Corpus ID: 237940097

Mixed-precision explicit stabilized Runge-Kutta methods for single- and multi-scale differential equations

  title={Mixed-precision explicit stabilized Runge-Kutta methods for single- and multi-scale differential equations},
  author={Matteo Croci and Giacomo Rosilho de Souza},
Mixed-precision algorithms combine lowand high-precision computations in order to benefit from the performance gains of reduced-precision without sacrificing accuracy. In this work, we design mixedprecision Runge–Kutta–Chebyshev (RKC) methods, where high precision is used for accuracy, and low precision for stability. Generally speaking, RKC methods are low-order explicit schemes with a stability domain growing quadratically with the number of function evaluations. For this reason, most of the… Expand

Figures and Tables from this paper


Perturbed Runge-Kutta methods for mixed precision applications
It is shown that Runge–Kutta methods can be designed so that certain costly intermediate computations can be performed as a lower-precision computation without adversely impacting the accuracy of the overall solution. Expand
Performance Evaluation of Mixed-Precision Runge-Kutta Methods
A FORTRAN code is developed to solve a nonlinear system of ordinary differential equations using the mixed precision additive Runge-Kutta (MP-ARK) methods on IBM POWER9 and Intel x86 64 chips to efficiently produce accurate solutions with significant reductions in runtime. Expand
A stabilized Runge-Kutta-Legendre method for explicit super-time-stepping of parabolic and mixed equations
This work builds temporally first- and second-order accurate super-time-stepping methods around the recursion relation associated with Legendre polynomials and proves that the newly-designed RKL1 and RKl2 schemes have a very desirable monotonicity preserving property for one-dimensional problems - a solution that is monotone at the beginning of a time step retains that property at the end of that time step. Expand
Convergence properties of the Runge-Kutta-Chebyshev method
SummaryThe Runge-Kutta-Chebyshev method is ans-stage Runge-Kutta method designed for the explicit integration of stiff systems of ordinary differential equations originating from spatialExpand
Algebraic error analysis for mixed-precision multigrid solvers
The theory presented here is the first forward error analysis in the energy norm of iterative refinement and the first rounding error analysis of multigrid in general. Expand
Explicit stabilized multirate method for stiff stochastic differential equations
This work introduces a stochastic modified equation whose stiffness depends solely on the "slow" terms and devise a multirate method which overcomes the bottleneck caused by a few severely stiff terms and recovers the efficiency of stabilized schemes for large systems of nonlinear SDEs. Expand
Discretization-error-accurate mixed-precision multigrid solvers
It is shown that the quantization error incurred by simply storing the matrix in any fixed precision quickly begins to dominate the total error as the discretization is refined, and while iterative refinement is susceptible to quantization errors during the residual and update computation, the V-cycle used to compute the correction in each iteration is much more resilient, and continues to work if the system matrices in the hierarchy become indefinite due toquantization. Expand
High order explicit methods for parabolic equations
This paper discusses explicit embedded integration methods with large stability domains of order 3 and 4. The high order produces accurate results, the large stability domains allow some reasonableExpand
Effects of round-to-nearest and stochastic rounding in the numerical solution of the heat equation in low precision
This paper studies the accumulation of rounding errors in the solution of the heat equation via Runge-Kutta finite difference methods using round-to-nearest (RtN) and stochastic rounding (SR), and proves that for SR the global rounding errors are only $O(u\Delta t^{-1/4})$ in 1D and are essentially bounded (up to logarithmic factors) in higher dimensions. Expand
Accelerating the Solution of Linear Systems by Iterative Refinement in Three Precisions
The results suggest that on architectures for which half precision is efficiently implemented it will be possible to solve certain linear systems up to twice as fast and to greater accuracy, as well as recommending a standard solver that uses LU factorization in single precision. Expand