Why can a dark telescope be too bright?
At night, a telescope mirror looks dark. It gathers the faint light of a distant galaxy and seems to contribute nothing of its own. An infrared detector sees a different scene: the warm mirror, supports and housing emit radiation too. Their own light can overwhelm the signal we want, even though our eyes cannot see it.
An infrared telescope is cooled to reduce its own thermal emission and unwanted detector signals. The required temperature depends on wavelength and detector material: the mirror, housing and different detectors need not be equally cold. Atmospheric background, solar heating and heat from electronics each require their own solutions.
We will follow the signal from its source to the data in two examples: ESO’s ground-based VISIR instrument and Webb’s space-based MIRI. From the article on light and noise, we need the idea that background produces random counts too. From the article on materials, we need the idea that optical shape and coatings depend on operating conditions. Equations appear in visible optional branches.
From Herschel's thermometer to a sky survey
In 1800, William Herschel split sunlight with a prism and compared the heating of thermometers. A thermometer beyond the red edge of the visible band warmed up too. This revealed radiation invisible to the eye. It was not yet a sensitive camera capable of photographing distant nebulae. NASA: infrared waves.
Early infrared measurements relied on heating thermocouples and bolometers. A bolometer turns absorbed energy into a temperature change that can be measured electrically. Early instruments were mainly suitable for bright targets. Photoconductive detectors and very cold bolometers later opened access to much fainter sources. One important step was Frank Low’s helium-cooled germanium bolometer in the early 1960s. Progress depended on more than aperture: it required detecting a small signal against the instrument’s own background. A participant’s retrospective: Okuda, 2019, §§1–2.1.
The next step was taking a sensitive cooled instrument beyond the atmosphere. IRAS, launched in January 1983, surveyed more than 96% of the sky before its mission ended in November, using four infrared bands: 12, 25, 60 and 100 μm. It demonstrated the value of a systematic infrared survey rather than observing only an individual star already known in visible light. Official IRAS Explanatory Supplement, I.A.
What are we looking for in infrared light?
Infrared names a range of electromagnetic wavelengths, rather than a single way of producing them. Near-infrared radiation can be starlight or sunlight reflected by a surface. Mid- and far-infrared observations are particularly useful for thermal emission from cooler bodies and dust. Spectra also contain molecular and atomic features that cannot be reduced to a smooth temperature curve.
Sub-band boundaries vary between fields. Here we adopt the explicitly conventional division in Okuda’s review: near infrared 0.8–5 μm, mid infrared 5–30 μm, far infrared 30–1000 μm. These are guides, not universal boundaries for every instrument. Webb covers near and mid infrared; it does not observe the whole far-infrared range.
A wavelength choice answers a particular question:
- What is hidden inside a dust cloud? At some infrared wavelengths, dust attenuates light less than in the visible. We can see some young stars behind a cloud; the cloud does not become transparent at every wavelength.
- Where is the cool material? Dust absorbs shorter-wavelength light and emits at longer wavelengths. Its infrared spectrum helps estimate temperature and composition within a model.
- Which gases and solid particles are present? MIRI distinguishes, for example, signatures of molecular hydrogen and silicate dust. This requires spectral measurements and comparisons, rather than guessing from an image’s colours.
- What were distant galaxies like? Expansion of the Universe makes light emitted at shorter wavelengths arrive at longer ones. Some ultraviolet and visible light originally emitted by ancient galaxies reaches us in the infrared.
NASA: infrared astronomy goals, MIRI spectra in Stephan’s Quintet.
More than target light reaches the detector
A ground observation begins before light reaches the primary mirror. The atmosphere absorbs some radiation, transmits some and emits its own. Light then reflects from optics, passes a filter and reaches the sensitive array. Along the way, emission from the warm telescope and other surfaces in the detector’s view joins the signal.
Cooling a camera solves only part of the problem. If a warm primary mirror and the atmosphere remain in front of it, their background still enters the image. The engineer therefore asks more than “how cold is the instrument?” The crucial question is “which surfaces can a sensitive pixel see in this band?” A cold camera looking at a warm housing edge may perform worse than a well-baffled design.
The VISIR manual gives a characteristic contrast: atmosphere around 253 K, telescope around 283 K, most cold instrument optics around 29 K and detectors around 9 K. These are values from a particular description, not our live telemetry. Kelvin temperature starts at absolute zero; 300 K is about 27 °C. ESO VISIR User Manual, §2.4 and §3.
Cooling changes brightness and the spectrum's position
A warm body emits across a broad range. As it cools, its total thermal emission weakens and the spectral peak moves to longer wavelengths. What matters to an instrument is how much unwanted light remains inside its own observing band. Comparing one surface at 10 μm with another at 100 μm and calling the ratio a “cooling effect” would be misleading.
Below is our model of four surfaces that differ only in temperature. They all have the same assumed, wavelength-independent emissivity ε=0.05, their ability to emit compared with an ideal blackbody. This greybody is a teaching approximation. It is not a measured Webb or VISIR coating.
| Temperature · K | Bλ maximum · μm | 8–13 μm integral · W m⁻² sr⁻¹ |
|---|---|---|
| 300 | 9.66 | 2.3549e+0 |
| 200 | 14.49 | 2.3594e-1 |
| 80 | 36.22 | 1.8481e-5 |
| 40 | 72.44 | 8.1023e-12 |
In the 8–13 μm band, cooling the model surface from 300 to 80 K reduces its band-integrated radiance from about 2.35 to 0.0000185 W m⁻² sr⁻¹, roughly a factor of 127 thousand. Area, field of view, transmission and detector response have not yet been included. This reduces one assumed surface’s contribution; it does not promise the same improvement in the complete telescope’s sensitivity.
Real emissivity depends on material, wavelength, angle and surface condition. A mirror coating with high reflectance in the relevant band helps pass useful light through the system and limit its own emission there. A visibly shiny surface need not behave the same way in the far infrared. NIST: radiance definitions and Planck’s law.
Branch · With equationsOptional calculation; the main idea that a cold surface produces less background in the same band is already explained above.
How the curves and fixed-band comparison were calculated
In vacuum, Planck’s law specifies blackbody spectral radiance per unit wavelength. Let λ be in metres, T in kelvins, h=6.62607015×10⁻³⁴ J s, c=299792458 m/s and k=1.380649×10⁻²³ J/K:
B_(λ)(λ,T) = (2h(c)^(2))/((λ)^(5))(1)/(exp[hc/(λkT)] − 1),; L_(λ) = εB_(λ).
In this form, Bλ has units W m⁻² sr⁻¹ m⁻¹: energy per second, projected area, solid angle and metre of spectral interval. For the graph per micrometre, multiply by 10⁻⁶. This is not power already collected by the aperture.
The peak of Bλ specifically, rather than another spectral quantity, satisfies:
λ_(max)T = b_(λ),; b_(λ) ≈ 2.89777 × (10)^( − 3) m K,; λ_(max)(300 K) ≈ 9.66 μm,; λ_(max)(80 K) ≈ 36.22 μm.
If we plot Bν per unit frequency, the peak differs: spectral intervals have different widths under a change of variable. We cannot simply take the frequency of the Bν maximum and call c/ν the Bλ maximum.
The comparison below the graph integrates the same band:
L_(8 − 13)(T) = ((∫)_(8 μm))^(13 μm)εB_(λ)(λ,T) dλ,; (L_(8 − 13)(300 K))/(L_(8 − 13)(80 K)) ≈ 1.274 × (10)^(5).
Here ε=0.05 for both surfaces, and the integral has units W m⁻² sr⁻¹. The σT⁴ law concerns the entire spectrum emitted into a hemisphere; it does not replace this band integral. To predict a particular telescope’s counts, we must also integrate over visible surfaces and directions, including transmission, filters and quantum efficiency. NIST: equations 2–3, printed p.2, CODATA constants.
VISIR: observing from Earth works, but not at every wavelength
The atmosphere is not equally opaque throughout the infrared. A dry high-altitude site offers usable transmission windows. At Paranal, VISIR works in windows including N, roughly 8–13 μm, and Q, roughly 17–25 μm. Absorption bands occur within these intervals; water vapour and the atmospheric path length are especially important. Weather and target elevation affect whether an observation is useful. ESO, §2.1, fig.1: atmospheric model conditions.
The telescope collects the light, while the attached VISIR instrument contains separate cold imaging and spectroscopy channels. In the imager, the input image feeds a collimator that forms the required beam. A cold stop in a pupil image plane restricts the view so the detector receives less unwanted radiation from outside the useful aperture. A filter selects the band and a reflective camera forms the image on the array. ESO, §3.1, fig.7.
A cryostat maintains a cold environment. Vacuum reduces heat transfer through gas; shields limit thermal radiation; supports and wires are chosen with thermal conduction in mind. Vacuum does not remove radiation, however, and wires and supports remain paths for incoming heat. A refrigeration stage must remove the residual load. VISIR uses closed-cycle cryocoolers; the whole building is not cooled to detector temperature.
The warm atmospheric background remains strong. The secondary mirror rapidly alternates directions on the target and beside it, called chopping. Subtracting the frames reduces their common background. Moving the telescope as well, called nodding, helps remove residuals associated with different optical paths. This is a carefully planned observing strategy; subtracting an arbitrary neighbouring photograph cannot replace it. ESO, §§2.4–2.5, fig.5.
Branch · Noise and statisticsOptional numerical example; the following space-cooling explanation does not require statistics.
Why a zero average background still makes noise
Let S be the expected number of detected source photons and B the background in one exposure. The on-target frame has mean S+B; an equal-duration empty-sky exposure has mean B. Assume independent Poisson counts and unit quantum yield, omitting drift, read noise and correlations.
\begin{aligned} D&=N_{\rm on}-N_{\rm off},\\ \mathbb E[D]&=S,\\ \mathrm{Var}(D)&=(S+B)+B=S+2B,\\ \mathrm{SNR}&=\frac{S}{\sqrt{S+2B}}. \end{aligned}
For S=100 and B=10000, the mean difference is indeed 100, but its standard deviation is about 142 and SNR about 0.71. Subtracting the mean background does not by itself make a weak source reliably measured. Averaging m independent, equal off exposures gives variance S+B+B/m; the background’s own fluctuations in the on frame remain.
This explains why reducing background before it reaches the detector helps even with good calibration. Real infrared instruments also have other noise sources, and electron yield is not always one per photon. STScI: photon noise and the approximation’s limits.
Webb: space removes the atmosphere, not solar heating
Beyond the atmosphere, its absorption and local thermal background disappear. Sunlight still heats the spacecraft, electronics produce heat and supports conduct it. The space around L2 does not automatically refrigerate a telescope: in vacuum there is no surrounding air to blow through radiator fins.
Webb keeps a five-layer sunshield between the warm illuminated side and cold optics. Separated coated membranes reflect part of the incoming energy and reradiate absorbed heat, including through open gaps at the edges. Thermal isolation reduces the flow into the cold side. Correct orientation is essential: the telescope cannot turn its exposed optics towards the Sun at will. NASA: sunshield construction.
Cryocooler: Heat transfer · cold stage → warm stage → outward radiation.
A radiator emits heat towards an available cold direction. A cryocooler transfers heat from a cold stage to another stage, from which it ultimately escapes. A shield mainly reduces incoming heat. These functions complement one another; they are not three names for the same component.
Webb’s near-infrared HgCdTe detectors can operate around 37 K with passive cooling. MIRI’s mid-infrared Si:As detectors require a nominal temperature ≤6.7 K. Passive cooling is insufficient here, so a dedicated helium cryocooler operates. A warm compressor assembly, precooling stages and a final cold stage are connected by tubing; heat is picked up near the instrument and carried away. Helium circulates in a closed loop. That does not imply unlimited mechanical or electronic lifetime. NASA: cryocooler, STScI: MIRI detectors.
Do not say “all of Webb is at 6.7 K.” MIRI’s engineering description separately treats an optical module around 7 K, its surrounding shield around 23 K, the ISIM environment around 40 K and warm electronics around 300 K. These are different nodes of a thermal design. Wright et al., 2015, §§V and VII.
Why not simply put an infrared filter in front of an ordinary camera?
A filter can select wavelengths, but cannot make an insensitive array detect them. An absorbed photon must produce a measurable change: heating a bolometer or generating mobile charge carriers in another detector type. Material and structure determine which photons produce a useful response.
HgCdTe combines mercury, cadmium and tellurium; changing its composition tunes spectral sensitivity. Such photodiodes serve Webb’s near-infrared instruments. Si:As in MIRI is arsenic-doped silicon with a specialised impurity-band conduction structure, IBC. This is a different mechanism suited to longer wavelengths at very low temperature. VISIR also uses Si:As IBC, but its particular detector and operating mode are not identical to MIRI’s. NASA: detector structure and composition, ESO, §3.4.
An array is assembled from different parts too. The sensitive semiconductor layer and silicon readout integrated circuit are manufactured separately; corresponding pixels are joined with indium microcontacts. This produces a hybrid array: one layer absorbs light, while the other reads the electrical response. It is then installed in a focal-plane module with mechanical mounting, a thermal link and protection against stray illumination. A detector is therefore more than a piece of cold silicon. NASA: detector architecture.
Heating can create mobile charges even without the desired light, producing dark current. Cooling reduces this unwanted contribution and helps stability. It does not arbitrarily change a material’s spectral properties or eliminate every pixel defect, read noise or memory of earlier illumination.
Branch · Detector physicsAn additional equation explains material choice; the manufacturing sequence below can be read without it.
Lower photon energy requires a different sensitive structure
A photon’s energy decreases as wavelength increases:
E = (hc)/(λ),; E(2 μm) ≈ 0.620 eV,; E(10 μm) ≈ 0.124 eV.
For an interband photodiode, the bandgap matters; an appropriate HgCdTe composition allows it to be tuned. Explaining every infrared detector through that gap alone would be wrong: Si:As uses impurity states, while a bolometer measures heating instead. Low signal energy also makes control of thermal excitation more demanding. Actual currents and efficiency must come from detector characterisation, rather than E=hc/λ alone. STScI: MIRI structure and limits.
Building a system that works after cooling
Components are warm in the workshop but will work in the cold. If materials contract differently, focus, mirror shape and pupil position change. A perfect room-temperature image is therefore insufficient. What matters is a predictable operating shape in the cold.
Webb’s mirrors use a lightweight beryllium substrate stable under the required conditions. After machining and polishing, its shape was measured cold, then corrected so that the proper surface would emerge at operating temperature. Vacuum deposition added a thin gold layer with a protective SiO₂ layer above it; coating required further verification. Gold was chosen for its relevant spectral properties. This does not mean “all infrared mirrors are always gold”: MIRI’s internal optics, for example, use a different material construction. NASA: making Webb’s mirrors.
In MIRI, similar aluminium alloys in the optical bench and optical components help them contract consistently. Precision reference surfaces, pins and interfaces locate modules; thermal ageing improves component stability. Composite supports limit heat arriving from the warmer structure. Separate conductive straps deliberately carry heat towards the cold head. “Every component should conduct heat poorly” would therefore be the wrong rule. MIRI team, §§II–V.
Integration has stages with verifiable outputs:
- Design the spectral and thermal paths. Define bands, surfaces visible to the detector, coatings, heat loads and required temperature stages.
- Manufacture and check modules separately. Measure optical shape, mount geometry, detector behaviour and material stability.
- Assemble the warm system with its cold state predicted. Check focus, pupil, mechanical interfaces, wiring and thermal links.
- Cool it in a vacuum test facility. Check temperatures, residual heat loads, cold focus and response to known sources. Look for contamination or unwanted edge emission.
- Compare before and after mechanical loading. For a space instrument, verify that vibrations and thermal cycles have not damaged alignment.
- Calibrate the output data. Measure dark signal, pixel nonuniformity, linearity and spectral response; retain conditions and uncertainties with calibration.
The MIRI engineering paper describes a 1600-hour test campaign on the flight optical module in 2011, with a simulated cold environment. This was the instrument team’s historical test. This article’s author has not built a cryostat or performed a physical test; the work here checks documentation and teaching calculations. Wright et al., §III.3 and IX.
What an infrared design solves
A telescope can interfere with an observation because its own warm surfaces emit in the same band as the target. A detector does not know in advance which photon came from a galaxy and which from a warm housing edge. Design therefore starts with the wavelength range and sensitive material, then limits unwanted views, reduces heat inflow, cools the necessary stages and calibrates the remaining signal.
Check your understanding: if VISIR’s ground-based array is cooled further, will the warm atmosphere’s background disappear? No: that background source remains in front of the instrument. What if we remove the atmosphere by flying into space? The atmospheric contribution disappears; solar heating, internal heat and dark current still require engineering solutions.
Ground infrared astronomy works in suitable windows with an appropriate strategy. Space infrared astronomy opens other bands and lower backgrounds, but needs shielding, heat rejection and sometimes active cryocooling. For more on the precise shape of large segmented mirrors, read the available article on giant mirrors and optical correction.