Why does an X-ray telescope look like nested tubes?

A radio telescope may have a large metal dish. A visible-light telescope has a broad mirror facing the stars. Yet the front of an X-ray telescope often resembles dozens of thin tubes nested inside one another. Did engineers forget how to make ordinary mirrors for shorter wavelengths?

X-rays are reflected efficiently at very small angles to a surface. Two correctly shaped successive surfaces direct them to a focus, while nested shells enlarge the narrow annular area that collects light. Geometry, coatings and precision all depend on photon energy.

We will follow a photon and see how XMM-Newton mirrors were made, then compare them with Chandra and NuSTAR. Two earlier ideas are enough: signal and resolution are different properties, and a mirror substrate and its coating have different jobs. Mathematical details appear in visible optional branches. Chandra: the mirror assembly.

First, get the detector above the atmosphere

X-ray astronomy investigates hot gas and high-energy particles: shocks in supernova remnants, gas in galaxy clusters and matter near compact objects. During accretion, matter falls toward a neutron star or black hole and can emit X-rays. This is light from matter around the black hole, not from inside its horizon. Spectral lines, continuum emission and brightness changes reveal composition, temperature, motion and physical processes; not every X-ray photon has a thermal origin. NASA: the X-ray sky.

Earth’s atmosphere absorbs cosmic X-rays. Taking an ordinary telescope up a dry mountain is insufficient. In 1962 a rocket experiment by Riccardo Giacconi’s group detected Scorpius X-1 beyond the Solar System. Events were counted with a detector lacking today’s large focusing optics. A brief flight revealed the source; orbit provided time for systematic study. NASA: rocket observations and accretion.

Uhuru, launched in 1970, surveyed the sky with proportional counters. Einstein, launched in 1978, applied a focusing Wolter-I telescope to cosmic sources and extended objects. This was a transition from “where is the excess count rate coming from?” to a much more detailed image. Rocket telescopes had already obtained earlier X-ray images of the Sun: 1978 did not mark the beginning of every kind of X-ray imaging. HEASARC: Uhuru, Einstein, NASA: focusing optics history, §2.1.

Measure the grazing angle from the surface

X-rays are electromagnetic radiation too, but their interaction with matter differs from that of visible light. An ordinary mirror struck nearly head-on will not return the required fraction of light into the telescope: photons are mostly absorbed or transmitted, depending on material, thickness and energy. Reflectivity can rise substantially when a ray travels almost along the surface.

The angle between a ray and the surface itself is the grazing angle α. It is small; the angle to the normal, perpendicular to the surface, is then almost 90°. Confusing these two angles reverses the meaning of a drawing. The reflected ray obeys the equal-angle law, but the material and wavelength determine the reflected energy fraction. The geometric law alone does not guarantee high efficiency. LBNL: grazing-incidence reflectivity.

For a single-layer coating, higher energies generally require smaller angles. Atomic absorption edges introduce structure, so one should not draw the same smooth curve for every metal. Soft X-rays have lower energies and hard X-rays higher ones; the precise boundary depends on the field and instrument. A practical distinction is that optics efficient at 1 keV may lose almost all useful signal at 10 keV and the same angle. NASA: energy and grazing reflection, §2.2.

Gold: energy and grazing angle change reflectivity

R · intensity fraction00.250.50.7511246810Photon energy · keV
  • α=0.25°
  • α=0.5°
  • α=1°
  • α · angle to surface
Fresnel calculation, s-polarization, smooth semi-infinite vacuum–Au interface; α measured from the surface. Three curves use 143 unchanged LBNL/CXRO f₁/f₂ rows from 1.01586 to 9.91555 keV, ρ=19.3 g/cm³. Lines only join tabulated samples: fine absorption-edge shape and location are not recovered. One interaction, not two reflections or a mission effective area. No finite film, nickel substrate, roughness, contamination or detector. Linear axes. Source and conditions.

Branch · With formulasAn ideal smooth vacuum–bulk-gold interface, not a measurement of a finished XMM mirror.

How energy enters the reflectivity calculation

Photon energy E and vacuum wavelength λ are related by:

λ = (hc)/(E),; λ[nm] ≈ (1.239842)/(E[keV]).

Here h is Planck’s constant and c the speed of light; 1 keV is one thousand electronvolts. At 1 keV the wavelength is about 1.240 nm, and at 10 keV 0.1240 nm. Energy is converted to joules for the calculation. NIST: SI/CODATA constants.

Write the complex refractive index as n = 1 − δ − iβ: δ describes the real part’s departure from unity and β absorption in this phase convention. Both depend on energy and material. For weak absorption and small angles, the critical grazing angle is approximately α_(c) ≈ √(2δ) in radians. This explains a trend, but does not supply an actual reflectivity curve near absorption edges. CXRO: optical constants and limits of the atomic model, critical angle.

The graph above uses the Fresnel equation for s-polarization, with the electric field perpendicular to the plane of incidence:

R_(s) = (∣(sin⁡α − √((n)^(2) − (cos⁡)^(2)α))/(sin⁡α + √((n)^(2) − (cos⁡)^(2)α))∣)^(2).

It uses 143 unchanged gold f₁/f₂ table rows from 1.01586 to 9.91555 keV, a density of 19.3 g/cm³ and molar mass of 196.966569 g/mol. The table determines δ and β; they were not chosen to produce a desired curve. Lines join finite tabulated samples and do not recover fine edge structure. CXRO: Au data, LBNL: equation and density, NIST: Au atomic mass.

At 4.97377 keV and α=0.5°, the model gives R≈0.755 for one reflection. Two identical reflections leave R²≈0.570. At 9.91555 keV and the same angle, these become about 0.144 and 0.0206. A finished telescope can have different angles at its two reflections; roughness, finite layer thickness, contamination and geometric losses also matter. Even 250 nm of gold on nickel is not identical to the semi-infinite interface used here.

A paraboloid and hyperboloid: two reflections to one focus

In Wolter I, the front inner surface is part of a paraboloid of revolution. Rays parallel to the axis head toward its focus after the first reflection. Before reaching it, however, they encounter the second surface, part of a hyperboloid. It intercepts the converging beam and redirects it toward a nearer final focus. These surfaces are coaxial and share one geometric focus. NASA: Wolter-I-like geometry, §2.3.

One paraboloid can focus on-axis rays too. The second surface is needed for useful imaging and a compact nested arrangement, rather than because “one reflection cannot focus at all.” Off-axis aberrations remain; a real instrument does not have equally perfect sharpness throughout its field. Two arbitrary conical tubes do not form an exact paraboloid–hyperboloid pair.

Two reflections in an exact teaching Wolter-I pair

Optical axisParallel on-axis beam →P · paraboloidH · hyperboloidF · detector-202510x · arbitrary unit (same for y)
Our ideal confocal geometry: F=(10,0), Fₚ=(20,0) beyond the frame; junction (0,±2). P:−2≤x≤0; H:0≤x≤2. The same arbitrary unit has equal size on both axes; teaching angles are larger than typical flight angles. Two upper and two lower on-axis rays are shown. Intersections, normals, direction and constant path are independently checked; reflectivity, diffraction and off-axis PSF are omitted. Not an XMM/Chandra prescription. NASA-CR-162072 supplies the confocal surface forms; our parameters are in the branch. Source and conditions.

Branch · GeometryAn ideal on-axis beam and our own teaching dimensions, not a reconstruction of a mission's optical prescription.

How our double-reflection trace was checked

Let x increase along the light’s travel, and y be the distance from the axis in a meridional section. The final focus is F=(10,0), the shared focus Fₚ=(20,0), and the surface junction (0,2), all in the same arbitrary unit. Define the parabola and left hyperbola branch by:

(y)^(2) = 4p(20 + p − x),; ((x − 15)^(2))/((a)^(2)) − ((y)^(2))/((b)^(2)) = 1.

Here p≈0.04987562, a≈4.95085611 and b²≈0.48902380. These coefficients follow uniquely from the stated foci and junction: equal distances to a focus and directrix define the parabola, and a constant difference of distances to two foci defines the hyperbola. The used sections are −2≤x≤0 and 0≤x≤2. The surface forms follow NASA-CR-162072, §II.A, printed p.3 / PDF9; the numbers and axis direction are ours.

At each point we apply the vector reflection law \mathbf d'=\mathbf d-2(\mathbf d\cdot\hat{\mathbf n})\hat{\mathbf n}, where d is a unit ray direction and n̂ a unit normal. After the first reflection we find the first admissible intersection ahead of the ray on the finite second surface, then compute the second reflection. The drawing does not connect arbitrarily chosen points.

Independent checks confirm equal angles to a numerically determined normal, arrival at F and a constant path from entrance plane x=−3 to F: about 13.19803903 units for different rays. The mirrored lower half gives the same result. Our teaching angles are larger than typical flight angles so the elements can be distinguished; both drawing axes use the same scale. This calculation does not establish efficiency at a given energy, off-axis PSF or manufacturing tolerance.

Why one mirror sleeve is not enough

Look into the end of a grazing-incidence mirror. A ray near the wall may reflect twice and reach the detector. A ray through the large central opening may pass through without the required reflection at all. The outer diameter therefore does not mean the entire disc collects light. One channel’s useful entrance is a narrow annulus.

Smaller coaxial shells occupy the hollow space. Each has its own pair of surfaces and collects another annulus into the same focus. This fills the available volume: more nested shells add collecting area, as long as passages remain between them. Wall thickness, clearances and supports occupy space and obstruct light too. The number of shells cannot increase indefinitely. NASA: geometry and mass constraints, §2.3–2.4.

Nested shells add annuli, not full discs

Central openingShell × 3Passage between shellsSchematic support
Our axial view of three ideal teaching shells, R=1.2/1.6/2 arbitrary units. Colored annuli represent the geometric entrance of on-axis rays onto P from x=−2 to 0, not metal thickness; the central hole is not collecting area. Four schematic spokes show obstruction; they do not reproduce the 16 spokes and 58 shells of XMM. Real tolerances, thickness, vignetting and reflectivity are not computed. Source and conditions.

XMM-Newton launched on 10 December 1999. Each of its three X-ray telescopes has 58 shells and a focal length of 7.5 m. A photon does not reflect successively from all 58 mirrors. A useful event follows one pair of sections of its own shell. Chandra uses four nested pairs, rather than the same set of 58. XMM Handbook: launch and instruments, ESA: XMM design, Chandra: HRMA.

Branch · CalculationOur geometric example omits obstruction and energy losses; these dimensions are not attributed to XMM.

How much area does an annulus add?

For outer radius rₒ and inner radius rᵢ, the entrance annulus has area:

A_(ring) = π((r_(o))^(2) − (r_(i))^(2)).

At rₒ=20 cm and rᵢ=19 cm this is 39π≈122.5 cm². A full disc of radius 20 cm would have 400π≈1256.6 cm², more than ten times as much. The mirror surface area along the tube is not this projection onto the entrance plane either. In a real telescope, tracing both reflections determines the annuli; supports and adjacent shells may block part of the passage.

XMM: first make the optical surface on a mandrel

A thin flexible shell is difficult to polish and hold without deformation. XMM used replication: first make an accurate outer surface on a rigid master, the mandrel, then transfer it to the finished mirror’s inner surface. This is a particular electroformed-nickel technology, not a mandatory method for every X-ray mission.

  1. Mandrel. Coat an aluminum blank with hard Kanigen nickel–phosphorus. Shape and superpolish the required parabolic and hyperbolic sections. Check figure and microroughness separately: a precise overall shape can have a poor small-scale surface, and a smooth surface can have the wrong overall figure.
  2. Gold. Deposit the reflective layer onto the cleaned mandrel in vacuum. For this design it is 250 nm thick. The X-rays do not need to pass through the thick aluminum master.
  3. Nickel shell. Electrochemically build nickel over the gold to provide the mirror’s mechanical form. Thickness cannot be arbitrary or uneven: internal stress and thickness variations can bend the released part.
  4. Separation. Controlled temperature changes and different adhesion allow release. Gold comes away with nickel and becomes its inner reflecting layer. The mandrel does not fly inside the telescope. Rough extraction could spoil the figure at this very step.

ESA’s team documented this route and its technological problems during XMM manufacture. It is a professional process involving controlled coatings, chemistry and metrology, not a recipe for a home electroforming bath. ESA: Manufacture and integration / Mirror technology development.

Replication transfers gold from the mandrel to the shell

1 · Master2 · Build-up3 · After separationNi–PAlNi–PAlNiAuNiAuMandrel removedFuture inner surface · AuLight enters · inner side
  1. GeometryAxial profile and roundness → figure errors.
  2. MicroroughnessSeparate small-scale metrology → scattering.
  3. IntegrationFree versus mounted shell → support distortion.
  4. X-rays and environmentKnown energies in vacuum; repeat after vibration and thermal cycles.
Functional sections of the XMM process after ESA Bulletin 89 (1997), not thicknesses to scale or a chemical-processing recipe. Al master/Ni–P first, then Au and electroformed Ni. After separation, Au remains on the inner side of Ni; the Ni–P-coated master is removed. The arrow means separation, not an optical ray. Figure, roughness and mounting checks are separated below. Source and conditions.

Preserve the figure during nesting and mounting

After release, the shell’s own figure and small-scale surface are checked. XMM shells are then nested from smallest to largest, aligned on a vertical optical bench and bonded by their entrance edges to the support spokes. The suspension distributes loads during this operation so that gravity does not create an accidental figure. After mounting, optical performance is compared with that of the freely supported shell. ESA: Mirror-shell integration.

XMM’s support has 16 spokes and is made of Inconel, an alloy whose thermal expansion is close to that of the nickel shells. Its job is to retain relative mirror positions, not merely carry weight. If a mount expands differently as temperature changes, it can turn an accurate surface into a distorted one. Spoke obstruction and mechanical stability belong to the same trade-off. ESA: technical section and support.

Separate three problems. Figure errors change ray directions and blur the image. Microroughness scatters some light; its effect depends on the spatial scales of irregularities and photon energy. An unsuitable coating reflects the required band poorly even with the correct geometry. A single result, “the surface looks shiny,” cannot replace these checks. Particles and molecular films further change actual throughput. LBNL: smooth interface and roughness, Chandra: surfaces and calibration.

Chandra and NuSTAR: different substrates and coatings

Chandra has polished Zerodur glass-ceramic shells coated with reflecting iridium over an intermediate chromium binding layer. Here the accurate surface is made on the glass-ceramic itself, rather than releasing a nickel replica from a mandrel. Flat witness samples are coated alongside the mirrors, then tested separately for X-ray reflectivity. This helps distinguish coating properties from figure errors in the large assembly. Chandra: fabrication and witness flats, §4.1.

NuSTAR, operating at 3–79 keV, needs hard-X-ray optics. Its mirrors use a conical approximation to Wolter I and thermally formed thin glass segments. The segments sit on precision graphite spacers and are bonded with epoxy. Radial positions are referenced to a common axis to prevent errors accumulating from layer to layer. This is another manufacturing route, with its own sharpness limits. NuSTAR: observatory, Craig et al.: Fabrication of the NuSTAR flight optics.

NuSTAR coatings contain many alternating thin layers. Waves reflected at their interfaces can add in phase and enhance useful reflectivity. The flight modules use Pt/C and W/Si, platinum/carbon and tungsten/silicon. An older general NuSTAR page lists Pt/SiC; the specific flight-module publication and the authors’ coating record confirm Pt/C. A development or spare module’s material should not be assigned to the flight modules. Christensen et al.: flight coatings.

For a periodic multilayer, first-order enhancement approximately requires λ = 2dsin⁡α, where d is the thickness of a layer pair and α the angle to the surface. At 40 keV and α=0.25°, this gives d≈3.552 nm. This is a teaching substitution, not a NuSTAR coating recipe: real stack thicknesses vary with depth, and refraction, absorption and interface roughness matter. One constant period does not provide equal reflectivity throughout a broad band. LBNL: multilayer mirrors and Bragg’s law.

After the focus, you still need a detector and calibration

Mirrors map directions in the sky to locations in the focal plane. The detector converts an absorbed X-ray photon into an electrical signal. In a suitable operating mode, position, time and signal size are used to reconstruct an event; its energy has finite precision. Cosmic particles, noise and several photons overlapping during readout can interfere with that reconstruction. Smaller pixels cannot repair the wrong mirror figure.

In XMM-Newton, one telescope’s open path leads to EPIC-pn. In the other two, part of the flux reaches EPIC-MOS and part is dispersed by gratings toward RGS spectrometers. These are different functions: mirrors focus, gratings separate wavelengths and CCDs register charge. Camera response, optical response and the whole chain have different limits. XMM Handbook: light paths, EPIC.

Effective area asks what ideal collecting area would yield the same useful event rate at a given energy. It includes losses, so it depends on energy, direction and whether the detector, filter and grating are included. One cannot compare the combined area of all three cameras with that of one bare mirror module without specifying these details. XMM Handbook: on-axis effective area, off-axis vignetting.

Branch · Precise definitionsThis equation shows the budget's structure; it is not a new mission calibration table.

Three different quantities: area, image width and energy resolution

A simplified on-axis budget can be written:

\begin{gathered} A_{\mathrm{eff}}(E)=\sum_i A_{\mathrm{geo},i}\,Q_i(E),\\ Q_i(E)=R_{P,i}(E)R_{H,i}(E)\,\eta_i(E). \end{gathered}

Qᵢ is the product of all included survival/detection probabilities for channel i. A_geo,i is the unobstructed entrance area of channel i; R_P,i and R_H,i are the two reflected fractions; η_i includes all other included losses or detection probabilities. Do not count the same obstruction twice in both A_geo and η. Nonuniform angles across the aperture require averaging or ray tracing rather than one coefficient. The detector factor is omitted for mirror-only area. Chandra: geometric and effective areas, §4.2.2.

The PSF is the image distribution of a point source. FWHM measures the core width at half its maximum height. HEW, or HPD, is the diameter of the circle enclosing half the point image’s flux. Faint broad wings can change FWHM little while increasing HEW substantially. HEW here is neither a radius nor another name for FWHM. NASA: imaging metrics, §2.4.

XMM’s table lists FWHM=4.3″ and HEW=16.8″ for MOS1, on-axis, at 1.5 keV, in orbit. These are two descriptions of one chain, not contradictory “resolutions.” Pixels, readout mode and event pile-up must also be considered in this context. Energy width ΔE in keV or eV is a separate spectral measurement property and cannot be added to angular arcseconds. XMM Handbook, table2.

How do engineers prove the assembled telescope retained its performance?

Testing begins before assembly: surface profiles and roundness expose figure errors, microscopic metrology examines small irregularities and coated witness samples establish spectral reflectivity. Comparing performance before and after mounting reveals support-induced distortion. These tests are not interchangeable.

Next comes an X-ray test facility in vacuum with known energies, measuring a point image, flux and effective area. Mechanical and thermal-vacuum tests reproduce loads and temperature conditions; repeated optical testing reveals any instrument change. XMM’s qualification module contained 21 X-ray-quality shells and 37 mass-representative dummies. Its result should not be presented as the specification of all 58 flight-quality shells. ESA: optical and environmental testing.

The test facility has its own geometry. A star supplies an almost parallel beam, whereas a terrestrial source is at finite distance. Gravity also acts on flexible shells. Ground measurements therefore become an orbital prediction through a validated model, refined by calibration. Chandra’s documentation explicitly distinguishes these conditions. A lower ground-test flux does not always mean a damaged coating. Chandra: ground calibration and model, §4.2.1.

Here we checked documentation, our own geometric trace and teaching reflectivity calculations. The author has not physically manufactured shells, deposited coatings or tested an X-ray telescope. Mission measurements belong to their teams; a computer model is not our own physical experiment.

What the nested shells now mean

A shell lets a ray encounter an inner surface at a small angle. A second correctly shaped section directs it to the final focus. Nesting adds collecting annuli; material, coating, figure and support determine which photons actually produce useful events.

You can now test three intuitive answers. Does a photon reflect from all 58 XMM shells? No: a useful path has two reflections on its own shell. Does more area always mean a narrower spot? No: collecting signal and blurring an image are different properties. Is a harder metal enough for harder X-rays? No: optical constants, angles and layer structure matter, not the everyday meaning of “hard.”

Compare this path with the ultraviolet telescope and STIS detector. Both require the entire photon path to work together; in an X-ray instrument, that requirement is visible in the shape of the entrance itself. At still higher energies, methods can reconstruct direction without an ordinary focusing objective, a separate subject in gamma-ray astronomy.