, and V filters.The source is detected (∼ 5σ) only in the U V W 1 filter, which has the longest exposure time (5000 s).We used the omichain command to reduce the data.We used omdetect to determine the source count rates and the ommag task to determine the instrumental magnitudes.We converted the instrumental magnitudes to the AB system using the standard zeropoints2 .

SALT Optical Spectroscopy
Longslit spectroscopic observations were conducted using the 10-m class Southern African Large Telescope (SALT; Buckley et al. 2006) equipped with the Robert Stobie Spectrograph (RSS; Burgh et al. 2003;Kobulnicky et al. 2003;Smith et al. 2006).An initial spectrum was obtained on 2022 April 26 using the low resolution PG0300 grating at a grating tilt of 5.75 • with slit width 1.25 ′′ .The spectrum covered the wavelength range 3200 Å to 9000 Å with a resolution of R ≈ 400 at the central wavelength (6210 Å).Additional frame-transfer spectroscopic observations were conducted on 2022 May 2, 10, and 27 using the high resolution PG0900 grating at a grating tilt 14.75 • with a slit width of 1.25 ′′ .The frametransfer spectra covered 4060 Å to 7120 Å with a resolution of R ≈ 1200 at the central wavelength (5620 Å).The exposure time for the frame-transfer spectra is 100 s.A log of SALT observations is provided in Table A3.
The spectra were reduced using the PySALT package (Crawford et al. 2010), which includes bias subtraction, flat-fielding, amplifier mosaicing, and a process to remove cosmetic defects.The spectra were wavelength calibrated, background subtracted, and extracted using standard IRAF3 procedures.We obtained a relative flux calibration of all spectra using spectrophotometric standard stars.For the observation obtained on 2022 April 26, we used the standard star HILT600, and for the frame-transfer observations we used the star LTT4364.The frame-transfer observations were further continuum normalized, and used to create trailed spectra.The spectra are discussed in §3.4.

SAAO High-speed Photometry
We obtained high-speed photometry using the Sutherland High speed Optical Camera (SHOC; Coppejans et al. 2013) mounted on the South African Astronomical Observatory (SAAO) 1-m telescope.Observations took place on 2022 May 4, 5, and 8 and 2022 June 1, 2, 7, and 8.The observations were performed in the clear and g ′ r ′ i ′ filters with exposure times of either 5 or 10 s (see Table A3).The weather was clear on all nights except 2022 May 8, and we ignore these data in the rest of our analysis.
The CCD images were reduced using the TEA-Phot package (Bowman & Holdsworth 2019), which was specifically designed to work with SHOC data cubes.TEA-Phot includes the subtraction of median combined bias images and a median combined flat-field correction to reduce the CCD images, and then uses a comparison star to produce a differential light curve of the target through adaptive elliptical aperture photometry.We used 2MASS 16542993-4337457 (J = 12.467) as the comparison star for TEA-Phot.We discuss the resulting lightcurves in §3.1.

Swift/XRT
We retrieved the fully calibrated Swift data products from the NASA High Energy Astrophysics Science Archive Research Center (HEASARC) server.A log of the X-ray observations is given in Table A1.All observations of J1654 with the X-ray Telescope (XRT; Burrows et al. 2005) were carried out in PC mode for a total of 10.3 ks.We re-reduced the data using xrtpipeline within HEASoft v6.29c with CALBD version 20210915.We then used the sosta task within ximage to determine the source count rates, and corrected the rates for point-spread function (PSF) losses.We extracted the source spectra from circular regions of radius 20-pixels (1 pixel = 2.36 ′′ ) and the background spectra were extracted within a nearby region of the same size.The spectra were grouped to a minimum of 1 count per bin.Using the Swift/XRT data products generator4 we derive an XRT enhanced position (Evans et al. 2009) of RA, DEC (J2000) = 16 h 54 m 23 s .40,−43 • 37 ′ 44.0 ′′ with an uncertainty of 2.5 ′′ .

Swift/BAT
The source does not appear in the Swift Burst Alert Telescope (BAT; Barthelmy et al. 2005) catalogs (Krimm et al. 2013;Oh et al. 2018;Baumgartner et al. 2013).However, as a check, we re-analyzed data obtained on 2020 June 18 (ObsID: 03110780002) with an exposure of 3974 s in survey mode.We performed the standard BAT survey data analysis using the HEASoft tool (CALDB version 20171016), batsurvey v6.16, which reports the count rate at the source location in eight energy bands between 14 − 195 keV.The analysis does not find any significant detections.The overall signal-to-noise ratio during this time period in the 14 − 195 keV band is ∼ 0.5, which is consistent with a background noise fluctuation.
To estimate the flux upper limit, we create a spectrum file using the count rate information, and generated the corresponding instrumental response files using the HEASoft tool, batdrmgen.The upper limit was determined using XSPEC v12.12.1 assuming a power-law model with photon index Γ = 2. Specifically, we adopted the npegpow model, which is the model used in the BAT GRB catalogs (Sakamoto et al. 2011;Lien et al. 2016).This analysis yields a flux upper limit of < 7.0 × 10 −10 erg cm −2 s −1 (90% confidence level) in the 14 − 195 keV energy range.
Furthermore, we performed a search using the BAT transient monitor (Krimm et al. 2013) pipeline and found no significant detections or outbursts (15−50 keV) from the source over the lifetime of the Swift mission (∼18 yr).

NuSTAR
On 2020 August 2, we carried out a NuSTAR ToO observation of J1654 for 26 ks.We used standard tasks within the NuSTAR Data Analysis Software pipeline (NuSTARDAS) within HEASoft v6.29c with CALDB version 20220131 to extract lightcurves and spectra from a circular region of size 30 ′′ for both FPMA and FPMB.The source region is covered by stray light, and we carefully chose the background region (60 ′′ ) to reflect this issue.Spectra were grouped to a minimum of 15 counts per bin.The arrival time of photons were barycentercorrected5 to the solar system using the Chandra source localization (Anderson et al. 2014).

XMM-Newton
We analyzed an archival XMM-Newton observation covering the field of J1654 (Table A1) with a total exposure of 15 ks.The observation were performed in Full Frame mode with the medium filter.We analyze and produce separate spectra for the three XMM-Newton detectors (PN/MOS1/MOS2).The data were reduced and analyzed using tasks with the Science Analysis System (SAS v18.0.0) software using the latest CCF as of March 2022.We extracted the source photons in a circular region identified by the eregionanalyse task.The background photons were taken from an annulus surrounding the source with radii of 60 ′′ and 100 ′′ , respectively.Spectra were grouped to a minimum of 1 count per bin.The response matrix and ancillary response files were obtained using the rmfgen and arfgen tasks, respectively.The event arrival times were barycenter Table 1.Photometry of the optical/infrared counterpart to J1654.The magnitudes m λ are reported in the AB magnitude system.We report photometry of the source in both the high-and low-states when available, as well as the difference between photometry obtained in the two states.The photometry is not corrected for interstellar reddening.Upper limits are reported at the 3σ level.

Source
Filter High-state Low-state 3. RESULTS

Optical Lightcurves and Timing Analysis
Our initial high-speed photometry lightcurves exhibited characteristic features of polars, such as dips, flares, and flickering, with amplitudes of up to 1 mag.In order to search for an orbital period, we employed a Lomb-Scargle period search with Gatspy (Vanderplas 2015) to identify significant coherent pulsations in the differen-tial lightcurves produced by TEA-Phot (Figure 2).We built periodograms for each individual night (Figure 2) as well as a combined periodogram after normalizing the lightcurves to the same scale (Figure 3).We identify a significant signal at a period of 2.87 hr (10359 ± 111 s), which we interpret as the orbital period.The significance of this signal is at the > 5σ level.We also identify a harmonic of this period.No other significant lower frequency signals were identified in the periodogram at expected beat frequencies for IP systems.
We note that an independent analysis of data from the Transiting Exoplanet Survey Satellite was carried out by Takata et al. (2022) (their source G596) which finds a period of 0.12 d (2.88 hr).Our observations are in agreement with their result.

X-ray Lightcurves and Timing Analysis
We compiled available archival X-ray observations of J1654 (Table A1) in order to produce a long term Xray lightcurve of the source.We display this lightcurve, covering a time period from 2007 to 2022, in Figure 4 (top).Based on our analysis, we identify that the XMM-Newton observation from 2012 August 20 was obtained during a low-state (Warner 1999), while all other available X-ray observations, including ROSAT (1990) and ASCA, detected the source during a high-state.We discuss a comparison between the NuSTAR high-state and XMM-Newton low-state in §3.4.1.
We searched the NuSTAR and XMM-Newton observations for coherent pulsations using both Z 2 statistics (Buccheri et al. 1983) and a Lomb-Scargle frequency analysis (Scargle 1982).We searched both the 3 − 10 keV and 3 − 79 keV NuSTAR lightcurve (Figure 4; bottom), and the 0.3 − 3 keV and 0.3 − 10 keV XMM-Newton lightcurve.For the NuSTAR data we searched both the FMPA and FMPB data separately, as well as the combined events file.We likewise searched both PN, MOS1, MOS2, and the combined events files for XMM-Newton for both energy ranges.No significant signal was identified in our analysis.While the NuS-TAR observation does indeed cover about 5.3 orbital cycles, we did not find any significant signal at 2.87 hr in either dataset, nor at any other frequency.We are able to set a 3σ upper limit to the RMS pulsed fraction of < 9% in the 3 − 79 keV energy range using the NuSTAR data (FPMA/FPMB) and < 25% in the 0.3 − 10 keV energy range using the XMM-Newton data (PN/MOS1/MOS2).
The lack of other signals in the data besides the orbital period at 2.87 hr identified in our optical data ( §3.1) is in agreement with the independent analysis of Takata et al. (2022).Takata et al. (2022)  periods at shorter or longer values than the orbit, which would have been expected for an IP system.

Spectral Energy Distribution
We compiled all available ultraviolet, optical, and infrared photometry of J1654 (Table 1) in order to build the source spectral energy distribution (SED).We discovered significant variability in the UV and optical filters by ∼ 1 − 2 mag, with the largest variability amplitude in the UV.The variability was identified as the transition between a high-and low-state, as shown in Figure 5.In the high-state, the UV/optical emission from the accretion column dominates, but during periods of low accretion we observe the contribution of the secondary star, peaking in the near-infrared.An additional component, interpreted as cyclotron emission, is observed at longer wavelengths (3 − 5µm from Spitzer /GLIMPSE).At UV wavelengths, we begin to observe either the contribution from the WD, modeled as a simple blackbody, or emission from the accretion column that is fainter due to a lower accretion rate during those periods.Given the sparse UV data during the low-state it is not possible to disentangle between those contributions.Further UV observations would be critical to constrain these possibilities.
The low-state SED is well sampled in the optical and infrared (Table 1), allowing us to identify the spectral type of the secondary star.We applied the ARIADNE (Vines & Jenkins 2022) code to perform dynamic nested sampling with dynesty (Speagle 2020) in order to determine the best fit secondary star parameters over a range of stellar models (e.g., Kurucz 1993;Castelli & Kurucz 2003;Husser et al. 2013).We find an effective temperature of 4700 +800 −200 K, surface gravity log g = 4.7 +0.6 −0.4 , stellar radius R = 0.32 +0.07 −0.04 R ⊙ , metallicity [Fe/H]= −0.12 ± 0.15, a significant extinction A V = 4.9 +1.1 −0.7 mag, and distance of 454 ± 20 pc, which is in good agreement with the Gaia estimate (Bailer-Jones et al. 2018, 2021).A corner plot of these results is displayed in Figure 9 and the best fit is shown in Figure 5 (blue solid line).Overall this is consistent with a star of spectral type K3.5V.
The orbital period and i − K color also favor a mainsequence donor star of late spectral type, but instead closer to spectral types between M3V to M5V (Knigge 2006).Without optical spectroscopy of the secondary star during the low-state we cannot rule out other spectral types, but note the SED clearly favors a late spectral type.Moreover, using a photometric distance estimate based on the K-band apparent magnitude (Bailey 1981;Warner 1987;Knigge et al. 2011), we find a lower limit to the distance of d ∼ > 300 pc, which is consistent with the Gaia result (Bailer-Jones et al. 2021).The estimate is a lower limit given that we cannot rule out an additional emission component (e.g., a circumbinary dust disk or cyclotron radiation; Brinkworth et al. 2007;Harrison & Campbell 2015), besides the secondary, contributing to the observed K-band flux.
We note that the ARIADNE model only represents one possible range of solutions.For example, the H-band photometric point (Figure 5) is underestimated in this model, though this could also be due to the source not fully being in the low-state at the time of this measurement.As an additional check on the secondary star spectral type, we test whether a less extinct M dwarf can explain the low-state SED.We performed a "byeye" check, comparing Kurucz models (Kurucz 1993) to the SED.We find an appropriate match for a M2V dwarf at a distance of 460 pc, with temperature T ∼ 3500 K, surface gravity log g ∼ 4.7, metallicity log Z/Z ⊙ ∼ 1, and radius R ∼ 0.3R ⊙ (see, e.g., Parsons et al. 2018) that is extinct by A V ∼ 0.5 mag.This model is also shown in Figure 5 (gray solid line).With the available constraints to the spectral type, we cannot exclude either of the two scenarios (e.g., K vs M dwarf), but can conclude that a late-spectral type is the most likely.
In order to determine the contribution from cyclotron radiation in the infrared, and in particular at Spitzer wavelengths, we applied the cyclotron models from Potter et al. (2002).These calculations utilize the shock structures from Cropper et al. (1999) and the cyclotron opacity and radiative transfer from Meggitt & Wickramasinghe (1982); Wickramasinghe & Meggitt (1985).The model is described by the WD mass, accretion rate, and magnetic field strength.The detailed morphology of the cyclotron emission, in particular the cyclotron humps, will depend on the WD mass and accretion rate.However, the wavelength at which the cyclotron emission peaks is dominated by the magnetic field strength, with less influence from the WD mass and accretion rate.In the case of J1654, we are limited by the fact that we do not detect phase-dependent cyclotron humps in our optical spectroscopy, which would be required to obtain robust limits on the WD mass and accretion rate.Instead, we simply re-scale the model (for a specific B field) to match the peak of the excess IR emission in Figure 5.
We find that if the infrared excess is due to cyclotron radiation a low magnetic field strength of B ≲ 3.5 MG is required.Any larger magnetic field would push the excess towards the optical.We have simply shown that if the main cause of the IR emission is cyclotron, that the magnetic field strength must be low to match the IR peak.We cannot, however, rule out the contribution of a circumbinary disk at these wavelengths (Brinkworth et al. 2007;Harrison & Campbell 2015).A circumbinary disk may reasonably explain the large extinction inferred from the SED.In either case, the lack of cyclotron humps in our phase-dependent optical spectra imply that small magnetic field strength would be consistent with the excess from Spitzer.Despite the uncertainty in the spectral type of the secondary star, due to the similarity of the models (Figure 5), it does not change our estimate of the magnetic field.
We further note that the detection of the source in the U V W 1 filter by the XMM-Newton/OM during the low-state may indeed be the contribution of the WD.We consider this in both of the scenarios for the secondary star spectral type.In Figure 5, the dotted blue line shows the contribution of a thermal blackbody with T = 120, 000 K at d = 460 pc added on top of the K dwarf model that has a large extinction (A V ∼ 4.9 mag).The high temperature is a direct result of the large extinction implied by the secondary star SED.In the M dwarf scenario, which implies a low extinction (A V ∼ 0.5 mag), we find a temperature of T = 12, 000 K. These scenarios are quite different, and without further UV observations it is difficult to favor one over the other.An additional possibility is that the U V W 1 detection during the low-state is still due to the accretion disk, and simply fainter due to a lower accretion rate.SED of the ultraviolet, optical, and infrared counterpart to J1654.The solid black line connects photometry (diamonds) obtained during the high-state, whereas the dashed black line (and squares) shows the low-state.The data are not corrected for extinction; downward triangles represent 3σ upper limits.The thick, solid blue line shows the best fit stellar model from ARIADNE (K dwarf), and the solid gray line is a less extinct M dwarf (see §3.3).The dotted blue line represents the addition of a WD with T = 120, 000 K and radius R = 0.01R⊙, modeled as a simple blackbody, extinct by AV ∼ 4.9 mag.The dotted gray line represents a WD with T = 12, 000 extinct by AV ∼ 0.5 mag.The red dashed line is the addition of cyclotron emission from a polar with a magnetic field strength of 3.5 MG.

Optical Spectroscopy
The initial optical spectrum of J1654 obtained with SALT ( §2.1.3)is displayed in Figure 6.We detected a flat continuum with a slight blue slope, peppered with emission features.There are no obvious features related to the secondary star in the mean spectrum.In total, we identify 29 high significance emission lines, including hydrogen lines of the Balmer and Paschen series, HeI (λ3487,3889,4026,4144,4387,4471,4921,5875,6678,7065,7281),HeII (λ4511,4686,5411), CaII K, MgI, OI, and the λ4640 Bowen blend (a CIII+NIII line complex).The emission features are narrow (< 1000 km s −1 ).This suggests emission from a hot spot, as opposed to the velocity-broadened inner regions of an accretion disk.
We use the observed line fluxes ( to expectation based on optically thin "case B" recombination (Pottasch 1984), and is, therefore, indicative of emission from an optically thick accretion region (Stockman et al. 1977;Williams 1980;Schwope et al. 1997), typical of CVs.We further identify the observed ratio of Hβ/HeII λ4686 ≈ 1.0 as evidence that J1654 is a polar (Williams & Ferguson 1982;Williams 1989;Silber 1992;Warner 2003).
If we assume A V ≈ 4.9 mag as derived in §3.3, we can correct the line fluxes using a Cardelli et al. (1989) Milky Way extinction law.This yields Hα/Hβ ≈ 0.3, Hγ/Hβ ≈ 2.5, and Hβ/HeII λ4686 ≈ 0.76.However, A V ≈ 4.9 mag is well above the A V ≈ 0.13 − 0.35 mag implied by Galactic dust maps over this distance range (Amôres & Lépine 2005;Amôres et al. 2021), which would yield ratios of order unity.We further note that the spectra do not appear significantly reddened nor are there significant NaI absorption lines, which would suggest a lower extinction than A V ≈ 4.9 mag.The order unity line ratios, when not de-reddened, further favor a lower extinction value.
Using the frame-transfer spectroscopy, we measured the radial velocities of Hα, Hβ, Hγ, HeII λ4686, and HeI λ5876 by fitting Gaussian functions using the lmfit package (Newville et al. 2016) to derive the central wavelength, amplitude, and FWHM.The radial velocities of these lines as a function of phase are shown in Figure 7.The HeI λ5876 and HeII λ4686 lines display a qualitatively similar behavior to the Balmer lines.This likely implies that the emission region is the same.

X-ray Spectral Analysis
We performed a time-averaged spectral analysis of the Swift/XRT, XMM-Newton (MOS1/MOS2/PN), and NuSTAR (FPMA/FPMB) spectra using XSPEC v12.12.0 (Arnaud 1996).We applied the ISM abundance table from Wilms et al. (2000) and the photoelectric absorption cross-sections presented by Verner et al. (1996).Spectral fits were performed in the 3 − 20 keV energy range for NuSTAR (the background dominates above 20 keV) and 0.3 − 10 keV for Swift and XMM-Newton.We determined the best fit parameters by minimizing the Cash statistic (Cash 1979).We began by modeling the time-averaged PC mode spectrum for all XRT observations (10.3 ks).Individual shorter exposures yield large model uncertainties, and the time-averaged spectrum allows us to better constrain the parameters.The best fit absorbed power-law model (con*tbabs*pow) has N H = (2.4 ± 1.4) × 10 21 cm −2 and Γ = 1.46±0.25.This yields a time-averaged unabsorbed flux F X = (2.56 ± 0.30) × 10 −12 erg cm −2 s −1 in the 0.3 − 10 keV energy band.For these parameters, we derive an energy conversion factor (ECF) of 5.9 × 10 −11 erg cm −2 cts −1 , which we use to convert the count rates in individual exposures to an unabsorbed flux (Figure 4; Top Panel).
As all of the Swift observations occur within the high-state, we performed a joint fit with the NuS-TAR data, which was also obtained during the high-state (Figure 4; Top Panel).Based on the source's classification as a mCV (Takata et al. 2022), we tested three physically motivated models: i ) thermal bremsstrahlung radiation (constant * tbabs * bremss), likely caused as the magnetically funnelled accretion stream shocks above the WD poles, ii ) a cooling flow model (constant * tbabs * mkcflow) that selfconsistently accounts for both bremsstrahlung radiation and line cooling (Mushotzky & Szymkowiak 1988), and, lastly, iii ) a model for bremsstrahlung radiation in the post-shock structures (PSRs) of magnetized WDs (constant * tbabs * ipolar; Suleimanov et al. 2016Suleimanov et al. , 2019)).This last model is parametrized by the WD mass M WD and relative magnetospheric radius R m (hereafter referred to as the fall-height).Following Shaw et al. (2020), we froze the fall-height to R m = 1000R WD .In the mkcflow model, we froze the redshift and lowT (hereafter T low ) parameters to their minimum values (10 −7 and 0.0808 keV, respectively).
Each of these models provides a good description of the broad-band spectrum (Cstat = 266 − 298 for 313 dof).The results are recorded in Table 3, and we display the thermal bremsstrahlung radiation model in Figure 8.We note that the mkcflow and ipolar models yield consistent measurements for the WD mass.The measurement of the shock temperature T max = 21 ± 3 keV in the mkcflow is consistent with a WD mass of M WD = 0.59 ± 0.06M ⊙ (Mukai 2017), which is the same as determined by ipolar (M WD = 0.58 ± 0.05M ⊙ ).However, if cyclotron cooling is important (for high magnetic field strengths; B > 20 MG), then these models underestimate the WD mass.We therefore treat these measurements as a lower limit to the true WD mass.We do note, however, that our analysis in Section 3.3 suggests a low magnetic field strength.
We note that there is a hint of residuals in the 6 − 7 keV region of the Fe lines (see Figure 8), which can be reduced with the addition of a Gaussian emission line (gauss) with fixed energy 6.7 keV and fixed line width σ = 0 keV.Using the bremsstrahlung model, we set an equivalent width upper limit of < 210 eV.This constraint is within the range of reasonable values for other mCVs (Romanus et al. 2015).
We performed a similar analysis of the XMM-Newton data (PN/MOS1/MOS2) obtained in the low-state.The data are sparse with almost no source photons at > 3 keV, and, therefore, we choose to only apply the powerlaw and bremsstrahlung models.We obtain a best fit for an absorbed power-law (Cstat=816 for 824 dof) with a low Galactic hydrogen column density N H = (8±5)×10 20 cm −2 and hard photon index Γ = 1.6±0.2.We derive an unabsorbed flux of (8.8 ± 1.3) × 10 −14 erg cm −2 s −1 in the 0.3 − 10 keV energy range, which is a factor of ∼ 30× less than the flux observed with XRT.The low-state spectrum yields a smaller hydrogen column density, consistent with increased absorption during the high-state, possibly due to the accretion column.We performed a test by applying the best-fit high-state model (only allowing the normalization to vary), and found that the increased absorption in the high-state led to a model which under-predicted the low-state spectrum below 1 keV (consistent with a smaller absorption in the low-state).We further note that there is no requirement of a soft X-ray excess, generally described by a blackbody component with T bb < 100 eV (Ramsay et al. 2001;Ramsay & Cropper 2004).

DISCUSSION
Polars are often luminous soft X-ray sources.The number of known polars expanded greatly thanks to ROSAT (e.g., Burwitz et al. 1997Burwitz et al. , 1998;;Reinsch et al. 1999).However, their systematic identification effort was limited to high Galactic latitude for practical reasons (Thomas et al. 1998;Schwope et al. 2002), among which is the presence of many coronal (i.e., soft) sources dominating the 0.1 − 2.5 keV ROSAT band at low latitudes.First discovered by ROSAT, J1654 was identified through DGPS observations as an unclassified, relatively hard source coincident with a bright Gaia star.Based on these qualifiers, we observed the source with a variety of X-ray and optical instruments to aid in its classification.J1654 is yet another example of an X-ray (0.3 − 10 keV) selected polar CV, initially discovered with ROSAT in soft X-rays, with a classification confirmed through optical spectroscopy and photometry (e.g., Thomas et al. 1998;Schwope et al. 2002;Beuermann et al. 2017Beuermann et al. , 2020Beuermann et al. , 2021)).However, J1654 lies at low Galactic latitude, b ≈ 0.01 deg, demonstrating the importance of X-ray surveys, such as the Swift DGPS, with broader energy coverage that can discriminate polars from coronal Xray sources based on their hardness ratios.
The properties of J1654 outlined in the previous sections are characteristic of polars.The X-ray (0.3 − 10 keV) luminosity varies between L X = (6.5 ± 0.8) × 10 31 and (2.3 ± 0.4) × 10 30 erg s −1 in the high-state and lowstate, respectively, assuming a distance of 460 pc.These luminosities suggest an accretion rate in the range of (1 − 2) × 10 −11 M ⊙ yr −1 .At these distances, the hard X-ray luminosity from Swift/BAT is less than < 2×10 34 erg s −1 in the 14 − 195 keV energy band.This is consistent with the range of hard X-ray luminosities of known polars (Suleimanov et al. 2022).
In the low-state, the the hydrogen column density N H,low = (8±5)×10 20 cm −2 yields a Galactic extinction of A V = 0.36 ± 0.23 mag (Güver & Özel 2009).This is consistent with the low value of A V ≈ 0.13 − 0.35 mag (Amôres & Lépine 2005) derived from Galactic dust maps, but is at odds with the best fitting SED (see §3.3).However, X-ray spectra obtained during the high-state yield hydrogen column densities in the range N H,high = (2 − 5) × 10 21 cm −2 , which leads to a significantly larger A V ≈ 1 − 2 mag (Güver & Özel 2009).We therefore cannot rule out that there is significant extinction intrinsic to the source environment (e.g., a circumbinary disk).Further soft X-ray (0.3−10 keV) observations, preferably during the high-state, with Swift, XMM-Newton, or eROSITA would be required to better constrain the column density.
Our analysis of the low-state SED implies a secondary star of late-spectral type, consistent with the CV classification.We identify two main possibilities: i ) a K dwarf with a large extinction A V ∼ 4.9 mag and ii ) an M dwarf with a smaller extinction A V ∼ 0.5 mag.On balance we prefer the latter scenario for multiple reasons.Firstly, the M dwarf scenario has better agreement with the inferred extinction values using the low-state X-ray spectra (N H,low ), as well as Galactic dust maps.Secondly, a smaller value of the extinction has better agreement with Balmer decrements of order unity in our optical spectra, matching predictions for CVs.Thirdly, interpreting the U V W 1 detection in the low-state as emission from a WD leads to an estimate of a WD temperature T = 12, 000 K, which is consistent with typical temperatures of accreting WDs (Szkody et al. 2010;Pala et al. 2017).
Our classification of the source as a polar agrees with the interpretation of Takata et al. (2022), who classified the source as a candidate polar, though they did not exclude an IP classification.In this work, we further exclude the possibility that the source is an IP and solidify the polar classification for the following reasons.First, both our analysis and that of Takata et al. (2022) identify only a single period of 2.87 hr.In the polar interpretation, this period is P orb = P spin .However, if the system is interpreted as an IP then the spin period must be less than the orbital period, and the 2.87 hr period must be one or the other.While 2.87 hr can be the orbital period of an IP, the phase-folded TESS light curve presented by Takata et al. (2022) (their Figure 17) shows a primary maximum lasting half a cycle, with a sharp rise and sharp decline.This is atypical compared to other IPs.Given that much of optical light of an IP is from the partial accretion disk, which is more or less axially symmetric, there is no ready explanation for such an orbital modulation.In fact, optical photometric variability on the orbital period is usually of low amplitude  in IPs, and is more gradual in waveform, unless there is an eclipse, to the point that the large majority of orbital period determinations are determined from optical spectroscopy (for examples of the period determinations in IPs see, e.g., Falomo et al. 1987;Hellier & Sproats 1992;Haberl et al. 1994;Remillard et al. 1994;Allan et al. 1999;Norton et al. 2002;Zharikov et al. 2002;de Martino et al. 2006;Scaringi et al. 2011;Aungwerojwit et al. 2012;Halpern et al. 2018;Gorgone et al. 2021).We must also consider the possibility that the 2.87 hr period is the spin period of the WD.This conflicts with known IP systems, where all but 3 of the 71 confirmed IPs6 have spin periods shorter than 1 hr.Therefore, this interpretation would make J1654 the longest spin period IP, significantly longer than the previous record holder (RX J2015.6+3711 at 7196 s, or just under 2 hrs; Halpern & Thorstensen 2015;Coti Zelati et al. 2016).Note that RX J2015.6+3711 has an orbital period of 12.76 hr and that the spectral features of the secondary are readily detectable in optical spectra during its normal (high accretion rate) state (Halpern et al. 2018).In general, if J1654 is an IP with a 2.87 hr spin period, this would likely require an orbital period significantly longer than a few hours.In contrast, the low-state SED indicates a small mass donor consistent with the size of the Roche lobe for a ∼ 3 hr orbital period CV (Knigge et al. 2011).Therefore, we conclude that the interpretation of J1654 as a polar is the most probable explanation, and that it matches all available information for the source.
With an orbital period of 2.87 hr, J1654 lies within the CV period gap.This is a common property of polars, which do not exhibit a prominent period gap (Wick-ramasinghe & Wu 1994;Webbink & Wickramasinghe 2002).This is thought to be due to reduced magnetic breaking affecting the evolution of polars, see, e.g., Li et al. (1994); Belloni et al. (2020).We discuss the validity of this interpretation for J1654 in more detail below.
We are confident that, if the infrared excess from Spitzer is due to cyclotron radiation, the magnetic field has to be rather low (B ≈ 3.5 MG).This is also supported by the fact that larger magnetic field strengths begin to push the excess towards the optical, where we would expect to observe cyclotron bumps in our SALT spectra (e.g., Figure 6).
Furthermore, based on the X-ray luminosity we infer an accretion rate of ∼ 10 −11 M ⊙ yr −1 .While some adjustment is needed to account for the accretion power that goes into cyclotron emission, the correction is likely small due to the fact that we do not detect a luminous soft X-ray component (Ramsay & Cropper 2004).On the other hand, the inferred rate is based on a single measurement during a high-state, and the long-term average (over high-and low-states) is likely somewhat, but not greatly, lower.This is because Figure 4 suggests that this polar is usually found in a high-state (4 out of 5 epochs since 2007; grouping XRT observations of a similar time into the same epoch).This should be compared to the number of 21 out of 37 polars found in a high-state by Ramsay et al. (2004).
In any case, the accretion rate is comparable to the expected values for systems whose evolution is driven purely by gravitational wave radiation (e.g., Knigge et al. 2011).This is consistent with the reduced magnetic braking model for the evolution of polars (Li et al. 1994;Belloni et al. 2020), in which the magnetic braking mechanism is ineffective for polars due to the WD magnetic field.However, there is a tension between this conclusion and the low magnetic field (B ≈ 3.5 MG) that was obtained by modeling the infrared excess as cyclotron emission: can such a low field nevertheless reduce magnetic braking, and if so, should we not see a similar effect in IPs?
Further optical and infrared observations, including infrared spectroscopy, and a more secure determination of the magnetic field, is highly desirable.In particular, optical and infrared spectroscopy to identify the phasedependent cyclotron humps are required before a detailed modeling of the WD mass, accretion rate, and magnetic field can be performed.Infrared spectroscopy may either confirm or exclude cyclotron emission as the nature of the IR excess, and is therefore strongly encouraged.

CONCLUSIONS
We have presented the results of the DGPS multiwavelength follow-up campaign of 1RXS J165424.6-433758 using Swift, NuSTAR, SAAO, and SALT.The source displays a hard X-ray spectrum in the highstate characterized by thermal bremsstrahlung radiation with temperature kT = 10.1 ± 1.2 keV and luminosity ∼ 6.5 × 10 31 erg s −1 .The low-state luminosity, revealed by archival XMM-Newton data, is a factor of ∼ 30× fainter.
Moreover, we identify the ultraviolet, optical, and infrared counterpart using archival Chandra observations.A SALT spectrum of the source revealed a variety of emission lines of hydrogen and helium, and demonstrated an inverse Balmer decrement.High-speed photometry with the 1m SAAO over multiple nights in 2022 May and June allowed us to derive an orbital period of 2.87 hr.The X-ray, ultraviolet, optical, infrared, and, in particular, emission line, properties allow us to classify J1654 as a polar.This classification is in agreement with the independent analysis of Takata et al. (2022).

ACKNOWLEDGMENTS
The authors thank the anonymous referee for their careful review and constructive feedback.Here we show the corner plot derived using ARIADNE (Vines & Jenkins 2022).The best fit model is displayed in Figure 5.

Figure 1 .
Figure 1.RGB image of the field of J1654 using three UVOT filters (left: red = uvw1, green = uvm2, and blue = uvw2) and three VVV filters (right: red = Ks, green = H, and blue = J).The archival Chandra localization (1.0 ′′ ; 95% CL) is displayed by the magenta circle.In the top right a zoom in on the position of the counterpart is shown.North is up and East is to the left.

Figure 3 .Figure 4 .
Figure 3. Lomb-Scargle periodogram of the combined optical data obtained with the SAAO.The orbital period Ω (main peak) and its harmonic are observed.The inset shows a zoom-in on the frequency corresponding to the orbital period of 2.87 hr, and its aliases.
Figure5.SED of the ultraviolet, optical, and infrared counterpart to J1654.The solid black line connects photometry (diamonds) obtained during the high-state, whereas the dashed black line (and squares) shows the low-state.The data are not corrected for extinction; downward triangles represent 3σ upper limits.The thick, solid blue line shows the best fit stellar model from ARIADNE (K dwarf), and the solid gray line is a less extinct M dwarf (see §3.3).The dotted blue line represents the addition of a WD with T = 120, 000 K and radius R = 0.01R⊙, modeled as a simple blackbody, extinct by AV ∼ 4.9 mag.The dotted gray line represents a WD with T = 12, 000 extinct by AV ∼ 0.5 mag.The red dashed line is the addition of cyclotron emission from a polar with a magnetic field strength of 3.5 MG.
B. O. would like to acknowledge the significant contribution of Nicholas Gorgone to DGPS operations and analyses during his dissertation research.B. O. acknowledges useful discussions with Jeremy Hare.B. O. and C. K. acknowledge support from NASA Grant 80NSSC20K0389.This work is based on observations obtained with the Southern African Large Telescope under programme 2021-2-LSP-001.This work made use of data supplied by the UK Swift Science Data Centre at the University of Leicester.This research has made use of the XRT Data Analysis Software (XRTDAS) developed under the responsibility of the ASI Science Data Center (ASDC), Italy.This work made use of data from the NuSTAR mission, a project led by the California Institute of Technology, managed by the Jet Propulsion Laboratory, and funded by the National Aeronautics and Space Administration.This research has made use of the NuSTAR Data Analysis Software (NuSTARDAS) jointly developed by the ASI Space Science Data Center (SSDC, Italy) and the California Institute of Technology (Caltech, USA).This research has made use of data and/or software provided by the High Energy Astrophysics Science Archive Research Center (HEASARC), which is a service of the Astrophysics Science Division at NASA/GSFC.The scientific results reported in this article are based on observations made by the Chandra X-ray Observatory.This research has made use of 1RXS J165424.6-433758data obtained from the 4XMM XMM-Newton serendipitous stacked source catalogue 4XMM-DR11s compiled by the institutes of the XMM-Newton Survey Science Centre selected by ESA.Facilities: Swift/XRT, Swift/UVOT, NuSTAR, XMM-Newton, SALT, SAAO Software: HEASoft 6.29c, XRTDAS, NuSTARDAS, SAS, XSPEC (Arnaud 1996), IRAF (Tody 1986), Astropy (Astropy Collaboration et al. 2013), ARIADNE (Vines & Jenkins 2022), dynesty (Speagle 2020)

Figure 9 .
Figure 9. Corner plot of secondary star parameters derived from modeling the low-state SED with ARIADNE.
likewise did not identify any PowerFigure 2. Optical lightcurves of J1654 from the 1-m SAAO telescope obtained on different nights, labeled by their Barycentric Julian Date (BJD).The photometry is measured as differential instrumental magnitudes, dependent on the observing conditions on each night.The inset displays the periodogram for each night.

Table 2 )
to compute the Balmer decrements.The line fluxes were determined by modeling each line as single Gaussian on top of the continuum emission.The ratios Hα/Hβ ≈ 1.36 and Hγ/Hβ ≈ 1.16 have not been corrected for the uncertain Galactic extinction, but still signify an inverted Balmer decrement.The inverse Balmer decrement is in excess Figure6.Mean spectrum of J1654 obtained with SALT.We have labeled the primary emission features: Balmer lines in blue, HeI lines in red, HeII lines in orange, and other features in black, including the Bowen Blend (BB; NIII/CIII).Chips gaps are located at 5000 Å and 8250.Gray vertical regions mark telluric features.

Table 2 .
Spectral features observed in the mean SALT spectrum from 26 April 2022.The line fluxes have not been corrected for extinction.

Table 3 .
Results of our X-ray spectral fitting.The high-state data includes our Swift and NuSTAR (FPMA and FPMB) spectra, while the low-state includes only the archival XMM-Newton (PN, MOS1, and MOS2) data.

Table A3 .
Log of SALT and SAAO observations of J1654.The modes refer to longslit (LS) spectroscopy, frame-transfer (FT) spectroscopy, or high-speed photometry (HSP) with an exposure time of either 5 or 10 s.