The geometry of Markov diffusion generators (Q5930239)
From MaRDI portal
scientific article; zbMATH DE number 1587668
Language | Label | Description | Also known as |
---|---|---|---|
English | The geometry of Markov diffusion generators |
scientific article; zbMATH DE number 1587668 |
Statements
The geometry of Markov diffusion generators (English)
0 references
24 February 2002
0 references
These notes form the summary of a course given at the ETH Zürich in 1998, they consist in four main sections and present some recent ideas and results of D. Bakry and his collaborators. On a Riemannian manifold with Ricci curvature bounded from below by a constant \(R\), the gradient \(\nabla\), the Laplacian \(\Delta\) and the \(\Gamma_2\) operator satisfy the inequality \[ \Gamma_2 (f) \geq R |\nabla f |^2 + {1\over n} (\Delta f)^2. \] From this observation, an abstract property of curvature and dimension, noted \(\text{CD}(R,n)\), is introduced for the generator \(L\) of an abstract Markov semigroup \((P_t)_{t\geq 0}\), and written as \[ \Gamma_2 (f) \geq R \Gamma (f) + {1\over n} (L f)^2, \] where the gradient norm \(|\nabla f|^2\) is replaced by the carré du champ \(\Gamma (f)\). This allows in particular to consider fractional dimensions. The property \(\Gamma_2 (f) \geq R \Gamma (f)\), noted \(\text{CD}(R,\infty)\), tells that \(L\) is of curvature \(R\) and can be translated equivalently to the semigroup \((P_t)_{t\geq 0}\) as \(\Gamma (P_t f) \leq e^{-2Rt} P_t (\Gamma f)\). After the introduction of these notions, the first section ends with a presentation of Poincaré, Sobolev and logarithmic Sobolev inequalities. The second section starts with a derivation of logarithmic Sobolev inequalities for the heat kernel measure associated to \((P_t)_{t\geq 0}\), under the curvature hypothesis \(\text{CD}(R,\infty)\). Under the same assumption, an infinite-dimensional analog of the Lévy-Gromov isoperimetric inequality is then derived as \[ {\mathcal U}(P_t f) \leq P_t \left( \sqrt{{\mathcal U}^2 (f) + {1-e^{-2Rt}\over R} \Gamma (f) }\right), \] with isoperimetric function \({\mathcal U}= \Phi' \circ \Phi^{-1}\), where \(\Phi\) is the distribution function of the standard Gaussian measure on \({\mathbb R}\). In the third section it is shown how Sobolev inequalities for abstract Markov generators can be obtained from the condition \(\text{CD}(R,n)\), and how these inequalities can imply classical results on Riemannian manifolds such as Bishop's volume comparison theorem and Myers's diameter theorem, as well as eigenvalue comparison theorems and an abstract analog of the Topogonov-Cheng theorem. The last section deals with a general one-parameter family of equivalent Sobolev inequalities that contain the Nash inequalities, and also entropy-energy inequalities as a limiting case. The equivalence between logarithmic Sobolev inequalities and hypercontractivity, as well as other applications to heat kernel bounds, are presented in the remaining of this section, which ends with a result stating that a manifold satisfying a Sobolev inequality with optimal Euclidean constant is isometric to \({\mathbb R}^n\).
0 references
logarithmic Sobolev inequalities
0 references
Markovian semigroups
0 references
isoperimetry
0 references
comparison theorems
0 references
heat kernel bounds
0 references