Abstract
The purpose of this paper is to prove a priori error estimates for the completely discretized problem of the dual mixed formulation for the heat diffusion equation in a polygonal domain. We complete the discretization of the problem (Farhloul et al., Functional Analysis and Evolution Equations. The Günter Lumer Volume, p. 240, Birkhäuser, Basel, 2007) in time by using the Crank-Nicolson scheme and we show the existence, the stability and a priori error estimates for the solution of the completely discretized problem.
Access provided by Autonomous University of Puebla. Download conference paper PDF
Similar content being viewed by others
Keywords
- Dual mixed finite element method
- Heat diffusion equation
- Singularities
- Grids refinements
- A priori error estimates
2010 MSC:
1 Introduction
The aim of this paper is to study the completely discretized problem of the dual mixed formulation for the heat evolution equation in a polygonal domain Ω of \(\mathbb {R}^{2}\). Let us note that we have dealt in [2] with the mixed dual semi-discretized formulation in space. In this work, we will complete the discretization of the problem in time by using the Crank-Nicolson scheme. For this completely discretized problem, we are going to demonstrate the existence and the uniqueness of the solution. Then, we will show the stability of this scheme. Finally, we will prove, under some conditions of meshing refinement in the reentrant corner of the polygonal domain, a priori error estimates of order 1 in space, and 2 in time for the solution of the completely discretized problem when using the Crank-Nicolson implicit scheme.
2 The Model Problem
Let Ω be a bounded polygonal domain of \(\mathbb {R}^{2}\). In particular the boundary of Ω: ∂ Ω \({ \equiv \Gamma } = \cup _{j=1}^{N}\overline {{\Gamma }}_{j}\) for some \(N\in \mathbb {N}\), where Γj is an open segment of a straight line of the plane \(\mathbb {R} ^{2},\;\forall \) j = 1, 2, …, N. As it is well known, the geometric singularities of the domain (the angles) induce in general singularities on the solution of the Cauchy problem with Dirichlet boundary condition for the heat diffusion equation (see for example the books of P. Grisvard [3, 4]). As shown in [4] and [3], we may suppose without harming to the generality that Ω has only one nonconvex angle, in other words one reentrant corner, and that its vertex is located at the origin. In the following, we denote by ω (ω > π) the measure of that angle. For a fixed T > 0, let us set Q:= Ω×]0, T[ and let us denote by Σ := Γ×]0, T[ the lateral boundary of the cylinder Q. We introduce the following weighted Sobolev space (see [3], definition 8.4.1.1 and lemma 8.4.1.2 p. 388):
which is a Hilbert space for the norm
where the semi-norm |.|2,α,Ω is defined by
r denotes the distance to the origin of \(\mathbb {R}^{2}\). Let us recall that:
Let us consider the Cauchy problem for the heat diffusion equation in Ω up to time T: given the right-hand side \(f\in L^{2}\left ( 0,T;L^{2}(\Omega )\right ) \) and the initial condition , find weak solution of:
We have seen in Proposition 3 of [2], that Problem (7.1) admits a unique solution:
with \(\alpha \in \left ] 1-\frac {\pi }{\omega },1\right [ .\)
In the following, we will consider the additional unknown heat flux \(\vec {p} :=\vec {\nabla }u\) (rigorously speaking, the heat flux density vector is the opposite [10]). We have proved in [2, Theorem 4] that the couple \((\vec {p},u)\) ∈ L 2(0, T;H(div, Ω)) × H 1(0, T;L 2( Ω)) is the unique solution of the dual mixed formulation:
3 The Completely Discretized Problem
Let us consider a family of triangulations \((\mathcal {T}_{h})_{h>0}\) on \( \overline {\Omega }\). For a triangle K belonging to the triangulation \( \mathcal {T}_{h}\), let us denote by h K the diameter of K and by ρ K the interior diameter of K, i.e. the diameter of the largest of disc included in K. As in Theorem 8.4.1.6 p. 392 of [3], we suppose that the family of triangulations \((\mathcal {T}_{h})_{h>0}\) has the property that \(\max \limits _{K\in \mathcal {T}_{h}}\frac {h_{K}}{\rho _{K}}\;\) is bounded by a positive constant independent of the parameter h; in that case, one says usually that the family of triangulations is regular (see for example [1] (17.1) p. 131). In accordance with the tradition (see [1] remark 17.1 p. 131) the same letter h may have also another significance: it may denote instead: \(h=:\max _{K\in \mathcal {T}_{h}}h_{K}\). The true significance of h is always clear from the context.
For the discretization in space, let us recall that we are using the approximant spaces:
where \(RT_{0}(K):=P_{0}(K)^{2}\oplus P_{0}(K)\,\binom {x_{1}}{x_{2}}\) denotes the real vector space of dimension three of the so called Raviart-Thomas vectorfields of degree 0 on the triangle K (RT 0(K) is denoted D 1(K) in [7, p. 550]), and \(P_{0}\left ( K\right ) \) is the real vector space of dimension one of the constant functions on the triangle K. For the complete discretized problem, we use the time subdivision of the interval \(\left [ 0,T\right ] \) into N sub-intervals \(\left [ t_{n-1},t_{n}\right ] ~(N\geq 2)\), such that: 0 = t 0 ≤⋯ ≤ t n ≤⋯ ≤ t N = T. Δt = t n − t n−1 denotes the fixed time step. Let us denote by \(u_{h}^{n}\) the approximation of the temperature at time t n = n Δt in M h. For the approximation of \(\frac { \partial u}{\partial t}\ \)at the time t n, we use the following formula:
4 The Crank-Nicolson Scheme
Before giving the complete discretized mixed formulation for the heat diffusion equation by the Crank-Nicolson scheme, we will first define new variables. We set,
Let us consider the following problem:
Let us note that in (7.4) appear \(\vec {p}_{h}^{\;n- \frac {1}{2}},u_{h}^{n-\frac {1}{2}},u_{h}^{n}\) and \(u_{h}^{n-1}.\) Thus, we can choose as unknowns \(\vec {p}_{h}^{\;n-\frac {1}{2}}\) and \(u_{h}^{n}\) for n ≥ 1. Another alternative is to consider that the unknowns are the \(\vec { p}_{h}^{\;n}\) for n ≥ 0 and \(u_{h}^{n}\) for n ≥ 1. \(u_{h}^{0}\) denotes the initial condition that is known and \(\vec {p}_{h}^{\;0}\) is exceptionally defined by the equation:
This last choice of unknowns presents the following advantages:
-
Traditional unknowns.
-
Symmetry of “\(\vec {p}\)” and “u” in the problem.
Proposition 7.1
Problem (7.4) admits one and only one solution
Proof
Let us start by demonstrating the unicity. Therefore, let us show that if \(( \vec {p}_{h}^{\;n},u_{h}^{n})\in X_{h}\times M_{h}\) verifies :
then \((\vec {p}_{h}^{\;n},u_{h}^{n})=0.\) By taking \(\vec {q}_{h}=\vec {p} _{h}^{\;n-\frac {1}{2}}\) in the first equation of (7.6) and \(v_{h}=u_{h}^{n-\frac {1}{2}}\) in the second equation of (7.6), we obtain:
On the other hand we have:
on the condition of having already demonstrated that \(u_{h}^{n-1}=0.\) So let us take n = 1 in (7.8) and since \(u_{h}^{0}=0\) (initial condition), consequently \(u_{h}^{1}=0\), and \(\vec {p}_{h}^{\;\frac {1}{2}}=0\) by Eq. (7.7) the second term in the left-hand side of Eq. (7.7) being also a priori non-negative due to [6].
According to (7.5), \(u_{h}^{0}=0\) implies that \(\vec {p} _{h}^{\;0}\) = 0. Knowing already that \(\vec {p}_{h}^{\;\frac {1}{2}}=0,\) it follows that \(\vec {p}_{h}^{\;1}=0\). Thus \(\vec {p}_{h}^{\;2}=0\) and \(u^{2}_{h}=0\) by (7.7) and (7.8) with n = 2 and so on for every n ≥ 1.
For the existence, it is well known by the Riesz Representation Theorem applied to Eq. (7.5) that \(\vec {p} _{h}^{\;0} \) exists (see Remark 1.5.2 [5, p. 31]). According to the system (7.4) with n = 1, and with the aim of constructing \(\vec {p}_{h}^{\;1}\) and \(u_{h}^{1}\), \(u_{h}^{0}\) and \(\vec { p}_{h}^{\;0}\) being known, we have
Let Φh: X h × M h ↦ \(X_{h}^{\prime }\times M_{h}^{\prime }\), be the application defined by:
Let us prove that it is an isomorphism. Since Φh is linear of X h × M h in its dual, and since the two spaces X h × M h and \(X_{h}^{\prime }\times M_{h}^{\prime }\) have the same dimension, it suffices to show that Φh is injective to prouve its bijectivity. Let \(\left ( \vec {p}_{h}^{\;1},u_{h}^{1}\right ) \) such that:
In (7.10), let us take \(\vec {q}_{h}=\vec {p}_{h}^{\;1}\) and \(v_{h}=u_{h}^{1}\) in (7.11), it follows that :
We obtain \(\vec {p}_{h}^{\;1}=0\) and \(u_{h}^{1}=0.\) It follows from that the injectivity and thus the bijectivity of Φh. So it suffices to apply \(\Phi _{h}^{-1}\) to the couple of linear forms defined by the two members of the right hand side of the system of Eqs. (7.9). In a similar way, we construct \(\left ( \vec {p}_{h}^{\;2},u_{h}^{2}\right ) \) by considering the system (7.4), with n = 2 and so on. □
4.1 Stability
Now we will give the stability result of the Crank-Nicolson scheme.
Theorem 7.1
Suppose \(\Delta t\leq \frac {1}{2}\) . There exists a constant c > 0 independent of h such that:
and
Proof
For the proof of (7.13), we use in particular the discrete inequality of Gronwall [6, p. VI-9], with:
For the complete demonstration of (7.13) see the proof of Theorem 1.5.15 [5, pp. 52–54]. For the proof of (7.14), see the proof of Proposition 1.5.16 [5, p. 55]. □
4.2 Error Estimates on the Temperature and on the Heat Flux Density Vector
In order to demonstrate the results related to the error estimate, proceeding similarly as in the semi-discrete case [2], we decompose the error: \(u(t_{n})-u_{h}^{n}\) into the sum of \(u(t_{n})-\tilde {u}_{h}(t_{n})\) and \(\tilde {u}_{h}(t_{n})-u_{h}^{n}\), where \((\tilde {\vec {p}}_{h}(t_{n}),{\tilde {u}}_{h}(t_{n}))\) ∈ X h × M h is the solution of the “elliptic projection problem” at time t n : find \((\tilde {\vec {p}}_{h}(t_{n}),{\tilde {u}}_{h}(t_{n}))\) ∈ X h × M h solution of,
The elliptic problem (7.15) being true for n and n − 1, making the sum, we obtain:
We may rewrite the Crank-Nicolson scheme, with \(u_{h}^{0}\;,\) given by:
Let us set \(\theta _{h}^{n}:=u_{h}^{n}-\tilde {u}_{h}(t_{n})\) and \(\vec { \varepsilon }_{h}^{\;n}:=\vec {p}_{h}^{\;n}-\tilde {\vec {p}}_{h}(t_{n})\). Then, by subtraction we obtain the following system:
Proposition 7.2
Assume \(f,\frac {df}{dt}\in H^{1}(0,T;L^{2}(\Omega ))\) and as well as Then
Proof
Let w ∈ H 1(0, T;L 2( Ω)) ∩ L 2(0, T;L 2( Ω)) be the solution of the heat diffusion equation
Set and .
By integrating the equation \(\frac {dw}{dt}(s)=\Delta w(s)+\frac {d^{2}f}{ dt^{2}}(s),\) ∀′ s ∈ [0, T] from 0 to t, we obtain:
Then, by simplifying the 2 members we obtain that v is the solution of the following Cauchy problem:
According to the proof of Proposition 8 in [2], we have \(v= \frac {du}{dt}.\) Thus \(\frac {d^{2}u}{dt^{2}}=\frac {dv}{dt}=w\in H^{1}(0,T;L^{2}(\Omega )).\) And consequently u ttt ∈ L 2(0, T;L 2( Ω)). □
Theorem 7.2
Under the hypotheses of Proposition 7.2 , there exists a constant c > 0 independent of h and of k such that:
Proof
The first step of this demonstration is to bound \(\left \Vert \theta _{h}^{n}\right \Vert \) in terms of \(\left \Vert \theta _{h}^{n-1}\right \Vert \) and \(\left \Vert \omega ^{n}\right \Vert .\) For this, let’s take \(v_{h}=\theta _{h}^{n}+\theta _{h}^{n-1}\) in the second equation of the system (7.16) and \(\vec {q}_{h}=\vec {\varepsilon }_{h}^{\;n}+\vec {\varepsilon }_{h}^{\;n-1}\) in the first. We have:
And as we have :
where, with analogy to the book of Vidar Thomée [8], “R h()” denotes here the component in M h of the couple of X h × M h “elliptic projection of” (cf definition 6 of [2]). Thus,
because Δu(t) + f(t) = u t(t), for every t > 0.
Let us set \(\widetilde {\omega }^{n}:=\widetilde {\omega }_{1}^{n}+ \widetilde {\omega }_{2}^{n}+\widetilde {\omega }_{3}^{n},\) with:
Then, it follows from (7.19) and (7.18), that,
Thus, we obtain:
As
we deduce that :
And so, a fortiori we get
Thus,
so that it suffices to bound \(\widetilde {\omega }^{n}.\) Let us start with \(\widetilde {\omega }_{2}^{n}\). By definition we have
Using Taylor’s formula, we get
Hence, at the time t n−1, we have
Let us consider the difference of these two above equalities, we obtain
It follows that
Thus we have demonstrated that
Now, let us try to bound \(\left \Vert \widetilde {\omega } _{3}^{n}\right \Vert { }_{0,\Omega }\). The Taylor formula gives:
By summing these two above equalities we obtain:
So, by applying the Laplace operator Δ, we obtain
Consequently,
We still have to bound \(\left \Vert \widetilde {\omega } _{1}^{n}\right \Vert { }_{0,\Omega }\). Let us recall that
Thus, by using Proposition 12 in [2, p. 252], there exists a constant c > 0 independent of h such that:
Consequently
According to the inequality (7.24), we have that:
recalling that \(\theta _{h}^{0}=u_{h}^{0}-\tilde {u}_{h}(0)=0.\) By using inequalities (7.26), (7.27) and (7.28), we get
Consequently recalling that \(\widetilde {\omega }^{n}=\widetilde {\omega } _{1}^{n}+\widetilde {\omega }_{2}^{n}+\widetilde {\omega }_{3}^{n},\) we obtain
And since we can replace Δu tt(s) by u ttt(s) − f tt(s) in the above inequality.
Finally, we get the following inequality:
□
Theorem 7.3
Let \(\{\mathcal {T}_{h}\}\) be a regular family of triangulations on \( \overline {\Omega },\) satisfying the properties (i) and (ii) of Proposition 9 of [ 2 , p. 250]. Under the hypotheses of Proposition 7.2 , and for \(\alpha \in \left ] 1-\frac {\pi }{w},1\right [ ,\) there exists a constant c > 0 independent of h such that for every n ≥ 1, we have
Proof
It suffices to apply the triangular inequality:
By the inequality (7.17) and the inequality 5.6 of Proposition 9 in [2], we obtain the result. □
Similarly to the bound obtained in the implicit case, and in order to demonstrate the error estimate on \(\vec {p}_{h}^{\;n}\), we need an analogous result to Proposition 1-5-10 of [5, p. 42], but for here \(\widetilde {\omega }^{n}:=\widetilde {\omega }_{1}^{n}+\widetilde {\omega }_{2}^{n}+\widetilde {\omega }_{3}^{n}\).
Proposition 7.3
Let us suppose that f ∈ H 1(0, T;L 2( Ω)) and
Proof
Let us consider the Crank-Nicolson scheme for the mixed method, written in the following form:
By subtracting member by member from the first equation of (7.34), the first equation defining the elliptic projection:
we obtain,
where \(\vec {\varepsilon }_{h}^{\;n}:=\vec {p}_{h}^{\;n}-\tilde {\vec {p}} _{h}(t_{n})\) and \(\theta _{h}^{n}:=u_{h}^{n}-\tilde {u}_{h}(t_{n})\). Equation (7.35) being true for n and n − 1, by making the difference member by member and by dividing by the time step, we obtain:
Taking \(\vec {q}_{h}=\vec {\varepsilon }_{h}^{\;n}+\vec {\varepsilon } _{h}^{\;n-1} \) in Eq. (7.36), we obtain:
By the equalities (7.16) and (7.19), it follows that:
∀v h ∈ M h. In particular, if we choose v h = 1K, for any \(K\in \mathcal {T}_{h}\), we obtain
since \(\overline {\partial }\theta _{h}^{n}=\frac {\theta _{h}^{n}-\theta _{h}^{n-1}}{\Delta t}\in M_{h}.\) From (7.37) and (7.38), follows that
By dividing both sides of the above inequality (7.41) by the time step Δt, we obtain: \(\bar {\partial }\Vert \vec {\varepsilon }_{h}^{\;n}\Vert ^{2}\leq \Vert \widetilde {\omega }^{n}\Vert ^{2}\) . □
Corollary 7.1
Under the hypotheses of Proposition 7.2 , there exists a constant c > 0 independent of h and Δt such that:
Proof
By inequality (7.31), we have:
Thus:
By (7.29), we have \(\Delta t\left \Vert \widetilde { \omega }_{2}^{j}\right \Vert { }_{0,\Omega }\leq \frac {\Delta t^{2}}{8} \int _{t_{j-1}}^{t_{j}}\left \Vert u_{ttt}(s)\right \Vert { }_{0,\Omega } ds\).
It follows from a similar calculus using the inequality of Cauchy-Schwartz:
And also for (7.30), we obtain:
Now, by using Proposition 7.3, we get
It follows that by summing up these inequalities, by (7.5) and \(u_{h}^{0}=\tilde {u}_{h}(0):\)
Since \(\ \widetilde {\omega }^{n}=\widetilde {\omega }_{1}^{n}+\widetilde { \omega }_{2}^{n}+\widetilde {\omega }_{3}^{n}\), it follows that
From inequalities (7.46), (7.45) and (7.44) the assertion follows. □
In conclusion, we get the following result:
Theorem 7.4
Under the hypotheses of Proposition 7.2 , let \(\{\mathcal {T}_{h}\}\) be a regular family of triangulations on \(\overline {\Omega },\) satisfying the properties (i) and (ii) of Proposition 9 of [ 2 , p. 250]. For \( \alpha \in \left ] 1-\frac {\pi }{w},1\right [ ,\) there exists a constant c > 0 independent of h such that for every n ≥ 1, we have
Proof
By the triangular inequality:
By Farhloul et al. [2, (5.5) p. 250]:
From this estimate and the preceding corollary, the result follows. □
5 Conclusion
In this paper we have demonstrated that the completely discretized problem of the mixed formulation for the heat equation using the Crank-Nicolson scheme for time discretization admits one and only one solution. By refining the meshings according to Raugel’s rules near the reentrant corners [9], we have established optimal order of convergence for this completely discretized dual mixed method for the heat diffusion equation in a polygonal domain. And this by using for the spatial discretization, Raviart–Thomas vectorfields of degree 0 for the heat flux density vector, locally constant functions for the scalar field of temperatures, and the Crank-Nicolson scheme for the discretization in time.
References
Ciarlet, P.G.: Basic error estimates for elliptic problems. In: Ciarlet, P.G., Lions, J.L. (eds.) Handbook of Numerical Analysis, vol. II, pp. 17–351. North-Holland, Amsterdam (1991)
Farhloul, M., Korikache, R., Paquet, L.: The dual mixed finite element method for the heat diffusion equation in a polygonal domain I. In: Amann, H., Arendt, W., Hieber, M., Neubrander, F., Nicaise, S., von Below, J. (eds.) Functional Analysis and Evolution Equations. The Günter Lumer Volume, pp. 239–256. Birkhäuser, Basel (2007)
Grisvard, P.: Elliptic Problems in Nonsmooth Domains. Monographs and Studies in Mathematics, vol. 24. Pitman, Boston (1985)
Grisvard, P.: Singularities in Boundary Value Problems. Recherches en Mathématiques Appliquées [Research Notes in Applied Mathematics], vol. 22. Masson, Paris; Springer, Berlin (1992)
Korikache, R.: Mixed finite element methods for parabolic equations. PhD thesis, Polytechnic University Hauts-De-France (2007)
Lions, J.-L.: Cours d’Analyse numérique, Cours de l’Ecole Polytechnique, Promotion 1972. Hermann, Paris (1974)
Roberts, J.E., Thomas, J.-M.: Mixed and hybrid methods. In: Ciarlet, P.G., Lions, J.L. (eds.) Handbook of Numerical Analysis, vol. II. Finite Element Methods (Part 1), pp. 523–639. Elsevier, North-Holland, Amsterdam (1991)
Thomée, V.: Galerkin Finite Element Methods for Parabolic Problems. Springer Series in Computational Mathematics, vol. 25. Springer, Berlin (2006)
Raugel, G.: Résolution numérique par une méthode d’éléments finis du problème de Dirichlet pour le laplacien dans un polygone. C.R.A.S. Paris 286, 791–794 (1978)
Taine, J., Petit, J.-P.: Heat Transfer. Prentice Hall, Englewood Cliffs (1993)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Korikache, R., Paquet, L. (2022). The Completely Discretized Problem of the Dual Mixed Formulation for the Heat Diffusion Equation in a Polygonal Domain by the Crank-Nicolson Scheme in Time. In: Barrera, D., Remogna, S., Sbibih, D. (eds) Mathematical and Computational Methods for Modelling, Approximation and Simulation. SEMA SIMAI Springer Series, vol 29. Springer, Cham. https://doi.org/10.1007/978-3-030-94339-4_7
Download citation
DOI: https://doi.org/10.1007/978-3-030-94339-4_7
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-94338-7
Online ISBN: 978-3-030-94339-4
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)