You are on page 1of 9

# i

i
“main” — 2013/11/8 — 17:18 — page 147 — #1

i

Tema

i

Tendˆencias em Matem´atica Aplicada e Computacional, 14, N. 2 (2013), 147-155
www.scielo.br/tema
doi: 10.1590/S2179-84512013005000004

Finite Difference Preserving the Energy Properties
of a Coupled System of Diffusion Equations
A.J.A. RAMOS1 and D.S. ALMEIDA Jr.2*
Received on April 4, 2012 / Accepted on July 5, 2013

ABSTRACT. In this paper we proved the exponential decay of the energy of a numerical scheme in finite
difference applied to a coupled system of diffusion equations. At the continuous level, it is well-known that
the energy is decreasing and stable in the exponential sense. We present in detail the numerical analysis of
exponential decay to numerical energy since holds the stability criterion.
Keywords: diffusion equations; finite difference; numerical exponential decay.

1

INTRODUCTION

In this work we consider the numerical solutions in finite difference applied to the following
coupled system of diffusion equations:
φt − D1 φxx + α(φ − ψ) = 0, 

× (0, T ),

(1.1)

ψt − D2 ψxx + α(ψ − φ) = 0, 

× (0, T ),

(1.2)

t > 0,

(1.3)

∀x ∈ ,

(1.4)

φ(0, t ) = φ(L , t ) = ψ(0, t ) = ψ(L , t ) = 0,
φ(x, 0) = φ0 (x),

ψ(x, 0) = ψ0 (x),

where  = (0, L), D1 > 0 and D2 > 0 are the diffusion coefficients and α > 0 is the coupling parameter. The initial-boundary value problem (1.1)-(1.4) appears in dispersion processes
between species. To more detail see Murray [5].
An important non-linear functional concerning to the system system (1.1)-(1.4) is its energy. It
is given by  


1 L
2
2
E(t ) :=
|φ(x, t )| + |ψ(x, t )| d x, ∀t ≥ 0.
(1.5)
2 0
*Corresponding author: Dilberto da Silva Almeida J´unior
1 Mestre em Matem´atica pela Universidade Federal do Par´a, Instituto de Ciˆencias Exatas e Naturais – Universidade
Federal do Par´a. Rua Augusto Corrˆea, 66075-110 Bel´em, PA, Brasil. E-mail: ramos@ufpa.br
2 Professor Adjunto II da Universidade Federal do Par´a, Instituto de Ciˆencias Exatas e Naturais – Universidade Federal
do Par´a. Rua Augusto Corrˆea, 66075-110 Bel´em, PA, Brasil. E-mail: dilberto@ufpa.br

i

i
i

i

4) is well-posed in Hilbert space H = L 2(0. ψ0) ∈ H there exists a unique solution (φ. Apl.1)-(1. we showed that this numerical energy preserves the exponential decay such as continuous case.1)-(1. 2 dt 0 0 (2. we showed that E(t ) obeys the energy dissipation law. Multiplying (1. one has E(t ) ≤ E(0). 2 ENERGY PROPERTIES In this section.3). T ].4). we proved the energy dissipation property. it follows that  L  L  L 1 d |φ|2 d x + D1 |φx |2 d x + α (φ − ψ)φd x = 0. the energy (1. L 2(0.5) satisfies the energy dissipation law. we focus on numerical analysis of the energy properties of a numerical scheme in finite difference applied to the system (1.1) 0 Taking into account the homogeneous Dirichlet boundary conditions (1.3). ψ) ∈ C([0. N. Indeed. (2. (1. we derived the numerical scheme in finite difference and its numerical energy. one has E(t ) ≤ E(0).4) Tend.3) (2. ∀t ≥ 0. in diffusion problems. The energy E(t ) in (1. 14.i i “main” — 2013/11/8 — 17:18 — page 148 — #2 i 148 i FINITE DIFFERENCE PRESERVING THE ENERGY PROPERTIES OF A COUPLED SYSTEM Naturally. we finished with the conclusions. 0 0 (2. L 2(0. Then.2) and (2. we obtain  L  L  L 1 d |ψ|2d x + D2 |ψx |2 d x + α (ψ − φ)ψd x = 0. In section 4. That is to say.5) suggests that the system (1. the existence and uniqueness of solutions can be assured by using the semigroups theory (see [4]). More precisely. ∀t ≥ 0. Comput. The rest of the paper is organized as follows: In section 2.6) Moreover. T ].1) by φ and integrating on (0. 2 dt 0 0 0 Adding up (2. In that direction. In section 3. L) × L 2 (0. Proof. In particular. we obtain    L  L 1 d |φ|2 + |ψ|2 d x = − D1 |φx |2 + D2 |ψx |2 + α|φ − ψ|2 d x.. we have the following Theorem: Theorem 2. we get  L  L  L φt φd x − D1 φxx φd x + α (φ − ψ)φd x = 0. L)) ∩ C 1 ([0. 2 (2013) i i i i . In section 5. for initial data (φ0 . we present some numerical results.2) 2 dt 0 0 0 Analogously.1 (Energy dissipation). L)). Another important property concerning to the energy E(t ) is its decay to zero as t → ∞ (see Murray [5]). L). Mat. L).

1 ≤ j ≤ J (3. 2 (2013) i i i i . Given J. 0) The choice of forward Euler relies in the questions of numerical stability. φ 0j = φ(x j . 3. we use the standard finite difference method to analyze the qualitative properties of system (1. 0 ≤ n ≤ N . E(t ) ≤ E(0). Being consistent and stable. Di > 0. to questions on exponential decay. ∀t ≥ 0. we will apply the finite difference method to the system (1.. 1. Comput.5) ∀n.1)-(1. ∂t u nj := u n+1 − u nj j t . Apl.7) is consistent and the truncation error is O(t. x 2 ). N. Moreover.6) ∀ j. Indeed. Therefore. (3. Mat. 1 ≤ j ≤ J (3.4).1)-(1. . (3. (3. ψ0n = ψ Jn+1 = 0 = 0. 3 NUMERICAL FORMULATION IN FINITE DIFFERENCE In this section.4)-(3. . i = 1. 14. We assume the standard numerical operators applied to diffusion problems. (3. ∀ j.i i “main” — 2013/11/8 — 17:18 — page 149 — #3 i i ´ RAMOS and ALMEIDA JUNIOR 149 from where we have d E(t ) ≤ 0. Next. 2. . 2. . dt since α > 0. 0 ≤ j ≤ J.4).3) Our problem consists in to find φ nj e ψ nj satisfying ∂t φ nj − D1 ∂ x ∂x φ nj + α(φ nj − ψ nj ) = 0. 2. ∀t ≥ 0.7). 0). 6] the respective convergence. (3. 1.4)-(3. We use the following operators in finite difference to the function u: ∂x ∂ x u nj := u nj+1 − u nj + u nj−1 x 2 .2) where x j = j x and tn = nt for j = 0. Tend.7) φ0n = φ nJ +1 = 0. ∀ j. . It is easy to see that system (3.1) 0 = t0 < t1 < · · · < tn = nt < · · · < t N < t N+1 = T. ψ 0j = ψ(x j . J + 1 and n = 0. N + 1. N ∈ N we set h = x = L J +1 and t = T N+1 and we introduce the nets 0 = x 0 < x 1 < · · · < x j = j x < · · · < x J < x J +1 = L .4) ∂t ψ nj − D2 ∂ x ∂x ψ nj + α(ψ nj − φ nj ) = 0.1) shows that E(t ) is decreasing along time t . it follows by Theorem of Lax [2. . we use energy arguments to identify the stability criterion to the system (3. .  Theorem (2. . Next. it is wellknown that the numerical solutions obtained for classical diffusion equations (single diffusion equation) converge if r := t /x 2 ≤ 1/2. we refer the readers to Murray [5].

15) j =0 respectively. 0).13) = θ(x j .4)-(3. . it is the numerical counterpart of energy in (1. We will show that it is given by En := x J    |φ nj |2 + |ψ nj |2 . The first of them is given by ∂t ωnj − ∂ x ∂x ωnj = 0. N + 1. 0 ≤ j ≤ J. we note that n = x n+1 − E E J  j =0 (|θ n+1 |2 − |θ nj |2 ) = x j J  (θ n+1 + θ nj )(θ n+1 − θ nj ). 1.4)-(3.7). N + 1. The energies are given by E n := x J  |ωnj |2 and n := x E j =0 J  |θ nj |2 . . 2. ∀n. For 1 − 2r − 2αt ≥ 0 where r := t /x 2 ≤ 1/2. we obtain the numerical energy concerning to the numerical equations (3. . (3.5). . Moreover.1 i FINITE DIFFERENCE PRESERVING THE ENERGY PROPERTIES OF A COUPLED SYSTEM Numerical Energy In this section. . 2 (2013) i i i i .16) Our first result concerning to the numerical energy is given by following Theorem: Theorem 3. . .. ∀n = 0. Then. ∀ n = 1. (3. 0).4)-(3. 1 ≤ j ≤ J (3.i i “main” — 2013/11/8 — 17:18 — page 150 — #4 i 150 3. N . 0 ≤ n ≤ N (3. Apl. First. we have En := n En + E . . (3. Making this. N. An analysis similar can be made to numerical solutions of (3. ∀n.7). N + 1. 1.17) j =0 Tend. . ∀n = 0. holds En ≤ E0 .9) ω0n = ωnJ +1 = 0.12) = 0.8) j =0 Note that En is composed by norms in the space l 2 . (3. .14) = θ 0j θ Jn+1 where ωnj = φ nj + ψ nj and θ nj = φ nj − ψ nj . 2 (3. 0 ≤ n ≤ N (3. 0 ≤ j ≤ J. j j (3.7). we obtain two systems from (3.4) are limited by initial data. . .1)-(1. Mat. N . ∀ j. 14.10) ∀ j.11) ω0j = ω(x j . 1 ≤ j ≤ J (3. For the sake of simplicity. In Theorem we used the energy method to show that the solutions of (1. θ0n ∀ j. N . Proof. ∀ j. we assume the particular case D1 = D2 = 1.1. and the other one is ∂t θ nj − ∂ x ∂x θ nj + 2αθ nj = 0. Comput.

Apl.18) into (3. 14. Comput. 2 (2013) i i i i . Moreover.12) θ n+1 − θ nj = r(θ nj+1 − 2θ nj + θ nj−1) − 2αt θ nj . (3. we have J  j =0 1  n+1 2 (|θ j | + |θ nj+1 + θ nj−1 |2) 2 J θ n+1 (θ nj+1 + θ nj−1 ) ≤ j ≤ j =0 J  (3. taking into account the inequality ab ≤ (a 2 + b2)/2.i i “main” — 2013/11/8 — 17:18 — page 151 — #5 i i ´ RAMOS and ALMEIDA JUNIOR 151 and for r = t /x 2 we obtain from (3. (3. having in mind the homogeneous Dirichlet boundary conditions (3.21) |θ nj+1 − θ nj |2 . we have the following identities: J J   θ nj (θ nj+1 − 2θ nj + θ nj−1 ) = − |θ nj+1 − θ nj |2 .19) j =0 Now.18) Substituting (3. j (3. Mat. j j =0 Tend.20) j =0 J  θ n+1 θ nj j j =0 = j =0 J   θ nj j =0 = J   j =0 + r(θ nj+1 − 2θ nj + θ nj−1 ) − 2αt θ nj  θ nj  |θ nj |2 + rθ nj (θ nj+1 − 2θ nj + θ nj−1 ) − 2αt |θ nj |2 = (1 − 2αt ) J  |θ nj |2 j =0 −r J  j =0 (3. N.13)..17) we obtain n = x n+1 − E E J  j =0 = rx   n n n n n (θ n+1 + θ ) r(θ − 2θ + θ ) − 2αt θ j j +1 j j −1 j j J  j =0 (θ n+1 + θ nj )(θ nj+1 − 2θ nj + θ nj−1 ) j − 2αt x J  θ nj (θ n+1 + θ nj ) j j =0 = rx J  j =0 θ nj (θ nj+1 − 2θ nj + θ nj−1 ) + rx − 2(r + αt )x J  J  j =0 θ n+1 θ nj − 2αt x j j =0 θ n+1 (θ nj+1 + θ nj−1 ) j J  |θ nj |2 .22) (|θ n+1 |2 + |θ nj |2 ).

. More precisely.20)-(3. we obtain n+1 − E n ≤ − rx E J  j =0 |θ nj+1 − θ nj |2 + rx − 2(r + αt )(1 − 2αt )x J  (|θ n+1 |2 + |θ nj |2) j j =0 J  |θ nj |2 − 2αt x j =0 + r(2r + 2αt )x J  j =0 J  |θ nj |2 j =0 |θ nj+1 − θ nj |2 . (3.4)-(3. Therefore   4αγ t n 0 n  E ≤ 1− E .2. and then n+1 − E n ) ≤ − 4αt (1 − r − αt ) E n (1 − r)( E − r(1 − 2r − 2αt )x J  j =0 |θ nj+1 − θ nj |2 . . Taking 1 − 2r − 2αt ≥ 0 we choose γ := 1 − r − αt ≥ 0 and then   n . .i i “main” — 2013/11/8 — 17:18 — page 152 — #6 i 152 i FINITE DIFFERENCE PRESERVING THE ENERGY PROPERTIES OF A COUPLED SYSTEM Substituting (3. we consider the case D1 = D2 = 1. N + 1. 2. ∀ n = 1. Finally. . . . .7). Theorem 3. 1.19). 2E = E + E ≤ 1 − E + E0 ≤ E 1−r from where we have En ≤ E0 .2  Uniform Exponential Decay In this section. 2. n ) ≤ −4αγ t  n+1 ≤ 1 − 4αγ t E n+1 − E En ⇒ E (1 − r)( E 1−r since r < 1. N . If 1 − 2r − 2αt ≥ 0 then En is exponentially stable. N + 1. 1−r n = 0.22) into (3. N . we present a result on exponential decay to the energy of the system (3. We assure that numerical scheme preserves the important property on exponential decay such as continuous case. Here. 3. . we can written assuring that E   4αγ t n 0 n n n  0 + E 0 ≤ 2E0 . .23) n is decreasing for γ = 1 − r − αt ≥ 0.

Apl. Comput. N.. 14. − 8 sin2 πx t n 2 En ≤ E0 e x2 . Tend. 2 (2013) i i i i . Mat.

0) = 3 sin(3π x j ). Tend.045s and 20 divisions in space and 38 in time..4)-(3. Finally.i i “main” — 2013/11/8 — 17:18 — page 153 — #7 i ´ RAMOS and ALMEIDA JUNIOR i 153 Proof.24) and (3. . We use the following data: L = 1. For the other one. J + 1. n ≥ 0 and we obtain θ nj = [1 − (λk + 2α)t ]n sin(kπ x j ). . we present numerical results obtained with the scheme (3. . ∀ k. j = 1. Otherwise. T = 0. Finally.24) (see [1]). Comput. Apl..26).. ψ(x j . 2 and then n = E 0 |1 − (λk + 2α)t |2n ≤ E 0 |1 − (λ1 + 2α)t |2n ≈ E 0 e−2n(λ1 +2α)t . . In Figures 1 and 2 we see the correct behavior of the solutions φ nj and ψ nj . 14..2) result α < 0. On the other hand.14).7). We consider the uncoupled systems (3.12)-(3. they are increasing and free of numerical oscillations because the stability criterion is obeys. we note that En is stable in the exponential sense since that stability criterion (see Theorem 3. 444 · 103 . ≤ (E 0 + E Therefore. . we take the inequalities (3. where λ1 = 4 x 2 sin2 πx 2 (3. Now. According stability restrictions given by Theorems (3. . we obtain En ≤ E0 e−2nλ1 t .1) and (3. 6). E n ≤ E 0 e−2nλ1 t . J . we assume the decomposition given by θ nj = X j T n . it follows immediately that n = |1 − (λk + 2α)t |2n x E J  sin2 (kπ x j ) = j =0 π |1 − (λk + 2α)t |2n . 2. .. 4  NUMERICAL EXPERIMENTS In this section. N. from where we have 0 n ≤ e−2nλ1 t E 0 + e−2nt (λ1 +2α) E 2En = E n + E 0 )e−2nλ1 t = 2E0 e−2nλ1 t . significative changes are showed in Figures 3 and 4 for α takes of order 103. 2.e. . 2. That is to say.26) since 1 − 2r − 2αt ≥ 0.9)-(3. 2 (2013) i i i i . k = 1. to obtain the estimative of exponential decay. i. numerical instabilities occur (see Fig. j = 1. Mat. To initial data we use φ(x j . 0) = sin(π x j ).  where λk = 4 x 2 sin2 kπx 2 (3. J + 1.2) prevails. . looking to the Figure 5.11) and (3.25)  . The first of them is exponentially stable. E (3.

765 · 103 Tend.02 0.3 0 1 0.02 0.025 tn Energy 6 0 0. 2 (2013) i i i i .5 0.015 0.8 0.045 0 0 0.9 1 0. Apl.03 0.02 E −2 0.04 0. 765 · 103 6 0 −2 0.015 0.5 0. 444 · 101 n n Numerical solution: φj Numerical solution: ψj 3 3 4 4 2 2 2 φn j 2 1 ψn j 0 1 0 0 0 −2 −2 −1 −4 1 0.6 φn 0.5 0. 765 · 103 Energy 5 5 4 4 n n E 3 3 2 2 1 1 0.03 xj −2.5 −1 −3 1 0.01 0 t n Figure 2: α = 0. N.05 0.5 j 2 3 0.04 0..03 0.02 0.01 0 Figure 4: α = 0.5 0.035 t n Figure 5: α = 0.045 t n Figure 6: α = 0.04 0.02 0.025 0.04 0.2 −1.05 −2 0.6 0.03 0.035 0.2 −2 0.005 0.03 0.4 0.i i “main” — 2013/11/8 — 17:18 — page 154 — #8 i 154 i FINITE DIFFERENCE PRESERVING THE ENERGY PROPERTIES OF A COUPLED SYSTEM n n Numerical solution: φj Numerical solution: ψj 2.01 0.7 0.5 0 0 −1 −0.1 0.005 0.04 0.05 0. 444 · 101 0.04 0.8 1.01 0.5 −1 −4 1 0.05 0 0. Comput.01 0 0 xj t n Figure 1: α = 0.4 0.5 0. 14. Mat. 444 · 101 0.5 2 1 1 ψn j 0.02 0.01 0 0 x tn j Figure 3: α = 0.03 0 x j 0.5 0.

Laborat´orio Nacional de Computac¸a˜ o Cient´ıfica – LNCC – Rio de Janeiro. Rivera. it is well-known that implicit schemes are free of stability criterion (unconditionally stable). 2nd Ed. [3] J. RESUMO. Palavras-chave: equac¸o˜ es de difus˜ao. SIAM (2004). Interdisciplinary Applied Mathematics. [5] J. “Estabilizac¸a˜ o de Semigrupos e Aplicac¸o˜ es”. 2 (2013) i i i i . e´ bem conhecido que a energia do sistema e´ decrescente e exponencialmente est´avel.C. (2005). Mayers. Cambridge University Press.D.W. Tveito & R.. Neste trabalho. [6] K. Murray. S´erie de M´etodos Matem´aticos. Ao n´ıvel da dinˆamica do cont´ınuo. We illustrated this property by means of some numerical experiments. 14. (1984). “Numerical Solution of Partial Differential Equations”. Another numerical schemes can be used in order to obtain the exponential decay. decaimento exponencial num´erico. Mat. Strikwerda. Smith. REFERENCES [1] A.M. 18 (2003). Oxford Applied Mathematics and Computing Science Series. Springer Vol. (2007). n´os provamos a propriedade de decaimento exponencial da energia num´erica associada a um particular esquema num´erico em diferenc¸as finitas aplicado a um sistema acoplado de equac¸o˜ es de difus˜ao.E. “Numerical Solution of Partial Differential Equations: Finite Difference Methods”.F. Aqui n´os apresentamos em detalhes a an´alise num´erica de decaimento exponencial da energia num´erica desde que obedecido o crit´erio de estabilidade. 2nd Ed.D. [4] J. Comput. “Finite Difference Schemes and Partial Differential Equations”.. “Introduction to partial differential equations: A computation approach”. Tend. For example.i i “main” — 2013/11/8 — 17:18 — page 155 — #9 i ´ RAMOS and ALMEIDA JUNIOR 5 i 155 CONCLUSION In this paper we showed in detail the numerical analysis of exponential decay of the solutions of a numerical scheme in finite difference applied to coupled systems of diffusion equations. Therefore. [2] J. implicit schemes can be used in this context to preserve the exponential decay. diferenc¸as finitas. Winther. This scheme preserves the exponential decay since holds the stability criterion. Springer-Verlag (1998). N. Apl. Morton & D. “Mathematical Biology II – Spatial Models and Biomedical Applications”.