R. FERRETTI Dipartimento di Matematica, Universita di Roma “La Sapienza”, 00185 Rome, Italy
and B. LARROUTUROU Cermics, INRIA, Sophia-Antipolis, (Received 20 September
06560 Valbonne, France
1990; received for publication
13 March 1991)
Key words and phrases: Travelling waves, error estimates, combustion models.
1. INTRODUCTION THE MOST fundamental
problem in combustion theory concerns a steady planar pre-mixed flame in an infinite tube. In the simplest case where one assumes a one-step chemical mechanism and a unit Lewis number and when one uses the classical isobaric approximation, the steady (or travelling-front) solution of the reacting flow equations is described by the following simple model (see e.g. [3, 6, 7, lo]): in IR,
--uM + CU’ = g(u) i U(---co) = 0,
u(+w)
= 1.
(1.1)
Here, the C?’ function u is the normalized temperature defined on all of IR; 0 (resp. 1) is the normalized temperature of the fresh mixture (resp. of the burnt gases) at the left (resp. right) end of the infinite tube. The unknowns in (1.1) are both the temperature field U(X) and the positive scalar c, which measures the normalized gaseous mass flux across the steady flame: in some sense, c plays the role of an eigenvalue in this problem. In the differential equation (1.1) the terms u”, CU’ and g(u) represent respectively heat conduction, heat convection by the flow, and heat source released by the chemical reaction. We will assume that the function g has an ignition temperature 8 [see (2.11) below]. For numerical purposes, one has to approximate the solution (u, c) of (1.1): one considers a bounded domain [-a, a] (with a “large enough”) and solves for a truncated approximation of (1 .l). Several such bounded-domain problems are considered by different authors (see, e.g. [8, 9, 12]), all being of the following form: one seeks for u, E (?‘(-a, a) and c, E IF?satisfying: -u; subject
to one of the following
to one of the following
+ c, U:, = g(u,)
boundary
conditions
-U:, (-a)
+ c, u, (-a)
boundary
conditions &(a)
= 0
in (-a,
a),
(1.2)
at x = -a:
= 0
or 2.4,(-a) = 0,
(1.3)
at x = a: or u,(a) 1177
= 1,
(1.4)
1178
and to the additional
R. FERRETTIand B. LARROUTUROU
condition: u,(O) = 8.
(1.5)
Condition (1.5) is added in order to keep c, unknown in (1.2)-(1.5) (see the discussions of this point in [l, 51). Some of the problems (1.2)-(1.5) [depending on the boundary conditions (1.3) and (1.4) which are actually used] have been studied from a mathematical point of view. One can then show that the solution (u,, c,) of (1.2)-(1.5) converges as a tends to +w to a solution (u, c) of (1.1) which in addition satisfies u(O) = 8. In fact, investigating a bounded-domain problem of type (1.2)-(1.5) and studying the limiting behaviour as a tends to +co is a good way of proving the existence of a solution to the original problem (1.1) (see [2-5, 111). Because the existing convergence proofs rely on compactness arguments, they do not provide any estimate of the errors Ic, - cl and IIu, - ~11. It is precisely our purpose in this work to derive such error estimates, whose interest for numerical calculations is clear. Under the appropriate conditions on the nonlinear function g, we will prove that the convergence of there exist a positive constant K and an explicitly known positive (u, 9c,) to (u, c) is exponential: constant k such that, for a large enough, the following holds: Ic, - cl I KePka,
IIu, - ~ll~z~-~,~)I Keek”.
(1.6)
The precise hypotheses and results are stated in theorems 3.1, 3.2, 3.6 and 4.2 below. The paper is organized as follows. We first recall, in Section 2, some known results about problem (1.1). The new mathematical results, and in particular error estimates of the form (1.6) are presented in Sections 3 and 4. Our main result is presented in Section 4, where we show that using a new boundary condition at x = a [instead of one of the conditions (1.4)] accelerates the convergence of the corresponding truncated solution to (u, c) (see theorem 4.2 below). Lastly, we illustrate the various convergence (or divergence) results by showing some numerical examples in Section 5. 2. SOME
We recall in this section
several known
KNOWN
results
RESULTS
about
2. I. A comparison principle We begin with a comparison principle for solutions with (1.1). This result was proved in [2, 31.
problem
(1.1) and related
of initial value problems
questions.
(IVP) associated
PROPOSITION 2.1. Let g, and g, be two Lipschitz-continuous functions, and let x0, 6, cl, c2, 01~,CY~ be given real numbers. For i = 1,2, consider the solutions ut and u2 of the forward IVP:
-Ui’
I
+ CjUi' = gi(Uj)
ui(xlJ = 6,
If:
g1
5
on [x0, +w),
&(x(J) = a;. g2
on
14
+a),
Cl zc21
011 2 cY2> 0, then ut (x) r u,(x)
for any x > x,, such that u; > 0 and ui > 0 on [x0, x).
(2.1)
(2.2)
(2.3) (2.4)
Error estimates
for travelling
1179
wave solutions
Remark 2.1. The proof of this result, which we do not present in detail here, relies on inverting the functions ui and u2 in the intervals where they are monotone increasing. Setting x;(s) = u;l(s) (i.e. i(;(xi(s)) = s) for i = 1, 2 and for s in some interval [a, S’), we can define Zi(S) = (d/ds)xi(s) and vi(s) = Zi(S)~l = u,!(xi(s)). Then, the functions yi satisfy:
(2.5) and one can show that y, 2 y, in [a, S’], whence [2, 31 for the details.
zi I z2, x, 5 x2 and u1 1 u2. We refer to
Remark 2.2. With the assumptions of proposition 2.1, one can also show that, if one of the two inequalities (2.3) or (2.4) is strict, then U,(X) > uZ(x) for any x > x0 such that u; > 0 and u; > 0 on [x0, x). We can also state a similar
result for the backward
IVP.
PROPOSITION 2.2. Let g, and g, be two Lipschitz-continuous functions, and let x0, 6, cl, c2, a,, 01~ be given real numbers. For i = 1,2, consider the solutions ui and u2 of the backward IVP: -Ul’ + ciu;