Effect of inhomogeneities on the expansion rate of the Universe

Worth your time, but read it for the mean, not the variance. The deliverable here is a genuine second-order calculation of ⟨θ⟩ for irrotational dust in synchronous gauge, with the averaging measure done properly — Eq. (7) and the second-order prescription in Eq. (24b), where a first-order quantity picks up the −3⟨ψ⁽¹⁾O⁽¹⁾⟩ correction. That bookkeeping is what makes ⟨ρ⟩⁻¹d⟨ρ⟩/dt = −⟨θ⟩ come out cleanly at second order (Eq. 34), and it is the part I'd actually cite. The result is a few parts in 10⁵, the integral is dominated by q ~ 0.1–1, and it is UV-insensitive. Appendix A shows the answer doesn't depend on the normalization of g⁽²⁾, and Appendix B redoes it in Poisson gauge and gets the same thing. That cross-check is not decoration: the observation there that 'intrinsic' second-order terms in one gauge become first-order-squared in another is exactly why Räsänen's partial calculation (their Eq. B23) is incomplete, and the paper says so without heat. Credit for that. The claim of being the first complete second-order treatment looks right to me; Hui–Seljak was an order-of-magnitude estimate and Räsänen was partial. On the variance, the stress-test note lands. Eq. (56) is the honest filtered variance and is finite for any finite R. To argue the variance is dominated by an 'IR-singular part' the paper drops the filter and writes Eq. (60), which as printed diverges at both ends — ∫dk₁/k₁⁵ Δ² is IR-log for HZ, but ∫dk₂/k₂ Δ² is UV-divergent, so the 'IR part' isn't cleanly isolated. Eq. (74) then re-imposes k_H as a UV cutoff as well as k_MIN, which is a choice, not a derivation. The paper doesn't show the filtered integral reduces to Eq. (60) in the relevant limit. To their credit they flag this, note the third-order AC cross term could carry its own IR-singular piece that needn't cancel the B² term, and defer the strong claims to Ref. [6]. So the abstract's 'sensitive to the physical significance of the unknown spectrum' (abstract:s2) is honest framing, not overreach. Bottom line: the mean result is a real, checkable contribution and the variance is a flagged open question. Send it out; the referees should press on the Eq. (56)→(60) reduction.

The mean-value result (Eq. 50/66, Fig. 1) is the paper's robust deliverable: the integral is dominated by linear scales q~0.1–1, is UV-convergent, and is cross-checked in the Poisson gauge (Appendix B). The variance claim is weaker. The full filtered variance is Eq. (56), which is finite for any finite R because of W²(kR). To argue that the variance is 'dominated by the IR-singular part' for a red spectrum, the paper drops the filter and writes Eq. (60), then re-imposes both an IR cutoff k_MIN and a UV cutoff k_H (Eqs. 74–75). This decomposition is not derived from Eq. (56): Eq. (60) as printed carries both an IR divergence (∫dk₁/k₁⁵ Δ² ~ ∫dk₁/k₁ for HZ) and a UV divergence (∫dk₂/k₂ Δ² ~ ∫dk₂ k₂³ for HZ), so the 'IR part' is not cleanly isolated. The paper does not show that the filtered integral reduces to Eq. (60) in the relevant limit, nor that the neglected filtered pieces are subdominant. The authors acknowledge the cutoff dependence and defer strong claims to Ref. [6], so this does not undermine the mean result, but the variance statement is less controlled than the mean and rests on an unproven reduction.

The reader correctly identifies the variance/cutoff issue as the weakest assumption, and I agree the mean result is solid: the two-gauge check (Appendix B) and the UV-insensitivity argument (integral peaked at q~0.1–1) give independent support for the ~10⁻⁵ mean. My concern is narrower and more technical than the reader's: it is not the physical choice of cutoff per se, but the unproven step of extracting Eq. (60) from the filtered Eq. (56) and then re-imposing cutoffs. This is acknowledged in spirit by the authors (Sec. IV, Sec. V, and the deferral to Ref. [6]), so it does not change the verdict. Two minor points worth noting but not load-bearing: (i) Eq. (50) appears to carry 1/(a²H²) where the derivation from Eq. (41) and Eq. (49c) gives a²H²; Eq. (66), which is what is actually evaluated, is dimensionally consistent, so this is a presentation typo rather than a physics error; (ii) the four-derivative terms are only checked to have zero mean in the Poisson gauge, not fully cross-checked, but they are subdominant for R = H₀⁻¹. Neither affects the central mean-value claim. The paper does not overclaim: it states the mean correction is small and treats the super-Hubble variance as an open question, so ACCEPT with the variance read as conditional is the right call.

If the mean result holds, it fixes the size of the second-order correction to the expansion rate for the terms that can be computed in this formalism, and it rules out a large homogeneous effect from those terms — a useful, quantitative negative result for the backreaction literature. The paper's strengths are real and worth naming: the averaging prescription is stated explicitly rather than left implicit; the result is cross-checked in a second gauge (Appendix B); Appendix A gives a clean argument that the answer is independent of the unphysical normalization of the second-order metric; and the paper engages directly with the published claim of Räsänen [5], reproducing his Eq. (11) as Eq. (B23) before explaining which terms he omitted. The mean correction is a falsifiable, order-of-magnitude statement. The variance claim in the abstract is appropriately hedged and is presented as conditional on the status of super-Hubble modes, with the more speculative extension explicitly deferred to Ref. [6].

The paper asks a deceptively simple question: if the universe is not perfectly smooth, does its average expansion rate differ from the Friedmann value $H = sqrt(kappa^2 rho / 3)$, and by how much? The authors carry out the first complete second-order calculation in synchronous (comoving) coordinates for a matter-dominated universe, taking care to average with the proper volume element $sqrt(gamma) d^3x$ and to identify the quantity that actually governs how the mean density dilutes. They find the mean correction is a few parts in $10^5$: small, and well behaved at short distances, which undercuts an earlier suggestion that short-wavelength modes could produce an apparent acceleration. The scatter about that mean is a different story, because on the scale of the Hubble radius it is dominated by a term proportional to $phi nabla^2 phi$ whose variance grows without bound as more super-Hubble modes are included. The size of that scatter is therefore set by physics outside the observable universe rather than by anything measured inside it.

What carries the argument. The load-bearing device is a pair of choices. The first is the $sqrt(gamma)$-weighted spatial average $langle O rangle = int d^3x sqrt(gamma), O / int d^3x sqrt(gamma)$ on constant conformal-time slices. The second is the continuity equation $a^{-1} langle rho rangle' / langle rho rangle = -langle theta rangle$, which defines the scale factor $a_V$ that actually tracks the dilution of matter through $dot a_V / a_V equiv langle theta rangle / 3$. Keeping the $sqrt(gamma)$ factor is what makes the second-order bookkeeping close: without it the density-dilution equation does not come out in this simple form and the correction $langle delta theta rangle / 3H$ would be misidentified. The other ingredient is the choice of synchronous (comoving) gauge, where the fluid four-velocity is trivially $u^mu = a^{-1}(1, vec{0})$, so $theta$ reduces to the trace $Gamma^i_{0i}$ and the second-order metric potentials carry no nonlocal terms. A Poisson-gauge recomputation in Appendix B reproduces the same answer, but only if the intrinsically second-order potentials $varphi^{(2)}$ and $psi^{(2)}$ are retained; dropping them, as earlier treatments did, breaks agreement between the two gauges.

Core claim. Expanding the metric and the expansion scalar $theta = nabla_mu u^mu$ to second order, the authors show that the averaged expansion rate of a perturbed matter-dominated universe obeys $langle theta rangle = 3H(1 + langle delta theta rangle / 3H)$ with $H equiv sqrt(kappa^2 langle rho rangle / 3)$, and they compute $langle delta theta rangle / 3H$ explicitly in terms of the peculiar gravitational potential $phi$. The result, Eqs. (41) and (50), contains a first-order piece averaged with the $sqrt(gamma)$ measure plus second-order pieces involving $phi nabla^2 phi$, $phi_{,i} phi_{,i}$, $(nabla^2 phi)^2$ and $phi_{,ij} phi_{,ij}$. Evaluated with a Harrison-Zel'dovich spectrum and a BBKS transfer function, the mean is a few parts in $10^5$ at the present epoch, scales as $a/a_0$, receives most of its contribution from wavenumbers around $0.1$ to $1,h,mathrm{Mpc}^{-1}$, and is essentially insensitive to the ultraviolet cutoff. The variance behaves differently: the linear term gives a scatter falling as $R^{-2}$ and dominates the mean out to roughly $3 times 10^3,h^{-1},mathrm{Mpc}$, while the second-order two-derivative term has an infrared-divergent variance scaling as $ln(k_H/k_{rm MIN})$. That variance is about $10^{-5}$ for a Hubble-radius cutoff, and reaches order unity only if the spectrum extends roughly $10^{18.8}$ e-folds beyond the present Hubble radius, or about 676 e-folds for a slightly red spectrum with $n = 0.94$. An appendix shows the answer is independent of the normalization of $langle g^{(2)}_{mu nu} rangle$, since second-order metric terms enter only through spatial gradients.

The premise it rests on. The variance result assumes the infrared divergence can be cut off at or near the present Hubble radius, meaning that modes larger than our observable universe are either unphysical or still small enough for perturbation theory to hold.

For the terms that can be computed exactly at second order, the mean effect of inhomogeneity on the expansion rate is a few parts in $10^5$, far too small to account for the observed late-time acceleration, so dark energy is not displaced at this order.

The short-wavelength contributions are ultraviolet-safe: the term earlier work flagged as potentially large is subdominant once the $sqrt(gamma)$ averaging is performed correctly.

The scatter in the expansion rate across different Hubble volumes is dominated by the ordinary linear term out to about $3 times 10^3,h^{-1},mathrm{Mpc}$.

If super-Hubble modes have physical standing, the second-order variance grows in the infrared and the perturbative expansion fails once the local potential becomes of order one, opening a route to a nonperturbative correction to the Friedmann equation.

The size of the effect depends on the super-Hubble spectral index: with $n = 0.94$, unit variance requires about 676 e-folds beyond the Hubble radius, which becomes a statement about how long inflation must have lasted for the effect to matter.

The calculation reframes the backreaction question as one about boundary conditions: because the divergent piece has the form $phi nabla^2 phi$, its variance is fixed by the amplitude of the potential on scales no local observer can see, so no measurement confined to the Hubble volume can settle it.

A third-order calculation is the natural next step. The paper itself notes that terms such as $langle A phi rangle langle C phi^3 rangle$ could carry an equally singular infrared piece, and no symmetry argument is offered that would make it cancel the second-order divergence.

If the super-Hubble variance is real and of order unity in some regions, observers in different Hubble volumes would infer different expansion histories from identical local physics, which is in principle a statistical selection effect testable against observed large-angle isotropy.

The demonstration that intrinsic second-order potentials must be kept for gauge consistency is a methodological point with reach beyond this paper: any second-order cosmological observable compared across gauges needs the same care.

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review