The Ultimate Fate of the Cosmos: Heat Death, the Big Rip, and Vacuum Decay

SpaceAug 12, 202612 min

A tiny figure stands on a vast open book, gazing up as its blank final pages lift and dissolve into stars.

For almost thirty years, the story of how the universe ends has had a villain: dark energy, the thing making cosmic expansion speed up. The popular version says the acceleration is winning and every ending on offer is horrible. The first half of that was the safe bet until the last two years of data. Now I think the most interesting fact in cosmology is that the acceleration might be easing off. I also think “horrible” is the wrong emotion for all of it.

There are three main candidate endings. The universe runs down, it gets torn apart, or the vacuum itself collapses into something else. What decides between them comes down to two numbers. One describes how dark energy behaves as space stretches. The other is the mass of the Higgs boson. Both have moved in interesting ways recently.

§The default ending is a very long cooldown

Start with the boring one, because it’s the baseline and it’s less boring than its name. Heat death is not the universe getting hot. It’s the universe running out of differences. Every useful process, from a star burning to a cell dividing, runs on a gradient: something hot next to something cold, something concentrated next to something spread out. Entropy is the measure of how far those gradients have been used up. Heat death is the state where there’s nothing left to run downhill.

The best map of how we get there is still a 1997 paper by Fred Adams and Gregory Laughlin, which splits the future into eras by powers of ten.^[1]1. Fred Adams & Gregory Laughlin, A Dying Universe: The Long Term Fate and Evolution of Astrophysical Objects, Reviews of Modern Physics 69 (1997). The era boundaries and the black-hole evaporation scaling, roughly 10⁸³ years for a million-solar-mass hole and growing as the cube of the mass, are theirs. Their proton lifetime of about 10³⁷ years is an assumption, not a measurement. Stars keep forming and shining until roughly 10¹⁴ years. Then comes the Degenerate Era, out to about 10³⁰ years, a universe of white dwarfs, neutron stars and brown dwarfs slowly cooling. If protons decay, which nobody has seen happen, they take ordinary matter with them, and the Black Hole Era runs from about 10³⁸ to 10¹⁰⁰ years. Black holes are the last things with any structure, and even they leak away through Hawking radiation. A black hole the mass of the Sun takes about 10⁶⁵ years to evaporate. One as heavy as a whole galaxy, 10¹¹ Suns, takes about 10⁹⁸. After that is the Dark Era: photons, neutrinos, a few stray particles and almost unimaginable amounts of empty space.

Numbers like these stop meaning anything to the human brain at around the third exponent, so here’s a way to feel them. Squash the entire 13.8-billion-year history of the universe so far into a single day. On that clock, stars keep shining for about another twenty years. The Degenerate Era then lasts about 2 × 10¹⁷ years, ten million times longer than the real universe has existed. And the last black holes evaporate after about 10⁸⁷ years. Even after compressing all of cosmic history into one day, the universe’s remaining lifespan is still a number with no everyday comparison.

§One parameter decides whether it rips

Heat death assumes dark energy behaves like Einstein’s cosmological constant: a fixed energy density for every cubic metre of space, forever. Cosmologists describe dark energy by its ratio of pressure to energy density, a number called $w$ :

p = w\,\rho c^2 \qquad\Rightarrow\qquad \rho \propto a^{-3(1+w)}

Here $a$ is the scale factor, the size of the universe relative to today. The second equation is the one worth staring at. Ordinary matter has $w = 0$ , so its density falls as $a^{-3}$ : double the size, divide by eight. Anything with $w < -1/3$ speeds up the expansion. Exactly $w = -1$ makes the exponent zero: the density stays the same no matter how much space grows. That’s the cosmological constant, and it gives a universe that accelerates gently forever while everything in it cools off.

Now push $w$ below −1. The exponent turns positive, and the more space expands, the denser the dark energy gets. Physicists call this phantom energy, and in 2003 Robert Caldwell, Marc Kamionkowski and Nevin Weinberg worked out what it does.^[2]2. Robert Caldwell, Marc Kamionkowski & Nevin Weinberg, Phantom Energy and Cosmic Doomsday (2003). The 22-billion-year figure is from their text for w = −3/2; their table uses slightly different inputs and gives a later date. The order of the destruction is the same either way. The density runs away to infinity in a finite time, and on the way there it overpowers every force holding anything together. They gave a neat approximation for how long we’d have:

t_{\text{rip}} - t_0 \approx \frac{2}{3}\,|1+w|^{-1}\,H_0^{-1}\,(1-\Omega_m)^{-1/2}

Plug in their example of $w = -3/2$ , a Hubble time $H_0^{-1}$ of about 14 billion years, and matter making up 30 percent of the total. You get 4/3 × 14 × 1.2, or roughly 22 billion years. The order of events is what makes the Big Rip famous. The Milky Way comes apart about 60 million years before the end. The Solar System lets go of its planets about three months before. The Earth explodes about half an hour before. Atoms come apart in the final 10⁻¹⁹ of a second.

That’s the scariest ending, and its whole case rests on one inequality: $w$ has to stay below −1. So the obvious question is what the measurements say.

§DESI says the accelerator may be easing off

The Dark Energy Spectroscopic Instrument maps the positions of galaxies and quasars to measure baryon acoustic oscillations, a frozen ripple from the early universe with a known physical size. Seen at different distances, it works as a ruler for how the expansion has changed over time. In March 2025, using its first three years of data on more than 14 million objects, DESI reported something nobody expected. A dark energy that changes over time fit better than a cosmological constant, at 3.1σ combining DESI with the cosmic microwave background. Adding supernova measurements, it rose to between 2.8σ and 4.2σ, depending on which supernova catalogue was used.^[3]3. DESI Collaboration, DESI DR2 Results II: Measurements of Baryon Acoustic Oscillations and Cosmological Constraints, Phys. Rev. D 112, 083515 (2025).

The models usually fitted to this write dark energy’s behaviour as a simple straight-line function of the scale factor:

w(a) = w_0 + w_a\,(1 - a)

The preferred region has $w_0 > -1$ and $w_a < 0$ . Here’s what that means in plain terms. Take the fit from the recalibrated Dark Energy Survey supernovae combined with DESI and the CMB: $w_0 = -0.803$ , $w_a = -0.72$ .^[4]4. Popovic et al., The Dark Energy Survey Supernova Program: A Reanalysis of Cosmology Results and Evidence for Evolving Dark Energy with an Updated Type Ia Supernova Calibration, MNRAS (2026). Today, $w$ sits above −1. In the distant past it was around −1.5, phantom territory. Dark energy, on this reading, was once stronger and has been weakening. That’s the reverse of what a Big Rip needs. If it holds, the acceleration isn’t winning. It’s tiring.

Now the caveats, and they’re serious. First, 3σ is not discovery. Physics uses 5σ for a reason, and 3σ results have a long record of fading away. Second, the most dramatic number has already come down. The 4.2σ figure came from the Dark Energy Survey’s five-year supernova sample. When that team recalibrated their data, with better cross-calibration against white dwarfs, a retrained light-curve model and a fixed numerical approximation, the significance dropped to 3.2σ. The authors described the result as equivalent to odds of about 5 to 1, and called it only a weak preference. Third, in July 2026 DESI published a new analysis of the Lyman-alpha forest, the absorption patterns in quasar light. Its central value moves back toward the cosmological constant, and the preference for evolving dark energy is now 2.7σ with the CMB, 3.2σ with supernovae too.^[5]5. DESI Collaboration, DESI DR2 Results IV: Alcock-Paczyński Measurements from the Lyman Alpha Forest and Cosmological Constraints (July 2026), and the collaboration’s own summary, New DESI DR2 Lyman-alpha Results Shed Light on Dark Energy. On the full survey’s timing: DESI Reaches Mapping Milestone (April 2026). The full five-year survey, which already counts more than 47 million galaxies and quasars, is due to report on dark energy in 2027.

And a caveat people skip: $w_0 + w_a(1-a)$ is a fitting function, not a law of nature. It’s a straight line drawn through data at redshifts we can see. Taken literally into the future, the fit above puts $w$ above −1/3 by the time the universe has doubled in size, which would switch the acceleration off. That’s a fun calculation and a meaningless forecast. You don’t predict the next decade of a stock from a line through the last one.

Some theorists have gone further anyway. Hoang Nhan Luu, Yu-Cheng Qiu and Henry Tye fit the DES and DESI data with a model in which an ultralight field sits on top of a negative cosmological constant. In that model the expansion eventually reverses into a Big Crunch, and their benchmark gives the universe a total lifespan of about 33 billion years.^[6]6. Hoang Nhan Luu, Yu-Cheng Qiu & S.-H. Henry Tye, The Lifespan of our Universe (2025), accepted by JCAP. The 33 billion years is a total lifespan, so it counts the 13.8 billion already gone. That’s one model with loose parameters, not a forecast. It still shows how much a few sigma can rearrange the options. The ending that sounded most like science fiction, the crunch, is back on the table. The ending that sounded most dramatic, the rip, is the one the new data push hardest against.

§The ending that needs no dark energy at all

The third candidate doesn’t care about any of this. Vacuum decay comes from the Higgs field, and it’s the one ending particle physics hands you rather than cosmology.

The Higgs field has a potential, a curve of energy against field strength, and our universe sits in a valley of it. The question is whether that valley is the lowest point anywhere, or just a dip on a hillside with a deeper valley somewhere at enormous field values. The answer depends on how the Higgs self-coupling changes with energy. That in turn depends sensitively on two measured masses: the Higgs, at about 125 GeV, and the top quark. When the Higgs was found, a 2012 analysis excluded full stability at 98 percent confidence for Higgs masses below 126 GeV.^[7]7. Degrassi et al., Higgs mass and vacuum stability in the Standard Model at NNLO (2012), which excluded absolute stability at 98% confidence for a Higgs lighter than 126 GeV. We appear to live right next to the boundary, on the metastable side. Our vacuum isn’t the bottom. It’s a ledge.

A ledge can be left by quantum tunnelling. Somewhere, at some moment, a small region tunnels into the lower state. It forms a bubble of true vacuum, which grows at close to the speed of light and rewrites the laws of particle physics inside it.^[8]8. The wall reaches nearly the speed of light almost at once, because the energy released by the true vacuum dwarfs the barrier it tunnelled through. See “False vacuum”, following Sidney Coleman & Frank De Luccia, “Gravitational effects on and of vacuum decay”, Physical Review D 21, 3305 (1980). Nothing could see it coming, because the warning would travel no faster than the wall. In 2018 Anders Andreassen, William Frost and Matthew Schwartz did what they called the first complete calculation of the lifetime. Their answer was 10¹³⁹ years, with a 95 percent lower bound of 10⁵⁸.^[9]9. Anders Andreassen, William Frost & Matthew Schwartz, Scale Invariant Instantons and the Complete Lifetime of the Standard Model, Phys. Rev. D 97, 056006 (2018). Most of the spread comes from the measured top quark mass and the strong coupling constant.

Two things are worth saying about that. First, it assumes the Standard Model holds all the way up to very high energies, with no new particles in between. Any new physics could stabilise the vacuum or make it worse, and we have no idea which. Second, look at the number. 10¹³⁹ years is 10³⁹ times longer than the whole Black Hole Era. If the estimate is right, vacuum decay isn’t the thing that ends the universe. It’s the thing that cleans up a universe that ended long before.

§Terrifying is the wrong word

Which brings me to the hook I don’t accept. I get why these endings are pitched as terrifying. Earth exploding half an hour before the Rip is a great image. But fear is a response to a threat, and a threat needs someone who could plausibly face it. Take even the most aggressive scenario on the board, the one current data disfavour. It plays out over more than twenty billion years, several times longer than the Earth has existed. The others play out on timescales where “long” stops being a useful word.

What I actually feel reading this material is closer to vertigo, the good kind. A big part of that is the method. We are working out the last chapter of everything from the size of a ripple in galaxy positions, the brightness of a couple of thousand exploding stars, and the mass of one particle, measured at CERN to a fraction of a percent. The difference between a gentle fade and a violent rip comes down to whether one number sits a little above or below −1. And that number moved this year, because someone re-checked a calibration against white dwarfs.

The real story of 2025 and 2026 isn’t that the ending got scarier. It’s that the ending stopped being settled. For twenty years the answer was a cosmological constant and a slow fade, and that made a good enough story that it started to feel like fact. Now there’s a 3σ crack in it that could close up in 2027 or widen into the biggest result about dark energy since 1998. I’d rather live in the universe where that question is open than in one where we already knew.

^Fred Adams & Gregory Laughlin, A Dying Universe: The Long Term Fate and Evolution of Astrophysical Objects, Reviews of Modern Physics 69 (1997). The era boundaries and the black-hole evaporation scaling, roughly 10⁸³ years for a million-solar-mass hole and growing as the cube of the mass, are theirs. Their proton lifetime of about 10³⁷ years is an assumption, not a measurement.
^Robert Caldwell, Marc Kamionkowski & Nevin Weinberg, Phantom Energy and Cosmic Doomsday (2003). The 22-billion-year figure is from their text for w = −3/2; their table uses slightly different inputs and gives a later date. The order of the destruction is the same either way.
^DESI Collaboration, DESI DR2 Results II: Measurements of Baryon Acoustic Oscillations and Cosmological Constraints, Phys. Rev. D 112, 083515 (2025).
^Popovic et al., The Dark Energy Survey Supernova Program: A Reanalysis of Cosmology Results and Evidence for Evolving Dark Energy with an Updated Type Ia Supernova Calibration, MNRAS (2026).
^DESI Collaboration, DESI DR2 Results IV: Alcock-Paczyński Measurements from the Lyman Alpha Forest and Cosmological Constraints (July 2026), and the collaboration’s own summary, New DESI DR2 Lyman-alpha Results Shed Light on Dark Energy. On the full survey’s timing: DESI Reaches Mapping Milestone (April 2026).
^Hoang Nhan Luu, Yu-Cheng Qiu & S.-H. Henry Tye, The Lifespan of our Universe (2025), accepted by JCAP. The 33 billion years is a total lifespan, so it counts the 13.8 billion already gone.
^Degrassi et al., Higgs mass and vacuum stability in the Standard Model at NNLO (2012), which excluded absolute stability at 98% confidence for a Higgs lighter than 126 GeV.
^The wall reaches nearly the speed of light almost at once, because the energy released by the true vacuum dwarfs the barrier it tunnelled through. See “False vacuum”, following Sidney Coleman & Frank De Luccia, “Gravitational effects on and of vacuum decay”, Physical Review D 21, 3305 (1980).
^Anders Andreassen, William Frost & Matthew Schwartz, Scale Invariant Instantons and the Complete Lifetime of the Standard Model, Phys. Rev. D 97, 056006 (2018). Most of the spread comes from the measured top quark mass and the strong coupling constant.