Three-Torus Universe in Cosmology

• The three-torus universe is a compact, flat cosmological model defined by T³ topology with opposite faces identified, yielding finite volume and periodic boundaries.
• It produces unique observable signatures by affecting CMB correlations, suppressing large-scale modes, and enabling topology searches via matched circles and spatial correlations.
• Quantum and gravitational studies on T³ reveal discrete field modes, Casimir energy effects, and modified Newtonian potentials due to its nontrivial, multiply connected topology.

A three-torus universe is a cosmological model in which the comoving spatial section is the compact flat manifold $T^3=S^1\times S^1\times S^1$, so that space is locally Euclidean but globally multiply connected and finite in volume. In FLRW language, the spatial manifold is a constant-curvature 3-manifold $M=\widetilde M/\Gamma$; for the toroidal case the local metric remains that of a flat model, while opposite faces of a fundamental domain are identified, so that crossing one face re-enters through the opposite face. This combination of local flatness and nontrivial global topology is the defining feature of the three-torus universe and underlies its observational, field-theoretic, and quantum-gravitational significance (Roukema, 2010, Aurich, 2014).

1. Topological definition and geometric structure

In the three-torus model, physical space is not infinite $\mathbb{R}^3$ but a compact quotient with periodic identifications. One may describe it as a Euclidean parallelepiped or cube with opposite faces identified, as a tiling of the simply connected covering space by copies of a fundamental domain, or as the product $T^3=S^1\times S^1\times S^1$. The local geometry is unchanged relative to a flat FLRW model: small triangles, geodesics, and local dynamics retain Euclidean metric properties, while the distinction appears only at global scale (Roukema, 2010, Aurich, 2014).

This global structure is naturally encoded by the quotient relation

$M=\widetilde M/\Gamma,$

with $\Gamma$ the covering or holonomy group. In the observational topology literature, it is useful to distinguish the ordinary comoving separation $d(q_1,q_2)$ in the universal cover from the topology-aware minimum distance

$$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$

For topology searches one excludes the identity element and uses

$$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$

so that pairs of sky points that are angularly distant may nevertheless be comovingly near once the torus identifications are applied (Aurich, 2014).

Compactness has immediate cosmological consequences. If the Universe is compact, then no physical structure larger than the fundamental domain can exist, and the longest fluctuation modes are absent or suppressed. This is one of the main reasons three-torus models were historically connected to the observed lack of very-large-scale CMB structure. Within that broader FLRW comparison, the three-torus and the Poincaré dodecahedral space were presented as the two best candidates based on observations available at the time, while theoretical criteria did not unanimously select one over the other: the residual gravity effect favored $S^3/I^*$, whereas a measure-space argument with $M=\widetilde M/\Gamma$0 treated as derived favored compact flat spaces almost surely (Roukema, 2010).

2. Cosmic microwave background signatures and empirical status

The most direct cosmological signature of a three-torus universe arises from the fact that two CMB photons arriving from very different sky directions may originate from points on the last-scattering surface that are actually nearby in comoving space after topological identifications. In the simply connected case the standard statistic is the angular correlation

$M=\widetilde M/\Gamma$1

but for a multiply connected space one can instead compare the ordinary spatial correlation

$M=\widetilde M/\Gamma$2

with the topology-induced correlation

$M=\widetilde M/\Gamma$3

If the assumed torus size and orientation are correct, then under ideal conditions $M=\widetilde M/\Gamma$4 should reproduce $M=\widetilde M/\Gamma$5, even though the pixel pairs entering the two averages are different. Matched circles appear here as the special case $M=\widetilde M/\Gamma$6, so the spatial-correlation method generalizes the matched-circles construction rather than replacing it (Aurich, 2014).

A detailed Planck 2013 analysis implemented this program for a cubic 3-torus using the SMICA temperature map, the Union mask U73, and torus simulations calibrated to the Planck 2013 best-fit $M=\widetilde M/\Gamma$7CDM model. The comoving radius to last scattering was taken as $M=\widetilde M/\Gamma$8, so the last-scattering diameter is

$M=\widetilde M/\Gamma$9

The simulations used 100 realizations of a cubic 3-torus with side length $\mathbb{R}^3$0, $\mathbb{R}^3$1 eigenmodes from the first $\mathbb{R}^3$2 eigenvalues, spherical expansion up to $\mathbb{R}^3$3, HEALPix $\mathbb{R}^3$4, Gaussian smoothing $\mathbb{R}^3$5, and 40 bins equally spaced in $\mathbb{R}^3$6 up to $\mathbb{R}^3$7. In Sachs-Wolfe-only realizations, $\mathbb{R}^3$8 almost perfectly traces $\mathbb{R}^3$9 for the correct torus, whereas in full Boltzmann simulations the match deteriorates substantially, though the correct orientation still remains enhanced relative to a wrong one (Aurich, 2014).

The search over torus size and orientation produced a notable candidate. After exploring $T^3=S^1\times S^1\times S^1$0 random points in the four-dimensional parameter space $T^3=S^1\times S^1\times S^1$1, with $T^3=S^1\times S^1\times S^1$2, the strongest candidate was found near

$T^3=S^1\times S^1\times S^1$3

For this configuration, the Planck 2013 topological correlation was described as “very close to the mean” topological correlation seen in the torus simulations with the correct orientation, and a similar signal appeared in WMAP 9-year data and at higher Planck resolution. The preferred orientation was summarized by the six Galactic directions where the torus generators intersect the last-scattering sphere: $T^3=S^1\times S^1\times S^1$4 A nearby secondary maximum at $T^3=S^1\times S^1\times S^1$5 had nearly the same orientation, indicating that the signal occupies a range of $T^3=S^1\times S^1\times S^1$6 around $T^3=S^1\times S^1\times S^1$7–$T^3=S^1\times S^1\times S^1$8 rather than an isolated point (Aurich, 2014).

The statistical interpretation remained cautious. The same study stressed that the spatial-correlation method has a high false-positive rate; an earlier estimate based on simply connected $T^3=S^1\times S^1\times S^1$9CDM simulations suggested a false-positive rate of roughly $M=\widetilde M/\Gamma,$0–$M=\widetilde M/\Gamma,$1. Moreover, Planck 2013 topology analyses found no convincing signal from the matched-circles test or from off-diagonal correlation and covariance-matrix methods. The torus-like configuration therefore remained striking but uncorroborated, and the author’s conclusion was that there exists a stable and unusually good torus candidate in the temperature data, but not robust evidence that the Universe is actually a cubic 3-torus (Aurich, 2014).

3. Quantum fields, Casimir energy, and inflationary scenarios

Quantum field theory on a three-torus differs from the simply connected case because compactness discretizes spatial momenta. For a scalar field on $M=\widetilde M/\Gamma,$2, the mode labels are integer triples $M=\widetilde M/\Gamma,$3, and the effective frequencies are

$M=\widetilde M/\Gamma,$4

This produces a genuine zero mode $M=\widetilde M/\Gamma,$5, whose dynamics differs qualitatively from the nonzero modes. In the particle-creation analysis of an expanding toroidal universe, nonzero modes exhibit the expected adiabatic shutoff around $M=\widetilde M/\Gamma,$6 and subsequently dilute as $M=\widetilde M/\Gamma,$7, while the zero mode is uniquely sensitive to the initial time $M=\widetilde M/\Gamma,$8. Under the physically adopted choice $M=\widetilde M/\Gamma,$9, both nonzero-mode and zero-mode created-particle energy densities remain negligible compared with the total cosmic energy density (Fornal, 2012).

The same compactness implies Casimir energies that scale as $\Gamma$0 or, in the notation of the later Planck-size torus scenario, $\Gamma$1. In one semiclassical model the Universe is assumed to begin at the Planck time

$\Gamma$2

with Planck-size linear scale

$\Gamma$3

and no matter or radiation before inflation. The pre-inflationary expansion is then driven solely by Casimir energies of the fields present on compact $\Gamma$4, with radiation-like equation of state $\Gamma$5 and

$\Gamma$6

For the preferred shape-moduli minimum, the Casimir density simplifies to

$\Gamma$7

and matching the Planck-time critical density

$\Gamma$8

requires

$\Gamma$9

substantially larger than the quoted Standard Model mismatch $d(q_1,q_2)$0 (Fornal, 25 Mar 2026).

That same scenario derives a present-size relation controlled mainly by the number of inflationary e-folds $d(q_1,q_2)$1 and only weakly by the reheating density drop $d(q_1,q_2)$2. For the benchmark choice

$d(q_1,q_2)$3

the torus grows from

$d(q_1,q_2)$4

to

$d(q_1,q_2)$5

while the present density becomes

$d(q_1,q_2)$6

The model combines this with a Planck lower bound

$d(q_1,q_2)$7

and a low-$d(q_1,q_2)$8-suppression-preferred range

$d(q_1,q_2)$9

to isolate a combined interval

$$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$0

The paper presents this as a proof-of-principle scenario rather than established cosmology (Fornal, 25 Mar 2026).

4. Quantum-gravitational realizations

In nonperturbative quantum gravity, the three-torus has been studied extensively in 4D Causal Dynamical Triangulations. With spatial topology fixed to

$$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$1

and periodic Euclidean time, the triangulated spacetimes have effective topology $$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$2. The central result is that the quantum geometry fluctuates around a dynamically generated background geometry whose large-scale behavior is described by an effective minisuperspace action. Unlike the familiar $$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$3 case, there is no de Sitter-like droplet; instead the average spatial volume profile is approximately constant, $$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$4, with macroscopic volume at all times (Ambjørn et al., 2017).

The effective toroidal action inferred from covariance-matrix and transfer-matrix methods is

$$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$5

with $$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$6 close to $$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$7. Its salient feature is the absence of the classical curvature term $$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$8 present for $$d_\Gamma(q_1,q_2):=\min_{\gamma\in\Gamma} d(q_1,\gamma(q_2)).$$9; the remaining power-law potential is interpreted as a quantum correction. The kinetic term has the sign measured in the Euclidean effective action, opposite to the sign of the naive Euclidean minisuperspace reduction, reflecting the entropic reversal of the conformal-mode instability in CDT (Ambjørn et al., 2017).

A broader CDT phase-diagram study then argued that replacing spatial $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$0 by spatial $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$1 does not alter the qualitative phase structure nor, as far as current evidence shows, the order of the key phase transitions. Analogues of phases $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$2, $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$3, $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$4, and $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$5 persist, the phase-transition lines are only slightly shifted, and the semiclassical phase survives; the toroidal distinction is that the average spatial-volume profile in phase $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$6 is flat rather than Euclidean de Sitter. This was presented as evidence for a form of topological universality of CDT results (Gizbert-Studnicki, 2019).

In quantum cosmology, the no-boundary proposal yields a qualitatively different result on $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$7 than on $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$8. For slow-roll inflation on the 3-sphere, the standard no-boundary wavefunction exponentially favors a small universe, whereas on the 3-torus one must sum over an infinite $$d_{\mathrm{topo}}(q_1,q_2):=\min_{\gamma\in\Gamma'} d(q_1,\gamma(q_2)),\qquad \Gamma':=\Gamma\backslash\{\mathrm{id}\},$$9 family of smooth fillings. After this arithmetic sum and integration over torus moduli, the resulting norm is

$S^3/I^*$0

so the catastrophic exponential $S^3/I^*$1 bias is replaced by a power-law dependence. For the solvable exponential slow-roll potential used there, this gives

$S^3/I^*$2

using the quoted observational bound $S^3/I^*$3. The same framework computes torus-moduli corrections to the rotationally averaged CMB power spectrum and finds them at most of order $S^3/I^*$4, well below cosmic variance (Godet, 6 May 2026).

5. Local gravity, compactness, and topology-induced exterior fields

The three-torus universe also modifies local gravitational physics. In the weak-field exterior of an isolated non-rotating star in a finite flat $S^3/I^*$5 universe, the naive method-of-images sum for the Newtonian potential diverges, because the associated Poincaré or Epstein series at the relevant power is not absolutely convergent. On a compact torus the Poisson equation also obeys a zero-mode integrability condition, so one cannot simply periodize the $S^3/I^*$6 potential (Steiner, 2016).

Regularization by Appell’s triply periodic zeta function or by Epstein zeta functions yields a triply periodic solution of the modified Poisson problem and forces an additional uniform term. The corresponding first-order field equation contains the topology-induced constant

$S^3/I^*$7

which depends on the source mass and torus volume. This term is interpreted as a topology-induced dark energy with

$S^3/I^*$8

The exterior is therefore not vacuum in the usual asymptotically flat sense, and the paper states explicitly that for this problem the vacuum equations $S^3/I^*$9 have no nontrivial solutions (Steiner, 2016).

The toroidal identifications also break global isotropy. Even for a spherical source, the exact multipole expansion contains no dipole but always a nonvanishing quadrupole, so Birkhoff’s theorem does not hold in the ordinary form. In the cubic case, the leading metric component is approximately Schwarzschild–de Sitter-like,

$M=\widetilde M/\Gamma$00

up to smaller topology-induced multipoles. On that basis the paper conjectures that black holes do exist in a toroidal universe and that their weak-field far geometry is well approximated by this topology-modified exterior field, although no exact nonlinear black-hole solution is constructed (Steiner, 2016).

6. Related constructions, common confusions, and current assessment

Several technically important but conceptually distinct lines of work are often grouped with three-torus cosmology. In $M=\widetilde M/\Gamma$01-dimensional Euclidean gravity, the “Euclidean torus universe” has spacetime topology $M=\widetilde M/\Gamma$02, not $M=\widetilde M/\Gamma$03, and can be quantized by Chern-Simons and quantum-group methods. In that setting the physical Hilbert space is identified with $M=\widetilde M/\Gamma$04, the gauge-invariant observables form two commuting copies of the Heisenberg algebra, and the modular group $M=\widetilde M/\Gamma$05 acts unitarily on states. This is a rigorous torus-universe model, but it is a lower-dimensional analogue rather than a $M=\widetilde M/\Gamma$06-dimensional three-torus cosmology (Meusburger et al., 2010).

A related lower-dimensional CDT program studies $M=\widetilde M/\Gamma$07-dimensional spacetimes foliated by $M=\widetilde M/\Gamma$08 slices. There the torus is valuable because it introduces genuine global shape variables in addition to the scale factor. Periodic toroidal simulations show no de Sitter-like condensation into a stalk-and-blob profile, and fluctuation analysis reconstructs an effective action with kinetic term

$M=\widetilde M/\Gamma$09

again with the sign reversal associated with entropic contributions. This suggests that compact torus topology can qualitatively reshape the semiclassical background even in lower-dimensional nonperturbative gravity (Budd et al., 2013).

Other constructions are even further from standard cosmological $M=\widetilde M/\Gamma$10. A point scatterer on a flat three-dimensional torus produces perturbed eigenfunctions that become uniformly distributed in configuration space at high energy; this is directly about quantum dynamics on a 3-torus, but not about cosmological evolution (Yesha, 2012). Rigid quotients of complex three-dimensional tori $M=\widetilde M/\Gamma$11 classify compact complex orbifolds of real dimension six, not ordinary spatial 3-tori (Gleissner et al., 2024). A “spacetime triple wormhole” obtained by spherically inverting an embedded flat 3-torus produces an unbounded, asymptotically flat three-neck wormhole, not a compact toroidal universe (Herr et al., 25 May 2026). A constrained-particle model on an embedded 3-torus uses Lagrange-multiplier and Wess-Zumino variables as additional phase-space directions and interprets them as “two extra dimensions,” but this is a gauged mechanical toy model rather than a relativistic cosmology (Nejad et al., 2015).

Several misconceptions therefore require explicit separation. A three-torus universe is not a literal doughnut embedded in a higher-dimensional Euclidean space; the statement is topological, not extrinsic. Nor does local flatness imply infinite spatial volume: $M=\widetilde M/\Gamma$12 is compact and flat-compatible simultaneously. Conversely, a torus-like signal in a CMB topology search is not equivalent to a detection of cosmic toroidal topology, because the best current Planck-era torus candidate was found by a method with a substantial false-positive rate and was not corroborated by matched circles or covariance-based searches (Aurich, 2014).

The present research picture is therefore differentiated rather than uniform. Observationally, the three-torus remains viable but unestablished; in Planck temperature data it yields an intriguing and stable cubic candidate, yet not compelling evidence (Aurich, 2014). Theoretically, it is one of the principal compact FLRW candidates, and under the $M=\widetilde M/\Gamma$13-derived measure argument compact flat spaces occur almost surely, even though the residual gravity criterion instead favors the Poincaré dodecahedral space (Roukema, 2010). In semiclassical and nonperturbative quantum gravity, the three-torus supports coherent effective dynamics, distinct background geometries, and in the no-boundary setting even a qualitatively improved inflationary probability measure (Ambjørn et al., 2017, Gizbert-Studnicki, 2019, Godet, 6 May 2026). The combined implication is not that cosmology has converged on $M=\widetilde M/\Gamma$14, but that the three-torus universe remains a mathematically controlled, observationally testable, and theoretically persistent candidate for global spatial topology.