Professional Documents
Culture Documents
Strongly Coupled Parabolic and Elliptic Systems Existence and Regularity of Strong and Weak Solutions 1st Edition Dung Le
Strongly Coupled Parabolic and Elliptic Systems Existence and Regularity of Strong and Weak Solutions 1st Edition Dung Le
https://ebookmeta.com/product/optimal-regularity-and-the-free-
boundary-in-the-parabolic-signorini-problem-1st-edition-
donatella-danielli/
https://ebookmeta.com/product/numerical-methods-for-elliptic-and-
parabolic-partial-differential-equations-2nd-edition-peter-
knabner/
https://ebookmeta.com/product/beyond-sobolev-and-besov-
regularity-of-solutions-of-pdes-and-their-traces-in-function-
spaces-cornelia-schneider/
https://ebookmeta.com/product/geometric-properties-for-parabolic-
and-elliptic-pde-s-vincenzo-ferone-tatsuki-kawakami-paolo-salani-
futoshi-takahashi-eds/
Elliptic Extensions in Statistical and Stochastic
Systems 1st Edition Makoto Katori
https://ebookmeta.com/product/elliptic-extensions-in-statistical-
and-stochastic-systems-1st-edition-makoto-katori/
https://ebookmeta.com/product/elliptic-regularity-theory-by-
approximation-methods-london-mathematical-society-lecture-note-
series-edgard-a-pimentel/
https://ebookmeta.com/product/china-east-asia-and-the-european-
union-strong-economics-weak-politics-1st-edition-tjalling-
halbertsma-jan-van-der-harst/
https://ebookmeta.com/product/dynamically-coupled-rigid-body-
fluid-flow-systems-1st-edition-banavara-n-shashikanth-2/
https://ebookmeta.com/product/dynamically-coupled-rigid-body-
fluid-flow-systems-1st-edition-banavara-n-shashikanth/
Dung Le
Strongly Coupled Parabolic and Elliptic Systems
De Gruyter Series in Nonlinear
Analysis and Applications
|
Editor in Chief
Jürgen Appell, Würzburg, Germany
Editors
Catherine Bandle, Basel, Switzerland
Alain Bensoussan, Richardson, Texas, USA
Avner Friedman, Columbus, Ohio, USA
Mikio Kato, Nagano, Japan
Wojciech Kryszewski, Torun, Poland
Umberto Mosco, Worcester, Massachusetts, USA
Louis Nirenberg, New York, USA
Simeon Reich, Haifa, Israel
Alfonso Vignoli, Rome, Italy
Volume 28
Dung Le
Strongly Coupled
Parabolic and
Elliptic Systems
|
Existence and Regularity of Strong and Weak Solutions
Mathematics Subject Classification 2010
Primary: 35K59, 35J62, 35D35; Secondary: 35B65, 74G25
Author
Prof. Dr. Dung Le
University of Texas at San Antonio
Department of Mathematics
One UTSA Circle
San Antonio, TX 78249
USA
ISBN 978-3-11-060715-4
e-ISBN (PDF) 978-3-11-060876-2
e-ISBN (EPUB) 978-3-11-060717-8
ISSN 0941-813X
www.degruyter.com
|
To my beloveds Thu Nguyen, Uyen Le and Andrew Le.
Preface
This book aims at researchers who are interested in reaction diffusion systems with
couplings occurring in both reaction and diffusion parts. The main theme of the book
is to demonstrate the crucial role of a priori estimates of the BMO norm of solutions in
the local and global existence theory of strong/weak solutions. Several examples are
included and show the applicability of the theory to cross diffusion models in math-
ematical biology/ecology. The content of this book grows out of our published works
and extends well beyond the scope of those by new added findings.
I extend my appreciation and thanks to the production staff at De Gruyter who
managed to put this book together. They have been so effective and professional.
https://doi.org/10.1515/9783110608762-201
Contents
Preface | VII
1 Introduction | 1
Bibliography | 183
Index | 185
1 Introduction
Since the beginning of the twentieth century, scalar reaction diffusion equations, with
constant and then variable coefficients, have been proposed both by applied scientists
and pure mathematicians to study physical phenomena. It was soon observed that
most of the equations modeling physical phenomena without excessive simplification
are nonlinear. They were extensively studied with great success in models involving
one unknown function u describing pressure, temperature, density of species. . . The
equation takes the general form
https://doi.org/10.1515/9783110608762-001
2 | 1 Introduction
The interaction among the components of (1.0.2) occurs only in the reaction terms
as one assumes that the diffusion (or movement) of each u i is not sensitive to those
of the others. This of course limits the applications of the theory as scientists already
raised such questions in the early 50’s of the last century (e.g., see [44]). Taking this
into consideration, because the movement of a species is reflected by its gradient, one
should replace (1.0.2) by
m
(u i )t = div( ∑ a i,j (u)Du j ) + f i (u, Du) , (x, t) ∈ Ω × (0, T0 ) . (1.0.3)
j=1
Introducing the m × m matrix A(u) = [a ij (u)] and the vector f ̂(u, Du) = [f i (u,
Du)]m
i=1 , we can write (1.0.1)–(1.0.3) in the vectorial form
It is clear that the scalar (1.0.1) is included in the above when m = 1, and (1.0.2) is also
a special case of (1.0.4) if A(u) is a diagonal matrix diag[a i (u)]. We refer to (1.0.4) in
this case as a weakly coupled system and a strongly coupled one if A(u) is a full matrix.
The strongly coupled system (1.0.4) appears in many physical applications, for
instance, Maxwell–Stephan systems describing the diffusive transport of multicom-
ponent mixtures, models in reaction and diffusion in electrolysis, flows in porous me-
dia, diffusion of polymers, or population dynamics, among others. A popular model
in mathematical biology, that has been widely investigated together with its simpli-
fied variances in the last few decades, is the Shigesada–Kawasaki–Teramoto system
(see [44]), which consists of two reaction-diffusion equations with variable cross-dif-
fusion and quadratic nonlinearities.
Another important issue is the global existence problem when the local one is es-
tablished: will there be a unique solution solving (1.0.4) for all T0 > 0? One can choose
to work with either weak or strong solutions. In the first case, the local existence of a
weak solution can be achieved via Galerkin, time discretization or variational methods
but its regularity (e.g., boundedness, Hölder continuity of the solution and its higher
derivatives) is then an open and serious issue. Several works have been done along
this line to improve the early work [14] of Giaquinta and Struwe and establish partial
regularity of bounded weak solutions to (1.0.4). Uniqueness of weak solutions for the
Cauchy problem is still an open question so that the meaning of global existence of a
weak solution is vaguely defined in literature.
Otherwise, if strong solutions are considered then their existence can be estab-
lished via semigroup theories as in the works of Amann [1, 2]. Combining with inter-
polation theories of Sobolev spaces, Amann established local and global existence of
a strong solution u of (1.0.4) under the assumption that one can control ‖u‖W 1,p (Ω,ℝm )
for some p > n. His theory did not seem to apply to the case where f ̂ has superlinear
growth in Du. In addition, the estimate of W 1,p norm when p > n is comparable to
that of Hölder norms, which is very hard to achieve.
It is always assumed that the matrix A(u) is elliptic in the sense that there exist
two scalar positive continuous functions λ1 (u), λ2 (u) such that
If there exist positive constants c1 , c2 such that c1 ≤ λ1 (u) and λ2 (u) ≤ c2 then
we say that A(u), or the corresponding system, is regular elliptic. If 0 < λ1 (u) and
λ2 (u)/λ1 (u) ≤ c2 , we say that A(u) is uniform elliptic. In addition, if λ1 (u) tends to
zero (respectively, ∞) when |u| → ∞ then we say that A(u) is singular (respectively,
degenerate) at infinity.
In both forementioned approaches, the assumption on the boundedness of u must
be the starting point because the techniques rely heavily on the fact that A(u) is reg-
ular elliptic in the above terminology. As we mentioned earlier, for strongly coupled
systems like (1.0.4) invariant/maximum principles are generally unavailable so the
boundedness of the solutions is already a hard problem and perhaps a very restric-
tive hypothesis. One usually needs to use ad hoc techniques on the case by case ba-
sis to show that u is bounded (see [23, 41]). Even so, for bounded weak solutions we
know that they are only Hölder continuous almost everywhere (see [14]). In addition
(see [21]), there are counter examples for systems (m > 1) which exhibit solutions that
start smoothly and remain bounded but develop singularities in higher norms in finite
times.
In this book, we will work with strong solutions and present a rather simple and
unified approach making use of fixed point theories to obtain the local and then
global existence of strong solutions of uniform elliptic (1.0.4), including the degener-
ate/singular cases. The framework of fixed point methods, the Leray–Schauder fixed
point theorem in particular, is fairly standard and consists of two steps. The first step
4 | 1 Introduction
is fairly standard; strong solutions of the system can be viewed as fixed points of a
compact map T on a suitable Banach space X. This map is embedded in a continu-
ous family of related compact maps T σ , σ ∈ [0, 1], on X. The second step is to show
that there is a uniform bound for the X norms of fixed points of T σ , and this is the
hardest and most crucial ingredient of the argument. This step is equivalent to the
establishment of some apriori bounds for higher order norm of strong solutions to
(1.0.4). We achieve this by combining a new local Gagliardo Nirenberg inequality and
a novel finite iteration argument. Our method then provides the existence results un-
der the weakest (and necessary) assumption that some suitable transformation of the
strong solution apriori have small BMO norms in small balls. This assumption and the
techniques set the sharp difference between this work and others known in literature.
The presentation here is an extensive generalization of our recent works [29, 30]
which established the local and global existence results for regularly uniform elliptic
(1.0.4) under the assumption that their strong solutions are a priori VMO (Vanishing
Mean Oscillation), but not necessarily bounded, and general structural conditions on
the data of (1.0.4) which are independent of x. We will allow A depend on x in this
work and assume only that A(x, u) is uniform elliptic.
Applications were presented in [30] when λ(u) has a polynomial growth in |u| and,
without the boundedness assumption on the solutions, so (3.0.1) can be degenerate as
λ(u) → ∞ when |u| → ∞. The singular case, λ(u) → 0 as |u| → ∞, was not discussed
there. Also, the reaction term f ̂ was assumed to have linear growth only in gradient
described in f.1) of F).
Most importantly, we will present here a very versatile VMO assumption on u. If
the system is singular, one may not see immediately or prove that u is VMO. However, it
is possible that K(u) is VMO for some suitable and carefully chosen map K : ℝm → ℝm .
In Chapter 2 we will establish one of the key ingredients of the proof in this book:
the local weighted Gagliardo–Nirenberg inequalities involving BMO norm with gen-
eral measure. These new inequalities extensively generalize those initially reported
in [29] and allow us to replace the VMO assumption on a strong solution u of (1.0.4)
there by a more versatile one in order to obtain the key apriori estimates: K(u) is VMO
for some general map K. Inequalities of this type, of course, have their own inter-
ests in the theory of functional spaces and they beautifully serve our main purpose
in this book. The general inequalities allow us to deal with singular uniform elliptic
systems with suitable and careful choice of the map K. We also show that the inequal-
ities in [29, 30] are just special (and simple) cases of those discussed here.
Chapter 3 is the longest one and the heart of this book. It provides the core frame-
work for the existence theory in the following chapters. The main existence result The-
orem 3.2.3 of the chapter relies on the two main conditions M.0) and M.1). We discuss
the solvability of (1.0.4) and show that the assumption on the smallness of BMO norm
of K(u) for any strong solution u of (1.0.4) for some suitable map K (see the condition
M.1) is sufficient (and also necessary) for the existence of strong solutions. We will also
consider very mild integrability conditions in M.0) but no boundedness assumption
1 Introduction | 5
of solutions will be presumed The setting of fixed point argument will be discussed
in Section 4.2.1. The required a priori estimates for possible strong solutions will be
proved in details in Section 4.2.2. We will establish a general energy estimate for the
gradient Du for any strong solution u and then use the Gagliardo Nirenberg inequali-
ties in the previous chapter to obtain a better estimate which allows us to perform an
iteration argument in the proof of our main result.
In Section 4.2.2, the reader will see the advantage of our approach in choosing
to work with strong solutions throughout this book. Thanks to the differentiability of
these solutions we can differentiate the system in order to derive a strong energy esti-
mate for Du. This is not possible if one chooses to deal with weak solutions. Also, we
should mention that the Gagliardo–Nirenberg inequalities, which involve with second
derivatives of the solutions, are not available for weak solutions.
On the other hand, we will have to assume the spectral gap condition SG) in addi-
tion to the structural conditions on the uniform (1.0.4). Roughly speaking, this requires
that the ratio between the ellipticity ‘constants’ of the matrice A(u) is not too big. We
only need this assumption in order to obtain the energy estimate. Otherwise, there are
counterexamples showing that strong solutions can blow up (or cease to exist) in fi-
nite time (see [31]). This also partly explains why Moser’s iteration technique for scalar
equations cannot be extended to the vectorial case.
The existence result of (1.0.4) in Theorem 3.2.3 is established under the most gen-
eral assumption that there is a suitable map K such that K(u) is VMO so that some
version of the local Gagliardo–Nirenberg inequalities in Chapter 2 is usable for our
purpose. However, the construction of such map is quite delicate as we present by ex-
amples in Section 3.5. Under slightly stronger structural assumptions, we will show in
Section 3.4 another energy estimate for Du and that will allow us to take K to be the
identity map, and the condition M.1) will then be considerably simplified.
We conclude Chapter 3 by discussing several examples where the main theorems
are applicable in Section 5.1. Motivated by the SKT system (1.0.5), we will introduce the
following generalized SKT model
This system can be thought of as the stationary (1.0.4) whose solutions are time in-
dependent. The proof of the main existence results in this chapter principally follows
6 | 1 Introduction
those of Chapter 3 and is somewhat simpler, of course due to the absence of the tempo-
ral derivative. Nevertheless, there are significant differences in the settings of the fixed
point argument and a priori estimates in the two cases; and we present the details to
keep this chapter as independent as possible.
In Section 4.4, we discuss several examples concerning the solvability of (1.0.8).
The absence of the temporal derivative u t allows us to consider examples with much
more general structural conditions and they don’t have to be the stationary ones as-
sociated to the generalized (SKT) system (1.0.7) considered in Section 5.1. In addition,
the spectral gap condition SG) for the parabolic case, which was void if N ≤ 2, is now
not needed if N ≤ 4.
We restrain ourself from discussing the elliptic counterpart of generalized (SKT)
system
−∆(P(u)) = f ̂(u, Du)
in this chapter because this topic will be investigated at great length in the next chap-
ter. In Chapter 5, we will study the parabolic/elliptic generalized (SKT) systems and
present several results extending known existence results for the traditional (SKT) un-
der a very weak assumption that some a priori integrability of solutions are given.
Concerning solutions to the generalized evolution SKT systems, we will discuss the
existence of weak solutions and their uniqueness and VMO regularity when the self
diffusivities are all zero, that is the system can be singular in Chapter 5. This model was
inspired by the combination between the SKT systems and porous media type scalar
equations which have been actively investigated in the last few decades, starting with
the work by Friedman, Brezis and Crandall (see [6, 12] and the references therein). To
the best of our knowledge, this is the first time such a singular and degenerate settings
are combined in one model and treated in a unified framework as in this book. The
regularity theory of weak solutions for these systems is still a wildly open problem
and the best we can do so far is to prove that they are VMO, at least when N = 2.
The elliptic counterpart of the generalized evolution SKT systems was also studied
in the chapter. Thanks to the special form of the diffusion part, we present another sim-
pler way to obtain L p (Ω) norm estimates for Du so that the theory and assumptions in
Chapter 4 can be greatly simplified and independent from the discussion here. In fact,
we can drop the spectral gap condition SG) and ESG) in the preceding Chapter 3 and
Chapter 4 so that the theory here applies to all dimension N to establish the existence
of a weak solution. Moreover, the regularity of this weak solution is greatly improved:
it is in fact Hölder continuous.
Chapter 6 is devoted to the study of solutions to the general elliptic system (1.0.4).
In recent years, much effort has been made to the investigation of steady states, be-
sides the obvious zero solution, of the (SKT) system (1.0.5) by several ad hoc methods.
The existence of a stronng solution to (1.0.4) is already established in Chapter 4 but, as
we see from the example of (SKT) sytem, this solution may be the trivial or other semi
trivial solutions which can be observed by simple inspections and thus the result in
1 Introduction | 7
Chapter 4 is not interesting in this case. We then face with a new question of finding
nontrivial solutions besides these semi trivial ones. This will be the main topic of the
chapter.
We will present a systematic approach to the problem by using fixed point index
theory to obtain necessary conditions for such nontrivial solutions to exist. The con-
tent of the chapter heavily relies on that of our recent work [30] and we add some new
clarifications as well as developments in applications. In particular, we would like to
draw the reader attention to the end of the chapter where, in connection with Chap-
ter 5, we will briefly discuss the existence of nontrivial solutions to the singular (SKT)
systems.
Finally, we present some necessary facts, which may be elementary to the experts,
in the appendix. These facts are in fact fairly important in our proof but briefly men-
tioned in order to keep the flow of our main ideas in the proof.
Basic notations:
Concerning the basic notations throughout this book, we mention some. Ω is a
bounded domain with smooth boundary ∂Ω in ℝN for some integer N ≥ 1. In dealing
with parabolic problem like (1.0.4) we work with vector valued functions (or matrix-
valued) function
The temporal and k-order spatial derivatives of u are denoted by u t and D k u respec-
tively. The functional spaces, which u and the data of the system belong to, are the
standard ones in literature.
For an integer k we shall denote by C k (Ω) the space of functions having contin-
uous derivatives up to and including the order k; and with C∞ (Ω) the space of in-
̄ consists of functions in C k (Ω),
finitely differentiable functions in Ω. The space C k (Ω)
whose derivatives up to order k can be continuously extended to the boundary ∂Ω.
For α ∈ (0, 1) we will denote by C k,α (Ω) the subspace of C k (Ω) consisting of functions
whose k-order derivatives are Hölder continuous with exponent α.
As usual, W 1,p (Ω, ℝm ), p ≥ 1, will denote the standard Sobolev spaces whose
elements are vector valued functions u : , Ω → ℝm with finite norm
Following [25], for some q, r ≥ 1 and T0 > 0 we denote by Vq,r (Q, ℝm ) the Banach
space of vector valued functions on Q = Ω × (0, T0 ) with finite norm
where 1
r
T0 q
r
q
‖v‖q,r,Q := (∫ (∫ |v(x, t)| dx) dt) .
0 Ω
8 | 1 Introduction
If the target space ℝm is clear from the context we will omit it from these notations
and simply write them by W 1,p (Ω) and Vq,r (Q).
In our statements and proofs, we use C, C1 , . . . to denote various constants which
can change from line to line but depend only on the parameters of the hypotheses in
an obvious way. We will write C(a, b, . . .) when the dependence of a constant C on its
parameters is needed to emphasize that C is bounded in terms of its parameters. We
also write a ≲ b if there is a generic constant C, which is clear from the context, such
that a ≤ Cb. In the same way, a ∼ b means a ≲ b and b ≲ a.
2 Interpolation Gagliardo–Nirenberg inequalities
In this chapter we will discuss one of the key ingredients of the proof of our existence
theory in this book: the local weighted Gagliardo–Nirenberg inequalities involving
BMO norm with general measure.
In [42, 43], for any p ≥ 1 and C2 scalar function u on ℝN , N ≥ 2, global and local
Gagliardo–Nirenberg inequalities of the form
{
{ u t = div(A(u, Du)) + f ̂(u, Du) (x, t) ∈ Q = Ω × (0, T0 ) ,
{
{ u(x, 0) = U0 (x) x∈Ω (2.0.2)
{
{
{u = 0 on ∂Ω × (0, T0 ) .
We will prove new general global and local versions of (2.0.1) and the results in [29]
in many aspects. Roughly speaking, we will establish inequalities of the following
type: for any p ≥ 1 and any C2 map U : Ω → ℝm ,
https://doi.org/10.1515/9783110608762-002
10 | 2 Interpolation Gagliardo–Nirenberg inequalities
Let μ be a measure on Ω. For any μ-measurable subset A of Ω and any locally μ-inte-
grable function U : Ω → ℝm we denote by μ(A) the measure of A and U A the average
of U over A. That is,
1
U A = ∫ U(x) dμ = ∫ U(x) dμ .
A μ(A) A
We now recall some well-known notions from harmonic analysis.
A function f ∈ L1 (μ) is said to be in BMO(μ) if
We then define
‖f‖BMO(μ) := [f]∗,μ + ‖f‖L1 (μ) .
For γ ∈ (1, ∞) we say that a nonnegative locally integrable function w belongs to
the class A γ or w is an A γ weight if the quantity
γ−1
[w]γ := sup (∫ w dμ) (∫ w1−γ dμ) is finite . (2.1.2)
B⊂Ω B B
Here, γ = γ/(γ − 1) and the supremum is taken over all cubes B in Ω. For more details
on these classes we refer the reader to [11, 40, 45].
We say that Ω and μ support a Poincaré–Sobolev inequality if the following holds:
PS) There are σ ∈ (0, 1) and τ ∗ ≥ 1 such that for some q > 2 and q∗ = σq < 2 we have
1 1
1 q q∗
(∫ |u − u B |q dμ) ≤ C PS (∫ |Du|q∗ dμ) (2.1.3)
l(B) B τ∗ B
2.1 Hypotheses on the measures | 11
for some constant C PS and any cube B with side length l(B) and any function u ∈
C1 (B).
Here and throughout this chapter, we write B R (x) for a cube centered at x with side
length R and sides parallel to the standard axes of ℝN . We will omit x in the notation
B R (x) if no ambiguity can arise. We denote by l(B) the side length of B and by τB the
cube that is concentric with B and has side length τl(B).
We have the following remark on the validity of the assumption PS).
Remark 2.1.1. Suppose that μ is doubling and supports a q∗ -Poincaré inequality [19,
eqn. (5)]: there are some constants C P , q∗ ∈ [1, 2] and τ ∗ ≥ 1 the inequality
1
q∗
∫ |h − h B | dμ ≤ C P l(B) ( ∫ |Dh|q∗ dμ) (2.1.4)
B τ∗ B
holds true for any cube B with side length l(B) and any function h ∈ C1 (B).
Assume also that for some s > 0 μ satisfies the following inequality:
r s μ(B r (x))
( ) ≲ , (2.1.5)
r0 μ(B r0 (x0 ))
where B r (x), B r0 (x0 ) are any cubes with x ∈ B r0 (x0 ). If q∗ = 2 then [19, Section 3]
shows that a q∗ -Poincaré inequality also holds for some q∗ < 2. Thus, we can assume
that q∗ ∈ (1, 2).
If q∗ < s then [19, 1) of Theorem 5.1] establishes (2.1.3) for q = sq∗ /(s − q∗ ). Thus,
q > 2 if s < 2q∗ /(2 − q∗ ). This is the case if we choose q∗ < 2 and close to 2. If s = q∗ ,
[19, 2) of Theorem 5.1] shows that (2.1.3) holds true for any q > 1. On the other hand,
if q∗ > s then [19, 3) of Theorem 5.1] gives a stronger version of (2.1.3) for q = ∞. In
particular, the Hölder norm of u is bounded in terms of ‖Du‖L q∗ (μ) .
Thus, in order to justify PS), we can assume that Ω, μ support a q∗ -Poincaré in-
equality for some q∗ ∈ (1, 2) and (2.1.5) is valid for some s > 0.
Remark 2.1.2. A typical choice of ω0 that satisfies the Hardy type inequality (2.1.7)
γ
is ω0 (x) = d Ω2 (x) for some constant γ. We now recall the following Hardy inequality
proved by Necas (see also the paper by Lehrbäck [33] for much more general versions):
γ−q γ
∫ |u(x)|q d Ω (x) dx ≤ C∫ |Du(x)|q d Ω (x) dx , γ <q−1. (2.1.8)
Ω Ω
γ−2
Because |Dω0 | ∼ d Ω2 (x), (2.1.8), with q = 2, shows that (2.1.7) holds true for γ < 1.
γ
We will also check the condition (2.1.6) of M) for dμ = ωdx ∼ d Ω dx. If B r is far
away from ∂Ω, we have μ(B r ) ≲ r N . Near the boundary, as ∂Ω is C1 , we easily see that
μ(B r ) ≲ r N+γ . If N + γ ≥ n and r is bounded then μ(B r ) ≤ Cr n . This is the case because
Ω is bounded. Thus, for any γ > −N, we define n = min{N, N + γ} ∈ (0, N] to see that
μ(B r ) ≤ Cr n for some constant C, which is bounded in terms of diam(Ω).
In this section we will state our main global and local weighted Gagliardo–Nirenberg
interpolation inequalities. The local version and its consequences will be one of our
main vehicles for the solvability of strong solutions in this book.
Throughout this section we will always use the following notations and hypothe-
ses.
P.1) Let K : dom(K) → ℝm be a C1 map on a domain dom(K) ⊂ ℝm such that K U (U)−1
exists for U ∈ dom(K). Assume that the map 𝕂(U) := (K U (U)−1 )T is differentiable.
Let Φ, Λ : dom(K) → ℝ+ be C1 positive functions. Assume that
W) Let
W(x) := Λ p+1 (U(x))Φ−p (U(x)) for x ∈ Ω . (2.2.3)
Assume that [W α ] β+1 is finite for some α > 2/(p + 2) and β < p/(p + 2).
and furthermore
Theorem 2.2.1. Assume M), P.1)–P.2) and W). Suppose that the integrals in (2.2.4)–
(2.2.7) are finite. Then there are constants C, C([W α ]β+1 ) for which
ω∗ ≡ 1 in Ω∗ and ω∗ ≤ 1 in Ω . (2.2.9)
L.1) ω0 ∈ C1 (Ω) and for dμ = ω20 dx and some n ∈ (0, N] we have μ(B r ) ≤ Cr n .
L.2) The measure ω20 dx supports the Poincaré–Sobolev inequality (2.1.3) in PS). In
addition, ω0 also supports a Hardy type inequality: For any function u ∈ C10 (B)
Theorem 2.2.2. Suppose L.0)–L.2) and P.1). Let U : Ω̄ → ℝm be a C2 map. For any ω1 ∈
L1 (Ω) and ω1 ∼ ω20 we define dμ = ω1 dx and accordingly recall the definitions (2.2.4)–
(2.2.7) and introduce
Then, for any ε > 0 there are constants C, C([W α ]β+1 ) such that
I1,∗ ≤ εI1 + ε−1 CC2∗ [I2 + I1̄ + C([W α ]β+1 )[I2 + I1̂ + I1̄ + I0,∗
̆ ]. (2.2.13)
The proof of these two theorems will be postponed to Section 2.4 and Section 2.5 at the
end of this chapter. We will move on and discuss various special versions of the above
two theorems in the next section.
14 | 2 Interpolation Gagliardo–Nirenberg inequalities
Our first consequence of Theorem 2.2.1 is the case where Φ = Λ and K is the identity
map. We then have the following main inequality in [29].
Proof of Corollary 2.3.1. For Φ = Λ and K(U) = U we see that 𝕂(U) = Id, the identity
matrix, satisfies P.1). The boundary conditions for U here show that P.2) is verified.
−p
Also, the weight W := Λ p+1 Φ U = Φ. The integrals in (2.2.4) and (2.2.5) are exactly
those defined here. Because 𝕂(U) = Id, we have 𝕂U (U) = 0 so that, from the defini-
tions (2.2.5) and (2.2.7), I1̂ = I1̄ . Finally, with ω ≡ 1, I0̆ = 0. We then have (2.3.3) from
(2.2.8).
In particular, if Φ = Λ ≡ 1 and K is the identity map then I1̂ = I1̄ = 0. Theorem 2.2.1,
with general measure dμ = ωdx, then shows that
I1 ≤ C‖U‖2BMO(μ) [I2 + I0̆ ] , (2.3.4)
where
I1 := ∫ |DU|2p+2 dμ, I2 := ∫ |DU|2p−2 |D2 U|2 dμ , (2.3.5)
Ω Ω
Proof. If k = 0, the corollary is (2.3.4). We need only consider the case k > 0. We argue
as in the proof of Corollary 2.3.1, without the assumption ω ≡ 1 so that I0̆ ≠ 0, to get
We consider the integral I1̄ defined by (2.2.5). As |Φ U (U)| ∼ (λ0 + |U|)k−1 , we can
take
I1∗ := I1̄ = ∫ ϕ2k (U)|DU|2p+2 dμ , ϕ k (U) := (λ0 + |U|)k−1 . (2.3.12)
Ω
We apply (2.3.11) again, with Φ(U) being ϕ k (U) and the integrals being redefined
accordingly, to get
I1∗ ≤ C‖U‖2BMO(μ) [I2∗ + I∗1̄ + C([ϕ αk (U)]β+1 )(I2∗ + I∗1̄ + I∗0̆ )] . (2.3.13)
Because ϕ k (U) ∼ Φ(U)δ with δ = (k − 1)/k < 1, by the definition of weights and
Hölder’s inequality we have [ϕ αk (U)]β+1 ≲ [Φ α (U)]δβ+1 . On the other hand, because
λ0 > 0, ϕ k (U) ≤ C(λ0 )Φ(U) for some constant C(λ0 ). Thus, I2∗ ≤ C(λ0 )I2 and I∗0̆ ≤
C(λ0 )I0̆ . Using these facts in (2.3.13) and then applying the result in (2.3.11), we see
that (2.3.11) holds again with I1̄ being defined as in (2.3.12) with ϕ k (U) being ϕ k−1 (U) =
(λ0 + |U|)k−1 .
Hence, assume that k = k 0 + κ for some integer k 0 = 0, 1, . . . and κ ∈ [0, 1) then
we can repeat the above argument k 0 + 1 times to get
The last integral can be estimated by (2.3.4), using the definition (2.3.5). Again, because
λ0 > 0, we can find a constant C(λ0 ) such that I1∗∗ ≤ C(λ0 )‖U‖2BMO(μ) [I2 + I0̆ ]. Using
this in (2.3.14), we obtain (2.3.8). The proof is complete.
16 | 2 Interpolation Gagliardo–Nirenberg inequalities
Theorem 2.3.4. Assume as in Theorem 2.2.1. Assume further that there is a constant C0
such that
|𝕂U | ≤ C0 , (2.3.16)
−1 −1
|Λ U | ≤ C0 Φ , |Φ U |Φ ≤ C0 |Λ U |Λ . (2.3.17)
Proof. From the assumption |Λ U | ≤ C0 Φ in (2.3.17) and the definition (2.2.5), it is clear
that I1̄ ≤ I1 .
We now look at I1̂ , the integral of (|Φ U |𝕂| + Φ|𝕂U |)2 |DU|2p+2 . First of all, because
𝕂 ≲ ΛΦ−1 , (2.3.17) implies |Φ U |𝕂| ≲ |Φ U |ΛΦ−1 ≲ |Λ U | ≲ Φ. Together with (2.3.16), we
now see that |Φ U |𝕂| + Φ|𝕂U | ≲ Φ and thus have I1̂ ≲ I1 .
The inequality (2.3.18) then follows from (2.2.8) of Theorem 2.2.1.
The local version Theorem 2.2.2 then implies the following corollary.
Corollary 2.3.5. Assume (2.3.16) and (2.3.17). Using the definitions (2.2.11) and (2.2.12)
̆ , we have
for I1,∗ and I0,∗
̆ ]] .
I1,∗ ≤ εI1 + ε−1 C‖K(U)‖2BMO(Ω) [I2 + I1 + C([W α ]β+1 )[I2 + I1 + I0,∗ (2.3.19)
Remark 2.3.6. In applications, we usually have Λ(U) = √λ(U) for some C2 function λ.
Furthermore, we also have Φ(U) = |Λ U (U)|. In this case, the first inequality in (2.3.17)
is clear. We look at the second one: |Φ U |Φ−1 ≤ C0 |Λ U |Λ−1 . It is clear that
|Φ U | |λ UU | |λ U |2 √λ |λ UU | |λ U | |Λ U | |λ U |
∼( + 3/2 ) = + , ∼ .
Φ √λ λ |λ U | |λ U | λ Λ λ
|𝕂U | ≤ C0 ⇔ |K −1 −1
U (U)K UU (U)K U (U)| ≤ C 0 .
Proof of Corollary 2.3.7. We apply (2.3.18) of Theorem 2.3.4 to this case. We consider
Λ(U) = (ε + |U|)k and Φ(U) ∼ |Λ U (U)| and validate the assumptions of Theorem 2.3.4.
Concerning the condition (2.3.16), as K(U) = [log(ε + |U i |)]m i=1 so that K U (U) =
diag[(ε + |U i |)−1 ] and K UU = diag[−(ε + |U i |)−2 |UU ii | ]. It is clear that (2.3.16) is justified.
Next, for Λ(U) = (ε + |U|)k and Φ(U) = |Λ U (U)| we let λ(U) = (ε + |U|)2k in Re-
mark 2.3.6 and see that |λ UU (U)|λ(U) ∼ |U|4k−2 ∼ |λ U (U)|2 . We see that the assumption
(2.3.17) is verified.
Hence, (2.3.18) of Theorem 2.3.4 applies here. As Φ(U) ∼ (ε + |U|)k−1 , we have
W = Λ p+1 Φ−p ∼ (ε + |U|)k+p . We then obtain (2.3.23) from (2.3.18) and the proof is
complete.
Let us discuss the connection between the two quantities ‖K(U)‖BMO(μ) , [W α ]β+1 . We
consider the case m = 1. As W ∼ |k|−p (ε+|U|)k+p , we have [W α ]β+1 = [(ε+|U|)α(k+p) ]β+1
and
[log(Wα )]∗,μ = α|k + p|[log(ε + |U|)]∗,μ .
Via a simple use of Jensen’s inequality, it is well known (e.g., [17, Chapter 9]) that
‖ log w‖BMO ≤ [w]A q for 1 < q ≤ 2. In our case, q = β + 1 < 2 so that ‖ log W α ‖BMO ≤
[W α ]A q . Thus, the term [log(ε +|U|)]BMO(μ) can be controlled by [W α ]β+1 . However, this
type of result is not helpful for our purpose in this book.
On the other hand, if log W is BMO then we also know that W is a weight. We
recall the John–Nirenberg inequality (e.g., [40] or [17, Corollary 7.1.7]): if μ is doubling
then for any function v there are constants c1 , c2 , which depend only on the doubling
constant of μ, such that
c1
|v−v B |
∫ e [v]∗,μ dμ ≤ c2 . (2.3.24)
B
We then have the following result.
It is clear that if ‖ log(ε + |U|)‖BMO(μ) is sufficiently small then (2.3.25) and (2.3.26) imply
To prove Corollary 2.3.8 we have the following estimate for [Wα ]β+1 .
Lemma 2.3.9. For any scalar function w on Ω and α, β > 0 and c1 , c2 as in (2.3.24), we
have
[log(w)]∗,μ ≤ c1 βα −1 ⇒ [w α ]β+1 ≤ c2 .
1+β
(2.3.27)
Proof. For any β > 0 and any scalar function v we know that e v is an A β+1 weight with
1+β
[e v ]β+1 ≤ c2 [17, Chapter 9] if
− 1β (v−v B )
sup ∫ e(v−v B ) dμ ≤ c2 , sup ∫ e dμ ≤ c2 . (2.3.28)
B B B B
One of the key ingredients of our proof is the duality between the Hardy space H 1 (μ)
and BMO(μ) space. If μ is the Lebesgue measure then this is the famous Fefferman–
Stein theorem [8]. It is well known that the norm of the Hardy space can be defined
by
‖F‖H 1 (Ω) = ‖F‖L1 (Ω) + ‖M∗ F‖L1 (Ω) .
Here,
M∗ F(y) = sup ∫ ϕ(x)f(x)dx ,
ϕ Ω
where the supremum is taken over all ϕ ∈ C 1 (ℝN ), which has support in some cube
Q ⊂ ℝN centered at y with side length l(Q) and satisfies
For a general measure μ, this theory was generalized by Tolsa and we will dis-
cuss parts of his works in Appendix A. Roughly speaking, Tolsa showed that there is a
constant cH depending on n, N and C μ such that we can similarly define the norm of
2.4 Proof of Theorem 2.2.1 | 19
where
M Φ̆ F(y) = sup ∫ ϕ(x)f(x)dμ(x) ∈ L1 (μ) . (2.4.2)
ϕ∈Φ̆ y Ω
∫ F dμ = 0 . (2.4.3)
Ω
We will also use the definition of the centered Hardy–Littlewood maximal operator
acting on functions F ∈ L1loc (μ),
We also note here the Muckenhoupt theorem for nondoubling measures. By [40,
Theorem 3.1], we have that if w is an A q (μ) weight then for any F ∈ L q (μ) with q > 1,
In particular,
∫ M(F)q dμ ≤ C μ ∫ F q dμ . (2.4.7)
Ω Ω
We start the proof of Theorem 2.2.1. First of all, let W = K(U). We then have DU =
K U (U)−1 DW so that, from the definition of 𝕂(U) = (K −1 U ) , |DU| = ⟨𝕂⟩(U)DU, DW.
T 2
where
ℙ(U) := Φ2 (U)Λ−1 (U)𝕂(U) . (2.4.9)
Applying integration by parts to the last integral in (2.4.8), we see that the bound-
ary integral is zero because of the boundary assumption (2.2.2) in P.2). We then have
I1 = −∫ ⟨div(|DU|2p Λ(U)ωℙ(U)DU), W⟩ dx .
Ω
I1 = −∫ ⟨G, W⟩ dμ . (2.4.10)
Ω
∫ G dμ = ∫ div(|DU|2p Λ(U)ωℙ(U)DU) dx = 0 .
Ω Ω
We will establish bounds for ‖G‖L1 (μ) , ‖M Φ̆ G‖L1 (μ) and show that
1 1 1 1 1 1
‖G‖Φ̆ ≤ C [I22 + I1̄ 2 + C([W α ]β+1 )[I1̂ 2 + I22 + I0̆ 2 ]] I12 . (2.4.11)
Once this is proved, we obtain from (2.4.10) and (2.4.4) of Lemma 2.4.1 that I1 ≤
C‖K(U)‖RBMO(μ) ‖G‖Φ̆ . As we are assuming that μ is doubling, ‖K(U)‖RBMO(μ) ∼
‖K(U)‖BMO(μ) . We then deduce
1 1 1 1 1 1
I1 ≤ C‖K(U)‖BMO(μ) [I22 + I1̄ 2 + C([W α ]β+1 )[I22 + I1̂ 2 + I0̆ 2 ]] I12 ,
which yields (2.2.8) of Theorem 2.2.1 via a simple use of Young’s inequality. The proof
will then be complete.
To prove (2.4.11), we first estimate ‖M Φ̆ G‖L1 (μ) and note that
M Φ̆ G(y) ≤ sup ∫ ϕG dμ = sup ∫ ϕg dx ,
ϕ∈Φ̆ y Ω ϕ∈Φ̆ y Ω
where
g := Gω = div(|DU|2p Λ(U)ωℙ(U)DU) . (2.4.12)
Therefore, we need to establish that there are constants C, C([W α ]β+1 ) for which
1 1 1 1 1 1
∫ sup ∫ ϕg dx dμ ≤ C [I22 + I1̄ 2 + C([W α ]β+1 )[I1̂ 2 + I22 + I0̆ 2 ]] I12 . (2.4.13)
Ω ϕ∈Φ̆ y Ω
Proof. We use integration by parts (the boundary integral is zero because ϕ ε ∈ C10 (B ε ))
to get
∫ ϕ (x)g1 dx = ∫ Dϕ ε (x)ℙ(U)(h − J 0,ε )|DU|p+1 dμ
B ε (y) ε B ε (y)
C
≤ ∫ |h − h B ε (y) ||ℙ(U)||DU|p+1 dμ . (2.4.18)
ε B ε (y)
Here, we used the property of Dϕ ε , as ϕ ε ∈ Φ̆ y , that |Dϕ ε | ≲ ε−n−1 , and the assump-
tion M) that μ(B ε ) ≲ ε n .
Note that (2.2.1) implies |ℙ(U)| ≤ Φ2 (U)Λ−1 (U)|𝕂(U)| ≲ Φ(U). This and a simple
use of Hölder’s inequality for q > 2 yield that the last integral in (2.4.18) is bounded
by
1 1
C q q
(∫ |h − h B ε (y) |q dμ) (∫ [Φ(U)|DU|p+1 ]q dμ) .
ε B ε (y) B ε (y)
Using the definition of maximal functions (2.4.5) and combining the above esti-
mates, we get from (2.4.18)
sup ∫ ϕ ε g1 dx ≤ C [Ψ1 (y) + Ψ2 (y)] Ψ3 (y) , (2.4.20)
̆ Ω
ϕ ε ∈Φ
1
q
where Ψ i (y) = (M(F i i (y))) qi with q1 = q2 = q∗ and q3 = q and
Because q i < 2 (as q > 2 and q∗ = qσ < 2), Muckenhoupt’s inequality (2.4.7)
implies
1 1 1
2 2 2 2
q
(∫ Ψ i2 dμ) = (∫ M(F i i ) qi dμ) ≤ C μ (∫ F 2i dμ) .
Ω Ω Ω
Therefore, applying Hölder’s inequality to (2.4.20) and using the above estimates and
the notations (2.2.4) and (2.2.5), we obtain (2.4.17).
Remark 2.4.3. We remark that (2.4.19) is the only place where we need the assump-
tion PS) that Ω, μ support a Poincaré–Sobolev inequality.
We now turn to g2 .
r+1 r(p + 1) − 1
α(r) = , β(r) = .
rp + r + 1 r(p + 1) + 1
1
Then there is C([W α(r) ]β(r)+1 ) ∼ [Wα(r)]β(r)+1
α(r)(p+1)
such that
1 1 1 1 1
∫ sup ∫ ϕ ε g2 dx dμ ≤ C ([W α(r) ]β(r)+1 ) [I1̂ 2 + I1̄ 2 + I22 + I0̆ 2 ] I12 . (2.4.21)
Ω ϕ ε ∈Φ̆ y Ω
Proof. From (2.4.16), we have div V2 ≤ C(J 1 + J 2 + J 3 ) for some constant C and
J 1 := ω|ℙU ||DU|p+2 J 0,ε , J 2 := ω|ℙ(U)||DU|p |D2 U|J 0,ε ,
J3 := Dω|ℙ(U)||DU|p+1 J 0,ε ,
B ε B ε
If r1 > 1/(p+1) we apply Hölder’s inequality to the last integral to get the following
estimate for J 0,ε :
1 r ∗1 1
r1 (p+1)
∗
J 0,ε ≤ (∫ W r1 (p+1) dμ) (∫ f pr1 dμ) . (2.4.22)
Bε Bε
Because ϕ ε (x) ≲ ε−n ∼ μ(B ε )−1 so that we can use Hölder’s inequality to get for any
s > 1,
1 1
∫ ϕ J dx ≤ (∫ L s dμ) s (∫ L s dμ) s J .
ε 1 ∗ 0,ε
Ω Bε Bε
2.4 Proof of Theorem 2.2.1 | 23
−sp s
We write L s = Λ−s Φ (p+1) Φ p+1 |DU|s and use Hölder’s inequality to get for any r >
1/(p + 1) the following estimate:
1 r∗ 1
s −s sp s rs(p+1)
(∫ L dμ) ≤ (∫
s
|Λ| Φ r∗ r∗ (p+1) dμ) (∫ f sr dμ) .
Bε Bε Bε
We now choose s, r, r1 such that s = sr = pr1 and sr < 2. This is the case if
r ∈ ( p+1
1
, 1), s = (r + 1)/r then s = r + 1 and r1 = (r + 1)/p > 1/(p + 1). Let α(r) = r∗ (p+1)
1
1
r∗
and β(r) = r ∗1 s
. With such a choice of s, r, r1 we have
−α(r) β(r)
C1,r = sup (∫ W α(r) dμ) (∫ W β(r) dμ) , (2.4.25)
B⊂Ω B B
where the supremum is taken over all cubes B in Ω. Clearly, the definition of weight
(3.2.2) implies
ν
[w]ν+1 = sup (∫ w dμ) (∫ w− ν dμ)
1
for all ν > 0 . (2.4.26)
B⊂Ω B B
Because q = 2/(rs) > 1, we can apply (2.4.7) to the integrals on the right and then
use the definitions of L∗ , f, I1̂ to see that
r ∗1
∫ sup ∫ ϕ ε J 1 dx dμ ≲ C1,r ‖L∗ ‖2 ‖f‖2 . (2.4.28)
Ω ̆ Ω
ϕ ε ∈Φ y
Concerning the term ‖L∗ ‖2 , we recall the definitions L∗ = ℙU ΛΦ−1 |DU|p+1 and ℙ(U) =
Φ2 (U)Λ−1 (U)𝕂(U). It is clear that |ℙU |ΛΦ−1 is bounded by
Here, we used (2.2.1) to get Φ|Λ U |Λ−1 |𝕂| ≲ |Λ U |. Therefore, from the definition of I1̂ ,
1 1
I1̄ , ‖L∗ ‖2 ≲ I ̂ 2 + I ̄ 2 . We then obtain from (2.4.28) that
1 1
1 1 1
∫ sup ∫ ϕ ε J 1 dx dμ ≲ C (C1,r ) (I1̂ 2 + I1̄ 2 ) I12 . (2.4.29)
Ω ϕ ε ∈Φ̆ y Ω
We repeat the argument for J 1 . Note that |ℙ(U)| ≤ Φ(U), by (2.2.1), and therefore
−sp s
L s ≤ Λ−s Φ (p+1) Φ p+1 |DU|s . We have the following inequality:
1 r∗ 1
s −s sp s rs(p+1)
(∫ L dμ) ≤ (∫
s
|Λ| Φ
r∗ r∗ (p+1) dμ) (∫ f sr
dμ) .
Bε Bε Bε
The same argument in (2.4.29) for J1 with the new definition of L∗ yields
r ∗1 1 1
∫ sup ∫ ϕ ε J 2 dx dμ ≲ C1,r ‖L∗ ‖2 ‖f‖2 = C(C1,r )I22 I12 . (2.4.31)
Ω ̆ Ω
ϕ ε ∈Φ y
Combining the estimates (2.4.29), (2.4.31), (2.4.32) and (2.4.27) (for the constant
C1,r ), we derive (2.4.21).
2.4 Proof of Theorem 2.2.1 | 25
Proof. Recall that G := gω−1 ((2.4.12)), thus ‖G‖L1 (μ) = ‖g‖L1 (dx). We write g =
div(B|DU|2p DU) with B = ωΛ(U)ℙ(U). First of all,
we see that a simple use of Hölder’s inequality as in the proof of Lemma 2.4.4, treating
the last factor ωΛ|Du|p as J 0,ε , implies
1 1 1 1
∫ |B U ||DU|2p+2 dx ≤ C[I1̂ 2 + I22 + I0̆ 2 ]I12 .
Ω
Thus, if α > 2/(p + 2) and β < p/(p + 2) then for r close to 1 we have α(r) < α and
β(r) > β. Hence, by choosing r close to 1 and using (2.4.34) and the open-end property
of weights, we see that
α(r)
[W α(r)]β(r)+1 ≲ C[W α ]β+1
α
. (2.4.35)
1 1
As C([W α(r) ]β(r)+1 ) ∼ [W α(r) ]β(r)+1
α(r)(p+1)
≲ [W α ]β+1
α(p+1)
, we can replace the constant
1
C([W α(r) ]β(r)+1 ) in (2.4.33) by a multiple of [W α ]β+1 and obtain (2.4.11). Namely,
α(p+1)
1 1 1 1 1 1
‖G‖Φ̆ ≤ C [I22 + I1̄ 2 + C([W α ]β+1 )[I1̂ 2 + I22 + I0̆ 2 ]] I12 .
26 | 2 Interpolation Gagliardo–Nirenberg inequalities
This gives
I1 ≤ C‖K(U)‖2BMO(μ) [I2 + I1̄ + C([W α ]β+1 )[I2 + I1̂ + I0̆ ]] . (2.4.36)
The proof of the theorem is complete.
Remark 2.4.7. The only place we use the assumption PS) is (2.4.19). We just need to
assume that PS) holds true for h = Λ(U)|DU|p−1 DU and some measure μ satisfying
M). Combining this with [19, Theorem 5.1] (Remark 2.1.1), which deals only with a pair
u, Du, we need only that some Poincaré inequality (2.1.4) holds for the pair h, Dh. That
is, we do not need (2.1.4) to hold for any h but the function h = Λ(U)|DU|p−1 DU in the
consideration.
Applying integration by parts to the last integral, where the boundary integral is
zero because ω∗ ∈ C10 (Ω), we obtain
I1,∗ ≤ −∫ ⟨G, W⟩ dμ ,
Ω
−2
∗ ω 0 ℙ(U)DU)ω 0
and W = K(U).
2
where G := div(|DU|2p Λ(U)ω
We now follow the proof of Theorem 2.2.1 to establish a similar version of (2.4.11),
with dμ = ω20 dx and G being defined above, to complete the proof. We see that (2.4.11)
holds true if (2.4.13) does. We then need only establish a similar version of (2.4.13).
Again, we can write g = g1 + g2 with g i = div V i , setting
Since ω∗ ≤ 1 we can discard it in the estimates for g1 after the use of integration by
parts (2.4.18) in the proof of Lemma 2.4.2. Because the measure μ supports a Poincaré–
Sobolev inequality (2.1.3), we can repeat the argument in the proof of Lemma 2.4.2 to
obtain the same estimate for the integral in (2.5.2) with i = 1. Similarly, we drop ω∗
in J i ’s, with the exception of J 3 , in the proof of Lemma 2.4.4 to estimate the integral in
(2.5.2) with i = 2. Therefore,
1 1 1 1 1 1
∫ sup ∫ ϕ ε g dx dμ ≤ [I22 + I1̄ 2 + C([W α ]β+1 )[I1̂ 2 + I22 + I∗̆ 2 ]] I12 .
Ω ϕ ε ∈ Φ̆ Ω
Here, the term I∗̆ , replacing I0̆ in (2.4.21), comes from the estimate for
to obtain the following version of (2.4.32) (with C1,r being replaced by [Wα ]β+1 ):
1 1
∫ sup ∫ ϕ ε J 3 dx dμ ≤ C([W α ]β+1 )I∗̆ 2 I12 ,
Ω ϕε Ω
Because |D(ω∗ ω20 )|2 ≲ |Dω∗ |2 ω40 + ω2∗ |Dω0 |2 ω20 , we have
We then have
The constant C depends on C PS , C μ and C H . In the same way, Lemma 2.4.5 gives a
similar estimate for ‖G‖L1 (μ) . Hence,
1 1 1 1 1 1
‖G‖Φ̆ ≤ C [I22 + I1̄ 2 + C([W α ]β+1 )[I1̂ 2 + I22 + I0,∗
̆ 2 ]] I 2 .
1
We then apply Lemma 2.4.1 as before and use Young’s inequality to prove (2.3.7)
for the case dμ = ω20 dx.
Finally, if ω1 ∼ ω20 then the integrals in (2.3.7) with respect to the two measures
are comparable, because Dω1 , Dω0 are not involved, so that (2.3.7) holds true as well.
The proof is complete.
Remark 2.5.1. For simplicity we assumed in L.0) that ω∗ ∈ C10 (Ω). More generally, we
need only that u = ω∗ Λ|DU|p ∈ C10 (Ω) so that the Hardy inequality can apply in (2.5.3).
3 The parabolic systems
In this chapter, for any T0 > 0 and bounded domain Ω with smooth boundary in ℝN ,
N ≥ 2, we consider the following parabolic system of m equations (m ≥ 2) for the
unknown u : Q → ℝm , where Q = Ω × (0, T0 ):
{
{ u t − div(A(x, u)Du) = f ̂(x, u, Du) , (x, t) ∈ Q ,
{
{ u = 0 or ∂u
∂ν = 0 on ∂Ω , t ∈ (0, T0 ) , (3.0.1)
{
{
{u(x, 0) = U0 (x) x∈Ω.
The first fundamental problem in the study of (3.0.1) is the local and global exis-
tence of its solutions. One can decide to work with either weak or strong solutions. In
the first case, the existence of a weak solution can be achieved via Galerkin, time dis-
cretization or variational methods but its regularity (e.g., boundedness, Hölder con-
tinuity of the solution and its higher derivatives) is then an open and serious issue.
Several works have been done along this line to improve on the early work [14] of
Giaquinta and Struwe and establish partial regularity of bounded weak solutions to
(3.0.1).
Otherwise, if strong solutions are considered then their existence can be estab-
lished via semigroup theories as in the works of Amann [1, 2]. Combining this with
interpolation theories of Sobolev’s spaces, Amann established local and global ex-
istence of a strong solution u of (3.0.1) under the assumption that one can control
‖u‖W 1,p (Ω,ℝm ) for some p > n. His theory did not apply to the case where f ̂ has a super-
linear growth in Du.
In both aforementioned approaches, the assumption on the boundedness of u
must be the starting point because the techniques rely heavily on the fact that A(x, u)
is regular elliptic. For strongly coupled systems like (3.0.1), as invariant/maximum
principles are generally unavailable, the boundedness of the solutions is already a
hard problem. One usually needs to use ad hoc techniques on a case by case basis to
show that u is bounded [23, 41]). Even so, for bounded weak solutions we know that
they are only Hölder continuous almost everywhere [14]. In addition, there are counter
examples by John and Stará [21] for systems (m = 2) that exhibit solutions that start
smoothly and remain bounded for all time but develop singularities in higher norms
in finite times.
In our recent work [29, 30], we chose a different approach making use of fixed-
point theory and discussed the existence of strong solutions of (3.0.1) under the weak-
https://doi.org/10.1515/9783110608762-003
30 | 3 The parabolic systems
est assumption that they are a priori VMO (vanishing mean oscillation), but not nec-
essarily bounded, and general structural conditions on the data of (3.0.1) that are in-
dependent of x. We assumed only that A(u) is uniformly elliptic. Applications were
presented in [30] when λ(u) has a polynomial growth in |u| and, without the bound-
edness assumption on the solutions, so (3.0.1) can be degenerate as λ(u) → ∞ when
|u| → ∞. The singular case, λ(u) → 0 as |u| → ∞, was not discussed there. Also, the
reaction term f ̂ was assumed to only have linear growth in gradient.
In this book, we will establish much stronger results than those in [29] under more
general assumptions on the structure of (3.0.1) as described in A) and F) of Section 3.1.
Besides the minor fact that the data can depend on x, we allow further that:
– A(x, u) can be either degenerate or singular as |u| tends to infinity.
– f ̂(x, u, Du) can have a superlinear growth in Du as in f.2). Even quadratic growth
in gradient is considered in the low-dimensional case.
– No a priori boundedness of solutions is assumed but a very weak a priori integra-
bility of strong solutions of (3.0.1) is considered.
Most remarkably, the key assumption in [29, 30] that the BMO norm of u is small in
small balls will be replaced by a more versatile one in this paper: K(u) is has small BMO
norm in small balls for some suitable map K : ℝ m → ℝm . This allows us to consider
the singular case where one may not be able to estimate the BMO norm of u but can
estimate that of K(u). Examples in applications when the systems are singular will be
provided later in Section 3.5 where |K(u)| ∼ log(|u|).
One of the key ingredients in the proof in [29, 30] is the local weighted Gagliardo–
Nirenberg inequality involving BMO norm [29, Lemma 2.4]. In this book, we make use
of a new version of this inequality developed in Chapter 2 replacing the BMO norm of
u by that of K(u). Combining this with a new and nontrivial iteration argument, start-
ing with a very weak integrability assumption on the solutions to (3.0.1), we establish
uniform estimates for their derivatives so that a homotopy argument in fixed-point
theories can be used here to provide the existence of strong solutions.
We organize this chapter as follows. We describe the general structural conditions
of (3.0.1) in Section 3.1. These conditions will be referred to throughout the book, with
the exception of Chapter 5. In Section 3.2 we state the main result of this chapter, The-
orem 3.2.3, and present its proof in Section 3.3. A simpler case of this theorem, which
however finds many applications, will be discussed in Section 3.4. Applications of
these theorems will be given in Section 3.5 for the general SKT models (see also Chap-
ter 5 for more general results).
It is always assumed, with the exception of Chapter 5, that the matrix A(x, u) is elliptic
in the sense that there exist two scalar positive continuous functions λ1 (x, u), λ2 (x, u)
3.1 The main structural conditions | 31
such that
λ1 (x, u)|ζ |2 ≤ ⟨A(x, u)ζ, ζ⟩ ≤ λ2 (x, u)|ζ |2 (3.1.1)
for all x ∈ Ω, u ∈ ℝm , ζ ∈ ℝmN .
If there exist positive constants c1 , c2 such that c1 ≤ λ1 (x, u) and λ2 (x, u) ≤ c2
then we say that A(x, u) is regular elliptic. If 0 < λ1 (x, u) and λ2 (x, u)/λ1 (x, u) ≤ c2 ,
we say that A(x, u) is uniform elliptic. In addition, if λ1 (x, u) tends to zero (respec-
tively ∞) when |u| → ∞ then we say that A(x, u) is singular (respectively degenerate)
at infinity.
Without the boundedness assumption on the solutions of (3.0.1), we will consider
the cases when A(x, u) is uniformly elliptic and degenerate/singular at infinity. The
following structural conditions on the data of (3.0.1) will be assumed throughout the
book.
A) A(x, u) is C 1 in x ∈ Ω, u ∈ ℝm and there exist a constant C∗ > 0 and scalar C1
positive functions λ(u), ω(x) such that for all u ∈ ℝm , ζ ∈ ℝmN and x ∈ Ω,
A(x, u)ζ = [Σ m m
j=1 a ij ζ j ]i=1 , ⟨A(x, u)ζ, ζ⟩ = Σ m
i,j=1 a ij ⟨ζ i , ζ j ⟩ .
The positivity of λ then implies that for any bounded set K ⊂ ℝm there are positive
constants λ∗ (K), λ∗∗ (K) such that
Combining (3.1.4) and AR), we see that A(x, u) is regular elliptic for bounded u. When
|u| → ∞, the matrix A(x, u) can be degenerate/singular.
Concerning the reaction vector f ̂(x, u, Du), which may have linear or subquadratic
growth in Du, we assume the following condition.
32 | 3 The parabolic systems
and
{|∂ x f ̂(x, u, ζ )| ≤ C[λ(u)|ζ | + f(u)]|Dω(x)| ,
{
{
̂
{|∂ u f (x, u, ζ )| ≤ C[|λ u (u)||ζ | + |f u (u)|]ω(x) ,
{
{ ̂
{|∂ ζ f (x, u, ζ )| ≤ Cλ(u)|ω(x) ;
or that
f.2) f ̂ has a superlinear growth in ζ . There is δ ∈ (0, 2/N) and λ uu (u) exists such
that
|f ̂(x, u, ζ )| ≤ C|λ u (u)||ζ |1+δ ω(x) + f(u)ω(x) , (3.1.7)
{
{ |∂ x f ̂(x, u, ζ )| ≤ C[|λ u (u)||ζ |1+δ + f(u)]|Dω(x)| ,
{
{ |∂ f ̂(x, u, ζ )| ≤ C[|λ uu (u)||ζ |1+δ + |f u (u)|]ω(x) ,
{ u
{ ̂
{|∂ ζ f (x, u, ζ )| ≤ C|λ u (u)||ζ | |ω(x) .
δ
By a formal differentiation of (3.1.6) and (3.1.7), one can see that the growth conditions
for f ̂ naturally imply those of its partial derivatives in the above assumptions. The
condition (3.1.8) is obviously verified if λ(u) has a polynomial growth in |u|.
We state the main results of this paper in this section. The key assumption of these
results is some uniform a priori estimate for the BMO norm of K(u) where K is some
suitable map on ℝm and u is any strong solution to (3.0.1). To begin, we recall some
basic definitions in harmonic analysis.
Let ω ∈ L1 (Ω) be a nonnegative function and define the measure dμ = ω(x)dx.
For any μ-measurable subset A of Ω and any locally μ-integrable function U : Ω → ℝm
we denote by μ(A) the measure of A and U A the average of U over A. That is,
1
U A = ∫ U(x) dμ = ∫ U(x) dμ .
A μ(A) A
3.2 Preliminaries and main results | 33
We then define
‖f‖BMO(Ω,μ) := [f]∗,μ + ‖f‖L1 (Ω,μ) .
γ−1
[w]γ,Ω := sup (∫ w dμ) (∫ w1−γ dμ) is finite . (3.2.2)
B⊂Ω B B
Here, γ = γ/(γ − 1). For more details on these classes we refer the reader to [40, 45]. If
the domain Ω is specified we simply denote [w]γ,Ω by [w]γ .
To begin, as in [29] with A independent of x, we assume that the eigenvalues of the
matrix A(x, u) are not too far apart. Namely, for C∗ defined in (5.0.3) of A) we assume
SG) (N − 2)/N < C−1 ∗ .
Here C∗ is, in a certain sense, the ratio of the largest and smallest eigenvalues of
A(x, u). This condition seems to be necessary as we deal with systems [31].
First of all, we will assume that the system (3.3.1) satisfies the structural conditions
A) and F). Additional assumptions then follow so that the local weighted Gagliardo–
Nirenberg inequality of Chapter 2 can apply here.
K) There is a C1 map K : ℝm → ℝm such that 𝕂(u) = (K u (u)−1 )T exists and 𝕂u ∈
L∞ (ℝm ). Furthermore, for all u ∈ ℝm ,
{
{ u t − div(A(x, u)Du) = f ̂(x, u, Du) , (x, t) ∈ Ω × (0, T0 ) ,
{
{ u = 0 or ∂u
∂ν = 0 on ∂Ω × (0, T 0 )
(3.2.4)
{
{
{ u(x, 0) = U0 (x) .
We embed this system in the following family of systems:
{
{ u t − div(A(x, σu)Du) = f ̂(x, σu, σDu) , (x, t) ∈ Ω × (0, T0 ), σ ∈ [0, 1] ,
{
{ u = 0 or ∂u
∂ν = 0 on ∂Ω × (0, T 0 )
(3.2.5)
{
{
{ u(x, 0) = U0 (x) .
For any strong solution u of (3.2.5) we will consider the following assumptions.
34 | 3 The parabolic systems
M.0) There exists a constant C0 such that for some r0 > 1 and β 0 ∈ (0, 1)
M.1) For any given μ 0 > 0 there is positive R μ0 sufficiently small in terms of the con-
stants in A) and F) such that
Furthermore, for Wp (σ, x, τ) := λ p+ 2 (σu)|λ u (σu)|−p and any p such that 1 − 1/p <
1
2
C−1
∗ we have sup τ∈(0,T 0 ) [W p ] 4 ≤ C 0 .
3
3
Remark 3.2.1. The conditions (3.2.7) and (3.2.8) are not needed if f ̂ is independent of
x. Also, in many physical models, the ellipticity ‘constant’ λ(u) in A) and the ‘reaction’
f(u) have polynomial growths like λ(u) ∼ (λ0 + |u|)k and f(u) ∼ |u|k0 λ(u) for some
positive constants λ0 , k and k 0 . In these cases, the condition (3.2.6) is clearly true if
‖u‖L l0 (Ω) ≤ C0 for some l0 > k 0 . The conditions in (3.2.7) then come from this if we
choose β 0 to be sufficiently small. Furthermore, in connection with M.1), especially
q
when K(u) = u, we see that (3.2.10) implies |u| is BMO so that |u| ∈ Lloc (Ω) for any
q > 1 and hence M.1) implies M.0), with the exception of (3.2.9).
Remark 3.2.2. The condition on W p in M.1) allows us to make use of the Gagliardo–
Nirenberg inequalities in Chapter 2 for any p ≥ 1. In fact, from the basic properties of
A γ weights, we have
μ0 (x 0 )∩Ω
It follows that the condition W) on the weight W in Chapter 2 holds and the Gagliardo–
Nirenberg inequalities are available.
3.
One day as she was eating a hurried luncheon she noticed a young
man reading by the window. His hair was ash blonde, brushed
glossily back, his face thin, sensitive, and browned by the sun. When
he smiled at Terese, the waitress, his teeth were milk-white, and very
regular. His eyes should have been blue, but were of a dark, velvety
brown. An extraordinary good-looking boy, she thought, with an air of
refinement, of race. He looked up and caught her eye; immediately
she looked down.
She had seen him before, she fancied, but where? Then she
remembered the young man who had stared at her in the train. It
was strange she should meet him again.
She saw him often afterwards in the gardens, walking hatless, with
his head held high. He never went to the Casino, and seemed very
gay and happy. It was easy to see he was well off, and had not a
care in the world. Once he passed her as she was on her way to her
room, but shyness came over her and she did not glance at him. He
looked so proud; he must be at least the son of an English lord. Why
then should she, daughter of a French head-waiter and an English
bar-maid, be even on bowing terms with him?
Then something happened that quite drove him out of her thoughts.
For ten days she had been playing her system without even a loss,
gaining nearly a thousand francs. Her winnings so far had more than
paid her modest expenses. When she entered the Casino on
Christmas morning she had four bills of a thousand francs each in
her bag. She had also a letter from Jeanne. Jeanne knew of such a
nice little shop on the Boulevard Raspail. It would be empty by the
January term and, if Margot was willing, they would each put in two
thousand francs and take it. Jeanne wanted an answer at once.
Margot was very happy. She would tell Jeanne to take the shop, and
she herself would return to Paris shortly after the beginning of the
new year. She was sorry to think of leaving Monte Carlo, and to give
up roulette; the keen shifts and stresses of the game intrigued her
and she loved that moment of emotion just before the ball dropped.
Then the thought came to her: Why not experience a moment of
more intense emotion than she had ever known? She had a
thousand francs of the bank’s money that she did not absolutely
need. Why not risk it? If she could win with bills of a hundred, why
not with notes of a thousand? She watched the table until the
opportunity came. She placed five hundred francs on the second
dozen, and five hundred on the third, then with an air of unconcern
fell to regarding one of the pictures on the wall. It was a painting of
Watteau-like delicacy, representing autumn; falling leaves, gallants
and ladies of the court....
“Rien ne va plus.”
Would the ball never drop? She heard it knocking about among the
diamond-shaped brass projections. Then silence, and ... zero.
Oh, what a fool she had been! For the first time in weeks she had
forgotten to cover zero. And for the first time in weeks she had
encountered it. She hated the calm croupier who raked in her
thousand francs. There was something so ruthless, so inexorable in
the way he did it. A dull rage filled her. She seemed to be impelled by
something stronger than herself. She took from her bag a second
note of a thousand francs and played it as before. No, she would not
stake on zero. The chances of it repeating were a thousand to one....
Zero! again!
It could not be possible! As she saw another thousand swept away
she felt physically sick. She sat down on a lounge, dazed, stunned.
The impassive croupiers seemed suddenly to become mocking
satyrs, the great guilded hall, pitiless, cruel. She watched a little
hunch-backed croupier spin the wheel by its brass handle; he flipped
the ivory sphere in the other direction in a careless, casual manner.
The girl started up. It was as if she were an automaton, moved by
some force outside of her will. Taking a third thousand franc bill from
her bag, she staked it in the same way as before. No use to stake on
zero this time. The chance of its coming up a third time was a million
to one. She saw the ball go scuttling among the brass knobs; she
heard a great murmur from the gazing crowd; all eyes turned
admiringly to the little hunchback who tried to look as if he had done
it on purpose.... ZERO!
She walked away. A bitter recklessness had seized her. She took out
her remaining thousand franc bill. She would risk it anywhere,
anyhow. A red haired man was coming in at the door. That was an
inspiration. She would play on the red and leave it for a paroli. She
went over to the nearest croupier and handed him her bill.
“Rouge, please.”
But the croupier misunderstood. He put the bill on black, looking at
her for approval. After all, what did it matter? Let it remain on black.
She nodded and black it was. Once more the ball whizzed dizzily
round and dropped into its slot. Rouge.
She had lost. In less than four minutes she had lost four thousand
francs. She pulled down her veil and walked out of the gambling
rooms. Her legs were weak under her and she felt faint. She sat
down on a bench in the atrium. It could not be true! She must have
dreamt it. She opened her hand-bag of shabby black leather and
searched feverishly. All she found was about thirty francs.
She was broke.
CHAPTER SIX
DERELICT
1.
2.
3.
Some days later she sat in her room staring at her face in the cheap
mirror. There were dark circles around her weary eyes. Her cheeks
were thinned to pathetic hollows, her mouth drooped with despair
and defeat. The Casino had beaten her. She was sick, weak,
nervously unstrung. Try as she would she could not get back her old
healthy view of life; that was the worst of it. Gambling had poisoned
the very blood in her veins.
She had no money to take her back to Paris, even if she were willing
to go, and she felt she would rather die than write and ask for help.
Then to take up the burden of labour again, the life of struggle
without hope and with misery to crown it all, ... Ah! she knew it so
well. She had seen too many of her comrades fight and fall. Must
she too work as they worked, until her strength was exhausted and
she perished in poverty?
There was death, of course! Only last week a young girl, after
pawning all her trinkets, shot herself under the railway bridge. She
would do better than that; she had some little white powders.
Then there was the compromise. Why not? Who under the
circumstances would dare to tell her that death was better than
dishonour? And yet ... she hated to think of doing it. She preferred to
steal. Funny, wasn’t it? Her sense of morality was curious. She
would rather be a thief than a harlot.
But she had no chance to be a thief. It would have to be the other
thing. Rising she put rouge on her ghastly cheeks then rubbed it off
again. No, not just yet! She would ask the young man for the five
hundred francs. If he demanded the quid pro quo she would beg him
to wait until to-morrow. Then she would go to the Casino and risk all.
If she won she would return him his money, and say she had
changed her mind. If she lost ... well, there was the white powder....
She would ask him at once. How dark and silent the house was.
Room fourteen was on the floor below. Softly she crept down the
shabby stairs. She had to put on her cloak; she shivered so.
That was his door. She hesitated, inclined to turn back. Perhaps he
had gone out. Her heart was beating horribly and the hand she put
out trembled. She knocked. There was no answer. Softly she tried
the handle of the door.
END OF BOOK ONE
BOOK TWO
The Story of Hugh
CHAPTER ONE
THE UNHAPPIEST LAD IN LONDON
1.
2.
Mr. Ainger, the cashier, sat on his high stool, and looked down at a
slim lad, twisting a shabby cap. Mr. Ainger was a tall man of about
fifty, his hair grey, his face fine and distinguished. It was said that in
his spare time he wrote.
“Well, my boy,” he said kindly, “what do you call yourself?”
“Hugh Kildair.”
The gaze of Mr. Ainger became interested. He noted the dark eyes
that contrasted so effectively with the light wavy hair, the sensitive
features, the fine face stamped with race. Centuries of selection, he
thought, had gone to the making of that face.
“A romantic name. So, my boy, you are making a start with us. I don’t
know that it’s what you would choose if you had any say in the
matter. Probably, you’d rather have been a corsair or a cowboy. I
know I would at your age. However, very few of us are lucky enough
to do the things we’d like to do. Life’s a rotten muddle, isn’t it?”
“Yes, sir.”
“Well, my young friend, I do not know if the horizon of your ambition
is bounded by cheese, if it inspires you with passion, with
enthusiasm. Still you might have made a worse choice. You might
have been in oils and varnishes, for instance, or soap. Imagine
handling those compared with that exquisite ivory curd—transmuted
by bovine magic from the dew and daisies of the field. I tell you
there’s romance in cheese; there’s even poetry. I’m sure a most
charming book could be written about it. Pardon me, but you’re not
by any chance thinking of writing a book about cheese, are you?”
“No, sir.”
“Glad to hear it. Now I think of it I might as well do it myself,—a
whimsical Belloc-sort of book with glimpses of many lands. But
there. Let us return to the subject of your future. All I can say is: Do
your best; we’ll do the rest. Now go; and believe me, our
discriminating gaze is upon you.”
In the years that followed, although he saw little of Mr. Ainger, he
was conscious of a protective and sympathetic eye. As for the work
he did not dislike it. It was pleasant in the cool gloom of the
warehouse where cheeses of all shapes and colours made strange
lights and shadows. He had more liberty too, than he would have
had in the office. He was able to make pen and ink sketches of his
companions in his spare moments. At the end of every month he
handed over his pay to Aunty who returned him a trifle for pocket-
money.
At the beginning of his fifth year his salary was raised to fifty pounds.
On the day he received his first instalment he did not return to
Balmoral Circus. Instead he went to a small room in Hammersmith,
carrying his few belongings in a cricket bag. He then wrote to Aunty,
saying he was “on his own,” and he would never see her again.
At last, at last he was free.
3.
How hard that first winter was! Fifty pounds went much further in
those days than it does now, but even then he had to go without
many needful things. An overcoat, for one. You can picture him a tall,
thin pale youth, with a woollen comforter and a shabby suit far too
small for him. He was often cold and hungry. A cough bothered him.
One day Mr. Ainger came down to see him.
“Hullo, young man. You haven’t written that book yet?”
“No, sir.”
“I am surprised. Assailed as you are by a dozen pungent odours do
you not realize that under the cork-trees of Corsica the goats browse
on the wild thyme in order that those shelves may be replenished
with green veined Roquefort; that cattle bells jingle in the high vivid
valleys of the Jura to make for us those grind-stone like masses of
cavernous Gruyère; sitting here are you not conscious of a rhythm
running through it all, of a dignity, even of an epic—cheese?”
“Well,” he went on, “I’ve come to hale you from all this source of
inspiration to a more sordid environment. There’s a spare stool in the
counting-house I think you might ornament.”
“I’ll be glad of a change, sir.”
“Good. By the way, where are you living?”
“Hammersmith, sir.”
“Ah, indeed, I have a cottage on the river. You must come and see
me.”
A fortnight later he took Hugh to his little villa. It was the only real
home the lad had even seen, and was a revelation to him. Mrs.
Ainger was the first sweet woman he had ever met, and he
immediately worshipped her. There were two fine boys and a most
fascinating library.
The Aingers had a great influence on Hugh’s development. Through
them he met a number of nice fellows and instinctively picked up
their manners. He played football, cricket, and tennis,—at which
games he was swift and graceful, but somewhat lacking in stamina.
He studied French, and Mr. Ainger was at great pains to see that he
had a good accent. But best of all, he was able to attend an art
school in the evenings and satisfy a growing passion for painting.
Then the war broke out. He went to France with the First Hundred
Thousand. In the wet and cold of the trenches he contracted
pneumonia and his recovery was slow. As soon as he was well
again, he was transferred to the transport service and drove a
camion in the last great struggle. When he was demobilized he
returned to the office at a comfortable salary.
Everything looked well now, everything but his health. He suffered
from a chronic cold and was nearly always tired.
Then one raw day in early Spring he saw a poor woman throw her
child over the Embankment.
“She was quite close to me,” he told Mr. Ainger afterwards, “so of
course I went in. It was instinctive. Any other chap would have done
the same.
“Well, I grabbed the kid and the kid grabbed me, and there I was
treading water desperately. But it was hard to keep afloat; and I
thought we must both go down. I remember I felt sorry for the little
beast. I didn’t care a hang for myself. Then just as I was about to
give up, they lifted us into a boat. There was a crowd and cheering,
but I was too sick to care. Some one took me home in a taxi and my
landlady put me to bed.”
The chill that resulted affected his lungs. All winter he had fits of
coughing that made him faint from sheer exhaustion. He awoke at
night bathed in cold sweat. In the morning he was ghastly, and rose
only by a dogged effort. One forenoon, after a hard fit of coughing
Mr. Ainger said to him:
“Cold doesn’t seem to improve.”
“No, sir.”
“By the way, ever had any lung trouble in your family?”
“Yes, sir. My mother, I’ve been told, died of it.”
“Look here, take the afternoon off and see our doctor.”
The doctor was a little bald, rosy man. He looked up at Hugh’s nigh
six feet of gaunt weariness.
“You’re not fit to be out, sir. Go home at once. I’ll see you there.”
So Hugh went to his bed, and remained in it all summer.
4.
One day in late October he lay on his bed staring drearily at the
soiled ceiling, and wondering if in all London there was a lad more
unhappy than he.
“A lunger,” he thought bitterly. “Rotten timber! A burden to myself and
others. Soon I must take up the fight again and I’m tired, tired. I want
to rest, do nothing for a year or two. Well, I won’t give in. I’ll put up a
good scrap yet. I’ll——”
Here a knock came at the door. It was the little doctor cheery and
twinkling.
“Hullo! How’s the health to-day?”
“Better, doctor; I’ll soon be able to go back to the office.”
The doctor laughed: “If you remain in London another six months
you’ll be a dead man.”
“What would you have me do?”
“Go away. Live in a warm climate. Egypt, Algeria, the Riviera.”
“And if I go away how long will I live?”
“Oh, probably sixty years.”
“Quite a difference. Well, doctor, I expect I’ll have to stick it out here.
You see, I’ve no money, no friends. Even now I’m living on the
charity of the firm. They’ve been awfully decent, but I can’t expect
them to go on much longer.”
“Have you no relatives?”
“None that I know of. I’m absolutely alone in the world.”
“Well, well! We’ll see about it. Surely something can be done. Don’t
get down-hearted. Everything will come out all right.”
The little doctor went away, and Hugh continued to stare at the
soiled ceiling. There came to him a desperate vision of palms and
sunshine. But that was not for him. He must stay in this raw bleak
London and perish as many a young chap had perished....
Next morning came another knock at the door. It was Mr. Ainger.
“Well, my lad, how are you feeling?”
“A little better. I hope I’ll soon be able to get back to my ledger.”
“Nonsense, my boy! You’ll never come back. You’re expected to
hand in your resignation. The doctor holds out no hope. You can’t go
on drawing on your salary indefinitely.”
Hugh swallowed hard. “No, that’s right. You’ve treated me square. I
can’t complain.”
“Complain, I should say not; look here....”
With that Mr. Ainger took from his pocket a sheaf of crisp Bank of
England notes and began to spread them out on the bed.
“Twelve of them. Ten pounds each. All yours. We collected sixty
pounds in the office and the firm doubled it. And now you’re going to
eternal sunshine, to blue skies, to a land where people are merry
and sing the whole day long. You’ve escaped the slimy clutch of
commerce. Gad! I envy you!”
“Do you really, sir?”
“Yes. I wanted to live in Italy, Greece, Spain; to roam, to be a
vagabond, to be free. But I married, had children, became a slave
chained to the oar. One thing though,—my boys will never be square
pegs in round holes. They’ll have the chance I never had.”
“Perhaps it’s not too late.”
“No, perhaps not. Perhaps some day I’ll join you down there.
Perhaps when I get things settled, I’ll live under those careless skies
where living is rapture. I’ll get back by own soul. I’ll write that book,
I’ve tried all my life to write. Perhaps ... it’s my dream, my dream....”
Mr. Ainger turned abruptly and went out, leaving Hugh staring
incredulously at the counterpane of notes that covered his bed.
CHAPTER TWO
THE CALL OF THE BLOOD
1.
PINES packed the vast valley, climbing raggedly to the pale grey
peaks. Sometimes the mountains swooped down in gulch and butte
of fantastic beauty. The pines were pale green in the sunshine, the
soil strangely red. There was a curious dryness, a hard brilliance
about it all.
As Hugh looked from the train window he had a feeling of home-
coming. It was as if his ancestors had lived in this land; as if in no
other could he thrive so well.
“I’m feeling heaps better,” he thought. “Only let me get six months in
these jolly old pine forests, living like a wood-cutter. The life of
nature, that’s what I need to make a new man of me. Ah! this is my
country. I’m here now; and here I’ll stay.”
Looking at that sky so invincibly blue, that soil so subjugated by the
sun, it seemed hard to believe that elsewhere there could be fog and
cold and sleet. Here the sunshine was of so conquering a quality, it
was difficult to think of sullen lands that could resist it.
Again Hugh felt that sense of familiarity: “I’m a son of the sun,” he
exulted; “a child of the sun-land.”
So absorbed was he that a rasping voice at his side almost startled
him.
“The verdure here is profligate, ain’t it?”
The speaker was a rusty, creaky man smoking a rank cigar. He had
a bony nose, and a ragged moustache. He wore a dusty bowler hat
and a coat with a collar of hard-bitten musk-rat.
“The pines do seem to thrive,” said Hugh.
“Pines is very tendatious,” observed the shabby man. “Very
saloobrious too.”
“Indeed,” said Hugh. “Are you a health-seeker?”
“No, sir. Not ’ealth,—wealth. I’m a man with a system, I am. The
finest system on the Riviera.”
“I wish mine was. It’s rather dicky.”
“Oh, I wasn’t referrin’ to my corporationus system. It’s my system at
roulette. Allow me....”
He handed Hugh a rather soiled card on which was engraved: