An analysis of seven DESI Full-Shape constraints reports a modest upward trend in Ω_m(z), the matter-density parameter tracked across different redshifts. The fitted slope was 0.022 ± 0.012, a 1.8σ result. That is a signal worth watching, but it does not yet settle whether the parameter changes with redshift or is consistent with a constant value. Nor does it directly detect new physics: the significance describes a model fit, not a causal finding.
The result comes from a combination of DESI releases. The analysis pairs a DR2 Lyman-alpha Full-Shape constraint at an effective redshift of 2.33 with earlier DR1 Full-Shape constraints spanning effective redshifts from 0.295 to 1.491. The main comparison between a changing and constant model used seven tomographic constraints, meaning measurements arranged across redshift slices. The document is an arXiv preprint, version 1, dated 20 August 2026.
The latest point changes the picture
The slope did not carry the same statistical weight in every version of the fit. Using the DR1 constraints without the DR1 Lyman-alpha value, the estimate was 0.021 ± 0.029, or 0.7σ. Adding that DR1 value gave 0.015 ± 0.015, or 1σ. Replacing it with the DR2 Lyman-alpha constraint produced the reported 0.022 ± 0.012 slope at 1.8σ. The sequence shows how strongly the result depends on which Lyman-alpha constraint is used.
The question is therefore not simply whether a line can be fitted through the measurements, but whether that line is a better description than a constant value. The authors compare a constant model, Ω_m(z)=c, with a linear model, Ω_m(z)=mz+c, using the tomographic constraints. They also compare the models with minimum chi-square, the Akaike information criterion, its small-sample correction, and Bayesian evidence calculated with uniform priors.
On the current data, the answer is mixed. The reported minimum-chi-square difference was Δχ²_min = −3.5, while ΔAIC was −1.4 and the small-sample-corrected ΔAICc was 0.7. Bayesian evidence was B10 = 1.5, a weak preference for the line. Taken together, the authors judge the line and constant models statistically indistinguishable rather than decisively selecting the increasing trend.
The dark-energy reading depends on the model
The possible cosmological meaning is more conditional still. Under the paper’s Om(z) diagnostic, an increasing Ω_m(z) maps to what cosmologists call phantom dark energy, with w(z)<−1, while a decreasing trend maps to quintessence, with w(z)>−1. In the wCDM fit, w was below −1 at 1.7σ. But when the analysis allowed the enlarged w0w_aCDM parameter space, it recovered ΛCDM within 1σ. The apparent tension therefore changes with the chosen dark-energy parameterization.
The fitted parameter ranges show why the paper avoids a firm physical conclusion. For wCDM, the reported 68% credible intervals were Ω_m = 0.325 (+0.016/−0.017) and w0 = −1.130 (+0.075/−0.081). For w0w_aCDM, they were Ω_m = 0.336 (+0.019/−0.026), w0 = −1.04 (+0.17/−0.14), and wa = −0.8 (+1.2/−1.3). The intervals are broader in the second fit and depend on the selected parameterization.
Future precision could sharpen the choice
The preprint then asks what a more precise DESI result could do. In its conditional DR2 forecast, the slope becomes 0.022 ± 0.011 at 2σ. The corresponding differences were Δχ²_min = −4.8, ΔAIC = −2.8 and ΔAICc = −0.6, while B10 = 2.2. All four measures weakly favored the line in that forecast, but the outcome depends on the assumed future errors and central-value behavior.
That forecast is accompanied by a distribution of 104 simulated mocks. Their median slope significance was 2.1 ± 0.6σ; median ΔAIC was −2.5 (+2.2/−2.9), median ΔAICc was −0.2 (+2.2/−2.9), and median B10 was 2.2 (+6.8/−1.4). In the same mock distribution, 40% exceeded B10 > 3 and 14% exceeded B10 > 10, thresholds corresponding to moderate and strong evidence. These percentages describe simulated outcomes, not risks attached to an observed event.
Under the assumed error contraction for the final DESI release, the forecast becomes more favorable to a rising trend. Across 104 mocks, median slope significance was 2.9 ± 1.0σ, with median ΔAIC = −6.6 (+4.9/−6.8) and ΔAICc = −4.4 (+4.9/−6.8). The median Bayesian evidence was B10 = 13 (+369/−12); 70% of mocks exceeded B10 > 3 and 53% exceeded B10 > 10. Those figures are a conditional forecast of moderate-to-strong support, not evidence that the future release will produce it.
The unresolved question is persistence
Several constraints keep the result open. The direct analysis is built from a small set of DESI Full-Shape constraints, and current model tests do not clearly separate the line from the constant model. The dark-energy reading changes with parameterization, while the forecasts depend on assumed error reductions and central-value behavior. The paper says independent observables must converge on the same trend before a physical conclusion can be drawn, because a single observable cannot by itself exclude observational or modelling systematics.
For now, the defensible news is narrower than a discovery claim: DESI’s current Full-Shape constraints contain a 1.8σ upward Ω_m(z) slope, while formal comparisons find only weak preference for adding that trend. The forecast is that improved precision could raise the signal to a median 2.9σ and make Bayesian support stronger if the central pattern persists. Whether that happens, and whether the trend survives independent checks, remains unresolved.
Paper data and sources
Original title: A Prediction for DESI Full-Shape: Increasing Tomographic $Ω_m(z)$ Trend
Authors: Eoin Ó Colgáin, M. M. Sheikh-Jabbari
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text