Slow-Roll Inflation and the Inflaton Field

Slow-roll inflation is the deepest idea in modern cosmology that still fits on a single line of physics: a scalar field whose potential energy vastly outweighs its kinetic energy behaves, for a while, like a cosmological constant, and a cosmological constant makes space expand exponentially. That field is the inflaton $\phi$ , minimally coupled to gravity, and while it creeps down a nearly flat potential $V(\phi)$ its stress-energy has equation of state $w \approx -1$ . The scale factor runs away as $a(t)\propto e^{Ht}$ — a quasi-de Sitter epoch that, in a fraction of a second, stretches a sub-atomic patch into a region larger than the observable universe. Everything in this article is a consequence of one dynamical simplification. When the potential is flat enough, the field equation $\ddot\phi + 3H\dot\phi + V' = 0$ is dominated by Hubble friction: the acceleration $\ddot\phi$ is negligible and the field drifts at terminal velocity $\dot\phi\approx -V'/(3H)$ , while the Friedmann equation reduces to $H^2\approx V/(3M_{\rm Pl}^2)$ . Two dimensionless slow-roll parameters, $\epsilon$ and $\eta$ , built from derivatives of $V$ , decide whether this holds. Inflation runs while they are small and ends when $\epsilon$ climbs to one. But inflation is more than a fix for the horizon, flatness, and monopole problems. It is the leading theory of where cosmic structure came from. Quantum fluctuations of the inflaton are stretched beyond the shrinking comoving Hubble radius and frozen into a nearly scale-invariant, Gaussian, adiabatic spectrum of curvature perturbations, with amplitude and tilt fixed by $V$ and its slopes. The measured spectral index $n_s\approx 0.965$ and the bound $r\lesssim 0.036$ are among the sharpest tests any theory of the early universe faces. We build the background equations, define the slow-roll hierarchy, derive the observables, and grind through the canonical models.

Key takeaways

Cosmic inflation is a burst of accelerated, quasi-de Sitter expansion driven by the inflaton $\phi$ , whose potential energy $V(\phi)$ dominates its kinetic energy while it slowly rolls down a nearly flat potential.
The dynamics collapse to $H^2\approx V/(3M_{\rm Pl}^2)$ plus a friction-limited field equation $3H\dot\phi\approx -V'$ , valid whenever the slow-roll parameters $\epsilon$ and $|\eta|$ are both $\ll 1$ .
$\epsilon_V=\tfrac12 M_{\rm Pl}^2 (V'/V)^2$ measures the potential's steepness and controls the deviation from perfect de Sitter; inflation ends when $\epsilon_V\simeq 1$ .
Total expansion is counted in $e$ -folds $N=\int H\,\mathrm{d}t$ ; erasing the horizon and flatness problems needs $N\gtrsim 50$ – $60$ .
Vacuum fluctuations frozen at horizon exit seed a nearly scale-invariant curvature spectrum with tilt $n_s-1=-6\epsilon_V+2\eta_V$ and tensor-to-scalar ratio $r=16\epsilon_V$ — the numbers the CMB measures.
When slow roll fails the inflaton oscillates and decays, reheating the universe and igniting the hot Big Bang.
Key idea: A scalar field whose potential energy dominates its kinetic energy acts as a nearly constant vacuum energy, driving quasi-exponential expansion; the flatter the potential (small $\epsilon$ , $\eta$ ), the closer to perfect de Sitter and the longer inflation lasts.

Where this lives today

Slow-roll inflation is now a precision science practiced at the boundary of two experimental frontiers. On the scalar side, the Planck satellite fixed the spectrum to $n_s=0.9649\pm0.0042$ and $\mathcal{A}_s\approx2.1\times10^{-9}$ , and its detection of a red tilt ( $n_s<1$ at more than $8\sigma$ ) is a genuine confirmation of a generic inflationary prediction — a perfectly scale-invariant spectrum is now excluded. On the tensor side, the race is to detect the primordial B-mode polarization sourced by inflationary gravitational waves; BICEP/Keck combined with Planck has driven the bound to $r\lesssim0.036$ , and next-generation experiments — the Simons Observatory, CMB-S4, and the LiteBIRD satellite — target $r\sim10^{-3}$ , precisely the regime where Starobinsky and $\alpha$ -attractor models predict a signal. A detection would be a direct measurement of the inflationary energy scale $V^{1/4}\sim10^{16}$ GeV, the only known probe of grand-unification-scale physics. Meanwhile the open problems remain sharp: the initial-conditions and measure problems of eternal inflation, the trans-Planckian censorship and swampland conjectures that question whether the flat potentials inflation needs can arise in quantum gravity at all, and the search for primordial non-Gaussianity — whose amplitude $f_{\rm NL}$ single-field slow roll predicts to be unobservably small, so that any detection would overturn the simplest picture entirely.

The physical picture

Picture a ball rolling on a nearly flat, gently tilted surface immersed in thick honey. Hubble expansion is the honey: instead of accelerating, the ball settles to a slow terminal velocity fixed by the local slope. Because it barely moves, its height — the potential energy $V$ — stays almost constant, and a constant energy density gravitates exactly like a cosmological constant, driving exponential expansion. The universe keeps inflating as long as the surface stays flat, meaning the slope-to-height ratio ( $\sim\sqrt\epsilon$ ) and the curvature ( $\eta$ ) remain tiny. When the ball finally reaches the steep wall near the minimum, $\epsilon$ grows to one, the honey can no longer hold it, and it plunges and oscillates — that plunge is reheating. The first figure is that landscape.

Figure

The inflaton potential V(φ): on the flat plateau ε and η are both ≪ 1, so V ≈ const gives H² ≈ V/(3M_Pl²) and w ≈ −1 (quasi-de Sitter); the field creeps at terminal velocity φ̇ ≈ −V′/(3H); inflation ends where ε = 1, after which φ oscillates about the minimum and reheats the universe

Now think in terms of the comoving Hubble radius $1/(aH)$ . During inflation $aH$ grows, so this radius shrinks, and a fixed comoving scale that started inside the horizon is pushed outside it. Fluctuations frozen at horizon crossing become the classical seeds of galaxies; after inflation the Hubble radius grows again and scales re-enter, which is why the CMB shows correlations across regions that were never in causal contact in a purely decelerating universe. The near-constancy of $H$ during slow roll is exactly what makes the spectrum nearly scale-invariant — modes of very different wavelength all exit the horizon when $H$ has almost the same value, so they carry almost the same amplitude. The second figure is the payoff: the concrete numbers that turn a Lagrangian into a falsifiable prediction.

Figure

The observational punchline in the (n_s, r) plane: the Planck+BICEP/Keck data prefer n_s = 0.9649 ± 0.0042 and bound r ≲ 0.036, which excludes the quadratic model V = ½m²φ² (r ≈ 0.13 at N = 60) but is comfortably fit by the Starobinsky R² plateau (r ≈ 0.003)

The core equations

Inflation begins from an action. The inflaton is a single scalar minimally coupled to Einstein gravity, and the whole theory follows from varying this functional — the gravitational part gives the Friedmann equations, the matter part gives the field's stress-energy and equation of motion.

S = \int \mathrm{d}^4x\,\sqrt{-g}\left[\frac{M_{\rm Pl}^2}{2}R - \frac{1}{2}g^{\mu\nu}\partial_\mu\phi\,\partial_\nu\phi - V(\phi)\right] \qquad (1)

where: $M_{\rm Pl}$ — reduced Planck mass, $M_{\rm Pl}=(8\pi G)^{-1/2}\approx 2.4\times10^{18}\,$ GeV; $R$ — Ricci scalar of the metric $g_{\mu\nu}$ ; $\phi$ — the inflaton field; $V(\phi)$ — its self-interaction potential energy density; $g$ — determinant of the metric. Reading: gravity plus a scalar with a potential, nothing more exotic. For a homogeneous field $\phi=\phi(t)$ the stress-energy is that of a perfect fluid. The kinetic term supplies pressure, the potential supplies tension; their competition sets the equation of state and hence whether gravity pushes or pulls.

\rho_\phi = \tfrac{1}{2}\dot\phi^2 + V(\phi),\qquad p_\phi = \tfrac{1}{2}\dot\phi^2 - V(\phi) \qquad (2)

where: $\rho_\phi$ — energy density; $p_\phi$ — pressure; $\tfrac12\dot\phi^2$ — kinetic energy density (gradients vanish for a homogeneous field); $w\equiv p_\phi/\rho_\phi$ — equation-of-state parameter. Reading: when $\tfrac12\dot\phi^2\ll V$ the pressure is $\approx -V\approx -\rho$ , so $w\to -1$ and the field mimics vacuum energy. Varying the matter action for $\phi$ in an expanding FLRW background gives its equation of motion — a damped oscillator equation in which cosmic expansion supplies the damping.

\ddot\phi + 3H\dot\phi + V'(\phi) = 0 \qquad (3)

where: $H\equiv \dot a/a$ — Hubble parameter; $3H\dot\phi$ — Hubble friction (expansion drains the field's motion); $V'\equiv\mathrm{d}V/\mathrm{d}\phi$ — potential gradient (the restoring force). Reading: the field rolls like a marble in syrup, its speed set by the balance of slope against friction. The gravitational side of the action gives the two Friedmann equations for a flat FLRW universe sourced by the inflaton fluid.

H^2 = \frac{1}{3M_{\rm Pl}^2}\left(\tfrac12\dot\phi^2 + V\right),\qquad \frac{\ddot a}{a} = -\frac{1}{3M_{\rm Pl}^2}\left(\dot\phi^2 - V\right) \qquad (4)

where: $\ddot a/a$ — acceleration of the scale factor, positive (inflation) when $V>\dot\phi^2$ ; $H^2$ — squared expansion rate, sourced by the total energy density. Reading: potential energy accelerates the expansion, kinetic energy decelerates it. The slow-roll parameters package "how flat is the potential" into two dimensionless numbers. They are the small quantities in which every inflationary observable is expanded.

\epsilon_V \equiv \frac{M_{\rm Pl}^2}{2}\left(\frac{V'}{V}\right)^2,\qquad \eta_V \equiv M_{\rm Pl}^2\,\frac{V''}{V} \qquad (5)

where: $\epsilon_V$ — first slow-roll parameter (steepness), requires $\epsilon_V\ll1$ during inflation and hits $\epsilon_V\simeq1$ at the end; $\eta_V$ — second slow-roll parameter (curvature), $|\eta_V|\ll1$ keeps $\ddot\phi$ negligible; $V''=\mathrm{d}^2V/\mathrm{d}\phi^2$ — the inflaton's effective mass-squared. Reading: $\epsilon$ is about slope, $\eta$ is about curvature; both must be small. When both parameters are small, equations (3) and (4) simplify to their slow-roll forms — the two relations that make inflation analytically tractable.

3H\dot\phi \approx -V',\qquad H^2 \approx \frac{V}{3M_{\rm Pl}^2} \qquad (6)

where: $\approx$ — leading order in $\epsilon_V,|\eta_V|\ll1$ . Reading: drop $\ddot\phi$ against friction and drop the kinetic energy against $V$ ; the field's speed and the expansion rate are then read straight off the potential. The amount of inflation is counted in $e$ -folds. Converting the time integral to a field integral with the slow-roll relations turns $N$ into a pure functional of the potential.

N \equiv \int_{t}^{t_{\rm end}} H\,\mathrm{d}t = \int_{\phi_{\rm end}}^{\phi} \frac{V}{M_{\rm Pl}^2\,V'}\,\mathrm{d}\phi = \frac{1}{M_{\rm Pl}}\int_{\phi_{\rm end}}^{\phi}\frac{\mathrm{d}\phi}{\sqrt{2\epsilon_V}} \qquad (7)

where: $N$ — number of $e$ -foldings between field value $\phi$ and the end of inflation; $\phi_{\rm end}$ — field value where $\epsilon_V=1$ . Reading: flatter potentials (small $\epsilon_V$ ) accumulate more $e$ -folds per unit field range; observable scales left the horizon at $N\approx 50$ – $60$ . Finally, the observables. The amplitude, tilt, and tensor content of the primordial spectrum are all leading-order functions of $V$ evaluated at horizon crossing.

n_s - 1 = -6\epsilon_V + 2\eta_V,\qquad r = 16\epsilon_V,\qquad \mathcal{A}_s = \frac{1}{24\pi^2}\frac{V}{M_{\rm Pl}^4\,\epsilon_V} \qquad (8)

where: $n_s$ — scalar spectral index, measured $n_s\approx0.965$ ; $r$ — tensor-to-scalar ratio, bounded $r\lesssim0.036$ (BICEP/Keck + Planck); $\mathcal{A}_s$ — scalar power spectrum amplitude, $\approx 2.1\times10^{-9}$ (Planck). Reading: measure two numbers, $n_s$ and $r$ , and you have measured the first two derivatives of the inflaton potential. These observables are the coefficients of a nearly power-law curvature spectrum, quoted at a chosen pivot scale.

\mathcal{P}_\mathcal{R}(k) = \mathcal{A}_s\left(\frac{k}{k_*}\right)^{n_s-1} \qquad (9)

where: $\mathcal{P}_\mathcal{R}$ — dimensionless power spectrum of the comoving curvature perturbation $\mathcal{R}$ ; $k$ — comoving wavenumber; $k_*$ — pivot scale, typically $0.05\,\mathrm{Mpc}^{-1}$ . Reading: a tilt $n_s<1$ means slightly more power on large scales than small — the "red" spectrum the CMB sees.

Derivation AAccelerated expansion and the first slow-roll parameter

The goal is to prove that inflation, defined by $\ddot a>0$ , is exactly equivalent to the equation-of-state condition $w<-1/3$ and to the Hubble slow-roll parameter $\epsilon_H<1$ , and then to show that in the slow-roll regime this dynamical $\epsilon_H$ reduces to the purely potential quantity $\epsilon_V$ . This is what licenses us to test inflation using only $V$ and its derivatives, and it pins down precisely what "the end of inflation" means. We start from the acceleration equation and rewrite everything in terms of $H$ and its time derivative.

Step 1. The second Friedmann equation reads $\ddot a/a = -\tfrac{1}{6M_{\rm Pl}^2}(\rho+3p)$ , so $\ddot a>0$ demands $\rho+3p<0$ , i.e. $w<-1/3$ . Why: acceleration is not about energy being present but about it being sufficiently negative-pressure; ordinary matter and radiation ( $w=0,\,1/3$ ) always decelerate.

Step 2. Since $H=\dot a/a$ , differentiating gives $\ddot a/a=\dot H+H^2$ . Factoring out $H^2$ yields $\ddot a/a=H^2(1+\dot H/H^2)$ . Why: recasting acceleration in terms of $H$ removes the scale factor entirely and exposes the single combination $\dot H/H^2$ that controls everything.

Step 3. Define the Hubble slow-roll parameter $\epsilon_H\equiv -\dot H/H^2$ , so that $\ddot a/a = H^2(1-\epsilon_H)$ ; accelerated expansion is then simply $\epsilon_H<1$ , and $\epsilon_H\to0$ is exact de Sitter. Why: this makes "how close to de Sitter" a single measurable number, and identifies $\epsilon_H=1$ as the exact instant acceleration stops.

Step 4. Differentiate the Friedmann equation and use Klein–Gordon to find $\dot H=-\dot\phi^2/(2M_{\rm Pl}^2)$ , hence $\epsilon_H=\dot\phi^2/(2M_{\rm Pl}^2 H^2)$ . Why: this reveals the physical content of $\epsilon_H$ — it is the fraction of the total energy that is kinetic, so de Sitter is the limit of vanishing kinetic energy.

Step 5. Insert the slow-roll relations $\dot\phi\approx -V'/(3H)$ and $3H^2\approx V/M_{\rm Pl}^2$ to get $\epsilon_H\approx \tfrac{M_{\rm Pl}^2}{2}(V'/V)^2=\epsilon_V$ . Why: the dynamical and potential definitions coincide at leading order, so inflation ends when $\epsilon_V\simeq1$ — a condition read directly off the shape of $V$ , without solving the equations of motion.

Go deeper · Derivation B · 1 mine-folds and observables for quadratic chaotic inflation

Derivation Be-folds and observables for quadratic chaotic inflation

The goal is to carry the simplest complete model, $V=\tfrac12 m^2\phi^2$ , all the way from potential to prediction: compute $\epsilon_V$ and $\eta_V$ , find where inflation ends, integrate the number of $e$ -folds, invert to get the field value at horizon crossing, and read off $n_s$ and $r$ . This worked-through chain is the template for every other potential, and it exposes both the strength (a clean $n_s$ ) and the fatal weakness (too much $r$ ) of large-field monomial inflation. Take $V=\tfrac12 m^2\phi^2$ , so $V'=m^2\phi$ and $V''=m^2$ .

Step 1. The slow-roll parameters are $\epsilon_V=\tfrac{M_{\rm Pl}^2}{2}(m^2\phi/\tfrac12 m^2\phi^2)^2 = 2M_{\rm Pl}^2/\phi^2$ and $\eta_V = M_{\rm Pl}^2 m^2/(\tfrac12 m^2\phi^2)=2M_{\rm Pl}^2/\phi^2$ , so here $\epsilon_V=\eta_V$ . Why: both fall as $\phi^{-2}$ , so the potential looks flat only at large field values — inflation lives out at super-Planckian $\phi$ .

Step 2. Inflation ends when $\epsilon_V=1$ , giving $\phi_{\rm end}=\sqrt2\,M_{\rm Pl}$ . Why: this fixes the lower limit of the $e$ -fold integral and confirms the excursion reaches down to the Planck scale in field space.

Step 3. Integrate equation (7) with $V/V'=\phi/2$ : $N=\int_{\phi_{\rm end}}^{\phi}\phi/(2M_{\rm Pl}^2)\,\mathrm{d}\phi = (\phi^2-\phi_{\rm end}^2)/(4M_{\rm Pl}^2)$ . Why: the integral is elementary for a monomial, which is exactly why this model is the standard teaching example.

Step 4. Substitute $\phi_{\rm end}^2=2M_{\rm Pl}^2$ to get $N=\phi^2/(4M_{\rm Pl}^2)-\tfrac12$ , so $\phi^2=4M_{\rm Pl}^2(N+\tfrac12)$ and $\epsilon_V=1/(2N+1)$ . Why: this inverts the relation so every observable can be written as a function of $N$ alone, the quantity fixed by when a given CMB scale left the horizon.

Step 5. At $N=60$ : $\epsilon_V\approx0.0083$ , giving $n_s=1-4\epsilon_V=1-2/(N+\tfrac12)\approx0.967$ and $r=16\epsilon_V=8/(N+\tfrac12)\approx0.13$ . Why: the tilt sits right on the data, but $r\approx0.13$ overshoots the bound $r\lesssim0.036$ by nearly fourfold — which is why pure $m^2\phi^2$ inflation is now observationally dead.

Go deeper · Derivation C · 1 minThe curvature power spectrum from horizon crossing

Derivation CThe curvature power spectrum from horizon crossing

The goal is to show heuristically why the primordial curvature spectrum has amplitude $\mathcal{P}_\mathcal{R}\propto H^2/\dot\phi^2\propto V/\epsilon_V$ , so that flatter potentials at fixed energy produce larger perturbations, and to reproduce the boxed result that ties $\mathcal{A}_s$ to $V$ and $\epsilon_V$ . The argument has three moves: quantify the frozen field fluctuation in de Sitter, convert it to a curvature perturbation using the background clock, and re-express the answer through slow roll. We take each mode at the moment it crosses the horizon, $k=aH$ .

Step 1. A light scalar in quasi-de Sitter space acquires a nearly scale-invariant spectrum of fluctuations of amplitude $\delta\phi\sim H/2\pi$ per mode at horizon exit. Why: the expansion rate $H$ sets the only scale, so dimensional analysis alone fixes the size of the frozen fluctuation — this is the Gibbons–Hawking temperature of de Sitter reappearing as a field variance.

Step 2. The comoving curvature perturbation is $\mathcal{R}=-(H/\dot\phi)\,\delta\phi$ , the field fluctuation measured in units of how long the background takes to traverse $\delta\phi$ . Why: $\mathcal{R}$ is the quantity conserved on super-horizon scales, so freezing it at horizon exit lets us carry it untouched across the unknown reheating epoch to the CMB.

Step 3. Square and combine: $\mathcal{P}_\mathcal{R}=(H/\dot\phi)^2(H/2\pi)^2 = H^4/(4\pi^2\dot\phi^2)$ . Why: the transfer factor $H/\dot\phi$ is large when the field barely moves, so a nearly static inflaton — the flat-potential limit — is precisely what amplifies the perturbations.

Step 4. Use $\dot\phi^2=2M_{\rm Pl}^2 H^2\epsilon_H$ and then $H^2=V/(3M_{\rm Pl}^2)$ to get $\mathcal{P}_\mathcal{R}=\tfrac{1}{8\pi^2}H^2/(M_{\rm Pl}^2\epsilon_H)=\tfrac{1}{24\pi^2}V/(M_{\rm Pl}^4\epsilon_V)$ . Why: this is the boxed observable — the amplitude is set by the energy scale $V$ divided by the steepness $\epsilon_V$ , so measuring $\mathcal{A}_s\approx2.1\times10^{-9}$ fixes one combination of the inflaton potential.

Limiting cases

The slow-roll description is an expansion in $\epsilon$ and $\eta$ ; its regimes of validity and breakdown are laid out below.

Case	Result	Meaning
Perfect de Sitter	$\epsilon\to0,\ \eta\to0$	$V$ exactly flat, $\dot\phi\to0$ , $H$ constant, $a\propto e^{Ht}$ , $w=-1$ . An idealized limit — inflation would never end.
Slow-roll regime	$\epsilon\ll1,\ \|\eta\|\ll1$	Quasi-de Sitter: $H$ nearly constant, $\ddot\phi$ negligible. Observable CMB scales exit the horizon here, $\sim50$ – $60$ $e$ -folds before the end.
End of inflation	$\epsilon=1$	$\ddot a=0$ ; accelerated expansion ceases. Beyond this the field rolls fast, kinetic energy grows, and $w\to+1$ (kination) transiently.
Reheating / oscillation	$\epsilon,\,\|\eta\|\gtrsim1$	Slow roll fails; $\phi$ oscillates about the minimum, behaving on average like pressureless matter ( $w\approx0$ ) and decaying into radiation.

Common trapSmall ε and small η are the same condition

They are not. Two distinct smallness requirements are in play, and conflating them is the most common slow-roll error. $\epsilon_V\ll1$ (the potential is shallow relative to its own height) is what makes the expansion accelerate; it equals $-\dot H/H^2$ and reaching $\epsilon_V=1$ is what ends inflation. $|\eta_V|\ll1$ (the potential's curvature, i.e. the inflaton mass-squared, is small) is a separate condition that keeps $\ddot\phi$ negligible so the slow-roll trajectory persists over many $e$ -folds. A potential can easily have tiny $\epsilon$ but large $|\eta|$ — a nearly flat but strongly curved bump — in which case slow roll breaks even though the expansion is still accelerating. Both must be small for the standard approximation, but it is $\epsilon\to1$ , never $\eta$ , that signals the end.

Worked examples

1 · Quadratic inflation from potential to inflaton mass (fully worked)

Chaotic inflation with $V=\tfrac12 m^2\phi^2$ produces $N=60$ $e$ -folds on the observable scales. Find the field value at horizon crossing, the predicted $n_s$ and $r$ , and fix the inflaton mass $m$ from the measured amplitude $\mathcal{A}_s=2.1\times10^{-9}$ .

Step 1. Field value at horizon crossing. From the derivation, $\phi^2=4M_{\rm Pl}^2(N+\tfrac12)=4M_{\rm Pl}^2(60.5)=242\,M_{\rm Pl}^2$ , so $\phi=\sqrt{242}\,M_{\rm Pl}=\mathbf{15.6\,M_{\rm Pl}}$ . The excursion is strongly super-Planckian — the defining feature (and headache) of large-field models.

Step 2. Tilt and tensors. With $\epsilon_V=\eta_V=1/(2N+1)=1/121=0.00826$ , the tilt is $n_s=1-4\epsilon_V=1-0.0331=\mathbf{0.967}$ and $r=16\epsilon_V=16/121=\mathbf{0.132}$ . The tilt is a bullseye on the data; the tensor amplitude is not.

Step 3. Fix the mass from $\mathcal{A}_s$ . Using $\mathcal{A}_s=\tfrac{1}{24\pi^2}V/(M_{\rm Pl}^4\epsilon_V)$ with $V=\tfrac12 m^2\phi^2$ and $\epsilon_V=2M_{\rm Pl}^2/\phi^2$ gives $V/\epsilon_V=m^2\phi^4/(4M_{\rm Pl}^2)$ . Substituting $\phi^2=242\,M_{\rm Pl}^2$ : $\mathcal{A}_s=\tfrac{1}{24\pi^2}\,m^2(242)^2 M_{\rm Pl}^2/(4M_{\rm Pl}^4)$ , so $(m/M_{\rm Pl})^2=24\pi^2\mathcal{A}_s\cdot4/(242)^2=3.4\times10^{-11}$ .

Step 4. Therefore $m=5.8\times10^{-6}\,M_{\rm Pl}\approx\mathbf{1.4\times10^{13}\ GeV}$ . The inflaton is a remarkably light field on the Planck scale — its tiny mass is what keeps the potential flat enough to inflate. Result: $\phi\approx15.6\,M_{\rm Pl}$ , $n_s\approx0.967$ , $r\approx0.13$ (excluded), $m\approx1.4\times10^{13}$ GeV.

2 · The Starobinsky plateau that survives

The Starobinsky ( $R^2$ ) model behaves at large field like a plateau with $n_s\approx1-2/N$ and $r\approx12/N^2$ . Evaluate its predictions at $N=55$ and compare with the quadratic model above.

Step 1. Spectral index: $n_s=1-2/N=1-2/55=\mathbf{0.9636}$ , within Planck's $n_s=0.9649\pm0.0042$ .

Step 2. Tensor ratio: $r=12/N^2=12/3025=\mathbf{0.0040}$ , roughly $33\times$ smaller than the quadratic model's $0.13$ and far under the bound $r\lesssim0.036$ .

Step 3. Interpretation: the plateau flattens as $\phi$ grows, so $\epsilon_V$ falls much faster than in a monomial, suppressing $r$ while keeping the same $n_s$ . This is why plateau models — Starobinsky, Higgs inflation, $\alpha$ -attractors — are the survivors of the Planck+BICEP era. Result: $n_s\approx0.964$ , $r\approx0.004$ — comfortably allowed.

3 · The energy scale of inflation from a future r detection

Suppose B-mode surveys eventually measure $r=0.01$ . What inflationary energy scale $V^{1/4}$ does that imply?

Step 1. Combine $\mathcal{A}_s=\tfrac{1}{24\pi^2}V/(M_{\rm Pl}^4\epsilon_V)$ with $r=16\epsilon_V$ to eliminate $\epsilon_V$ : $V=\tfrac{3}{2}\pi^2 M_{\rm Pl}^4\,r\,\mathcal{A}_s$ .

Step 2. Insert $M_{\rm Pl}=2.4\times10^{18}$ GeV, $\mathcal{A}_s=2.1\times10^{-9}$ , $r=0.01$ : $V^{1/4}\approx\mathbf{1.0\times10^{16}\ GeV}$ , essentially the grand-unification scale.

Step 3. Interpretation: because $V^{1/4}\propto r^{1/4}$ , even a modest $r$ pins the inflaton potential to GUT-scale energies — the only known observational window onto physics near $10^{16}$ GeV. Result: $V^{1/4}\approx10^{16}\,(r/0.01)^{1/4}$ GeV.

Physics of the portrait

Inflation is the rare cosmological idea with a nearly datable birth: a puzzle about magnetic monopoles, a young particle theorist, and a late-night realization that a supercooled false vacuum would blow the universe up exponentially.

Origins · 1 minAlan Guth and how this was discovered

Alan Guth

born 1947 · the inflationary universe

Alan Guth was a postdoc working on grand unified theories, not cosmology, when in late 1979 he saw that the exponential expansion driven by a metastable "false vacuum" would simultaneously dilute unwanted magnetic monopoles and flatten the universe — he wrote "SPECTACULAR REALIZATION" in his notebook. His 1981 paper introduced the word inflation and framed the horizon and flatness problems as its motivation, but his original model (later called "old inflation") ended by first-order bubble nucleation that never completed — the "graceful exit" problem. The fix came from Andrei Linde and, independently, Andreas Albrecht and Paul Steinhardt in 1982, who replaced tunneling with a scalar field slowly rolling down a nearly flat potential — new inflation, and then Linde's chaotic inflation, which is the slow-roll picture developed here. Meanwhile Alexei Starobinsky had, in 1980, written down an $R^2$ model that inflates for entirely gravitational reasons and remains the best-fitting model today. The decisive turn was the recognition — by Mukhanov and Chibisov, and worked out at the 1982 Nuffield workshop by Guth, Hawking, Starobinsky, Bardeen, Steinhardt, Turner and others — that inflation does not merely smooth the universe but seeds its structure: quantum fluctuations of the inflaton become the density perturbations that grow into galaxies and imprint the CMB. That prediction, of a nearly scale-invariant spectrum with a slight red tilt, was confirmed to exquisite precision by WMAP and Planck.

History

1980

— Starobinsky proposes an R² model that inflates from quantum-gravitational corrections

1981

— Guth names inflation and links it to the horizon, flatness, and monopole problems

1982

— Linde, and Albrecht and Steinhardt, introduce slow-roll (new) inflation, solving the graceful-exit problem

1982

— the Nuffield workshop establishes that inflaton fluctuations seed a nearly scale-invariant spectrum

1983

— Linde proposes chaotic inflation with V = ½m²φ², the archetypal large-field model

2018

— Planck measures n_s = 0.9649 ± 0.0042, favoring plateau models over monomials

Going further

Texts. Ryden, Introduction to Cosmology, 2nd ed., ch. 11 (Inflation) · Dodelson & Schmidt, Modern Cosmology, 2nd ed., ch. 6 · Carroll & Ostlie, An Introduction to Modern Astrophysics, 2nd ed., ch. 30 · Baumann, Cosmology (2022), ch. 4 (Cosmic Inflation) · Liddle & Lyth, Cosmological Inflation and Large-Scale Structure · Weinberg, Cosmology (2008), ch. 4 & 10 · Mukhanov, Physical Foundations of Cosmology, ch. 8. The key relations used throughout are: $\epsilon_H=\dot\phi^2/(2M_{\rm Pl}^2H^2)$ and its slow-roll match to $\epsilon_V$ ; the $m^2\phi^2$ results $N=\phi^2/(4M_{\rm Pl}^2)-1/2$ , $\phi=15.6\,M_{\rm Pl}$ , $n_s=0.967$ , $r=0.13$ at $N=60$ , and $m=1.4\times10^{13}$ GeV from $\mathcal{A}_s$ ; the Starobinsky formulas $n_s=1-2/N$ and $r=12/N^2$ ; the energy-scale relation $V^{1/4}=1.0\times10^{16}(r/0.01)^{1/4}$ GeV; and the scalar spectrum $\mathcal{P}_\mathcal{R}=H^4/(4\pi^2\dot\phi^2)$ . The quoted constants — $M_{\rm Pl}\approx2.4\times10^{18}$ GeV, $\mathcal{A}_s\approx2.1\times10^{-9}$ , $n_s=0.9649\pm0.0042$ , $r\lesssim0.036$ , $|\Omega_k|\lesssim0.005$ — are Planck 2018 / BICEP-Keck constraints and CODATA-derived values.

Prerequisites. Friedmann Equations and FLRW Cosmology · Scalar Field Theory in Curved Spacetime · Equation of State and Cosmic Fluids · Thermal History of the Universe. Leads to. Reheating and the Onset of the Hot Big Bang · Primordial Power Spectrum and CMB Anisotropies · Baryon Asymmetry and Baryogenesis · Thermal History of the Universe.

Ask the tutor: Ask me to work any inflaton potential — natural, hilltop, Higgs, or α-attractor — through the slow-roll machine and place it on the (n_s, r) plane against the latest data.

Why does the universe need to inflate at all — what problems does it solve?

Three classic puzzles of the standard hot Big Bang. The horizon problem: regions of the CMB on opposite sides of the sky share the same temperature to one part in $10^5$ , yet in a purely decelerating universe they were never in causal contact. The flatness problem: the curvature density $\Omega_k$ grows away from zero over time, so today's near-flatness demands absurd fine-tuning at early times. The monopole (relic) problem: grand unified theories overproduce magnetic monopoles that we never see. Inflation solves all three at once by expanding a tiny causal patch by a factor $e^N$ with $N\gtrsim60$ : this places today's observable universe inside one pre-inflationary horizon, drives $\Omega_k\to0$ as $e^{-2N}$ , and dilutes any relics to invisibility.

What exactly is the inflaton, and is it a known particle?

The inflaton is the scalar field whose potential energy drives inflation — but as of now it is a hypothetical field, not an identified Standard Model particle. It must be a spin-0 field with an extremely flat potential (a very small effective mass, $\sim10^{13}$ GeV in the quadratic model) and couplings that let it eventually decay into ordinary matter during reheating. Candidates have been proposed: the Higgs boson with a non-minimal coupling to gravity (Higgs inflation), a modulus or axion-like field from string theory, or the scalar degree of freedom hidden in $R^2$ gravity (Starobinsky). None is confirmed. Identifying the inflaton — connecting it to particle physics — is one of the central open goals of the field.

How many e-folds of inflation do we actually need, and why 60?

Solving the horizon and flatness problems requires roughly $N\gtrsim60$ $e$ -folds of expansion, but the precise number depends on the reheating temperature and the inflationary energy scale, typically landing in the range $N\approx50$ – $60$ for the scales we observe in the CMB. What matters observationally is not the total duration but when a given comoving scale crossed the horizon: the pivot scale $k_*=0.05\,\mathrm{Mpc}^{-1}$ exited about $50$ – $60$ $e$ -folds before the end of inflation, and it is $V$ evaluated at that moment that sets the measured $n_s$ and $r$ . This is why every prediction in this article is quoted "at $N=50$ – $60$ " — that window is the observable one.

Why does slow-roll inflation predict a spectrum that is "nearly" but not exactly scale-invariant?

A perfectly scale-invariant (Harrison–Zeldovich, $n_s=1$ ) spectrum would arise only in exact de Sitter, where $H$ is truly constant and every mode crosses the horizon under identical conditions. Real inflation is only quasi-de Sitter: $H$ decreases slowly as the field rolls, at a rate set by $\epsilon$ , and the potential curves, at a rate set by $\eta$ . So modes that cross the horizon later see a slightly smaller $H$ , giving them slightly less power — a red tilt. The precise result $n_s-1=-6\epsilon_V+2\eta_V$ makes the tilt a direct measurement of the slow-roll parameters, and Planck's finding $n_s=0.9649$ (rather than exactly 1) is a genuine detection of this tiny departure from de Sitter.

Do super-horizon perturbations really stay frozen, and how do they reach the CMB?

Yes — the key theoretical fact is that the comoving curvature perturbation $\mathcal{R}$ is conserved on super-horizon scales for adiabatic modes, independent of the unknown physics of reheating. Once a mode's wavelength is stretched beyond the Hubble radius during inflation, causal processes can no longer act across it, so its amplitude is locked in. It then rides out the entire reheating and radiation era untouched, and re-enters the horizon much later, during radiation or matter domination, where it acts as the initial condition for the acoustic oscillations we see in the CMB and for the growth of large-scale structure. This conservation law is what lets us predict CMB observables from inflation without knowing the messy details in between.

Can we ever directly detect inflationary gravitational waves?

Possibly — that is the single biggest experimental prize in the field. Inflation produces a stochastic background of primordial tensor perturbations (gravitational waves) with amplitude fixed purely by $H$ , hence by the energy scale $V^{1/4}$ . These are far too long-wavelength for interferometers like LIGO, but they leave a distinctive curl pattern, called B-mode polarization, in the CMB. BICEP/Keck and Planck have pushed the bound to $r\lesssim0.036$ ; the Simons Observatory, CMB-S4, and LiteBIRD aim for $r\sim10^{-3}$ . A detection would measure the inflationary energy scale directly and confirm that inflation was driven at GUT-scale energies — while continued non-detection steadily favors the low- $r$ plateau models over large-field ones.