Is the universe expanding, or are you shrinking?

Light is the fastest thing there is, but it isn't instantaneous: it covers 299,792 kilometres every second. Sunlight is about eight minutes old when it reaches you. Light from the Andromeda galaxy, about 2.5 million light-years away, set out before our species existed. A light-year is a distance, the distance light travels in a year (about 9.5 trillion km), and it contains the central fact of observational cosmology: looking far away means looking back in time.

The most distant galaxy confirmed so far with a spectrum is MoM-z14, at a redshift of 14.44, measured with the James Webb Space Telescope25. Its light set out when the universe was about 280 million years old, roughly 2% of its present age. Telescopes don't only show us where things are; they show us when they were. Every signal we receive is a message from the past, interpreted with instruments that exist here and now. That detail will matter.

Spread starlight out with a prism or a grating and you get a spectrum crossed by thin dark (or bright) lines. Each kind of atom absorbs and emits light only at its own fixed set of wavelengths, a pattern as distinctive as a barcode. Hydrogen's visible lines, the Balmer series, sit at 656.3, 486.1, 434.0 and 410.2 nanometres in the laboratory.

When astronomers find the same pattern in a distant galaxy, but with every line moved to longer wavelengths by the same factor, they call it a redshift:

Notice what's actually measured: a ratio. Light from the galaxy is compared with light from the same kind of atoms in a laboratory on Earth today. Hold on to that; Part II is built on it.

The first galaxy speeds came from Vesto Slipher at Lowell Observatory. His 1912 spectrum of Andromeda showed it approaching at about 300 km/s, but by 1917 he had measured 25 "spiral nebulae", and 21 of them were moving away, some at over 1,000 km/s1. Nobody yet knew these nebulae were separate galaxies.

Meanwhile theory was racing ahead. In 1917 Einstein applied general relativity to the whole universe and, assuming it must be static, added a "cosmological constant" Λ to hold it still2. In 1922 Alexander Friedmann showed that the equations naturally allow expanding or contracting universes3. In 1927 the Belgian priest and physicist Georges Lemaître went further. He derived an expanding universe and combined Slipher's velocities with Hubble's 1926 distance estimates to measure the rate, about 625 km/s per megaparsec4. His paper appeared in French in a little-read Belgian journal and went largely unnoticed.

In 1929 Edwin Hubble published distances to 24 galaxies alongside their velocities, most of them Slipher's6. The pattern was unmistakable: the farther a galaxy, the faster it recedes, in proportion. In 2018 the International Astronomical Union voted (78% in favour) to recommend calling it the Hubble–Lemaître law7:

Hubble's own value of the constant, about 500 km/s/Mpc, was roughly seven times too high, because of errors in the distance scale that took decades to fix. Modern values are 67–73 km/s/Mpc (more on that spread in Chapter 8). In everyday terms (taking the CMB-based 67.4 km/s/Mpc), a galaxy a billion light-years away recedes at about 20,700 km/s, and every light-year of intergalactic space grows by about 2 centimetres per second. Einstein dropped his cosmological constant once expansion was established, and reportedly called it his "biggest blunder"36. A Λ-like term would make a comeback in 1998.

The picture from general relativity isn't galaxies flying through space away from a central explosion. Space itself grows. On large scales galaxies stay at roughly fixed comoving positions, and every distance between them scales with a single number, the scale factor a(t):

Because every distance grows by the same factor, every galaxy sees every other galaxy receding at a speed proportional to distance. There's no centre and no special place, as the figure above shows for any galaxy you pick.

Two common misconceptions:

• Not everything expands. Atoms, people, planets, the solar system, the Milky Way and our Local Group of galaxies are held together by electromagnetic or gravitational forces and don't take part. The expansion's effect on them is utterly negligible17. Our neighbour Andromeda is actually approaching us at about 110 km/s (Slipher's 300 km/s also includes the Sun's own orbit around the Milky Way)18. Only the gaps between gravitationally unbound groups of galaxies grow.
• The analogies break somewhere. The inflating balloon (galaxies as stickers on its skin) and rising raisin bread (raisins drifting apart as the dough rises) capture "everyone recedes from everyone". But a balloon has an inside and an outside, and a loaf has a centre and an edge. As far as we can tell, the universe has neither.

Since recession speed grows with distance, at some distance it equals the speed of light. That distance, the Hubble radius c/H₀, is about 14.5 billion light-years (for the CMB-based 67.4 km/s/Mpc of Chapter 8). Galaxies with redshifts above about z ≈ 1.5 are receding faster than light today16.

This doesn't break relativity. Special relativity forbids anything from moving through space faster than light relative to a local observer. It puts no limit on how fast the distance between two far-apart objects can grow when space between them is stretching. We routinely see such galaxies, including some that were already receding faster than light when they emitted the light we now receive16. For the same reason, although the universe is 13.8 billion years old, the most distant matter we can in principle see is now about 46 billion light-years away: the space its light crossed kept stretching behind it.

Run the expansion backwards and everything gets closer together, denser and hotter. Lemaître himself proposed in 1931 that the universe began from "a single quantum", an idea he later called the "primeval atom"5. The modern version makes sharp predictions, and they came true:

• A leftover glow. In 1948 Ralph Alpher and Robert Herman predicted that the hot early universe should have left radiation that would today be only a few degrees above absolute zero (they estimated about 5 K)8. In 1964–65 Arno Penzias and Robert Wilson found it by accident, as unexplained microwave noise in a radio antenna that survived every check they could think of, even evicting nesting pigeons and cleaning out their droppings9, and shared the 1978 Nobel Prize15. This cosmic microwave background (CMB) was released about 380,000 years after the Big Bang, when the universe had cooled to about 3,000 K and became transparent. Its light has since been stretched about 1,100-fold. The COBE satellite showed it has the most perfect thermal ("blackbody") spectrum ever measured; its temperature today is 2.72548 ± 0.00057 K10. John Mather and George Smoot shared the 2006 Nobel Prize for this and for COBE's first map of its faint hot and cold spots.
• The first elements. In the first few minutes, nuclear reactions should have turned about a quarter of ordinary matter, by mass, into helium, plus traces of deuterium and lithium. Measurements give a primordial helium fraction of 0.245 ± 0.003 and a deuterium-to-hydrogen ratio of about 2.5 × 10⁻⁵, both matching the predictions12. (Lithium-7 comes out about three times lower than predicted, a long-standing puzzle most often attributed to lithium being destroyed in stars, though new physics hasn't been ruled out.)
• An age. The CMB combined with other data gives 13.8 billion years11.

"Big Bang" here means this hot, dense early phase, not an explosion at a point in space. What happened at the very beginning, if there was a beginning, is still an open question.

Gravity pulls, so everyone expected the expansion to be slowing down. Two teams set out to measure the slowdown with Type Ia supernovae: exploding white dwarf stars that briefly outshine their entire host galaxy and whose peak brightness can be calibrated from how quickly they fade and their colour. That makes them "standard candles". Know how bright one really is, and its apparent faintness tells you its distance.

In 1998 the High-z Supernova Search Team (16 distant supernovae) and the Supernova Cosmology Project (42) reached the same startling result: distant supernovae were fainter than expected, so the expansion is accelerating1314. Saul Perlmutter, Brian Schmidt and Adam Riess shared the 2011 Nobel Prize in Physics "for the discovery of the accelerating expansion of the Universe through observations of distant supernovae"15. Whatever drives the acceleration is called dark energy. In the standard model it behaves like Einstein's cosmological constant and makes up about 68–69% of the universe's energy budget11.

Today's supernova catalogues contain thousands of objects: Pantheon+ has 1,550, Union3 2,087, the Dark Energy Survey about 1,600, and the 2025 release of the Zwicky Transient Facility 3,628 nearby ones24. That will matter in Part III.

Very. The expanding, hot-Big-Bang picture (the "ΛCDM" model, for Λ plus cold dark matter) is supported by many independent lines of evidence that all fit together:

• Redshift grows with distance (the Hubble–Lemaître law): redshifts are now measured for tens of millions of galaxies, with independent distances for tens of thousands.
• Distant events appear slowed down. A supernova at redshift z should appear to unfold (1 + z) times slower. The Dark Energy Survey tested this with 1,504 supernovae and found the exponent b = 1.003 ± 0.011 for a (1 + z)^b law, matching the predicted 1 to about 1%19. Quasars, watched for two decades, show the same slow motion out to redshift about 4, with less precision19. "Tired light" ideas, in which light simply loses energy on the way, can't produce this.
• The CMB was hotter in the past, T = 2.725 K × (1 + z). This is confirmed in galaxy clusters out to z ≈ 0.9, to about 1% with molecules in a galaxy at z = 0.89 (5.13 ± 0.06 K, against 5.14 K predicted), and, using water molecules in a galaxy at z = 6.34, about 13 billion years ago20.
• Distant galaxies dim consistently with expansion once galaxy evolution is allowed for, and inconsistently with tired light (the Tolman surface-brightness test)21.
• The CMB's spectrum and pattern of hot and cold spots, the light-element abundances, the large-scale clustering of galaxies (baryon acoustic oscillations) and the way galaxies change with look-back time all fit the same model11.

There are real open questions, and they're about details:

• The Hubble tension. The expansion rate inferred from the CMB (67.4 ± 0.5 km/s/Mpc11) disagrees with the local distance ladder: about 73.5 ± 0.8, which its authors put about 7σ from the latest CMB prediction (Planck, ACT and SPT combined), and 5σ from one that uses galaxy maps (BAO) and nucleosynthesis but no CMB. Another team finds about 70.4 ± 1.9, between the two and consistent with both22. The cause is still debated.
• Is dark energy changing? The DESI survey reported hints that dark energy evolves over time. As of October 2026 the hint stands at roughly 3 to 3.5σ, below the 5σ discovery standard: recalibrating the supernova catalogues pulled the strongest estimates down from about 4σ and brought the different samples into agreement23.

Neither question challenges whether the universe expands; both are about how fast, and why.

No experiment measures a length against "the universe". Every measurement compares like with like: a distance with a ruler, a duration with a clock, a wavelength with light from a reference atom. The numbers that come out are dimensionless ratios. In a well-known 1962 paper, Robert Dicke made this a principle: the laws of physics must not depend on our choice of units, and only dimensionless combinations are physically meaningful2630. Today's best clocks live by this rule: a ratio of two clock frequencies is a pure number, which needs no definition of the second (Nº 9).

Our rulers and clocks are made of atoms, and atoms get their size and ticking rate from the masses of their particles. The size of a hydrogen atom (the Bohr radius) and the frequency of the light it emits are

So if the electron and every other particle grew heavier, every atom would shrink and every atomic clock would tick faster, all in step. If the strength of gravity measured in particle units (equivalently, the ratio of particle masses to the Planck mass) stayed fixed as well, every dimensionless constant of nature (α, mₑ/m_p, G m_p²/ħc, …) would be unchanged.

Suppose that tonight every material thing in the universe (every atom, ruler, clock and person) shrinks to half its size, and every atomic clock starts ticking twice as fast. Could you tell in the morning?

If light already in flight shrank too, and gravity kept in step (the Planck mass doubling along with every particle mass, as Chapter 9 requires), the answer is no. Every comparison you could make, between any two things, would give the same number as before, so the change would be invisible and physically meaningless.

But light already travelling through space isn't made of atoms; it carries no ruler of its own. If it keeps its wavelength while the atoms in your spectrograph shrink, then tomorrow the starlight arriving from distant galaxies has twice the wavelength of the light your own atoms emit. You'd measure a redshift, 1 + z = 2. Spread the shrinking smoothly over billions of years and older light shows bigger shifts. That is exactly the cosmological redshift of Part I.

Now the precise version. Take the standard expanding universe and change only the units: measure everything in units in which the space between galaxies stays fixed. Mathematically this is a Weyl (conformal) rescaling of the metric. For a spatially flat universe (ours is flat to within measurement precision11), the expanding metric

becomes, after rescaling by Ω = 1/a, the metric of flat, static space, ds̃² = −dη² + dχ². For all physics to stay identical, every mass scale in physics (particle masses, the QCD scale, the Higgs and dark-energy scales, and the Planck mass) must then grow in proportion to a: m ∝ a(η). Atoms shrink as 1/a, and atomic clocks tick faster in proportion to a. Every dimensionless observable is unchanged. This isn't a new theory but the same theory in different variables. Fred Hoyle and Jayant Narlikar wrote flat cosmology this way in 197241, and Christof Wetterich made the point vivid in 2013 with a model in which the universe even shrinks while all masses grow27. Figure 1 shows both descriptions side by side and measures each picture independently.

Redshift. In the static description, light keeps its wavelength in transit. What changed is the atoms. Atoms that emitted light long ago were lighter and therefore larger, so their light started out longer. Compared with today's smaller atoms it looks redshifted, by exactly the same factor as before:

That's the second reading of the barcode figure in Chapter 2: the same shift, explained by bigger atoms long ago instead of stretched light.

Slow-motion supernovae. Distant supernovae appear to unfold (1 + z) times slower. In the static description, atomic and nuclear processes in the past ran slower because particles were lighter, by exactly that factor. The prediction is identical. This is also why the shrinking picture is not "tired light", Fritz Zwicky's 1929 idea that light simply loses energy on its way35. Tired light fails this test and can't produce the CMB's perfect blackbody spectrum. The shrinking picture passes both, because it is mathematically the same as expansion.

Faster than light. In the static description the Hubble flow disappears: apart from small local motions (Andromeda still approaches us), galaxies sit at fixed positions. The "faster than light" recession speed is just how fast a fixed distance grows when expressed in rulers that keep shrinking. Turn on "Hubble radius" (the faster-than-light zone) in Figure 1 to see both readings at once.

The rate. Today's expansion rate, in Story 2's terms, means your rulers (and you) shrink by about one part in 14.5 billion per year. It's the mirror image of each light-year of space growing by about 2 cm per second.

Nothing forces us to put all the change into space (Story 1) or all of it into matter (Story 2). Any time-dependent conformal factor works, as long as the masses are rescaled to match. Figure 2 interpolates between the two. With Ω = a^−s, distances between galaxies scale as a^1−s and material lengths as a^−s, for any s (Figure 2 spans 0 to 1.5; past 1 the space between galaxies even shrinks, as in Wetterich's model). Every value of s gives the same measured numbers.

How to use it: the figure starts in the future (distances ×1.8). Drag the frame parameter s slider, or press "Sweep s". At a = 1 every description coincides, so the slider has nothing to change there.

If all these descriptions are equivalent, why does every textbook say "expanding"? Because it's the simplest. With particle masses held fixed, Einstein's equations take their standard form. In the static or mixed descriptions, the Planck mass becomes a time-dependent field and gravity has to be written as a "scalar–tensor" theory. It is formally of the type Carl Brans and Robert Dicke proposed in 196143, but with the one special setting (ω = −3/2) and with particle masses tied to the same field; then the extra field adds no new physics. It is just as correct, only more cumbersome. (Brans and Dicke's own theory, which keeps particle masses fixed, really is different physics, and Chapter 19's tests constrain it.) Choosing expansion is like choosing convenient coordinates or units. What was discovered (how measurable ratios change over cosmic time) is the same in every description; calling it "expansion" is the choice.

This isn't a fringe observation. Dicke's 1962 paper laid the groundwork26, Hoyle and Narlikar wrote down the static-space version of flat cosmology in 197241 (Narlikar and Arp revisited it in 199338, though they also used it for a contested claim, beyond the equivalence itself, that some quasar redshifts are not cosmological), Wetterich's 2013 paper made the case vividly27, and Nature covered it with the headline "Cosmologist claims Universe may not be expanding"29.

Distances are one-dimensional, but space is three-dimensional. Fill a box of space between galaxies with ruler cubes. When distances grow by a factor a, the number of cubes along an edge grows by a, across a face by a², and through the volume by a³. In Story 1 the box grows and the cubes stay the same; in Story 2 the box stays the same and the cubes shrink. Either way the count is identical, so the density of matter (galaxies per ruler-cube) falls as 1/a³ in both descriptions, just as standard cosmology requires.

Figure 3 shows one cell of a lattice of galaxies both ways. On the left, the bound galaxies keep their proper size while their separations grow as a. On the right, they keep their comoving positions while they, and every ruler, shrink as 1/a.

Run Story 2 backwards and something strange happens: particles get lighter instead of being squeezed together. The CMB photons keep their wavelength the whole time; what changed is us. Since the CMB was released, the energy scale of atoms has grown about 1,100-fold, so light that matched 3,000 K atoms back then reads as 2.7 K to ours. The "hot" early universe becomes one in which particle masses, and with them atomic and nuclear energy scales, were small compared with the energy of the unchanged light; early enough, they were tiny and particles behaved relativistically, exactly as in the hot Big Bang. Nucleosynthesis, the CMB and everything else in Chapter 6 come out the same.

In the simplest rewrite, the Big Bang becomes the moment when all particle masses approach zero. Wetterich went further. In a 2014 paper titled "Hot big bang or slow freeze?", he described the same observations as a very cold, slowly evolving universe in which particle masses grow and the gravitational constant falls while ordinary Newtonian attraction stays unchanged, and he built a model with no Big Bang singularity at all, spacetime being flat in the infinite past28. That's his interpretation, and it's debated. It shows how much of the "story" of the beginning depends on the description chosen.

It's natural to hope that a big enough dataset could settle the question. Could another thousand supernovae do it? No, and it's worth seeing why.

Both descriptions predict identical values for every dimensionless observable: every redshift, every brightness ratio, every time-dilation factor, every temperature ratio. More data shrinks the error bars on those values, but it can't separate two predictions that are the same number.

We already have the thousands of supernovae: Pantheon+ (1,550), Union3 (2,087), the Dark Energy Survey (about 1,600) and ZTF's 2025 release (3,628 nearby)24. They measure the expansion history a(t) superbly, which in Story 2 is the history of particle masses m(t). They confirm the (1 + z) time dilation to about 1%19, exactly as both descriptions require. More supernovae will sharpen the numbers, and may settle the Hubble tension or the DESI hint. But whatever they find will have both a Story 1 and a Story 2 reading.

Even watching the expansion happen wouldn't decide it. Redshifts should creep as the expansion rate changes: for gas at redshift 4, by the equivalent of about half a centimetre per second each year (this page's arithmetic). This Sandage–Loeb test is a flagship goal for ESO's Extremely Large Telescope; ESPRESSO has reached a few metres per second per year so far, with no signal yet44. When it is detected, Story 1 will read it as the expansion rate changing and Story 2 as the masses' growth rate changing; the measured number will be the same.

Some versions of the shrinking idea could be caught by experiment: the ones that change a dimensionless ratio, because not everything shrinks in step. (The shrinking may vary over time, and even from place to place, as long as every mass scale, the Planck mass included, follows it exactly; that is still only a change of units2640.)

• Different materials shrinking at different rates. Objects made of different materials would then fall at different rates, violating the equivalence principle, and constants of nature such as the fine-structure constant would drift.
• Matter shrinking relative to gravity's own yardstick. If particle masses changed relative to the Planck mass, from place to place or over time, gravity measured in atomic units would vary: an extra force would act on matter, light passing the Sun would be delayed differently from Einstein's prediction, and the Moon's orbit would slowly drift.

Physicists have tested exactly these effects, with stunning precision:

It's tempting to think the shrinking picture explains dark energy away. It doesn't. "Why is the expansion accelerating?" simply becomes "why do particle masses grow faster and faster, as measured by our own clocks?" (Our rulers' fractional shrinking rate, H, actually slows, just as the expansion rate does; what accelerates is the growth of the masses, m ∝ a.) Every open puzzle in cosmology (dark energy, dark matter, the Hubble tension, the DESI hint) translates one-to-one between the descriptions. Some physicists think the static frame still offers a fresh angle: Lucas Lombriser (2023) argues that the cosmological-constant problem seems to disappear there46. Since every measurement comes out the same, that is a new way of posing the question, not new evidence. The equivalence changes how we describe the universe, not what we still need to explain.

If no measurement can ever distinguish two descriptions, the standard view40 is that "which one is real?" isn't a physics question, in the same way that "is this temperature really in Celsius or Fahrenheit?" isn't a question about the weather. Expansion is the convenient description, not the uniquely true one.

"The universe is expanding" is one of the most confident sentences in modern science, and it is true. But it's true the way "the temperature is 20 degrees" is true: it relies on a choice of units we rarely notice making. Change the ruler and the same data describe a static universe of slowly shrinking matter, or anything in between.

That leaves a question worth keeping open: how many other "facts" about the world are really facts about the rulers we use to describe it?