- Abstract. We show that the best constants appearing in the weak type (1,1) inequalities satisfied by the centered Hardy-Littlewood maximal function associated to some finite radial measures, such as the standard gaussian measure, grow exponentially fast with the dimension.

1 Introduction

Let
$M$
be the centered maximal function associated to euclidean balls. The study of uniform
${L}^{p}$
bounds in the dimension
$d$
was initiated by E. M. Stein, who proved their existence ([St1] , [St2] , see also [St3] ) for every
$p>1$
. A motivation for this work comes from the desire to extend (at least some parts of ) harmonic analysis in
${\mathbb{R}}^{d}$
, to the infinite dimensional case. Stein's result was generalized to arbitrary balls by J. Bourgain ([Bou1] , [Bou2] , [Bou3] ) and A. Carbery ([Ca] ) when
$p>3/2$
. For
${\ell}_{q}$
balls,
$1\le q<\infty $
, D. Müller [Mu] showed that uniform bounds again hold for every
$p>1$
. With respect to weak type bounds, in [StSt] E. M. Stein and J. O. Strömberg proved, among other things, that the best constants in the weak type (1,1) inequality satisfied by
$M$
grow at most like
$O\left(d\right)$
, and asked if uniform bounds could be found.

Since then, there has been remarkably little progress on this question (see, for instance, [A] , [AV] for the case of cubes).

Here we study the weak type (1,1) problem for integrable radial densities defined via bounded decreasing functions, the canonical example being the standard gaussian measure.

This is a natural variant of Stein and Strömberg's question, given the growing interest in what has been termed “gaussian harmonic analysis”, where Lebesgue measure is replaced by the standard gaussian measure, and also because of the importance of gaussian measures and other probabilites in the infinite dimensional setting (see, for instance, [Bo] ). For the measures considered in this paper, instead of uniform bounds we have exponential increase:

If
${\mu}_{d}$
is a finite radial Borel measure on
${\mathbb{R}}^{d}$
defined by a bounded decreasing function
${f}_{d}$
, and if
${c}_{d}$
denotes the best constant appearing in the weak type (1,1) inequality satisfied by the associated maximal function
${M}_{\mu}$
, then for
$d$
large enough,
${c}_{d}\ge {\left(1+\frac{1}{\sqrt{d}}\right)}^{-1}{\left(\frac{2}{\sqrt{3}}\right)}^{d/6}.$

2 Notation and results

Given a Borel measure
$\mu $
on
${\mathbb{R}}^{d}$
, the associated centered maximal function
${M}_{\mu}$
is defined by
$${M}_{\mu}f\left(x\right):={sup}_{\{r>0:\mu B(x,r)>0\}}\frac{1}{\mu B(x,r)}{\int}_{B(x,r)}\left|f\right|d\mu ,$$
where
$B(x,r)$
denotes the euclidean closed ball of radius
$r>0$
centered at
$x$
(the choice of closed balls in the definition is mere convenience; using open balls instead does not change the value of
${M}_{\mu}f\left(x\right)$
). The boundary of
$B(x,r)$
is the sphere
$\mathbb{S}(x,r)$
. Sometimes we use
${B}^{d}(x,r)$
and
${\mathbb{S}}^{d-1}(x,r)$
to make their dimensions explicit. If
$x=0$
and
$r=1$
, we just write
${B}^{d}$
and
${\mathbb{S}}^{d-1}$
.

It is a consequence of Besicovitch covering theorem that there exists a constant
$c=c\left(d\right)$
, independent of
$\mu $
, such that for every
$f\in {L}^{1}({\mathbb{R}}^{d},\mu )$
and every
$\alpha >0$
, we have
$\alpha \mu \{{M}_{\mu}f\ge \alpha \}\le c\parallel f{\parallel}_{1}$
. In fact, by the Theorem in pg. 227 of [Su] we may take
$c=(2.641+o(1){)}^{d}$
. Of course, if we are interested in the best (lowest) such
$c$
, then
$c=c\left(\mu \right)$
will depend on
$\mu $
. Note that it makes no difference in the determination of the best constant if instead of the strict inequality
$\{{M}_{\mu}f>\alpha \}$
we use
$\{{M}_{\mu}f\ge \alpha \}$
.

The following trivial lemma has been upgraded to Proposition in order to highlight the fact that uniform bounds are not likely to exist when dealing with families of finite measures, due to the decay imposed by the condition
${\mu}_{d}\left({\mathbb{R}}^{d}\right)<\infty $
. We give
$\frac{a}{\mu B(y,r)}$
the obvious interpretation when
$a\ne 0$
and
$\mu B(y,r)=0$
.

Proposition 2.1.
Let
$\mu $
be a Borel measure on
${\mathbb{R}}^{d}$
, and let
${c}_{d}$
be the best constant appearing in the weak type (1,1) inequality satisfied by
${M}_{\mu}$
. Given any ball
$B(x,r)$
with
$\mu B(x,r)>0$
,
$${c}_{d}\ge {inf}_{y\in \mathbb{S}(x,r)}\frac{\mu B(x,r)}{\mu B(y,r)}.$$

Proof. Fix
$x$
and
$r$
so that
$\mu B(x,r)>0$
. By a standard approximation argument we may consider, instead of a function, the Dirac delta
${\delta}_{x}$
placed at
$x$
. Thus
${c}_{d}\ge {sup}_{\alpha >0}\alpha \mu \{{M}_{\mu}{\delta}_{x}\ge \alpha \}.$
Note that for some
$y\in \mathbb{S}(x,r)$
,
$\mu B(y,r)>0$
. Note also that the support of
$\mu $
has full measure, i.e.,
$\mu \left(\right(\text{supp}\mu {)}^{c})=0$
. Let
${\alpha}_{0}:={inf}_{y\in \mathbb{S}(x,r)}\frac{1}{\mu B(y,r)}$
. Clearly for every
$y\in \mathbb{S}(x,r)\cap \left(\text{supp}\mu \right)$
,
${M}_{\mu}{\delta}_{x}\left(y\right)\ge {\alpha}_{0}$
, and the same holds for every
$z=(1-t)x+ty$
,
$y\in \mathbb{S}(x,r)$
,
$0<t<1$
,
$z\in \text{supp}\mu $
, since
$x\in B(z,tr)\subset B(y,r)$
. It follows that
$B(x,r)\cap \left(\text{supp}\mu \right)\subset \{{M}_{\mu}{\delta}_{x}\ge {\alpha}_{0}\}$
, so
${c}_{d}\ge {\alpha}_{0}\mu B(x,r)$
, as claimed. □ Let
${f}_{d}:[0,\infty )\to [0,\infty )$
be a nonincreasing integrable function, not
$0$
a.e., let
${\sigma}^{d-1}$
denote the area on the unit sphere
${\mathbb{S}}^{d-1}$
, let
${\stackrel{~}{\sigma}}^{d-1}$
denote the normalized area on
${\mathbb{S}}^{d-1}$
(thus
${\stackrel{~}{\sigma}}^{d-1}$
is a probability), and let
${\lambda}^{d}$
be Lebesgue measure on
${\mathbb{R}}^{d}$
. Then, expressing a point in
${\mathbb{R}}^{d}$
as
$(r,x)$
,where
$r\in [0,\infty )$
and
$x\in {\mathbb{S}}^{d-1}$
, the function
${f}_{d}$
defines a nontrivial rotationally invariant (or radial) measure
${\mu}_{d}$
via
$${\mu}_{d}\left(A\right):={\int}_{A}f\left(\right|y\left|\right)d{\lambda}^{d}\left(y\right)={\int}_{0}^{\infty}{\int}_{{\mathbb{S}}^{d-1}}{\chi}_{A}(r,x)f\left(r\right){r}^{d-1}drd{\sigma}^{d-1}\left(x\right).$$
Theorem 2.2.
Let
${f}_{d}$
and
${\mu}_{d}$
be as above, and let
${c}_{d}$
be the best constant appearing in the weak type (1,1) inequality satisfied by
${M}_{{\mu}_{d}}$
. Then there exists a sequence
$\left\{{a}_{d}\right\}$
such that
${lim}_{d}{a}_{d}=1$
and for all
$d$
,
$${c}_{d}\ge {\left(1+\frac{{a}_{d}2\sqrt{2}}{\sqrt{3\pi d}}\right)}^{-1}{\left(\frac{2}{\sqrt{3}}\right)}^{d/6}.$$

Proof. Fix
$d$
. Given a unit vector
$v\in {\mathbb{R}}^{d}$
and
$\varepsilon \in [0,1)$
, the
$\varepsilon $
spherical cap about
$v$
is the set
$C(\varepsilon ,v):=\{\theta \in {\mathbb{S}}^{d-1}:\langle \theta ,v\rangle \ge \varepsilon \}$
. Note that spherical caps are just geodesic balls
${B}_{{\mathbb{S}}^{d-1}}(x,r)$
in
${\mathbb{S}}^{d-1}$
. In the special case
$v={e}_{1}=(1,0,...,0)$
,
$\varepsilon ={2}^{-1}$
, we have
$C({2}^{-1},{e}_{1})={B}_{{\mathbb{S}}^{d-1}}({e}_{1},\pi /3)$
. Next we remind the reader of some well-known facts that will be used in the sequel: i)
${\lambda}^{d}\left({B}^{d}\right):=\frac{{\pi}^{d/2}}{\Gamma (1+d/2)}$
; ii)
${\sigma}^{d-1}\left({\mathbb{S}}^{d-1}\right)=d{\lambda}^{d}\left({B}^{d}\right)$
; iii)
${lim}_{x\to \infty}\frac{{e}^{x}\Gamma (1+x)}{{x}^{x+1/2}}=\sqrt{2\pi}$
(Stirling's formula); iv)
${\sigma}^{d-1}\left({B}_{{\mathbb{S}}^{d-1}}\right(x,r\left)\right)={\sigma}^{d-2}\left({\mathbb{S}}^{d-2}\right){\int}_{0}^{r}{sin}^{d-2}tdt$
(see, for instance, (A.11) pg. 259 of [Gra] for a more general statement). Note that from i), ii), iii), iv) and the fact that
$cost\ge 1/2$
on
$[0,\pi /3]$
, we get the following upper bound on the normalized area of
$C({2}^{-1},{e}_{1})$
:
$${\stackrel{~}{\sigma}}^{d-1}C({2}^{-1},{e}_{1})\le 2\frac{{\sigma}^{d-2}\left({\mathbb{S}}^{d-2}\right)}{{\sigma}^{d-1}\left({\mathbb{S}}^{d-1}\right)}{\int}_{0}^{\pi /3}{sin}^{d-2}tcostdt$$
$$=\frac{2}{d}\frac{{\lambda}^{d-1}\left({B}^{d-1}\right)}{{\lambda}^{d}\left({B}^{d}\right)}{\left(\frac{\sqrt{3}}{2}\right)}^{d-1}={\left(\frac{\sqrt{3}}{2}\right)}^{d}\frac{{a}_{d}2\sqrt{2}}{\sqrt{3\pi d}},$$
where
${lim}_{d}{a}_{d}=1$
(the value of
${a}_{d}$
depends on Stirling's approximation).

The function
$h\left(R\right):=\frac{{\mu}_{d}B(0,R)}{{\mu}_{d}B(0,(\sqrt{3}/2\left)R\right)}$
is continuous, and
${lim}_{R\to \infty}h\left(R\right)=1$
by the finiteness of the measure. It follows that there is a largest real number
${R}_{1}$
such that
$h\left({R}_{1}\right)=(2/\sqrt{3}{)}^{d/6}$
, provided of course that the set
$\{h\ge (2/\sqrt{3}{)}^{d/6}\}$
is nonempty. We show next that this is always the case. Fix
${R}_{0}$
such that
${f}_{d}\left(0\right){\lambda}^{d}B(0,{R}_{0})=1$
(this is just a convenient normalization, any
${R}_{0}$
could be used), and suppose towards a contradiction that for every
$R>0$
,
$h\left(R\right)<(2/\sqrt{3}{)}^{d/6}$
.Select a positive integer
$k$
satisfying
$(\sqrt{3}/2{)}^{5dk/6}<{\mu}_{d}B(0,{R}_{0})$
. Using
$k$
times the assumption
$h<(2/\sqrt{3}{)}^{d/6}$
we get
$${\mu}_{d}B(0,{R}_{0})<(2/\sqrt{3}{)}^{dk/6}{\mu}_{d}B(0,(\sqrt{3}/2{)}^{k}{R}_{0})\le (2/\sqrt{3}{)}^{dk/6}f(0\left){\lambda}^{d}B\right(0,(\sqrt{3}/2{)}^{k}{R}_{0})$$
$$=(\sqrt{3}/2{)}^{5dk/6}f(0\left){\lambda}^{d}B\right(0,{R}_{0})<{\mu}_{d}B(0,{R}_{0}).$$
By the previous proposition together with rotational invariance, it is enough to check that there exists a sequence
$\left\{{a}_{d}\right\}$
with
${lim}_{d}{a}_{d}=1$
such that for all
$d$
,
$$\frac{{\mu}_{d}B(0,{R}_{1})}{{\mu}_{d}B({R}_{1}{e}_{1},{R}_{1})}\ge {\left(1+\frac{{a}_{d}2\sqrt{2}}{\sqrt{3\pi d}}\right)}^{-1}{\left(\frac{2}{\sqrt{3}}\right)}^{d/6}.$$
We split
${\mu}_{d}B({R}_{1}{e}_{1},{R}_{1})={\mu}_{d}\left(B\right(0,{R}_{1})\cap B({R}_{1}{e}_{1},{R}_{1}\left)\right)+{\mu}_{d}\left(\right(B(0,{R}_{1}{)}^{c}\cap B({R}_{1}{e}_{1},{R}_{1}\left)\right)$
and estimate each of the summands. Note that
$$B(0,{R}_{1})\cap B({R}_{1}{e}_{1},{R}_{1})\subset B({2}^{-1}{R}_{1}{e}_{1},{2}^{-1}\sqrt{3}{R}_{1}).$$
For every pair of points
$(x,y)$
with
$x\in B(0,{R}_{1})\backslash B({R}_{1}{e}_{1},{R}_{1})$
,
$y\in B({R}_{1}{e}_{1},{R}_{1})\backslash B(0,{R}_{1})$
, we have
$\left|x\right|<\left|y\right|$
, so
${f}_{d}\left(\right|x\left|\right)\ge {f}_{d}\left(\right|y\left|\right)$
. By the choice of
${R}_{1}$
and the preceding observation,
$${\mu}_{d}\left(B\right(0,{R}_{1})\cap B({R}_{1}{e}_{1},{R}_{1}\left)\right)\le {\mu}_{d}B({2}^{-1}{R}_{1}{e}_{1},{2}^{-1}\sqrt{3}{R}_{1})$$
$$\le {\mu}_{d}B(0,{2}^{-1}\sqrt{3}{R}_{1})=(\sqrt{3}/2{)}^{d/6}{\mu}_{d}B(0,{R}_{1}).$$
Let
$E$
be the (semi) cone in
${\mathbb{R}}^{d}$
defined by
${x}_{1}={3}^{-1/2}\sqrt{{x}_{2}^{2}+\cdot \cdot \cdot +{x}_{d}^{2}}$
, and let
${E}^{\prime}:=\{{x}_{1}\ge {3}^{-1/2}\sqrt{{x}_{2}^{2}+\cdot \cdot \cdot +{x}_{d}^{2}}\}$
denote the solid cone determined by
$E$
. Then
$$B(0,{R}_{1}{)}^{c}\cap B({R}_{1}{e}_{1},{R}_{1})\subset B(0,(2/\sqrt{3}{)}^{5}{R}_{1})\cap {E}^{\prime}.$$
By rotational invariance of the probability measure
$$\nu \left(A\right):=\frac{{\mu}_{d}\left(B\right(0,(2/\sqrt{3}{)}^{5}{R}_{1})\cap A)}{{\mu}_{d}\left(B\right(0,(2/\sqrt{3}{)}^{5}{R}_{1}))},$$
the
$\nu $
-measure of
${E}^{\prime}$
is just the normalized area of its intersection with the sphere, i.e.

$$\nu \left({E}^{\prime}\right)={\stackrel{~}{\sigma}}^{d-1}({E}^{\prime}\cap {\mathbb{S}}^{d-1})={\stackrel{~}{\sigma}}^{d-1}C({2}^{-1},{e}_{1}).$$
From the choice of
${R}_{1}$
and the upper bound on
${\stackrel{~}{\sigma}}^{d-1}C({2}^{-1},{e}_{1})$
we get
$${\mu}_{d}\left(\right(B(0,{R}_{1}{)}^{c}\cap B({R}_{1}{e}_{1},{R}_{1}\left)\right)\le \nu \left({E}^{\prime}\right){\mu}_{d}\left(B\right(0,(2/\sqrt{3}{)}^{5}{R}_{1}))$$
$$<{\stackrel{~}{\sigma}}^{d-1}C({2}^{-1},{e}_{1})(2/\sqrt{3}{)}^{5d/6}{\mu}_{d}(B(0,{R}_{1}))$$
$$\le {\left(\frac{2}{\sqrt{3}}\right)}^{5d/6}{\mu}_{d}\left(B\right(0,{R}_{1}\left)\right){\left(\frac{\sqrt{3}}{2}\right)}^{d}\frac{{a}_{d}2\sqrt{2}}{\sqrt{3\pi d}}$$
$$={\mu}_{d}\left(B\right(0,{R}_{1}\left)\right){\left(\frac{\sqrt{3}}{2}\right)}^{d/6}\frac{{a}_{d}2\sqrt{2}}{\sqrt{3\pi d}}.$$
Putting both estimates together, we obtain
$$\frac{{\mu}_{d}B(0,{R}_{1})}{{\mu}_{d}B({R}_{1}{e}_{1},{R}_{1})}\ge \frac{{\mu}_{d}B(0,{R}_{1})}{\left(1+\frac{{a}_{d}2\sqrt{2}}{\sqrt{3\pi d}}\right){\left(\frac{\sqrt{3}}{2}\right)}^{d/6}{\mu}_{d}B(0,{R}_{1})},$$
as desired. □

Remarks 2.3.
1. We briefly comment on the role that different assumptions play in the preceding proof. Finiteness of the measures was only used to show that for each
$d$
there is an
${R}_{d}>0$
such that
$h\left({R}_{d}\right)\ge (2/\sqrt{3}{)}^{d/6}$
and
$h\left(\right(2/\sqrt{3}{)}^{5}{R}_{d})\le (2/\sqrt{3}{)}^{d/6}$
.

So exponential dependency holds for families of radial measures, not necessarily finite, provided there exists a ball in each dimension with this kind of decay (clearly the proof can be adapted to decays lower than the one considered above).

Normalization of the measures is not an issue either: If
$c>0$
, then
${M}_{\mu}={M}_{c\mu}$
.

The radial assumption makes it easy to check the decay and to apply Proposition 2.1 , but as noted above, the existence of some ball with such decay is likely to be prevalent even among nonradial (finite) measures. Regarding the assumption of absolute continuity, it is a simple way to ensure that the dimension of
${\mu}_{d}$
goes to infinity as
$d\to \infty $
. To obtain (trivially) singular examples where there is dependendecy of
${c}_{d}$
on
$d$
, embed
${\mathbb{R}}^{d}$
, for instance, in
${\mathbb{R}}^{2d}$
. On the other hand, setting
${\mu}_{d}=\delta $
for every
$d$
, we have
${c}_{d}=1$
. Finally, we point out that there are some natural families of singular radial finite measures to which the preceding arguments do not apply, such as, for instance, area on
${\mathbb{S}}^{d-1}$
.

2. When dealing with concrete families of measures, the additional information may lead to more precise bounds. For instance, let
${\nu}_{d}\left(A\right):={\lambda}^{d}(A\cap {B}^{d})$
be Lebesgue measure on the unit ball in
${\mathbb{R}}^{d}$
. The argument used in the proof of Theorem 2.2 immediately gives
${c}_{d}\left({\nu}_{d}\right)\ge (2/\sqrt{3}{)}^{d}$
, but it is possible to do better by estimating the volume of the solid cap
${B}^{d}\cap \{{x}_{1}\ge {2}^{-1}\}$
and using Stirling's approximation. This easily yields
${c}_{d}\left({\nu}_{d}\right)\ge \frac{{a}_{d}\sqrt{\pi}}{\sqrt{6}}{\left(\frac{2}{\sqrt{3}}\right)}^{d}\sqrt{d},$
where
${lim}_{d}{a}_{d}=1$
.
References
Departamento de Matematicas y Computacion, Universidad de La Rioja, 26004 Logrono, La Rioja, Spain.

- Aldaz, J. M. A remark on the centered $n$ -dimensional Hardy-Littlewood maximal function. Czechoslovak Math. J. 50(125) (2000), no. 1, 103–112.
- Aldaz, J. M., Varona, J. L. Singular measures and convolution operators. To appear in Acta Math. Sinica.
- Bogachev, V. I. Gaussian measures. Mathematical Surveys and Monographs, 62. American Mathematical Society, Providence, RI, 1998.
- Bourgain, J. On high-dimensional maximal functions associated to convex bodies. Amer. J. Math. 108 (1986), no. 6, 1467–1476.
- Bourgain, J. On the ${L}^{p}$ -bounds for maximal functions associated to convex bodies in ${R}^{n}$ . Israel J. Math. 54 (1986), no. 3, 257–265.
- Bourgain, J. On dimension free maximal inequalities for convex symmetric bodies in ${R}^{n}$ . Geometrical aspects of functional analysis (1985/86), 168–176, Lecture Notes in Math., 1267, Springer, Berlin, 1987.
- Carbery, A. An almost-orthogonality principle with applications to maximal functions associated to convex bodies. Bull. Amer. Math. Soc. (N.S.) 14 (1986), no. 2, 269–273.
- Gray, A. Tubes. Addison-Wesley Publishing Company, Advanced Book Program, Redwood City, CA, 1990.
- Müller, D. A geometric bound for maximal functions associated to convex bodies. Pacific J. Math. 142 (1990), no. 2, 297–312.
- Stein, E. M. The development of square functions in the work of A. Zygmund. Bull. Amer. Math. Soc. (N.S.) 7 (1982), no. 2, 359–376.
- Stein, E. M. Three variations on the theme of maximal functions. Recent progress in Fourier analysis (El Escorial, 1983), 229–244, North-Holland Math. Stud., 111, North-Holland, Amsterdam, 1985.
- Stein, E. M. Harmonic analysis: real-variable methods, orthogonality, and oscillatory integrals. With the assistance of Timothy S. Murphy. Princeton Mathematical Series, 43. Monographs in Harmonic Analysis, III. Princeton University Press, Princeton, NJ, 1993.
- Stein, E. M.; Strömberg, J. O. Behavior of maximal functions in ${R}^{n}$ for large $n$ . Ark. Mat. 21 (1983), no. 2, 259–269.
- Sullivan, John M. Sphere packings give an explicit bound for the Besicovitch covering theorem. J. Geom. Anal. 4 (1994), no. 2, 219–231.