Model-independent reconstruction of Kaniadakis holographic dark energy with neural networks: constraints from Pantheon+SH0ES
📄 arXiv:2610.00194 · 📥 PDF · 2026-09-18 · physics.gen-ph
Authors: Shaima Abdullah Al Shanfari [arXiv · scholar] , Suresh Kumar Pathamuthu [arXiv · scholar]
📄 Abstract
Model-independent reconstruction of Kaniadakis holographic dark energy (KHDE), future event horizon IR cutoff, via ANNs trained on simulated Type Ia SNe. Mock Pantheon ($N=1048$) and Pantheon+SH0ES ($N=1701$, 77 Cepheid-calibrated) sets generated from fiducial KHDE ($c=1.15$, $β=0.30$, $Ω_m=0.30$, $H_0=69.6$ km/s/Mpc). Bootstrap deep ensemble mapped $z\toμ(z)$ plus intrinsic scatter; $H(z)$, $w(z)$ from analytic derivatives. Pantheon+SH0ES outperformed Pantheon: scatter $0.018$ vs $0.048$ mag; $H(z)$ precision $3.2\%$ vs $6.0\%$ ($0.1<z<1.2$). Cepheid calibration essential for $H_0$: $72.0\pm2.0$ vs $64.8\pm6.0$ uncalibrated. Direct fit and MCMC recovered injected $Ω_m$, $H_0$; $β$ unconstrained (signal $0.009$ mag in $μ(z)$). Applied to real Pantheon+SH0ES (1657 SNe, full stat+sys covariance), reproducing published $Λ$CDM benchmarks. KHDE fit: $Ω_m=0.254$, $H_0=73.12$ km/s/Mpc; $β<0.63$ (95\%). $Δ$AIC$=+3.8$, $Δ$BIC$=+14.6$: no preference for KHDE over $Λ$CDM. Run on Azure ML Studio.