How can we study a gamma ray that never reaches the ground?
A large telescope in Namibia looks into the night sky. Its mirror gathers visible and ultraviolet light, yet researchers call it a gamma-ray telescope. Is that a contradiction? Cosmic gamma rays usually interact with the atmosphere long before they could reach that mirror.
A ground-based gamma-ray observatory measures secondary signals from an atmospheric shower: a brief flash of Cherenkov light or particles that reach ground-level detectors. The geometry, timing and strength of these signals allow researchers to reconstruct the primary event’s direction and energy and separate it statistically from background. The atmosphere becomes part of a huge detector rather than just an obstruction in front of its optics.
We will follow a cosmic photon into an event in H.E.S.S., compare that route with the HAWC water array and examine the status of CTAO separately. A useful contrast is the previous lesson on space gamma-ray detectors, which try to measure the photon itself before an atmospheric shower develops. You can skip the visible mathematical branches: the main explanation remains complete. H.E.S.S.: technique and telescopes.
First came a flash; identifying a source took longer
In 1953, William Galbraith and John Jelley published a report of short pulses of light from the night sky correlated with cosmic rays. They used a photomultiplier, a sensitive light detector. The date refers to publication; it does not mean that a gamma-ray image of a particular nebula had already been obtained. The essential opportunity was that an atmospheric event left an optical trace accessible to an instrument on the ground. Original Nature publication.
The next difficulty was background: most flashes are produced by charged cosmic particles rather than the gamma rays being sought. Counting flashes alone is insufficient. A camera with many light-sensitive channels made it possible to measure the shape and orientation of a shower image. In 1989, the team operating the 10-metre Whipple telescope published a convincing detection of the Crab Nebula in TeV gamma rays, using a 37-pixel camera and image selection. This was a statistical comparison with background, not infallible identification of every photon. Weekes et al., ApJ 342, 379, 1989.
Systems such as H.E.S.S. added another capability: several telescopes observe the same atmospheric event from different positions. An elongated object that is hard to reconstruct from one view becomes easier with two or more. The original four H.E.S.S. telescopes formed a square of side 120 m; a larger telescope was later added at the centre. “Stereoscopic” means different geometrical views of one shower. This is a different task from combining wave phases in a radio interferometer. H.E.S.S.: array layout.
The primary gamma ray and the camera’s light are different signals
An energetic gamma ray can produce an electron and a positron in the atmosphere. These particles emit energetic photons through bremsstrahlung in matter; new photons can in turn create pairs. An electromagnetic shower develops: many interactions and secondary particles replace the single primary photon. Its axis approximately retains information about the original direction, while scattering and fluctuations in its development add uncertainty.
Charged shower particles can travel faster than the phase velocity of light in air and emit Cherenkov light. They still travel slower than c, the speed of light in vacuum. No relativistic speed limit is broken. The neutral primary gamma ray is not itself the charged particle leaving this Cherenkov trace. PDG: Cherenkov radiation, §34.7.1.
A real shower emits at different heights and positions. It produces a light pool on the ground, within which a telescope can collect part of the short flash. The H.E.S.S. overview gives a representative diameter of about 250 m and a duration of a few nanoseconds; these are not fixed dimensions of every event. One nanosecond is a billionth of a second. A large mirror is needed to capture enough of the sparse optical photons during this short interval. H.E.S.S.: shower and light pool.
Branch · With equationsOptional · safe to skip
Threshold, angle and a ring in an ideal model
Write the particle speed as v=βc and the refractive index as n. In a uniform medium at a chosen wavelength, the emission condition and the photon angle θ relative to the particle’s direction are:
βn > 1, cosθ = (1)/(βn).
The particle can exceed c/n, the wave’s phase velocity in the medium. In real material, n depends on wavelength; the speed of a wave packet and the geometry of the wavefront require separate treatment of dispersion. Our model takes n as constant and does not calculate all those effects. PDG: condition, angle and limits, equations 34.41–34.42.
For an electron with total energy E, including its rest energy mₑc²≈0.511 MeV:
β = √(1 − (((m_(e)(c)^(2))/(E)))^(2)),; K_(thr) = m_(e)(c)^(2)((1)/(√(1 − 1/(n)^(2))) − 1).
K is kinetic, not total, energy. Choose illustrative values n=1.0001 for an air-like medium and n=1.33 for water. These are not measured profiles of a particular observatory. The electron kinetic thresholds are approximately 35.625 MeV and 0.264 MeV, respectively. For an electron of total energy 1000 MeV, the angles are about 0.810° and 41.247°. A secondary electron’s low threshold in water is not the primary gamma-ray energy threshold of the entire HAWC observatory.
Imagine a vertical straight track, one emission height H and flat ground. A cone of fixed angle intersects the ground in a ring with radius R=H tan θ. For H=10000 m and our first n, R≈141.33 m. This is not a calculation of a uniformly filled light pool: that is produced by a developed shower with a distribution of emission heights, directions and particles. The figure compresses the vertical scale relative to the horizontal, so its drawn slope cannot be measured as the actual 0.810° angle.
The mirror collects an optical flash rather than focusing a gamma ray
Optically, this is a familiar reflector: the mirror collects secondary visible and ultraviolet light and forms an image on a camera. Its target, however, is a very short, faint image of an extended atmospheric shower, rather than a long exposure of a remote galaxy. The design therefore needs a large collecting area, a suitable field of view and a fast camera together. Such instruments are called imaging atmospheric Cherenkov telescopes, or IACTs.
In the primary description of the 12-metre H.E.S.S. I telescopes, the reflector consists of 380 circular panels of diameter 60 cm, each a small spherical mirror with nominal focal length 15 m. The panels follow a Davies–Cotton layout: their positions lie on a spherical support surface, while their orientations bring light to the camera. This does not mean that all panels are simply pieces cut from one giant spherical mirror surface.
There are 382 mounting positions, but two were reserved for the optical guide; this explains differing numbers in summaries. Around 107 m² of geometrical mirror area is also not the area of atmosphere from which gamma-ray events can be registered. Camera and support shadows reduce the useful optical area, while the effective gamma-ray detection area additionally depends on atmosphere, energy, geometry, triggering and selection. The lesson on signal and resolution explains the distinction. Primary H.E.S.S. description, §2 and §6.
Atmospheric telescopes do not all have this optical design. The larger H.E.S.S. II approximates a paraboloid to reduce the spread of light arrival times. The label “gamma-ray telescope” does not specify a single mirror shape or detector. H.E.S.S.: optical differences between the two types.
Test a panel before it becomes part of the dish
H.E.S.S. I uses worked glass panels with reflective aluminium and a protective quartz coating. Glass establishes the shape, a thin metal layer reflects the relevant light, and the protective layer helps the surface survive outdoors. The initial specification called for glass about 15 mm thick. This is not a chemical recipe for applying coatings: the published account explains selection and acceptance, not every factory operation. A separate lesson covers mirror shape, coating and tests.
Before installation, each panel was tested for spot size and specular reflectivity. On a bench near twice the focal length, about 30 m, light returned to the source region, where a photodiode scanned it. Filters allowed the response at different wavelengths to be compared. Independent monitoring of the source helped distinguish changes in lamp brightness from changes in a panel. The diameter enclosing 80% of the collected light measures concentration; it is not the reflectivity itself.
The initial specification required at least 80% specular reflectivity from 300–600 nm. The primary paper also reports panels that failed this requirement, especially near 300 nm, and changes to acceptance criteria. An original requirement must not become a claim that every delivered panel automatically met it. Bernlöhr et al.: materials, test bench and actual results, §3.
The panel is then installed on a three-point support. In this design, one point is fixed and the other two are associated with actuators. Sliding and articulated connections allow necessary movements: rigidly clamping everything could deform the optical surface through differential thermal expansion of glass and support. Actuators change the tilt but do not themselves know where the reflected light lands.
After installation, each panel produces a separate star image on the closed camera lid. An auxiliary CCD camera sees these spots, and the actuators bring them towards the target position. The alignment camera and the fast scientific camera have different jobs. Image quality and stability are then checked at different structural orientations. The reported thermal, humidity and optical tests were performed by the instrument team; no physical testing of our own is claimed here. Supports, assembly and checks, §5, H.E.S.S. alignment.
The camera must keep up with a flash lasting nanoseconds
The H.E.S.S. I camera uses photomultiplier tubes, PMTs. A photon can release an electron from a photocathode; successive dynodes multiply the charge into an electrical pulse. Light concentrators ahead of the PMTs help direct light onto the sensitive area and exclude unwanted directions. A “pixel” here is one such channel covering a particular angular field, not a pixel of an ordinary CMOS camera.
The camera can be assembled in modules. H.E.S.S. I has 960 PMTs in 60 drawers of 16 channels. A module combines photodetectors, readout and control so that it can be tested and serviced separately. Heat, power, joints, cables and channel reproducibility become part of the eventual image quality. A large pixel count alone does not ensure correct signals. H.E.S.S.: camera and modules.
During the 2015–2016 upgrade, the cameras received NECTAr-based readout. Analogue memory rapidly stores the recent signal history; after a trigger, the required segment is digitised. Sampling at 1 GHz means approximately one sample per nanosecond. This differs from the operating frequency of the ADC that subsequently reads the stored data and does not automatically give every photon a 1 ns timing accuracy. The described readout example takes a short window of 16 samples rather than the full memory. Electronics design, §2, completed upgrade and commissioning, §1–3.
Silicon photomultipliers, SiPMs, provide another option: arrays of microcells in which an avalanche discharge generates a measurable pulse. This is a different photodetector, not a new kind of primary gamma-ray interaction in the camera. FACT used 1440 such detectors. Their response depends on temperature and voltage, so a feedback system uses temperature sensors and current measurements to adjust the supply. Stability was checked through dark-count spectra. Assuming “silicon is always more stable” cannot replace these tests. FACT: construction and feedback, §I–II.
A trigger assembles a coincident event, not a long exposure
The night sky continuously sends optical photons into a camera. A short, consistent group of signals must be distinguished from random firings. A local trigger checks camera channels; the central array trigger compares signals from different telescopes.
The 2017 technical account of the upgraded H.E.S.S. cameras describes three levels: a single-pixel threshold, a condition on several pixels in a sector and coincidence between telescopes. Stereo events required at least two telescopes within an 80 ns window after accounting for delays. 80 ns is a parameter of that documented mode, not a universal duration of a Cherenkov flash. The larger CT5 could also record separate monoscopic events, so “every H.E.S.S. event must involve two telescopes” would be incorrect.
Signals travel along different electronic and optical paths to the central node. The team measured delays, including through fibre round-trip test signals. In addition to cable delays, alignment must account for observation geometry. Comparing raw timestamps alone can lose one real event or assemble a false coincidence. Performed calibration and trigger logic, §4–5.
- Pixel threshold
- ≥3 pixels / sector
- Camera trigger
- Delays → ≥2 telescopes
- Readout · 16 ns
| Channel | Time t · ns | Delay d · ns | t − d · ns |
|---|---|---|---|
| T1 | 100 | 0 | 100 |
| T2 | 220 | 120 | 100 |
Before correction: Δt = 120 ns · outside window.
After correction: Δt = 0 ns ≤ 80 ns · coincidence.
Branch · CalculationOptional · safe to skip
Timestamps of 100 and 220 ns can describe a coincidence
Take our own ideal example: signals reach the central node with timestamps of 100 and 220 ns, and the known path delays are 0 and 120 ns. These are arbitrary teaching values, not measurements of H.E.S.S. cables. Subtract the delays:
\begin{gathered} t'_i=t_i-d_i,\\ \Delta t=\max_i t'_i-\min_i t'_i. \end{gathered}
Both corrected timestamps are 100 ns, so Δt=0. Without correction the difference would be 120 ns. Under our teaching rule Δt≤80 ns, the first case passes and the second does not. A real system must account for resolution, calibration and shower-front shape; the particular electronics defines behaviour exactly at a window boundary. Coincidence improves reliability and supplies viewpoints, but it does not itself prove a gamma-ray origin.
Two images constrain one spatial axis
An array’s telescopes see different images of the same shower: its position, length and brightness in each camera differ. Each elongated image approximately defines an observation plane that can contain the shower axis. The intersection of two such planes constrains the axis in space. Researchers can then estimate where it intersects the ground: the shower core.
The celestial source, shower axis and ground location are different concepts. The axis direction points back towards a possible source; the core location helps establish how far each telescope was from the shower. Blurry shapes, unfavourable viewpoints or images clipped by a camera edge worsen reconstruction. More telescopes provide further constraints but do not remove every physical fluctuation.
Statistical images of gamma-ray and hadronic showers often differ, yet the geometry of an elongated image applies to both. Our diagram shows an exact projection of our own ideal straight axis. Its construction lines do not mean that every chosen point emits a Cherenkov photon directly towards both telescopes. It models geometrical inference, not the physical visibility of an entire shower. VERITAS: atmospheric-image geometry, H.E.S.S.: stereoscopy.
Branch · GeometryOptional · safe to skip
From camera coordinates to intersecting planes
Our ideal telescopes stand at T₁=(−60,0,0) m and T₂=(60,0,0) m with aligned coordinate axes. The axis intersects the ground at C=(0,40,0) m. At height z, its coordinates follow x=0.003z and y=40−0.002z. For checking, take three points at heights 6000, 10000 and 14000 m. These are specified teaching parameters.
After accounting for camera orientation, directions can be written in gnomonic coordinates:
u = (x − T_(x))/(z), v = (y − T_(y))/(z).
The figure’s axes show 1000u and 1000v, a gnomonic scale in milliradians, not physical sensor millimetres. At small angles it approximates an angular scale, but exact conversion requires an arctangent. A spatial straight line projects to a straight line on this plane. Its direction and position give a normal Nᵢ to the plane passing through the corresponding telescope:
\begin{gathered} \mathbf N_i\cdot(\mathbf r-\mathbf T_i)=0,\\ \hat{\mathbf s}\parallel\mathbf N_1\times\mathbf N_2. \end{gathered}
The cross product gives the direction of the intersection line. Choose its sign towards the sky; intersection with the ground returns the core (0,40,0) m, while the celestial gnomonic coordinate is (3,−2) in our displayed scale. A separate check confirms that all spatial points lie in both reconstructed planes.
If the planes nearly coincide, the direction is poorly constrained; exactly coincident planes have no unique solution. Real images require uncertainty estimates, weights, multiple viewpoints and detector response. Our ideal line omits shower scattering, width and brightness and is not a measured H.E.S.S. PSF.
Electronics, optics and air need different checks
A weaker electrical pulse could mean less light, a different PMT gain, a dirty mirror or a less transparent atmosphere. Without separate checks, those causes merge into a misleading change in the source energy.
The upgraded H.E.S.S. cameras required calibration of the baseline of each analogue memory cell, the short readout-window position and PMT gain. Known weak flashes allowed the pedestal peak to be distinguished from the one-photoelectron and subsequent peaks. Another system of uniform illumination, flat-fielding, compared relative channel responses, including the photocathode and light concentrator. Equal electronic gain does not guarantee equal collected light. Actual camera procedures, §2–3.
Star-image shape checks optics and alignment. Ring images of cosmic muons can also monitor an instrument’s optical throughput. Meanwhile, transparency along a multi-kilometre atmospheric path needs its own monitoring: clouds and aerosols can absorb or scatter light before it reaches a mirror. Observatories use or plan cloud monitors, radiometers, lidar measurements, stellar transparency measurements and weather data; the particular combination depends on the observatory. Electronics calibration does not remove a cloud above the telescope. Muon rings and optical calibration, H.E.S.S.: atmospheric-monitoring tasks.
Branch · Measurement chainOptional · safe to skip
Why is an ADC number not yet a gamma-ray energy?
Let Q be an integrated signal in digital ADC units, P the corresponding baseline and G the calibrated number of these units per photoelectron. A first charge estimate in photoelectron units is:
N_(pe) ≈ (Q − P)/(G).
Our example Q=1800, P=1000, G=80 gives 10 photoelectrons. That is not 10 primary gamma rays, 10 MeV or even the exact number of optical photons incident on the mirror: some were lost by the optics and some caused no photoelectric conversion.
After correcting channel differences and accounting for atmosphere and image geometry, shower parameters can be estimated. The relationship to primary gamma-ray energy requires calibrated response and modelling. This whole chain therefore cannot be replaced with one universal “ADC → TeV” conversion factor. From gain to flat-field, §3.
A hadronic shower can trigger several cameras too
Charged cosmic rays, including protons and nuclei, generate their own atmospheric showers. They also produce Cherenkov light, stereoscopic images and consistent timestamps. There is no external anticoincidence shell wrapped around kilometres of air.
Hadronic showers on average have different lateral structure and more penetrating particles, including muons, than electromagnetic showers initiated by gamma rays. Analysis uses shape, orientation, consistency between views and other features. “On average” matters: distributions overlap, and a rare fluctuation can resemble a useful event. Neither a compact ellipse nor one successful trigger independently proves the origin. VERITAS: image differences and geometrical information.
An astronomical result emerges when a testable excess above the expected background has the relevant direction, spectrum or time behaviour. This requires knowledge of exposure, observing conditions and selection. Rejecting more background helps, but overly strict selection also removes real gamma-ray events. The task is to estimate response and uncertainty rather than attach an infallible label to every event.
HAWC measures particles at ground level; light is produced in the water
Cherenkov telescopes are not the only ground-based route. At HAWC, high in the mountains of Mexico near Sierra Negra, the main array was designed as 300 water Cherenkov detectors. It measures shower particles that reach the ground. Inside the water, charged particles emit Cherenkov light detected by photomultipliers. No large mirror collects the atmospheric optical flash in this arrangement.
The 2017 HAWC account describes steel tanks containing light-tight inner bladders of purified water. Three peripheral 8-inch PMTs and one central 10-inch PMT stand near the bottom, with their sensitive surfaces facing upwards. A closed tank measures internally generated light without illumination by the daytime sky. The water array can therefore observe during the day and operates differently from an optical telescope on a clear, dark night. That is not a promise of uninterrupted operation regardless of any malfunction or weather. HAWC: construction and detection, §1.
Assembly must ensure watertightness, water purity, optical isolation, PMT positions and a well-understood electrical response. Pulse-to-charge conversion and channel timing offsets are then calibrated. The described system used the duration above two thresholds, time over threshold or ToT, to estimate charge and sent timestamps to processing. Different cables and electronics behaviour must not become a false shower-front tilt. HAWC: readout and calibration, §1–2.2.
The signal distribution across the ground estimates the core; timing estimates direction. A plane front is a first approximation, but the actual front is curved and particles far from the core can arrive late. Corrections affect the reconstructed direction. An isolated large signal far from the core may indicate a penetrating particle or hadronic subshower; selection is statistical again. The number of triggered PMTs relates to event size but also depends on geometry and conditions, so it is not an exact universal gamma-ray energy measurement. HAWC: core, direction and background, §2.3–2.6.
The comparison is useful: an IACT images faint light from a shower above the ground; HAWC samples its ground-level particles and measures their interactions in water. Both infer the primary event from secondary signals, but they have different apertures, fields of view, operating modes and uncertainties.
CTAO: completed structures still need commissioning
A large array needs different telescopes for different energies: at lower energies the light is scarce and a large mirror is needed; at higher energies events are rare, making broad coverage by many instruments valuable. This helps explain the different telescope sizes in the Cherenkov Telescope Array Observatory, CTAO design. A design energy range and expected sensitivity must be distinguished from measured performance of a particular commissioned part.
Status checked on 14 November 2026. The official CTAO release of 15 October 2026 reported the inauguration of a subarray of four 23-metre LSTs on La Palma. The same document describes subsequent commissioning and integration into CTAO operations. The ceremony marking completion of these telescopes therefore cannot be presented as completed commissioning of the entire planned two-site observatory.
The earlier LST-1 had already produced scientific results during commissioning; there is no contradiction. An individual instrument can observe and test methods while the larger system is still being built and accepted. To assess a claim, always look for its date, the particular part of the array and whether it describes a design expectation, a test result or routine observing. CTAO: official dated status of the LST subarray.
The atmosphere hides the photon but preserves measurable traces
The initial contradiction is resolved. A H.E.S.S. mirror receives secondary optical light; a camera rapidly records the shower image; a trigger associates signals into an event; several viewpoints and calibration allow reconstruction of the primary direction and energy. At HAWC, particles reach the water and generate light inside ground-level detectors. In both methods, the atmosphere converts the signal.
A large dish therefore does not mean that the gamma ray fell onto an ordinary mirror, and a Cherenkov cone does not mean a particle exceeded c in vacuum. To compare these routes with direct detection in space, return to Fermi LAT, the Compton cone and the coded mask. The route to light has changed, while the need to separate signal, background and instrument response remains.