The brightness no lens can increase

A magnifying glass will set paper alight. A larger one will do it faster. The obvious extrapolation is that a large enough lens, well enough made, will reach any temperature wanted, and it is wrong — not by a factor, and not because of any defect in the lens. There is a ceiling, it is 5,770 kelvin for sunlight, and 5,772 kelvin is the temperature of the surface of the Sun.

The coincidence between the achievable temperature and the source temperature is not a coincidence, and following it back is the cleanest route into a quantity that governs every optical instrument and is almost never mentioned in an optics course.

The quantity that cannot be reduced #

A lens forms an image, and the usual account of what it is doing stops at where the image is and how big.

Three rays are enough to find where an image is, and the magnification follows from the geometry — but magnification is not the quantity this essay is about. Enlarging an image spreads the same light over more area, so what is gained in size is lost in brightness exactly. The conserved quantity is the product, and no arrangement of glass changes it.

Take a bundle of rays crossing an aperture of area $A$ and filling a cone of half-angle $\theta$. The product

$$G = n^2 A \sin^2\theta$$

is the étendue of the bundle, and in a lossless system it cannot decrease. It can be wasted — a badly designed instrument throws light away and the étendue of what survives goes up — but no arrangement of surfaces reduces it.

Dividing the flux by the étendue gives the radiance, and the statement in its most useful form is that no passive optical system can increase the radiance of a beam. That is a conservation law of the same kind as the ones this collection keeps returning to — a quantity that survives an operation and therefore constrains its outcome, in the way the flux through a closed surface constrains what is inside it without anybody having to look. An image is never brighter than the object. A telescope gathers more light from a star because a star is a point and the image is diffraction-limited; it does not make the surface of the Moon look brighter than the Moon looks, and no eyepiece will. The same statement rules out the reverse trick as well: a curved mirror that cannot focus fails to make a small image and therefore fails to concentrate, and correcting it recovers only what the geometry allowed in the first place.

Why that puts a ceiling on temperature #

Concentration follows immediately. If a receiver accepts light out to a half-angle $\theta$, the largest area ratio between what is collected and what is delivered is $n^2/\sin^2\theta$. For sunlight in air that is 46,165, and the number comes entirely from the Sun’s angular radius.

The agreement to three figures is the argument. The optical bound was derived by counting rays and contains no thermodynamics; the temperature it delivers is the source’s own. Had the two disagreed, the direction of the disagreement would have decided which one was wrong, because an absorber hotter than the Sun with the Sun in view is a machine that can be run to produce work from a single temperature.

An engine between two bodies at the same temperature does no work, and that is the statement doing the work here. If a lens could make an image brighter than its source, a black surface at the focus would reach a temperature above the source’s — and a heat engine between the two would run on nothing. The optical limit is the second law in disguise, which is why no cleverness of design can get round it.

There is a second and more direct way to see the same thing, and it explains where the n² comes from. A receiver that can accept radiation out to an angle θ can also emit into that angle, and equilibrium is reached when what it absorbs equals what it emits. That balance is the bound, written as a statement about detailed balance rather than about rays.

The distances that do not matter #

An instinct worth dismantling is that concentration ought to depend on how far away the source is.

The distances do not matter, and it is worth seeing why. The Sun’s flux at Mercury is six and a half times what it is here — but its angular size is larger by the same factor in each dimension, so its brightness per unit solid angle is identical. That ratio is the invariant: flux falls as the inverse square and apparent area falls as the inverse square, and the quotient does not fall at all.

Moving closer to the Sun buys flux and buys nothing in concentration, because it enlarges the angular size of the source and therefore the smallest image the concentrator can make. Light does carry momentum and does push, and the push does grow as the inverse square on the way in; the brightness does not. The maximum achievable temperature at Mercury is the same 5,770 K as it is at Neptune. Only the collector area required differs, and it differs by exactly the inverse square.

That indifference to distance is the signature of a radiance argument, and it is worth carrying as a diagnostic: a quantity that survives the inverse-square law is a quantity about the source, not about the arrangement.

The quarter that a focusing mirror gets #

A parabolic dish aimed at the Sun forms an image of it at the focus, and the image has an angular size, so it has a physical size, so the concentration is finite. Computing it from the geometry gives a number well below the ceiling.

The shortfall is not a tolerance and no figuring recovers it. The reason is that an imaging concentrator insists on more than it needs: it requires that every ray from a given point on the source arrive at a given point on the receiver, and that requirement is far stronger than merely requiring the rays to arrive somewhere on the receiver. Imaging is an expensive constraint, paid for in étendue.

Aberration is a real defect and it is not this one. A spherical mirror does not bring parallel rays to a point and a parabolic one does, so the parabola is genuinely better — and curing the aberration does not help here at all, because the limit is set by the source’s angular size rather than by the sharpness of the optics. A perfect mirror and a poor one have the same ceiling.

The shape that takes all of it #

Give up the image and the ceiling becomes reachable. The design rule is the edge-ray principle: make sure that a ray entering at the extreme acceptance angle just grazes out of the exit, and everything inside that angle looks after itself.

A ray entering inside the acceptance angle emerges somewhere on the exit and nobody can say where; a ray entering outside it is turned round and sent back out of the front. That is the whole behaviour, and it is enough. The subject that grew out of the observation — non-imaging optics — is now what solar concentrators, luminaires and the light-collecting cones in front of scintillation detectors are designed with, and its founding fact is that insisting on an image costs a factor of four.

Three instruments the bound explains #

The rule is worth exercising on cases where it settles an argument that intuition gets wrong.

A solar furnace at Odeillo in the French Pyrenees uses a field of 63 heliostats and a paraboloid 40 metres across, and reaches about 3,500 K. That is well under the ceiling and the reason is entirely the imaging loss above compounded by the finite accuracy of the heliostats, each of which adds its own angular spread to the Sun’s. Every source of angular error at the input enlarges the effective size of the source, and the ceiling falls as the square of it.

A fibre-coupled lamp cannot be made brighter by choosing a better lens. The étendue of the fibre is its core area times the square of its numerical aperture, and the fraction of a lamp’s output that can be got into it is the ratio of that to the lamp’s own — usually a per cent or two, and not improvable. Replacing the lamp with a laser diode of the same power changes the answer by four orders of magnitude, because the diode’s étendue is tiny.

And a solar cell under concentration gains efficiency, which sounds like something for nothing and is not. The gain comes from the cell running at a higher photon flux and therefore a higher internal voltage, and it is bounded by the same 46,165: a cell under maximum concentration has its open-circuit voltage raised by kT·ln(46,165), which is 280 millivolts at room temperature and is the whole of the prize.

The same conservation law, for particles #

The bound has been stated for light and the derivation used only ray counting, which should provoke the question of whether light was needed at all. It was not, and the version for material particles is the same theorem with different words — discovered independently, named differently, and constraining a different industry.

A beam of particles is described by the positions and transverse momenta of its members, which is a volume in phase space. Liouville’s theorem says that volume is conserved under any evolution generated by a Hamiltonian — which covers every arrangement of static or slowly varying electric and magnetic fields anybody can build. Magnets are the lenses, and they obey the same rule for the same reason.

Accelerator physicists call the conserved quantity the emittance, and every consequence in this essay reads across without alteration. A beam can be squeezed to a smaller spot only by making it more divergent, in exact proportion. The luminosity of a collider is limited by the emittance of its beams, not by how good the final focusing magnets are. And no sequence of magnets, however clever, reduces it.

The escape is the same escape too, and this is the part worth carrying. Étendue can be beaten by a source that is not passive; emittance can be beaten by a process that is not Hamiltonian — one that removes entropy from the beam and puts it somewhere else. Every such process is called cooling.

Electron storage rings cool themselves for free: the electrons radiate as they bend, losing momentum in every direction and having it restored only along the beam axis by the accelerating cavities, so the transverse spread damps away. Antiproton machines cannot use that, and van der Meer’s stochastic cooling instead measures the position of a passing slice of the beam, sends a correction across the ring ahead of it, and kicks it — a measurement and a feedback, which exports information and therefore entropy. It is what made a usable antiproton beam possible, and therefore the discovery of the W and Z, and it shared the 1984 Nobel Prize.

The pattern is worth stating once, in the general form. A conserved phase-space volume is a bound on every passive arrangement, and the only way past it is a process that does bookkeeping — measuring, dissipating, or radiating. The laser in the previous section, the fluorescent concentrator, the synchrotron’s own radiation and van der Meer’s feedback loop are four instances of one exception.

That the same quantity carries five names — étendue in illumination, the Lagrange invariant in lens design, throughput in spectroscopy, emittance in accelerators, phase-space volume in mechanics — is a fair measure of how often it has had to be rediscovered.

The trade a spectrometer cannot avoid #

The bound’s most consequential everyday application is not in concentrating anything. It is in the design of instruments that measure spectra, where it forces a choice between resolution and how much light gets in.

A grating spectrometer resolves by geometry: light enters through a narrow slit, the grating disperses it, and the finer the detail wanted the narrower the slit must be. But the slit is the aperture, so its width sets the étendue — and halving it to double the resolving power halves the light. Resolution is bought directly with throughput, at a fixed exchange rate, and no improvement in gratings or detectors alters the rate.

That is a serious constraint when the signal is faint, which in infrared spectroscopy it almost always is. The escape, found by Jacquinot in the 1950s, is to stop resolving with a slit.

A Fourier-transform spectrometer is a Michelson interferometer whose mirror is scanned; the spectrum is recovered from the interferogram rather than laid out in space. Its resolution is set by how far the mirror travels, which is a path difference and not an aperture, so the entrance aperture is free to be a circular hole. What limits that hole is only that rays at the extreme angle must not smear the fringes, which allows a solid angle of about $2\pi/R$ — enormously more than a slit of the same resolving power admits.

The gain is a factor of a hundred or two in étendue at equal resolution, and it is called the Jacquinot advantage. Together with the multiplex advantage — all wavelengths measured at once rather than scanned past a slit — it is why essentially every infrared spectrometer built since the 1970s is an interferometer, and why the technique took over an entire field of chemistry.

The general form of the trick is worth extracting, because it is not confined to spectroscopy. Resolution paid for out of a spatial aperture is paid for in étendue; resolution paid for out of a scan is not. The interferometer buys its resolving power with a mechanical travel and a computation, leaving the spatial phase-space volume to be spent on collecting light. Nothing has beaten the conservation law; the instrument has simply arranged not to spend its budget on the thing the grating spends it on.

A different limit, often confused with this one #

Optics has a second famous bound and the two have nothing to do with each other.

The resolution limit comes from diffraction and is often confused with this one, so it is worth separating them. Diffraction says a finite aperture cannot produce an arbitrarily small spot; this essay says an image cannot be brighter than its source. The first is about detail and improves as the aperture grows; the second is about brightness and does not improve at all.

The concentration limit contains no wavelength at all. It applies to a ray bundle of any colour, to sound, to neutrons and to a beam of particles, and it would be exactly the same in a world where light had no wave properties. The diffraction limit and the étendue limit happen to bite at similar places for visible light and a hand-sized lens, which is why they are so often run together, but they are limits on different things — one on the smallest spot, one on the greatest brightness — and they scale differently with everything.

Where the number 5,772 comes from #

One loose end is worth closing, because it makes the coincidence less mysterious.

The Sun’s 5,772 K is not an independent fact from the 1,361 watts per square metre arriving here. Given the flux and the angular size, the brightness temperature follows — and it is the same temperature a black surface at the focus of any optic would reach. The number in every textbook is therefore this essay’s ceiling, computed once and quoted ever since.

So the ceiling and the source temperature are computed from the same three measurements, and their agreement is a consistency check rather than a discovery. What the blackbody curve itself contributes is the constant: the spectrum that classical physics could not bring down is the reason a temperature and a flux are related by a fourth power at all, and the Stefan constant in that relation is where the 5,770 gets its last significant figure. What is not circular, and is the physics, is that the form of the bound — an area ratio limited by a sine squared — comes out of ray counting and lands exactly on the thermodynamic requirement. Two arguments with no shared apparatus reach the same place.

Where the model stops #

Only a passive system is bounded. A laser breaks the radiance limit trivially, because it is not passive: it has a population inversion, which is a negative temperature, and thermodynamics places no ceiling on concentrating a beam whose source is hotter than any body. Fluorescent concentrators do the same thing more gently, absorbing at one wavelength and re-emitting at another — the bound applies wavelength by wavelength and they move between wavelengths.

The receiver’s index matters and is easy to forget. The bound is $n^2/\sin^2\theta$ with $n$ measured at the receiver, so immersing the absorber in a medium of index 1.5 raises the ceiling by a factor of 2.25. That is real, it is used, and it is bounded in turn by the highest index available.

Nothing here is spectrally selective. A real absorber can be made black in the visible and shiny in the infrared, which raises its equilibrium temperature above the figure by suppressing its own re-emission. That does not break the bound — the bound is on concentration — but it does break the naive translation from concentration to temperature, and every good solar receiver exploits it.

Étendue is conserved and entropy is not. A lossless system conserves the phase-space volume of the beam; a real one scatters, absorbs and diffuses, all of which enlarge it irreversibly. That asymmetry is the optical face of the same arrow the counting argument gives to thermodynamics — the quantity can grow and cannot shrink, and a system that returned it to its original value would be running the second law backwards.

And the second law does not bound a rate. The ceiling says what temperature is reachable in equilibrium. It says nothing about how long a receiver takes to get there, how much power it collects, or what happens when its losses are included, all of which are the questions an engineer has.

What the pictures cannot show #

The étendue figure draws a beam as a bundle of rays with a sharp edge, and no real beam has one. Étendue is properly an integral over a phase-space volume, and drawing it as an area and a cone is a two-dimensional cartoon of a four-dimensional object. The cartoon is faithful about the conservation and silent about the shape.

Nor can the temperature figure show what a receiver at 5,770 K would be doing. Nothing survives there; the drawing is of an idealised black surface with no losses, no convection and no material, and the highest temperature actually reached by solar concentration is a little over 3,000 K.

Where this ladder goes next #

What has been established is that an optical system has a conserved quantity, that the conservation bounds concentration, and that the bound coincides with a thermodynamic one derived from an entirely different argument. The reason those two arguments meet is that light in equilibrium with matter is itself a thermodynamic system, with an entropy and a temperature — and a beam’s étendue is, to within constants, its entropy.

The habit worth carrying away is the question that separates a real limit from an engineering one. Does the bound contain any property of the apparatus? The diffraction limit contains a wavelength and an aperture, so a shorter wavelength and a bigger aperture improve it. The concentration limit contains neither, only the angular size of the source and the index at the receiver, so no improvement in the apparatus can touch it. A bound with no apparatus in it is a bound on the arrangement rather than on the components, and it will still be there when everything has been made perfectly.

What is left on this ladder is the entropy of radiation itself: what it means for a beam to have a temperature, why a beam of sunlight arriving at Earth is not in equilibrium with anything, and how much of the energy in it is available as work.