Computer Methods in Applied Mechanics and Engineering
Estimation of interpolation error constants for the P0 and P1 triangular finite elements
Introduction
The finite element method (FEM) is now recognized as a powerful numerical method for wide classes of partial differential equations. Furthermore, it also has sound mathematical bases such as highly refined a priori and a posteriori error estimations. In the classical a priori error analysis of FEM, interpolation errors are essential to derive final error estimates in various norms [7], [8], [10]. In this process, there appear various positive constants besides the standard discretization parameter h and norms (or seminorms), but it has been very difficult to evaluate such constants explicitly. For quantitative purposes, however, it is indispensable to evaluate or bound them as accurately as possible, because sharper estimates enable more efficient finite element computations. Thus such an evaluation has become progressively more important and has been attempted especially for adaptive finite element calculations based on a posteriori error estimation as well as for numerical verification by FEM [1], [4], [6], [7], [13]. In this paper, we will give a few fundamental results on some interpolation error constants of the most popular triangular finite elements.
More specifically, we give some results on interpolation error constants appearing in the popular P0 (piecewise constant) and P1 (piecewise linear) triangular finite elements. Essentially based on the paper of Babuška–Aziz [3], we analyze the dependence of several constants on the geometric parameters such as the maximum interior angle and the minimum edge length of the triangle more quantitatively than in [3]. Above all, the optimal constant (C3 in this paper) appearing in the H1 error estimate of the P1 interpolation of H2 functions over the unit isosceles right triangle is essential and frequently used, and it was explicitly evaluated firstly by Natterer [15]. On the other hand, this constant was shown to be closely related to the one (C1 in this paper) presented and effectively used by Babuška and Aziz in conjunction with the maximum angle condition [3]. More precisely, C1 gives an upper bound quite close to the optimal constant C3, and the relation between C3 and C1 was further discussed in [13], [18]. Thus a precise estimation of these two constants is very important, and a number of researchers have given bounds for these using various approximation methods including numerical verification, see e.g. [2], [11], [13], [14], [15], [18]. Furthermore, these constants can be also used to evaluate the interpolation error constants for the non-conforming P1 triangle, as will be mentioned later.
For the above Babuška–Aziz constant, we have succeeded in obtaining a value which is in a sense optimal. That is, by analytically solving an eigenvalue problem for the 2D Laplacian over the above triangular domain, we can show that the constant can be easily determined from a solution of the simple transcendental equation . In this process, we use the reflection (or symmetry) method [16]. Moreover, we have obtained some explicit relations for the dependence of such constants on the geometry of triangles. It is to be emphasized that they are consistent with the maximum angle condition in [3]. We also give some numerical and analytical results, the latter of which are based on asymptotic analysis. Thus our results can be effectively used in the quantitative a priori and a posteriori error estimations of the finite element solutions by the P1 triangular element and also those based on the P0 triangle. The former is of course the most classical and fundamental one, but still in frequent use, while the latter appears in some mixed finite element methods and implicitly on various occasions. Moreover, we also give some results for the non-conforming P1 triangle by using the constants for the P0 and the conforming P1 triangles.
Section snippets
Preliminaries
Let h, α and θ be positive constants such thatThen we define the triangle by with three vertices , and . From (1), AB is shown to be the edge of maximum length, i.e. , so that here denotes the medium edge length, although the notation h is often used as the largest edge length. A point on the closure of is denoted by . By an appropriate congruent transformation in R2, we can configure any triangle as
Dependence of constants on
This section is devoted to analysis of the effects of the maximum interior angle θ on ’s for fixed α. For , the well-known maximum angle condition was derived in [3]. However, the results reported there are not fully quantitative, so that we give here more quantitative estimates for the constants including .
To this end, let us introduce the following simple affine transformation between and :This transformation is a bit
Dependence of constants on
Up to now, we have given some basic results for dependence of error constants on h and θ. In this section, we will consider the dependence of such constants on α when . With this regard, we owe much the following results to the analysis by Babuška and Aziz [3]. In particular, the estimation below is an important consequence derived in [3] and also in [13], [18], and so we here call C1 the Babuška–Aziz constant. Theorem 2 For h = 1 and , are continuous positive-valued
Determination of C0 and C1
First let us determine C0 exactly. Actually, its exact value is already known, see e.g. [13], [14]. However, we here state the results with a proof, since the underlying idea is somewhat common to the more complicated case of C1. Theorem 3 With regard to C0, i.e., for α = 1 and , it holds that . Proof We will prove in two steps, each of which is based on rather well-known arguments and techniques. The triangular domain to be considered here is T. (1) Let Ω be a unit square domain:
Asymptotic behaviors of constants as
Moreover, we can analyze the asymptotic behaviors of the constants ’s as , cf. [12]. In particular, the right limit values ’s are given by zeros of certain transcendental equations (derived from eigenvalue problems of ordinary differential equations, ODE’s) in terms of the hypergeometric functions [20]. For example, is equal to the first positive zero of the Bessel function .
For the analysis, we use various techniques including compactness arguments. We will publish
Nonconforming P1 triangle
We have mainly considered the conforming P1 triangle, which can naturally construct subspaces of H1 space over the entire domain. But there also exists a non-conforming counterpart, which is also based on but uses as nodes the midpoints of edges or edges themselves [19]. Analysis of such an element is more complicated, since we must additionally evaluate the errors induced by the interelement discontinuity of the approximate functions. Still we can obtain some results for the interpolation
Numerical results
We performed numerical computations to see the actual dependence of various constants on α and θ. Here, we just show the results for , and by the P1 FEM with the uniform triangulation of the domain Tα. In such calculations, Tα is subdivided into a number of small congruent triangles with . The penalty method in [18] was also adopted to calculate approximately. The resulting approximate problems are matrix eigenvalue ones, and can be solved numerically if
Concluding remarks
We have obtained some explicit relations for the dependence of a few interpolation error constants on geometric parameters of triangular finite elements. In particular, we have succeeded in determining the Babuška–Aziz constant from a very simple equation. We can effectively utilize these results to give upper bounds of the a priori and a posteriori error estimates of finite element solutions based on the P1 and/or P0 approximate functions. To obtain more clear picture for the dependence of the
Acknowledgements
The authors would like to express their deepest appreciation to Prof. M.T. Nakao of Graduate School of Mathematical Sciences, Kyushu University and Prof. N. Yamamoto of Department of Computer Science, The University of Electro-Communication for acquainting them with the importance of the present problem and a number of references. The estimation was obtained through fruitful discussion with Prof. A. Kaneko of Ochanomizu University, who presented an estimation method in his textbook:
References (20)
- et al.
A Posteriori Error Estimation in Finite Element Analysis
(2000) Computable finite element error bounds for Poisson’s equation
IMA J. Numer. Anal.
(1982)- et al.
On the angle condition in the finite element method
SIAM J. Numer. Anal.
(1976) - et al.
The Finite Element Method and Its Reliability
(2001) - et al.
A comparison of finite element error bounds for Poisson’s equation
IMA J. Numer. Anal.
(1981) - et al.
Adaptive Finite Element Methods for Differential Equations
(2003) - et al.
The Mathematical Theory of Finite Element Methods
(2002) The Finite Element Method for Elliptic Problems
(2002)- et al.
Determination of the Babuška–Aziz constant for the linear triangular finite element
Japan J. Ind. Appl. Math.
(2006) - et al.
Numerical Methods for Elliptic and Parabolic Partial Differential Equations
(2003)
Cited by (39)
Shape optimization for the Laplacian eigenvalue over triangles and its application to interpolation error analysis
2023, Journal of Differential EquationsGuaranteed local error estimation for finite element solutions of boundary value problems
2023, Journal of Computational and Applied MathematicsInclusion method of optimal constant with quadratic convergence for H<inf>0</inf><sup>1</sup>-projection error estimates and its applications
2023, Journal of Computational and Applied MathematicsExplicit eigenvalue bounds of differential operators defined by symmetric positive semi-definite bilinear forms
2020, Journal of Computational and Applied MathematicsCitation Excerpt :Recently, compared to the classical analysis for qualitative error analysis of eigenvalue approximation, the research on explicit bounds of eigenvalues has become a new topic in the field of numerical analysis. Early work about explicit bound of eigenvalues based on the finite element method (FEM) can be traced back to the work of [1–4], where the upper bounds of various interpolation error constants are considered by estimating the first eigenvalue of the corresponding differential operator. In [5–7], the lower bounds for leading eigenvalues of differential operators are provided; see also the work of [8,9].
Explicit bound for quadratic Lagrange interpolation constant on triangular finite elements
2018, Applied Mathematics and ComputationA framework of verified eigenvalue bounds for self-adjoint differential operators
2015, Applied Mathematics and ComputationCitation Excerpt :For example, a good indicator for the error of approximate solutions requires the explicit error estimation for various interpolation operators. The estimation of error constants is reduced to eigenvalue problems of Laplace and biharmonic operators; see, [10,13]. In addition, verifying the solution for nonlinear differential equations requires eigenvalue bounds of the controlling differential operators; see, e.g., [17,19,22].