WorksPhysics and Astronomy
Beyond dust in local cosmography: perturbative Vlasov predictions
Abstract
The dust approximation used in cosmography neglects the velocity dispersion generated during structure formation. This dispersion produces isotropic and anisotropic pressure, whose gradients accelerate the mean matter flow even though individual particles follow geodesics. The acceleration introduces an intrinsic dipole into the local distance-redshift relation. We calculate this dipole and the accompanying changes in expansion and shear using linear Vlasov perturbation theory about an FLRW background, with a mean dispersion history motivated by halo estimates. The assumed FLRW background prevents an independent test of the cosmological principle; the numerical predictions also depend on the adopted LCDM cosmology. Pressure also changes the growth of structure, modifying the shear and the spatial gradients that govern how the distance anisotropy varies with redshift. The largest correction at the lowest order in redshift is to the shear, while changes in the expansion gradient can produce a larger dipole at the next order. The resulting intrinsic dipole is far smaller than the observed low-redshift dipole, whose behaviour nearest the observer is dominated by motion between the matter and CMB frames.
Introduction
The dust model assigns a unique matter velocity to each event and describes the flow well while neighbouring streams remain separate. As structure forms, particle streams falling into the same region pass through one another, and different velocities coexist at the same position. The resulting velocity dispersion produces isotropic and anisotropic pressure through the random particle motions, even though the particles do not collide. The generation and evolution of these pressures have been studied using velocity moments and the phase-space distribution itself [1, 2, 3, 4, 5, 6, 7, 8]. Individual dark-matter particles follow geodesics, but gradients of these pressures can accelerate their mean flow. Our companion paper, hereafter Paper I [9], derives this acceleration without assuming a background spacetime and shows how it enters local cosmography. Here we estimate its magnitude for a prescribed mean dispersion, together with the associated changes in expansion and shear.
The dust-based covariant cosmographic analysis of [10] uses sources in the range 0.01 < < 0.1 from the Cosmicflows-4 catalogue [11]. Extending the fit to include sources below produces systematic shifts in the fitted parameters and worsens the goodness of fit. Reference [10] associates this sensitivity with the limits of representing nearby galaxy motions by a smooth fluid. This motivates examining the pressure terms omitted by the dust model, beginning with their effects on the local cosmographic coefficients. We calculate the local coefficients here; we do not make the full distance prediction needed to refit the catalogue.
Covariant cosmography expresses the distance–redshift relation in terms of the local expansion, shear and 4-acceleration [12, 13, 14, 15, 16, 17]. Vlasov perturbation theory (VPT) [18, 19, 20, 21] describes the density, mean velocity and velocity cumulants in a perturbed FLRW universe, from which the kinematic quantities follow. The mean dispersion is generated by nonlinear structure formation; VPT evolves its spatial fluctuations. We use the halo-based dispersion history of [21] as an input, with its relation to the required spatial mean discussed in § 3.1.
We evolve linear fluctuations about this mean, compare successive cumulant truncations and identify the scales contributing after spatial smoothing. At zeroth order in redshift, the largest correction is a 0.3–1.3% change in shear relative to dust on the scales considered here. The acceleration dipole is much smaller. Changes in the expansion and shear gradients enter at the next redshift order, where the growth-induced dipole can exceed the shear correction. We calculate these contributions and their variation between observers, then relate them to the coefficients used in dust cosmographic fits.
Collisionless matter and local cosmography
We follow the terminology and covariant notation of Paper I [9]. The matter energy frame is defined by a vanishing energy flux, . Let be its 4-velocity, the spatial projector and the projected spatial derivative. An overdotted denotes differentiation along , with free indices projected when indicated by . Angle brackets on tensor indices denote the projected, symmetric and tracefree (PSTF) part. To leading non-relativistic order, the isotropic pressure and anisotropic pressure are density-weighted second moments of particle velocities about the local mean. We set in covariant equations and restore it in numerical expressions.
In this frame, the exact momentum conservation law of Paper I gives
where is the total matter energy density and . If is the particle mass and the number density, the rest-mass density through second order in particle speed is , so that
The acceleration of the mean flow is therefore fixed by gradients of its isotropic and anisotropic pressure. The energy and particle-number frames agree at this order, since their relative velocity first appears at third velocity order [9]. Thus either frame gives the same acceleration to the order used here.
At second velocity order, Paper I gives the energy and pressure equations
Here , and are the volume expansion, shear and vorticity. Expansion dilutes the pressure and shear generates anisotropic pressure from an initially isotropic dispersion. These equations contain no mechanism for generating dispersion from exactly cold single-stream initial data. A statistical description after stream crossing must supply the dispersion produced by the unresolved dynamics; we specify this input in the next section. The relation to the moment hierarchy, including the third moments omitted here, is derived in § 3.2.
The directional Hubble parameter is [9, 12, 16]
where points from the observer towards the source, with and , as in Paper I. The intrinsic Hubble dipole is and the quadrupole is . To see how these enter a distance measurement, write the local luminosity-distance expansion [9, 12, 15] as
which defines the directional deceleration parameter . Following [10], we describe the anisotropy of by
where is the sky average. Expanding the logarithm to first order about FLRW gives
Both terms are evaluated at the observer and are independent of source redshift at leading order. The acceleration contribution is an intrinsic property of the matter congruence. Motion of the observer relative to that frame adds a Doppler dipole whose leading low-redshift amplitude scales as , where is the observer’s velocity relative to the matter frame projected along [9, 16]. A constant intrinsic term cannot be represented by a single change in the observer velocity over a range of redshifts, although the two can be degenerate in a narrow shell. Holding the smoothing radius and observer velocity fixed, the ratio of the intrinsic dipole to the Doppler dipole decreases in proportion to .
These predictions refer to the mean matter flow. Individual galaxies can be approximately geodesic without the mean velocity of a multi-stream galaxy population defining a geodesic congruence. A comparison with a galaxy catalogue therefore also requires the relation between the matter and tracer frames (§ 5.3).
Vlasov perturbation theory
Velocity moments and mean dispersion
The Vlasov equation follows the distribution of collisionless particles in position and momentum. Taking velocity moments converts it into equations for density, mean velocity, dispersion and higher moments. These equations form a hierarchy because transport of a moment depends on the next one: for example, the pressure equation contains the divergence of the third velocity moment. A finite set of fluid equations therefore requires a prescription for the omitted moments.
VPT uses velocity cumulants to describe the same distribution in the non-relativistic, subhorizon limit [18, 20]. Let be a particle’s peculiar velocity and its mean at fixed position and time. This average integrates over the local velocity distribution, weighting each stream by its particle density.1 The second cumulant is the dispersion about that local mean,
The cumulant cutoff specifies the highest velocity rank whose fluctuations are evolved. At we retain density, mean velocity and all six components of , setting higher cumulants to zero. At this order the resulting equations agree with the non-relativistic limit of Paper I’s second-velocity-order closure [9]. At we also evolve the third cumulant, which describes the asymmetry of the local velocity distribution and transports dispersion. The relation between cumulants and central moments is given in appendix § 8.§ 1.
We evolve linear fluctuations about homogeneous mean cumulants in an FLRW spacetime, using the Vlasov–Poisson equations on subhorizon scales. The perturbative order and the cumulant cutoff are separate choices: each evolved cumulant is treated at linear order. The nonzero mean dispersion represents random particle motions generated by nonlinear collapse. Standard pressureless perturbation theory [23] sets all cumulants above the mean velocity to zero.
We use , where a prime denotes a derivative with respect to conformal time , the pressureless growing mode with , and . Statistical isotropy makes the mean dispersion tensor proportional to , with amplitude
The inner average is over particle momenta at fixed position, as in eq. (10). The outer average is over statistical realisations of the resulting local dispersion field; for a statistically homogeneous, ergodic ensemble, it can also be estimated by a spatial volume average. The normalised dispersion has dimensions of comoving length squared, with and at the retained velocity order. The scale marks where dispersion significantly changes the evolution of perturbations. When redshift is used as the argument, we write , and likewise for .
In an FLRW spacetime without dispersion production, random peculiar velocities of non-relativistic particles redshift as , giving . Together with rest-mass conservation, , this gives , the FLRW limit of eq. (4). In our normalisation the freely redshifting solution obeys . Nonlinear collapse and stream crossing generate further random motion, so the mean dispersion history must also include its production.
We prescribe using the fitted halo estimate of [21]. Halo density and velocity-dispersion profiles, combined with the halo mass function, supply an estimate of the random motions generated by nonlinear structure formation. The fiducial prescription includes halo outskirts out to and gives . We use this history as the background dispersion in the linear VPT equations; its construction is described in appendix B.1.
We use the calibration cosmology of [21]: flat CDM with , , , and . These are the parameters used to obtain the dispersion history. They differ from the observational best-fit values from Planck [25]. A different cosmology changes the density spectrum, growth and dispersion history together.
Second-moment evolution
To connect the dispersion tensor to the covariant hierarchy of Paper I, write its moments as , where labels the angular rank and counts additional factors of squared particle speed. The velocity order is , and the lowest moments are
Here . The second-velocity-order closure keeps ; the rank alone does not specify the velocity order [9, 22]. These momentum-space ranks are distinct from the sky multipoles. Dividing by the rest-mass density gives
The unnormalised are exactly the moments of Paper I; denotes their ratio to . Here is particle velocity measured in the matter frame, so these are central velocity moments at leading non-relativistic order. The physical spatial components of the second moment are the dispersion tensor of eq. (10).
At leading non-relativistic order in each moment, we divide the covariant hierarchy by and use rest-mass conservation, . The unnormalised second-moment equations (4)–(5) have damping ; dividing by reduces it to . In general, becomes , and a normalised moment of velocity order obeys
In a homogeneous, isotropically expanding flow with no sources, a central moment of velocity order therefore redshifts as . The damping is set by the total velocity order, including traces.
To combine the equations at we add the and members, the trace and tracefree rank-two equations. This gives us a single equation for the second moment , whose trace is and tracefree part is :
Here we have collected the two members at into , so both the tracefree third moment and its vector trace enter the pressure evolution. In its Newtonian limit, eq. (15) is eq. (36) of [18].
Appendix § 9.§ 2 gives the scalar system including the third cumulant, and appendix § 10 tests its contribution. The relativistic hierarchy and its angular decomposition are given in [9].
Linear scalar perturbations
We use Poisson gauge,
with comoving wavenumber , Fourier convention and . The lapse potential enters the Euler equation and describes the spatial curvature. Here , and denote physical spatial components at linear order, while acts on comoving coordinates. Covariant coordinate components of acceleration and shear are and , respectively. We write , , and . We denote the scalar amplitude of the anisotropic pressure by :
The Fourier divergence is . Both isotropic and anisotropic pressure contribute to the longitudinal acceleration. On subhorizon scales their relevant combination is
and the momentum conservation (Euler) equation becomes
The relativistic projected pressure gradient also contains . For the growing modes its contribution is suppressed by relative to the spatial pressure gradient, as are the pressure corrections to the Poisson equation. We retain the leading subhorizon terms; the relativistic equations and this reduction are given in appendix § 8.§ 2.
Rest-mass conservation gives in this limit. Combining it with the divergence of eq. (19) and yields
To evolve the pressure, we perturb about a spatially flat background carrying a homogeneous dispersion, so that . The anisotropic pressure, acceleration, shear, vorticity and third moment vanish in the background. Density, expansion and isotropic dispersion each have a background value and a perturbation. At this order and . For the scalar velocity , the expansion perturbation and shear are
The expansion perturbation and shear have the same velocity-divergence amplitude. When we linearise eq. (15) we therefore pick up an isotropic source through and an anisotropic one through , and by eq. (21) their sum is purely longitudinal:
The pressure has its own evolution equation and is coupled to the density. We write the scalar dispersion perturbation as
The decomposition agrees with eq. (58) of [18]. The term arises because the pressure is density weighted. With , the linear second-moment equations at are
Here multiplies the full longitudinal tensor , including its trace. The source in eq. (24) therefore generates alone. The independent isotropic mode is unsourced and is set to zero in the growing solution. These are the two scalar equations in eq. (64) of [18]. Differentiating eq. (23) then gives
Of the coefficient 3 multiplying , one comes from differentiating the density-weighting term and two from the evolution of the longitudinal dispersion . The term accounts for the specified time dependence of the mean dispersion. In VPT it is supplied by the ensemble-averaged hierarchy, including advection, pressure–velocity correlations and, when retained, third-cumulant correlations [18]. For freely redshifting dispersion, and . The prescribed history represents production of random motions by nonlinear structure formation through a nonzero , with today. A long-wavelength density perturbation can change the abundance and evolution of the structures producing those motions. Here we prescribe their mean production and omit its spatial modulation; density, velocity and dispersion perturbations still evolve through the coupled equations above. The resulting mean is maintained by production, whereas source-free homogeneous pressure in eqs. (4)–(5) redshifts freely. § 5.2 discusses the omitted modulation and appendix § 9.§ 3 quantifies its possible effect in an illustrative model.
Once and initial conditions selecting the growing mode are specified, the coupled density and pressure equations form a closed linear system. At early times, when , we select the dust growing mode and its associated pressure perturbation (eq. (B.5)). We solve for the spatial fluctuations retaining the full dependence on , without expanding in this parameter. We define the kernels by
where is the pressureless linear density field extrapolated to today. The factor is the scale-independent dust growth; the kernels contain the VPT scale dependence. For the density kernel approaches unity. Near , pressure gradients oppose gravitational infall and reduce the growing mode. The kernels in Figure 1 describe this collisionless form of Jeans suppression.
Figure 1. The prescribed mean dispersion and the linear VPT kernels. Top: and the three-dimensional rms random particle speed about the local mean, including internal halo motions. Bottom: and , where . Their leading large-scale departures are proportional to . Triangles mark and circles mark . Pale curves extend beyond ; dotted segments extend beyond , where the finite truncation can generate oscillations and sign changes.
At sufficiently large , the solution varies on a timescale short compared with the expansion and the evolution of . Equation (25) then reduces to , so its rapidly varying part satisfies . Neglecting the expansion and gravitational terms in eq. (20) gives
In this limit the truncated equation has oscillatory solutions. Following the relative motion and phase mixing of particle streams requires the higher members of the velocity hierarchy [18, 20]. Nonlinear density evolution becomes important first: defining by , we find , and at , and . The high- continuations illustrate the linear equations outside their predictive range.
We use for the acceleration spectrum and spatial and radial windows, and for every local Hubble and deceleration multipole in § 4. The latter involve pressure time derivatives, whose evolution contains the third cumulant explicitly. Both calculations set the mean fourth cumulant to zero. Appendix § 10 compares the truncations, and appendix § 13 identifies the short-wavelength contributions to each windowed variance.
Acceleration spectrum
We compare VPT with dust at a fixed present-day density field, so that both have the same today. Define
Here is the scale-dependent VPT growth, while remains the pressureless growth factor. The differences then measure changes in velocity and pressure for the same final density field. A comparison at fixed primordial amplitude would change the final density spectrum by and give the two models different present-day .
Using eq. (19), we define the acceleration coefficient by
The density equation supplies the consistency identity,
For pressureless evolution , and . At long wavelengths multiplies a density gradient in the same way as an effective sound-speed coefficient. Section § 5.2 and appendix § 11 compare it with the acceleration obtained by spatially coarse-graining dust [26, 27].
Restoring , the dimensionless acceleration is , with now expressed as a physical acceleration. Its transfer from is
The factor makes dimensionless when is expressed in . We define the spectrum by
We use the baryon-corrected Eisenstein–Hu no-wiggle transfer function [28], normalised to for the cosmology specified above (numerical details are in appendix § 9).
The coefficient gives the acceleration produced by a specified density gradient, but does not determine which scales dominate its variance. The contribution per logarithmic wavenumber interval is
As shown in Figure 2, appreciable power comes from modes near and above and . An unsmoothed amplitude would therefore depend on modes that this linear calculation does not control. We quote acceleration amplitudes only after specifying the window that selects the relevant scales.
Figure 2. Linear VPT acceleration with the prescribed mean dispersion. Left: at . Right: the dimensionless acceleration power amplitude . Triangles mark and circles mark ; pale curves begin at . The continuation requires nonlinear density evolution, and the finite cumulant truncation develops spurious oscillations and sign changes at still larger wavenumbers. The plotted is the square root of the variance of per unit . Values of order occur in the pale continuation outside the linear density regime.
Local expansion multipoles
Hubble dipole and quadrupole
We calculate these local cosmographic coefficients for the matter flow resolved on a specified spatial scale. Smoothing suppresses the short-wavelength contributions to the linear variance discussed in § 3.4. We use a Gaussian window, , and compare multipoles of different angular rank through their sky rms. At a given observer, this measures the variation of a multipole across the sky; its amplitude also varies between observers. We denote the VPT–dust difference at the same present-day density field by . For a rank- PSTF contribution , angular orthogonality gives [29]
Unless stated otherwise, we quote the square root of the ensemble average of this mean square over observers. § 4.3 gives the distribution of amplitudes among individual observers. For the leading acceleration,
The integration limits and numerical details are given in appendix § 9.
We use , where the direct contribution from nonlinear modes is negligible, and a finer radius . The latter matches the long-wavelength second moment of a real-space top-hat extending to (appendix § 12.§ 3). The matched windows retain different short-wavelength power, and a catalogue’s effective resolution also depends on its selection and estimator [30].
The shear follows from the velocity divergence. With and ,
Define the change in velocity growth by . The corresponding changes in the Hubble multipoles are
There is no Hubble octupole. The shear correction changes an existing dust multipole, and the acceleration supplies a new matter-frame dipole.
At these radii the Hubble-shear correction has sky rms and , or and of the dust shear rms. It is the largest correction at zeroth order in redshift. The acceleration dipole has sky rms and . The detailed amplitudes and truncation test are in Table 1.
| Contribution | Tensor rms in | Sky rms in | Dust sky rms | Fraction [%] | [%] |
| Hubble shear, | 0.30 | 0.015 | |||
| Acceleration, | 0 | — | 0.016 | ||
| Acceleration correction, | 0 | — | 0.077 | ||
| Growth dipole, | 0.48 | 0.020 | |||
| Deceleration quadrupole, | 2.91 | 0.026 | |||
| Octupole, | 0.48 | 0.020 | |||
| Full dipole, through | 0.50 | 0.020 | |||
| Full quadrupole, through | 0.29 | 0.015 | |||
| Hubble shear, | 1.28 | 0.069 | |||
| Acceleration, | 0 | — | 0.072 | ||
| Acceleration correction, | 0 | — | 0.339 | ||
| Growth dipole, | 2.17 | 0.093 | |||
| Deceleration quadrupole, | 12.36 | 0.119 | |||
| Octupole, | 2.17 | 0.093 | |||
| Full dipole, through | 2.19 | 0.093 | |||
| Full quadrupole, through | 1.26 | 0.068 |
Table 1. Local VPT–dust corrections at fixed present-day density, using . Tensor and sky rms are defined in (E.8) and (E.7). “Fraction” is the correction’s sky rms as a percentage of the dust sky rms; the last column gives the change on including the third cumulant. Terms through use . Full multipoles add the signed contributions before taking the rms.
Deceleration multipoles
The next term in the distance expansion brings in derivatives of the local flow. Expanding eq. (7) gives
The linear deceleration multipoles follow from Paper I [9] and [15]. At linear scalar order the shear constraint gives . The deceleration dipole is then
where is the background deceleration parameter. The expansion-gradient term is also present for dust. Velocity dispersion changes it by altering the growth of density and velocity perturbations. We label this change grow; terms containing or its derivatives explicitly are labelled A. Both arise from the same pressure dynamics and must be added with their relative signs. Writing in units , the acceleration contribution through this order is
Here , and similarly for . The projected time derivative includes changes in both the magnitude and direction of the acceleration.
The quadrupole and octupole are
These expressions use shear propagation to replace the electric Weyl tensor. The octupole has the same kinematic form as for dust and gains its Vlasov correction through the changed shear. The contributions to the dipole and octupole from changes in growth have the same velocity-growth source. Their scalar transfers are proportional, although their different angular ranks give different sky amplitudes and make their components uncorrelated at the same observer in an isotropic ensemble. The transfers and angular normalisations are collected in appendix E.
To compare the leading and first redshift-dependent terms at a representative depth, we average the local Taylor expression with the volume weight of a cumulative sample ending at . This replaces by . The growth contribution to the dipole has sky rms at and at . The acceleration transfer evolves slowly today, so its additional contribution is suppressed by approximately relative to the intrinsic dipole (eq. (42)). The quadrupole is dominated by the Hubble shear, with a small partial cancellation from deceleration.
Successive redshift orders can introduce further spatial gradients, bringing factors of . At comoving distance , the combination compares the distance traversed with the spatial variation scale of a mode. The spatial Taylor expansion requires for the modes that dominate the signal. A small value of alone does not ensure this, since can be large. Our variance-weighted estimates give of order unity or larger at both radii, for dust as well as Vlasov matter (appendix E.3). The local coefficients are well filtered against nonlinear modes, but extrapolating them across the sample introduces spatial variations that the first redshift orders cannot describe reliably. Comparison with a finite-distance fit requires the light-cone calculation discussed in section 5.2. Numerical and exact-spacetime studies likewise show why local derivatives and coefficients fitted across a finite region must be distinguished [31, 32, 33].
At fixed smoothing radius, decreases towards the observer. The CF4 cut at addresses a different approximation: representing very nearby galaxy velocities by a smooth flow [10]. The linear Vlasov corrections found here provide no explanation for the deterioration of that fit. We interpret the low-redshift sensitivity primarily as a limit of this smooth-flow cosmographic description.
Variation between observers
For Gaussian initial conditions, the components of a linear multipole in an orthonormal PSTF basis are Gaussian. Statistical isotropy gives them equal variance, so the norm follows a chi distribution with degrees of freedom. Figure 4 shows the resulting variation between random observers. The quadrupole includes its leading Hubble-shear term, while the octupole begins at ; sky rms puts them on a common angular normalisation.
Figure 4. Probability distributions of VPT–dust corrections at a random observer, with and Gaussian smoothing at . The horizontal axis is the sky rms of each contribution to , so different multipole ranks can be compared directly. The density is per logarithmic amplitude interval; shading marks the central 68% interval and dots mark the ensemble rms. The acceleration and growth dipoles are shown separately; the acceleration curve includes its small correction. The quadrupole includes Hubble shear and deceleration.
Comparing two contributions at the same observer uses their joint distribution. The acceleration and growth dipoles have correlation coefficient 0.83 for the 30 window. Their joint covariance allows us to ask whether the intrinsic acceleration could dominate the change in the ordinary growth dipole. This happens for only of random observers in the model (appendix § 12.§ 4), predominantly where the growth dipole happens to be small. The acceleration’s 99th-percentile sky amplitude is . For comparison, the measured CF4a dipole has sky rms (§ 5.1). These probabilities describe random positions; observers in haloes or Local Group-like environments require a distribution selected for those environments.
Implications for dust cosmography
The intrinsic dipole
Kinetic pressure changes the cosmographic expansion in two ways. It alters the growth of density and velocity perturbations, changing the shear and expansion gradients that already enter dust cosmography. It also accelerates the mean matter flow, producing an intrinsic Hubble dipole. The latter is absent for dust in its rest frame and is the most direct signature of the additional matter dynamics.
The measured CMB-frame CF4a dipole in [10], for , has Legendre coefficient (their table 3). For a dipole proportional to , the sky rms is
Motion of the local matter frame relative to the CMB produces a Doppler contribution proportional to at lowest order, which dominates near the observer. The intrinsic acceleration dipole calculated here is defined in the matter frame and supplies a constant term at that order. At larger redshifts, and higher cosmographic coefficients also contribute appreciably to the observed dipole [10]. The measured dipole is three to four orders of magnitude larger than our fiducial intrinsic dipole. Even the largest intrinsic amplitude across the tested dispersion histories remains more than a thousand times smaller.
The cosmographic fit in eq. (6.2) of [10] includes the shear and deceleration anisotropies but fixes the intrinsic Hubble dipole to zero. Pressure changes coefficients already present in that fit, as well as generating the additional dipole.
The growth correction is substantially larger than the acceleration dipole. At the finer radius, the local correction to has an rms component 0.22 along a fixed axis, about half the fitted uncertainty of 0.4 in [10]. This gives the scale of a possible correction to the fit. Predicting its fitted value requires the finite-distance calculation, with the redshift weights and common multipole axis used in that analysis.
The acceleration-dipole test proposed in [40] can be implemented by allowing to vary alongside and over , using the full boosted distance model. Their distinct constant, and dipole contributions could constrain an intrinsic dipole and test whether fixing biases the inferred velocity or deceleration dipole. Through , this would give an empirical constraint on the acceleration of the fitted flow. Restricting all multipoles to a common axis constrains the acceleration component along that axis; a bound on its full magnitude would also require fitting its direction. Interpreting the result as a matter-frame acceleration then depends on the tracer relation in § 5.3.
Velocity dispersion and nonlinear evolution
The intrinsic dipole remains far below the measured CF4a dipole throughout our tested range of mean dispersions. Its amplitude scales approximately with the mean dispersion, while including the third cumulant changes the observer amplitudes by less than one percent (Table 1). These variations leave the hierarchy of corrections found in linear theory intact.
Nonlinear structure can change the pressure perturbations even at fixed mean dispersion. A long density mode changes the abundance and internal motions of the structures producing random velocities, and hence the dispersion in that region. Our example in appendix § 9.§ 3 shows that this modulation can change the acceleration dipole by order unity. Calculating it requires the nonlinear pressure–density and pressure–velocity correlations. Smoothing the linear fields removes short-scale fluctuations but cannot recover these correlations.
The Vlasov pressure describes random particle velocities about the mean flow at each point. Spatial averaging also produces effective pressure from variations of that flow and unresolved gravitational fields [26, 27, 34]. Appendix § 11 compares the coefficients relating these accelerations to the same long-wavelength density gradient. With linear short modes above , the perturbative coarse-grained coefficient agrees with our fiducial present-day VPT value to about 5%. In virialised objects, kinetic and gravitational contributions can cancel in the slowly varying, long-wavelength acceleration [26]; their relation to the local Vlasov pressure requires a nonlinear calculation.
Our results describe local Taylor coefficients. Comparison with a catalogue fit requires distances along the light cone, including the observer, source and integrated terms, with the catalogue’s angular selection, redshift weights and fitted multipole directions. The gradient test in § 4.2 shows why this is necessary even at low redshift. Relating the matter flow to the observed galaxies is a further physical input, which we discuss next.
Galaxy tracers
A galaxy population can define a different mean flow from the matter energy frame. Following the relative-velocity notation of Paper I, we write its 4-velocity as , with . Its acceleration satisfies
The intrinsic Hubble dipole of this flow is . For galaxies whose mean velocity follows the matter energy frame, , it is the dipole calculated here. For a geodesic galaxy congruence, the relative velocity instead obeys at linear order about FLRW, and the intrinsic dipole vanishes.
Predicting for a selected galaxy population requires a model for the galaxies’ velocity bias and velocity dispersion [35, 36]. The catalogue selection then determines which population contributes to the observed mean flow. Together, the matter dynamics and this tracer model determine the acceleration that enters a prediction for galaxy distances.
Conclusion
We have calculated how velocity dispersion changes the local acceleration, shear and expansion gradients by connecting the covariant equations of Paper I to linear Vlasov perturbation theory. At zeroth order in redshift, the largest correction is to the Hubble-shear quadrupole: 0.3% and 1.3% of the dust shear rms at the two Gaussian radii. The intrinsic acceleration dipole remains more than a thousand times smaller than the measured CMB-frame CF4a dipole, which includes the much larger contribution from motion between the matter and CMB frames.
At the next redshift order, the growth-induced dipole can exceed the shear correction. Pressure therefore affects the deceleration anisotropies already present in dust cosmographic fits more strongly than it generates an intrinsic dipole.
The numerical amplitudes rely on the mean dispersion history estimated from halo structure and abundance in [21], which we adopt as an input to the linear VPT equations. Nonlinear evolution can change the pressure perturbations, and a prediction for a galaxy sample also depends on how its mean flow follows the matter (§ 5.3). The VPT calculation assumes an FLRW background and cannot independently test the cosmological principle; its numerical predictions also depend on the adopted CDM cosmology.
At fixed smoothing radius and observer velocity, the intrinsic and growth terms decrease relative to the Doppler dipole as and , respectively. The linear Vlasov corrections therefore provide no explanation for the deterioration of the CF4 fit below . We interpret this sensitivity primarily as a limit of the smooth-flow cosmographic description. The size of also shows that the local Taylor expansion lacks a small gradient parameter at our representative depth, for dust as well as Vlasov matter. Connecting these local coefficients to a catalogue fit requires a finite-distance calculation.
Covariant cosmography itself does not assume an FLRW background. CMB isotropy and the Copernican principle do not establish homogeneity when the fundamental observers accelerate [40]. The intrinsic-dipole test described in § 5.1 would measure for the fitted flow and establish how fixing affects the inferred velocity and deceleration dipole. It would directly test the geodesic-flow assumption used in arguments for the cosmological principle.
Acknowledgments
We thank Christian Marinoni for helpful comments on the manuscript. CC is supported by STFC (UK) grant ST/X000931/1. CC also acknowledges OpenAI for providing complimentary access to ChatGPT and Codex through the ChatGPT for Academic Researchers programme. The authors used OpenAI ChatGPT Astra 6.0 to assist with algebraic consistency checks, numerical implementation, figure preparation, literature comparisons, and manuscript drafting and editing. All analytical results were independently rederived and all numerical results reproduced using separate code developed with Anthropic’s Claude Opus 5.5. Code to reproduce the results is available on request from the authors. The authors take full responsibility for the calculations, interpretation and final text.
Velocity cumulants and relativistic equations
Velocity cumulants
Let denote the phase-space distribution, where is the comoving momentum per unit mass. It is normalised by . The cumulant generating function is
The zeroth cumulant gives , the first gives the bulk peculiar velocity and the second gives the velocity-dispersion tensor. Writing the fourth central moment as gives
Thus setting the connected fourth cumulant to zero leaves the products of second moments in the fourth central moment.
For scalar perturbations, each PSTF moment takes the form
and its divergence and projected tracefree gradient are
Both rules follow from . For , the divergence gives , as used in eq. (19).
Linear scalar equations in Poisson gauge
The matter-frame derivative of a scalar with nonzero background is at linear order. Our scalar conventions are
The kinematic relations are
The scalar momentum conservation equation therefore includes the projected background-pressure derivative,
The tracefree spatial Einstein equation fixes the potential difference. Substituting it into the trace equation gives
These fix the potential and anisotropic-pressure conventions [37]; no barotropic relation between and is assumed. Rest-mass conservation and momentum conservation give
Multiplying the Euler equation by gives . We differentiate the continuity equation and substitute this expression for . Using in the Hubble damping term, then subtracting from both sides, gives
On subhorizon scales, the Poisson relation cancels the square bracket. The remaining metric terms are smaller than the ordinary dust terms by . The pressure-dependent metric corrections and the projected contribution are also smaller than the spatial pressure contribution by this factor.2 Dropping these relativistic corrections and evolving the dispersion with eq. (24) gives the two equations used in section 3.3, at :
Here and are defined in eqs. (18) and (26). The pressure gradient opposes gravitational growth, while the evolving density and velocity generate the pressure perturbation.
Numerical implementation
We take the mean dispersion history from the halo-based fit of [21], evaluated using the dust growth factor in the adopted CDM background. For each wavenumber we then evolve the density, velocity and pressure perturbations about that mean, including the third cumulant where specified in section 3.3. We normalise the solutions to the same present-day density field, obtain the acceleration and cosmographic coefficients, and integrate their variances with the chosen windows.
The mean dispersion used in the calculation
Velocity dispersion is a local moment of the phase-space distribution, as in eq. (10). Nonlinear collapse and shell crossing generate random motion about the local mean flow. Reference [21] estimates the mean dispersion from halo velocity-dispersion profiles, using Navarro–Frenk–White density profiles [38] and a halo mass function calibrated to -body simulations. For their fiducial halo boundary of , the fitted history is
with , , , and . The halo estimate underlying this fit is [18, 21]
Here is the local dispersion tensor inside a halo. The integral extends over to , including the halo outskirts, and is normalised by the virial volume . The factor gives the fraction of matter in each halo mass interval. We adopt the fitted history in eq. (B.1) for every occurrence of and in the perturbation equations. It supplies the mean random motion generated by nonlinear structure; the coupled equations then calculate its effect on the density and velocity perturbations.3 At linear order we set , giving today and allowing us to neglect pressure in the background expansion. In the nonlinear field, also includes the correlation ; the fitted mean dispersion alone does not fix it.
Evolution equations and initial conditions
We use to write the evolution compactly, with , and define . This is the normalised velocity divergence, distinct from the covariant expansion . The pressure variable is from § 3.3. At its evolution also contains the longitudinal divergence of the third cumulant,
This term transports dispersion between neighbouring points. With the mean fourth cumulant set to zero, the system we integrate is [18]
The coefficients and are defined in § 3.3. At we set and solve the first three equations. The density and velocity equations describe gravitational growth opposed by pressure gradients; the last two describe how the pressure changes and is transported. Specifying fixes their time-dependent coefficients without specifying the perturbation .
We start at , when , and select the dust growing mode with its associated pressure perturbation. Writing , the initial data are
The initial condition is used only at . These expressions follow by taking , and at early times; the factors in their denominators combine the growth rates with the damping terms in (B.4).
The code integrates in , multiplying each right-hand side by , using an adaptive eighth-order Runge–Kutta method. Time derivatives needed for the deceleration multipoles are evaluated from the same equations. At the end we divide by to compare VPT and dust at fixed present-day density, as in (30).
We use the baryon-corrected Eisenstein–Hu no-wiggle spectrum [28], normalised to in the stated cosmology, with . The variance integrals use 4001 logarithmically spaced wavenumbers from to , including the cutoff values in Table 3. This upper limit lies below the first zero of for every cumulant order and dispersion normalisation tested, so the division by final density remains regular. The plotted kernels extend further to show their high- behaviour, outside the domain used for variances. Independent integration in physical moment variables checks the solutions and the amplitudes in Table 1. The radial integrals use 2401 redshifts over ; doubling this resolution changes the rms values at by less than fractionally.
| [] | Cumulative vector rms | Outer-shell vector rms |
| 0.267 | ||
| 0.36 | ||
| 0.5 | ||
| 0.75 | ||
| 1 | ||
| 1.5 |
Table 3. Cutoff dependence of source-position dipole-vector rms for , using . The lowest cutoff is and the second is . Converting either column to sky rms of divides it by .
Production in overdense regions
The mean history fixes how much dispersion is produced on average, but an overdense region may produce more random motion than an underdense one. To estimate the importance of this omitted effect, we keep the same and add an isotropic production term to the local dispersion equation. At this changes the pressure equation to
The dimensionless constant specifies how production changes with the long-wavelength density: positive values add production in overdense regions, and negative values reduce it. Our fiducial calculation has . This is a separate sensitivity test, with no fitted value assigned to .4 For , the acceleration sky rms at is respectively , and . Thus production correlated with the long mode can change its acceleration by order unity even when the mean dispersion is fixed. Spatial smoothing does not remove this contribution to the long mode.
Effect of the third cumulant
The third cumulant describes an asymmetry of the local velocity distribution and transports dispersion. We test its importance by solving eq. (B.4) with and without , keeping the mean history, present-day density spectrum and windows fixed. The mean fourth cumulant is zero in both calculations. This retains the products of second moments in the fourth central moment, as shown in eq. (A.2).
Figure 5 shows the fractional change in . It reaches about at and grows on shorter scales. The windows suppress most of this difference. For an rms amplitude , we quote the absolute percentage change
Figure 5. Fractional change in when the third cumulant is included. The mean fourth cumulant is set to zero; in the plot denotes its one-component value divided by the squared one-component dispersion. The change is 0.07% at , 0.9% at and 3.9% at . The pale continuation begins at .
The results are
| Window | |
| Gaussian, – | |
| real-space top-hat, – | |
| cumulative, | |
| outer shell, |
Table 4.
The local observer multipoles change by less than one percent at both Gaussian radii (Table 1). Their time derivatives come directly from the evolution equations in appendix § 9.§ 2.
This comparison tests the cumulant truncation within linear evolution. It does not test nonlinear mode coupling, which becomes important at , before . The Gaussian observer windows suppress these modes strongly. The top-hat and radial windows retain more power from them, as quantified in appendix § 13, and their amplitudes depend on the common upper cutoff.
Acceleration and coarse-grained dynamics
In the effective field theory of large-scale structure, spatial averaging replaces unresolved motions and gravitational fields by an effective pressure tensor [26, 27]. Even a dust flow with one velocity at each point acquires such a tensor when neighbouring points with different velocities are averaged together. Its divergence accelerates the smoothed, mass-weighted flow. Our local Vlasov pressure instead comes from different particle velocities at the same point. Spatially averaging that Vlasov flow would add the unresolved bulk-motion and gravitational contributions as well.
For a sharp comoving cutoff , let the smoothed velocity be the smoothed momentum density divided by the smoothed mass density. Its acceleration relative to the resolved gravitational field is . The part correlated with a long density mode has the form
Thus enters the long-mode Euler equation as an effective sound-speed coefficient. In EFT it includes short-scale physics that must be matched to a nonlinear calculation or data. We compare only the perturbative contribution from linear short modes. The separation scale is independent of the observational smoothing radius .
The required one-loop result is the ultraviolet limit of with Einstein–de Sitter spatial kernels, given in appendix B of [34]. Since is twice the cross-correlation of the linear and third-order density fields, its contribution from corresponds to
Here is the pressureless linear spectrum extrapolated to today. To recover the acceleration producing this density correction, we apply the CDM density evolution operator, with :
The second identity follows by differentiating and using . Equating gives
The superscript denotes the contribution quadratic in short-mode amplitude. Here is the three-dimensional variance of their linear bulk velocity. This expression keeps CDM growth with Einstein–de Sitter spatial kernels and reduces to in Einstein–de Sitter.
At we obtain
The perturbative coarse-grained value is 5.4% larger than the fiducial VPT value. Raising to lowers it to . This agreement concerns the specified linear short-mode estimate. The full nonlinear coefficient requires the kinetic and gravitational contributions of the unresolved structures, including their virial cancellation [26].
Observer multipoles and amplitudes
Multipole transfers
Pressure changes the observer multipoles in two linked ways. It alters the growth of density and velocity, changing the expansion and shear terms already present for dust, and it adds terms containing the acceleration explicitly. We label these contributions grow and , respectively, as in section 4.2 and Figure 3. We keep them separate to identify where the change enters the cosmographic expression, then add them to obtain the total correction.
Figure 3. VPT–dust corrections to the deceleration multipoles at . The upper panel shows changes in the ordinary growth and shear terms; the lower panel shows terms containing acceleration explicitly. There is no explicit acceleration term in the octupole at linear scalar order. Pale curves begin at and become lighter beyond , marked by triangles and circles respectively.
Writing the VPT–dust deceleration multipoles as
gives, at ,
Here is the difference between VPT and dust velocity growth at the same final density. The octupole contains only a shear gradient, so its correction is entirely in the grow contribution. The scalar transfers satisfy
as follows from their common velocity-growth source.
For the volume-weighted cumulative sample, we replace in the local Taylor expression by . Define as the scalar coefficient transferring to , with the PSTF tensor and the factor for odd displayed as in (E.1). The coefficients are
The signed contributions are combined before taking their rms, retaining their cross-correlations. The dust transfers use the pressureless growth factor , with ; the shear transfer is . The deceleration dipole and octupole have scalar transfers and . The dust quadrupole is
using the pressureless growth equation. Its two terms partly cancel today, giving about . This small dust coefficient explains why the fractional change in the deceleration quadrupole can be appreciably larger than the fractional change in the Hubble shear.
Angular normalisation
For the scalar PSTF basis, contracting one index with lowers the rank and multiplies the result by (appendix § 8.§ 1; [29]). Trace terms vanish against a PSTF tensor, so repeated contraction gives
The norms are , and for ranks one to three. Combining this contraction with the sky average (36) gives
The corresponding mean-square tensor norm is
Thus the tensor rms is converted to sky rms by . Together with eq. (E.3), this gives the growth-octupole to growth-dipole sky-rms ratio . For the leading acceleration, the conversion is
where is the rms magnitude of the smoothed acceleration vector and the spatial window must be the same on both sides.
Numerical amplitudes and resolution
Table 1 uses Gaussian radii of and . The finer radius comes from matching a top-hat extending to . Expanding the two windows at small gives
With , this gives , quoted as 13.4 throughout. The matching fixes their large-scale smoothing; their short-wavelength tails still differ.
The local terms use the cumulative volume weight, for which .
For each contribution with transfer , the rms value of the spatial expansion parameter is
The same variance weights determine the fraction above . The variance-weighted rms of ranges from 1.1 to 1.5 at and from 2.2 to 3.2 at . The corresponding dust transfers give 0.81–1.07 and 1.63–2.25 at the same two radii. Modes above supply less than 0.1% of the observer variances, so the gradient limitation is distinct from direct short-wavelength contamination of these local coefficients.
The growth correction to the deceleration dipole follows from its contribution to by multiplying by . At the two radii, its sky rms is 0.010 and 0.22, and its rms vector magnitude is 0.018 and 0.38. In an isotropic ensemble the rms component along a specified axis equals the sky rms. This fixed-axis quantity is the comparison used in § 5.1. At the full dipole correction has sky rms , rising to at the largest tested dispersion normalisation: about 3% and 6% of the measured CF4a sky rms. These use a cumulative weight ending where the CF4a sample begins, so they compare amplitude scales rather than identical estimators. At the opposing deceleration term reduces the shear quadrupole rms by 1.6%, and the explicit acceleration correction is of the leading dipole.
Cosmic variance
We use the chi distributions described in § 4.3, with the angular normalisation of eq. (36). Figure 4 shows probability density per logarithmic interval in sky amplitude, with shading for the central 68% interval.
The two dipoles have transfers and , as in eq. (E.4). Their covariance is obtained by replacing in eq. (E.7) with the product of these two transfers. At the correlation coefficient is 0.83, giving
Observers with mainly have a small growth dipole. The 99th percentile of the acceleration sky amplitude is , compared with the growth-dipole sky rms .
Spatial and radial windows
A window determines which spatial scales contribute to a quoted acceleration amplitude. We first smooth the field around the observer and compare it with the bulk velocity on the same scale. We then average the radial acceleration evaluated at source positions, where changes in direction and sign along the line of sight can cancel. These two calculations explain the spatial and depth dependence in Figure 6–Figure 8. The second is a test of the acceleration field’s spatial variation; a distance prediction also needs the remaining light-cone terms.
Figure 6. Present-day acceleration after spatial smoothing. Left: three-dimensional vector rms for Gaussian and real-space top-hat windows. Divide by for sky rms in . The shaded region marks –; equal radii in the two windows represent different resolutions. Right: fractions of the variance above (solid) and (dashed). The horizontal line marks 10%.
Figure 8. Acceleration dipole-vector rms for sources throughout the volume and in the outer shell . The orange curve is the bulk-flow Doppler scale , with . Filled circles mark ; the horizontal line marks . Pale acceleration curves receive more than 10% of their variance from . Divide by to obtain sky rms in .
Observer smoothing
For an isotropic spatial window , the smoothed acceleration obeys
Thus is the rms magnitude of the three-dimensional vector, and each Cartesian component has variance . We use
The second-moment match in eq. (E.10) associates the Gaussian radius with a top-hat radius . Their Fourier tails differ. Figure 6 shows the acceleration amplitude and the variance fractions above and . For Gaussian radii –, less than of the acceleration variance comes directly from . The top-hat retains more short-wavelength power: the corresponding fraction is – at –, at and at . The small-radius top-hat estimates depend strongly on the linear treatment of those modes.
The bulk velocity on the same spatial scale follows from the subhorizon continuity equation, which gives and hence
For top-hat smoothing, the two variances are
In Figure 7 we set to compare these two fields on a scale set by survey depth. At ,
Figure 7. Present-day matter fields after real-space top-hat smoothing on . The orange curve is the rms bulk velocity and the blue curve is the rms dimensionless acceleration . These are three-dimensional vector rms values; the acceleration contribution to sky rms in is . Their ratio is approximately 1.3% at , marked by the filled circles.
The acceleration-vector rms corresponds to sky rms in by eq. (E.9). The Gaussian value at is smaller because it filters more broadly than a top-hat with the same numerical radius. The Gaussian with gives , close to the top-hat value despite their different Fourier tails.
The corresponding Doppler dipole is enhanced by the inverse source redshift. At the outer radius, its characteristic amplitude is
At , , so the acceleration-to-Doppler ratio is only . Here the smoothing radius changes with depth. At fixed radius the intrinsic contribution instead falls relative to Doppler in direct proportion to , as discussed in § 2.
Source-position projection
Figure 8 compares the acceleration dipole averaged over all sources within a depth with that from sources in an outer shell. The broader average allows more cancellation between nearby and distant regions. We evaluate at the source positions, using a normalised redshift weight . The dipole vector of this sky pattern is
| Gaussian vector rms | Top-hat vector rms | |
| 5 | ||
| 10 | ||
| 20 | ||
| 40 |
Table 2. Spatially smoothed acceleration-vector rms at and . Dividing by gives sky rms in .
The factor 3 recovers from a constant vector field. For a longitudinal Fourier mode, the angular integral gives : the radial projection supplies the Bessel derivative. Inserting the transfer and averaging over the density field therefore gives
The radial integral is taken before squaring, so contributions from different distances can cancel. Since , a window at the observer recovers the unsmoothed vector variance. The sky rms of the projected dipole in is .
For constant comoving source density and equal source weights, the number of sources in a redshift interval is proportional to . We therefore use
with zero weight outside this interval. The cumulative sample uses and the outer shell uses .
At , their dipole-vector rms values are and , respectively. The outer shell has less radial cancellation and receives 59% of its variance from , compared with 16% for the cumulative sample. These estimates consequently depend on the extrapolation of linear evolution to short scales. Both projected acceleration amplitudes remain far below the bulk-flow Doppler scale shown in the figure.
Ultraviolet cutoff
The radial calculation uses . Table 3 shows its dependence on this cutoff at . The cumulative projection changes modestly above , whereas the outer-shell value changes substantially because it retains more short-wavelength power. The top-hat cumulant comparison is regulated at the same cutoff in appendix § 10.
References
- [1]T. Buchert and A. Dominguez, Modeling multi-stream flow in collisionless matter: approximations for large-scale structure beyond shell-crossing, Astron. Astrophys. 335 (1998) 395–402, [astro-ph/9702139].arxiv.org/abs/astro-ph/9702139
- [2]P. Valageas, Dynamics of gravitational clustering I. Building perturbative expansions, Astron. Astrophys. 379 (2001) 8, [astro-ph/0107015].
- [3]S. Pueblas and R. Scoccimarro, Generation of Vorticity and Velocity Dispersion by Orbit Crossing, Phys. Rev. D 80 (2009) 043504, [arXiv:0809.4606].DOI
- [4]A. Erschfeld and S. Floerchinger, Evolution of dark matter velocity dispersion, JCAP 06 (2019) 039, [arXiv:1812.06891].DOI
- [5]A. Erschfeld, S. Floerchinger, and M. Rupprecht, General relativistic non-ideal fluid equations for dark matter from a truncated cumulant expansion, Phys. Rev. D 102 (2020), no. 6 063520, [arXiv:2005.12923].DOI
- [6]A. Erschfeld and S. Floerchinger, Dark matter vorticity and velocity dispersion from truncated Dyson-Schwinger equations, JCAP 02 (2024) 053, [arXiv:2305.18517].
- [7]C. Nascimento and M. Loverde, Cosmological perturbation theory for large scale structure in phase space, JCAP 06 (2025) 002, [arXiv:2410.05389].DOI
- [8]H. Heisler, M. Sipp, and M. Bartelmann, Phase-space perturbation theory for cosmic large-scale structure, JCAP 04 (2026) 012, [arXiv:2512.07396].DOI
- [9]C. Clarkson and R. Maartens, Beyond dust in local cosmography: covariant kinetic theory, arXiv:2610.04284.arxiv.org/abs/2610.04284
- [10]B. Kalbouneh, C. Marinoni, R. Maartens, J. Bel, J. Santiago, C. Clarkson, M. Sarma, and J.-M. Virey, The anisotropic expansion rate of the local Universe and its covariant cosmographic interpretation, JCAP 03 (2026) 084, [arXiv:2510.02510].arxiv.org/abs/2510.02510
- [11]R. B. Tully, E. Kourkchi, H. M. Courtois, G. S. Anand, J. P. Blakeslee, D. Brout, T. de Jaeger, A. Dupuy, D. Guinet, C. Howlett, J. B. Jessen, D. Pomarède, L. Rizzi, D. Rubin, K. Said, D. Scolnic, and B. E. Stahl, Cosmicflows-4, Astrophys. J. 944 (2023) 94, [arXiv:2209.11238].DOI
- [12]J. Kristian and R. K. Sachs, Observations in cosmology, Astrophys. J. 143 (1966) 379–399.
- [13]M. A. H. MacCallum and G. F. R. Ellis, A Class of Homogeneous Cosmological Models. II. Observations, Commun. Math. Phys. 19 (1970) 31–64.
- [14]C. A. Clarkson, On the Observational Characteristics of Inhomogeneous Cosmologies: Undermining the Cosmological Principle or Have Cosmologists Put All Their EGs in One Basket, PhD thesis, University of Glasgow (2000), astro-ph/0008089.
- [15]A. Heinesen, Multipole decomposition of the general luminosity distance ‘Hubble law’ – a new framework for observational cosmology, JCAP 05 (2021) 008, [arXiv:2010.06534].DOI
- [16]R. Maartens, J. Santiago, C. Clarkson, B. Kalbouneh, and C. Marinoni, Covariant cosmography: the observer-dependence of the Hubble parameter, JCAP 09 (2024) 070, [arXiv:2312.09875].DOI
- [17]E. Pastén, Redshift Dipoles from Non-Geodesic Observer Congruences in Covariant Cosmology, arXiv:2603.20963.
- [18]M. Garny, D. Laxhuber, and R. Scoccimarro, Perturbation theory with dispersion and higher cumulants: Framework and linear theory, Phys. Rev. D 107 (2023), no. 6 063539, [arXiv:2210.08088].DOI
- [19]M. Garny, D. Laxhuber, and R. Scoccimarro, Perturbation theory with dispersion and higher cumulants: Nonlinear regime, Phys. Rev. D 107 (2023), no. 6 063540, [arXiv:2210.08089].DOI
- [20]M. Garny and R. Scoccimarro, Vlasov perturbation theory and the role of higher cumulants, Phys. Rev. D 111 (2025), no. 12 123509, [arXiv:2502.20451].arxiv.org/abs/2502.20451
- [21]M. Garny, D. Laxhuber, and R. Scoccimarro, Vlasov perturbation theory applied to ΛCDM, Phys. Rev. D 112 (2025), no. 2 023556, [arXiv:2505.02907].arxiv.org/abs/2505.02907
- [22]A. Lewis and A. Challinor, Evolution of cosmological dark matter perturbations, Phys. Rev. D 66 (2002) 023531, [astro-ph/0203507].DOI
- [23]F. Bernardeau, S. Colombi, E. Gaztañaga, and R. Scoccimarro, Large-Scale Structure of the Universe and Cosmological Perturbation Theory, Phys. Rept. 367 (2002) 1–248, [astro-ph/0112551].DOI
- [24]M. Buehlmann and O. Hahn, Large-Scale Velocity Dispersion and the Cosmic Web, Mon. Not. Roy. Astron. Soc. 487 (2019), no. 1 228–245, [arXiv:1812.07489].DOI
- [25]Planck Collaboration, N. Aghanim et al., Planck 2018 results. VI. Cosmological parameters, Astron. Astrophys. 641 (2020) A6, [arXiv:1807.06209]; erratum: Astron. Astrophys. 652 (2021) C4, doi:10.1051/0004-6361/201839910.
- [26]D. Baumann, A. Nicolis, L. Senatore, and M. Zaldarriaga, Cosmological Non-Linearities as an Effective Fluid, JCAP 07 (2012) 051, [arXiv:1004.2488].DOI
- [27]J. J. M. Carrasco, M. P. Hertzberg, and L. Senatore, The Effective Field Theory of Cosmological Large Scale Structures, JHEP 09 (2012) 082, [arXiv:1206.2926].DOI
- [28]D. J. Eisenstein and W. Hu, Baryonic Features in the Matter Transfer Function, Astrophys. J. 496 (1998) 605, [astro-ph/9709112].
- [29]T. Gebbie and G. F. R. Ellis, 1+3 Covariant Cosmic Microwave Background Anisotropies I: Algebraic Relations for Mode and Multipole Expansions, Annals Phys. 282 (2000) 285–320, [astro-ph/9804316], doi:10.1006/aphy.2000.6033.DOI
- [30]J. Adamek, C. Clarkson, R. Durrer, A. Heinesen, M. Kunz, and H. J. Macpherson, Towards cosmography of the local universe, Open J. Astrophys. 7 (2024) 118782, doi:10.33232/001c.118782, [arXiv:2402.12165].DOI
- [31]H. J. Macpherson and A. Heinesen, A theoretical prediction for the dipole in nearby distances using cosmography, Open J. Astrophys. 8 (2025) 150319, doi:10.33232/001c.150319, [arXiv:2507.01095].DOI
- [32]M. Sarma, C. Marinoni, B. Kalbouneh, C. Clarkson, and R. Maartens, Covariant cosmography in the presence of local structures: comparing exact solutions and perturbation theory, JCAP 04 (2026) 069, [arXiv:2510.03517].DOI
- [33]M. Hills and A. Heinesen, Cosmography with Λ-Szekeres models, JCAP 06 (2026) 016, doi:10.1088/1475-7516/2026/06/016, [arXiv:2601.16844].DOI
- [34]C. Nascimento, D. Jamieson, M. McQuinn, and M. Loverde, A semi-analytic estimate for the effective sound speed counterterm in the EFTofLSS, JCAP 02 (2025) 023, [arXiv:2410.11949].DOI
- [35]M. Biagetti, V. Desjacques, A. Kehagias, D. Racco, and A. Riotto, The Halo Boltzmann Equation, JCAP 04 (2016) 040, [arXiv:1508.07330].DOI
- [36]T. Matsubara, Velocity bias and the nonlinear perturbation theory of peaks, Phys. Rev. D 100 (2019), no. 8 083504, [arXiv:1907.13251].DOI
- [37]C.-P. Ma and E. Bertschinger, Cosmological perturbation theory in the synchronous and conformal Newtonian gauges, Astrophys. J. 455 (1995) 7–25, [astro-ph/9506072].DOI
- [38]J. F. Navarro, C. S. Frenk, and S. D. M. White, A universal density profile from hierarchical clustering, Astrophys. J. 490 (1997) 493–508, [astro-ph/9611107].
- [40]C. A. Clarkson and R. K. Barrett, Does the isotropy of the CMB imply a homogeneous universe? Some generalized EGS theorems, Class. Quant. Grav. 16 (1999) 3781–3794, [gr-qc/9906097], doi:10.1088/0264-9381/16/12/302.DOI







