Professional Documents
Culture Documents
Numerical Approximation of Hyperbolic Systems of Conservation Laws, 2nd Edition (Applied Mathematical Sciences Series) Edwige Godlewski
Numerical Approximation of Hyperbolic Systems of Conservation Laws, 2nd Edition (Applied Mathematical Sciences Series) Edwige Godlewski
https://ebookmeta.com/product/analysis-of-legal-argumentation-
documents-a-computational-argumentation-approach-translational-
systems-sciences-29-hayato-hirata/
https://ebookmeta.com/product/diophantine-approximation-and-the-
geometry-of-limit-sets-in-gromov-hyperbolic-metric-spaces-1st-
edition-lior-fishman/
https://ebookmeta.com/product/elliptic-regularity-theory-by-
approximation-methods-london-mathematical-society-lecture-note-
series-edgard-a-pimentel/
https://ebookmeta.com/product/mathematical-modeling-the-life-
sciences-numerical-recipes-in-python-and-matlab-1st-edition-n-g-
cogan/
Stochastic Approximation A Dynamical Systems Viewpoint
2nd Edition Vivek S Borkar
https://ebookmeta.com/product/stochastic-approximation-a-
dynamical-systems-viewpoint-2nd-edition-vivek-s-borkar/
https://ebookmeta.com/product/b-series-algebraic-analysis-of-
numerical-methods-john-c-butcher/
https://ebookmeta.com/product/weighted-polynomial-approximation-
and-numerical-methods-for-integral-equations-1st-edition-peter-
junghanns/
https://ebookmeta.com/product/lattice-rules-numerical-
integration-approximation-and-discrepancy-josef-dick-peter-
kritzer-friedrich-pillichshammer/
https://ebookmeta.com/product/conservation-of-architectural-
heritage-2nd-edition-antonella-versaci/
Applied Mathematical Sciences
Edwige Godlewski
Pierre-Arnaud Raviart
Numerical
Approximation
of Hyperbolic Systems
of Conservation Laws
Second Edition
Applied Mathematical Sciences
Volume 118
Series Editors
Anthony Bloch, Department of Mathematics, University of Michigan, Ann Arbor,
MI, USA
abloch@umich.edu
C. L. Epstein, Department of Mathematics, University of Pennsylvania,
Philadelphia, PA, USA
cle@math.upenn.edu
Alain Goriely, Department of Mathematics, University of Oxford, Oxford, UK
goriely@maths.ox.ac.uk
Leslie Greengard, New York University, New York, NY, USA
Greengard@cims.nyu.edu
Advisory Editors
J. Bell, Center for Computational Sciences and Engineering, Lawrence Berkeley
National Laboratory, Berkeley, CA, USA
P. Constantin, Department of Mathematics, Princeton University, Princeton, NJ,
USA
R. Durrett, Department of Mathematics, Duke University, Durham, CA, USA
R. Kohn, Courant Institute of Mathematical Sciences, New York University,
New York, NY, USA
R. Pego, Department of Mathematical Sciences, Carnegie Mellon University,
Pittsburgh, PA, USA
L. Ryzhik, Department of Mathematics, Stanford University, Stanford, CA, USA
A. Singer, Department of Mathematics, Princeton University, Princeton, NJ, USA
A. Stevens, Department of Applied Mathematics, University of Münster, Münster,
Germany
S. Wright, Computer Sciences Department, University of Wisconsin, Madison, WI,
USA
Founding Editors
F. John, New York University, New York, NY, USA
J. P. LaSalle, Brown University, Providence, RI, USA
L. Sirovich, Brown University, Providence, RI, USA
The mathematization of all sciences, the fading of traditional scientific boundaries,
the impact of computer technology, the growing importance of computer modeling
and the necessity of scientific planning all create the need both in education and
research for books that are introductory to and abreast of these developments. The
purpose of this series is to provide such books, suitable for the user of mathematics,
the mathematician interested in applications, and the student scientist. In particular,
this series will provide an outlet for topics of immediate interest because of the nov-
elty of its treatment of an application or of mathematics being applied or lying close
to applications. These books should be accessible to readers versed in mathematics
or science and engineering, and will feature a lively tutorial style, a focus on topics
of current interest, and present clear exposition of broad appeal. A compliment to
the Applied Mathematical Sciences series is the Texts in Applied Mathematics se-
ries, which publishes textbooks suitable for advanced undergraduate and beginning
graduate courses.
Numerical Approximation
of Hyperbolic Systems of
Conservation Laws
Second Edition
Edwige Godlewski Pierre-Arnaud Raviart
Laboratoire Jacques-Louis Lions Laboratoire Jacques-Louis Lions
Sorbonne University Sorbonne University
Paris, France Paris, France
Mathematics Subject Classification: 35L65, 35L67, 65M06, 65M08, 65M12, 76Nxx, 35L50, 35L60,
35Q35, 65Mxx, 35Q20, 35Q86, 76P05, 76W05, 80A32
This Springer imprint is published by the registered company Springer Science+Business Media, LLC,
part of Springer Nature
The registered company address is: 1 New York Plaza, New York, NY 10004, U.S.A.
Preface to the Second Edition
There was an obvious need to complete the first edition of this textbook
with the treatment of source terms. Thus, a new chapter (Chap. VII) has
been added, which also provides a few important principles concerning non-
conservative systems that are naturally introduced with the derivation of
well-balanced or asymptotic preserving schemes. Note that most theoretical
results are only referred to since it is out of scope to give detailed proofs;
these may be tricky and are often quite technical.
We took the opportunity of this second edition to include more examples
in the introduction chapter (now Chap. I), such as MHD, shallow water, and
flow in a nozzle, and to give some insights on multiphase flow models; this last
subject deserves a much longer treatment. Then we thought it is important
to emphasize the change of frame from Eulerian to Lagrangian coordinates
and the specificity of fluid systems. Additionally, the low Mach limit has been
addressed in the chapter devoted to multidimensional systems (now Chap. V)
with the final section introducing all Mach schemes.
For 25 years, there has been a tremendous lot of work dedicated to the
numerical approximation of hyperbolic systems, among which we choose to
introduce the relaxation approach, now at the end of Chap. IV and the case
of discontinuous fluxes, and interface coupling, a topic covered in Chap. VII.
Both subjects are treated in some specific outlines.
Then, some complements may be found here and there, such as recalling
some results of our earlier publication at the beginning of Chap. IV, or more
examples of systems of two equations in Chap. II.
We must finally confess that it took us some time to complete the work
of this second edition, for different reasons. In fact, most of this work was
achieved several years ago, which may explain why only few very recent results
are presented, some of them are just mentioned in the notes at the end of
each chapter, to give a hint and provide references where the subject is more
thoroughly treated.
v
Preface to the First Edition
1 The numbering of the chapters has changed in the second edition, the Introduction is
now Chap. I. Hence in what follows, Chap. I refers to what is now Chap. II and so on.
vii
viii Preface to the First Edition
simplest models of reacting flow, first the Chapman-Jouguet theory and then
the Z.N.D. model for detonation.
After this theoretical approach, we go into the numerical approximation of
hyperbolic systems by conservative finite-difference methods. The most usual
schemes for one-dimensional systems are developed in Chap. III, with special
emphasis on the application to gas dynamics. The last section begins with a
short account on the kinetic theory so as to introduce kinetic schemes.
Chapter IV is devoted to the study of finite volume methods for bidimen-
sional systems, preceded by some theoretical considerations on multidimen-
sional systems.
For the sake of completeness, we could not avoid the problem of boundary
conditions. Chapter V is but an introduction to the complex theory and
presents some numerical boundary treatment.
The authors wish to thank R. Abgrall, F. Coquel, F. Dubois, and particu-
larly T. Gallouet, B. Perthame, and D. Serre, from whom they learned a great
deal and who answered willingly and most amiably their many questions.
They owe thanks to the SMAI reading committee and to the reviewers,
who made very valuable suggestions.
The first author is grateful to all her colleagues who encouraged her in
completing this huge work, especially to H. Le Dret and F. Murat for so
often giving her their time, and to L. Ruprecht for her kind and competent
assistance in the retyping of the final manuscript; such friendly help was
invaluable.
I Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Definitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Fluid Systems in Eulerian and Lagrangian Frames . . . . . . . . . . 6
3 Some Averaged Models: Shallow Water, Flow in a Duct, and
Two-Phase Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4 Weak Solutions of Systems of Conservation Laws . . . . . . . . . . . 27
4.1 Characteristics in the Scalar One-Dimensional Case . . . 27
4.2 Weak Solutions: The Rankine-Hugoniot Condition . . . . 30
4.3 Example of Nonuniqueness of Weak Solutions . . . . . . . . 35
5 Entropy Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.1 A Mathematical Notion of Entropy . . . . . . . . . . . . . . . . . 37
5.2 The Vanishing Viscosity Method . . . . . . . . . . . . . . . . . . . 44
5.3 Existence and Uniqueness of the Entropy Solution in
the Scalar Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
ix
x Contents
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 749
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 831
I
Introduction
∂u ∂
d
+ fj (u) = 0, x = (x1 , . . . , xd ) ∈ Rd , t > 0, (1.1)
∂t j=1
∂xj
where ⎛ ⎞
u1
⎜ ⎟
u = ⎝ ... ⎠
up
is a vector-valued function from Rd × [0, +∞[ into Ω. The set Ω is called the
set of states and the functions
⎛ ⎞
f1j
⎜ .. ⎟
fj = ⎝ . ⎠
fpj
are called the flux functions. One says that system (1.1) is written in
conservative form.
In the sequel, we will also write the system (1.1) in the form
∂u
+ ∇ · f (u) = 0
∂t
where f stands for the matrix-valued function
f = (fij )1≤i≤p,1≤j≤d
and ∇· is the divergence operator (we will equivalently use the notation div)
d
∂
∇·f = fj .
j=1
∂x j
d
d
u dx + fj (u) nj dS = 0.
dt D j=1 ∂D
This balance equation has now a very natural meaning: the time variation of
D
u dx is equal to the losses through the boundary ∂D.
In all the following, we shall be concerned with the study of hyperbolic
systems of conservation laws, which we define in the following way. For all
j = 1, . . . , d, let
∂fij
Aj (u) = (u)
∂uk 1≤i,k≤p
be the Jacobian matrix of fj (u); the system (1.1) is called hyperbolic if, for
any u ∈ Ω and any ω = (ω1 , . . . , ωd ) ∈ Rd , ω = 0, the matrix
d
A(u, ω) = ωj Aj (u)
j=1
If, in addition, the eigenvalues λk (u, ω) are all distinct, the system (1.1) is
called strictly hyperbolic.
In fact, little is known about systems in more than one space variable unless
they are symmetrizable, i.e., there exists for all u ∈ Ω a symmetric positive-
definite matrix A0 (u) smoothly varying with u such that the matrices
For such systems, we shall study the Cauchy problem, or initial value prob-
lem (IVP): find a function u : (x, t) ∈ Rd × [0, ∞[→ u(x, t) ∈ Ω that is a
solution of (1.1) satisfying the initial condition
u , x<0
u0 (x) = (1.3)
ur , x > 0,
∂u ∂u ∂2u
+u −ν =0 (1.4a)
∂t ∂x ∂x2
was introduced in particular by Burgers as the simplest differential model
for a fluid flow and is therefore often called the (viscous) Burgers’ equation.
Though very simple, this equation can be regarded as a model for decaying
free turbulence (see Cole [325]). A number of authors have developed the
asymptotic theory of Navier-Stokes equations in terms of Burgers’ equation,
and it is thus often used in numerical tests [1059]. Burgers studied the limit
equation when ν tends to zero, which we write in conservation form
∂u ∂ u2
+ = 0. (1.4b)
∂t ∂x 2
Equation (1.4b) is the inviscid Burgers’ equation (or Burgers’ equation with-
out viscosity), which, for brevity, we shall simply call from now on Burg-
ers’ equation. It occurs in particular in wave theory to depict the distortion
of waveform in simple waves (see Lighthill [800, Sec. 2.9], Whitham [1188,
Sec. 2.8]).
We shall see that Burgers’ equation possesses all the features of a scalar
convex equation
∂u ∂
+ f (u) = 0, (1.5)
∂t ∂x
where f : R → R is a convex smooth function. In particular, the Cauchy
problem for Burgers’ equation may have discontinuous weak solutions even
for a smooth initial function u0 , and the solution of the Riemann problem
4 I. Introduction
is either a shock propagating or a rarefaction wave (see Fig. 4.2), both being
kinds of waves that are involved in the solution of the Riemann problem for
a system.
Remark 1.1. It will be very easy to derive the properties of Eq. (1.5) when the
flux f is concave, from the convex case. An example of a concave flux is illus-
trated by the classical LWR traffic model (for Lighthill-Whitham-Richards)
for which
ρ
f (ρ) = umax ρ(1 − ),
ρmax
where ρ ∈ [0, ρmax ] measures the density of cars and umax is a maximum
velocity.
Finally, it is worth mentioning that the Cauchy problem for (1.4a) has an
explicit solution, obtained using the Cole-Hopf transform
ϕx
u = −2ν .
ϕ
It eliminates the nonlinear term and transforms (1.4a) into the heat equation
∂ϕ ∂2ϕ
= ν 2,
∂t ∂x
for which explicit expressions of the solution are known. For details, we refer
to the original papers of Hopf [630], Cole [325], and Whitham (1974, Chap-
ter 4) [1188] where a thorough study of Eq. (1.4a) (including limit as ν → 0
and shock structure) can be found.
kr,w (s)/μw
f (s) = , (1.6)
kr,w (s)/μw + kr,o (s)/μo
where μ denotes the viscosity (supposed to be constant) and kr (s) the relative
permeability of a fluid (the indices w and o refer to water and oil, respec-
tively); the set of states is the interval [0, 1]. A typical example is obtained
by taking kr,w = s2 , kr,o (s) = (1 − s)2 , so that f writes
s2
f (s) = ,
s2 + (1 − s)2 μw /μo
p(v) = Av −γ
for some constants A = A(s) > 0 (depending on the constant entropy) and
γ ≥ 1.
System (1.7) is of the form
∂w ∂
+ f (w) = 0,
∂t ∂x
where
v −u
w= , f (w) = , Ω = {(v, u) ∈ R2 ; v > 0}.
u p(v)
and the system (1.7) is strictly hyperbolic provided that we assume p (v) < 0.
In that case, the matrix A(w) (which depends only on v) has indeed two real
distinct eigenvalues
λ1 = − (−p (v)) < λ2 = (−p (v)).
Besides the scalar case, the p-system (1.7) is the simplest nontrivial ex-
ample of a nonlinear system of conservation laws. Note that any nonlinear
wave equation
∂2g ∂ ∂g
− σ =0
∂t2 ∂x ∂x
can be put in the form (1.7) by setting
∂g ∂g
u= , v= , p(v) = −σ(v),
∂t ∂x
as one can easily check.
6 I. Introduction
System (1.7) is the simplest model for a fluid system in Lagrangian coordi-
nates. We now consider the full system of gas dynamics, an example that
appears to be fundamental in the applications, without assuming the flow is
isentropic, and in the general multidimensional case.
Example 2.1. The equations of gas dynamics in Eulerian coordinates. In Eule-
rian coordinates, the Euler equations for a compressible inviscid fluid (where
we neglect heat conduction) can be written in the following conservative form:
⎧
⎪ ∂ρ 3 ∂
⎪
⎪ + (ρuj ) = 0,
⎪
⎪ ∂t ∂x
⎪
⎪ j=1 j
⎪
⎨ ∂ 3 ∂
(ρui ) + (ρui uj + p δij ) = 0, 1 ≤ i ≤ 3, (2.1)
⎪
⎪ ∂t j=1 ∂xj
⎪
⎪
⎪
⎪
⎪
⎪ ∂ 3 ∂
⎩ (ρe) + ((ρe + p)uj ) = 0.
∂t j=1 ∂x j
the specific total energy. Equations (2.1) express respectively the laws of
conservation of mass, momentum, and total energy for the fluid.
Note that (2.1) may be equivalently written
⎧
⎪ ∂ρ
⎪
⎪ + ∇ · (ρu) = 0,
⎪
⎪ ∂t
⎪
⎪
⎨
∂
(ρu) + ∇ · (ρu ⊗ u + pI) = 0,
⎪ ∂t
⎪
⎪
⎪
⎪
⎪
⎪
⎩ ∂ (ρe) + ∇ · ((ρe + p)u) = 0
∂t
where I is the d × d identity matrix and u ⊗ u denotes the matrix-valued
function (u ⊗ u)ij = ui uj . More generally, for v ∈ Rp , w ∈ Rq , we define the
p × q matrix v ⊗ w by
(v ⊗ w)ij = vi wj , 1 ≤ i ≤ p, 1 ≤ j ≤ q.
p = p(ρ, ε),
p = (γ − 1)ρ ε, γ > 1.
Now, setting
qi = ρ u i , 1 ≤ i ≤ 3, E = ρ e,
the system (2.1) can be put into the general framework of system (1.1) if we
take
⎛ ⎞ ⎛ ⎞
ρ q1
⎜q1 ⎟ ⎜ p + q12 /ρ ⎟
⎜ ⎟ ⎜ ⎟
U=⎜ ⎟ ⎜
⎜q2 ⎟ , f1 (U) = ⎜ q1 q2 /ρ ⎟ ,
⎟ (2.2a)
⎝q3 ⎠ ⎝ q1 q3 /ρ ⎠
E (E + p)q1 /ρ
⎛ ⎞ ⎛ ⎞
q2 q3
⎜ q1 q2 /ρ ⎟ ⎜ q1 q3 /ρ ⎟
⎜ ⎟ ⎜ ⎟
f2 (U) = ⎜
⎜ p + q 2
2 /ρ ⎟
⎟ , f 3 (U) = ⎜ q2 q3 /ρ ⎟
⎜ ⎟ (2.2b)
⎝ q2 q3 /ρ ⎠ ⎝ p + q32 /ρ ⎠
(E + p)q2 /ρ (E + p)q3 /ρ
with
E |q|2
p = p ρ, − 2
ρ 2ρ
and if the set of states is
|q|2
Ω= (ρ, q = (q1 , q2 , q3 ), E); ρ > 0, q ∈ R , E −
3
>0 .
2ρ
∇ · B = 0. (2.6)
2 Fluid Systems in Eulerian and Lagrangian Frames 9
∇ · (∇ · (B ⊗ u − u ⊗ B)) = 0
If we assume that the flow is invariant with respect to (x2 , x3 ), the ideal
MHD equations read (using obvious notations)
10 I. Introduction
⎧
⎪ ∂ρ ∂
⎪
⎪ + (ρu) = 0,
⎪
⎪ ∂t ∂x
⎪
⎪
⎪
⎪ ∂ ∂
⎪
⎪ (ρu) + (ρu2 + p∗ ) = 0,
⎪
⎪ ∂t ∂x
⎪
⎪
⎪
⎪ ∂ ∂ Bx By
⎪
⎪ (ρv) + (ρuv − ) = 0,
⎪
⎪
⎪
⎪
∂t ∂x μ
⎪
⎪ ∂ ∂ Bx Bz
⎨ (ρw) + (ρuw − ) = 0,
∂t ∂x μ (2.7)
⎪
⎪
⎪
⎪ ∂By ∂
⎪
⎪ + (By u − Bx v) = 0,
⎪
⎪ ∂t ∂x
⎪
⎪
⎪
⎪ ∂Bz ∂
⎪
⎪ + (Bz u − Bx w) = 0,
⎪
⎪ ∂t ∂x
⎪
⎪
⎪
⎪ ∂ ∂ Bx
⎪
⎪ (ρe∗ ) + ((ρe∗ + p∗ )u − (Bx u + By v + Bz w)) = 0,
⎪
⎪ ∂t ∂x μ
⎪
⎩
Bx = constant.
We shall show in the next chapter that the ideal MHD system is indeed
hyperbolic (but not strictly hyperbolic) and symmetrizable in the whole set
of states Ω.
More generally, a number of nonlinear hyperbolic systems of conservation
laws arising in fluid mechanics can be written in the form
⎧
⎪ ∂ρ
⎪
⎨ ∂t + ∇ · (ρu) = 0,
(2.8)
⎪
⎪
⎩ ∂ (ρΦ) + ∇ · (ρΦ ⊗ u + g(ρ, Φ)) = ρs(ρ, Φ)
∂t
where ⎛ ⎞
Φ1
⎜ ⎟
Φ = ⎝ ... ⎠ , g = (gij )1≤i≤p−1,1≤j≤3 .
Φp−1
In (2.8), ρ is again the mass density of the fluid, u is its velocity while ρΦ
stands for the other conservative variables, and s is a source term. Note that,
in (2.8), we have distinguished the mass conservation equation from the other
conservation laws. Indeed, the equations of gas dynamics and the ideal MHD
equations fit into the framework of such a general system (2.8) with
u pI
Φ= , g(ρ, Φ) = (2.9)
e pu
and
2 Fluid Systems in Eulerian and Lagrangian Frames 11
⎛ ⎞
⎛ ⎞ 1
u p∗ I − B⊗B
⎜ μ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ −u ⊗ B ⎟
Φ = ⎜B⎟ , g(ρ, Φ) = ⎜ ⎟ (2.10)
⎝ρ⎠ ⎜ ⎟
⎝ 1 ⎠
e∗ p∗ u − (u · B)B
μ
respectively, and s = 0.
Let us now write the system (2.8) using Lagrangian coordinates. Consider
the differential system
dx
= u(x, t) (2.11)
dt
and, for all ξ ∈ R3 , we denote by t → x(ξ, t) the solution of (2.11) that
satisfies the initial condition
x(ξ, 0) = ξ. (2.12)
∂ϕ ∂
3
∂ ∂ϕ
(ϕ J) = J + ∇ · (ϕu) = J + (ϕuj ) .
∂t ∂t ∂t j=1
∂xj
Let us put this system in a more classical form. The first equation (of mass
conservation) gives
ρ J = ρ0 ,
where ρ0 (ξ) = ρ(ξ, 0). If we introduce the specific volume
1
τ= ,
ρ
we get
J = ρ0 τ
so that the one-dimensional analogue of (2.14)
∂J ∂u ∂u
=J =
∂t ∂x ∂ξ
becomes
∂τ ∂u
ρ0 − = 0,
∂t ∂ξ
and the second equation (2.15) gives
∂Φ ∂g
ρ0 + = ρ0 s.
∂t ∂ξ
Finally, we introduce a mass variable m such that
dm = ρ0 dξ.
2 Fluid Systems in Eulerian and Lagrangian Frames 13
Then the system (2.16) written in Lagrangian coordinates takes the simple
form ⎧
⎪ ∂τ ∂u
⎪
⎨ ∂t − ∂m = 0,
(2.19)
⎪ ∂Φ
⎪ ∂g
⎩ + = s,
∂t ∂m
with s = s(1/τ, Φ).
Remark 2.1. Let us have a look at the two-dimensional case. We now
write (2.8) in the form (without source for simplicity)
⎧
⎪
⎪ ∂ρ ∂ ∂
⎨ + (ρu) + (ρv) = 0,
∂t ∂x ∂y
⎪
⎪ ∂ ∂ ∂
⎩ (ρΦ) + (ρuΦ + g(ρ, Φ)) + (ρvΦ + h(ρ, Φ)) = 0
∂t ∂x ∂y
∂g ∂h ∂g ∂ξ ∂g ∂η ∂h ∂ξ ∂h ∂η
J( + ) = J( + + + )
∂x ∂y ∂ξ ∂x ∂η ∂x ∂ξ ∂y ∂η ∂y
∂g ∂y ∂g ∂y ∂h ∂x ∂h ∂x
= − − + )
∂ξ ∂η ∂η ∂ξ ∂ξ ∂η ∂η ∂ξ
∂ ∂y ∂x ∂ ∂y ∂x
= (g −h )+ (−g + h ).
∂ξ ∂η ∂η ∂η ∂ξ ∂ξ
∂x ∂y ∂x ∂y
Setting a = ,b = ,c = ,d = , we have J = ad − bc, and we
∂ξ ∂ξ ∂η ∂η
notice that the system in Lagrangian coordinates can be equivalently written
(dropping the bars)
14 I. Introduction
⎧ ∂τ ∂ ∂
⎪
⎪ ρ0 − (ud − vc) − (−ub + va) = 0,
⎪
⎪ ∂t ∂ξ ∂η
⎪
⎪
⎪
⎪ ∂Φ ∂ ∂
⎪
⎪
⎨ ρ0 ∂t + ∂ξ (gd − hc) + ∂η (−gb + ha) = 0,
⎪
⎪ ∂ a ∂ u
⎪
⎪ − = 0,
⎪
⎪ ∂t b ∂ξ v
⎪
⎪
⎪
⎪
⎩ ∂ c − ∂ u = 0.
∂t d ∂η v
we obtain the system of gas dynamics (2.4) which reads in Lagrangian coor-
dinates ⎧
⎪ ∂τ ∂u
⎪
⎪ − = 0,
⎪
⎪ ∂t ∂m
⎪
⎪
⎪
⎪ ∂u ∂p
⎪
⎪ + = 0,
⎪
⎪ ∂t ∂m
⎪
⎨
∂v
= 0, (2.20)
⎪ ∂t
⎪
⎪
⎪
⎪
⎪ ∂w
⎪
⎪ = 0,
⎪
⎪ ∂t
⎪
⎪
⎪
⎪
⎩ ∂e + ∂(pu) = 0
∂t ∂m
2
with p = p(τ, ε) = p(τ, e − u2 ). By setting
⎛ ⎞ ⎛ ⎞
τ −u
⎜u⎟ ⎜ p ⎟
⎜ ⎟ ⎜ ⎟ |u|2
V= ⎜ ⎟ ⎜ ⎟
⎜ v ⎟ , f (V) = ⎜ 0 ⎟ , Ω = V; τ > 0, u = (u, v, w) ∈ R , e− 2 > 0 ,
3
⎝w ⎠ ⎝ 0 ⎠
e pu
|B|2 1 τ |B|2
p∗ = p(τ, ε) + , e∗ = ε + (u2 + v 2 + w2 ) + .
2μ 2 2μ
∂u ∂
d
+ fj (u) = s(u), x = (x1 , . . . , xd ) ∈ Rd , t > 0, (2.23)
∂t j=1
∂xj
Example 2.6. The equations of reacting gas flows. Consider the more complex
example of a system of reacting gas flows. We assume that the system is
multicomponent, consisting of n reacting species. If we neglect again viscosity
and heat conduction, the equations of the flow are obtained by adjoining
to the Euler system of gas dynamics the equations governing the chemical
reactions between species
∂
(ρYi ) + ∇ · (ρYi u) = mi ωi , 1 ≤ i ≤ n − 1. (2.27)
∂t
In (2.27), Yi is the mass fraction; mi is the molar mass, also termed the
molecular weight, of the ith species; and ωi is the molar production rate of
the ith species with
n n
Yi = 1, mi ωi = 0, (2.28)
i=1 i=1
n
1 p
ε= Yi h0i +
i=1
γ(Y) − 1 ρ
Cv,i (resp. Cp,i ) denotes its specific heat at constant volume (resp. at con-
stant pressure). For more details on ideal gases, see Chap. III, Sect. 1.2. If
Cp,i
in addition γi = is independent of i, i.e., all the species have the same
Cv,i
adiabatic exponent γ, we obtain γ(Y) = γ.
Now, going back to the general case, we set
and
U fj 0
W= , gj = , 1 ≤ j ≤ 3, s=
ρY qj Y ρr
where we have used the notations (2.2a), (2.2b) of Example 2.1. We obtain
the system
∂W
+∇·g =s
∂t
which is indeed of the form (2.23) (note that we may write the flux with
ρY
Y= , i.e., in terms of the conservative variables).
Assuming slab symmetry, the equations of reacting gas flows become
⎧
⎪
⎪
∂ρ ∂
⎪
⎪ + (ρu) = 0,
⎪
⎪ ∂t ∂x
⎪
⎪
⎪
⎪ ∂ ∂
⎨ (ρu) + (ρu2 + p) = 0,
∂t ∂x (2.31)
⎪
⎪ ∂ ∂
⎪
⎪ (ρe) + ((ρe + p)u) = 0,
⎪
⎪ ∂t ∂x
⎪
⎪
⎪ ∂
⎪ ∂
⎩ (ρY) + (ρYu) = ρr.
∂t ∂x
In Lagrangian coordinates, the last equation (2.31) gives
∂
(ρY J) = Jρr
∂t
or equivalently by the mass conservation equation
2 Fluid Systems in Eulerian and Lagrangian Frames 19
∂Y
= r.
∂t
Hence, (2.31) gives in Lagrangian coordinates (dropping the bars)
⎧
⎪ ∂τ ∂u
⎪
⎪ − = 0,
⎪
⎪ ∂t ∂m
⎪
⎪
⎪ ∂u
⎪ ∂p
⎪
⎨ + = 0,
∂t ∂m
(2.32)
⎪
⎪ ∂e ∂(pu)
⎪
⎪ + = 0,
⎪
⎪ ∂t ∂m
⎪
⎪
⎪
⎪
⎩ ∂Y = r.
∂t
We next restrict ourselves to the simple case where we consider only two
species, the unburnt gas and the burnt gas. We denote by z the mass fraction
of the burnt gas so that 1 − z is the mass fraction of the unburnt gas. The
simplest model is obtained by assuming that on the one hand the unburnt
gas is converted to the burnt gas through an irreversible chemical reaction
with a production rate
r = r(ρ, p, z) (2.33)
and on the other hand that both unburnt and burnt gases behave as poly-
tropic ideal gases with the same adiabatic exponent γ so that the equation
of state reads
1 p
ε(ρ, p, z) = zh0b + (1 − z)h0u + .
γ−1ρ
Observe that h0b < h0u for an exothermic reaction. In this simple case, (2.27)
resumes to
∂
(ρz) + ∇ · (ρzu) = ρr.
∂t
If we assume that the kinetics of the chemical reaction obeys an Arrhenius
mechanism, we have
E∗
r = K(1 − z) exp − ,
RT
r = K(1 − z)α ,
with α = 1
2 or 1.
One can even consider systems of conservation laws more general than
(2.23), namely, systems of the form
20 I. Introduction
∂u
+ ∇ · f (x, t, u) = s(x, t, u), x ∈ Rd , t > 0, (2.34)
∂t
where the flux function f = (fj )1≤j≤d and the source term s are smooth
functions from Rd × [0, +∞[×Ω into (Rp )d and Rp , respectively.
Example 3.2. Flow in a duct with variable cross section. We now give another
example of a system for which the source term comes from an averaging
22 I. Introduction
We consider the integral of the first equation of (1.7) and get for the first
term
x+δx
∂t ρ(x, t)dx = ∂t ( ρ(x, t)dx) = ∂t A(ξ)ρ(ξ, t)dξ
ω ω x
where ρ(x, t) is the average density over the section S(x). Then we write
∇ · (ρu)dx = ρu.ndσ.
ω ∂ω
Now, the surface ∂ω is made of the two parallel sections, say Sx (resp. Sx+δx )
with outward normal −e1 (resp. e1 ), where e1 denotes the unit vector along
the x axis, and with length A(x) (resp. A(x + δx)) and a lateral surface which
we denote Σ, then
= + + .
∂ω Sx Sx+δx Σ
Following the same lines, we obtain the terms ∂t ρu + ∂x (ρu2 + p). However,
there is now an added contribution coming from the pressure term p e1 inte-
grated on the lateral surface Σ. The tangent vector to the upper part of Σ is
(1, A (x)/2)T and the normal to the upper part (resp. lower) is (−A (x)/2, 1)T
(resp. (−1, −A (x)/2)T ), and we add the equal contributions of the upper and
lower parts. Hence, if n denotes the unit normal vector (obtained by normal-
izing the normal vector (−A (x)/2, 1)T ) and dσ is the arc length measure
along Σ, we get
x+δx
1
p(x, t) e1 · n dσ = p(ξ, A(ξ)/2, t) + p(ξ, −A(ξ)/2, t) A (ξ)dξ.
Σ 2 x
yields
∂t (Aρv) + ∂x (A(ρuv)) − δp = 0,
with δp(x, t) = 12 (p(x, A(x)/2, t) − p(x, −A(x)/2, t). Now, if we add the as-
sumption that the flow is symmetric, the average velocity v vanishes, similarly
ρv = 0, and also the term corresponding to the pressure Σ p(x, t) e2 · n dσ =
0, we can then ignore the second momentum equation.
The energy equation comes as the other ones since we integrate the term
∇·((e+p)u), and again, only a term with u.n is involved. This term vanishes
on Σ and is easily computed on the cross sections S(x), S(x + δx).
Let us assume last that both pressure terms p and p can be replaced by
p(ρ, ε), and ρu2 by (ρu)2 /ρ̄, then dropping the bars for simplicity, we get the
system ⎧
⎪ ∂ ∂
⎪
⎪ (Aρ) + (Au) = 0,
⎪
⎪ ∂t ∂t
⎪
⎪
⎨
∂ ∂
(Aρu) + (A(ρu2 + p)) = pA (x), (3.3)
⎪
⎪ ∂t ∂x
⎪
⎪
⎪
⎪
⎪
⎩ ∂ (Aρe) + ∂ (A(e + p)u) = 0,
∂t ∂x
which also looks like the Euler system, provided we define a surfacic density
r = Aρ, with a source term which is generally called a geometric source term
(when A is not constant). However, we have to look more precisely at the
pressure law given in terms of conservative variables. Assuming for simplicity
24 I. Introduction
velocity and pressure at the interface, among which some simplifications are
usually done to lead to a tractable model. We do not go into details for which
we refer to [965, 1073], and give only some simple examples.
Assume for simplicity that the flow is isentropic. We first consider the
case where the two fields have an equal pressure p. It leads to a four-equation
model, which writes, with k = 1, 2 and α1 + α2 = 1,
⎧
⎪ ∂ ∂
⎪
⎨ ∂t (αk ρk ) + ∂x (αk ρk uk ) = 0,
(3.5)
⎪
⎪
⎩ ∂ (αk ρk uk ) + ∂ (αk ρk u2 ) + αk ∂ p + Δp ∂ αk = Mk ,
k
∂t ∂x ∂x ∂x
with equations of state ρi = ρi (p) and where we have noted Δp = p − pI , a
pressure default. Here pI denotes an interfacial pressure and Δp is some given
function of the state u = (αρ1 , (1 − α)ρ2 , αρ1 u1 , (1 − α)ρ2 u2 )T . In the right-
hand side, the Mk ’s are some source terms, satisfying M1 + M2 = 0, which
we assume do not contain derivatives of the variables: they model interfacial
forces (such as interface drag which is proportional to the relative velocity
ur = u1 − u2 ) or external forces (such as gravity). When Δp = 0, we get the
so-called one-pressure model. Note that the second equation in (3.5) can also
be written
∂ ∂ ∂
(αk ρk uk ) + (αk (ρk u2k + p)) − pI α k = Mk ,
∂t ∂x ∂x
which clearly resembles the second equation obtained in the preceding exam-
ple modeling the flow in a nozzle, with now αk (x, t) unknown and depending
also on t. System (3.5) is nonconservative, and it can be put in the form
∂ ∂
u + A(u) u = 0.
∂t ∂x
However, the resulting model may fail to be hyperbolic, in the sense that
A(u) does not necessarily have real eigenvalues for all physically acceptable
quantities. In particular, the one pressure model (Δp = 0) is not hyper-
bolic in general (not in the range of relative velocities in common physical
applications), so that, usely, one builds some “pressure correction” term Δp
proportional to u2r , where again ur = u1 − u2 is the relative velocity [896],
or a correction term involving derivatives (see [1130]). This system, though
simple, gives an example where the computation of the eigenvalues, which
are solution of a fourth-degree polynomial, cannot be achieved explicitly, and
the result is obtained by perturbation, using some Taylor expansions.
A related model keeps two different pressures, leading to the so-called five-
equation two-pressure model which writes, ignoring the external forces and
transfer terms between the phases,
26 I. Introduction
⎧
⎪ ∂ ∂
⎪
⎪ (αk ρk ) + (αk ρk uk ) = 0, k = 1, 2,
⎪
⎪ ∂t ∂x
⎪
⎪
⎨
∂ ∂ ∂
(αk ρk uk ) + (αk (ρk u2k + pk )) − pI αk = 0, k = 1, 2, (3.6)
⎪ ∂t
⎪ ∂x ∂x
⎪
⎪
⎪
⎪
⎪
⎩ ∂ α1 + uI ∂ α1 = δ,
∂t ∂x
with equations of state pi = pi (ρi ) and where uI , pI are the interfacial veloc-
ity and pressure which are supposed to be known functions of the state. A
frequent choice is uI = u2 , pI = p1 ; the model is often called the (isentropic)
model of Baer-Nunziato. The eigenvalues are explicitly computed [965], and
the system is hyperbolic (outside resonance). In the last equation, when the
source term δ is taken proportional to the relative pressure p1 − p2 , with a
coefficient which can be stiff, one speaks of a pressure relaxation term. At
least formally, the limit for zero relaxation time forces the equality of pres-
sures p1 = p2 and gives the four-equation two-fluid model (3.5); for the full
seven-equation model with energy, a hyperbolic five-equation reduced model
is derived in [890] via an asymptotic analysis for zero relaxation time (see
also [16]).
An equilibrium mixture model corresponds to the assumption that the two
fluids have equal velocity, and we have only one total momentum equation,
obtained by adding the momentum equations of each phase; it writes in the
barotropic case as a three-equation model
⎧
⎪ ∂ ∂
⎪
⎨ ∂t (αk ρk ) + ∂x (αk ρk u) = 0, k = 1, 2,
(3.7)
⎪
⎪
⎩ ∂ (ρu) + ∂ (ρu2 + p) = 0,
∂t ∂x
where ρ = k=1,2 αk ρk is the total density, and the system is closed with
equations of state for each fluid pk (ρk ) and equating the pressure p = p1 (ρ1 ) =
p2 (ρ2 ). We may equivalently replace one equation of conservation of the mass
of one phase by the sum of the two equations which writes
∂ ∂
ρ+ ρu = 0.
∂t ∂x
The resulting system (3.7) is hyperbolic, and the eigenvalues are explicit [279].
For modeling a phase transition, one can consider a relaxation source term
of the form ± 1ε (g1 − g2 ), gi denoting the chemical potential of phase i, that
can be added in the right-hand side of each equation of conservation of mass
(the sum of the term is zero so as to keep the conservation of total mass).
Then the limit system as ε → 0 is the (barotropic) Euler system (conservation
of total mass and momentum) with a pressure law which takes into account
the transition between the two phases; we refer to [238] (see also [602] for the
system with energy).
4 Weak Solutions of Systems of Conservation Laws 27
It is just an insight into the extreme variety of models for two-phase flow,
which, according to the modeling assumptions, differ by many aspects, in
particular the number of equations, the presence of nonconservative products
[1010], not mentioning the source terms, and also for which the hyperbolic
feature is not an obvious issue (see [346, 364, 693], and also [511] for mul-
tiphase flow modeling via Hamilton’s principle). Though we did not go into
details on the modeling assumptions, because we focus on giving different
examples of systems for which a general numerical approach is possible, it
should be emphasized that a given model is valid only in some flow configu-
ration and may fail to reproduce other configurations (see [965, 1073], for a
recent survey on multi-fluid models [195].
For the sake of simplicity, we now essentially concentrate in the following
on systems of the form (1.1) and occasionally on systems of the form (2.23)
since they cover a large range of physical situations. The treatment of source
terms will be given some consideration later on, in a special chapter.
Let us go back to the Cauchy problem (1.1), (1.2) for a general system of
conservation laws. We shall say that a function u : Rd × [0, ∞[−→ Ω is a clas-
sical solution of (1.1), (1.2) if u is a C 1 function that satisfies Eqs. (1.1), (1.2)
pointwise.
An essential feature of this problem is that there do not exist in general
classical solutions of (1.1), (1.2) beyond some finite time interval, even when
the initial condition u0 is a very smooth function. Let us illustrate this fact
by studying the simple case of a one-dimensional scalar equation.
and let us introduce the characteristic curves associated with (4.1). They are
defined as the integral curves of the differential equation
dx
= a(u(x(t), t)). (4.4)
dt
Proposition 4.1
Assume that u is a smooth solution of (4.1). The characteristic curves (4.4)
are straight lines along which u is constant.
Proof. Consider a characteristic curve passing through the point (x0 , 0), i.e.,
a solution of the ordinary differential system
⎧
⎨ dx = a(u(x(t), t))
dt
⎩
x(0) = x0 .
Then, the characteristics C1 and C2 drawn from the points (x1 , 0) and (x2 , 0),
respectively, have slopes m1 and m2 and intersect necessarily at some point P .
At this point P , the solution u should take both values u0 (x1 ) and u0 (x2 ),
which is clearly impossible. Hence, the solution u cannot be continuous at the
point P . Note that this phenomenon is independent of the smoothness of the
4 Weak Solutions of Systems of Conservation Laws 29
t=1
x
0 1
u
u(., 1)
u(., 0)
x
0
functions u0 and f . Indeed, using (4.5), we see that the two characteristics
intersect at time t if
Thus, unless the function x → a(u0 (x)) is monotone increasing, in which case
this equation has no positive solution t, we cannot define a classical solution
u for all time t > 0 (see Fig. 4.1). Moreover, one can determine the critical
time T ∗ up to which a classical solution exists and can be constructed by the
method of characteristics; T ∗ is given by
1 d
T∗ = − , α = min a(u0 (y)).
min(α, 0) y∈R dy
By using the method of characteristics, we can solve up to the time when the
characteristics intersect. We already know by (4.5) that the characteristic
passing through the point (x0 , 0) is given by
so that ⎧
⎪
⎪ x0 + t, x0 ≤ 0,
⎨
x(x0 , t) = x0 + t(1 − x0 ), 0 ≤ x0 ≤ 1,
⎪
⎪
⎩
x0 , x0 ≥ 1.
For t < 1, the characteristics do not intersect (see Fig. 4.1). Hence, given a
point (x, t) with t < 1, we draw the (backward) characteristic passing through
this point, and we determine the corresponding point x0
⎧
⎪ x − t, if x ≤ t < 1,
⎪
⎪
⎨
x−t
x0 = , if t ≤ x ≤ 1,
⎪
⎪ 1−t
⎪
⎩
x, if x ≥ 1.
∞ ∂u d
∂
− + fj (u) · ϕ dx dt
0 Rd ∂t j=1
∂xj
∞ ∂ϕ
d
∂ϕ
= u· + fj (u) · dx dt + u(x, 0) · ϕ(x, 0) dx,
0 Rd ∂t j=1
∂xj Rd
where the dot · denotes the Euclidean inner product on Rp . Thus, any classical
solution u of (1.1), (1.2) satisfies ∀ϕ ∈ C10 (Rd × [0, +∞[)p
∞ ∂ϕ
d
∂ϕ
u· + fj (u) · dx dt + u0 (x) · ϕ(x, 0) dx = 0. (4.7)
0 Rd ∂t j=1 ∂xj Rd
a weak solution of the Cauchy problem (1.1), (1.2) if u(x, t) ∈ Ω a.e. and
satisfies (4.7) for any function ϕ ∈ C10 (Rd × [0, +∞[)p .
By construction, a classical solution of problem (1.1), (1.2) is also a weak
solution. Conversely, by choosing ϕ in C∞ d p ∞
0 (R ×]0, ∞[) , where C0 (R ×]0,
d
∞
+∞[) is the space of C functions with compact support in R ×]0, ∞[, we
d
obtain that any weak solution u satisfies (1.1) in the sense of distributions on
Rd ×]0, ∞[. Moreover, if u happens to be a C 1 function, then it is a classical
solution. Indeed, let ϕ ∈ C10 (Rd ×]0, +∞[)p ; integrating (4.7) by parts gives
∞ ∂u d
∂
+ fj (u) · ϕ dx dt = 0,
0 Rd ∂t j=1
∂xj
so that (1.1) holds pointwise.
Next, if we multiply (1.1) by a test function ϕ ∈ C10 (Rd × [0, +∞[)p ,
integrate by parts, and compare with (4.7), we obtain
d
(u+ − u− )nt + (fj (u+ ) − fj (u− )) nxj = 0 (4.8)
j=1
∂ϕ
d
∂ϕ
0= u· + fj (u) · dx dt = + .
D ∂t j=1
∂xj D+ D−
Another random document with
no related content on Scribd:
contentedly in the thought that Heaven had given me, if only for one
little hour, a lover so loyal, true and brave. Ludovic, my love, my poor
starved heart thanks God for you.”
For an instant the word was at his lips which would have told her his
secret, for, surely, the opportunity was apt. Perhaps it was a feeling
that, in a higher sense, in that atmosphere so fully charged with
tenderness and love, the cold shock of the announcement would be
unfitting; perhaps, too, his sensitive, innate chivalry made him shrink
from taking advantage of that supreme moment. The very certainty
that the stroke must win held him back from making it. Anyhow it
passed, and when rapture allowed him speech it was of a still more
urgent matter, their escape. She told him it was for that she had
risked the message.
“The Baron does not say so, but I know I am destined for Krell. And
once there,” she shuddered, “I may say farewell to my hopes and to
my liberty, except on terms which are now forever impossible.”
He understood, and signified it by a kiss.
“There is no reason, I hope,” he said, “why we should not push on
again for Beroldstein. The longest and worst part of the way has
been travelled, and the end of our journey is now not so far off. With
a couple of hours’ start we could laugh at pursuit, and need not fear
the high roads to-night.”
“Then let us go, dearest,” she urged.
He smiled at the eagerness he loved. “Everything is arranged,” he
replied. “Ompertz is waiting with horses, and will ride with us. I fear,
though, we must leave Countess Minna behind this time. But she is
now safe from this fellow.”
A look of disappointment clouded Ruperta’s face. “Rollmar will visit
my sins on poor Minna’s head.”
“Her penance shall be of short duration, I promise you that,” Ludovic
assured her confidently. “She shall join you in a very few days.
Rollmar is too sensible to take a foolish and futile revenge. Indeed, it
is best; more, it is necessary. We have no horse for her.”
“And Minna hates riding, if you had. Well then, we must leave her. It
is easier now,” she added, with a loving look of confidence.
In a very few minutes preparations for the escape and the journey
were made. Ludovic extinguished the light, and, cautiously opening
the door, crept out, leading the way along the narrow passage, and
down the winding stairs, descending to the outer door by which his
guide had admitted him to the castle. No one was to be seen; the
door was unlocked; they passed out, and crossed an angular court-
yard to a massive stone door set in the outer wall. This, as Ludovic’s
conductor had shown him, was left merely bolted on the inside; at a
strong pull it swung slowly open, and they found themselves in a
passage cut through the rock and leading out into the wood.
Ludovic put his arm round Ruperta to help her along the rough path.
“Now for our faithful Ompertz and the horses,” he said
encouragingly. “He is near at hand. Another hour, dearest, will see
us miles away from this hateful place.”
They were now at the end of the cutting. It was with a delicious
sense of freshness and liberty that Ruperta felt the wind through the
trees blowing on her face. Her lover’s strong arm was round her—in
a few minutes the enemies of her happiness were to be given the
slip. There was just light enough to see the path; a stronger blast of
wind came through the wood, deadening the sound of another rush.
More quickly than they could realize it, they were surrounded by half
a dozen men who had suddenly sprung from their ambush. Before
Ludovic could put his hand to a weapon, he was seized by four
strong fellows, who held his arms firmly, and began to drag him back
to the castle. Ruperta, with all her spirit, was powerless to render him
any help. She herself had been captured by two men who, with less
violence, but equally insistent force, kept her from following.
But the dashing of her hopes, the sickening sense of the Count’s
treachery, made her desperate and reckless. She struggled furiously
with her captors, two tall, evil-looking ruffians who had, however,
evidently had orders to treat her with as much respect as their object
permitted. This was to take her back to the castle by another
entrance; but they found it not so easy. Ruperta resisted vigorously,
then, remembering that Ompertz might be near, she began calling
for help. It was but a faint hope, but, to her joy, she heard an
answering call which was followed by the welcome appearance of
the great dashing swashbuckler, who came through the wood with a
leap and uplifted sword, a very fury to the rescue.
Evidently the men thought so, for it was with no very confident air
that one of them released his hold on Ruperta, and, drawing his
sword, stood before her to keep Ompertz off. A dog might as well
have tried the repel the spring of an attacking lion. With a mighty
sweep his sword was sent flying among the trees, and it was only by
a smart backward spring that he cheated the soldier’s blade of its
second blow.
At the same moment Ruperta found herself free, her other captor
thinking less of his charge than of his skin, which was, indeed, just
then in jeopardy of damage. She quickly told her rescuer what had
happened. He just checked an oath of angry disappointment.
“I told him what to expect,” he said, savagely rueful. “But we both
hoped I might prove a false prophet. Oh,”—he set his teeth
ominously—“oh, for five minutes alone with this precious Count! He
should never tell another lie while I lived, or he.”
Ruperta entreated him to follow her lover and free him. He felt the
urgency of the move, yet hesitated.
“I dare not leave you, Princess, and if we go together”—he gave a
shrug—“I am only one to defend you against this gang of bandits. It
were better to see you into safety first.”
But she would not hear of abandoning Ludovic while there was a
chance of rescue. She too would go back; she had no fear.
Ompertz saw the true courage in her eyes, and no longer opposed
her wish. The two men had skulked away; they were scarcely worth
consideration now. The soldier gave his hand to Ruperta, and, sword
in the other, led her quickly along the passage to the stone door. It
was closed and fast bolted; the men had clearly taken their prisoner
through, and now had him safely lodged. Ompertz gave a kick at the
unyielding barrier.
“No hope of opening that fellow from outside,” he remarked, with a
baffled shake of the head. “And, Highness, let me tell you the sooner
for your sake we get out of this ugly trap the better. We should not
have a chance if these rascals took it into their heads to drop a few
lumps of rock down on ours.”
Although Ruperta had little fear of that awkward contingency, she
recognized the futility of staying there. Her heart was full of
indignation and a terrible anxiety for her lover. But hers was a nature
which rage and fear simply stirred into action; she would never bow
to the inevitable or confess herself beaten.
“Yes. Come back with me quickly,” she said, with sudden resolution.
Ompertz glanced at her and knew that the move was not prompted
by fear, at least for herself. They hurried back along the passage of
rock and into the wood.
“The horses are close by,” Ompertz said, in a tone of doubtful
suggestion.
“That is well; we may want them,” Ruperta replied, and he saw that
she had in her mind a plan of action.
“The Chancellor brought men—soldiers—with him? How many?”
“About eighty.”
“They are near?”
“Hard by, in the forest beyond the valley.”
“That is well,” she said. “I can trust myself to them. I am their
Princess. It is only their leaders who are so vilely treacherous.”
Ompertz looked a little dubious. “If they were all like me, Princess,
you might trust them to the death.”
“And you think I cannot rely upon them to protect me against the
false hearts and lying tongues of the cowards who threaten us? At
least I will try them.”
There was a rustling in the wood, and Count Irromar stood before
them.
CHAPTER XXVII
AN UNWISE MERCY