This paper shows that a usually neglected piece of quantum gravity — the running of the Gauss–Bonnet (Euler) term — can restore a natural de Sitter starting point for a class of inflation models built from curvature‑squared gravity. By keeping the scale dependence of that topological term in the one‑loop renormalization group, the authors find a unique balance that gives a stationary de Sitter saddle. That saddle supplies a very flat “hilltop” for the scalar degree of freedom (the scalaron) and leads to inflationary predictions compatible with current data.

The authors work with quadratic gravity, a theory whose action contains only terms quadratic in curvature rather than the usual Einstein–Hilbert term. Quantum effects make the coefficients of those curvature‑squared terms run with scale. Earlier work found viable inflation when one particular choice of renormalization scheme produced a maximum of one coupling. But that maximum disappears once the Euler trace anomaly is included. Here the team keeps the Gauss–Bonnet coefficient and calculates the standard one‑loop running including contributions from vectors, Weyl fermions, and conformally coupled scalars. They show that the Euler running balances the scale dependence of the R^2 term at a single coupling ratio, producing a de Sitter solution that is stationary both for the gravitational constraint and for the compact Euclidean action.

Physically, the de Sitter saddle appears as an extremely flat hilltop in the scalaron potential. The curvature of that hilltop is tiny: the paper quotes m^2/H^2 of order −5×10^−10 for a cosmic‑microwave‑background (CMB) normalized trajectory. The subsequent rolling toward lower curvature produces an inverse‑linear inflationary plateau. The last N*≈50–60 e‑folds of inflation in this scenario are independent of renormalization scheme. The predicted scalar spectral tilt is approximately n_s ≈ 1 − 4/(3N*), and the tensor‑to‑scalar ratio r depends on the number and type of matter fields that speed up the running. Across the viable parameter window the model predicts r ≳ 0.008 with 0.973 ≲ n_s ≲ 0.978.