Professional Documents
Culture Documents
Continuous Functions Volume 2 Jacques Simon Full Chapter PDF
Continuous Functions Volume 2 Jacques Simon Full Chapter PDF
Continuous Functions Volume 2 Jacques Simon Full Chapter PDF
Jacques Simon
Visit to download the full and correct content document:
https://ebookmass.com/product/continuous-functions-volume-2-jacques-simon/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
https://ebookmass.com/product/distributions-volume-3-1st-edition-
jacques-simon/
https://ebookmass.com/product/the-new-academic-2-edition-simon-
clews/
https://ebookmass.com/product/oceans-evolving-concepts-guy-
jacques/
https://ebookmass.com/product/core-java-volume-1-fundamental-
volume-2-twelfth-edition-cay/
Spatial Gems: Volume 2 John Krumm
https://ebookmass.com/product/spatial-gems-volume-2-john-krumm/
https://ebookmass.com/product/continuous-emission-monitoring-3rd-
edition-james-a-jahnke-2/
https://ebookmass.com/product/continuous-emission-monitoring-3rd-
edition-james-a-jahnke/
https://ebookmass.com/product/continuous-integration-ci-and-
continuous-delivery-cd-a-practical-guide-to-designing-and-
developing-pipelines-1st-edition-henry-van-merode/
https://ebookmass.com/product/the-parody-exception-in-copyright-
law-sabine-jacques/
Continuous Functions
To Claire and Patricia,
By your gaiety, “joie de vivre”, and femininity,
you have embellished my life,
and you have allowed me to conserve the tenacity
needed for this endeavor
Analysis for PDEs Set
coordinated by
Jacques Blum
Volume 2
Continuous Functions
Jacques Simon
First published 2020 in Great Britain and the United States by ISTE Ltd and John Wiley & Sons, Inc.
Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms and licenses issued by the
CLA. Enquiries concerning reproduction outside these terms should be sent to the publishers at the
undermentioned address:
www.iste.co.uk www.wiley.com
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
Introduction
Objective. This book is the second of six volumes in a series dedicated to the
mathematical tools for solving partial differential equations derived from physics:
Volume 3: Distributions;
Volume 5: Traces;
This second volume is devoted to the partial differentiation of functions and the
construction of primitives, which is its inverse mapping, and to their properties, which
will be useful for constructing distributions and studying partial differential equations
later.
Target audience. We intended to find simple methods that require a minimal level
of knowledge to make these tools accessible to the largest audience possible – PhD
candidates, advanced students1 and engineers – without losing generality and even
generalizing some standard results, which may be of interest to some researchers.
1 Students? What might I have answered if one of my MAS students in 1988 had asked for more details
about the de Rham duality theorem that I used to obtain the pressure in the Navier–Stokes equations?
Perhaps I could say that “Jacques-Louis L IONS, my supervisor, wrote that it follows from the de Rham
cohomology theorem, of which I understand neither the statement, nor the proof, nor why it implies the
result that we are using.” What a despicably unscientific appeal to authority!
This question was the starting point of this work: writing proofs that I can explain to my students for
every result that I use. It took me 5 years to find the “elementary” proof of the orthogonality theorem
x Continuous Functions
Originality. The construction of primitives, the Cauchy integral and the weighting
with which they are obtained are performed for a function taking values in a Neumann
space, that is, a space in which every Cauchy sequence converges.
We explicitly determine all primitives (Theorem 9.17) and construct one that
depends continuously on q (Theorem 9.18).
.
primitives of a field q. I needed a way to obtain Γ q d = 0
(Theorem 9.2, p. 194) on the existence of the
for every closed path Γ from the condition Ω q . ψ = 0 for every divergence-free ψ. It gave me the greatest
mathematical satisfaction I have ever experienced to explicitly construct an incompressible tubular flow (see
p. 184). Twenty-five years later, I am finally ready to answer any other questions from my (very persistent)
students.
Introduction xi
The properties established here for continuous functions will also be used to
extend them to integrable distributions in Volume 4, by continuity or transposition.
Indeed, one of the objectives of the Analysis for PDEs series is to extend integration
and Sobolev spaces to take values in Neumann spaces. However, it seemed more
straightforward to first construct distributions (in Volume 3) using just continuous
functions before introducing integrable distributions (in Volume 4), which play the
role usually fulfilled by classes of almost everywhere equal integrable functions.
Novelties. Many results are natural extensions of previous results, but the following
seemed most noteworthy:
— The construction of the topology of the space K(Ω; E) of continuous functions
with compact support using the semi-norms f K(Ω;E);q = supx∈Ω q(x)f (x)E;ν
indexed by q ∈ C + (Ω) and ν ∈ NE (Definition 1.17). This is equivalent to and much
simpler than the inductive limit topology of the CK (Ω; E).
— The fact that if a function f ∈ C(Ω) satisfies supx∈Ω q(x)|f (x)| < ∞ for every
q ∈ C + (Ω), then its support is compact (Theorem 1.22). This is the basis for defining
the semi-norms of D(Ω) in Volume 3.
— The concentration theorem for the integral and the construction of an
incompressible tubular flow (Theorems 8.18 and 8.17), which are key steps in our
construction of the primitives of a field taking values in a Neumann space, as it is
explained in the comment Utility of the concentration theorem, p. 186.
Prerequisites. The proofs in the main body of the text only use definitions and results
established in Volume 1, whose statements are recalled either in the text or in the
Appendix. Detailed proofs are given, including arguments that may seem trivial to
experienced readers, and the theorem numbers are systematically referenced.
Comments. Comments with a smaller font size than the main body of the text appeal to external results or
results that have not yet been established. The Appendix on Reminders is also written with a smaller font
size, since its contents are assumed to be familiar.
Historical notes. Wherever possible, the origin of the concepts and results is given
as a footnote2.
2 Appeal to the reader. Many important results lack historical notes because I am not familiar with their
origins. I hope that my readers will forgive me for these omissions and any injustices they may discover.
And I encourage the scholars among you to notify me of any improvements for future editions!
xii Continuous Functions
Olivier B ESSON, Fulbert M IGNOT, Nicolas D EPAUW, and Didier B RESCH also
provided many improvements, in form and in substance.
Pierre D REYFUSS gave me insight into the necessity of simply connected domains
for the existence of primitives with Poincaré’s condition, as explained on p. 209 in the
comment Is simple connectedness necessary for gluing together local primitives?
Joshua P EPPER spent much time discussing about the best way to adapt this work
in English.
Jacques S IMON
Chapdes-Beaufort
April 2020
Familiarization with Semi-normed Spaces
S PACES OF FUNCTIONS
B(Ω; E) space of uniformly continuous functions with bounded support . . . . . . 87
C(Ω; E) space of continuous functions . . . . . . . . . . . . . . . 3
Cb (Ω; E) space of bounded continuous functions . . . . . . . . . . . . 3
CK (Ω; E) space of continuous functions with support included in the compact set K ⊂ Ω . 6
C∇ (Ω; E d ) space of gradients of continuous functions . . . . . . . . . . . 205
C + (Ω) set of positive continuous real functions . . . . . . . . . . . . 14
C m (Ω; E) space of m times continuously differentiable functions, and the case m = ∞ 43, 44
Cbm (Ω; E) id. with bounded derivatives, and the case m = ∞ . . . . . . . . 43, 44
CKm (Ω; E) id. with support included in the compact set K ⊂ Ω, and the case m = ∞ . 43, 44
C m (Ω; E) space C m defined on the closure of a bounded open set . . . . . . . . 52
C m (Ω; U ) set of functions in C m taking values in the set U . . . . . . . . . . 54
C(Ω; E) space of uniformly continuous functions . . . . . . . . . . . . 4
Cb (Ω; E) space of bounded uniformly continuous functions . . . . . . . . . 4
CD (Ω; E) id. with support included in the compact subset D of Rd . . . . . . . 6
Cmb (Ω; E) space C m with uniformly continuous bounded derivatives, and the case m = ∞ 43, 44
K(Ω; E) space of continuous functions with compact support . . . . . . . . . 14
Km (Ω; E) id. m times continuously differentiable, and the case m = ∞ . . . . . 43, 44
O PERATIONS ON A FUNCTION f
f extension by 0E . . . . . . . . . . . . . . . . . . . 47
f˘ image under permutation of variables . . . . . . . . . . . . . 27
fˇ image under the symmetry x → −x of the variable . . . . . . . . . 151
f image under separation of variables . . . . . . . . . . . . . 23
τx f translation by x ∈ Rd . . . . . . . . . . . . . . . . . 77
Rn f global regularization . . . . . . . . . . . . . . . . . . 163
f μ function weighted by μ . . . . . . . . . . . . . . . . . 148
f ρn local regularization . . . . . . . . . . . . . . . . . . 157
f μ convolution with μ . . . . . . . . . . . . . . . . . . 149
f ⊗g tensor product with g . . . . . . . . . . . . . . . . . 129
f ◦T composition with T . . . . . . . . . . . . . . . . . . 69
xvi Continuous Functions
supp f support . . . . . . . . . . . . . . . . . . . . . 4
Lf or L ◦ f composition with the linear mapping L . . . . . . . . . . . . 55
D ERIVATIVES OF A FUNCTION f
f or df /dx derivative of a function of a single real variable . . . . . . . . . . 32
∂i f partial derivative: ∂i f = ∂f /∂xi . . . . . . . . . . . . . . 37
β
∂β f derivative of order β: ∂ β f = ∂1β1 . . . ∂d d f . . . . . . . . . . . 41
β positive multi-integer: β = (β1 , . . . , βd ), βi ≥ 0 . . . . . . . . . 37
|β| differentiability order: |β| = |β1 | + . . . |βd | . . . . . . . . . . . 37
∂0f derivative of order 0: ∂ 0 f = f . . . . . . . . . . . . . . . 37
∇f gradient: ∇f = (∂1 f, . . . , ∂d f ) . . . . . . . . . . . . . . 32
df differential . . . . . . . . . . . . . . . . . . . . 74
q field: q = (q1 , . . . , qd ) . . . . . . . . . . . . . . . . . 176
∇.q divergence: ∇ . q = ∂1 q1 + . . . ∂d qd . . . . . . . . . . . . . 183
∇−1 q primitive that depends continuously on q . . . . . . . . . . . . 211
q∗ explicit primitive: q ∗ (x) = Γ(a,x) q . d . . . . . . . . . . . 192
∂U boundary of U . . . . . . . . . . . . . . . . . . . 229
[u, v] closed segment: [u, v] = {tu + (1 − t)v : 0 ≤ t ≤ 1} . . . . . . . 34
L(E; F ) space of continuous linear mappings . . . . . . . . . . . . . 56
L (E1 × . . . × E ; F ) space of continuous multilinear mappings . . . . . . . . . 59
OTHER SETS
N∗ set of natural numbers: N∗ = {0, 1, 2, . . .} . . . . . . . . . . . 227
N set of non-zero natural numbers: N = {1, 2, . . .} . . . . . . . . . 227
Z set of integers: Z = {. . . , −2, −1, 0, 1, 2, . . .} . . . . . . . . . . 227
Q set of rational numbers . . . . . . . . . . . . . . . . . 227
R space of real numbers . . . . . . . . . . . . . . . . . 228
m, n integer interval: m, n = {i ∈ N∗ : m ≤ i ≤ n} . . . . . . . . . 227
m, ∞ extended integer interval: m, ∞ = {i ∈ N∗ : i ≥ m} ∪ {∞} . . . . . 227
(a, b) open interval: (a, b) = {x ∈ R : a < x < b} . . . . . . . . . . 228
[a, b] closed interval: [a, b] = {x ∈ R : a ≤ x ≤ b} . . . . . . . . . . 228
compact inclusion in Rd . . . . . . . . . . . . . . . . 80
⊂ algebraic inclusion
\ set difference: U \ V = {u ∈ U : u ∈ / V}
× product: U × V = {(u, v) : u ∈ U, v ∈ V }
∅ empty set
S PECIAL FUNCTIONS
det determinant . . . . . . . . . . . . . . . . . . . . 112
e exponential number . . . . . . . . . . . . . . . . . . 238
xviii Continuous Functions
T YPOGRAPHY
end of statement
end of proof or remark
Chapter 1
This chapter is dedicated to the properties of spaces of continuous functions taking values in a semi-
normed space E that we will need later. Definitions of the space C(Ω; E) of continuous functions, the
space C(Ω; E) of uniformly continuous functions and variants of these spaces are given in § 1.2. We
then compare these spaces (§ 1.3) and study their completions (§ 1.4) and metrizability properties (§ 1.5).
The space K(Ω; E) of functions with compact support is investigated in § 1.6. We also study continuous
extensions (§ 1.8), separation of variables (§ 1.9) and sequential compactness (§ 1.10) in these spaces.
These topics are more necessary than original, with the exception of our construction of the topology
of K(Ω; E) using the semi-norms f K(Ω;E);q = supx∈Ω q(x)f (x)E;ν indexed by q ∈ C + (Ω) and
ν ∈ NE . These semi-norms yield properties that are usually obtained using the inductive limit topology
of the CK (Ω; E). A useful tool for this, which is also new, is Theorem 1.23: if a family F of functions in
C(Ω) satisfies supf ∈F supx∈Ω q(x)|f (x)| < ∞ for every q ∈ C + (Ω), their supports are all included in
the same compact set.
1 History of the notion of semi-normed space. John VON N EUMANN introduced semi-normed spaces
in 1935 [59] (with a superfluous countability condition). He also showed [59, Theorem 26, p. 19] that
they coincide with the locally convex topological vector spaces that Andrey KOLMOGOROV had previously
introduced in 1934 [49, p. 29].
2 Continuous Functions
Any such space is said to be separated if u = 0E is the only element such that
uE;ν = 0 for every ν ∈ NE .
Caution. Definition 1.1 is general but not universal. For Laurent S CHWARTZ [67, p. 240], a semi-normed
space is a space endowed with a filtering family of semi-norms (Definition 2.21). This definition is
equivalent, since every family is equivalent to a filtering family [Vol. 1, Theorem 3.15]. For Nicolas
B OURBAKI [12, editions published after 1981, Chapter III, p. III.1] and Robert E DWARDS [32, p. 80], a
semi-normed space is a space endowed with a single semi-norm, which drastically changes the
meaning.
Let us define various notions relating to the continuity2 of a function taking values
in a semi-normed space. These are special cases of Definition 1.24 of a continuous
mapping from a semi-normed space into another.
(a) We say that f is continuous at the point x of Ω if, for every ν ∈ NE and > 0,
there exists η > 0 such that, if y ∈ Ω and |y − x| ≤ η, then
f (y) − f (x)E;ν ≤ .
(b) We say that f is uniformly continuous if, for every ν ∈ NE and > 0, there
exists η > 0 such that, if x and y belong to Ω and |y − x| ≤ η, then
f (y) − f (x)E;ν ≤ .
(c) We say that f is sequentially continuous at the point x of Ω if, for every sequence
(xn )n∈N in Ω such that xn → x in Rd , we have f (xn ) → f (x) in E.
2 History of the notion of continuous mapping. Augustin C AUCHY defined sequential continuity for a
real function on an interval in 1821 in [20]. Bernard Placidus Johann Nepomuk B OLZANO also contributed
to the emergence of this notion.
History of the notion of uniformly continuous function. Eduard H EINE defined the notion of uniform
continuity of functions on (a subset of) Rd in 1870 in [46]. This notion had previously been implicitly used
by Augustin C AUCHY in 1823 to define the integral of a real function [21, pp. 122–126] and was later
explicitly used by Peter D IRICHLET.
Spaces of Continuous Functions 3
(d) We say that f is bounded if its image f (Ω) = {f (x) : x ∈ Ω} is a bounded set
(of E), or in other words, if, for every ν ∈ NE ,
1.2. Spaces C(Ω; E), Cb (Ω; E), CK (Ω; E), C(Ω; E) and Cb (Ω; E)
(a) We denote by C(Ω; E) the vector space of continuous functions from Ω into E
endowed with the semi-norms and indexed by the compact sets K ⊂ Ω and ν ∈ NE ,
x∈K
(b) We denote by Cb (Ω; E) the vector space of continuous and bounded functions
from Ω into E endowed with the semi-norms and indexed by ν ∈ NE ,
x∈Ω
3 Theorem A.29. Theorems with numbers in the form A.n can be found in the Appendix.
4 History of the notion of function space. Bernhard R IEMANN introduced the concept of (infinite-
dimensional) function space in 1892 in his inaugural lecture On the Hypotheses Which Lie at the Bases
of Geometry [64, p. 276] (see the extract cited in [14, p. 176]).
4 Continuous Functions
In (a), the mapping f → supx∈K f (x)E;ν is a semi-norm on C(Ω; E), since the
upper envelope of a family of semi-norms is a semi-norm whenever it is everywhere
finite (Theorem A.6). This is indeed the case here, since, for every x, the mapping
f → f (x)E;ν is a semi-norm on C(Ω; E) and, for each f , supx∈K f (x)E;ν < ∞
because continuous functions are bounded on compact sets (Theorem A.34).
These spaces are written as C(Ω) and Cb (Ω), respectively, in the case where
E = R.
The topology with which we have endowed C(Ω; E) is said to be the topology of
uniform convergence on compact sets, and the topology of Cb (Ω; E) is said to be
the topology of uniform convergence.
(a) We denote by C(Ω; E) the vector space of uniformly continuous functions from Ω
into E.
(b) We denote by Cb (Ω; E) the vector space of uniformly continuous and bounded
functions from Ω into E endowed with the semi-norms of Cb (Ω; E).
We denote these spaces by C(Ω) and Cb (Ω), respectively, in the case where
E = R.
Absence of a topology on C(Ω; E). We shall not endow C(Ω; E) with semi-norms, since we do not
require them.
Next, let us define the support of a function (recall that denotes the closure).
supp f = {x ∈ Ω : f (x) = 0E } ∩ Ω.
def
Spaces of Continuous Functions 5
Let us state some properties of the support of a function defined on an open set5.
(a) Then supp f = Ω\O, where O is the largest open subset of Rd on which f = 0E ;
in other words, O is the interior of {x ∈ Ω : f (x) = 0E }.
5 Numbering of statements. The same numbering scheme is used for all statements – Definitions 1.1–1.5,
Theorem 1.6, Definition 1.7, etc. –, to make it easier to find a given result by number. For example, the
reader will struggle to find Theorems 1.1–1.5, because these numbers were assigned to definitions.
6 Continuous Functions
Justification. (a) The vector space CK (Ω; E) can be endowed with the semi-norms of
Cb (Ω; E) because every function f in this space is bounded, since f is zero outside
of K and bounded on K, noting that continuous functions are bounded on compact
sets (Theorem A.34).
(b) The vector space CD (Ω; E) can be endowed with the semi-norms of Cb (Ω; E)
because every function f in this space is bounded, since f is zero outside of D
and bounded on Ω ∩ D. Indeed, since D is precompact (Theorem A.19 (a)), the
subset Ω ∩ D is also precompact (Theorem A.20), and so is its image f (Ω ∩ D)
(Theorem A.33), which is therefore bounded (Theorem A.19 (a)).
We denote these spaces by CK (Ω) and CD (Ω), respectively, in the case where
E = R.
Caution about behavior at the boundary. When Ω is an open set, functions in CK (Ω; E) must be zero
on some neighborhood of the boundary ∂Ω, since K is included in Ω.
By contrast, functions in CD (Ω; E) do not necessarily vanish on a neighborhood of ∂Ω (regardless of
whether Ω is open, unless it is the whole of Rd ) when D is not included in Ω. To highlight this difference,
we have chosen notation that differentiates between arbitrary compact sets D and compact sets K that are
included in Ω.
Utility of the spaces CK (Ω; E) and CD (Ω; E). The space CK (Ω; E), or, more precisely, its
generalization CKm (Ω; E), will be useful for our study of distributions in Volume 3, since the space D(Ω)
∞ (Ω).
of test functions is the union of the CK
The space CD (Ω; E) will be useful for our study of the Cauchy integral of continuous functions (for
example in Theorem 4.22), since B(Ω; E) (Definition 4.7) is the union of the CD (Ω; E).
We say that the two families are equivalent if each family dominates the other. We also
say that they generate the same topology.
Terminology. The topology of E is the set of its open sets. We can say that two families of semi-norms
generate the same topology instead of saying that they are equivalent because equivalence of two families
of semi-norms implies equality of their open sets [Vol. 1, Theorem 3.4] and the converse also holds [Vol. 1,
Theorem 7.14 (a) and 8.2 (a), with L = T = Identity].
= F , if E = F , their
(b) We say that E is topologically equal to F , written E ↔
additions and multiplications coincide, and their families of semi-norms are
equivalent.
Let us now show that spaces of continuous functions are separated and compare
them.
(a) C(Ω; E), Cb (Ω; E), CK (Ω; E), Cb (Ω; E) and CD (Ω; E) are separated semi-
normed spaces.
(b) CK (Ω; E), Cb (Ω; E) and CD (Ω; E) are closed and hence sequentially closed
topological subspaces of Cb (Ω; E).
(c) ≡ CK (Ω; E) →
CK (Ω; E) ↔ ⊂ Cb (Ω; E) →
⊂ Cb (Ω; E) →
⊂ C(Ω; E).
(d) If Ω is bounded,
Cb (Ω; E) = C(Ω; E).
(e) If Ω is compact,
≡ Cb (Ω; E) ↔
Cb (Ω; E) ↔ = C(Ω; E).
Proof. (a) These spaces are semi-normed spaces (Definition 1.1) by construction.
(b) The vector spaces CK (Ω; E) and CD (Ω; E) are included in Cb (Ω; E), as we saw
earlier in the justification of Definition 1.7, and so is Cb (Ω; E). These spaces are
topological subspaces (Definition 1.9 (d)) of Cb (Ω; E) because they are endowed with
its semi-norms. Let us check that they are closed.
Closedness of CK (Ω; E) in Cb (Ω; E). Let us show that the complement is open. Thus,
let f ∈ Cb (Ω; E) \ CK (Ω; E). There exists x ∈ Ω \ K such that f (x) = 0E and hence
ν ∈ NE such that f (x)E;ν = a > 0. Therefore, g − f Cb (Ω;E);ν ≤ a/2 implies
g(x) > 0, and so g ∈ Cb (Ω; E) \ CK (Ω; E), showing that the complement is open.
Closedness of Cb (Ω; E) in Cb (Ω; E). Let us show that the complement is open. Thus,
let f ∈ Cb (Ω; E) \ Cb (Ω; E). There exists ν ∈ NE , a > 0 and, for all n ∈ N, xn
and yn in Ω such that |xn − yn | ≤ 1/n and f (xn ) − f (yn )E;ν ≥ a. Therefore,
g − f Cb (Ω;E);ν ≤ a/3 implies g(xn ) − g(yn )E;ν ≥ a/3, and so g ∈ Cb (Ω; E) \
Cb (Ω; E), showing that the complement is open.
(c) Identity CK (Ω; E) ↔≡ CK (Ω; E). Algebraic equality is satisfied because every
function in CK (Ω; E) is uniformly continuous on the compact set K by Heine’s
theorem (Theorem A.34) and therefore on the whole of Ω (since it vanishes on ∂K).
Topological identity follows because these two spaces are both endowed with the
semi-norms of Cb (Ω; E).
Inclusion CK (Ω; E) →
⊂ Cb (Ω; E). Algebraic inclusion follows from the fact that every
function in CK (Ω; E) is bounded on the compact set K (Heine’s theorem again) and
is therefore bounded on the whole of Ω. Topological inclusion follows because both
spaces are endowed with the semi-norms of Cb (Ω; E).
Inclusion Cb (Ω; E) →
⊂ Cb (Ω; E). This topological inclusion follows from the fact that
every uniformly continuous function is continuous, noting that Cb (Ω; E) is endowed
with the semi-norms of Cb (Ω; E).
Inclusion Cb (Ω; E) →
⊂ C(Ω; E). This topological inclusion follows from the fact that,
by Definition 1.3 of the semi-norms of C(Ω; E) and Cb (Ω; E), for every f ∈ Cb (Ω; E)
and ν ∈ NE and every compact set K ⊂ Ω,
(d) Case where Ω is bounded. In this case, Cb (Ω; E) = C(Ω; E) because, given
that Ω is precompact (Theorem A.23 (b)), uniformly continuous functions on Ω are
bounded. Indeed, their images are precompact (Theorem A.33) and hence bounded
(Theorem A.19 (a)).
(e) Case where Ω is compact. Equality Cb (Ω; E) = Cb (Ω; E) = C(Ω; E). This
algebraic equality follows from the fact that every continuous function on a compact
set is uniformly continuous and bounded on this set (Heine’s theorem once again,
Theorem A.34).
Equality Cb (Ω; E) ↔
= C(Ω; E). Here, we can take K = Ω in Definition 1.3 (a), which
gives f Cb (Ω;E);ν = f C(Ω;E);Ω,ν and hence C(Ω; E) → ⊂ Cb (Ω; E) (because
Cb (Ω; E) is endowed with the semi-norms of Cb (Ω; E)). The topological equality
follows because the inverse inclusion holds by (c).
Non-closedness of C(Ω) in C(Ω). If Ω is an open set (or more generally a set that is not compact),
C(Ω) is not (sequentially) closed in C(Ω). (1.2)
For example, the functions fn defined on (0, 1) by fn (x) = inf{n, 1/x} are uniformly continuous and,
as n → ∞, they converge in C((0, 1)) to the function f (x) = 1/x, which is not uniformly continuous.
10 Continuous Functions
Terminology. We named these spaces Neumann spaces in Volume 1 in honor of John VON N EUMANN,
who introduced them in 1935 [59]. Thus, readers will need to recall this definition before using it elsewhere.
If E is a Neumann space, then C(Ω; E), Cb (Ω; E), CK (Ω; E), Cb (Ω; E) and
CD (Ω; E) are also Neumann spaces.
Sequential completeness of C(Ω; E). Let (fn )n∈N be a Cauchy sequence in C(Ω; E).
By Definition 1.3 (a) of the semi-norms of C(Ω; E), for every compact set K ⊂ Ω,
every ν ∈ NE and > 0, there exists ∈ N such that
sup sup fn (x) − f (x)E;ν = sup fn − f C(Ω;E);K ν ≤ .
n≥ x∈K n≥
6 History of Theorem 1.12. In a series of unpublished but nonetheless hugely influential lectures, Karl
W EIERSTRASS showed that the limit of a sequence of sequentially continuous real functions that converge
uniformly is itself sequentially continuous. In other words, he showed that an inequality of type (1.3) implies
the sequential continuity of f . This is the key point in the proof of the sequential completeness of C(R).
Spaces of Continuous Functions 11
Sequential completeness of CK (Ω; E), Cb (Ω; E) and CD (Ω; E). Since every closed
topological subspace of a sequentially complete space is sequentially complete
(Theorem A.27), the subspaces CK (Ω; E), Cb (Ω; E) and CD (Ω; E) of Cb (Ω; E) are
sequentially complete (they are closed by Theorem 1.10 (b)).
Another proof of the sequential completeness of Cb (Ω; E). The proof above can be revisited with Ω
instead of K to show that the limit f thus obtained is uniformly continuous by using two sequences
(xm )m∈N and (ym )m∈N in Ω such that |xm − ym | ≤ 1/m instead of xm → x.
7 History of the notion of metrizable space. Maurice F RÉCHET gave the general definition of a metric
space in 1906 in [38].
12 Continuous Functions
Justification of the term “metrizable”. We speak of a metrizable space because any countable family, or
equivalently any sequence ( k )k∈N of semi-norms, may be associated with a metric d, that generates
the same topology, given by
u − vk
2−k
def
d(u, v) = .
k∈N
1 + u − vk
Strictly speaking, Definition 1.13 defines a separated countably semi-normable space. We prefer to speak
instead of a metrizable space by abuse of language because this equivalent notion is more familiar. A
metrizable space is, more precisely, a space that is “topologically equal to a metric space.”
(a) If E is metrizable, then Cb (Ω; E), CK (Ω; E), Cb (Ω; E) and CD (Ω; E) are
metrizable.
(c) If E is a normed space, then Cb (Ω; E), CK (Ω; E), Cb (Ω; E) and CD (Ω; E) are
normed spaces.
(c) If E is a normed space, the family of semi-norms of Cb (Ω; E) reduces to the norm
f Cb (Ω;E) = supx∈Ω f (x)E . Thus, by definition, the spaces CK (Ω; E), Cb (Ω; E)
and CD (Ω; E) are endowed with this norm as well.
8 History of Komogorov’s theorem. Andrey KOLMOGOROV showed in 1934 [49, p. 33] that a topological
vector space is normable if and only if there exists a bounded convex neighborhood of the origin, which
here is equivalent to the existence of a bounded open set.
Spaces of Continuous Functions 13
where B(x, 1/n) denotes the closed ball {y ∈ Rd : |y −x| ≤ 1/n}. As we shall check
in Lemma 1.15, every compact set K ⊂ Ω is included in one of the compact sets
Kn = Ωn1/n , so the family of semi-norms of C(Ω; E) (Definition 1.3 (a)) is equivalent
to the sub-family associated with the Kn , which is indexed by N × NE . If, moreover,
E is metrizable, by Definition 1.13, we can choose NE to be countable, in which case
N × NE is also countable (Theorem A.2), and so C(Ω; E) is metrizable.
Then, Ωr and Ωnr are open, the closure Ωnr is a compact set included in Ω, the
(Ωn1/n )n∈N forms an increasing open covering of Ω and every compact set K included
in Ω is included in one of the Ωn1/n .
The intersection Ωnr with the open ball {x ∈ Rd : |x| < n} is therefore also open
(Theorem A.10).
Finally, the sets Ωn1/n increase with n and form a covering of Ω. Every compact set
K ⊂ Ω is therefore, by Definition A.17 (a), included in one of these sets, and, finally,
in one of their closures Ωn1/n .
(a) If E is a Fréchet space, then Cb (Ω; E), CK (Ω; E), Cb (Ω; E) and CD (Ω; E) are
Fréchet spaces.
(c) If E is a Banach space, then Cb (Ω; E), CK (Ω; E), Cb (Ω; E) and CD (Ω; E) are
Banach spaces.
In particular:
Proof. (a), (b) and (c). This follows from the sequential completeness properties in
Theorem 1.12 and the metrizability and normability properties in Theorem 1.14.
(e) and (d). This follows from (a), (b) and (c), since R is a Banach space
(Theorem A.23 (a)) and hence a Fréchet space.
Metrizability of C(Ω). The properties of Theorems 1.14 (b) and 1.16 (c) and (d) still hold when:
Ω is the union of a sequence (Kn )n∈N of compact subsets of Rd such that,
for any compact set K ⊂ Ω, there exists n such that K ⊂ Kn .
This property is, for example, satisfied whenever Ω is open or closed, but not for arbitrary subsets of Rd .
Let us define the space of continuous functions with compact support9. The support
of a function was defined on page 4. We denote
We denote by K(Ω; E) the vector space of continuous functions from Ω into E with
compact support endowed with the semi-norms, indexed by q ∈ C + (Ω) and ν ∈ NE ,
x∈Ω
9 History of the semi-norms of K(Ω; E). The semi-norms in Definition 1.17, which are new to the best of
the author’s knowledge, provide a very simple way [S IMON, 75, forthcoming] of generating the inductive
limit topology of the CK (Ω; E) with which B OURBAKI endowed K(Ω; E) [11, Chapter III, § 1, p. 41].
Spaces of Continuous Functions 15
Justification. The continuous functions with compact support form a vector subspace
of C(Ω; E) because the support of the sum of two functions is included in the union
of their supports, which is compact if the supports of both functions are compact.
The mapping f → supx∈Ω q(x)f (x)E;ν is a semi-norm on K(Ω; E), since the
upper envelope of a set of semi-norms is itself a semi-norm whenever it is
everywhere finite (Theorem A.6). This is indeed the case here, since, on the one
hand, for every x, the mapping f → q(x)f (x)E;ν is a semi-norm on K(Ω; E) and,
on the other hand, for each f , supx∈Ω q(x)f (x)E;ν is finite, since q and f E;ν
are bounded on the support of f (because continuous functions are bounded on
compact sets by Theorem A.34).
Notation K(Ω; E). This notation was used by B OURBAKI [11, Chapter III, § 1, p. 40].
Benefit of the semi-norms of K(Ω; E). The semi-norms newly introduced in Definition 1.17 yield the
properties usually obtained from the inductive limit topology of the CK (Ω; E). In particular, the functions
in a bounded subset of K(Ω; E) all have supports included in the same compact set K included in Ω
(Theorem 1.23, which is also new). These semi-norms generate the inductive limit topology [S IMON, 75,
forthcoming] and are simpler.
Motivation of the study of K(Ω; E). We study the space K(Ω; E) because we will frequently need
continuous approximations with compact support of functions or distributions. More importantly, the
properties of K(Ω) stated in Theorems 1.22 and 1.23 are the basis of those of the space D(Ω) that will be
used to construct the space D (Ω; E) of distributions in Volume 3.
{x ∈ Ω : f (x) = 0E } ⊂ K = K ⊂ Ω,
CK (Ω; E) →
⊂ K(Ω; E) →
⊂ Cb (Ω; E).
Inclusion CK (Ω; E) →⊂ K(Ω; E). Let f ∈ CK (Ω; E). Its support is included in the
compact set K and hence is compact (Theorem 1.18), so f ∈ K(Ω; E). Furthermore,
for every q ∈ C + (Ω) and ν ∈ NE , we have, by Definitions 1.17 and 1.3,
f K(Ω;E);q,ν = sup q(x)f (x)E;ν ≤ sup q(x) sup f (x)E;ν = cq f Cb (Ω;E);ν ,
x∈Ω x∈K x∈Ω
where cq = supx∈K q(x), which is finite by Heine’s theorem (Theorem A.34). This
gives us the topological inclusion (Definition 1.9 (c)), since, by Definition 1.7 (a),
CK (Ω; E) is endowed with the semi-norms of Cb (Ω; E).
Inclusion K(Ω; E) →⊂ Cb (Ω; E). Let f ∈ K(Ω; E). The function f is continuous with
compact support, so it is uniformly continuous and bounded on its support, again
by Heine’s theorem, and hence bounded everywhere. Thus, f ∈ Cb (Ω; E). The
topological inclusion follows from the fact that, by Definition 1.4 (b), Cb (Ω; E) is
endowed with the semi-norms of Cb (Ω; E) and, for every ν ∈ NE ,
Let us observe that the topologies of C(Ω; E), Cb (Ω; E) and K(Ω; E) coincide on
CK (Ω; E).
Spaces of Continuous Functions 17
Then the topologies of C(Ω; E), Cb (Ω; E), CK (Ω; E) and K(Ω; E) coincide on
CK (Ω; E).
Proof. Let E be a Banach space. Observe that for all f ∈ K(Ω; E) and q ∈ C + (Ω),
the function qf belongs to K(Ω; E) and hence to Cb (Ω; E) by Theorem 1.19.
Furthermore, by the definition of their semi-norms (Definitions 1.17 and 1.3 (b)),
(the index ν is absent because E only has one semi-norm, namely its norm).
Now, let (fn )n∈N be a Cauchy sequence in K(Ω; E). By (1.4), this sequence is
Cauchy in Cb (Ω; E), which is sequentially complete (Theorem 1.12), so it has a limit f
in this space. By (1.5), for every q ∈ C + (Ω), the sequence (qfn )n∈N is Cauchy in
Cb (Ω; E) and therefore has a limit in this space, which must necessarily be qf . Hence,
qf ∈ Cb (Ω; E) and so, again by (1.5),
This implies that the support of the function f E is compact by Theorem 1.22,
which we will prove below. Indeed, this function f E is continuous, since
18 Continuous Functions
f (y)E − f (x)E ≤ f (y) − f (x)E (because norms are contractions, see
Theorem A.5). Since the supports of f and f E coincide, it follows that
f ∈ K(Ω; E). Therefore, once again by (1.5),
fn − f K(Ω;E);q = q(fn − f )Cb (Ω;E) → 0.
Hence, fn → f in K(Ω; E), which is therefore sequentially complete, i.e. a Neumann
space.
This is, in particular, true for K(Ω), since R is a Banach space (Theorem A.23 (a)).
It remains to be checked that the property (1.6) implies the compactness of the
support of f E . More precisely, we need to establish the following property10.
T HEOREM 1.22.– Let f ∈ C(Ω), where Ω ⊂ Rd , such that, for every q ∈ C + (Ω),
sup q(x) |f (x)| < ∞.
x∈Ω
Proof. Let
K = {x ∈ Ω : f (x) = 0}.
def
10 History of Theorem 1.22. This property, newly introduced here, is equivalent to a special case of a
result by Nicolas B OURBAKI relating to the inductive limit topology of the CK (Ω; E) which is specified in
footnote 11.
Spaces of Continuous Functions 19
q(zm ) |f (zm )| = m.
This contradicts the hypothesis that qf is bounded. Therefore, (1.7) is satisfied, that
is, K is included in Ω.
supp f = K ∩ Ω = K.
Let us generalize this result to a family of functions that will be useful later11.
T HEOREM 1.23.– Let F ⊂ C(Ω), where Ω ⊂ Rd , such that, for every q ∈ C + (Ω),
Then all the functions of F have their support included in the same compact subset
of Ω.
Proof. Let
K = ∪f ∈F supp f .
def
K ⊂ Ω.
11 History of Theorem 1.23. This new property is equivalent to a result by B OURBAKI [11, Chapter III,
§ 1, Proposition 2, (ii), p. 42]: the support of every function in a bounded subset of K(Ω; E) endowed with
the inductive limit topology of the CK (Ω; E) is included in some compact set K ⊂ Ω. The proof given by
B OURBAKI, which uses topological arguments (infinite direct topological sum and strict inductive limits),
is completely different from the proof given here.
20 Continuous Functions
Suppose that K is not included in Ω. Then, since in Rd the closure coincides with the
set of limits of convergent sequences (Theorem A.25 (b)), there exists z ∈ K \ Ω and
a sequence (ym )m∈N converging to z such that, for every m ∈ N, there exists fm ∈ F
such that ym belongs to the support of fm . By Definition 1.5 of the latter,
ym ∈ {x ∈ Ω : fm (x) = 0}.
Again by Theorem A.25 (b), there exists zm ∈ Ω such that |zm − ym | ≤ 1/m and
fm (zm ) = 0.
This contradicts the hypothesis that the qf are bounded as a whole. Thus, K is indeed
included in Ω.
Non-metrizability of K(Ω). If Ω is non-empty, then K(Ω) is not metrizable [S IMON, 75], despite being
the union of the normed spaces CKn (Ω), where the Kn are compact sets whose union is Ω. This
illustrates that the countable union of normed spaces is not necessarily metrizable. Similarly, K(Ω; E) is
not metrizable if E = {0}.
Characterization of K(Ω; E) when E is a normed space. Theorem 1.22 implies the following property
(by applying it to f E and using equality (1.5)). If E is a normed space:
f ∈ K(Ω; E) ⇔ f ∈ C(Ω; E) and qf is bounded for every q ∈ C + (Ω).
T HEOREM 1.25.– Let E and F be two separated semi-normed spaces with families
of semi-norms { E;ν : ν ∈ NE } and { F ;μ : μ ∈ NF }.
A linear mapping L from E into F is continuous if and only if, for every μ ∈ NF ,
there exists a finite subset N of NE and c ≥ 0 such that: for every u ∈ E,
which gives the desired inequality with c = 1/η. This still holds if supν∈N uE;ν is
zero, in which case L(u)F ;μ = 0, otherwise v = 2u/L(u)F ;μ would not satisfy
the derived property for v.
Proof. (a) Since, by Definition 1.4 (b), Cb (Ω; E) is endowed with the semi-norms of
Cb (Ω; E) (Definition 1.3 (b)),
f (x)E;ν ≥ a − .
Since f is uniformly continuous, by Definition 1.2 (b), there exists η > 0 such that,
for every y ∈ Ω satisfying |y − x| ≤ η,
But by the characterization of the closure Ω from Theorem A.11, there exists y ∈ Ω
such that |y − x| ≤ η. Therefore, (1.10) implies that
This holds for every > 0, which gives the reverse inequality in (1.9), which must
therefore be an equality.
Spaces of Continuous Functions 23
Isomorphism from Cb (Ω; E) onto Cb (Ω; E). Every function g in Cb (Ω; E) has an
extension g that is uniformly continuous on Ω (Theorem A.38, which may be applied
because Ω is sequentially dense in Ω by Theorem A.25 (b)). This function g is bounded
because the inequality g(x)E;ν ≤ bν in Ω may be continuously extended to Ω
(Theorem A.37 (a) and (b)). Therefore, g ∈ Cb (Ω; E) and g |Ω = g. Thus, the
mapping f → f |Ω is surjective. The equality (a) then shows that it is bijective and
bicontinuous.
Isomorphism from CD (Ω; E) onto CD (Ω; E). This isomorphism follows from the
previous one because, when D is closed, as is the case here, the support of g is included
in D if and only if the support of g is included in D (by Definition 1.5 of the support).
Identification. If we identify f with its restriction in Theorem 1.26, we obtain the following topological
equality:
If E is a Neumann space, Cb (Ω; E) ↔ ≈ Cb (Ω; E). (1.11)
Some care is required with identifications, as it is explained in the section Dangerous identifications of
Volume 1 [73, § 14.6, pp. 240–243].
(a) The mapping f → f is an isomorphism from C(Ω1 ×Ω2 ; E) onto C(Ω1 ; C(Ω2 ; E))
and from Cb (Ω1 × Ω2 ; E) onto Cb (Ω1 ; Cb (Ω2 ; E)).
(b) The mapping f → f is linear, continuous and injective from Cb (Ω1 ×Ω2 ; E) into
Cb (Ω1 ; Cb (Ω2 ; E)).
(c) For any two compact subsets D1 of Rd1 and D2 of Rd2 , the mapping f → f is
an isomorphism from CD1 ×D2 (Ω1 × Ω2 ; E) onto CD1 (Ω1 ; CD2 (Ω2 ; E)).
24 Continuous Functions
Let Q1 = {xn1 }n∈N ∪ {x1 }. The set Q1 × K2 is compact in Rd1 +d2 by the Borel–
Lebesgue theorem (Theorem A.23 (b)) because it is closed and bounded, so f is
uniformly continuous on this set by Heine’s theorem (Theorem A.34). Therefore, for
every > 0, there exists η > 0 such that x2 ∈ K2 and |(xn1 , x2 ) − (x1 , x2 )| ≤ η
together imply that f (xn1 , x2 ) − f (x1 , x2 )E;ν ≤ .
Regrouping the variables. Let h ∈ C(Ω1 ; C(Ω2 ; E)) and denote by f the function
obtained by regrouping its variables. Let ((xn1 , xn2 ))n∈N be a sequence in Ω1 × Ω2
converging to a point (x1 , x2 ) in Ω1 × Ω2 . The set Q2 = {xn2 }n∈N ∪ {x2 } is compact
and included in Ω2 . Therefore, by (1.12) (since h = f ), for every ν ∈ NE ,
f (xn1 , xn2 ) − f (x1 , xn2 )E;ν ≤ h(xn1 ) − h(x1 )C(Ω2 ;E);Q2 ,ν → 0,
since h is continuous and hence sequentially continuous (Theorem A.29 again).
Moreover,
f (x1 , xn2 ) − f (x1 , x2 )E;ν = (h(x1 ))(xn2 ) − (h(x1 ))(x2 )E;ν → 0,
since h(x1 ) ∈ C(Ω2 ; E). These two properties imply that
f (xn1 , xn2 ) − f (x1 , x2 )E;ν → 0.
This shows that f is sequentially continuous and hence continuous (Theorem A.29
again, since Rd1 +d2 is metrizable). In other words, f ∈ C(Ω1 × Ω2 ; E).
Hence, f → f and its inverse mapping, which are both linear, are continuous by the
characterization of continuous linear mappings from Theorem 1.25. This shows that
f → f is an isomorphism from C(Ω1 × Ω2 ; E) onto C(Ω1 ; C(Ω2 ; E)).
sup f (x1 )Cb (Ω2 ;E);ν = sup sup (f (x1 ))(x2 )E;ν
x1 ∈Ω1 x1 ∈Ω1 x2 ∈Ω2
= sup f (x1 , x2 )E;ν .
(x1 ,x2 )∈Ω1 ×Ω2
(b) 1. Continuity and injectivity in Cb (Ω1 × Ω2 ; E). Let f ∈ Cb (Ω1 × Ω2 ; E). For
every x1 ∈ Ω1 , f (x1 ) is uniformly continuous (since x2 → f (x1 , x2 ) is uniformly
continuous by restriction) and, for every y1 ∈ Ω1 and ν ∈ NE ,
f (y1 ) − f (x1 )Cb (Ω2 ;E);ν = sup f (y1 , x2 ) − f (x1 , x2 )E;ν . (1.14)
x2 ∈Ω2
By Definition 1.2 (b) of uniform continuity, for every > 0, there exists η > 0 such
that |(y1 , x2 ) − (x1 , x2 )| ≤ η implies f (y1 , x2 ) − f (x1 , x2 )E;ν ≤ . By (1.14),
|y1 − x1 | ≤ η therefore implies that f (y1 ) − f (x1 )Cb (Ω2 ;E);ν ≤ . This shows that
f is uniformly continuous from Ω1 into Cb (Ω2 ; E), since, by Definition 1.4 (b), the
latter is endowed with the semi-norms of Cb (Ω; E). Hence, f ∈ Cb (Ω1 ; Cb (Ω2 ; E)).
Case where Ω1 is compact. Let h ∈ Cb (Ω1 ; Cb (Ω2 ; E)) and f ∈ C(Ω1 × Ω2 ; E) the
function obtained by regrouping its variables with (c).
f (xn1 , xn2 ) − f (x1 , xn2 )E;ν ≤ h(xn1 ) − h(x1 )Cb (Ω2 ;E);ν → 0, (1.16)
f (x1 , xn2 ) − f (x1 , y2n )E;ν = (h(x1 ))(xn2 ) − (h(x1 ))(y2n )E;ν → 0, (1.17)
f (x1 , y2n ) − f (y1n , y2n )E;ν ≤ h(x1 ) − h(y1n )Cb (Ω2 ;E);ν → 0, (1.18)
respectively, since h ∈ Cb (Ω1 ; Cb (Ω2 ; E)), h(x1 ) ∈ Cb (Ω2 ; E), and, once again,
h ∈ Cb (Ω1 ; Cb (Ω2 ; E)). These three properties imply that
Case where Ω1 is bounded and E is a Neumann space. Let h ∈ Cb (Ω1 ; Cb (Ω2 ; E)).
Since E is a Neumann space, Cb (Ω2 ; E) is also a Neumann space (Theorem 1.12),
so, by Theorem 1.26 (b), h has an extension P h ∈ Cb (Ω1 ; Cb (Ω2 ; E)).
Since the domain Ω1 is bounded, Ω1 is closed and bounded and hence compact
in Rd1 , so the previous case provides k in Cb (Ω1 × Ω2 ; E) such that k = P h. The
restriction f of k to Ω1 ×Ω2 belongs to Cb (Ω1 ×Ω2 ; E) and satisfies f = h. Therefore,
f → f is once again an isomorphism from Cb (Ω1 ×Ω2 ; E) onto Cb (Ω1 ; Cb (Ω2 ; E)).
(c) Isomorphism in CD1 ×D2 (Ω1 × Ω2 ; E). We will proceed in three steps:
Separation of variables. If f ∈ CD1 ×D2 (Ω1 ×Ω2 ; E), then the support of f is included
in D1 and, for every x1 ∈ Ω1 , the support of f (x1 ) is included in D2 for all x1 . Hence,
f ∈ CD1 (Ω1 ; CD2 (Ω2 ; E)) by case (b).
Regrouping the variables. Let h ∈ CD1 (Ω1 ; CD2 (Ω2 ; E)) and f ∈ C(Ω1 × Ω2 ; E)
the function obtained by regrouping its variables with (c). Then the support of f is
included in D1 × D2 .
Thus, f ∈ CD1 ×D2 (Ω1 × Ω2 ; E), and the mapping f → f is a bijection from
CD1 ×D2 (Ω1 × Ω2 ; E) onto CD1 (Ω1 ; CD2 (Ω2 ; E)).
Isomorphism. The equality (1.13) then shows that it is an isomorphism of these spaces,
since CD (Ω; E) is endowed with the semi-norms of Cb (Ω; E) by Definition 1.7 (b).
Then:
(b) For any two compact subsets D1 of Rd1 and D2 of Rd2 , the mapping f → f˘ is
an isomorphism from CD1 (Ω1 ; CD2 (Ω2 ; E)) onto CD2 (Ω2 ; CD1 (Ω1 ; E)).
Proof. Let f ∈ C(Ω1 ; C(Ω2 ; E)). Regrouping its variables yields g ∈ C(Ω1 × Ω2 ; E)
(Theorem 1.27 (a)). Permuting its variables, i.e. writing g(x2 , x1 ) = g(x1 , x2 ), yields
g ∈ C(Ω2 × Ω1 ; E). Finally, separating the variables of g yields f˘ ∈ C(Ω2 ; C(Ω1 ; E))
(Theorem 1.27 (a) once again).
Similarly, the other properties can be deduced from the properties in Theorem 1.27.
28 Continuous Functions
Then
Notes. The hypothesis (1.20) means that F is uniformly equicontinuous [Vol. 1, Definition 10.5].
Optimality of Theorem 1.29. The conditions of this theorem are necessary and sufficient for every
sequence of F to have a convergent subsequence [S IMON, 75, forthcoming].
m+1
Utility of Theorem 1.29. We will use this theorem to show that every bounded sequence in CK (Ω) has
a convergence subsequence in CK
m (Ω) (Theorem 2.25) and, using this fact, that every bounded sequence
in D(Ω) has a convergent subsequence [Vol. 3], a result that will in turn be useful for the Schwartz kernel
theorem.
Compact versus sequentially compact The definitions of both notions of compactness in a semi-normed
space are recalled in the Appendix (Definitions A.17 and A.18). In spaces of continuous functions, we only
use sequential compactness, not compactness.
Note that there are separated semi-normed spaces where compact is neither stronger nor weaker than
sequentially compact [Vol. 1, Properties (2.6) and (2.7), p. 43]. If this holds in a given space E, then it also
holds in Cb (Ω; E).
12 History of the Arzelà-Ascoli theorem. In 1883 [4, pp. 545–549], Giulio A SCOLI gave a construction
with the essence of Theorem 1.29 in one dimension, namely d = 1. Cesare A RZELÀ showed the converse in
1889 [2]. Both of their manuscripts use geometric language that makes our statement difficult to recognize.
In 1895 [3, pp. 56–60], Cesare A RZELÀ stated Theorem 1.29 in one dimension. Maurice F RÉCHET
extended the result in 1906 to functions defined on a subset of a metrizable space in his thesis [38].
Another random document with
no related content on Scribd:
Defiles the verdant plain. Nor idle stand
The trusty slaves; with pointed spears, they pierce
Through their tough hides; or at their gaping mouths
An easier passage find. The king of brutes,
In broken roarings, breathes his last; the bear
Grumbles in death; nor can his spotted skin,
Though sleek it shine, with varied beauties gay,
Save the proud pard from unrelenting fate.
The battle bleeds; grim slaughter strides along,
Glutting her greedy jaws, grins o’er her prey.
Men, horses, dogs, fierce beasts of every kind,
A strange promiscuous carnage, drench’d in blood,
And heaps on heaps amass’d. What yet remain
Alive, with vain assault, contend to break
The impenetrable line: others, whom fear
Inspires with self-preserving wiles, beneath
The bodies of the slain for shelter creep;
Aghast they fly, or hide their heads, dispersed.
And now, perchance, had Heaven but pleased, the work
Of death had been complete; and Aurengzebe,
By one dread frown, extinguish’d half their race;
When lo! the bright sultanas of his court
Appear, and to his ravish’d eyes display
Those charms, but rarely to the day reveal’d.