You are on page 1of 3

“Lorentz2” — 2021/6/15 — 23:13 — page 1 — #1

Lorentz spaces
Much of science and engineering centers on with respect to the Lebegue measure. Notice that
understanding functions that describe physical, µf is always defined on R+ := [0, ∞). One can
chemical, or biological processes. In mathemat- select a representative among all equi-measurable
ics, providing such understanding organizes itself functions as
into the disciplines of real, complex, and func-
tional analysis. At the coarsest level, functions f ∗ (t) := inf{y : µf (y) ≤ t}, t ≥ 0,
are typically classified in terms of either their size,
or their smoothness, or self-similarities. The ear- which is also a non-increasing function defined on
liest measures of size were boundedness and in- R+ . The function f ∗ is called the decreasing re-
tegrability. Smoothness tries to understand how arrangment of f . It encodes all size properties of
the functions’ output depends on small variations f relative to Lp norms. For example,
in the input and has led to a myriad of notions Z Z∞
circling around differentiability. Self-similarities |f | dµ = [f ∗ (t)]p dt,
p
1 ≤ p < ∞.
include periodicity and fractal structures
Ω 0
The best known quantifications of size are
given by Lebesque Lp norms (formulated by M. In his semi-
Riesz). If f is a real valued measureable function nal paper [8],
defined on a measure space (Ω, µ), the normed George Lorentz
linear space Lp := Lp (Ω, µ) , 1 ≤ p ≤ ∞, con- suggested using
sists of all µ measurable functions f for which the the rearrange-
norm ment f ∗ as a
R general way to
( |f |p dµ)1/p , 1 ≤ p < ∞,

define spaces
kf kLp := Ω
ess sup |f (x)|, p = ∞.
 of functions de-
x∈Ω scribed by size.
The idea is to
is finite.
simply apply a
In their work, Hardy and Littlewood stressed
function norm
the fact that Lp is a rearrangement invariant
to f ∗ . He sug-
space. To explain what this means, we introduce Figure 0.1: A function
gested a cadre
for any µ measurable function f its distribution and its monotonic rear-
of possibilities.
function rangement.
The most im-

µf (y) := µ {x ∈ Ω : |f (x)| > y} , y ≥ 0. portant of these
turned out to be the Lorentz Lp,q := Lp,q (Ω, µ)
This function is non-increasing on R+ . All Lp spaces, described as all µ measurable functions f
norms of f are determined from µf via the iden- defined on Ω for which
tity
kf kLp,q :=
Z∞  R∞
kf kpLp = p y p−1 µf (y)dy, 1 ≤ p < ∞. { [t1/p f ∗ (t)]q dt }1/q , 1 ≤ p, q < ∞,
t
0 0
1/p ∗
supt>0 t f (t), 1 ≤ p < ∞, q = ∞,

Thus, for the purpose of measuring size by Lp
norms, two functions f, g which have distribution is finite. Notice that these spaces agree with Lp ,
functions equal almost everywhere have identi- 1 ≤ p < ∞, when we choose q = p but give a
cal Lp norms irrespective of their native measure fine gradation of spaces near Lp when p is fixed
spaces. and q varies. The most important of these vari-
Two measurable functions f and g are said to ants turns out to be the case q = ∞, which gives
be equi-measurable if µf (y) = µg (y), a.e. on R+ , the spaces Lp,∞ commonly referred to as weak

1
“Lorentz2” — 2021/6/15 — 23:13 — page 2 — #2

Lp spaces. The membership condition for weak L2 to itself will necessarily boundedly map Lp to
Lp is that Lp0 for 1 ≤ p ≤ 2. This result was later for-
mally put into interpolation theorems (such as
µ {x ∈ Ω : |f (x)| > y} ≤ Cy −1/p ,

y > 0, the Riesz–Thorin theorem) and started the study
of interpolation of operators. The Lorentz spaces
with C an absolute constant. played an important role in this new development
The subsequent impact of [8] was quite pro- in several important ways.
found in a range of disciplines including har-
monic analysis, approximation theory, partial
differential equations, and functional analysis. One of the main chapters of harmonic anal-
On the one ysis is the Calderón–Zygmund (CZ) program to
hand, the new understand the mapping properties of the funda-
Lorentz spaces mental operators that arise in the theory of differ-
Lp,q were in- ential equations. These include singular integrals
strumental in and maximal operators such as the well-known
explaining the Hardy–Littlewood maximal operator. The main
mapping prop- vehicle for proving mapping properties of opera-
erties of certain tors is the above theory of interpolation. A proto-
fundamental op- typical example in the CZ program is the Hilbert
erators of anal- transform. M. Riesz showed that this operator
ysis such as the mapped Lp into itself when 1 < p < ∞ but does
Fourier, Hilbert, not map L1 into itself. This is a typical exam-
and Riesz trans- ple where the ‘end point’ behavior of the opera-
forms. On tor is opaque. This end-point behavior is nicely
the other hand, filled in via the Lorentz spaces. In the case of the
they served as Hilbert transform, it turns out that it maps L1
the prototype Figure 0.2: The Lebesgue into weak L1 , i.e. into L1,∞ . While this result
for the devel- L1 norm corresponds to is interesting in and of itself, even more is true.
opment of new the area of the sub- Namely, any operator that boundedly maps L1
function spaces graph (for positive func- to weak L1 and L2 into itself will automatically
such as Besov tions), while the L1,∞ boundedly map Lp into itself for 1 < p ≤ 2. In
classes and ap- norm corresponds for the other words, to ascertain strong mapping prop-
proximation area of the maximal in- erties of linear operators, it is enough to verify
classes that re- scribed ”rectangle”. weak mapping properties on pairs of spaces and
main important deduce the strong mapping via interpolation for
to this day in intermediate spaces.
our understanding of numerical methods. Per-
haps, the most important consequence was the
development of a new branch of analysis called The subject of interpolation of linear opera-
interpolation of operators which helped unify tors began to organize itself into unified studies in
several mathematical disciplines. the late 1960s (see the treatises [1, 2]). Calderón
Regarding the latter point, the modern the- introduced his complex method of interpolation
ory of interpolation of operators can rightfully [4] and subsequently his treatment of weak inter-
be traced to the work of M. Riesz who was inter- polation via rearrangements [5]. At more or less
ested, among other things, in the mapping prop- the same time, the real method of interpolation
erties of the Fourier transform F. The Hausdorff– initiated by Lions and Peetre came to the fore-
Young theorem showed that F mapped Lp into front. The main theme of the real method of in-
Lp0 , 1 ≤ p ≤ 2, where p0 is the conjugate in- terpolation was to show that a linear operator T
dex given by 1/p + 1/p0 = 1. Riesz showed that which boundedly maps Banach spaces Xi → Yi ,
the Hausdorf–Young theorem is actually a conse- i = 0, 1 automatically maps X → Y for a stable
quence of a general principle that any linear op- of new spaces X, Y . These new spaces are de-
erator which boundedly mapping L1 to L∞ and fined via a K-functional which, in the simplest

2
“Lorentz2” — 2021/6/15 — 23:13 — page 3 — #3

case when X1 continuously embeds into X0 , is A similar connection occurs when engaging
smoothness spaces. In a development quite sim-
K(f, t) := K(f, t; X0 , X1 ) := ilar to the Lorentz spaces, O. Besov [3] intro-
 
inf kf − gkX0 + tkgkX1 , 0 < t < ∞. duced a fine gradation of smoothness spaces that
g∈X1 carry his name. The space Bqs (Lp (Ω)) consists
Examples of the new intermediate spaces X are of functions with smoothness of order s in Lp
the θ, q spaces Xθ,q = (X0 , X1 )θ,q consisting of with q playing the same role of fine gradation
all f ∈ X0 for which as in the case of Lorentz spaces. The connec-
tions with Lorentz spaces become even clearer
Z∞ when one realizes that these Besov spaces are in-
kf kqXθ,q := [t−θ K(f, t)]q dt/t, terpolation spaces between Lp and the Sobolev
0 space W r (Lp (Ω)) when r > s. Indeed, the K
0 < θ < 1; 0 < q < ∞, (0.1) functional for this pair is none other than the
modulus of smoothness ωr (f, t)p and so the norm
with the obvious modification for q = ∞. One in the Besov space takes the form (0.1). Besov
can rather easily show that the above operator T spaces are now commonly used in the study of
boundedly maps Xθ,q into Yθ,q for each θ ∈ (0, 1) solutions to partial differential equations and nu-
and 0 < q ≤ ∞. While the similarity between merical analysis of these equations. Finally, let
these spaces and the Lorentz spaces is appar- us mention that interpolation spaces, which can
ent, the connections have an even happier ending be viewed as descendants of the Lorentz spaces,
since when the measure µ is finite, are also a staple in approximation theory where
Zt they are used to characterize the functions which
K(f, t; L1 , L∞ ) = f ∗ (s) ds, t > 0, (0.2) possess a specified approximation order (see [7]).
0
Ronald DeVore
and
(L1 , L∞ )θ,q = Lp,q , θ = 1 − 1/p. (0.3)

References
[1] C. Bennett and R. Sharpley, Interpolation of Operators, Academic Press, N.Y., 1988.
[2] J. Bergh and J. Löfström, Interpolation Spaces, Springer, Berlin, 1976.
[3] O. V. Besov, On a certain family of functional spaces. Embedding and extension theorems, Dokl.
Akad. Nauk SSSR, 126 (1959), 1163–1165.

[4] A. P. Calderón, Spaces between L1 and L∞ and the theorem of Marcinkiewicz, Studia Math.26
(1966), 273–299.
[5] A. P. Calderón, Intermediate spaces and interpolation, the complex method, Studia Math. 24 (1964)
113–190.

[6] A. P. Calderon and A. Zygmund On the existence of certain singular integrals, Acta Math., 88(1952),
85–139 (1952).
[7] R. DeVore and G.G. Lorentz, Constructive Approximation, Grundlehren, Springer, New York, 1993.
[8] G. G. Lorentz, Some new functional spaces, Annals of Math., 51(1950), 37–55.

You might also like