Eternal Inflation

A race between two motions of φ

During slow-roll inflation, the inflaton field φ is doing two things at once. Classically, it is rolling down its potential V(φ) at a slow-roll velocity

φ̇_classical = − V′(φ) / (3 H)

Quantum mechanically, it is undergoing stochastic fluctuations of magnitude

δφ_quantum ~ H / (2π) per Hubble time

The two effects compete. In a Hubble time, the classical motion moves φ down by Δφ_classical = φ̇/H, while the quantum jitter shifts it randomly by ±H/(2π). The relevant ratio is

R = δφ_quantum / Δφ_classical = H² / (2π |φ̇|)

When R < 1, classical motion wins and the field reliably drifts toward the minimum. When R > 1, quantum motion dominates and the field jitters back up the potential with appreciable probability. The condition for eternal inflation is R ≳ 1.

When is R big enough for eternal inflation?

Slow-roll relations give H² = V/(3M_Pl²) and φ̇ = −V′/(3H), so

R = H² / (2π |φ̇|) ~ H³ / (2π |V′|) = (1/2π) · (V / (3 M_Pl²))^(3/2) · (3 M_Pl² / V′)

Combine with the slow-roll parameter ε = (M_Pl²/2)(V′/V)²:

R² ~ V / (12 π² M_Pl⁴ · ε)

Eternal inflation requires R > 1, which gives

V / (12 π² M_Pl⁴) ≳ ε

For high-scale inflation (V ~ M_Pl⁴ ε with ε ~ 0.01), R can be of order 0.1 — borderline eternal. For very flat plateaus (V/M_Pl⁴ ≫ ε), R ≫ 1 and eternal inflation is robust. The condition becomes weaker for higher inflation scales, but is generically satisfied for plateau-type potentials with very small ε.

Self-reproduction: the inflating background never depletes

Imagine starting with one Hubble volume in slow-roll inflation. In a Hubble time:

1. The volume's physical size grows by e (one e-fold).
2. It splits into ~e³ ≈ 20 sub-Hubble-volumes, each still inflating.
3. In each sub-volume, the inflaton has fluctuated by ±H/(2π), randomly.
4. In some fraction P_up of sub-volumes, the field fluctuates UP — the field reaches a higher value of V, with smaller ε, and these sub-volumes are 'more inflating' than the parent.
5. In some fraction P_down, the field rolls down past the end of inflation and reheats — these become bubble universes.
6. The number of inflating sub-volumes grows as e³ × P_up per Hubble time.

If e³ × P_up > 1 — equivalently, P_up > e⁻³ ≈ 0.05 — the inflating volume keeps growing. The condition is easily met for plateau potentials with R ≳ 0.5. Inflating regions reproduce faster than they decay into bubble universes; the multiverse becomes a fractal.

Bubbles of normal universe

Inside the eternally inflating background, bubbles of post-inflation universe nucleate continuously. There are two main mechanisms:

• False-vacuum tunneling (Vilenkin's original picture). If V(φ) has a metastable false vacuum, the field can tunnel through the barrier to the true vacuum via a Coleman-de Luccia instanton. A bubble of true vacuum nucleates and expands at near light speed.
• Slow-roll-driven nucleation (Linde's chaotic inflation picture). Even without a metastable vacuum, slow-roll regions where the field has happened to fluctuate downward can fall off the inflationary plateau, reheating and becoming bubble universes.

From inside any bubble, the bubble's wall is receding at near light speed — observers in the bubble see an apparently infinite, homogeneous, expanding universe. This is what we see, with one important caveat: inside a bubble, the universe is hyperbolic (negatively curved on average), but the observed flatness of our cosmos requires that our bubble be very old. The fact that we see flatness is consistent with bubble nucleation in models that allow for further slow-roll inside the bubble after nucleation.

A worked numerical example

Quantity	Value	Notes
Inflation scale H_inf	~10¹³ GeV	From r < 0.036 bound
Hubble time 1/H	~10⁻³⁷ s	One e-fold of inflation
Quantum fluctuation δφ	~10¹² GeV per Hubble time	H/(2π)
Classical drift φ̇/H	10¹¹ to 10¹² GeV (model-dep.)	√(2ε) M_Pl, with ε small
Eternal-inflation ratio R	0.5 to 10	Comparable to or larger than 1
Sub-volume reproduction rate	~e³ per Hubble time	~20 sub-volumes per e-fold
Bubble nucleation rate Γ	~e⁻³ to 10⁻⁵ per Hubble vol/time	Highly model-dependent
Net inflating volume growth	~e³(1 − Γ) per Hubble time	Continues forever if Γ < e³
Expected bubbles in 60 e-folds	~e¹⁸⁰ = 10⁷⁸	Each containing a possible cosmos

The numbers depend strongly on the inflation potential and model details. Generic plateau inflation has R ≳ 1, easily eternal. Specific small-field inflation models can avoid eternal inflation, but they require fine-tuning.

Multiverse pictures: ranked by speculation

Level (Tegmark)	What it means	Cosmological basis	Evidence
Level I	Beyond cosmological horizon	Standard inflation + observable scale	Required if inflation produces uniform CMB
Level II	Bubble universes with different vacua	Eternal inflation + landscape	Predicted; no direct evidence
Level III	Many-worlds quantum branches	Quantum mechanics (Everett)	Speculative interpretation of QM
Level IV	All mathematical structures exist	None (Platonic)	Philosophical proposal

Eternal inflation is the Level II multiverse, the one most directly grounded in mainstream physics. It is not an additional hypothesis layered on inflation — it is a generic consequence of slow-roll inflation. To avoid it, you need to either (1) find an inflation model with R < 1 by fine-tuning, (2) introduce dynamics that terminate inflation before eternal regime sets in, or (3) reject the entire inflationary picture. Most cosmologists accept some version of Level II eternal inflation, with deep disagreement about its physical interpretation.

The measure problem: why we can't extract probabilities

An eternally inflating universe produces infinitely many bubble universes. Even if we knew the physics of each — say, that 10% of bubbles have a cosmological constant compatible with our value — we cannot in general compute probabilities because of infinities.

Concretely, ask: 'What is the probability that an observer sees Λ = 10⁻¹²² in Planck units?' To answer it, you need to count observers in bubbles with that Λ relative to observers in bubbles with different Λ. But there are infinite numbers of both. The ratio depends on which observers you count first.

Different 'measures' on the multiverse give different answers:

• Proper time cut-off. Stop the universe at some proper time and count finite quantities. Gives a 'youngness paradox': observers should overwhelmingly find themselves in young, low-temperature bubbles. Conflicts with observation.
• Volume cut-off (comoving). Count using comoving coordinates. Tends to overweight late-nucleated bubbles.
• Causal-patch / causal-diamond measure. Count only inside the future light cone of a fixed point. Eliminates infinities cleanly. Gives predictions that match observation for Λ but is hard to motivate from first principles.
• Pocket-based measure (Vilenkin, Garriga). Average over types of bubbles weighted by some prior.

No consensus. The measure problem is widely seen as the most pressing technical obstacle to extracting predictions from eternal inflation. Without a measure, we cannot say 'inflation predicts our Λ to be small' — we can only say 'inflation produces a multiverse in which small Λ is one possibility'.

History: Vilenkin 1983, Linde 1986, and the rise of the multiverse

The story starts with Coleman and de Luccia's 1980 paper on quantum tunneling in de Sitter spacetime. They showed that false-vacuum bubble nucleation in inflation produces an inhomogeneous, fractured spacetime — exactly the 'graceful exit' problem of Guth's original 1981 inflation paper.

Alexander Vilenkin in 1983 (Phys. Rev. D 27, 2848) recognised the flip side: while bubbles nucleate, the inflating background between them reproduces itself exponentially fast. The result is an inflating universe that never globally ends, peppered with bubble universes that locally reheat into normal cosmologies. He used this in 1983 to propose a 'tunneling wave function of the universe' that selects values of the cosmological constant.

Andrei Linde, working in parallel on his own chaotic inflation model, came to the same conclusion in 1986 (Phys. Lett. B 175, 395) from a different angle: even smooth slow-roll inflation has the same self-reproducing property because of quantum fluctuations. His paper introduced the phrase 'eternally existing self-reproducing inflationary universe' and gave the first explicit calculation of the eternal-inflation regime in slow-roll models.

For about a decade these ideas were treated as theoretical curiosities. Then the string theory landscape — Bousso and Polchinski's 2000 'multiverse with many vacua' picture, refined by Susskind and others in 2003 — gave eternal inflation a concrete mechanism: an eternally inflating universe populates the landscape's 10⁵⁰⁰ vacua over cosmic time. The 1998 supernova measurement of dark energy at exactly the value Weinberg predicted from anthropic reasoning seemed to validate the framework. Today eternal inflation + landscape is the standard speculative cosmology of fundamental physics, with sharp critics (Steinhardt, Penrose, others) and enthusiastic supporters (Linde, Vilenkin, Susskind, Bousso).

The critique: is it physics?

Critics have raised serious objections:

• Untestability. Bubbles outside our own are causally disconnected; no signal can pass between them. The multiverse as a whole is not directly observable.
• Measure problem. Without a consistent measure, the multiverse cannot make falsifiable probabilistic predictions.
• Initial conditions. Eternal inflation requires that inflation started somewhere; that requirement smuggles back the 'initial conditions' problem that inflation was supposed to solve.
• Penrose's entropy concern. Roger Penrose argues that the low-entropy initial state required for inflation is itself extremely improbable, and eternal inflation does not solve this problem; it merely defers it.
• Steinhardt's bubble nucleation rate. Paul Steinhardt has argued that the rate of large bubble nucleation in eternal inflation is so high that observable consequences should appear in the CMB. The fact that we see no such signatures is, in his view, evidence against typical inflation models.

Proponents respond: eternal inflation is a logical consequence of slow-roll inflation, which is independently supported by Planck-quality CMB observations. It is not an additional speculative layer — it is what slow-roll inflation predicts when you take it seriously. Rejecting eternal inflation requires rejecting the inflation picture itself or finding fine-tuned models that avoid the eternal regime.

Common pitfalls

• Treating eternal inflation as an alternative to inflation. Eternal inflation is what generic slow-roll inflation becomes. It is not a competing theory.
• Assuming all bubble universes have the same physics. In the string-landscape picture, different bubbles can have different cosmological constants, gauge couplings, particle masses. Whether this is physically realised depends on what string theory actually says.
• Confusing eternal inflation with eternal inflation in the past. A theorem by Borde, Guth, and Vilenkin (2003) shows that any inflating spacetime must have had a beginning — eternal inflation is eternal forward, not backward.
• Thinking the multiverse explains the cosmological constant. The anthropic argument requires a measure; without one, the multiverse doesn't predict Λ — it just allows it.
• Looking for bubble collisions in the wrong way. Direct CMB searches for bubble-collision signatures (circular cold/hot spots) are model-dependent. Null results constrain bubble nucleation rates but don't rule out eternal inflation generally.