arXiv subjects

Tommaso Buvoli

Publications and source records attributed to Tommaso Buvoli.

12 recordsLinked to original sources

arXiv2025

Nonlinear Splitting for Gradient-Based Unconstrained and Adjoint Optimization

High dimensional and/or nonconvex optimization remains a challenging and important problem across a wide range of fields, such as machine learning, data assimilation, and partial differential equation (PDE) constrained optimization. Here we consider gradient-based methods for solving unconstrained and constrained optimization problems, and introduce the concept of nonlinear splitting to improve accuracy and efficiency. For unconstrained optimization, we consider splittings of the gradient to depend on two arguments, leading to semi-implicit gradient optimization algorithms. In the context of adjoint-based constrained optimization, we propose a splitting of the constraint $F(\mathbf{x},θ)$, effectively expanding the space on which we can evaluate the ``gradient''. In both cases, the formalism further allows natural coupling of nonlinearly split optimization methods with acceleration techniques, such as Nesterov or Anderson acceleration. The framework is demonstrated to outperform existing methods in terms of accuracy and/or runtime on a handful of diverse optimization problems. This includes low-dimensional analytic nonconvex functions, high-dimensional nonlinear least squares in quantum tomography, and PDE-constrained optimization of kinetic equations, where the total number of high-dimensional kinetic solves is reduced by a factor of three compared with standard adjoint optimization.

math.OC↗

arXiv2025

A New Class of Runge-Kutta Methods for Nonlinearly Partitioned Systems

This work introduces a new class of Runge-Kutta methods for solving nonlinearly partitioned initial value problems. These new methods, named nonlinearly partitioned Runge-Kutta (NPRK), generalize existing additive and component-partitioned Runge-Kutta methods, and allow one to distribute different types of implicitness within nonlinear terms. The paper introduces the NPRK framework and discusses order conditions, linear stability, and the derivation of implicit-explicit and implicit-implicit NPRK integrators. The paper concludes with numerical experiments that demonstrate the utility of NPRK methods for solving viscous Burger's and the gray thermal radiation transport equations.

math.NA↗

arXiv2025

Multirate Runge-Kutta for Nonlinearly Partitioned Systems

Multirate integration is an increasingly relevant tool that enables scientists to simulate multiphysics systems. Existing multirate methods are designed for equations whose fast and slow variables can be linearly separated using additive or component-wise partitions. However, in realistic applications, this assumption is not always valid. Building on the recently developed class of nonlinearly partitioned Runge-Kutta (NPRK) methods, we develop a framework for multirate NPRK (MR-NPRK) that allows for arbitrary nonlinear splittings of the evolution operator. We discuss order conditions, formalize different types of coupling between timescales, and analyze joint linear stability of MR-NPRK methods. We then introduce a class of 2nd- and 3rd-order methods, referred to as ``implicitly-wrapped'' multirate methods, that combine a user-specified explicit method for integrating the fast timescale with several slow implicit stages. These methods are designed to be algorithmically simple with low memory costs and minimal operator evaluations. Lastly, we conduct numerical experiments to validate our proposed methods and show the benefits of multirating a nonlinear partition.

math.NA↗

arXiv2025

Order Conditions for Nonlinearly Partitioned Runge-Kutta Methods

Recently a new class of nonlinearly partitioned Runge--Kutta (NPRK) methods was proposed for nonlinearly partitioned systems of autonomous ordinary differential equations, $y' = F(y,y)$. The target class of problems are ones in which different scales, stiffnesses, or physics are coupled in a nonlinear way, wherein the desired partition cannot be written in a classical additive or component-wise fashion. Here we use rooted-tree analysis to derive full order conditions for NPRK$_M$ methods, where $M$ denotes the number of nonlinear partitions. Due to the nonlinear coupling and thereby mixed product differentials, it turns out the standard node-colored rooted-tree analysis used in analyzing ODE integrators does not naturally apply. Instead we develop a new edge-colored rooted-tree framework to address the nonlinear coupling. The resulting order conditions are enumerated, provided directly for up to 4th order with $M=2$ and 3rd-order with $M=3$, and related to existing order conditions of additive and partitioned RK methods. We conclude with an example which shows how the nonlinear order conditions can be used to obtain an embedded estimate of the state-dependent nonlinear coupling strength in a dynamical system.

math.NA↗

arXiv2024

One-sweep moment-based semi-implicit-explicit integration for gray thermal radiation transport

Thermal radiation transport (TRT) is a time dependent, high dimensional partial integro-differential equation. In practical applications such as inertial confinement fusion, TRT is coupled to other physics such as hydrodynamics, plasmas, etc., and the timescales one is interested in capturing are often much slower than the radiation timescale. As a result, TRT is treated implicitly, and due to its stiffness and high dimensionality, is often a dominant computational cost in multiphysics simulations. Here we develop a new approach for implicit-explicit (IMEX) integration of gray TRT in the deterministic S$_N$ setting, which requires only one sweep per stage, with the simplest first-order method requiring only one sweep per time step. The partitioning of equations is done via a moment-based high-order low-order formulation of TRT, where the streaming operator and first two moments are used to capture the asymptotic stiff regimes of the streaming limit and diffusion limit. Absorption-reemission is treated explicitly, and although stiff, is sufficiently damped by the implicit solve that we achieve stable accurate time integration without incorporating the coupling of the high order and low order equations implicitly. Due to nonlinear coupling of the high-order and low-order equations through temperature-dependent opacities, to facilitate IMEX partitioning and higher-order methods, we use a semi-implicit integration approach amenable to nonlinear partitions. Results are demonstrated on thick Marshak and crooked pipe benchmark problems, demonstrating orders of magnitude improvement in accuracy and wallclock compared with the standard first-order implicit integration typically used.

math.NA↗

arXiv2023

Exponential Runge-Kutta Parareal for Non-Diffusive Equations

Parareal is a well-known parallel-in-time algorithm that combines a coarse and fine propagator within a parallel iteration. It allows for large-scale parallelism that leads to significantly reduced computational time compared to serial time-stepping methods. However, like many parallel-in-time methods it can fail to converge when applied to non-diffusive equations such as hyperbolic systems or dispersive nonlinear wave equations. This paper explores the use of exponential integrators within the Parareal iteration. Exponential integrators are particularly interesting candidates for Parareal because of their ability to resolve fast-moving waves, even at the large stepsizes used by coarse propagators. This work begins with an introduction to exponential Parareal integrators followed by several motivating numerical experiments involving the nonlinear Schrödinger equation. These experiments are then analyzed using linear analysis that approximates the stability and convergence properties of the exponential Parareal iteration on nonlinear problems. The paper concludes with two additional numerical experiments involving the dispersive Kadomtsev-Petviashvili equation and the hyperbolic Vlasov-Poisson equation. These experiments demonstrate that exponential Parareal methods offer improved time-to-solution compared to serial exponential integrators when solving certain non-diffusive equations.

math.NA↗

arXiv2022

Additive Polynomial Time Integrators, Part I: Framework and Fully-Implicit-Explicit (FIMEX) Collocation Methods

In this paper we generalize the polynomial time integration framework to additively partitioned initial value problems. The framework we present is general and enables the construction of many new families of additive integrators with arbitrary order-of-accuracy and varying degree of implicitness. In this first work, we focus on a new class of implicit-explicit polynomial block methods that are based on fully-implicit Runge-Kutta methods with Radau nodes, and possess high stage order. We show that the new fully-implicit-explicit (FIMEX) integrators have improved stability compared to existing IMEX Runge-Kutta methods, while also being more computationally efficient due to recent developments in preconditioning techniques for solving the associated systems of nonlinear equations. For PDEs on periodic domains where the implicit component is trivial to invert, we will show how parallelization of the right-hand-side evaluations can be exploited to obtain significant speedup compared to existing serial IMEX Runge-Kutta methods. For parallel (in space) finite-element discretizations, the new methods can achieve orders of magnitude better accuracy than existing IMEX Runge-Kutta methods, and/or achieve a given accuracy several times times faster in terms of computational runtime.

math.NA↗

arXiv2021

IMEX Runge-Kutta Parareal for Non-Diffusive Equations

Parareal is a widely studied parallel-in-time method that can achieve meaningful speedup on certain problems. However, it is well known that the method typically performs poorly on non-diffusive equations. This paper analyzes linear stability and convergence for IMEX Runge-Kutta Parareal methods on non-diffusive equations. By combining standard linear stability analysis with a simple convergence analysis, we find that certain Parareal configurations can achieve parallel speedup on non-diffusive equations. These stable configurations all posses low iteration counts, large block sizes, and a large number of processors. Numerical examples using the nonlinear Schrodinger equation demonstrate the analytical conclusions.

math.NA↗

arXiv2021

On the Stability of Exponential Integrators for Non-Diffusive Equations

Exponential integrators are a well-known class of time integration methods that have been the subject of many studies and developments in the past two decades. Surprisingly, there have been limited efforts to analyze their stability and efficiency on non-diffusive equations to date. In this paper we apply linear stability analysis to showcase the poor stability properties of exponential integrators on non-diffusive problems. We then propose a simple repartitioning approach that stabilizes the integrators and enables the efficient solution of stiff, non-diffusive equations. To validate the effectiveness of our approach, we perform several numerical experiments that compare partitioned exponential integrators to unmodified ones. We also compare repartitioning to the well-known approach of adding hyperviscosity to the equation right-hand-side. Overall, we find that the repartitioning restores convergence at large timesteps and, unlike hyperviscosity, it does not require the use of high-order spatial derivatives.

math.NA↗

arXiv2021

Exponential Polynomial Block Methods

In this paper we extend the polynomial time integration framework to include exponential integration for both partitioned and unpartitioned initial value problems. We then demonstrate the utility of the exponential polynomial framework by constructing a new class of parallel exponential polynomial block methods (EPBMs) based on the Legendre points. These new integrators can be constructed at arbitrary orders of accuracy and have improved stability compared to existing exponential linear multistep methods. Moreover, if the ODE right-hand side evaluations can be parallelized efficiently, then high-order EPBMs are significantly more efficient at obtaining highly accurate solutions than exponential linear multistep methods and exponential spectral deferred correction methods of equivalent order.

math.NA↗

arXiv2020

Constructing Polynomial Block Methods

The recently introduced polynomial time integration framework proposes a novel way to construct time integrators for solving systems of first-order ordinary differential equation by using interpolating polynomials in the complex time plane. In this work we continue to develop the framework by introducing several additional types of polynomials and proposing a general class of construction strategies for polynomial block methods with imaginary nodes. The new construction strategies do not involve algebraic order conditions and are instead motivated by geometric arguments similar to those used for constructing traditional spatial finite differences. Moreover, the newly proposed methods address several shortcomings of previously introduced polynomial block methods including the ability to solve dispersive equations and the lack of efficient serial methods when parallelism cannot be used. To validate our new methods, we conduct two numerical experiments that compare the performance of polynomial block methods against backward difference methods and implicit Runge-Kutta schemes.

math.NA↗

arXiv2015

A Class of Exponential Integrators Based on Spectral Deferred Correction

We introduce a new class of arbitrary-order exponential time differencing methods based on spectral deferred correction (ETDSDC) and describe a simple procedure for initializing the requisite matrix functions. We compare the stability and accuracy properties of our ETDSDC meth- ods to those of an existing implicit-explicit spectral deferred correction scheme (IMEXSDC). We find that ETDSDC methods have larger accuracy regions and comparable stability regions. We conduct numerical experiments to compare ETD and IMEX spectral deferred correction schemes against a competing fourth-order ETD Runge-Kutta scheme. We find that high-order ETDSDC schemes are the most efficient in terms of function evaluations and overall speed when solving partial differential equations to high accuracy. Our results suggest that high-order ETDSDC schemes are well-suited to work in conjunction with spectral spatial methods or other high-order spatial discritizations. Addi- tionally, ETDSDC schemes appear to be immune to severe order reduction, a problem which affects other ETD and IMEX schemes, including IMEXSDC.

math.NA↗