Andrea Agazzi's home page

Andrea Agazzi (agazzi at
Griffith research assistant professor
Mathematics Department, Duke University
Physics Building, 120 Science Drive
27705 Durham, NC

Research Interests

I am interested in applied probability theory, more specifically in interacting particle systems for real world applications. I have worked on scaling limits for models of chemical reaction networks, focusing on the relations between their dynamics and their structure. More recently, I have started to work on the dynamics of scaling limits of machine learning algorithms seen as interacting particle systems.

Publications and Preprints (see also vitae or Google Scholar Page)

  1. Large deviations with Markov jump processes with uniformly diminishing rates, with JL. Andreis, M. Renger, R. Patterson, arXiv:2102.13040
  2. Global optimality of softmax policy gradient with single hidden layer neural networks in the mean-field regime, with J. Lu, International Conference on Learning Representations, 2021 (to appear)
  3. Temporal Difference Learning with nonlinear function approximation in the lazy training regime, with J. Lu, arXiv:1905.10917
  4. Seemingly stable chemical kinetics can be stable, marginally stable or unstable, with J. Mattingly, Comm. Math. Sci. 18 (6), 1605 - 1642 (2020)
  5. Large Deviations Theory for Markov Jump Models of Chemical Reaction Networks, with A. Dembo and J.-P. Eckmann, Ann. Appl. Prob. 28 (3), 1821-1855 (2018)
  6. On the Geometry of Chemical Network Theory: Lyapunov Function and Large Deviations Theory, with A. Dembo and J.-P. Eckmann, J. Stat. Phys. 172 (2), 321-352 (2018)
  7. The Colored Hofstadter Butterfly for the Honeycomb Lattice, with G. M. Graf and J.-P. Eckmann, J. Stat. Phys. 156 (3), 417-426 (2014)


  1. Stochastic calculus (MATH 545.01)