How Will the Universe End?

Somewhere in a laboratory in a Swiss valley, a machine strikes protons together at nearly the speed of light, and physicists hold their breath — not because they fear discovering something, but because for one uneasy decade after the Higgs boson's mass was pinned down, some calculations suggested the vacuum we live in might not be the true, lowest-energy state of the universe. If that were so, then somewhere, someday, a bubble of true vacuum could nucleate and expand at the speed of light, rewriting the laws of physics as it swallowed galaxies whole, arriving at Earth without warning, faster than any light that could tell us it was coming. This is not science fiction. It is one candidate ending, seriously entertained in the technical literature, for a universe that seems otherwise so serenely, boringly stable.

That such an apocalypse can be written down in the equations of quantum field theory, argued over in journals, and neither confirmed nor ruled out — this is the strange position cosmology now occupies. We have measured the universe's expansion with exquisite precision. We know it is accelerating. We do not know why. And that single unresolved fact — the true nature of dark energy — cascades forward through billions and trillions of years to determine whether the cosmos ends in ice, in fire, in a rip, or in a catastrophic reset of physical law itself.

The question of how the universe ends sounds like metaphysics dressed up as physics, the kind of thing better suited to a campfire than a colloquium. But it is a real, working problem in cosmology, constrained by real data, and it turns entirely on a number we have measured only crudely: how the density of dark energy evolves with time.

svg[The three classic destinies of an expanding universe, depending on dark energy's equation of state $w$]

The observational case for something behaving like Λ is genuinely strong. Planck's measurements of the cosmic microwave background, combined with baryon acoustic oscillation surveys, constrain $w$ today to be $-1.03 \pm 0.03$ or so, depending on exactly which datasets are combined — consistent with exactly $-1$ to within a few percent. That is a triumph of observational cosmology: a number describing the fate of everything, measured to a few percent precision, from light that has traveled billions of years and structure imprinted in the arrangement of galaxies across the observable sky.

And yet a few percent is not zero percent. That residual uncertainty is the entire subject of this essay.

The trouble is twofold, and the two troubles compound each other.

First, the observational trouble: a growing, genuinely unsettled tension in the data. In 2024 and 2025, the Dark Energy Spectroscopic Instrument (DESI) — a survey mapping the three-dimensional positions of tens of millions of galaxies and quasars — released results suggesting that a simple constant $w = -1$ might not be the best fit after all. When DESI's baryon acoustic oscillation measurements are combined with supernova compilations (Pantheon+, Union3, or DES-SN5YR) and CMB data, the preferred fit in an extended model — one that allows $w$ to evolve with redshift according to a simple two-parameter form,

with $a$ the cosmic scale factor — comes out with $w_0 > -1$ today and $w_a < 0$, meaning dark energy may have been more negative (more "phantom-like") in the past and is drifting toward less negative, less exotic behavior now, perhaps even threatening to cross into decelerating territory in the future. Depending on the combination of datasets used, this departure from a pure cosmological constant is reported at somewhere between two and four standard deviations — intriguing, persistent across successive DESI data releases, but not yet the five-sigma gold standard that would let cosmologists call it a discovery. It could be new physics. It could be an unaccounted-for systematic in how supernova luminosities are calibrated, or subtle evolution in the supernova progenitor population, or an unmodeled effect in the BAO analysis. The community is, at the time of writing, genuinely divided, and that division is healthy: this is exactly the stage at which real science looks less like a settled textbook and more like people arguing productively over a whiteboard.

Second, and deeper: even if we perfectly measured $w(z)$ across the entire observable history of the universe, we would still be extrapolating into a future we cannot directly probe, using physics we do not understand at a fundamental level. This is where the cosmological constant problem rears its head — the notorious mismatch, by something like 120 orders of magnitude, between the vacuum energy density predicted by quantum field theory (summing zero-point energies of all fields) and the tiny value actually observed driving cosmic acceleration. Nobody has a persuasive first-principles derivation of why Λ has the value it has, or indeed why it should be constant, or a slowly rolling scalar field's potential energy (quintessence), or something stranger tied to modifications of gravity itself on cosmic scales. Without an underlying theory, extrapolating the equation of state trillions of years into the future is a bit like extrapolating a stock's price from three days of ticker data: technically you can draw the line, but you have no warrant to trust it far from where you measured it.

There is a third and stranger failure mode, orthogonal to the expansion-history question entirely: the vacuum itself might not be stable. This comes from particle physics, not cosmology. The measured mass of the Higgs boson (about 125 GeV) and the mass of the top quark sit, according to Standard Model renormalization group calculations, disconcertingly close to a boundary in parameter space separating "vacuum absolutely stable forever" from "vacuum metastable — long-lived, but not the true minimum of the Higgs potential." Current central-value calculations place us just barely in the metastable regime, meaning quantum tunneling could someday, somewhere, nucleate a bubble of true vacuum with different physical constants (different particle masses, possibly no stable atoms) that would expand at light-speed and be utterly unstoppable and utterly undetectable in advance. The calculated lifetime of this metastability, if it's real, vastly exceeds the current age of the universe — this is not a near-term worry — but the fact that our best current framework leaves this door open, unresolved by anything except better measurements of the top quark mass and Higgs self-coupling, is itself a startling gap in a theory otherwise considered spectacularly successful.

Four broad scenarios dominate serious discussion, and it's worth being honest that they are not exactly symmetric in status: one is the default, cautious extrapolation of established physics; the others range from plausible-but-unconfirmed to exotic-but-logically-consistent.

Heat death (the "boring" default). If dark energy really is a cosmological constant, exactly $w=-1$ forever, the universe expands exponentially in the far future. Distant galaxies redshift beyond detectability; within about $10^{14}$ years, the last stars burn out; over inconceivably longer timescales, black holes evaporate via Hawking radiation; protons may or may not decay (this too is unknown — grand unified theories generically predict proton decay, but experiments like Super-Kamiokande have pushed the lower limit on the proton lifetime past $10^{34}$ years without a detection). Eventually, on timescales that make even a black hole's evaporation ($10^{100}$ years for the largest ones) seem instantaneous, the universe approaches maximum entropy: a cold, dark, thin de Sitter space with a horizon and, in some analyses, only sparse Hawking-like radiation as a residual "temperature." This is heat death in its modern form, and it is the future built into vanilla ΛCDM. It has the virtue of requiring no new physics beyond what's already in the standard model, and the vice of being unfalsifiable on any human timescale — a "boring" answer that is also, notably, the null hypothesis that all current data statistically favor, even as recent results poke at it.

The Big Rip. If $w < -1$ — phantom dark energy — the energy density of the vacuum actually grows over time rather than staying fixed. The expansion accelerates so violently that it eventually overwhelms every bound structure in the universe in reverse order of scale: first superclusters unbind from each other, then galaxies unbind from their neighbors, then the Milky Way itself is torn apart, then solar systems, then planets, and in the final $10^{-19}$ seconds before a mathematical singularity, atoms themselves are ripped apart, all in a finite, calculable future time — a paper by Robert Caldwell, Marc Kamionkowski, and Nevin Weinberg in 2003 first worked out this scenario rigorously and gave it its dramatic name. Phantom dark energy is not obviously forbidden by any deep principle, but it is deeply uncomfortable theoretically: field-theoretic models producing $w < -1$ tend to violate the null energy condition and often harbor ghost instabilities — negative-kinetic-energy degrees of freedom that generically wreck the theory's consistency, allowing runaway particle-antiparticle production from the vacuum. Nonetheless, as noted above, some recent DESI-based fits to $w_0$–$w_a$ models place $w$ mildly on the phantom side in the recent past, keeping this scenario alive in the literature, if unconfirmed and theoretically awkward.

The Big Crunch. If dark energy is not constant but decaying — rolling down some potential in a quintessence-like model, and eventually crossing to positive energy density's cousin or simply fading away relative to matter — the acceleration could stop and reverse. Ordinary matter and dark matter, diluting more slowly than a decaying dark energy component, could reassert gravitational dominance, halting the expansion and pulling the universe back down into contraction: a mirror image of the Big Bang, ending (in the classical picture) in a singularity, or, in more speculative cyclic models entertained by theorists like Paul Steinhardt and Neil Turok, bouncing into a new expansion phase entirely — an oscillating universe, ancient in concept (versions go back to informal Big Bang-era speculation and Richard Tolman's 1930s thermodynamic analysis) but revived in modern brane-cosmology language in the 2000s. This scenario currently has less direct observational support than the mild phantom-leaning hint from DESI, but it remains a logically live possibility precisely because we don't know the underlying dynamics driving $w(z)$, and some quintessence models naturally produce exactly this kind of late-time reversal.

Vacuum decay. Entirely independent of how the expansion evolves, there's the possibility that the electroweak vacuum itself is metastable, and that quantum tunneling — or, some analyses suggest, an assist from black hole seeding of the transition — nucleates a bubble of true vacuum somewhere in the universe. This bubble wall would expand at essentially the speed of light, converting everything it touches to a different vacuum state with different particle masses and interactions, arriving with no warning whatsoever, since nothing outruns the wall to tell us it's coming. This is not, as of current calculations, expected to happen anytime soon — the estimated tunneling rate implies a lifetime enormously longer than the current age of the universe — but it is a live theoretical possibility, discussed seriously by particle theorists (a landmark 2012 calculation by Giuseppe Degrassi and collaborators, following up on earlier work by others including Isidori, Rychkov, and Strumia, sharpened just how close the Standard Model sits to this stability boundary), and it depends sensitively on measured quantities — the top quark mass in particular — that current colliders measure only to about half a percent precision, not quite tight enough to settle the question definitively.

It is worth pausing to notice how differently these four scenarios are supported. Heat death is the conservative extrapolation of the best-fit model to existing data — not glamorous, but currently favored if you weight by Occam's razor and treat the DESI hints cautiously pending confirmation. The Big Rip has a specific, live (if contested) observational foothold in recent large-scale-structure surveys. The Big Crunch is logically consistent and has historical pedigree but currently weaker direct support than the phantom-leaning hint. Vacuum decay is a wildcard from a completely different branch of physics, resting on Standard Model parameters measured to sub-percent precision that happen to sit tantalizingly near a stability boundary — worth taking seriously, but also a reminder that "close to a boundary" in a logarithmically sensitive calculation is not the same as "known to be on one side of it."

There is no single crucial experiment here, in the way a particle physicist might point to a specific collision energy where a predicted resonance must appear or not. Instead, settling this question is a war of attrition against statistical uncertainty, fought on several fronts simultaneously.

The most direct front is exactly what DESI is already doing, and what its successors will do at greater scale: mapping the three-dimensional distribution of tens of millions of galaxies and quasars across cosmic time, using the baryon acoustic oscillation scale — a "standard ruler" imprinted in the clustering of matter by sound waves in the pre-recombination plasma — to trace the expansion history $H(z)$ directly. DESI's full multi-year dataset, expected to be complete later this decade, roughly triples the data behind the current results; the Vera C. Rubin Observatory's Legacy Survey of Space and Time, beginning full science operations in the mid-2020s, will deliver an independent handle through weak gravitational lensing and its own supernova and large-scale-structure programs; ESA's Euclid satellite, already returning data, targets the same expansion history through lensing and galaxy clustering out to higher redshift. If the current $w_0$–$w_a$ hint is real, these surveys, combined, should push the significance from the current "intriguing" 2–4 sigma toward a decisive 5-sigma detection or firm exclusion within roughly the next five to ten years. If it evaporates as more data comes in — as several previous "anomalies" in cosmology have, historically, done — that itself will be an important, if less thrilling, result: a tightened confidence that $w=-1$ really is exact, at least across the redshift range we can probe.

A second front is theoretical: building a compelling, falsifiable model of why dark energy might evolve, rather than simply fitting a phenomenological two-parameter curve to data. Quintessence models — a slowly rolling scalar field, conceptually similar to the inflaton field thought to have driven cosmic inflation — make specific predictions for how $w(z)$ should behave, and some of these are now being tested directly against the DESI-preferred trajectory. Modified-gravity alternatives, where the acceleration isn't due to a new energy component at all but a large-scale breakdown of general relativity, make different predictions again, often testable through how the growth of structure (not just the expansion rate) evolves — a place where the Rubin Observatory's lensing maps could prove decisive, since general relativity and many modified-gravity models diverge specifically in how expansion history and structure growth relate to one another.

For the vacuum-decay question, the settling measurement is comparatively mundane and already underway: refining the measured top quark mass and Higgs self-coupling at the LHC and its planned successors (the High-Luminosity LHC upgrade, and further out, proposed machines like the Future Circular Collider), since the Standard Model's stability calculation is exquisitely sensitive to these two numbers. A shift of even a few hundred MeV in the top mass can move the calculation from "metastable" to "stable" or deeper into "metastable," which is a strange thing to contemplate — that the fate of the vacuum may hinge on a measurement made with particle detectors the size of cathedrals, refined collision by collision.

And there's a sobering asymmetry worth naming honestly: heat death, the Big Rip, and the Big Crunch are all, in principle, questions we can make real observational progress on, because their signatures are written into $H(z)$ across redshifts we can actually observe. Vacuum decay is different — even perfect knowledge of Standard Model parameters would only tell us the calculated metastability lifetime under specific assumptions (including, notably, the absence of new physics beyond the Standard Model at higher energies, which could change the calculation substantially). We may never get better than a probabilistic, model-dependent statement about that particular doom.

It would be easy to wave this off as cosmic trivia — humanity, and very likely all life, and very likely the Sun and Earth themselves, will be long gone tens of billions of years before any of these endings plays out in earnest, so why does the distinction matter?

It matters first because the equation of state of dark energy is not a footnote to physics; it is one of the two or three deepest open problems in the field, intimately tangled with the cosmological constant problem — arguably the largest quantitative discrepancy between theory and observation in the history of science. A definitive detection that $w \neq -1$, confirmed and reproduced, would be a firework going off in theoretical physics: it would rule out the simplest possible explanation for cosmic acceleration and demand a genuinely new field, or a revision of gravity itself on cosmological scales, joining the ranks of revolutions like the discovery of the expansion itself. Conversely, ruling out any evolution to exquisite precision would sharpen the cosmological constant problem to a fine point, forcing theorists to explain not just why vacuum energy is so absurdly small, but why it is so precisely, permanently constant — a fact that would cry out for either a symmetry we haven't found, or an anthropic explanation drawing on the deeply contested idea of a multiverse of vacua with different constants, of which ours happens to be bio-friendly.

It matters, too, because these scenarios are genuinely different physical futures with different implications for what kinds of structures, and — some cosmologists have speculated, more informally — what kinds of information-processing, life, or intelligence could exist arbitrarily far into the future. Freeman Dyson's classic 1979 paper on the long-term survival of intelligent life explored exactly this in a matter-dominated, non-accelerating cosmology; the discovery of dark energy in 1998 forced a wholesale revision of those calculations, since an eternally accelerating universe eventually places almost everything beyond one another's cosmic horizon, cutting off communication, energy transfer, and — in most analyses — any hope of indefinitely sustained complexity, in a way a merely expanding-forever-but-decelerating universe would not.

And it matters, finally, as a case study in how science handles genuine uncertainty at the frontier. The public conversation around dark energy often collapses into false certainty in one direction or another — either "the universe will expand forever, case closed" or breathless "scientists say universe might rip apart" headlines chasing a single new paper. The actual state of the field is more interesting than either: a hard-won, well-tested standard model, under productive stress from new data, with several serious and mutually exclusive futures still on the table, and a research community racing — literally, with billion-dollar telescopes — to find out which one is real.

The discovery that set this whole question in motion belonged to two teams and, eventually, a shared 2011 Nobel Prize in Physics: Saul Perlmutter, leading the Supernova Cosmology Project, and Brian Schmidt and Adam Riess, leading the High-Z Supernova Search Team. Their 1998 papers — arriving at the same startling conclusion independently, using the same technique of Type Ia supernovae as standard candles — are as close to a single, sharp turning point as cosmology has had since the discovery of the cosmic microwave background in 1965.

history

• 1917 — Einstein introduces the cosmological constant to permit a static universe.
• 1929 — Hubble establishes the expansion of the universe from galaxy redshifts.
• 1998 — Two supernova teams discover the expansion is accelerating.
• 2024–2025 — DESI reports hints that dark energy's equation of state may deviate from $w=-1$.

There is something almost vertiginous about a question this large resting, ultimately, on the third decimal place of a single cosmological parameter. Everything — whether galaxies a hundred billion years from now will drift alone in an eternally darkening sky, or be ripped violently from their own constituent atoms, or be gathered back into a fiery point, or persist in a universe that could, without warning, be replaced by a different vacuum with different laws — depends on whether $w(z)$ is a flat line or has the barest hint of a slope, and on where a Higgs potential's second minimum truly sits.

This is, in the best sense, an open frontier rather than a closed mystery awaiter a clever trick: it is being worked on right now, with data arriving in real time, by people who genuinely do not know the answer and say so in their papers. That combination — enormous stakes, genuine uncertainty, and active, well-funded, methodical pursuit — is rare, and it's worth savoring rather than resolving prematurely into false confidence in either direction.

What remains wide open:

• Is the DESI-hinted deviation from $w=-1$ real, or a statistical fluctuation or unmodeled systematic that will fade with more data?
• If dark energy does evolve, is it a rolling scalar field (quintessence), a sign of modified gravity, or something without precedent in current theoretical frameworks?
• Can any theory explain the value of the cosmological constant from first principles, resolving the 120-orders-of-magnitude problem, or is an anthropic, multiverse-based explanation the best we will ever get?
• Is the Standard Model vacuum truly metastable, and if so, does new physics beyond the Standard Model stabilize it or leave the danger intact?
• Could gravitational-wave observatories, 21-cm cosmology, or future CMB experiments (like CMB-S4) break the current degeneracies faster than galaxy surveys alone?

Ask the tutor: If the next decade of data confirms that dark energy's equation of state really does drift away from $w=-1$, what would that imply for the cosmological constant problem — would a dynamical explanation make the fine-tuning puzzle easier to solve, or harder?