The growth rate of the spectral radius of the upwind discontinuous Galerkin method

For linear advection on a uniform periodic mesh, the upwind discontinuous Galerkin (DG) method approximates the solution by a polynomial of degree $p$ on each cell and couples neighboring cells through the upwind flux. The resulting operator, scaled by the ratio of the cell width to the advection speed, has eigenvalues on the curve $|f_{p+1}(z)|=1$, where $f_{p+1}$ is the $[p/(p+1)]$ Pad\'e approximant of $e^{-z}$, the rational function of degree $p$ over degree $p+1$ whose Taylor series agrees with that of $e^{-z}$ to the highest possible order. Krivodonova and Qin observed for $p<100$ that the leftmost real point $-x_p$ of the curve satisfies $x_p\ge1.5(p+1)^{1.75}$ and conjectured that the spectral radius of the scaled operator, which is at least $x_p$ for an even number of cells, grows proportionally to $(p+1)^{1.75}$. Because the spectral radius limits the time step of explicit time integration, explicit time integration with a time step chosen from a bound proportional to $(p+1)^{1.75}$ can be unstable for large $p$ if the spectral radius grows faster. We determine the growth of the largest modulus $\rho_p$ on the curve, an upper bound on the spectral radius because every eigenvalue lies on the curve. The number $x_p$, which is at most $\rho_p$, lies between $2p^2/\log p$ and $5p^2/(2\log p)$ for every $p\ge2$, and $x_p\log p/p^2$ tends to $2$. The largest modulus $\rho_p$ equals $x_p$ for $p\le64$ and lies between $x_p$ and $p^{1/p}x_p$ for $p\ge64$, so $\rho_p\log p/p^2$ also tends to $2$. For an even number of cells, the spectral radius therefore lies between $x_p$ and $\rho_p$, so the conjectured growth is false, whereas the bound $x_p\ge1.5(p+1)^{1.75}$ that Krivodonova and Qin observed holds for every $p\ge0$ in the sharper form $x_p>1.5591(p+1)^{1.75}$. We bound integral representations of the Pad\'e numerator and denominator analytically for large $p$ and, for the remaining $p$, check in exact and in interval arithmetic finitely many inequalities that imply the bounds.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24422596
Primary Topic
Advanced Numerical Methods in Computational Mathematics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The growth rate of the spectral radius of the upwind discontinuous Galerkin method

강성우
Zenodo (CERN European Organization for Nuclear Research)
Advanced Numerical Methods in Computational Mathematics
preprint

The growth rate of the spectral radius of the upwind discontinuous Galerkin method

강성우
preprint en

Abstract

For linear advection on a uniform periodic mesh, the upwind discontinuous Galerkin (DG) method approximates the solution by a polynomial of degree $p$ on each cell and couples neighboring cells through the upwind flux. The resulting operator, scaled by the ratio of the cell width to the advection speed, has eigenvalues on the curve $|f_{p+1}(z)|=1$, where $f_{p+1}$ is the $[p/(p+1)]$ Pad\'e approximant of $e^{-z}$, the rational function of degree $p$ over degree $p+1$ whose Taylor series agrees with that of $e^{-z}$ to the highest possible order. Krivodonova and Qin observed for $p<100$ that the leftmost real point $-x_p$ of the curve satisfies $x_p\ge1.5(p+1)^{1.75}$ and conjectured that the spectral radius of the scaled operator, which is at least $x_p$ for an even number of cells, grows proportionally to $(p+1)^{1.75}$. Because the spectral radius limits the time step of explicit time integration, explicit time integration with a time step chosen from a bound proportional to $(p+1)^{1.75}$ can be unstable for large $p$ if the spectral radius grows faster. We determine the growth of the largest modulus $\rho_p$ on the curve, an upper bound on the spectral radius because every eigenvalue lies on the curve. The number $x_p$, which is at most $\rho_p$, lies between $2p^2/\log p$ and $5p^2/(2\log p)$ for every $p\ge2$, and $x_p\log p/p^2$ tends to $2$. The largest modulus $\rho_p$ equals $x_p$ for $p\le64$ and lies between $x_p$ and $p^{1/p}x_p$ for $p\ge64$, so $\rho_p\log p/p^2$ also tends to $2$. For an even number of cells, the spectral radius therefore lies between $x_p$ and $\rho_p$, so the conjectured growth is false, whereas the bound $x_p\ge1.5(p+1)^{1.75}$ that Krivodonova and Qin observed holds for every $p\ge0$ in the sharper form $x_p>1.5591(p+1)^{1.75}$. We bound integral representations of the Pad\'e numerator and denominator analytically for large $p$ and, for the remaining $p$, check in exact and in interval arithmetic finitely many inequalities that imply the bounds.

Zenodo (CERN European Organization for Nuclear Research)
Advanced Numerical Methods in Computational Mathematics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
The growth rate of the spectral radius of the upwind discontinuous Galerkin method — 강성우 · Zenodo (CERN European Organization for Nuclear Research) (2026) | TGRS Research Map | TGRS