IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 1, 2002 207
A Dispersion Analysis for the Finite-Element Method
in Time Domain With Triangular Edge Elements
Agostino Monorchio, Member, IEEE, Enrica Martini, Member, IEEE, Giuliano Manara, Senior Member, IEEE, and
Giuseppe Pelosi, Fellow, IEEE
Abstract—A numerical dispersion analysis for the finite-element
(FE) method in time domain (TD) is presented. The dispersion re-
lation is analitically derived by considering a time-harmonic plane
wave propagating through an infinite uniform mesh consisting of
equilateral triangular elements. The effect of the time step on the
numerical dispersion is investigated and it is shown that, if linear
tangential-linear normal (LT-LN) edge-basis functions are used,
there exists a time-step value that minimizes the deviation of the
dispersion relation from the ideal linear case. In particular, the
analysis performed shows that this optimum time step holds for any
propagation direction of the plane wave within the mesh and, virtu-
ally, for any frequency, strongly enhancing numerical accuracy of
the FE-TD method. As a working example, we choose to compare
the numerically computed TE modes of two-dimensional guiding
structures with the corresponding analytical values; to this end, an
efficient procedure for the computation of the eigenfrequencies is
proposed, allowing us to avoid TD data processing.
Index Terms—Discrete time systems, finite-element time-domain
(FE-TD) method, numerical dispersion, wave propagation.
I. INTRODUCTION
I
T is well known that an electromagnetic wave propagating
through a finite-element (FE) mesh will experience some
numerical dispersion, yielding a progressive phase error in the
problem solution. Such a phenomenon has been analyzed for
the finite-element method (FEM) in frequency domain for both
nodal [1], [2] and edge [2]–[4] basis functions; some results rele-
vant to the FE solution of first-order Maxwell equations in time
domain (TD) have also been presented [5], [6]. However, al-
though a TD analysis is required to simulate transient electro-
magnetic fields, the numerical dispersion in the TD-FEs has not
been thoroughly investigated.
In this letter, we analyze the numerical dispersion in
two-dimensional (2-D) FE-TD procedures using edge basis
functions. Edge elements have revealed particularly suited
for solving electromagnetic problems automatically ensuring
the continuity of the tangential components of the fields. A
comprehensive survey of the properties of edge elements can
be found in [7], where curl-conforming edge basis functions of
different orders are considered.
Manuscript received November 6, 2002; revised December 5, 2002.
A. Monorchio and G. Manara are with the Department of Information
Engineering, University of Pisa, I-56126 Pisa, Italy (e-mail: a.monorchio@
iet.unipi.it; g.manara@iet.unipi.it).
E. Martini is with the Department of Information Engineering, University of
Siena, I-53100, Siena, Italy (e-mail: martini@dii.unisi.it).
G. Pelosi is with the Department of Electronics and Telecommunications,
University of Florence, I-50134 Florence, Italy (pelosi@det.unifi.it).
Digital Object Identifier 10.1109/LAWP.2002.807962
The analysis performed in this paper reveals that if linear
tangential-linear normal (LT-LN) edge basis functions are used
on a uniform hexagonal mesh, there exists a time step value
that allows us to increase the convergence rate of the disper-
sion relation at any propagation direction of the wave within
the mesh and for any frequency, strongly enhancing numerical
accuracy of the FE-TD method. This is not the case if constant
tangential-linear normal (CT-LN) edge basis functions are used.
From a practical point of view, for a given TD formulation, the
time step can be used as an additional parameter, with respect
to its frequency formulation counterpart, to control the accuracy
of the procedure. Numerical results relevant to the analysis of
closed guiding structures are presented to verify the theory.
II. FE-TD FORMULATION
The specific FE-TD formulation adopted relies on the
discretization of the second-order vector wave equation
with proper boundary conditions. The equation describes a
time-harmonic electric field in an isotropic and loss-free region
(1)
where and are the permeability and the permittivity of the
medium, respectively, and is an impressed current density.
The region of interest is subdivided into discrete elements,
where the field is expanded by using a proper set of vector basis
functions and a Galerkin testing procedure is employed. By
assuming perfectly electric or magnetic conducting boundaries
and by setting , the above equation is finally numerically
expressed as
T S (2)
where denotes the unknown coefficients vector,
T ,S ,
and are the vector basis functions used for expanding the
field.
As far as the discretization in time is concerned, we resort to
the Beta–Newmark formulation [8], for which the above equa-
tion can be cast in the following form:
M N P (3)
where
M T S N T S P M (4)
1536-1225/02$17.00 © 2002 IEEE