Fabio Durastante

Sede ufficiale: LARGO BRUNO PONTECORVO, 5, 56127 PISA

Email: fabio.durastante@unipi.it

Telefono: 050 2213232

Sito web: https://fdurastante.github.io/

Profilo

Ruolo: Ricercatore a tempo determinato L.240/2010

Struttura: Dipartimento di Matematica

Settore scientifico-disciplinare: Analisi Numerica MATH-05/A



Didattica

Attività didattica

Incarichi di responsabilità didattica di moduli/insegnamenti

Incarichi di co-docenza in moduli/insegnamenti



Ricevimento

Modalità: Il ricevimento è disponibile sulla piattaforma online BBB oppure in presenza in ufficio. Inviare una mail all'indirizzo istituzionale per prenotarsi (disponibili anche altri orari).

Luogo: BBB: https://hausdorff.dm.unipi.it/b/fab-e7l-8ph-nwh Ufficio: Dipartimento di Matematica, Ufficio 107, piano terra.

Orario: Giovedì dalle 9.00 alle 10.45

Ricerca

Interessi di ricerca

My research centers on the efficient numerical solution of extremely large, sparse linear systems—matrices so vast that they can contain billions or even trillions of unknowns, yet sparse enough to admit specialized iterative algorithms. I develop and analyze Krylov subspace methods alongside algebraic and geometric multigrid preconditioners that adapt dynamically to sequences of related problems, achieving robust convergence at scales previously out of reach. Translating these advances into practice, I have architected high-performance software toolkits that exploit CPU and GPU parallelism—demonstrating, for instance, how cutting-edge coarsening strategies and polynomial smoothers can accelerate industrial-scale fluid dynamics codes on the world’s leading supercomputers.
Extending beyond classic linear systems, I investigate matrix functions and rational Krylov techniques, which enable the rapid evaluation of fractional powers, exponentials, and other transformations of large matrices. By leveraging low-rank corrections to update existing factorizations, these methods open new possibilities in time-dependent simulations, control applications, and network dynamics. In parallel, I explore the numerical treatment of fractional differential equations—modeling nonlocal, memory-dependent processes—by designing adaptive discretizations for Caputo-type derivatives, short-memory approximations, and specialized preconditioners that keep anomalous diffusion and wave-propagation simulations both accurate and tractable.
My interests also span matrix-theoretic approaches to complex networks and optimal control of PDEs. I have introduced walk-based centrality measures grounded in Mittag-Leffler functions to capture nuanced connectivity patterns, and I have studied nonlocal diffusion on graphs through fractional Laplacians of variable order. In the realm of PDE-constrained optimization, I analyze the spectral structure of saddle-point systems and devise optimize-then-discretize schemes—combining quasi-Newton solvers like L-BFGS with bespoke preconditioners—to tackle inverse problems and design challenges in engineering. Throughout all these endeavors, I prioritize clean, modular code—written in Fortran, C, Python, and MATLAB—and leverage MPI, OpenMP, and CUDA to ensure that theory, software, and real-world applications converge seamlessly at the exascale.
 

Pubblicazioni