de Sitter Space

The solution: empty space with a positive Λ

Take Einstein's equations in vacuum, with no matter and no radiation but a non-zero cosmological constant:

R_μν − (1/2) g_μν R + Λ g_μν = 0

The maximally symmetric solution with Λ > 0 is de Sitter space. "Maximally symmetric" means it has the largest possible isometry group for a four-dimensional spacetime — ten independent Killing vectors, the same as Minkowski space, but realised as SO(4,1) instead of the Poincaré group. The Riemann curvature is fully determined by Λ alone: R_μνρσ = (Λ/3)(g_μρ g_νσ − g_μσ g_νρ). There are no preferred points, no preferred directions, no preferred times. There are no waves to chase, no clumps to circle. Empty space, curved by itself.

And yet it expands. The cosmological constant on the left side of the equation acts as if it were a vacuum stress tensor on the right with energy density ρ_Λ = Λc⁴/(8πG) and pressure p_Λ = −ρ_Λ c². The Friedmann equations applied to that stress give a constant H and an exponentially growing scale factor. Curvature alone, with no fuel, is enough to make the geometry blow up forever.

The metric in coordinates that matter

de Sitter space is one geometry but it admits several useful coordinate patches, each foliating it differently. The three most common are:

Coordinates	Metric form	Covers	Use
Flat (cosmological)	ds² = −dt² + e2Ht δij dxidxj	Half the manifold	Inflation, FLRW limit
Static (Schwarzschild-like)	ds² = −(1 − H²r²)dt² + dr²/(1 − H²r²) + r²dΩ²	Causal patch of one observer	Horizon thermodynamics
Global (closed-slicing)	ds² = −dτ² + (1/H²) cosh²(Hτ) dΩ²₃	Entire manifold	Embedding, conformal diagrams

The flat slicing is what cosmologists actually use. The static patch makes the cosmological horizon at r = 1/H manifest. The global slicing reveals that de Sitter space is topologically R × S³ — a contracting then re-expanding three-sphere when traced from past to future infinity.

The hyperboloid embedding

The cleanest way to see what de Sitter space is is to embed it in one higher dimension. Take five-dimensional Minkowski space with one time and four spatial directions, with metric η_AB = diag(−1, +1, +1, +1, +1). de Sitter space of radius ℓ = √(3/Λ) is the hyperboloid

−X₀² + X₁² + X₂² + X₃² + X₄² = ℓ²

This single-sheeted hyperboloid inherits its metric from the ambient Minkowski space. Constant time slices in the global coordinates are three-spheres of radius ℓ cosh(τ/ℓ): the universe contracts from infinite size at τ = −∞, reaches a minimum size ℓ at τ = 0, and re-expands to infinity. The SO(4,1) isometry group is exactly the group of Lorentz rotations of the ambient five-dimensional space that preserve the hyperboloid.

The embedding is the cleanest visual answer to "what shape is de Sitter space?" It is a hyperboloid in a fictitious extra dimension, with constant intrinsic curvature 1/ℓ². For Λ ≈ 1.1 × 10⁻⁵² m⁻² (the value measured in our universe) the radius ℓ ≈ 1.6 × 10²⁶ m, or about 17 billion light-years.

The cosmological event horizon

An observer in de Sitter space sees the universe accelerate away from them, with comoving objects receding faster the further away they are. At a critical radius the recession speed reaches c and beyond it the recession exceeds c — light from objects beyond can never reach the observer. That radius is the cosmological event horizon

r_H = c / H_dS

This is fundamentally different from a black hole horizon, but shares many features:

• Observer-centric. Every comoving observer has their own horizon — there is no single global horizon. Two observers separated by ten billion light-years see horizons that hardly overlap.
• A one-way membrane. Light emitted from inside r_H by the observer can eventually escape to infinity (in the static patch's redefined "infinity"); light from outside cannot reach them.
• Generated by a Killing vector. In static coordinates, ∂_t is a Killing vector that becomes null at r = 1/H, just as ∂_t in Schwarzschild becomes null at r = 2GM/c². Surface gravity is κ = c²H.
• Has entropy and temperature. Like the Schwarzschild horizon, it follows the Bekenstein-Hawking area law.

For our universe today, with H₀ ≈ 67 km/s/Mpc, the present Hubble radius is c/H₀ ≈ 14.4 billion light-years. The future de Sitter horizon — where H asymptotes once matter and radiation have completely thinned out — settles slightly outside this, at about 16 billion light-years.

Gibbons-Hawking temperature and entropy

In 1977 Gary Gibbons and Stephen Hawking applied the same Euclidean path-integral argument that gave Hawking radiation for black holes to the de Sitter horizon. The static patch metric in imaginary time has a conical singularity at r = 1/H unless the time coordinate is periodic with period β = 2π/H. By the basic relation β = ℏ/(k_B T), this gives

T_GH = ℏ H / (2π k_B)

For our universe, plugging in H₀ ≈ 67 km/s/Mpc:

T_GH ≈ 2.7 × 10⁻³⁰ K

This is about 30 orders of magnitude below the CMB. In practical terms it is undetectable, dwarfed by every other thermal source in the universe. But it is conceptually enormous: an empty de Sitter universe is not at absolute zero. It is intrinsically thermal, with the horizon supplying real photons at a real (if vanishingly small) temperature.

During inflation, when H was perhaps 10¹³ GeV ≈ 10²⁶ K, T_GH was correspondingly stupendous. Those quantum fluctuations stretched by the exponential expansion are the standard explanation for the primordial density perturbations seen in the CMB. The Gibbons-Hawking temperature of inflation is literally the source of the structure we see today.

The entropy of the de Sitter horizon follows the Bekenstein-Hawking area law:

S_dS = A_H / (4 ℓ_p²) = 3π c³ / (ℏ G Λ)

For our Λ this is S ≈ 10¹²². The most entropy the observable universe will ever contain, by a huge margin, is associated with its own future event horizon.

Inflation as quasi-de-Sitter

Cosmic inflation is the canonical application of de Sitter geometry in physics. The idea is that for a brief window — perhaps 10⁻³⁵ s to 10⁻³² s after the Big Bang — the energy density of the universe was dominated by a slowly rolling scalar field φ whose potential V(φ) acted like an effective cosmological constant Λ_eff = 8πG V(φ)/c⁴. The scale factor expanded by roughly e⁶⁰ in a tiny fraction of a second, smoothing out curvature, diluting any pre-existing relics, and seeding the seeds of structure with quantum fluctuations stretched to cosmic scales.

Inflation is "quasi-de-Sitter" because pure de Sitter has H exactly constant and the expansion would never end. To match observation, inflation must terminate. The standard slow-roll mechanism does this by letting V(φ) drift slowly downhill, so H drops gradually until the slow-roll parameter

ε = (1/2) (V′/V)² M_Pl²

reaches 1, at which point the kinetic energy of φ becomes comparable to its potential and the de Sitter approximation breaks. The departure from exact dS is quantitatively encoded in the spectral tilt of primordial perturbations: n_s − 1 ≈ −6ε + 2η. Planck's CMB measurement of n_s = 0.9649 ± 0.0042 thus measures the slow-roll parameters directly — it is one of the cleanest tests of the inflationary picture.

The future of our universe is de Sitter

Run the Friedmann equations forward with our measured Ω_m ≈ 0.31 and Ω_Λ ≈ 0.69. Matter density drops as a⁻³, radiation as a⁻⁴, but Λ stays constant. After a few tens of billions of years, matter and radiation are negligible compared to Λ, and the universe asymptotes to exact de Sitter expansion. Some quantitative milestones, taking today as t = 0:

Time	State of universe	Effective geometry
t = 0 (today)	Ωm = 0.31, ΩΛ = 0.69	Transition to dS
~10 Gyr	Ωm ≈ 0.1, dark energy dominant	Nearly dS, accelerating
~100 Gyr	All galaxies outside Local Group recede past horizon	Essentially dS
~10¹² yr	Stars exhausted, only black holes, white dwarfs, brown dwarfs	Exact dS to ppm
~10¹⁰⁰ yr	Supermassive black holes evaporated by Hawking radiation	Empty de Sitter

By around 100 Gyr, every galaxy outside the gravitationally bound Local Group will have receded beyond our cosmological horizon. Far-future astronomers — if any exist — will see only the Local Group and an otherwise dark, empty, accelerating sky. The CMB will redshift below detectability. The remaining observable universe will be vanishingly small. Our era of cosmic accessibility is brief on the de Sitter timescale.

de Sitter's cousin: anti-de-Sitter

The other maximally symmetric vacuum solution of Einstein's equations is anti-de-Sitter (AdS), obtained by taking Λ < 0 instead of Λ > 0. Where dS has positive curvature and an exponentially expanding scale factor, AdS has negative curvature and a finite-distance timelike boundary at infinity. Light rays reach the boundary in finite affine parameter; massive geodesics oscillate radially in finite proper time.

Property	de Sitter	Anti-de-Sitter
Λ	Λ > 0	Λ < 0
Curvature	+ constant	− constant
Isometry group	SO(4,1)	SO(3,2)
Spatial slices	S³ (closed) or R³ (flat)	R³ with negative curvature
Boundary at ∞	Spacelike (future + past)	Timelike
Observed in nature?	Yes — late-time cosmology	No
Holographic dual	dS/CFT (conjectured)	AdS/CFT (well-established)

AdS appears mostly as a theoretical workbench, because the AdS/CFT correspondence — Maldacena's 1997 conjecture that string theory on AdS₅ × S⁵ is dual to N = 4 super Yang-Mills on the AdS boundary — is the only setting in which we have a fully-defined non-perturbative theory of quantum gravity. dS, by contrast, is the geometry our universe is actually approaching. The mismatch is awkward: the cleanest theoretical tools are for the wrong sign of Λ.

dS/CFT and the holography puzzle

Following the success of AdS/CFT, a dS/CFT correspondence was proposed by Strominger, Witten and others around 2001. The conjecture is that quantum gravity on a d-dimensional de Sitter background is dual to a (d−1)-dimensional Euclidean CFT living on the future spacelike infinity I⁺ of de Sitter. The CFT would compute the wave function of the universe via a "Hartle-Hawking" path integral over geometries that asymptote to I⁺.

Several features make dS holography much harder than its AdS sibling:

• The boundary is in the future, not at fixed spatial infinity. Observers cannot send signals to it and back, so the analogue of asymptotic scattering states is unclear.
• No supersymmetric vacuum. dS breaks supersymmetry, removing the most powerful tool for analytic control.
• Apparently finite entropy. The horizon area gives S ~ 10¹²² for our universe. If this counts independent microstates, the Hilbert space of dS quantum gravity is finite-dimensional — a feature with no analogue in AdS.
• No known string-theory landscape. Constructing stable dS vacua in string theory ("the dS swampland conjecture") is famously controversial; some authors argue no metastable dS exists in any consistent theory of quantum gravity.

Whether de Sitter space admits a genuinely complete quantum description remains one of the deepest open problems in theoretical physics. Banks and Fischler, following the finite-entropy clue, have argued for a finite-dimensional Hilbert space of dimension exp(A/4ℓ_p²) ≈ exp(10¹²²) for our universe. If correct, this would imply that the asymptotic future of the cosmos has only finitely many distinguishable quantum states — a sharp departure from ordinary quantum field theory and a constraint that any successful theory of dS quantum gravity must respect.

Origin: de Sitter 1917 and the long rehabilitation

The solution was found within a year of general relativity's birth. In 1917 Albert Einstein, looking for a static universe, introduced the cosmological constant Λ as a repulsive term to balance gravitational attraction in a uniform matter distribution. Willem de Sitter, working in Leiden, solved the same modified Einstein equations in a different limit: matter density ρ = 0, with only Λ present. Out came a vacuum solution that, despite being empty, was not static. In de Sitter's original "static" coordinates the solution appeared time-independent, but he and Lemaître soon realised that test particles inserted into it would not stay at rest; they would recede exponentially.

Einstein objected on Machian grounds. To him, a universe with no matter could not be a sensible spacetime — the geometry should be sourced by matter, not by Λ alone. The two argued by correspondence for years. Hubble's 1929 discovery that distant galaxies recede with v ∝ r reframed the question: a non-static universe was no longer obviously wrong. Lemaître, then Friedmann, then a generation of cosmologists incorporated Λ and matter together in the FLRW framework. After Hubble, Einstein famously called Λ his "greatest blunder" — premature, as it turned out, because the same Λ was rehabilitated in spectacular fashion in 1998 when Riess, Perlmutter, and Schmidt's supernova surveys revealed cosmic acceleration. The simplest model consistent with the data has Λ > 0, which means the asymptotic future of the actual universe is the de Sitter solution that de Sitter found in 1917.

Common pitfalls

• Confusing de Sitter horizon with a black hole horizon. The de Sitter horizon is observer-centric (every observer has their own) and sits at proper distance c/H. A black hole horizon is global (one for the spacetime) and sits at the Schwarzschild radius 2GM/c². They share thermodynamic features but are physically distinct surfaces.
• Thinking "exponential expansion" is the same as "physical things moving faster than light". Comoving objects are at rest in their local frame; what grows is the proper distance between them. Special relativity's speed limit applies locally, not to cosmological recession velocities.
• Treating de Sitter and inflation as identical. Inflation is quasi-de-Sitter — H is slowly varying, not exactly constant. Pure dS would never end and would not produce the observed slight tilt n_s < 1 of the primordial spectrum.
• Forgetting the embedding has signature. The natural ambient space for dS is (−,+,+,+,+) Minkowski, not Euclidean. The hyperboloid is single-sheeted, and the SO(4,1) isometry is a Lorentz group, not a rotation group.
• Ignoring the difference between de Sitter and anti-de-Sitter. Λ > 0 versus Λ < 0 changes everything: spatial topology, asymptotic structure, boundary location, holography, observed sign in nature. They are not minor variants of each other.