Cosmographic Reconstruction

• Cosmographic reconstruction is a technique that directly derives key cosmic observables, such as the Hubble rate and luminosity distance, from data without assuming a specific dynamical model.
• It employs various series expansions, rational approximants like Padé and Chebyshev, and non-parametric methods including Gaussian processes to manage convergence and uncertainties.
• Extensions into chrono-cosmography enable the reconstruction of the Universe’s four-dimensional density field through Bayesian tomographic inversions and Hamiltonian Monte Carlo methods.

Cosmographic reconstruction is a family of inference procedures that reconstructs observable properties of the Universe directly from data, with minimal commitment to a specific dynamical model. In one established usage, it is a kinematic program: the scale factor, the Hubble rate, and cosmological distances are expanded around the present epoch and the coefficients are estimated from supernovae, BAO, Hubble data, quasars, or related probes. In another usage, the term extends to “chrono-cosmography,” where Bayesian forward models or tomographic inversions reconstruct the four-dimensional history of the density or gravitational-potential field from galaxy surveys or CMB anisotropies. Across these usages, the common structure is an observation-to-field or observation-to-kinematics map with explicit uncertainty propagation (Mehrabi et al., 2021, Jasche et al., 2014, Chung et al., 2 Jul 2025).

1. Kinematic basis and cosmographic variables

In standard cosmography, the expansion history is described without assuming a particular dark-energy fluid or modified-gravity Lagrangian. The central objects are the present-day Hubble, deceleration, jerk, and snap parameters,

$$H(t)\equiv\frac{1}{a}\frac{da}{dt},\qquad q(t)\equiv-\,\frac{1}{aH^2}\frac{d^2a}{dt^2},\qquad j(t)\equiv\frac{1}{aH^3}\frac{d^3a}{dt^3},\qquad s(t)\equiv\frac{1}{aH^4}\frac{d^4a}{dt^4},$$

with present-day values $H_0,q_0,j_0,s_0$. The Hubble function is then expanded around $z=0$, for example as

$$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$

and the derivatives are re-expressed in terms of the cosmographic parameters (Mehrabi et al., 2021, Capozziello et al., 2022).

The luminosity distance admits a corresponding low-$z$ expansion. In flat FLRW, one form used in the literature is

$$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$

The same coefficients enter alternative distance definitions such as the photon-flux distance $d_F$ and angular-diameter distance $d_A$, which have distinct low-redshift series but all reduce at first order to $d_{L;F;A}\sim z/H_0$ (Capozziello et al., 2022, Capozziello et al., 2014).

A basic technical limitation is the convergence domain of the $z$-series. One strand of the literature states that a straightforward Taylor expansion in the standard redshift $H_0,q_0,j_0,s_0$0 converges only for $H_0,q_0,j_0,s_0$1, motivating the reparametrization

$H_0,q_0,j_0,s_0$2

for which $H_0,q_0,j_0,s_0$3 as $H_0,q_0,j_0,s_0$4. The $H_0,q_0,j_0,s_0$5-expansion mitigates, but does not remove, convergence and truncation problems at high redshift (Mehrabi et al., 2021).

2. Reconstruction strategies beyond the basic Taylor series

Because truncated Taylor series become inaccurate at $H_0,q_0,j_0,s_0$6, several reconstruction strategies have been developed. Rational cosmography replaces a polynomial truncation by Padé or Chebyshev approximants. The general Padé form is

$H_0,q_0,j_0,s_0$7

with coefficients chosen to match the Taylor data up to order $H_0,q_0,j_0,s_0$8. Chebyshev rational approximants similarly replace monomials by Chebyshev polynomials $H_0,q_0,j_0,s_0$9 (Capozziello et al., 2022).

In model-independent comparisons using Pantheon SNIa and Hubble data, cosmographic Taylor reconstruction has been contrasted with Gaussian Processes and Genetic Algorithms. In that comparison, the cosmographic approach was reported to be “not exact enough,” with Hubble-data estimates of $z=0$0 and $z=0$1 more than $z=0$2 away from the best result of $z=0$3CDM, while GP and GA remained within $z=0$4. The same analysis states that Taylor cosmography provides biased results, especially at higher redshifts, whereas GP and GA yield smoother and more flexible reconstructions to $z=0$5 (Mehrabi et al., 2021).

Non-parametric Gaussian-process reconstruction has also been applied directly to $z=0$6, $z=0$7, and their derivatives. For the combined CC+SNIa+BAO+RSD dataset, one reported result is

$z=0$8

with all reconstructed functions compatible with $z=0$9CDM within $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$0 (Mukherjee, 2022).

At still higher redshift, orthogonalized logarithmic polynomials have been introduced using $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$1 or $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$2. In that framework, the fifth-order logarithmic series was reported to fit the Hubble diagram up to $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$3, with the fifth-order term $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$4 significant and the sixth-order coefficient consistent with zero. The same study reports a strong tension, at $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$5, between the concordance model and the Hubble diagram at $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$6, dominated by quasars at $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$7 (Bargiacchi et al., 2021).

3. Reconstruction of dark energy, modified gravity, and exact kinematic closures

A major application of cosmographic reconstruction is the recovery of effective gravitational or dark-energy sectors from kinematic data. A common strategy is to adopt a cosmographic parametrization of $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$8, express the geometric scalar $$H(z)=H_0+\left.\frac{dH}{dz}\right|_{0}z+\frac12\left.\frac{d^2H}{dz^2}\right|_{0}z^2+\cdots,$$9, $z$0, or $z$1 in terms of $z$2 and its redshift derivatives, rewrite the modified Friedmann equations as an ODE for $z$3, impose boundary conditions at $z$4, and then invert $z$5 scalar to obtain $z$6, $z$7, or $z$8 (Capozziello et al., 2022).

For $z$9 gravity in flat FRW, the Ricci scalar is written as

$$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$0

Imposing the local conditions $$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$1 and $$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$2, one analysis derived

$$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$3

Using Union 2.1 SNe and BAO at $$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$4, the same work reported $$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$5 from low-$$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$6 SNe, and for the luminosity-distance SN+BAO fit,

$$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$7

It further states that the pressure of the curvature dark-energy fluid is slightly lower than that associated with the cosmological constant, and that $$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$8 is compatible with slight late-time dynamics in the curvature sector (Capozziello et al., 2014).

For $$d_L(z)=\frac{1}{H_0}\Bigl[ z+\tfrac12(1-q_0)z^2-\tfrac16(1-q_0-3q_0^2+j_0)z^3+\tfrac1{24}(2-2q_0-15q_0^2-15q_0^3+5j_0+10q_0j_0+s_0)z^4+\mathcal O(z^5) \Bigr].$$9 cosmology, a model-independent Maclaurin expansion around $d_F$0 has been related directly to $d_F$1. In one Monte Carlo analysis based on Union 2.1, OHD, and an HST prior, the 68% credible best-fit values for model B were

$d_F$2

$d_F$3

with $d_F$4. The higher derivatives were interpreted there as tiny but nonzero departures from $d_F$5CDM (Aviles et al., 2013).

Cosmography has also been used to reconstruct scalar-field potentials directly. In a quintessence analysis, the slope and curvature variables

$d_F$6

were expressed as algebraic functions of $d_F$7, leading to a local expansion

$d_F$8

Using recent cosmographic fits, that work reported examples such as

$d_F$9

for Pantheon+, and

$d_A$0

for a DESI DR2 Chebyshev reconstruction (Chakraborty et al., 25 Mar 2026).

An exact kinematic closure is obtained when the jerk is assumed constant. In that case,

$d_A$1

with the lower bound $d_A$2 required for real exponents. In the flat $d_A$3CDM limit, $d_A$4, $d_A$5, $d_A$6, and the standard Hubble function is recovered. A joint OHD+BAO+Pantheon fit in this framework gave

$d_A$7

at 68% C.L. (Amirhashchi et al., 2018).

4. Chrono-cosmography of the large-scale structure

In a different but related usage, cosmographic reconstruction denotes the inference of spacetime structure itself rather than only the background expansion. In the BORG framework, the problem is posed as the recovery of an initial density contrast field $d_A$8 at cosmic scale factor $d_A$9, given a galaxy redshift catalogue $d_{L;F;A}\sim z/H_0$0 at $d_{L;F;A}\sim z/H_0$1, through the posterior

$d_{L;F;A}\sim z/H_0$2

The prior is a zero-mean Gaussian random field with covariance $d_{L;F;A}\sim z/H_0$3 defined by a fiducial power spectrum $d_{L;F;A}\sim z/H_0$4, and the forward model is second-order Lagrangian perturbation theory (2LPT) (Jasche et al., 2014).

The galaxy data are modeled voxelwise as an inhomogeneous Poisson process,

$d_{L;F;A}\sim z/H_0$5

with survey response

$d_{L;F;A}\sim z/H_0$6

The inference accounts for luminosity-dependent bias, noise calibration, survey geometry, and selection effects; six absolute-magnitude bins each carry their own power-law bias index $d_{L;F;A}\sim z/H_0$7 and noise level parameter $d_{L;F;A}\sim z/H_0$8 (Jasche et al., 2014).

Because $d_{L;F;A}\sim z/H_0$9 lives in a $z$0-dimensional space, BORG uses Hamiltonian Monte Carlo. In the SDSS DR7 application, the reconstructed volume is a cubic box of side $z$1 on a $z$2 grid, corresponding to $z$3 resolution and $z$4 parameters. Over $z$5 HMC samples, each requiring $z$6 2LPT evaluations, the method produces posterior means and standard deviations for the present density field, reconstructs large-scale velocities, and generates a full time series $z$7 from $z$8 to $z$9. This was presented as the first quantitative inference of plausible formation histories of the dynamic large-scale structure underlying the observed galaxy distribution (Jasche et al., 2014).

5. Tomographic inversion of the observable Universe

Tomographic cosmographic reconstruction extends the program to line-of-sight integrals. For the integrated Sachs–Wolfe effect, the observed anisotropy is modeled as a light-ray transform of the scalar potential,

$H_0,q_0,j_0,s_0$00

or, in the straight-ray formulation,

$H_0,q_0,j_0,s_0$01

The inverse problem is to recover the 3D potential $H_0,q_0,j_0,s_0$02 from noisy, partial-sky measurements of $H_0,q_0,j_0,s_0$03 (Chung et al., 2 Jul 2025).

The mathematical treatment uses Sobolev spaces and Fourier integral operator calculus. The main stable-inversion theorem states that $H_0,q_0,j_0,s_0$04 and a second back-projection $H_0,q_0,j_0,s_0$05 uniquely determine the relevant Cauchy data in the visible ring $H_0,q_0,j_0,s_0$06, with a stability estimate in Sobolev norms. Numerically, the work discretizes $H_0,q_0,j_0,s_0$07 on an $H_0,q_0,j_0,s_0$08 grid with $H_0,q_0,j_0,s_0$09, uses $H_0,q_0,j_0,s_0$10 time steps, and reconstructs by LSQR, $H_0,q_0,j_0,s_0$11-FISTA, or IGMRF regularization. Reported errors are $H_0,q_0,j_0,s_0$12 for full data and $H_0,q_0,j_0,s_0$13–$H_0,q_0,j_0,s_0$14 for partial data with regularization, with stable performance under noise up to $H_0,q_0,j_0,s_0$15 (Chung et al., 2 Jul 2025).

A related phase-space program reconstructs the joint evolution of background and growth observables in

$H_0,q_0,j_0,s_0$16

Using a three-parameter Padé approximant for $H_0,q_0,j_0,s_0$17 in $H_0,q_0,j_0,s_0$18, a semi-cosmographic equation of state $H_0,q_0,j_0,s_0$19, and SDSS IV BAO+RSD data, that approach reconstructs phase trajectories and identifies three special redshifts: $H_0,q_0,j_0,s_0$20 The same analysis reports departures from $H_0,q_0,j_0,s_0$21CDM at low redshift, including $H_0,q_0,j_0,s_0$22 for $H_0,q_0,j_0,s_0$23 at $H_0,q_0,j_0,s_0$24 and $H_0,q_0,j_0,s_0$25 above $H_0,q_0,j_0,s_0$26 by $H_0,q_0,j_0,s_0$27 (Chavan et al., 17 Jun 2025).

6. Limitations, tensions, and disputed conclusions

The principal methodological limitation is truncation and convergence. The literature repeatedly emphasizes that $H_0,q_0,j_0,s_0$28-series cosmography becomes inaccurate beyond $H_0,q_0,j_0,s_0$29, that higher derivatives are poorly constrained, and that propagated errors grow rapidly for $H_0,q_0,j_0,s_0$30. This is why Padé, Chebyshev, logarithmic-polynomial, GP, and GA reconstructions have been introduced (Capozziello et al., 2022, Mehrabi et al., 2021).

The empirical conclusions are not uniform. One fourth-order luminosity-distance analysis of Union, Constitution, and Union 2 SNe found a considerable probability for $H_0,q_0,j_0,s_0$31, namely $H_0,q_0,j_0,s_0$32 with the $H_0,q_0,j_0,s_0$33-redshift expansion and $H_0,q_0,j_0,s_0$34 with the fourth-order $H_0,q_0,j_0,s_0$35-expansion restricted to $H_0,q_0,j_0,s_0$36. In that reconstruction, $H_0,q_0,j_0,s_0$37 could describe a transient acceleration, with a transition around $H_0,q_0,j_0,s_0$38 and a peak acceleration near $H_0,q_0,j_0,s_0$39–$H_0,q_0,j_0,s_0$40; the same work states that fourth order is essential because a cubic truncation cannot produce such behavior (Guimarães et al., 2010).

By contrast, Padé-based analyses do not always support a high-redshift tension. One study using mock and real Hubble diagrams of SNIa, QSOs, GRBs, and BAO concluded that Padé cosmographic approaches do not reveal any cosmographic tension with the standard model and that this conclusion is confirmed by AIC. The same work reported that Padé reconstructions are sufficiently suitable even at high redshift, whereas the inferred $H_0,q_0,j_0,s_0$41 from the snap parameter can exceed that inferred from $H_0,q_0,j_0,s_0$42 and from Planck-$H_0,q_0,j_0,s_0$43CDM values in redshift-bin analyses (Pourojaghi et al., 2022).

Model-specific Padé reconstructions can also suppress purported dynamical features. In VCDM, a Padé $H_0,q_0,j_0,s_0$44 analysis with cosmic chronometers, DESI BAO, and Type Ia supernovae reported

$H_0,q_0,j_0,s_0$45

and found that a previously claimed transition feature is not observed within the Padé reconstruction, suggesting sensitivity to parametrization. The same study states that VCDM effectively mimics $H_0,q_0,j_0,s_0$46CDM at the background level when constrained cosmographically (Bhoi et al., 9 Jun 2026).

Taken together, these results show that cosmographic reconstruction is not a single estimator but a methodological class. Its outputs depend on the chosen expansion variable, truncation order, rational approximant, regularization scheme, and dataset combination. The literature therefore treats cross-validation across multiple reconstruction schemes as a central requirement, especially once the analysis moves beyond low redshift or from background kinematics to tomographic field inference (Mehrabi et al., 2021, Chung et al., 2 Jul 2025).