Summary

5XMM-DR15 is the fifth generation catalogue of serendipitous X-ray sources from the European Space Agency's (ESA) XMM-Newton observatory, and has been created by the XMM-Newton Survey Science Centre (SSC) in collaboration with the XMM-Newton Science Operations Centre (SOC) on behalf of ESA. For this fifth generation of catalogues, only a stacked catalogue of sources (with the individual detections and upper limits) is provided. This stacked catalogue uses improved stacking methodology which allows all the individual observations to be included, contrary to stacked versions prior to 5XMM-DR15.

The catalogue contains source detections drawn from a total of 14616 XMM-Newton EPIC observations made between 2000 January 19 and 2024 October 14; all datasets included were publicly available by 2024 October 31 but not all public observations are included in this catalogue. This is due to some observations having very poor signal to noise or processing issues. The net area of the catalogue fields taking account of the substantial overlaps between observations is ~1397 deg².

5XMM-DR15 contains 818656 unique X-ray sources and 2578752 X-ray detections or upper limits above the processing likelihood threshold (column STACK_DET_ML) of 6. Almost half of all sources (411307) have more than one detection in the catalogue (up to 98 repeat observations in the most extreme case).

The catalogue distinguishes between extended emission and point-like detections. Parameters of detections of extended sources are only reliable up to the maximum extent measure of 80 arcseconds. There are 42669 detections of extended emission, only about half of the number in 4XMM-DR14, but twice the number of 'clean' extended sources in 4XMM-DR14. This can be understood by the improved signal to noise in 5XMM-DR15 which is a stacked catalogue, which ensures that extended sources are more reliably identified. Indeed more than 60% of the extended sources in 5XMM-DR15 (25845) are identified as clean (SUM_FLAG < 3).

Due to intrinsic features of the instrumentation as well as some shortcomings of the source detection process, some sources are considered to be spurious or their parameters are considered to be unreliable. It is recommended to use a flag as filters to obtain what can be considered a 'clean' sample. There are 764140 out of 818656 sources that are considered to be clean (i.e., summary flag < 3).

For 408694 detections, EPIC time series and 408901 detections, EPIC spectra were automatically extracted during processing, and a χ²-variability test was applied to the time series. This is a significant increase since 4XMM-DR14 as these products are now extracted for detections with 50 EPIC counts, whereas 100 EPIC counts were previously required. 12330 detections in the catalogue are considered variable, within the timespan of the specific observation, at a probability of 10^-5 or less based on the null-hypothesis that the source is constant. Of these, 10907 have a summary flag  < 3. On the long-term, 41187 sources have a variability of a factor five or greater.

The median flux in the total photon energy band (0.2 - 12 keV) of the catalogue is ~ 1.3 × 10^-14 erg cm^-2 s^-1; in the soft energy band (0.2 - 2 keV) the median flux is ~ 2.9 × 10^-15, and in the hard band (2 - 12 keV) it is ~ 7.6 × 10^-15. The flux values from the three EPIC cameras are, overall, in agreement to ~ 10% for most energy bands. The median positional accuracy of the catalogue point source detections is generally < 1.52 arcseconds (with a standard deviation of 1.41).

1. Introduction

Pointed observations with the XMM-Newton Observatory detect significant numbers of previously unknown 'serendipitous' X-ray sources in addition to the proposed target. Combining the data from many observations thus yields a serendipitous source catalogue which, by virtue of the large field of view of XMM-Newton and its high sensitivity, represents a significant resource. The serendipitous source catalogue enhances our knowledge of the X-ray sky and has the potential for advancing our understanding of the nature of various Galactic and extragalactic source populations.

The 5XMM-DR15 catalogue is the seventeenth publicly released XMM-Newton X-ray source catalogue produced by the XMM-Newton Survey Science Centre (SSC) consortium. It follows the 1XMM (released in April 2003), 2XMMp (July 2006), 2XMM (August 2007), 2XMMi (August 2008), 2XMMi-DR3 (April 2010), 3XMM-DR4 (July 2013), 3XMM-DR5 (April 2015), 3XMM-DR6 (July 2016), 3XMM-DR7 (June 2017) and 3XMM-DR8 (May 2018), 4XMM-DR9 (December 2019), 4XMM-DR10 (December 2020), 4XMM-DR11 (August 2021), 4XMM-DR12 (July 2022), 4XMM-DR13 (June 2023) and 4XMM-DR14 (July 2024) catalogues: 2XMMp was a preliminary version of 2XMM. 2XMMi and 2XMMi-DR3 are incremental versions of the 2XMM catalogue.

The 5XMM-DR15 catalogue is about 18% larger than the 4XMM-DR14 catalogue in terms of sources (126547 more sources) and almost twice the number of sources in 4XMM-DR14s, as observations with no overlap were not considered in previous versions of the stacked catalogue. In terms of the number of X-ray sources, it is 88% of the eROSITA DR1 catalogue that covers half of the sky (Merloni et al. 2024) and more than twice the number of sources and detections that are in the Chandra source catalogue version 2.1 (Evans et al. 2010). 5XMM-DR15 complements deeper Chandra and XMM-Newton small area surveys, probing a large sky area at the flux limit where the bulk of the objects that contribute to the X-ray background lie. The 5XMM-DR15 catalogue provides a rich resource for generating large, well-defined samples for specific studies, utilising the fact that X-ray selection is a highly efficient (arguably the most efficient) way of selecting certain types of object, notably active galaxies (AGN), clusters of galaxies, interacting compact binaries and active stellar coronae. The large sky area covered by the serendipitous survey, or equivalently the large size of the catalogue, also means that 5XMM-DR15 is a superb resource for exploring the variety of the X-ray source population and identifying rare source types.

The production of the 5XMM-DR15 has been undertaken by the XMM-Newton SSC and XMM2ATHENA consortia in collaboration with the XMM-Newton Science Operations Centre in fulfillment of one of its major responsibilities within the XMM-Newton project. The catalogue production process has been designed to fully exploit the capabilities of the XMM-Newton EPIC cameras and to ensure the integrity and quality of the resultant catalogue through rigorous screening of the data.

5XMM-DR15 is based on the pipeline configurations 21.51. This pipeline version contains many changes with respect to the pipeline used to make the previous major version of the catalogue, 4XMM. The main changes to the EPIC processing include an empirical correction the MOS effective area to align with the pn effective area, along with a correction to the pn effective area above 3.0 keV to align EPIC pn to NuSTAR, which has the advantage of carrying out calibration without the mirror module and is therefore more accurate, an update to the CCD layout in the LINCOORD current calibration file (CCF) to align the source positions with the pn camera source positions (see Webb, Traulsen et al. 2026), introducing an evolving Energy Conversion Factor (ECF) with time for the MOS cameras, extracting spectra and lightcurves for each detection when there are more than 50 EPIC counts (previously 100 EPIC counts were required), and new source detection techniques developed for stacked source detection, which involves first fitting the position, extent and common flux and spectral parameters to each detection using the ECF in spectral fitting before maximum likelihood fitting and determining the final source parameters and extracting the variability information through point spread function (PSF) photometry for each detection. More information on these changes can be found in Webb, Traulsen et al., (2026), currently the draft version available.

As in previous versions of the stacked catalogue, the SRCID is not propagated from previous versions.

2. User Guide for 5XMM

The extensive User Guide (UG) for the 2XMM catalogue still describes many of the details of the data processing and compilation approach applicable to the 5XMM-DR15 catalogue. However, a significant number of changes to the processing have been implemented for 5XMM and these are described in the 5XMM-DR15 documentation and the Section 3 below. For convenience, Table 1, which gives the energy band definitions, is repeated here.

Table 1:  Energy bands used in 5XMM-DR15

Basic energy bands:	1	=	0.2 -   0.5 keV
	2	=	0.5 -   1.0 keV
	3	=	1.0 -   2.0 keV
	4	=	2.0 -   4.5 keV
	5	=	4.5 - 12.0 keV
Broad energy bands:	6	=	0.2 -   2.0 keV		soft band, no images made
	7	=	2.0 - 12.0 keV		hard band, no images made
	8	=	0.2 - 12.0 keV		total band

3.   5XMM-DR15 -- key changes with respect to previous versions

3.1 Data selection

XMM-Newton observations considered for inclusion in the 5XMM-DR15 catalogue were those with ODFs available for processing up to 2024 October 14 and all were publicly available as of 2024 October 31. After allowing for a small number of observations which failed in processing for a variety of reasons, Table 2.1 gives the list of the final 14616 observations which are included in the 5XMM-DR15 catalogue.

3.2 Data Processing

Data processing for the 5XMM-DR15 catalogue was based on the SAS version 21 and carried out with the pipeline version 21.51 and the latest set of current calibration files at the time of processing (November and December 2024). This new version includes a number of improvements compared to previous versions. Improvements to the EPIC (and RGS) effective areas were made using an empirical correction from MOS to pn (and RGS to pn), as well as a further correction above 3 keV to align the pn to the NUSTAR spectral fits. A correction was also included to update the MOS CCD positions to improve the astrometry. The main data processing steps used to produce the 5XMM data products were similar to those outlined in (Webb et al. 2020, Rosen et al. 2016, Watson et al. 2009) and described on the SOC webpages. For all the 5XMM data, the observation data files were processed to produce calibrated event lists. The optimised background time intervals were identified and using them, the filtered exposures (taking into account exposure time, instrument mode, etc.), multi-energy-band X-ray images, and exposure maps were generated. The initial detections were made on single observations, using simultaneously all images and bands, one to five, from the three cameras when available, see Table 1. The probability, and corresponding likelihood, were computed from the null hypothesis that the measured counts in the search box result from a Poissonian fluctuation in the estimated background level. A detection mask was made for each camera that defines the area of the detector which is suitable for source detection. An initial source list was made using a ‘box detection’ algorithm. This slides a search box (20'' x 20'') across the image defined by the detection mask. Sources were cut-out using a radius that was dependent on source brightness in each band, and these areas of the image where sources had been detected were blanked out. The source-excised images, normalised by the exposure maps, and the corresponding masks are convolved with a Gaussian kernel to create the background map (Traulsen et al. 2019). A second box-source-detection pass was then carried out, creating a new source list, this time using the background maps (‘map mode’) which increased the source detection sensitivity compared to the first pass. The box size was again set to 20'' x 20''. A maximum likelihood fitting procedure was then applied to the sources to calculate source parameters in each input image, by fitting a model to the distribution of counts over a circular area of radius 60'' (Watson et al. 2009). 1.19 million detections were made before the stacking procedure. These detections were then used to produce detection level spectra and lightcurves, if more than 50 EPIC counts per detection were detected, where previously 100 EPIC counts were required. This resulted in almost 409000 spectra and lightcurves, an increase of 10% with respect to 4XMM-DR14. For the catalogue of sources (5XMM-DR15), the exposures were stacked and source detection was carried out by first fitting the position, extent and common flux and spectral parameters to each detection using the ECF in spectral fitting before maximum likelihood fitting and determining the final source parameters and extracting the variability information through point spread function (PSF) photometry for each detection. Automatic and visual screening procedures were carried out to check for any problems in the data products.

3.3 Stacking and source detection

Source detection on XMM-Newton EPIC observations uses maximum-likelihood fits under Cash statistics as described for example by (Watson et al. 2009, Traulsen et al. 2019). With 5XMM, we introduce a revised approach to stacked source detection in order to handle all 5XMM data from single observations to 99 directly overlapping observations. During the source-detection step, we assume that the flux of each source remains constant over all exposures and that its spectrum in the five standard energy bands can be described by a simple model. We choose an absorbed power-law as the spectral model which is a reasonable approximation to most XMM-Newton sources (Watson et al. 2009). Under these assumptions, the equations of the maximum-likelihood detection take the same form and the same degrees of freedom irrespective of the number of exposures in which a source is fitted. The degrees of freedom are the source coordinates, the mean source flux, and the spectral parameters -- column density and power-law index -- if the source is fitted as point-like, and additionally the radius of the extent model, if the source is fitted as extended. The results of the five-band spectral fit in the detection step are given in the catalogue in the columns with prefix ``STACK_''.

The photon flux is related to the measured count rates in each input image to source detection by energy conversion factors (ECFs). In the new XMM-Newton source detection, the ECFs for each fitted pair of spectral parameters, for each fitted detector position, and for each instrumental setup (EPIC/pn, MOS1, MOS2 with their respective filters) are extrapolated on the fly over a grid of pre-compiled values. They cover column densities between 10^19-23 cm^-2 and power-law indices between 0 and 5. Time-dependence of the EPIC instrumental cross-calibration is taken into account over six different epochs.

Once a source is reliably detected with a log-likelihood STACK_DET_ML ≥ 6, the assumptions of constant flux and power-law spectrum are dropped, and image-level count rates and related parameters are determined by forced PSF photometry at the detected source position and extent radius. During PSF photometry, the count rate in each contributing image is treated as a free fit parameter: the method used in source detection in the previous Serendipitous Source Catalogues from EPIC data. The photometry results are given in the catalogue in the RATES, FLUX, DET_ML and related columns without the prefix STACK.

For each fit parameter, the lower and upper confidence limit are calculated, searching for the parameter values for which the minimum Cash statistics value plus one is reached. For an efficient and robust search, the source-detection task emldetect employs the so called false-position method, which is a numerical bracketing approach. If the calculation of an error component does not converge, this component is now set to undefined in all cases. Previously, a count-rate dependent fall-back value was used for coordinates, extent, and count rates. The total 1 σ error on a parameter is the arithmetic mean of the lower and the upper error if both are defined. If one component does not converge, the other component is taken as the total error. In addition to the total errors, 5XMM also includes the asymmetric upper and lower errors on the image coordinates, the extent radius, and the spectral fit parameters STACK_FLUX, STACK_NH, and STACK_GAMMA.

3.4 Systematic position error

The systematic uncertainty of the 5XMM-DR15 astrometry is estimated using a statistical approach based on the cross-matching of the X-ray sources with an external catalogue with accurate positions. The adopted methodology is similar to that described in Section 6.2 of Merloni et al. (2024). It is assumed that the X-ray positional errors are symmetric in the direction of the right ascension and declination and are described by the normal distribution. Under these assumptions the probability of a radial offset, r, of an X-ray source from its true position is given by the Rayleigh distribution with parameter σ that represents the astrometric standard deviation in the right ascension or declination direction. We assume that σ has a statistical (σ_stat) and a systematic (σ_sys) component

σ = (σ_stat² + σ_sys²)^0.5    (1)

with σ_stat=RADEC_ERR/√2, i.e. the statistical error is approximated by the RADEC_ERR parameter estimated by the detection chain. The systematic uncertainty is inferred at the population level by modeling the angular separation distribution between the positions of X-ray sources and an external catalogue with vanishing astrometric errors. The linear part at large angular distances represents chance alignments and a pronounced peak at small separations corresponds to true associations. Modeling the observed number of pairs at a given angular separation can constrain the fraction of X-ray sources with true associations in the external catalogue, the sky density of the external catalogue, X-ray source positional uncertainty and hence σ_sys².

The total number of X-ray verses external catalogue pairs at a given angular separation bin θ is a Poisson variate with expectation value λ(θ) = N_rand(θ) + N_assocθ. Therefore the likelihood of the model can then be expressed as the product of the Poisson probabilities at each angular separation bin, see Webb, Traulsen et al. (2026)

The modeling assumes that the positional uncertainty of an X-ray source is given by Equation 1 and that the systematic uncertainty is the same for all sources. Although σ_sys depends on the number of X-ray photon counts of a source, this dependence is weak and therefore assuming a single catalogue-wide value for this parameter is an acceptable approximation, see Webb, Traulsen et al. (2026). The external astrometric catalogue used was quasars from Gaia and unWISE Data (Shu et al. 2019). We only consider Gaia/unWISE sources with probability of being a quasar >0.8 and g-band magnitude <20.5 mag. The latter criterion is adopted to minimise variations in the sky density of quasar candidates because of the variable depth of the Gaia survey as a result of the scanning law of the mission. We limit the 5XMM catalogue to sources with emldetect detection likelihood EP_DET_ML>15 (to increase the purity of the sample), that are not spatially extended (parameter EXTENT=0), are not close to CCD gaps or the edges of the field of field of view (PN_MASKFRAC>0.9 or M1_MASKFRAC>0.9 or M2_MASKFRAC>0.9), have quality flags that do not indicate issues during the detection (SUM_FLAG=0) and lie outside the Galactic plane (Galactic latitude >30°). For this sample, we infer σ_sys=0.88±0.01 arcsec.

This σ_sys is larger than the one derived for 4XMM-DR10s by Traulsen et al. (2020), because the 5XMM data were not rectified astrometrically when producing DR15. The astrometric correction will be included in DR16 to further improve the source positions.

3.5 Long term variability

5XMM-DR15 is a stacked catalogue, containing all of the sources detected following the stacking of overlapping observations, but also includes the individual detections in each of the contributing observations and non-detections when no detection was made. This provides the user with long-term XMM-Newton variability over the 25 years of data used to construct the catalogue. These data can be visualised in lightcurves produced for each source.

However, to increase the timeframe over which variability can be examined and to increase the number of data points for each source, the STONKS algorithm was implemented (Quintin et al. 2024). This algorithm uses a master catalogue constructed from data from a variety of different observatories. To create the 5XMM-DR15 catalogue, this master catalogue was generated in February 2026 using the most recent versions of the XMM-Newton catalogue (4XMM-DR14, Webb et al. 2020), Chandra Source Catalogue version 2.1 (Evans et al. 2010), the Living Swift/XRT point-source catalogue (Evans et al. 2023), the eROSITA eRASS1 catalogue (Merloni et al. 2024), the XMM-Newton Slew survey catalogue version 3, XMMSL3 (Saxton et a.l. 2008), the two ROSAT catalogues, 2RXS (Boller et al. 2016) and WGACAT (White et al. 1994). We also generated upper limits for the XMM-Newton non-detections using the RapidXMM (Ruiz et al. 2022) version of HILIGT (Saxton et al. 2022). Matching was done on a two by two basis using an algorithm based on (Budavary & Szalay 2008) and implemented in NWAY (Salvato et al. 2018), see Quintin et al. (2024) for the details of the algorithm. We ensured that the fluxes estimated were comparable by converting each flux detection to a single, common energy band. The common band we chose was the 0.1-12 keV band, as it contains the energy bands of every one of the missions we used and then assumed an absorbed powerlaw spectra, with parameters Gamma = 1.7 and n_H = 3 x 10²⁰ cm^-2.

The pessimistic variability ratio was calculated, taking the ratio of the highest flux point minus the 1 σ error and the lowest flux point plus the 1 σ error. Alternatively, in the case of an upper limit, the difference is calculated using the 3 σ upper limit and the highest flux point minus the 1 σ error. We provide significant long-term variability for sources that have a ratio of five or greater. The variability is calculated for the detections made with the standard pipeline before stacking. This is provided in the column 'APPROX_SOURCE_VAR'. There are 41187 detections with a variability ratio of five or greater, with the highest reaching a ratio of 78000. The mean variability is a factor 80.

3.6 Spectral fitting

The procedure to select, merge and analyse the 5XMM spectra is similar to Viitanen et al 2025, with some changes in the procedure used to merge the spectra, and in the output quality flags. Standard PPS processing of individual observations includes spectral extraction for detections with more than 50 EPIC counts, with a corresponding background spectrum. For each of the stacked sources, the associated individual detections are checked for extracted spectra. Each spectrum is checked for a strictly positive number of total counts (in the extracted detection spectrum, including source and background), background counts (in the extracted background spectrum), and net counts (calculated by subtracting the background counts from the total counts, after scaling by the relative extraction areas), in the 0.2-12~keV band. If any of these conditions are not fulfilled, the spectrum is discarded from further processing.

The selected spectra are then separated by instrument (pn or MOS). Spectra from each instrument are merged. The procedure to decide which spectra to merge for each source has been simplified with respect to that in Viitanen et al 2025. The spectra are sorted in decreasing signal-to-noise ratio (defined as the net counts divided by twice the total counts minus the re-scaled background counts) and the cumulative signal-to-noise ratio of the spectra with higher or equal signal-to-noise ratio than the one under consideration is calculated. Spectra are only merged down to the point in which the maximum cumulative signal-to-noise ratio is reached. The merging is done using the SAS task epicspeccombine. The merged total and background spectra for each instrument are then re-binned to have one or more counts per bin. Spectral fitting and modelling were done with Xspec (Arnaud 1996) through the Python interface together with the Bayesian X-ray Analysis (BXA) tool (Buchner et al 2014), which connects Xspec to the nested-sampling package UltraNest (Buchner et al 2021). The spectral models were implemented in Xspec and explored with BXA. To speed up the Bayesian fitting, we first performed a quick Levenberg--Marquardt fit to obtain approximate best-fit values, and then reduced the prior volume explored by UltraNest by centring the prior bounds on those preliminary estimates. For the power-law model used here, the prior range for N_H was set to the preliminary best-fit value, clipped to the interval [0.001,10.0] (in units of 10²² cm^-2), the prior range for Γ was set to the preliminary best-fit value ±1.0, clipped to [1.0,3.0], and the prior range for log₁₀(Flux) was set to the preliminary best-fit value ±2, clipped to [-15,-9]. This substantially reduced the prior volume explored by UltraNest and accelerated convergence. The posterior probability distributions were then obtained from the BXA/UltraNest sampling and stored as chains for further summary and export.

The spectra were fitted with an absorbed powerlaw (in Xspec notation cflux * phabs * zpowerlw) to the merged spectra. This model has three free spectral parameters: the flux (FLUX in the catalogue, observed flux not corrected for absorption), the column density (NH, not constrained by the column density of our Galaxy in the direction of the source) and the spectral slope (GAMMA). In addition, when pn and MOS spectra are fitted jointly, we include an inter-instrument normalisation parameter (IIN), implemented as a multiplicative constant, with the pn normalisation fixed to unity and the MOS normalisation left free. Future versions of the catalogue will include spectral fitting with other models as well. All spectral fitting was performed using the Cash statistic (Cash 1979), which is appropriate for Poisson-distributed data and particularly effective in the low-count regime. Unfortunately, the Cash statistic does not provide a goodness-of-fit (GoF) indicator. This was estimated by fitting the merged background spectra with an empirical, camera-specific background model. The GoF was determined by re-binning the background spectra to have at least 20 counts per bin, and then the χ² of the best fit Cash model was compared to the expected value for an equivalent number of degrees of freedom (see Viitanen et al 2025). Fits with probabilities p<0.01 were discarded and excluded from further analysis. The second step, for the merged spectra whose corresponding background spectra passed the previous filter, was fitting the pn and MOS spectra using a combined source+background model, in which all background-shape parameters were fixed to the best-fit values obtained in the initial background-only fit, leaving only the background normalisation free to vary. This approach ensures consistency and mitigates overfitting. The method of Buchner et al 2014 was used to estimate the GoF of the source+background fits. The p-values were estimated using a permutation test. For each source, we generated 1000 resampled datasets by randomly redistributing the combined data+model counts into two equal-size subsamples, allowing each energy-bin count to originate from either the observed or modelled spectrum. For each resampling, we computed the corresponding Kolmogorov-Smirnov (KS) statistic. The p-value was then defined as the fraction of permutations yielding a KS statistic larger than that of the original data–model comparison. Models with KS p-values ≥0.01 were considered acceptable fits.

Catalogue values provided include the median and the 5 and 95% percentiles, degrees of freedom and the KS GoF p$-value (PVALUE) of the source+background fit. The INFO parameter links to the list of spectra included in the merged spectrum (designated by their observation identification OBS_ID and source number (SRCNUM). A flag (FLAG) is provided, with the following possible values:

0 : no issues detected
1 : zero or negative source counts
2 : zero or negative source counts (also implies ≤0 net counts)
3 : could not create merged spectrum
4 : source+background or background fit failed
5 : poor goodness-of-fit, with KS p-value <0.01 (or pre-fit p-value <0.01 if KS is unavailable)
6 : photon index pegged, with PhoIndex median within 0.05 of the hard prior limits ≤ 1.05 or ≥ 2.95, for priors in the range [1.0,3.0])
7 : poor goodness-of-fit and photon index pegged
8 : NH pegged, with median NH ≥ 9.5 (near the upper cap of 10.0, in units of 10²² cm^-2)
9 : poor goodness-of-fit and NH pegged
10: photon index pegged and NH pegged
11: poor goodness-of-fit, photon index pegged, and NH pegged

3.7 Optical monitor (OM) products

The XMM-OM observes the sky simultaneously with the X-ray instruments onboard XMM-Newton. For 5XMM, XMM-OM counterparts to X-ray sources are drawn from version 6.2 of the XMM-Newton Serendipitous Ultraviolet Source Survey (XMM-SUSS) catalogue (Page et al. 2012). The XMM-SUSS is compiled from images obtained through the six primary photometric filters of XMM-OM, which have effective wavelengths from 2120 Å (UVW2) to 5430 Å (V). XMM-SUSS 6.2 includes sources detected in any of the six optical and ultraviolet photometric filters.

For XMM-OM counterparts, 5XMM contains the corresponding source ID in XMM-SUSS 6.2, the match-probability to the X-ray source (using an NWAY-like algorithm, Salvato et al. 2018) and the following information for each and every XMM-OM passband in which the counterpart is detected: AB magnitude and magnitude uncertainty, a quality flag, an extended flag, a χ² value and the degrees of freedom for which it is calculated. The AB magnitude and uncertainty provided for each band is a weighted mean of the measurements over all XMM-Newton observations in which the source is detected, and the corresponding magnitude uncertainty. The quality flag is an integer equivalent to a binary number in which each bit corresponds to a different quality issue; a bit is set to 1 when a data quality concern is identified or otherwise set to 0. Sources with the highest quality in the corresponding photometric band will thus have a value of 0. These flags are :

The meaning of the quality flags (columns OM_filter_QUALITY_FLAG) in 5XMM are as follows:

bit 0 (value 1)	source on a bad pixel
bit 1 (value 2)	source on a readout streak
bit 2 (value 4)	source on a smoke-ring
bit 3 (value 8)	source on a diffraction spike
bit 4 (value 16)	source affected by Mod-8 pattern
bit 5 (value 32)	source within the central enhancement
bit 6 (value 64)	source near a bright source
bit 7 (value 128)	source near the edge
bit 8 (value 256)	point source within an extended source
bit 9 (value 512)	weird source (bright pixel)
bit 10 (value 1024)	multiple exposure values within photometry aperture
bit 11 (value 2048)	the source is affected by the reduced sensitivity patch [**]
bit 12 (value 4096)	the source is too bright (rate > 0.97 c/frame)

The extended flag is set to 0 if the counterpart is consistent with the shape of a point source in the corresponding band or 1 if the source appears extended. Note that it is possible for a source to appear point-like in one passband but have measurable extent in another. Where a source has been detected in multiple XMM-Newton observations in the same XMM-OM passband a χ² value is computed for the sequence of magnitude measurements compared to a single, constant magnitude. The corresponding degrees of freedom is one less than the number of measurements in that band. The χ² divided by the degrees of freedom can be used as an indicator as to whether there is evidence for variability between observations in that XMM-OM passband. Where variability is suggested by the χ², the individual measurements can be consulted in XMM-SUSS 6.2. Caution is advised in inferring variability from χ² when the corresponding quality flag is other than 0, or for sources which appear point-like in some XMM-Newton observations and extended in others, because the photometry is measured differently for extended and point-like sources (see Page et al. 2012).

XMM-OM counterparts have been classified probabilistically into Galactic and extragalactic source types and this classification information is included in 5XMM; see Section Classification for more details.

3.8 Redshifts

Photometric redshifts (photo-z) were calculated by selecting all 5XMM sources classified as AGN (see Section Classification) and outside the galactic plane (|b| > 20°). We used the optical and NIR-MIR photometry provided in their SEDs (Section Products) and calculated photometric redshifts for these sources. We used two different algorithms for estimating redshifts: TPZ, a machine learning algorithm (Carrasco Kind & Brunner, 2013) using the training sample described below MLZ-TPZ is a machine learning algorithm based on a supervised technique with prediction trees and random forest. The photometric redshifts and the corresponding cross-validation of the results was done through the photo-z pipeline developed for 5XMM, which includes a k-fold cross-validation method for evaluating the accuracy and reliability of our method, the selection of the optimal feature set for photo-z calculations using a Recursive Elimination Feature algorithm, and the quality evaluation of the individual photo-z, by using the shape of the redshift probability distribution given by TPZ. We also used the LePhare alogorithm, a template fitting algorithm (Arnouts et al. 1999, Ilbert et al. 2006), an SED template fitting code, well adpated to find high-redshift sources that would be otherwise missed by TPZ, since the results of machine learning methods are limited to the redshift range of the corresponding training sample (redshifts below 3.5 in our case). Moreover, LePhare allows us to estimate redshifts for sources with only partial photometry in the optical or infrared bands. We used two different sets of templates for LePhare, depending on the optical morphological classification of the sources. For extended objects we used the templates proposed by Salvato et al. 2009, 2011 for the COSMOS survey. For point-like objects we used the eFEDS templates (Salvato et al. 2022).

We compiled a large spectroscopic sample of X-ray selected extragalactic sources that can be used for the training and cross-validation of the machine learning and template fitting algorithms we used for calculating photometric redshifts. The training sample was selected from the second version of the Millions of Optical-Radio/X-ray Associations (MORX) Catalogue (Flesch 2024). We selected sources with secure spectroscopic redshifts, with an X-ray counterpart and classified as AGN or galaxies. We defined four different subsamples based on the photometry available in these large area optical surveys: SDSS sample (~55,000 sources), PanSTARRS sample (~47,000 sources), SkyMapper sample (~6000 sources), and DES sample (~14,000 sources). Ancillary photometry in the near- (from the 2MASS, UKIDSS and VHS surveys) and mid-infrared (AllWISE catalogue) was included if available.

This training sample is an order of magnitude larger than those previously used in similar efforts for estimating photometric redshifts for X-ray sources using machine learning techniques (e.g. Mountrichas et al. 2017, Ruiz et al. 2018). A variety of validation techniques for the photometric redshifts were carried out (see Webb, Traulsen et al. 2026 for more details). The spectral redshifts are provided in the 5XMM catalogue under the column REDSHIFT_ZSP. 31831 spectral redshifts are provided in 5XMM-DR15 and 154734 photometric redshifts (REDSHIFT_TPZ_Z_BEST and REDSHIFT_LPH_Z_BEST, along with the confidence limits on these redhifts and a link to further information on distance determination per source).

3.9 Classification

Both the X-ray sources and the optical / ultraviolet sources in XMM-SUSS 6.2 have undergone a classification using an adapted version of the Naive Bayes classifier CLAXBOI (Tranin et al. 2022). For the X-ray sources this algorithm uses the XMM-Newton X-ray properties of each source such as the hardness ratios, spectral fits (with a power law, but also an APEC model), along with the X-ray to r-band flux ratio, the X-ray to W1 infra-red band ratio when these complementary data are available, the maximum X-ray variability, the X-ray luminosity when the distance is known from Gaia or the Glade+ catalogue (Dalyà et al. 2022) and the distance to the centre of the galaxy in case of extra-galactic sources associated with a galaxy. For the X-ray sources, the most-likely classifications are given in the column CLASSX_CLASS. These are AGN, star, Galactic X-ray binary, cataclysmic variable, background AGN, extra-galactic X-ray binary and extended sources. We also provide an outlier classification, for the case when none of the above source-types matches the source. Seven further columns provide the probability attributed to each classification. This allows the user to make an informed decision about the reliability of the classification. For the X-ray sources, there are 556337 AGN, 119661 stars, 26100 Galactic X-ray binaries, 1276 cataclysmic variables, 49969 background AGN, 22732 extra-galactic X-ray binary and 42581 extended sources. The higher the outlier value (maximum 10), the more likely the source does not fit any of the designated categories, but see also Tranin et al. (2022). 49404 sources have an outlier value greater than five, 3943 have the best classification as AGN, 17834 have the best classification as a star, 15554 as a Galactic X-ray binary, 3657 as extra-galactic X-ray binaries and 2394 as extended sources. This could imply that these are extreme types of each classification, or indeed, different objects.

For the XMM-OM sources only three source classes were retained, quasar (QSO), galaxies and stars. The most probable classification is given in the 'CLASSOPT_CLASS' column. Again the probability attributed to the three classes for each source is provided in the three subsequent columns. A total of 201536 sources have an optical classification. There are 66306 QSO, 29971 galaxies and 105259 stars. Of the 20564 sources with both an X-ray classification of star and an XMM-OM classification, all of the sources are classified as stars, implying that the classification is reliable. The other classifications are more complicated to compare as an AGN does not necessarily have the same definition as a quasar, however, of the X-ray sources classified as an AGN and with an XMM-OM classification, two-thirds are classified as a quasar.

3.10 Hot areas in the detector plane

Warm pixels on a CCD (at a few counts per exposure) are too faint to be detected as such by the automatic processing, but can either push faint sources above detection level, or create spurious sources when combined with statistical fluctuations. This is an intrinsically random process, not visible over a short period of time, but which creates hot areas when projecting all sources detected over 24 years onto the detector plane.

We addressed this by projecting all sources onto CCD coordinates PN/M1/M2_RAWX/Y, keeping only sources above the detection threshold with the current instrument alone. In that way, we can distinguish hot areas coming from different instruments. We proceeded to detect hot pixels or columns in each CCD, using a similar method to the SAS task embadpixfind. For more information see Webb et al. (2020). Many of the warm pixels were not present at the beginning of the mission, and some appear for a short amount of time. So we tested each hot area for variability using revolution number, and the same Kolmogorov-Smirnov-based algorithm used to detect segments of bright columns, compared to the reference established over all sources on all CCDs and all instruments. This resulted in a revolution interval for each hot area.

Sources on a hot area for a particular instrument and within the corresponding revolution interval are flagged with flag 10 (PN_FLAG, M1_FLAG or M2_FLAG) as T (true) and then propagated to the SUM_FLAG to indicate a possibly spurious detection/source). In version 5XMM-DR15 this flagging was done after source detection, but from versions DR16, this step will be carried out before the stacked source detection. Below we provide the quality flags in 5XMM.

Table 3.1:  Meaning of the characters in the quality flags PN_FLAG, M1_FLAG, M2_FLAG, EP_FLAG and their distribution in 5XMM-DR15.

Flag	Description	EP		PN		M1		M2
all		818656	100%	753847	100%	692614	100%	777900	100%
0	No warning issued	537679	66%	550693	73%	594018	86%	679272	87%
1	PSF coverage $<$ 50%	160445	20%	97178	13%	42025	6%	41749	5%
2	Near a bright point-like source	4090	0%	3875	1%	3588	1%	3876	0%
3	Near a bright extended source	60955	7%	56180	7%	54894	8%	58519	8%
4	Extended near a bright point source	728	0%	683	0%	629	0%	675	0%
5	Extended near a bright extended source	12201	1%	8203	1%	9917	1%	11088	1%
6	Extended, significant in one band	6146	1%	5467	1%	5644	1%	5862	1%
7	Extended, flag 4, 5, or 6	14605	2%	11699	2%	12654	2%	13694	2%
8	On a bad pixel or CCD area	24433	3%	24342	3%	170	0%	34	0%
9	Near a bad CCD area	65913	8%	65711	9%	476	0%	268	0%
10	On a warm CCD pixel	13662	2%
11	Flagged during visual screening	54516	7%

Table 3.2: Meaning of the SUM_FLAG values

Value	Description
0	Good
1	if the warning flags EP_FLAG 1, 2, 3 or 9 set to true but not 7, 8, 10 or 11
2	if the possibly-spurious or warm pixel flags EP_FLAG 7, 8 or 10 set to true but not the manual flag 11
3	if the manual flag EP_FLAG 11 is set to true but not the spurious or warm pixel flags 7, 8 or 10
4	if the manual flag 11 as well as one of the spurious or warm pixel flags 7, 8 or 10 are set to true

The default value of every flag is F for False. When a flag was set it means it has been changed to T for True.

The task dpssflag sets all flags except the camera-specific flags (i.e., flags 2,3,4,5,6,7) on the summary row (EPIC band 8) which are then propagated backwards to the individual cameras and bands.

4. Catalogue content and organisation

This section summarises the organisation of the catalogue and gives details of all the columns. Known problems with parameters presented in the catalogue or with products associated with it are listed in Sec. 6.

There are 421 columns in the catalogue; they are grouped together and explained in the links below.

For each observation there are up to three cameras with one or more exposures which were merged when the filter and submodes were the same (2XMM UG, Sec. 2.2). The data in each exposure are accumulated in several distinct energy bands (Table 1). Camera-level measurements can further be combined into observation-level parameters. Consequently, the detection/source parameters can refer to some or all of these levels: on the observation level there are mean parameters of the source (prefix 'EP'); on the camera level the data for each of the three cameras (where available) are given (prefix 'PN', 'M1', or 'M2'), and on the energy band level the energy-dependent details of the detection parameters are given (indicated by a 'b' in the column name where b = 1,2,3,4,5,8). Detections (r upper limits) from individual observations were stacked to provide source data.. Also provided in 5XMM is one line comprising the information related to the stacked unique source. Not all data is given for either source or detection level information.

The column name is given in capital letters, the FITS data format in brackets and the unit in square brackets. If the column originates from a SAS task, the name of the task is given to the right hand side and a link is set to the SAS package documentation with which the data in the 5XMM catalogue was processed. It should be pointed out that the SAS used for the bulk reprocessing (for 5XMM) was from manifest pipeline version 21.51, which is based on SAS 21. A description of the column and possible cross-references follow.

Details of the columns:

Part 1:	7 columns: Identification of the source
	This includes the basic identifiers, IAU name, source number from the pipeline processing, number of observations in the stack for a unique source, the number of contributing observations for which the source position is inside the field of view, and the number of individual instrument observations of the unique source.
Part 2:	27 columns: Coordinates
	The position of the source is the result of the simultaneous fit and considered to be the same in all contributing observations and images, while the number of source counts are determined separately per image. They are given in equatorial, galactic and image coordinate systems in the RA, DEC, symmetrical and elliptical errors, LII, BII, and X_IMA, Y_IMA columns, respectively. The CCD number and raw x and y positions per instrument. Also given are the distance to the nearest neighbour and the number of sources fitted simultaneously.
Part 3:	241 columns: Source parameters
	The parameters of the source detection as derived from the SAS tasks emldetect.
Part 4:	7 columns: Detection flags
	This part lists the flags to qualify the detections. The summary flag, which gives an overall assessment for the detection, is followed by particular flags for each camera. A flag each is given if there exists at least one time series or one spectrum for this source.
Part 5:	34 columns: Detection variability
	This part gives variability information for those detections for which time series were extracted and for XMM-Newton observation to observation variability as a function of energy bands.
Part 6:	9 columns: Details of the observation and exposures
	This part provides details on the first and last observation in a stack, the revolution that the observation was taken in, the submodes and the filters of each observation.
Part 7:	95 columns: Added value products
	This part provides source classifications, multiwavelength counterparts, fits to the full extracted spectrum, very long-term variability and redshift information for extragalactic sources.

5. Catalogue Properties

This section summarises the main properties of the catalogue but does not provide a detailed analysis. A comprehensive evaluation of the catalogue is presented in the 5XMM catalogue papers Webb, Traulsen et al., (2026), draft version available currently.

5.1 Overview

The catalogue contains source detections drawn from 14616 XMM-Newton EPIC observations made between 2000 January 19 and 2024 October 14 and which were publicly available by 2024 November 30. Net exposure times in a stack range from < 1000 up to 2.7 million seconds. Figure 5.1 shows the distribution of fields on the sky.

The sky area of the catalogue observations corrected for field overlaps is ~1397 deg².

The catalogue contains 2578752 X-ray detections or upper limits with total-band (0.2 -12 keV) likelihood values  ≥  6. These are detections of 818656 unique X-ray sources, that is, 411307 X-ray sources have multiple detections in separate observations (up to 98 observations). Of the 818656 X-ray sources, 42699 are classified as extended with 25845 of these being in regions considered to be 'clean' (SUM_FLAG  < 1).

5.2 Data quality

As part of extensive quality evaluation for the catalogue, each field has been visually screened. Regions where there were obvious deficiencies with the automatic source detection and parameterization process were identified and all detections within those regions were flagged (cf. 2XMM UG, Sec. 3.2.6 but importantly, note Section 3.11). Such flagged detections include clearly spurious detections as well as detections where the source parameters may be unreliable. For most uses of the catalogue it is recommended to use SUM_FLAG as a filter to obtain what can be considered a 'clean' sample.

Note that no attempt is made to flag spurious detections arising from statistical fluctuations in the background.

5.3 Sensitivity and Photometry

Figure 5.2 presents, for each of the three cameras, the distributions of the stacked fluxes in the soft, hard and total band. These give an indication of the limiting flux available in the catalogues for each of the bands.

5.4 Astrometry

Considerable improvements have been made to improve the astrometry for 5XMM and these are detailed in Webb, Traulsen et al., (2026), draft version available currently.

5.5 Products

Spectral Energy Distributions (SEDs)

Spectral energy distributions (SEDs) are provided for 5XMM-DR15 sources, produced using ARCHES, a tool developed under The Astronomical Resource Cross-matching for High Energy Studies (ARCHES) framework under the EU Horizon ARCHES project (Motch et al. 2017). The ARCHES algorithm (Pineau et al. 2017) performs a simultaneous cross-match of all selected archival catalogues, computing a probability for every possible combination of catalogue entries. These probabilities are derived from the likelihood that sources from different catalogues share the same sky position, taking into account their astrometric uncertainties. The method is based exclusively on positional information and associated astrometric uncertainties, without the inclusion of photometric or other auxiliary source properties in the matching likelihood.

For each combination of catalogue entries, excluding high optical density regions such as the LMC, SMC, and M31, a cross-match probability is computed based on the likelihood that all sources share the same true sky position. This process led to the removal of 19 stacks comprising 708 observations. The final association probability further depends on a prior, representing the likelihood that an X-ray source has a genuine counterpart in a longer-wavelength catalogue. This prior is estimated empirically from observed association rates while accounting for spurious matches. The method assumes ideal conditions, such as accurate astrometry, negligible systematics, and uniform source densities, while ignoring effects like proper motion or blending. To reduce these limitations, additional preprocessing steps, such as clustering by source density, are operated.

As for 4XMM (Webb et al. 2020) we selected a number of archival catalogues to cover the largest sky area and energy bands from the UV to radio. We grouped catalogues by wavelength coverage to produce single master UV, optical and infrared catalogues and perform a statistical cross-match of four catalogues (X-ray to IR). Some of the catalogues are all sky (e.g. Gaia DR3, 2MASS), while other cover only partially the areas covered by the 5XMM-DR15 stacks, notably GALEX in the UV. Probabilities depend on the source density of each of these catalogues which in turn depends on the depth and the covered area covered by each catalogue and the depth. We therefore calculated the intersected area covered by each of the several groups of stacks with each of the catalogues using mocpy and set the area to the corresponding value.

The set of SEDs are available as individual FITS files and graphical output for the three highest probability SEDs via the SEDFinder service and through a link in the catalogue in the INFO_COUNTERPARTS column.

Auxiliary stack data

For each stack, eleven combined files in FITS format are provided: * six mosaic images in the full 0.2–12.0 keV band and for each of the standard bands 0.2–0.5 keV, 0.5–1.0 keV, 1.0–2.0 keV, 2.0–4.5 keV, 4.5–12.0 keV, * three coverage maps indicating un-exposed (0) and exposed (1) pixels, the number of observations covering a sky pixel, and the number of exposures covering a sky pixel, * and an un-vignetted and a vignetting-corrected exposure map.

Auxiliary source images

Four types of auxiliary images are produced for the stacked catalogue and provided in the XSA interface:

1. a broad-band X-ray image in the 0.2–12.0 keV range,
2. a three-colour X-ray image in the energy bands 0.2–1.0 keV, 1.0–2.0 keV, 2.0–12.0 keV, corresponding to the 4XMM standard energy bands 1 plus 2, 3, and 4 plus 5,
3. an optical finding chart, taken from the first survey that covers the source position among Pan-STARRS g-band imaging, SDSS g-band imaging, DESI Legacy Imaging Surveys, or broad-band data of the ESO Digitized Sky Survey,
4. and, if the source has non-zero flux during at least two observations, a long-term light curve, showing the stacked EPIC flux value and the EPIC flux during the contributing observations. Different plot symbols are used to indicate tentative short- and long-term variability. The stacked flux is plotted with a filled circle, if the probability VAR_PROB is 10^–5 or lower. It indicates whether the source fluxes are consistent with being constant or not.

All images have a side length of 10 arcminutes, are centered at the X-ray source position, and are annotated with information on the position and detection quality.

Figure 5.3 : Top left : a broad-band X-ray image in the 0.2–12.0 keV range. Top right: a three-colour X-ray image in the energy bands 0.2–1.0 keV, 1.0–2.0 keV, 2.0–12.0 keV, corresponding to the 4XMM standard energy bands 1 plus 2, 3, and 4 plus 5. Bottom right: an optical finding chart, taken from the first survey that covers the source position among Pan-STARRS g-band imaging, SDSS g-band imaging, DESI Legacy Imaging Surveys, or broad-band data of the ESO Digitized Sky Survey. Bottom left: a long-term light curve, showing the stacked EPIC flux value and the EPIC flux during the contributing observations.