X-ray properties of Galactic HMXBs corrected by interstellar absorption

Marina Belén Badaracco · Professors: Dr. Matías Zaldarriaga & Dr. Nahuel Miron Granese
Work carried out with Claude Code 2.1.278 (Opus 5, effort high) · 28 September 2026

The column density measured towards a high-mass X-ray binary is a sum: part interstellar, part produced by the donor's wind and the wake trailing the compact object. Only the second is a property of the binary, and it is not constant — it varies around the orbit. This project measures it as a function of orbital phase, homogeneously, across the Galactic sample.

The gap this fills

Fortin et al. (2023) compile 152 Galactic HMXBs with 111 Gaia DR3 counterparts and distances, but tabulate no X-ray flux, column density or spectral slope. Neumann et al. (2023) give fluxes in five bands and one column density, taken from the Swift catalogue, where it came from interpolating a power law through hardness ratios rather than from a spectral fit. For a population lying in the Galactic plane this is the gap that matters: below about 2 keV an HMXB spectrum is shaped more by absorption than by emission.

Doroshenko et al. (2024) make the separation possible with a pan-Galactic cumulative-reddening cube: queried at the Bailer-Jones distance it returns the interstellar column, and the remainder of a fitted N_H is the local one. The subtraction cannot be done with point values — the distance has an asymmetric posterior, the cube its own calibration uncertainty, the conversion depends on the abundance table — so a Monte Carlo over distance, cube and conversion is required for the residual to have an honest error bar.

Why the catalogue columns cannot simply be used. A catalogue publishes a best-fit column and at most a 68 % interval; the probability distribution behind that number is not published, neither its shape nor its covariance with the photon index. Subtracting a distribution from a point estimate with a symmetric error bar does not propagate the error, it invents one — and the result is then binned in orbital phase, where the scatter between bins is the signal. The fits have to be ours, made in XSPEC with each posterior kept rather than summarised.

Where the work stands

Quantity	Value
Systems with at least one X-ray counterpart (the working sample)	89
— with a CSC 2.1.1 source / a 5XMM-DR15 source / both	61 / 60 / 32
Systems with a published orbital period	72
Published phase-zero epochs read by hand from the literature	57
— distinct conventions among them	10
Systems brought onto a common convention (periastron passage)	33
Systems with a usable orbital phase on their observations	31
Observations carrying a phase	134 of 140
Reprocessed observations (Chandra / XMM-Newton)	194 / 193
Broad-band images with source detection (Chandra / EPIC)	193 / 561
Stage 1: Chandra detections reaching the published statistic	105 of 115

Bringing the orbital phases onto one convention

This is the part with no established recipe. The 57 epochs are of ten kinds — periastron passage, mean longitude 90°, mid-eclipse, conjunctions, outburst and optical maxima, an ellipsoidal minimum, a spectroscopic phase zero, and eight origins chosen only to fold a light curve. A phase measured from one is not the phase measured from another, so any statement comparing phases between systems is empty until they are referred to one point.

They are referred to the periastron passage. An epoch of mean longitude 90° converts exactly and needs only the longitude of periastron; a conjunction or a mid-eclipse requires the true anomaly at that configuration and then Kepler's equation. A third route needs no geometry: Fornasini et al. (2023) state that the Type I outbursts of Be X-ray binaries happen quasi-periodically at or near periastron, and that their recurrence is how the orbital period of such a system is measured — so for a Be XB an epoch of maximum flux is a periastron passage. Whether a system qualifies is decided by the catalogue's own class column, not by judgement: five of the ten outburst epochs are adopted, the other five belong to supergiants and SFXTs and are not.

One trap is worth stating because getting it wrong would exchange the answer with its opposite. A longitude of periastron describes somebody's orbit, and the two possibilities differ by exactly 180°: pulse timing gives the compact object's, radial velocities of the companion give the donor's. The frame is recorded per system, decided from what each paper actually quotes.

What the sample looks like

Mosaic of the 14 observations of SS 433 in time order — **SS 433** — 14 observations in time order. It has a superior-conjunction epoch and no published longitude of periastron, so it cannot be placed on the common convention and is ordered by MJD instead.

Mosaic of the 17 observations of LS I +61 303 in phase order — **LS I +61°303** — 17 observations from phase 0 to ~0.99.

Mosaic of the 41 observations of PSR B1259-63 in phase order — **PSR B1259−63** — 41 observations in phase order, the largest set in the sample.

Each tile is one observation, centred on the source and marked with a 10″ circle; the number above it is the orbital phase, or the MJD where none exists. The colour scale is raw counts per image pixel, uncorrected for exposure or vignetting, square-root stretched to the 99.5th percentile of that tile alone, so brightness is not comparable between tiles. In a three-colour tile red, green and blue are the counts of EMOS1, EMOS2 and EPN. Mosaics are produced by Observations/Images/ds9_phase_mosaic.sh, one command per system.

What has been checked, and what has not

Every statement in the summary is traced in work/PROVENANCE.md to the file and function that produced it, with the check that was run beside it and an explicit note of what was not checked. A few examples of the second kind, so the shape of the record is clear: no confidence limits are computed on our own Stage 1 parameters, so the comparison is our best fit against the catalogue's interval and not interval against interval; the ten Chandra detections that never reach the published statistic have not been diagnosed; apsidal motion is not applied; and the systematic that rides on the five Be outburst epochs, the width of the outburst, is not quantified.

The images are not all pointings at their system. The observation lists hold every observation whose field covers the position, and the source lies beyond 6 arcmin off-axis in 69 of the 193 Chandra observations and beyond 10 in 32. An empty tile can mean the source was not detected, or that the telescope was aimed elsewhere. They are kept and studied: a large off-axis angle belongs in the error budget of a measurement, not in a cut on the sample.

Repository layout, environment and reproduction steps are in README.md. The bulk data — 367 GB of reprocessed observations and images, the 5XMM-DR15 catalogue, and the source papers — are not in the repository; the README says how to obtain each and which script rebuilds it.