Non-Markovian Cosmology with Causal Memory Kernels

Abstract

We investigate a non-Markovian extension of Friedmann cosmology in which the present expansion rate retains information about the past cosmic evolution through a causal memory kernel, with \LCDM{} recovered in the limit of vanishing kernel amplitude, $A\to0$. We show that five kernels (Linear, Power Law, Exponential, Gaussian and Oscillatory) arise as cosine transforms of the spectral densities of an unresolved hidden sector, and that the Exponential and Linear kernels also follow independently from Israel-Stewart causal viscosity and from Deser-Woodard non-local gravity, respectively. We constrain the kernels with 30 Cosmic Chronometer (CC) measurements and with $H(z)$ derived from the DESI DR1 and DR2 baryon acoustic oscillation (BAO) data, using multi-start $χ^{2}$ minimisation followed by Markov chain Monte Carlo sampling, and compare them with \LCDM{} through the Akaike and Bayesian information criteria. The Linear amplitude is consistent with zero ($A=0.002\pm0.030$ for CC+DR2). The Power-law, Exponential and Gaussian kernels are observationally degenerate: the data select sharply peaked kernels that constrain only an integrated amplitude $\Aeff\simeq0.01$--$0.03$, equivalent to a uniform $0.5$--$1.5\%$ rescaling of $H(z)$, and they are disfavoured relative to \LCDM{} ($Δ{\rm AIC}=+2.5$ to $+3.6$). Only the Oscillatory kernel improves the fit appreciably ($\dchi=4.5$--$8.1$), but it requires a negative damping rate, which is incompatible with the causal derivations, and it is not favoured once CC data are included ($Δ{\rm BIC}=+4.4$ to $+6.2$)