Pointwise convergence of certain continuous-time double ergodic averages

The uploaded paper (CDKR) states and proves that for jointly measurable, measure-preserving R^2-actions ((s,t),x) ↦ S^s T^t x, the averages A_N(f1,f2)(x) = (1/N)∫_0^N f1(S^t x) f2(T^{t^2} x) dt converge for µ-a.e. x whenever p,q ∈ (1,∞] with 1/p+1/q ≤ 1; this is Theorem 1 in the PDF and matches the problem exactly . The proof uses (i) a lacunary subsequence reduction (2.2)–(2.3) , (ii) a weak L1 maximal inequality (2.4) to reduce to dense subspaces , (iii) a real-variable L^2→L^1 estimate—Proposition 2 (2.1)—and Calderón transference (Lemma 3) to control the difference f1(S^{t+δ}x)−f1(S^t x) times f2(T^{t^2}x) by N^{−γ} , and (iv) the continuous-time polynomial ergodic theorem [Bergelson–Leibman–Moreira, Thm. 8.31] to settle the long-scale limit (2.8) . The harmonic-analytic input is a curved triangular Hilbert transform model (Theorem 4) combined with a low-frequency parabolic averaging bound to obtain δ^γ gain in (3.2)–(3.9) . In contrast, the model’s Phase-2 writeup misstates key parts of the CDKR argument: it asserts a dyadic-block oscillation/variation inequality of the form ∑_k ||sup_{N∈[2^k,2^{k+1})}|B_N−B_{2^k}| || ≤ C that is neither claimed nor proved in CDKR (their Proposition 2 is an L^1-in-N bound for a shift-difference integrand, not a supremum-over-N variational estimate) . It also describes an eigenfunction-based reduction for the T-flow that CDKR do not use; instead, CDKR decompose with respect to the S-flow (2.6), reduce to (2.7), and then invoke the continuous-time polynomial result for (2.8) . Thus the paper’s theorem and proof are correct and complete for the stated range, while the model’s proof outline contains incorrect attributions and steps, even though its final conclusion (that CDKR’s Theorem 1 settles the problem) is true.