The Cosmological Principle

The cosmological principle says that the universe admits no preferred spatial position or direction at the level of its large-scale model. In its exact form it is a symmetry assumption: spatial slices are homogeneous and isotropic relative to a chosen family of observers. In observational work it is usually a statistical approximation, supported only above specified scales and after selection effects are modeled. Keeping those statements separate is essential: the sky seen by one observer cannot by itself prove that every spatial point is equivalent.

Required background. Pseudo-Riemannian Metrics and Signature supplies Lorentzian metrics and causal cones; Affine and Variational Geodesics supplies the Geodesics and Jacobi Fields background; this page uses its timelike-worldline, affine-null-ray, and proper-time material for the observer and light-cone geometry, not Jacobi-field theory.

Helpful background. The Einstein Equation supplies the Einstein equations and metric–matter dynamics, while Relativistic Matter and Compact Objects supplies fluid stress energy and observer congruences.

Three different symmetry statements

A spatial geometry is homogeneous if an isometry can carry any point to any other point. It is isotropic about a point if rotations fixing that point act in every spatial direction. Homogeneity does not imply isotropy: a Bianchi I universe may be spatially homogeneous while expanding at different rates along three axes. Isotropy about one point does not imply homogeneity either: a spherically symmetric but radially varying universe can look isotropic to its central observer.

Isotropy about every point is much stronger. In three dimensions it forces the spatial curvature to have the constant-curvature form

$${}^{(3)}R_{ijkl} =\mathcal K(t)\left(h_{ik}h_{jl}-h_{il}h_{jk}\right), \qquad {}^{(3)}R=6\mathcal K(t),$$

where $h_{ij}$ is the spatial metric. The time dependence here records the choice of physical spatial metric. After writing $h_{ij}=a_{\rm dim}^2(t)\gamma_{ij}$ with a dimensionful scale factor and normalized reference curvature $\kappa=+1,0,-1$, one has $\mathcal K(t)=\kappa/a_{\rm dim}^2(t)$. The congruence kinematics and symmetry classification are developed in Ellis and van Elst 1999, §2.1, pp. 6–9, and §§5.1–5.2, pp. 33–38 (PDF).

Real cosmological data motivate a weaker statement. A random field is statistically homogeneous and isotropic when its joint probability law is invariant under the isometry group of the chosen homogeneous, isotropic reference slice. For a flat reference slice this reduces to invariance under Euclidean translations and rotations; spherical and hyperbolic slices have their corresponding isometry groups. A particular realization still contains galaxies, voids, and anisotropic structures. Statistical symmetry therefore refers to an ensemble or to suitably smoothed estimators, not to a claim that the observed density is constant point by point.

From spatial symmetry to an FLRW model

Choose a timelike fundamental congruence $u^a$ and assume that it is orthogonal to the preferred slices. Hypersurface orthogonality already sets its vorticity to zero. Exact spacetime isotropy then eliminates the projected acceleration vector and the trace-free shear tensor. If the matter fields share the same symmetry, their energy flux and anisotropic stress also vanish; alternatively, that statement follows for the total stress tensor after Einstein’s equation is imposed. The purely kinematical content reduces to one scalar expansion,

$$\theta=\nabla_a u^a=3H, \qquad H=\frac{\dot a_{\rm dim}}{a_{\rm dim}}.$$

With cosmic time $t$ along $u^a$, the resulting line element can be written

$$ds^2=-dt^2+a_{\rm dim}^2(t) \left[d\chi^2+S_\kappa^2(\chi)d\Omega_2^2\right],$$

where $S_\kappa(\chi)=\sin\chi$, $\chi$, or $\sinh\chi$ for normalized $\kappa=+1,0,-1$. This geometric reduction does not use Einstein’s equation. The field equation enters only when matter is supplied and $a_{\rm dim}(t)$ is determined. Tong constructs the three reference-space coordinate forms and writes the FLRW line element in Tong 2019, §§1.1.1–1.1.2, pp. 7–13 (PDF).

The local FLRW metric does not determine global topology. A flat section may be Euclidean space or a compact flat quotient, and spherical or hyperbolic sections also admit quotients. Generic identifications preserve local constant curvature but destroy global isotropy, global homogeneity, or both; for example, a flat three-torus is homogeneous but not globally isotropic. An exact global cosmological principle therefore restricts the allowed identifications more strongly than the local curvature calculation does.

What one past light cone can support

An observer receives information on a past light cone, not on an entire constant-time slice. Near-isotropy of the cosmic microwave background, source counts, lensing fields, or expansion observables is therefore evidence only after foregrounds, masks, calibration, source evolution, peculiar motion, and the observer’s location have been addressed. Agreement between independent tracers is more discriminating than the appearance of a single smooth map.

There is correspondingly no estimator-independent “homogeneity scale.” As one representative test, the WiggleZ analysis used counts in spheres and the correlation dimension $D_2(r)$, defining its transition by $D_2=2.97$, within one percent of the homogeneous value $3$. Across four tracer redshift bins it found transition radii of roughly $70$–$80\,h^{-1}\,\mathrm{Mpc}$, where $h\equiv H_0/(100\,\mathrm{km\,s^{-1}\,Mpc^{-1}})$, with selection effects, tracer bias, redshift-space distortions, and covariance tested using mocks. Those numbers belong to that sample and threshold; changing the tracer, statistic, tolerance, or model changes the reported scale Scrimgeour et al. 2012, §§3–6, pp. 4–16, especially Table 4, p. 11 (Open PDF).

The Ehlers–Geren–Sachs result illustrates both the power and the limitation of such evidence. In the original one-component Einstein–Liouville system, a massless collisionless gas is itself the gravitational source and its distribution is exactly isotropic relative to a timelike unit field throughout an expanding region. Theorem 1 gives Robertson–Walker geometry when that mean flow is geodesic or irrotational; the exceptional case with both acceleration and vorticity nonzero is not covered. Ehlers, Geren, and Sachs 1968, §1, printed pp. 1344–1345, Theorem 1, p. 1348, and §5, p. 1349 also stress the equilibrium and causal limitations. We observe the radiation field at one event along one worldline, not at every fundamental worldline. Applying the theorem to our universe therefore additionally uses a Copernican premise: we are not at a specially symmetric location. Approximate isotropy requires a separate almost-EGS argument with control of the relevant derivatives, while additional matter requires the hypotheses of an appropriate extension of the exact theorem. Neither modification follows simply by replacing “isotropic” with “nearly isotropic”; see the explicit small-multipole and derivative assumptions in Ellis and van Elst 1999, §8.5.1, pp. 71–72, Eqs. (280)–(284) (Open PDF).

That premise is testable indirectly rather than logically guaranteed. Radial and transverse distance measures, redshift drift, remote-scattering probes, and light-cone consistency relations can constrain special-center or strongly inhomogeneous alternatives. Each test still brings a matter model, selection function, and null-propagation calculation. The scientifically useful conclusion is consequently modest: an FLRW background is a remarkably economical, well-tested zeroth-order model on declared large scales, while inhomogeneities and observer-dependent light-cone effects must be restored for precision work.

A scale-explicit working principle

Before invoking the cosmological principle, state four items: the observer congruence, the spatial or statistical quantity being smoothed, the smoothing scale, and the tolerated anisotropy. This prevents three common errors: treating local structure as a refutation of a large-scale statistical model, treating one isotropic sky as a proof of homogeneity, and treating FLRW as a consequence of Einstein’s equation rather than a symmetry reduction to which the equation is later applied.

The next pages make that working principle concrete. FLRW geometry implements the exact idealization; comoving observers define its cosmic time; Bianchi models retain homogeneity while relaxing isotropy; and the observed-light-cone chapter returns to the limits of inferring spatial structure from radiation.

Exercises

Consider a spherically symmetric density field $\rho=\rho(r)$ around an observer at $r=0$. Explain why this observer measures angular isotropy but why the field is not homogeneous unless $\rho$ is constant.

Solution

Every rotation about $r=0$ leaves $r$ and hence $\rho(r)$ unchanged, so the central observer has no preferred angular direction. A translation that moves the origin changes the value and radial gradient of $\rho$ whenever $d\rho/dr\neq0$. Points at different radii are then not related by a symmetry, so the field is inhomogeneous. This is the basic counterexample to inferring homogeneity from isotropy about one location.

References

• Ehlers, Jürgen, P. Geren, and Rainer K. Sachs. “Isotropic Solutions of the Einstein–Liouville Equations.” Journal of Mathematical Physics 9 (1968): 1344–1349. DOI. Public publisher PDF.
• Ellis, George F. R., and Henk van Elst. “Cosmological Models.” In Theoretical and Observational Cosmology, edited by Marc Lachièze-Rey, 1–116. Dordrecht: Springer, 1999. DOI; Open PDF.
• Scrimgeour, Morag I., et al. “The WiggleZ Dark Energy Survey: The Transition to Large-Scale Cosmic Homogeneity.” Monthly Notices of the Royal Astronomical Society 425 (2012): 116–134. DOI; Open PDF.
• Tong, David. Cosmology. University of Cambridge Part II Mathematical Tripos lecture notes, Michaelmas Term 2019. Official PDF.