Horizon Problem

The puzzle in one sentence

In the standard hot Big Bang, the universe expands from an extremely dense initial state. At about 380,000 years it cools enough for free electrons to combine with protons into neutral hydrogen, and the photons that had been scattering off the plasma stream freely toward us. We see those photons today as the cosmic microwave background — a near-perfect 2.7255 K blackbody covering the whole sky. The puzzle is that two points on the surface of last scattering separated by, say, 90° in our sky have never been able to exchange a single photon. The light that could most efficiently equalise their temperatures has to travel from one to the other through the expanding plasma. From the time of the Big Bang up to recombination, the maximum distance light can travel is the particle horizon, and at recombination it subtends only about 1.5° on the sky from our vantage. Yet the two patches agree in temperature to one part in 10⁵.

Either the universe began with its temperature already finely tuned to be the same across all the disconnected patches we now see, or the standard radiation-era expansion history is incomplete.

Putting numbers on it

The cleanest way to make the horizon problem precise is to compute two angular scales on the last-scattering surface and compare them. The first is the particle horizon — the comoving distance light has travelled from the Big Bang to recombination at redshift z_* ≈ 1090:

d_H(t_*) = ∫₀^{t_*} c dt'/a(t')

For a radiation-dominated universe, a(t) ∝ t^{1/2}, the integral is finite and yields d_H ≈ 2c t_* / a(t_*). Plugging in t_* ≈ 380,000 yr and accounting for the precise expansion history, this projects onto an angular size on the sky of

θ_H ≈ 1.5°

The second is the angular size of the entire visible CMB — essentially 180° in any direction. The ratio gives the number of causally disconnected patches we see:

N_patches ≈ (4π) / (π θ_H²) ≈ 4 × 10⁴

So we observe roughly forty thousand patches of CMB sky that, in the standard hot Big Bang, never had time to thermalise with one another — and they all share the same temperature to within 18 µK out of 2.7255 K. That is the horizon problem.

Historical development

The horizon problem was first articulated cleanly by Charles Misner in 1969 in the context of "mixmaster" anisotropic cosmologies — he was trying to identify what was anomalous about the apparently isotropic universe we observe. Robert Dicke pointed out a related cluster of puzzles, including the flatness problem, in his 1969 Jayne Lectures. Through the 1970s these were known together as the "smoothness problem" and were taken seriously by cosmologists; the standard response was to assume nature had chosen smooth initial conditions, and to move on.

Alan Guth changed that in late 1979. While a postdoc at SLAC working on grand-unified-theory phase transitions, he realised that a brief period of false-vacuum-dominated exponential expansion would automatically inflate any small, thermalised patch into a region much larger than the observable universe. The horizon problem would dissolve: the entire CMB sky descends from one homogeneous region. His original "old inflation" paper appeared in 1981. It had a graceful-exit problem (bubbles of true vacuum never percolated), which Andrei Linde, Andreas Albrecht, and Paul Steinhardt fixed in 1982 with the "new" or slow-roll inflation that remains the canonical model today. The framework has since been refined into the Starobinsky, R²-inflation, Higgs-inflation, and α-attractor classes, but the horizon-problem-solving mechanism — exponential expansion of a causal patch — is identical across all of them.

How inflation makes the puzzle go away

The mechanism is mechanical. Before inflation begins, take a small region of space whose physical size is well within the Hubble radius at that epoch — say a few Planck lengths across, perhaps 10⁻²⁹ m. In this region, the matter and radiation are in thermal equilibrium because they have been interacting at light speed since t = 0. Now switch on inflation: a scalar field (the inflaton) sits at high potential energy and drives the metric to expand as a(t) ∝ exp(H t), where H is the Hubble parameter during inflation. After N e-folds of expansion, the original region is bigger by a factor of e^N.

For the horizon problem we need the inflated patch to encompass our present observable universe, of comoving radius about 14 Gpc. Tracking the expansion factor back through standard radiation- and matter-dominated eras to the end of inflation gives a requirement

N ≳ 60 e-folds   (energy scale dependent, but robust)

which corresponds to a stretch factor of about 10²⁶. After inflation ends, the inflaton oscillates around its potential minimum and decays into ordinary matter and radiation — this is the reheating epoch — and the universe enters the standard hot Big Bang expansion. From that point on, the entire observable universe is descended from one originally-thermalised patch, which is why opposite ends of the CMB sky have the same temperature now.

A worked example: counting the patches

To see why the horizon problem is so sharp, work through the numbers explicitly for a standard ΛCDM cosmology with no inflation.

1. Last-scattering time. Recombination occurs at z_* = 1089.92, t_* = 379,000 yr (Planck 2018).
2. Particle horizon at last scattering. In a radiation-then-matter universe, d_H(t_*) ≈ 2 c t_eq^{1/2} t_*^{1/2} where t_eq is matter-radiation equality. Numerically d_H ≈ 290 kpc comoving.
3. Comoving distance to last scattering. From here to last scattering: D_LS ≈ 14.0 Gpc comoving.
4. Angular subtense. θ_H ≈ d_H / D_LS ≈ 290 / 14,000 ≈ 0.0207 rad ≈ 1.19°. Pre-Planck estimates often quoted 2°; the precise figure depends on whether you use the matter-only or full ΛCDM expansion history.
5. Total sky patches. 4π sr / π θ_H² ≈ 4π / (π × 0.0207²) ≈ 4 / 4.3 × 10⁻⁴ ≈ 9.3 × 10³ patches — roughly 10⁴.
6. Required coincidence. Each of these patches independently started at 2.7255 K to within 18 µK. The probability of that arising by chance, if each patch had a thermal temperature drawn independently from any reasonable distribution, is astronomically small.

The pre-inflationary patch that becomes our universe under inflation has a size at the start of inflation given by H_inf⁻¹ — the Hubble radius during inflation. For the GUT-scale energy density ρ_inf ~ (10¹⁵ GeV)⁴, the Hubble parameter is H_inf ~ 10¹³ GeV ~ (10⁻²⁹ m)⁻¹. Stretching by e^{60} ≈ 10²⁶ gives a final patch size of order 10⁻³ m at the end of inflation, which then expands by the standard hot Big Bang factor of about 10²⁸ to reach our present observable universe radius of about 10²⁶ m. The bookkeeping works out.

What the CMB itself shows us

The angular power spectrum of CMB temperature fluctuations C_ℓ peaks at the first acoustic peak ℓ ≈ 220, which corresponds to an angular scale of about 1° — the sound horizon at recombination, smaller than the particle horizon by a factor √3 (the sound speed in the photon-baryon plasma was c/√3). On angular scales larger than the particle horizon at recombination, ℓ ≲ 50, the temperature fluctuations are not produced by acoustic physics on the last-scattering surface (which had no time to operate). They were imprinted earlier, on super-horizon scales, and have the nearly scale-invariant Harrison-Zeldovich spectrum predicted by inflation — n_s = 0.9649 ± 0.0042 from Planck 2018, with the small red tilt being one of inflation's most specific signatures.

The Sachs-Wolfe plateau at low multipoles — the flat region of the C_ℓ spectrum for ℓ ≲ 50 — is the direct observational consequence of correlated super-horizon perturbations. Without inflation or some equivalent pre-recombination dynamics, those correlations should not exist. They are the empirical reason cosmologists take inflation seriously.

Comparison: horizon, flatness, and monopole problems

Puzzle	What is fine-tuned	Standard prediction	Observed value	Inflation fix
Horizon	CMB temperature uniformity	~10⁴ causally disconnected patches	ΔT/T ≈ 10⁻⁵ across whole sky	One thermalised patch → all of observable universe
Flatness	Ω near 1 today	|Ω − 1| grows with time	Ω_K = 0.001 ± 0.002 (Planck)	Curvature exponentially diluted
Monopole	GUT-era defect density	n_M ~ ρ_GUT / m_M ~ overclosure	None observed, < 1 per Hubble volume	Volume expansion dilutes density by 10⁷⁸
Origin of structure	Seeds of CMB anisotropies	No causal explanation	n_s = 0.965 ± 0.004	Quantum vacuum fluctuations stretched
Isotropy on large scales	Suppression of tensor anisotropies	Mixmaster oscillations expected	CMB isotropic to 10⁻⁴	Cosmic no-hair theorem in de Sitter
Defect dilution generally	Cosmic strings, domain walls	Should dominate from phase transitions	String tension Gµ < 10⁻⁷	All defects diluted away

Alternatives to inflation

Although inflation is the consensus solution, several alternatives have been proposed that also address the horizon problem.

• Varying speed of light (VSL). Albrecht and Magueijo (1999), Moffat (1993). If c was much larger in the early universe, the particle horizon at recombination could be enormously bigger — enough for full thermalisation. The theoretical price is breaking Lorentz invariance at a fundamental level. No consistent Lorentz-violating theory has gained broad acceptance.
• Cyclic and bouncing models. Steinhardt-Turok (2002), ekpyrotic models. The universe bounces from a contracting phase. Thermalisation happens in the previous cycle, and the horizon problem is exported to that earlier epoch. Whether this counts as solving the puzzle is debated.
• Variable Newton's constant. Some quintessence models allow G_N to differ in the early universe in ways that effectively rescale the horizon. Severely constrained by BBN and CMB measurements.
• String-gas cosmology. Brandenberger-Vafa (1989). A pre-Big Bang gas of strings at the Hagedorn temperature could have been thermal across a very long causal time, then dropped into the radiation era. Predicts a slightly different CMB power-spectrum tilt than inflation; not currently favored over inflation but not ruled out.
• Conformal cyclic cosmology. Penrose (2010). Each aeon ends in a heat-death phase that is conformally identical to the Big Bang of the next aeon. CMB temperature uniformity emerges from the previous aeon's thermalisation. Highly speculative; supposed observational signatures (concentric circles in the CMB) have been claimed and disputed.

None of these has the predictive economy of inflation. Inflation explains the horizon problem and simultaneously the flatness, monopole, and origin-of-structure problems with a single mechanism. The alternatives must work harder on each puzzle separately.

Common misconceptions

• "The CMB is uniform because we are at the centre." No. There is no centre. The CMB looks uniform from every point because the universe is homogeneous on scales larger than the particle horizon. The horizon problem is precisely about why this is so given that no causal mechanism in standard cosmology can have established the homogeneity.
• "Inflation says the universe expanded faster than light, violating relativity." Relativity does not forbid the expansion of space to carry distant points apart faster than c — that is true of dark energy today as well. What it forbids is signals propagating faster than c through space. Inflation does not violate this; it stretches space itself.
• "The horizon problem is solved as soon as you assume the universe was always smooth." That assumption is precisely the fine-tuning we are trying to explain. Calling it a solution is begging the question.
• "Inflation has been confirmed." Indirectly, through the nearly-scale-invariant adiabatic power spectrum, super-horizon correlations, and geometric flatness — yes. Directly, through a primordial gravitational-wave signal (B-mode polarisation) — no. The current bound r < 0.036 from BICEP/Keck-Planck 2021 rules out high-energy-scale slow-roll models but is consistent with low-energy-scale inflation.
• "The CMB is from the Big Bang." The CMB is the relic photon background emitted at recombination, 380,000 years after the Big Bang. The Big Bang itself was not a flash of light from one place; it was an epoch covering the entire universe at once.
• "All patches of CMB share exactly the same temperature." They share the same mean temperature, but the 10⁻⁵ fluctuations carry all the cosmological information. Inflation explains both the uniformity and the spectrum of fluctuations.

Where we stand now

The horizon problem remains the cleanest argument for inflation in the early universe. The numbers are unambiguous: in the standard hot Big Bang, ~10⁴ causally disconnected patches on the sky agree in temperature to one part in 10⁵, which requires either an extraordinary coincidence in the initial conditions or a pre-recombination dynamical mechanism. Inflation provides such a mechanism with 50–60 e-folds of de Sitter-like expansion and additionally predicts the observed scale-invariant adiabatic perturbation spectrum, the near-flat geometry of the universe, and the absence of GUT-era defects. As of 2026, Planck 2018 + BICEP/Keck 2021 jointly constrain the tensor-to-scalar ratio r < 0.036 at 95% confidence, which is consistent with most slow-roll models and is the primary observational handle on the energy scale of inflation. LiteBIRD (launch ~2032) aims to push to r ~ 10⁻³; CMB-S4 (this decade) and the Simons Observatory complement the space-based searches. A B-mode detection would be the smoking gun for inflation's resolution of the horizon problem; a non-detection over a few more decades would push some models out and motivate alternatives more seriously.