Let X be a separable Hilbert space endowed with a non-degenerate centred Gaussian measure γ and let λ 1 be the maximum eigenvalue of the covariance operator associated with γ. The associated Cameron-Martin space is denoted by H. For a sufficiently regular convex function U : X → R and a convex set Ω ⊆ X, we set ν := e −U γ and we consider the semigroup (T Ω (t)) t≥0 generated by the self-adjoint operator defined via the quadratic formwhere ϕ, ψ belong to D 1,2 (Ω, ν), the Sobolev space defined as the domain of the closure in L 2 (Ω, ν) of D H , the gradient operator along the directions of H.A suitable approximation procedure allows us to prove some pointwise gradient estimates for (T Ω (t)) t≥0 . In particular, we show thatfor any p ∈ [1, +∞) and f ∈ D 1,p (Ω, ν). We deduce some relevant consequences of the previous estimate, such as the logarithmic Sobolev inequality and the Poincaré inequality in Ω for the measure ν and some improving summability properties for (T Ω (t)) t≥0 . In addition we prove that if f belongs to L p (Ω, ν) for some p ∈ (1, ∞), thenwhere Kp is a positive constant depending only on p. Finally we investigate on the asymptotic behaviour of the semigroup (T Ω (t)) t≥0 as t goes to infinity.
GRADIENT ESTIMATES ON INFINITE DIMENSIONAL CONVEX DOMAINS3 therein) and coupling methods (see for example [15,16,39]). On the other hand, in infinite dimensional Wiener spaces some partial results are also available. In the case of a Gaussian measure γ and Ω = X, the classical Mehler's representation formulawhere the equality has to be meant componentwise, (see [10, Proposition 1.5.6]). Again for the Gaussian measure γ on a convex subset Ω, in [11, Theorem 3.1] it is proved that |D H T (t)f | H ≤ e −t T (t)|D H f | H for any smooth function f . In this case, the idea consists in approximating the parabolic problem with a sequence of finite dimensional parabolic problems and using the factorisation of the Gaussian measure. Clearly, this approach does not work in our case since our measure in general does not decompose as a product of measures on orthogonal subspaces. Finally, the case of a weighted Gaussian measure is also considered in [20] where a version of (3) is proved when Ω = X and the H-derivative is replaced by the Fréchet one. We point out that, in this latter case, the proof of the gradient estimate is based on purely stochastic techniques. Hence, taking account of the existing literature, estimate (3) represents a generalisation of all the above results and the purely analytical proof we proposed, inspired by an idea due to Bakry andÉmery (see [4] and [38]), is a novelty in the proofs of gradient estimates.As announced, the pointwise gradient estimate (3) has several interesting consequences. First of all it yields that the semigroup T Ω (t) is smoothing, in the sense that it is bounded from L p (Ω, ν) into D 1,p (Ω, ν), for any p ∈ (1, ∞) and t > 0 as the estimatereveals. Due to the fact that the Sobolev embedding theorems fail to hold when we replace the Lebesgue measure with another general m...