Measuring Dark Energy

The Expansion History

The universe is expanding. About a quarter of a century ago, distant supernovae showed that the expansion is speeding up; whatever speeds it up is called dark energy. The simplest explanation is a “cosmological constant”: energy whose density does not change with time. If dark energy changes, the expansion rate in each era differs from what a constant predicts.

DESI measures distances: at different redshifts, that is, at different times in the past, how far away galaxies are and how fast the universe was expanding then, reaching back about 11 billion years. In three years of data (DR2), each redshift bin’s distance is measured to between 0.45 % and 1.54 %.

seven redshift bins · 0.45⁠–⁠1.54 % per bin (DR2) · about 11 billion years back

A Ruler Frozen into the Galaxies

The early universe was a hot plasma carrying sound waves. By a redshift of about 1060 the photons stopped dragging ordinary matter along, the waves stalled, and they left one fixed distance: about 147 megaparsecs (Mpc), about 480 million light-years. These are baryon acoustic oscillations, BAO for short.

The distance is printed in how galaxies are spread: pairs of galaxies about 147 Mpc apart are a little more common than pairs at other separations. It is a statistical effect. A single galaxy has no visible ring round it; the red ring on the plaque only marks the ruler’s length. In 2005 the Sloan Digital Sky Survey measured it with 46,748 galaxies, and the 2dF survey reported it the same year. Since the ruler’s length is known, how large it looks at each redshift gives the distance.

about 147 Mpc ≈ 480 million light-years · measured in 2005 (SDSS, 46,748 galaxies, 3.4σ)

Redshift in a Spectrum

A galaxy’s redshift comes from its spectrum. The longer light travels, the more the expansion of the universe stretches it, and the further its spectral lines move towards the red. The redshift z is (observed wavelength − original wavelength) ÷ original wavelength.

Each fibre carries one galaxy’s light to a spectrograph, which measures how far its lines have moved. Galaxies forming stars give off a pair of oxygen-ion lines near 372.7 nm; in a galaxy at redshift 1 they appear at about 745 nm, and DESI can find them out to redshift 1.6.

z = (λ observed − λ original) ÷ λ original · [O II] 372.7 nm → about 745 nm at z = 1

At the Top of the Telescope

DESI is mounted on the Mayall telescope on Kitt Peak, Arizona. Its 4-⁠metre primary mirror reflects the light up to the prime focus at the top of the tube; DESI’s first lens is about 8.7 m above the mirror, and from it to the focal surface runs a 2.43 m train of lenses. With the new cage and vanes, the top end’s mass budget is 10,700 kg.

Here the top end is taken off and laid on its side at 1:4: six lenses in a steel barrel, the focal plane and its enclosure behind them, and the hexapod that hangs the barrel from the cage’s middle ring. A piece is cut away at the front and top, the cut faces painted red. A 1.75 m person stands beside it.

4 m primary · first lens about 8.7 m above it · 2.43 m from C1 to the focal surface

Six Lenses

The six lenses take the light from the primary and bring a field 3.2° across (about 8 deg², about 40 full Moons) to a sharp focus, changing the focal ratio from f/2.8 to about f/3.9. The largest, C1, is 1.14 m across and weighs 201 kg; C4 is 1.03 m and 237 kg; the six weigh 864 kg in all.

The middle two are the atmospheric dispersion corrector: two borosilicate lenses ground with a wedge of about 0.25°, each in a large bull gear, turning in opposite directions. When the telescope looks low in the sky, the air acts as a weak prism and draws each star into a tiny rainbow streak; turned to the right angles, the two wedges pull it back.

C1 1.14 m · C4 1.03 m · a 3.2° field, about 8 deg² · 864 kg of glass

Ten Petals

The focal surface is 812 mm across, slightly convex, and divided into ten 36° wedges called petals. Each petal is an 8.6 kg block of aluminium with 514 holes, each pointed along the chief ray at its spot: 502 hold robots, 12 hold fiducials that glow blue. The ten hold 5,020 robots.

Each petal’s 500 fibres run to one spectrograph; 2 more go to a sky-⁠monitor camera. A guide or focus camera sits at each petal’s outer corner: six guide, four sense the wavefront. The whole focal plane is closed in an enclosure held at 11 °C.

10 × 502 = 5,020 robots · 5,000 fibres to the spectrographs · an 812 mm focal plane

One Robot, Two Arms

Each robot is a thin rod about 29 cm long and 8.3 mm across, screwed into its hole in the petal like a spark plug. At the front are two axes: the central θ axis turns the whole head, and the φ axis, about 3 mm off centre, turns an arm about 3 mm long that holds the fibre at its end. The fibre can reach any point in a circle 12 mm across.

Each axis has its own motor, 4 mm across, geared down about 337 times. Neighbours stand as close as 10.4 mm, so the circles they can reach overlap: each robot pulls its arm in, turns, then reaches out again, to keep clear of the others. The fibre’s core is 107 µm across, about 1.5 arcseconds on the sky.

arms of about 3 mm each · a 12 mm patrol circle · 10.4 mm apart at the closest · 107 µm ≈ 1.5″

A New Field in Two Minutes

An exposure usually lasts 1,000 seconds. When it ends, the telescope turns to the next field and all 5,020 robots move their fibres to new galaxies. The robots are actually moving for only 8 to 12 seconds; with the anti-⁠collision planning, the commands and the checks, the gap between exposures is about 2 minutes, 118 s at best.

The check: the fibres are lit from the spectrograph end, so their tips shine as small points; the fibre-⁠view camera behind the primary mirror, about 12 m from the focal surface, photographs them, works out the errors, and the robots correct once. After the correction they are off by about 6 to 9 µm, against a requirement of 10 µm.

expose about 1,000 s · move 8⁠–⁠12 s · about 2 min between fields · 6⁠–⁠9 µm off

Nearly Fifty Metres of Fibre

Each petal’s fibres gather into one cable, and each of the ten cables is 47.5 m long; with the piece of about 3 m inside each robot, the run from fibre tip to spectrograph is nearly 50 m. The cables follow the vanes to the telescope’s upper ring, down the truss, through both of the telescope’s axes and through a wall into the room that holds the spectrographs.

Each cable holds about 530 fibres, a few spares more than are used, wound round a strength member like an undersea cable. About 90 % of the light gets through. The cables break off on the floor of the model: their real length does not fit.

ten cables · 47.5 m each · about 530 fibres each · about 90 % throughput

Ten Spectrographs, Thirty Cameras

At a cable’s far end, 500 fibres line up in a slit 121 mm long, where the light enters a spectrograph. Two dichroics split it into three channels: blue 360⁠–⁠593 nm, red 560⁠–⁠772 nm and near-⁠infrared 747⁠–⁠980 nm. Each has its own grating, camera and CCD; the blue CCDs are cooled to 163 K, the red and near-⁠infrared to 140 K.

The ten spectrographs stand in two rows of five on a rack 8.5 m long, in a clean room held at 20 °C. Only one column of the rack is drawn here: two spectrographs. After each exposure, reading them out takes 60 seconds.

10 × 3 = 30 cameras · 360⁠–⁠980 nm · resolution about 2,000⁠–⁠5,500

A 3D Map

The main survey began on 14 May 2021. The first year (DR1) measured BAO with over 6 million galaxies and quasars; three years (DR2) with over 14 million, from redshift 0.1 to 2.1, plus the Lyman-⁠α forest in the spectra of more than 820,000 quasars: a row of absorption lines that distant hydrogen gas leaves in a quasar’s light, which takes the measurement out to a redshift of about 2.3.

The planned five-⁠year survey was finished on 15 April 2026 with more than 47 million galaxies and quasars and more than 20 million stars, above the planned 34 million. Berkeley Lab calls it the largest high-⁠resolution 3D map of the universe so far. The plaque on the stand is this page’s sketch: a thin slice, with the Earth at its left point.

DR1: over 6 million · DR2: over 14 million · five years: over 47 million (April 2026)

Is Dark Energy Changing?

Two numbers describe dark energy here: w₀, its equation of state today, and wₐ, how fast that changes over time; a cosmological constant has w₀ = −1 and wₐ = 0. DESI’s distances are put together with the cosmic microwave background (CMB) and supernovae to find which w₀ and wₐ fit best.

Three years of data prefer a dark energy that changes, by an amount that depends on the data used: DESI alone 1.7σ, with the CMB 3.1σ, and with one of three supernova samples as well, 2.8σ, 3.8σ or 4.2σ (2.5σ to 3.9σ in the first year). The Dark Energy Survey (DES) later recalibrated its supernovae, and that combination fell from 4.2σ to 3.2σ. A discovery in physics takes 5σ; this is a hint, not yet a conclusion.

DR2: DESI 1.7σ · with the CMB 3.1σ · with supernovae 2.8⁠–⁠4.2σ · recalibrated 3.2σ · none at 5σ

What Would Settle It

DESI’s dark-⁠energy results from the full five years are expected in 2027; the survey has been extended to 2028, to cover 17,000 deg² and aim for 63 million galaxies and quasars. More supernovae are coming too: DESI’s paper names ZTF and the Rubin Observatory’s survey, whose camera is the plate “The Largest Digital Camera”.

If the hint grows with more data, dark energy is not a constant, and the universe will go on expanding differently from what a constant predicts; if it fades, the cosmological constant stays the simplest explanation. For the other unseen ingredient, dark matter, see the plate “Hunting Dark Matter”.

five-⁠year results about 2027 · extended to 2028 · 17,000 deg² · aiming for 63 million

Taken Apart

The corrector, the hexapod, the cage, the focal plane, its enclosure, the fibre cables and the spectrographs come apart one by one.

Point at a part (tap it on a phone) and it lifts on its own, with its name; the rest fade.