DNS and role of round-off error: Two-Dimensional Taylor-Green vortex problem
Tapan K. Sengupta 1, Vajjala K. Suman 2,3 · 2025
Read DNS and role of round-off error: Two-Dimensional Taylor-Green vortex problem by Tapan K. Sengupta 1, Vajjala K. Suman 2,3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-18696.txt.
Introduction
Suman and Sengupta ask whether the tiniest rounding errors can steer a fluid flow. Their 2025 paper replays a classic vortex test case with direct simulation at extreme accuracy. A fourth-order time scheme and a Fourier space method left rounding as the only error source. They ran the case in both double and quadruple precision.
The two runs split in kind, not just in detail. Quadruple precision took a different route into turbulence and decayed differently afterward. The team also spotted a receptivity phase never reported before. Their claim is strong: rounding noise alone can shape how a flow breaks down. The result warns simulation users that machine precision is itself a physical actor in sensitive flows.
Catalog record
| CREATOR | Tapan K. Sengupta 1, Vajjala K. Suman 2,3 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-18696
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
DNS and role of round-off error: Two-Dimensional Taylor-Green vortex problem Tapan K. Sengupta 1, Vajjala K. Suman 2,3 Abstract The role of round-off errors on the receptivity and instability of fluid flows are conclusively established for the first time using high accuracy simulations of the benchmark two-dimensional (2D) Taylor-Green vortex problem using double and quadruple precisions. Employing the fourth order Runge-Kutta (RK4) method for temporal discretization and Fourier pseudospectral method for spatial discretization enables unprecedented accuracy necessary for controlling all forms of errors, except the remaining round-off error. Results clearly show that adopting quadruple precision results in a qualitatively different receptivity route to turbulence and its subsequent decay compared to double precision. Another important observation is the identification of the receptivity phase which has never been reported before. Present study not only establishes the singular role of round-off errors but also has potential ramifications on receptivity and instability of flows due to precision of simulation. † † affiliation: 1 Professor (Retired), High Performance Computing Laboratory, IIT Kanpur, UP-208 016, India 2 Computational & Theoretical Fluid Dynamics Division, CSIR-NAL, Bangalore, India 3 Department of Aerospace Engineering, IIT Kanpur, U.P.-208 016, India 1 Introduction In scientific computing, errors are committed as one approximates various operators while discretizing and are called the truncation error. In solving the Navier-Stokes equation, one approximates the convection and diffusion operators by different methods. A good yard-stick has been adopted by expressing different methods in the spectral plane, as is customary ( Vichnevetsky & Bowles 1982 ). It must be emphasized that the early practice of treating spatial and temporal discretizations separately are incomplete, and instead space and time discretizations are analyzed together for model convection equation ( Sengupta 2013 ), convection-diffusion equation ( Suman et al. 2017 ) etc., and the generic approach is known as the global spectral analysis (GSA) which has been reviewed in Sagaut et al. 2023 . With the help of GSA, it is now possible to understand the contributions of truncation error into explaining dispersion (phase errors, q-waves ) and dissipation errors for various phenomenon including the Gibbs’ phenomenon. GSA also explains anti-diffusion which leads to catastrophic numerical instability shown by Vajjala et al. 2020 . Both finite volume and finite element methods have been studied using GSA for convection equation in Sengupta 2013 . In the context of representing spatial derivatives, the Fourier spectral method ( Canuto et al. 1991 ; Gottlieb & Orszag 1977 ) is most accurate up to all resolved length scales i.e., Nyquist limit ( Sengupta 2013 ). In contrast to the study of error dynamics due to truncation error, there are only a very few studies of error dynamics due to round-off error explained by GSA. The reason for this is due to the fact that in most situations truncation error dominates over round-off error. Thus, to study the effects of round-off error, one needs to carefully design a numerical study, as it is done here, to remove the other potential sources of errors, other than the round-off error. Another goal here is to establish whether the effects of the round-off error is epistemic or aleatoric. Round-off error is the difference of results obtained by an algorithm with exact arithmetic and that produced by the same algorithm using finite-precision. Thus, round-off error is due to truncated representation of real numbers during the arithmetic operations for numerical computations. It is conjectured that round-off error may accumulate, implying more refined calculations suffer more due to round-off error. It is noted that for ill-conditioned problems, such as the case of accumulating round-off error ( Chapman 2013 ; Chapra 2012 ). Thus in scientific computing, representation of real numbers with finite precision is one source of round-off error. Also, certain problems prone to physical/numerical instability can be influenced by such errors. For this reason, present investigation is performed on the disturbance growth for the Taylor-Green vortex (TGV) problem ( Taylor & Green 1937 ). Existence of an analytical equilibrium solution prompted Taylor and Green to solve the TGV problem as a perturbation series in time to explain transition to turbulence. Goldstein also extended this perturbation series analysis by expanding it in terms of the Reynolds number ( Re ) - which is used as the inverse of the kinematic viscosity ( \nu ). The singularity in time and Re , prompted researchers to interpret singularities to imply turbulence. Due to the existence of an analytical periodical solution in 2D, one can use Fourier basis functions for spatial discretization ( Orszag 1971a ; Orszag 1971b ; Canuto et al. 1991 ; Gottlieb & Orszag 1977 ; Brachet et al. 1983 ; Brachet 1991 ; Brachet et al. 1992 ) for enhanced accuracy. Fourier method is adopted to minimize truncation error in representing convection and diffusion terms, and initially Morf et al. 1980 ; Brachet et al. 1983 ; Brachet et al. 1992 numerically investigated the TGV problem by solving the Euler equation. Brachet et al. 1983 also investigated the three-dimensional (3D) viscous flow problem for the generation of small-scale structures by vortex stretching in the resulting turbulence. It is important to highlight the accuracy of the pseudo-spectral methods, in the context of the GSA to calibrate methods. Such methods are used in recent times, including the DNS of homogeneous isotropic turbulence by Buaria et al. 2020 , who used 12288^{3} periodic grids. Such DNS uses a forcing term, following the work of Rogallo 1981 , which also requires hyperviscosity for numerical stabilization ( Sengupta et al. 2022 ). This is mandatory as it uses a two-stage Runge-Kutta (RK2) method for time advancement. The RK time integration method has been thoroughly investigated with the canonical convection and convection-diffusion equations by Sengupta et al. 2022 which show that the RK4 method is significantly superior. Thus, there is no need to add numerical diffusion to eliminate numerical instability. Further details about these issues have been discussed in greater details for the double precision simulation of 2D TGV problem using Fourier spectral -RK4 method in Sengupta et al. 2024 . The same methodology is used here for the double and quadruple precision simulations of the 2D TGV problem to investigate the role of precision on the receptivity. The truncation error is absent, i.e. no dissipation and dispersion error, as the use of Fourier spectral method is equivalent to having a box filter. The use of 128 Fourier modes is adequate to also eliminate any error due to Gibbs’ phenomenon. The only major source of error for the presented simulation is related to aliasing caused by the convection terms, and can be controlled by choosing correct de-aliasing techniques. For example, using the 3/2-rule of zero-padding of Canuto et al. 1991 , one partially circumvents aliasing error. A correct zero-padding has been proposed with 2-rule in Sengupta 2004 that ensures complete removal of aliasing error. The TGV problem in 2D or 3D form has allowed the study of the primary disturbance growth of the unsteady equilibrium flow by using near-spectral accuracy compact schemes by using nonuniform grids, reported in Sengupta et al. 2018 ; Gau & Hattori 2014 . The 2D TGV problem has been investigated by Sengupta et al. 2018 with the help of the disturbance