How can a telescope find a source without an ordinary objective?
Imagine a photographer receiving a list instead of a photograph: “At this second, these silicon strips fired and this much energy was deposited in these crystals.” Where could a picture of the sky come from? A space gamma-ray telescope starts with records rather like these.
A gamma-ray telescope infers direction from the effects of a photon interacting with matter: particle tracks, energies in successive interactions or the shadow of a coded mask. Many events, together with calibration and a background model, allow a source to be reconstructed. There is no ordinary eyepiece, but measurable traces preserve information about the light.
We will follow this process through Fermi LAT, then compare it with COMPTEL and SPI on INTEGRAL. Their materials and reconstruction methods differ. It helps to remember how the amount of signal differs from resolution. You can skip the mathematical branches; the main explanation does not depend on them. NASA: the LAT instrument.
First, the detector had to rise above the atmosphere
Most cosmic gamma-ray photons do not reach Earth’s surface in their original form: they interact with the atmosphere. Direct measurements therefore need rockets, high-altitude balloons or satellites. Orbit permits long observations, but does not remove the background: charged particles fly around the spacecraft, while Earth and the spacecraft’s own materials can also produce gamma rays. Ground-based observatories exist, but measure the consequences of atmospheric particle showers, a different observing method. NASA: observing platforms, air-shower detectors.
On 27 April 1961, Explorer 11 carried the first gamma-ray telescope into orbit. Counters and charged-particle rejection were already part of its design: the story began with identifying rare events, rather than a large glass objective. Compton Gamma Ray Observatory, launched in April 1991, combined four different instruments; COMPTEL imaged the sky using Compton events, while EGRET used pair conversion with an earlier detector technology. Fermi launched on 11 June 2008; its LAT uses a silicon tracker. This is continuity in physical principle, rather than identical construction. HEASARC: Explorer 11, CGRO, a historical account from 1998, the LAT team, §1.
Why collect such complicated records? For pulsars, researchers look for periodicity in arrival times; for gamma-ray bursts, rapid changes in flux and spectrum; for active galactic nuclei, radiation from high-energy particles, including those in jets. Nuclear lines, such as the aluminium-26 decay line near 1.809 MeV, help investigate the production of elements. Direction, energy and time answer different parts of the question; a bright point alone does not explain the physical process. NASA: the gamma-ray sky, COMPTEL and nuclear lines.
Choose the design for the interaction and energy
Gamma rays are electromagnetic light, with high energy per photon. A household lens or ordinary mirror does not provide the required broadband image in these instruments’ energy ranges. This should not become an absolute claim that “a gamma-ray photon can never be reflected”: X-ray and lower gamma-ray ranges partly overlap, and specialised optical methods have their own narrow conditions. Here we examine three methods used in actual instruments.
| Example | What information is measured? | How is direction obtained? |
|---|---|---|
| Fermi LAT, an overview range of about 20 MeV to beyond 300 GeV | Pair tracks and shower energy | Reconstruct the initial particle directions, with corrections |
| COMPTEL, approximately 0.8–30 MeV for imaging | Two positions and two energy deposits | One event defines a circle of possible sky directions |
| SPI, a coded mask | Distribution of counts across a detector plane | Compare the shadow with responses for different directions |
These ranges describe particular instruments, rather than three impermeable boundaries in nature. Different interactions are possible at the same energy; designers choose materials, geometry and event selection so useful information can be measured reliably. A gamma-ray telescope can also be primarily a spectrometer, like SPI, while reconstructing images. NASA: LAT, NASA: COMPTEL, §III, ESA: SPI.
LAT turns a photon into a pair of measurable tracks
In LAT, a gamma-ray photon can produce an electron and a positron in the field of a nucleus. The positron has the same mass as the electron and the opposite electric charge. “Can” matters: not every photon interacts in the right place and passes the subsequent selection.
A thin tungsten foil increases the probability of conversion. Behind it sits a silicon strip detector: a charged particle crosses the silicon and creates electric charge carriers, while electronics record which strips fired. A sequence of coordinates at different heights allows a track to be found. The neutral gamma-ray photon itself leaves no such continuous ionisation trail before conversion.
A single strip plane primarily gives one coordinate. Two perpendicular orientations measure x and y. Each LAT module has 18 such measurement pairs of planes, meaning 36 silicon layers rather than 18 individual silicon sheets. The first 12 positions have thin converters, the next four have thicker ones, and the final two have no tungsten converter. They continue measuring the track before the calorimeter. LAT team: §2.2.1 and Table 2.
Branch · With equationsOptional · you can skip this
Why is 1.022 MeV insufficient without a momentum condition?
Two rest masses require at least approximately 1.022 MeV, but accounting for energy alone is insufficient. A free photon in vacuum cannot turn into this pair by itself while conserving energy and momentum: its invariant mass is zero, while a system of two massive particles has a positive invariant mass. In matter, the nucleus participates in momentum balance and recoils.
Let a nucleus of mass M initially be stationary and remain the same nucleus after the event. In units with c=1, the initial system’s invariant is M²+2MEγ. At threshold, the minimum invariant mass of the final system is M+2mₑ. Thus:
(M)^(2) + 2ME_(thr) = (M + 2m_(e))^(2),; E_(thr) = 2m_(e)(1 + (m_(e))/(M)).
Here masses are expressed as rest energies in MeV; mₑ=0.51099895069 MeV. For a heavy nucleus the correction is small, but necessary in principle. The roughly 20 MeV lower boundary of LAT’s overview energy range is set by the whole instrument and its selection, rather than the pair-production threshold. PDG: kinematics, §49.1–49.2, NIST: CODATA 2022.
Build a tracker without spoiling direction with excess material
Thicker tungsten might seem always better: conversion happens more often. Yet the electron and positron undergo multiple scattering in the material. If the first measurement comes after a substantial deflection, the track retains less information about its initial direction. Active silicon is therefore placed close to the converter, while passive supports are kept light. LAT’s thin front and thicker back sections balance the number of useful events against the width of their angular distribution.
LAT’s technical description gives a front converter thickness of 0.03 radiation lengths X₀ and a back thickness of 0.18 X₀. A radiation length characterises the interaction of high-energy particles with a material; it is not the geometrical length of a ray inside the telescope. These fractions allow the amount of material and its shower effects to be compared. They do not mean all photons are absorbed at a sharp depth. LAT team: §2.2.1, Table 2.
Assembly starts with tested silicon sensors. For LAT, 400 μm thick sensors were bonded four at a time edge to edge with epoxy, with corresponding strips joined by fine wire bonds. This creates a long sensor ladder; four ladders form a layer. The strip centre-to-centre pitch is 228 μm. That is a dimension of the electrode structure, not the telescope’s finished angular resolution.
Sensors and foils are mounted on lightweight composite trays: carbon face sheets, a vented aluminium honeycomb core and carbon sidewalls. A module has 19 trays, with adjacent trays rotated by 90°. An x/y measurement pair is formed between adjoining trays, so “18 pairs of planes” and “19 trays” are consistent. Side-mounted electronics and flexible cables take signals out without filling the entire entrance with heavy circuit boards. Channels and relative coordinates must be checked after each stage: a mechanically intact stack does not by itself guarantee a correct track. Fabrication and layout, §2.2.1.
The calorimeter measures energy through light in crystals
After the tracker, particles enter the calorimeter. An electromagnetic shower develops in the material: bremsstrahlung and subsequent interactions distribute energy among many particles. In a scintillator, some deposited energy becomes a brief light signal. Photodiodes turn the light into electric charge; amplifiers and digitisation produce numbers for reconstruction.
Each LAT calorimeter module contains 96 optically isolated CsI(Tl) crystals, caesium iodide doped with thallium. Eight layers of 12 long crystals alternate in orientation by 90° to reveal the shower’s spatial development. PIN photodiodes at both ends of each crystal measure light; the difference between their signals helps estimate where along the crystal energy was deposited. This readout differs from the silicon coordinate strips above. NASA: LAT subsystems, the primary calorimeter description, §2.2.2.
Crystals are checked for light response, optically separated, fitted with photodiodes and assembled into modules. Conversion factors between signal and energy, and dependence on particle position, are then needed. Arbitrary voltages cannot simply be added and called MeV. Figure 7 of the technical paper shows the LAT team’s measurement of light asymmetry using muons: actual calibration data, rather than an assumption that both ends always see equal light.
Even the calorimeter does not retain all the energy of every shower. Some can escape through the sides or bottom; the algorithm uses shower shape to correct for energy leakage. “Energy deposited in the crystals” and “estimated initial photon energy” are therefore different quantities. Their relationship is checked with known beams and response models, while uncertainty remains part of the result. LAT team: §2.2.2, §2.5.1 and Figs. 7–8.
Anticoincidence separates an entering particle from a useful event
A cosmic proton or electron also leaves a track and deposits energy. LAT surrounds its tracker with 89 plastic scintillator tiles forming the anticoincidence detector, ACD. A charged particle crossing this shell on entry produces light. Fibres collect the light and carry it to photomultiplier tubes. The event is matched in time and position with the internal track; an entering neutral photon is expected to have a different pattern. This is a selection clue rather than an infallible particle identity card. NASA: ACD.
There is a trap: a genuine gamma-ray shower can send secondary radiation back upwards, called backsplash. It can produce an ACD signal and incorrectly veto a useful event. The shell was therefore segmented into tiles, and analysis considers the region near the extrapolated incoming track. “Any tile fired, reject everything” would lose some high-energy signal.
Tiles overlap in one direction, while scintillating fibre ribbons cover gaps in the other. This addresses actual seams without requiring a solid heavy armour. Final selection also uses track quality, the conversion vertex and shower shape. Some gamma-ray background cannot be distinguished by the ACD simply through electric charge, such as photons produced by particle interactions nearby. LAT team: §2.2.3 and §2.4.
- Conversion
- γ → W → pair → shower; the entrance ACD may remain quiet.
- Charged particle
- Entrance ACD plus a matched internal track indicate background.
- Backsplash
- Shower → ACD; matching the tile and track helps retain the γ.
Branch · Geometry and momentumOptional · you can skip this
Why not simply average the two track angles?
Track directions must be weighted by their momenta, rather than simply averaging two angles. For a particle with total energy E, including its rest energy, in units with c=1:
p=\sqrt{E^2-m_e^2},\qquad \mathbf P=\mathbf p_++\mathbf p_-.
If nuclear recoil could be neglected, the direction of the sum P would approximate the photon direction. The full balance actually includes nuclear momentum q:
E_\gamma\hat{\mathbf z}=\mathbf P+\mathbf q.
Our figure uses total energies of 40 and 60 MeV and angles of +10° and −5°. Their average angle is 2.5°, while the momentum sum points at about 0.992°. The exact incoming direction in the constructed event is 0°; nonzero recoil of the heavy nucleus accounts for the difference. With a chosen nuclear rest energy of 170000 MeV, the required photon energy is 100.000010745 MeV, rather than exactly the 100 MeV sum. The complete balance is checked for all three final bodies.
This is our ideal kinematic event with deliberately visible angles, rather than a typical LAT record or a simulation of its accuracy. It omits multiple scattering, missing strips, radiation losses and sensor errors, which further complicate actual reconstruction. LAT also has no magnetic spectrometer measuring each track’s charge sign from curvature in this figure. The e⁺/e⁻ labels assign roles within the model. PDG: energy and momentum, LAT: tracker limitations.
Modules must work as one measuring instrument
LAT has 16 tracker–calorimeter pairs in a 4×4 array on an aluminium support. The ACD covers the top and sides, while electronics combine timing signals. Heat is carried to radiators: a stable measurement requires working sensors, power, thermal paths and known geometry together. The support does not focus gamma rays, but deformation can spoil the matching of coordinates. LAT team: §2.2.
A trigger decides when to read out the instrument: for example, coordinated hits in successive planes indicate a passing particle. Onboard processing rejects much of the background, while more detailed ground processing combines tracks, energies and anticoincidence signals. The output is an event record with estimates of direction, energy, time and uncertainties. Celestial coordinates also require spacecraft orientation and time references. This marks the boundary between detector and space platform: no single silicon layer knows where a pulsar is in the sky. DAQ and data processing, §2.2.4 and §2.6.2.
Checks are separated by the errors they reveal. Channel tests identify noise or a dead strip; passing muons check coordinates and efficiency; a known beam checks energy, showers and selection; repetition after environmental stresses looks for changes during integration. At CERN in 2006, the team tested a calibration unit comprising two tracker and three calorimeter modules, plus some ACD tiles. This does not mean the entire flight LAT was placed in an accelerator beam. During assembly and environmental testing, the team also recorded cosmic rays and used muon samples for checks. These are the mission team’s documented tests; we have performed no laboratory measurement ourselves. LAT: §2.5.1.
COMPTEL obtained a cone of directions from two interactions
At lower energies, Compton scattering can be used. A photon transfers some energy to an electron, then travels on in a different direction. Measuring the first and second interaction positions, along with both energy deposits, allows the photon’s deflection angle to be calculated.
COMPTEL had two detector assemblies separated by 1.5 m: seven modules of NE 213A liquid scintillator above and 14 NaI(Tl) crystals below. In the desired event type, the photon first scatters above and is then absorbed below. Photomultipliers read the light; ratios of their signals help locate the interactions. Plastic anticoincidence shields, time of flight and pulse shape reject some unwanted events. This is more than “LAT with less silicon”: different geometry and materials measure a different mechanism. NASA: COMPTEL technical description, §III.B.
The two assemblies need a fixed separation and known relative positions. COMPTEL’s lower crystals were mounted on a support plate, with most electronics on a separate platform outside the detector assembly. Two small scintillation detectors containing weak cobalt-60 sources sat between the assemblies for in-flight calibration. Their own signals tagged calibration events; ordinary celestial-event selection required these signals to be absent. A known check was thus separated from the desired cosmic radiation, rather than being added to the sky image. COMPTEL: layout, calibration and selection, §III.B.1–2.
The line between two interactions gives the direction of the already scattered photon. The initial direction can lie at the calculated angle anywhere around this axis in space. For directions towards the source in the sky, the axis must be reversed, from the lower absorber towards the first interaction. This produces a cone of possible directions and an event circle on the celestial sphere. Many circles from one source intersect in a consistent region. Actual errors turn circles into bands, while background and multiple sources require statistical analysis. One ideal circle is not an already located point.
Branch · With equationsOptional · you can skip this
How do two energies specify a 60° angle?
Let E₁ be the energy transferred in the first scatter and E₂ the scattered photon’s energy, fully absorbed in the second detector. For a free electron initially at rest:
E_(γ) = E_(1) + E_(2),; cosθ = 1 − m_(e)(c)^(2)((1)/(E_(2)) − (1)/(E_(1) + E_(2))).
All energies must use the same units, here MeV; mₑc²≈0.511 MeV. For a teaching photon of 1 MeV turning by 60°, E₁≈0.49456 MeV and E₂≈0.50544 MeV. Substitution returns cos θ=0.5. Energy is conserved, and the electron’s recoil momentum satisfies its relativistic mass–energy–momentum relation.
If the second photon is not fully absorbed, the interaction order is mistaken or the event is an accidental coincidence, this angle may be wrong or physically impossible. Bound-electron motion and detector resolution also broaden the ideal circle. The equation explains the mechanism; it does not replace COMPTEL calibration. NIST: the Compton formula, §3, Eq. 1, COMPTEL: measured quantities.
A coded mask records direction as a shadow
Another method does not require two successive interactions of one photon to measure its angle. A mask with opaque and open regions is placed above a detector plane. A source in one direction casts a particular shadow pattern; another direction shifts the pattern. Many counts are used to compare the measured distribution with those expected for different source positions.
SPI on INTEGRAL had an original plane of 19 germanium detectors and a mask about 1.7 m above it. High-purity germanium supplies spectral information; it is kept at cryogenic temperatures to reduce leakage current and noise. A single photon can have a well-measured energy, but the identity of the detector that fired does not by itself specify where that photon came from. Direction is encoded in a statistical shadow from many photons. The number 19 describes the original construction, rather than making a claim about the present number of working detectors. ESA: SPI, SPI Observer Manual 2022, pp. 8–12.
SPI’s germanium crystals are mounted in aluminium capsules on a cold beryllium plate inside a cryostat. Preamplifiers, amplification and pulse-height measurements turn the pulse into an energy channel, while electronics add timing. The mask has a Nomex honeycomb support, skins and a titanium ring; its 127 hexagonal cells contain 63 tungsten elements 3 cm thick and 64 largely open regions. The support also absorbs light, so “open” does not mean exactly 100% transmission at every energy. SPI’s active side and rear shields use BGO scintillator, rather than simply copying LAT’s plastic tiles. Their signals help reject background. SPI construction and electronics, §2.
Assembly requires the relative positions of mask and detectors, thermal control, working energy channels and a known directional response. A warped mask or incorrect geometry can shift a reconstructed source; an incorrect energy scale spoils a spectral line. These are different defects, so checking that “the counter clicks” is insufficient.
Branch · Geometry and calculationOptional · you can skip this
How can seven cells find a direction in a teaching model?
For parallel rays in a plane section, the shadow displacement depends on the mask–detector separation L and angle θ from the normal:
Δx = Ltanθ.
Our example uses L=10 cm, a cell pitch of 1 cm and a shadow shifted by two cells: Δx=2 cm and θ≈11.310°. The angle describes photons propagating down and right; the direction towards the source is opposite. Drawn rays pass through open cells and stop at closed ones. The repeating pattern extends beyond the displayed detector edges, so no photon is artificially transferred from one edge to the other.
Take our cyclic pattern 1110100. With a uniform background of 3 and a source contribution of 10 per open cell, the expected counts, before random fluctuations, are 3, 3, 13, 13, 13, 3, 13. Subtract the known background, then compare the remainder with each trial shift: add where the trial mask is open and subtract where it is closed. The seven results are 0, 0, 40, 0, 0, 0, 0. The maximum at shift 2 recovers the correct angle.
This is an ideal one-dimensional cyclic model with known background and no random fluctuations, partial coding or energy-dependent transmission. Its seven cells are not SPI’s actual hexagonal pattern. Real instruments combine shadows from multiple sources, changing backgrounds and unequal detector responses. Reconstruction must account for these; an attractive peak in our model does not establish the same mission accuracy. ESA: shadows and reconstruction, §2.6.
Single-event accuracy differs from source localisation
In LAT, even photons from an ideal point source reconstruct to positions distributed around its location. This distribution is described by the point spread function, PSF. Its width depends on energy, incidence direction, conversion location and selected event class. An individual photon’s direction estimate does not become exact because a map draws it as a tiny dot. LAT: tracker and PSF, §2.2.1 and §2.5.
Source localisation uses many events together: their distribution, energies, background, exposure and response model. A source’s centre can sometimes be determined far more precisely than a single event’s direction. Yet an arbitrary PSF width cannot simply be divided by the square root of photon count without conditions: uneven background, nearby sources, PSF tails and systematic errors change the problem. COMPTEL similarly combines many circles, while SPI combines shadows and different pointings.
Read a gamma-ray map by asking: what energy range and time interval does it cover, how were background and exposure treated, and what is the positional uncertainty? Colour can encode intensity or energy on a chosen scale; it is not a colour an eye would see. A spectral line and continuum provide different information from a pulsar’s or burst’s light curve. One instrument produces several complementary kinds of data. NASA: the gamma-ray sky, COMPTEL: event circles.
Direction survives in traces; the image must be reconstructed
The original list of detector hits now has meaning. In LAT, tungsten initiates conversion, silicon measures track coordinates, CsI(Tl) supplies the shower’s energy and spatial response, the ACD helps reject charged particles, and electronics and processing combine the measurements. In COMPTEL, two interactions define a circle of possible directions; in SPI, a mask encodes direction in a shadow. Calibration and a background model turn these traces into a testable astronomical conclusion.
For comparison, return to an X-ray telescope with grazing-incidence mirrors: there, optical surfaces create much of the required geometry; here, interactions and reconstruction do the work. A further variation is how a ground-based gamma-ray observatory makes part of the atmosphere into its detector.