While cosmology headlines often fixate on the Hubble tension, a quieter but foundational discrepancy exists in Big Bang nucleosynthesis (BBN), specifically the predicted versus observed primordial deuterium abundance. A new preprint (not yet peer-reviewed) from Timothy Launders and colleagues dated April 2026 goes beyond conventional theoretical cross-section calculations by deploying Gaussian process (GP) regression directly on laboratory nuclear data. This fully data-driven technique avoids assuming specific functional forms, allowing experimental results to dictate the behavior of the three reactions that most strongly influence deuterium destruction: d(d,n)³He, d(d,p)t, and d(p,γ)³He.

Using the Planck satellite's baryon density as input, the authors predict 10^5 × D/H = 2.442 ± 0.040 in standard BBN. This value lies 1.70σ below the Cooke et al. (2018) astronomical measurement derived from high-resolution spectroscopy of 10–15 quasar absorption systems in distant, nearly pristine gas clouds formed when the universe was roughly 1–2 billion years old. When the analysis switches to a higher baryon density inferred from a joint fit to Planck, ACT DR6, and SPT-3G CMB data, the discrepancy grows to 1.98σ.

This work synthesizes and extends three key lines of research. It aligns with first-principles nuclear calculations (consistent with updated predictions from Pitrou et al. 2018 on BBN reaction rates) yet diverges from earlier BBN analyses that relied on low-degree polynomial fits to the same nuclear data. The preprint explicitly validates both approaches and finds that polynomials systematically over-predict deuterium yields, an important bias previous coverage and many cosmological papers overlooked. It also connects to Planck 2018 cosmological parameters and recent ground-based CMB experiments (ACT DR6, SPT-3G), which independently tighten constraints on the baryon density and therefore on BBN predictions.

Methodologically, the study applies GP regression to all available experimental S-factor data for the cited reactions, then propagates the resulting uncertainty into a BBN abundance calculation. The authors validate that GPs produce unbiased predictions with well-calibrated error bars through controlled tests on mock data. Limitations are transparently addressed: the ±0.040 uncertainty is limited by the density and precision of existing lab measurements, particularly sparse in the astrophysically relevant 0.1–0.6 MeV energy window. Unrecognized systematic errors in older nuclear experiments could still shift the central value. The sample of experimental data points, while the best currently available, remains finite and energy-gap limited.

Analytically, this result sharpens BBN as a precision test of the standard cosmological model and its early-universe assumptions. Deuterium is exponentially sensitive to the baryon-to-photon ratio; a lower predicted abundance at fixed baryon density implies faster burning and potentially different expansion history or particle content than assumed. The growing tension with CMB-inferred baryon density does not yet rise to discovery level but fits a broader pattern of mild conflicts between early-universe probes. If confirmed with tighter nuclear data, it could point to new physics—such as extra relativistic species, varying fundamental constants, or decaying particles altering the Hubble rate during the first minutes—or simply demand renewed accelerator time on those d+d reactions.

By minimizing theoretical nuclear input and exposing the shortcomings of polynomial extrapolations used for decades, the preprint delivers a cleaner primordial deuterium benchmark. It does not resolve existing cosmological tensions but makes them more diagnostically powerful, underscoring that further progress now depends as much on terrestrial nuclear physics as on larger telescopes.