Abstract
The well known result of Bourgain and Kwapień states that the projection \(P_{\le m}\) onto the subspace of the Hilbert space \(L^2\left( \Omega ^\infty \right) \) spanned by functions dependent on at most m variables is bounded in \(L^p\) with norm \(\le c_p^m\) for \(1<p<\infty \). We will be concerned with two kinds of endpoint estimates. We prove that \(P_{\le m}\) is bounded on the space \(H^1\left( {\mathbb {D}}^\infty \right) \) of functions in \(L^1\left( {\mathbb {T}}^\infty \right) \) analytic in each variable. We also prove that \(P_{\le 2}\) is bounded on the martingale Hardy space associated with a natural double-indexed filtration and, more generally, we exhibit a multiple indexed martingale Hardy space which contains \(H^1\left( {\mathbb {D}}^\infty \right) \) as a subspace and \(P_{\le m}\) is bounded on it.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The Rademacher functions \(\left( r_i\right) _{i\in {\mathbb {N}}}\) generate a well studied subspace of \(L^p[0,1]\), which we identify with \(L^p\left( {\mathbb {Z}}_2^{\mathbb {N}}\right) \). In particular by Khintchine inequality
for \(0<p<\infty \), and \(\overline{\mathrm {span}}\left( r_i:i\in {\mathbb {N}}\right) \) is complemented in \(L^p\) for \(1<p<\infty \) but not for \(p=1\). We will index the Walsh system by finite subsets of \({\mathbb {N}}\), i.e.
The number |A| is called the mutiplicity of \(w_A\). Analogous problems for Walsh functions of finite multiplicity have been resolved independently by Bonami [2] and Kiener [13]. Namely, the inequality
holds true, and the orthogonal projection \(P_m\) onto \(\overline{\mathrm {span}}\left( w_A:|A|\le m\right) \) is bounded if and only if \(1<p<\infty \). Some lower estimates for \(P_m\) are also known, see [12] or [20] for a detailed discussion on this subject.
In [4], Bourgain generalized these results to a setting in which \(\left( {\mathbb {Z}}_2,\left\{ \emptyset ,\{0\},\{1\},\{0,1\}\right\} ,\frac{1}{2}\#\right) \) is replaced with an arbitrary probability space \(\left( \Omega ,{\mathcal {F}},\mu \right) \). To be more precise, let \(\left( \Omega ^\infty ,{\mathcal {F}}^{\otimes \infty },\mu ^{\otimes \infty }\right) \) be the infinite product space. Any \(f\in L^2\left( \Omega ^\infty ,{\mathcal {F}}^{\otimes \infty },\mu ^{\otimes \infty }\right) \) can be decomposed in a unique way into a series
where \(f_{i_1,\ldots ,i_m}\in L^2\left( \Omega ^m\right) \) is mean zero in each of its m arguments. Thus, \(P_A\) and \(P_m\) defined by
are mutually orthogonal orthogonal projections. In the case of \(\Omega ={\mathbb {Z}}_2\), the image of \(P_A\) is just the one-dimensional space spanned by \(w_A\), so the above definition of \(P_m\) coincides for with the projection onto Walsh functions of multiplicity m. In [4] Bourgain proved that for \(1\le p<\infty \),
which is a direct generalization of (1.3). Moreover, he proved that \(P_m\) is bounded on \(L^p\) if and only if \(1<p<\infty \), with norm smaller than \(c_p^m\) where \(c_p\lesssim \frac{ {\hat{p}}^\frac{5}{2}}{\log {\hat{p}}}\) and \({\hat{p}}=p\vee \frac{p}{p-1}\).
It turns out that the projections \(P_m\) have a well established probabilistic interpretation. In [14], Kwapień connected them to the notion of Hoeffding decomposition, which originated from Hoeffding’s work [11]. More precisely, elements of the image of \(P_m\) are what is called generalized canonical U-statistics and the decomposition \(f=\sum _m P_m f\) plays a crucial role in the proofs of many theorems concerning U-statistics. For more information, we refer the reader to [16]. Kwapień provided a shorter proof of Bourgain’s result about boundedness of \(P_m\), with a better constant \(c_p\lesssim \frac{{\hat{p}}}{\log {\hat{p}}}\).
Let us decribe the main results of this paper, which give certain endpoint estimates for \(P_m\). One of them (Theorem 4.5 in the text) is obtained by restricting the domain of \(P_m\). For exact definition of \(H^1_{\mathrm {all}}\), see Sect. 2.
Theorem A
\(P_m\) is bounded on the subspace \(H^1_{\mathrm {all}}\left( {\mathbb {T}}^\infty \right) \) of \(L^1\left( {\mathbb {T}}^\infty \right) \) consisting of functions analytic in each variable.
We also find a norm stronger than \(L^1\) and weaker than \(L^p\) (\(p>1\)), in which \(P_m\) is bounded. The detailed construction is described in Sect. 5.2.
Theorem B
For any \(m\in {\mathbb {N}}\), there is a partition of the family of finite subsets of \({\mathbb {N}}\) into \({\dot{\bigcup }}_{i\in {\mathcal {I}}}{\mathcal {A}}_i\) such that the norm
is between \(L^1\) and all \(L^p\) (\(p>1\)) and \(P_m\) is bounded in this norm.
It is worth noting that Theorem A translates directly to the space \({\mathcal {H}}^1\) of Dirichlet series, i.e. the closure of polynomials of the form \(\sum _{n=1}^N b_n n^{-s}\) in the norm
The Bohr lift, dating back to [1], is the map
where \(a_k=b_n\) for n having the prime number factorization \(n=\prod _{j}p_j^{k_j}\). It is an isometry between \(H^1_{\mathrm {all}}\left( {\mathbb {T}}^\infty \right) \) and the space \({\mathcal {H}}^1\) of Dirichlet series. Thus, our result is equivalent to the fact that the projection from \({\mathcal {H}}^1\) onto
is bounded. For a more detailed exposition of Dirichlet series and their relation to polydisc Hardy spaces, see [21].
The paper is organized as follows. In Sect. 2 we introduce necessary notation and definitions. In Sect. 3, we provide a new simple proof of the historic \(L^p\) boundedness result. The proof of the estimate \(\left\| P_m\right\| \le \left( \mathrm {e}\left\| P_1\right\| \right) ^m\) is done by means of a combinatorial identity expressing \(P_m\) in terms of tensor products of \(P_1\). In Sect. 4, we show that the same argument carries over with little modification showing boundedness of \(P_m\) on \(H^1_{\mathrm {all}}\left( {\mathbb {T}}^\infty \right) \). In Sect. 5.1, we define, purely in terms of square functions and not referring to analyticity, a multiple indexed martingale Hardy space \(H^1\left[ {\mathcal {T}}_m\right] \) of functions on \(\Omega ^\infty \) that admits a bounded action of \(P_m\). It turns out that if \(\Omega ={\mathbb {T}}\), there is a subspace \(H^1_{m\text { last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \) of \(L^1\left( {\mathbb {T}}^\infty \right) \), much bigger than \(H^1_{\mathrm {all}}\left( {\mathbb {T}}^\infty \right) \), on which the \(L^1\) norm is equivalent to \(H^1\left[ {\mathcal {T}}_m\right] \) norm. The arguments rely heavily on \(L^1\) square function theorem for Hardy martingales and decoupling inequality of Zinn. We present two proofs of the latter in Sect. 1.
2 Preliminaries
Probability spaces and conditional expectations In all of the text, \(\left( \Omega ,{\mathcal {F}},\mu \right) \) will be a probability space. We will equip sets of the form \(\Omega ^I\), where I is an at most countable index set, with the product measure \(\mu ^{\otimes I}\) defined on \({\mathcal {F}}^{\otimes I}\). In case we are only concerned with the cardinality of I, we will write \(\Omega ^n\), where n is a natural number or \(\infty \). By the natural filtration on \(\Omega ^{\mathbb {N}}\) we mean the filtration \(\left( {\mathcal {F}}_n:n=0,1,\ldots \right) \), where \({\mathcal {F}}_k\) is generated by the coordinate projection \(\omega \mapsto \left( \omega _1,\ldots ,\omega _k\right) \) and denote \({\mathbb {E}}_k={\mathbb {E}}\left( \cdot \mid {\mathcal {F}}_k\right) \). In general, for a subset A of the index set, \({\mathcal {F}}_A\) will be the sigma algebra generated by the coordinate projection \(\omega \mapsto \left( \omega _i\right) _{i\in A}\) and \({\mathbb {E}}_A={\mathbb {E}}\left( \cdot \mid {\mathcal {F}}_A\right) \). In more explicit terms, measurability with respect to \({\mathcal {F}}_A\) is equivalent to being dependent only on variables with indices belonging to A and the conditional expectation operator \({\mathbb {E}}_A\) integrates away the dependence on all other variables, so that the formulas
are satisfied (with the convention that sequences indexed by A and \({\mathbb {N}}{\setminus } A\) are merged in a natural way into a sequence indexed by \({\mathbb {N}}\)). It will often be convenient to identify a function f defined on \(\Omega ^A\) with an \({\mathcal {F}}_A\)-measurable function \(\Omega ^I\ni \omega \mapsto f\left( \left( \omega _i\right) _{i\in A}\right) \). In order to save space, we will often write \(\mathrm {d}x\) instead of \(\mathrm {d}\mu (x)\) whenever the measure is implied by context.
Tensor products Let \(1\le p<\infty \). For \(f_k\in L^p\left( \Omega _k\right) \), we will denote by \(\bigotimes _{k=1}^n f_k\) the function on \(\prod _k \Omega _k\) satisfying
Because of separation of variables, we have \(\left\| \bigotimes _k f_k\right\| _{L^p\left( \prod _k \Omega _k\right) }= \prod _k \left\| f_k\right\| _{L^p\left( \Omega _k\right) }\). This way we actually define an injection of the algebraic tensor product \(\bigotimes _k L^p\left( \Omega _k\right) \) into \(L^p\left( \prod _k \Omega _k\right) \), the image of which is dense.
Let \(X_k\) be subspaces (by a subspace we always mean a closed linear subspace) of \(L^p\left( \Omega _k\right) \). By \(\bigotimes _k X_k\) we will denote the subspace of \(L^p\left( \prod _k \Omega _k\right) \) spanned by functions of the form \(\bigotimes _k f_k\), where \(f_k\in X_k\), and the norm is inherited from \(L^p\left( \prod _k \Omega _k\right) \) (care has to be taken, as \(\bigotimes _k X_k\) is not determined solely by \(X_k\) as Banach spaces, but rather by the particular way they are embedded in \(L^p\left( \Omega _k\right) \)). If \(T_k: X_k\rightarrow L^p\left( \Omega _k\right) \) are bounded operators, then we can define an operator \(\bigotimes _k T_k: \bigotimes _k X_k\rightarrow L^p\left( \prod _k \Omega _k\right) \) by the formula
and easily check that the property
is satisfied. Indeed, \(\bigotimes _k T_k= \prod _k \mathrm {id}_{L^p\left( \prod _{j\ne k}\Omega _j\right) }\otimes T_k\), and any operator of the form \(\mathrm {id}\otimes T\) has norm bounded by \(\Vert T\Vert \), because \(\left( \mathrm {id}\otimes T\right) f\left( \omega _1,\omega _2\right) = T\left( f\left( \omega _1,\cdot \right) \right) \left( \omega _2\right) \).
Fourier transform Let \({\mathbb {T}}\) be the interval \([0,2\pi )\) equipped with addition modulo \(2\pi \) and normalized Lebesgue measure \(\mathrm {d}\mu = \frac{\mathrm {d}x}{2\pi }\). We will be exclusively dealing with Fourier transforms of functions on \({\mathbb {T}}\) or some power of \({\mathbb {T}}\). Since the group dual to \({\mathbb {T}}\) is \({\mathbb {Z}}\), the dual group to the product \({\mathbb {T}}^{\mathbb {N}}\) is the direct sum \({\mathbb {Z}}^{\oplus {\mathbb {N}}}\) (i.e., integer-valued sequences that are eventually 0), on which we define the Fourier transform by
Hardy spaces of martingales and analytic functions By \({\mathbb {D}}\) we denote the unit disk in the complex plane. We can identify \({\mathbb {T}}\) with the unit circle by the map \(t\mapsto \mathrm {e}^{it}\). For \(N\in {\mathbb {N}}\), the space \(H^1({\mathbb {D}}^N)\) is defined as the space of functions analytic in the polydisc \({\mathbb {D}}^N\) such that the norm
is finite. It is well-known [10] that such a function has an a.e. radial limit \(f\left( t_1,\ldots ,t_n\right) =\lim _{r\rightarrow 1} F\left( r\mathrm {e}^{it_1},\ldots ,r\mathrm {e}^{it_n}\right) \) on the distinguished boundary \({\mathbb {T}}^N\) and F can be recovered from f by convolution with a Poisson kernel. This sets a one-to-one correspondence between \(H^1\left( {\mathbb {D}}^N\right) \) and the space
We also can define \(H^1_{\mathrm {all}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \) in the same manner as in (2.8), but care has to be taken, since these functions are can no longer be extended analytically to \({\mathbb {D}}^{\mathbb {N}}\) in general (hence the shorthand \(H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) \), which we will sometimes use, is an abuse of notation). Later we will use two more \(H^1\) spaces, namely \(H^1_{\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \) (also called Hardy martingales) and \(H^1_{m\,\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \), which we will define as follows.
In the space \(H^1_{m\,\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \) we allow characters of the form \(\mathrm {e}^{i\langle n,t\rangle }\), where \(\left| \mathrm {supp}\,n\right| <m\) and \(n_j\ge 0\) for all j.
Now we recall the definition of a martingale Hardy space and some related inequalities. A standard reference in this matter is [9]. Let \(\left( {\mathcal {F}}_n\right) _{n=0}^\infty \) be an arbitrary filtration on a probability space \(\left( \Omega ,{\mathcal {F}},\mu \right) \), where \({\mathcal {F}}\) is generated by \(\bigcup {\mathcal {F}}_n\). We denote \({\mathbb {E}}_k={\mathbb {E}}\left( \cdot \mid {\mathcal {F}}_k\right) \), \(\Delta _0={\mathbb {E}}_0\), \(\Delta _{k}={\mathbb {E}}_k-{\mathbb {E}}_{k-1}\) for \(k\ge 1\), and define the square function and maximal function of f respectively by
This allows us to define the martingale Hardy space.
Definition 2.1
The space \(H^1\left[ \left( {\mathcal {F}}_n\right) _{n=1}^\infty \right] \) is a function space on \(\Omega \) with the norm
We will make use of three following classical martingale inequalities.
Theorem 2.2
(Burkholder, Gundy [7] for \(1<p<\infty \); Davis [8] for \(p=1\)) For \(1\le p<\infty \),
Theorem 2.3
(Burkholder [6]) For \(1<p<\infty \),
Theorem 2.4
(Stein [24]) For \(1<p<\infty \) and an arbitrary sequence \(\left( f_n\right) _{n=0}^\infty \),
Definition 2.5
A martingale atom is a function of the form
where
Theorem 2.6
Let \(f \in H^1\left[ \left( {\mathcal {F}}_n\right) _{n=1}^\infty \right] \) be of mean 0. Then there are atoms \(a_1,a_2,\ldots \) and scalars \(c_1,c_2,\ldots \) such that
and
Theorem 2.7
[Fefferman] The dual space to \(H^1\left[ \left( {\mathcal {F}}_n\right) _{n=1}^\infty \right] \) is \(BMO\left[ \left( {\mathcal {F}}_n\right) _{n=1}^\infty \right] \), where
where the duality is given by \(\langle f,g\rangle =\lim _{n\rightarrow \infty }{\mathbb {E}}\left( {\mathbb {E}}_n f{\mathbb {E}}_n g\right) \).
Vector-valued inequalities For a Banach space B, by \(L^p\left( S,B\right) \) we denote the Bochner space of strongly measurable B-valued random variables equipped with the norm
(or, equivalently, the closed span of functions of the form \((f\otimes v)(x)= f(x)v\), where \(f\in L^p(S)\) and \(v\in B\), in the \(L^p(S,B)\) norm). For an operator T between subspaces of \(L^p\left( S_1\right) \) and \(L^p\left( S_2\right) \) and a linear operator \(F:B_1\rightarrow B_2\) we can define \(T\otimes F\) and the algebraic tensor product by \(\left( T\otimes F\right) \left( f\otimes v\right) = T(f)\otimes F(v)\), but this construction does not necessarlily produce a bounded operator on the closure. The main tool for obtaining vector-vlaued extensions of inequalities will be the following lemma, which for \(I_1,I_2\) being singletons is due to Marcinkiewicz and Zygmund [18] (in this case \(\lesssim \Vert T\Vert \) can be replaced with \(\le \Vert T\Vert \)).
Lemma 2.8
Let \(X_i\subset L^1\left( S_i,\ell ^2\left( I_i\right) \right) \) for \(i=1,2\), B be a Hilbert space and \(T:X_1\rightarrow X_2\) be bounded. Then \(T\otimes \mathrm {id}_B: X_1\otimes B\rightarrow X_2\otimes B\), where \(X_i\otimes B\) is treated as a subspace of \(L^1\left( \Omega _i,\ell ^2\left( I_i,B\right) \right) \), is bounded with norm \(\lesssim \Vert T\Vert \).
Proof
Without loss of generality, B is finite-dimensional, say \(B=\ell ^2\left( J\right) \) for some finite J. Let \(X_1\otimes \ell ^2\left( J\right) \ni f=\left( f_j\right) _{j\in J}\), so that \(f_j\in X_1\). Let also \(r_j\) for \(j\in J\) be Rademacher variables. Then, applying \(\ell ^2\left( I_2\right) \)-valued Khintchine inequality,
\(\square \)
Hoeffding decomposition Now we define the main object of our interest. In order to avoid technicalities with convergence in strong operator topology, we will work in a finite product of \(\Omega \) (all the results extend automatically to \(\Omega ^\infty \) by density). We will see in a moment that any function \(f\in L^1\left( \Omega ^n\right) \) can be decomposed in a unique way as
where \(P_{i_1,\ldots ,i_m}f\left( x_1,\ldots ,x_n\right) \) depends only on \(x_{i_1},\ldots ,x_{i_m}\) and is of mean 0 with respect to each of \(x_{i_1},\ldots ,x_{i_m}\) (equivalently, \(P_A f\) is \({\mathcal {F}}_A\)-measurable and is orthogonal to all \({\mathcal {F}}_B\)-measurable functions for \(B\subsetneq A\)). This decomposition has been studied in [4, 14]. In particular, \(P_{i_1,\ldots ,i_m}\) are pairwise orthogonal orthogonal projections. Let
and \(U_m\) be the range of \(P_m\). It is known [4, 14] that \(P_m\) is bounded on \(L^p\left( \Omega ^n\right) \), \(1<p<\infty \), with norm independent on n, but this is not true for \(L^1\left( \Omega ^n\right) \).
One of the possible ways to prove the existence of the above decomposition in \(L^2\left( \Omega ^n\right) \) is as follows. First we define the subspace
for each \(m\ge 0\). The sequence of subspaces \(U_{\le 0},U_{\le 1},\ldots ,U_{\le n}\) is increasing, so by putting
we obtain a decomposition
into an orthogonal direct sum of \(U_m\). We will denote the orthogonal projection onto \(U_m\) by \(P_m\).
A more explicit formula for \(P_m\) can be obtained. For \(A\subset [1,n]\), let
where \(\mathrm {id}\) and \({\mathbb {E}}\) are understood to act on \(L^2(\Omega )\), and let \(U_A\) be the range of the projection \(P_A\). It is easy to see that
and, since the subspaces \(U_B\) are mutually orthogonal,
Moreover
and consequently
Decoupling inequalities We are going to present a special case of a theorem of J. Zinn [26], which will be one of the most important tools.
Theorem 2.9
For \(k=1,\ldots ,N\), let \(f_k\) be a function on \(\Omega ^k\). Then
We will provide two new proofs of the above in Sect. 1. Below, we state two corollaries obtained by iterating Zinn’s inequality.
Corollary 2.10
For \(1\le a< b\le N\), let \(f_{a,b}\in L^1\left( \Omega ^N, {\mathcal {F}}_{[a,b]}\right) \). Denote \(\left( x_a,\ldots ,x_b\right) \) by \(x_{[a,b]}\). Then
Proof
Let \(F_b\in L^1\left( \Omega ^N, {\mathcal {F}}_b,\ell ^2\right) \) be defined by \(\left( F_b\right) _a= f_{a,b}\) for \(a<b\) and 0 otherwise. Then, by Theorem 2.9 applied for functions \(\left\| F_b\right\| _{\ell ^2}\),
Analogously, by setting y as fixed, and applying Theorem 2.9 with reversed order of variables (which we can do, because we are dealing with finite sums),
as desired. \(\square \)
Corollary 2.11
For all \(i=\left( i_1,\ldots ,i_m\right) \) such that \(i_1<\ldots <i_m\), let \(f_i\) be an \({\mathcal {F}}_{\left[ 1,i_1-1\right] \cup \left\{ i_1,\ldots ,i_m\right\} }\)-measurable function on \(\Omega ^{\mathbb {N}}\). Then, treating each \(f_i\) as a function on \(\Omega ^{\left[ 1,i_1-1\right] }\times \Omega ^{\left\{ i_1,\ldots ,i_m\right\} }\),
where \(y^{(1)},\ldots ,y^{(m)}\) are variables in \(\Omega ^{\mathbb {N}}\).
Proof
Let us fix \(k\in \{1,\ldots ,m\}\) and for each \(j\in {\mathbb {N}}\) define a function \(\varphi _j\) on \(\Omega ^{\left[ 1,j\right] }\times \left( \Omega ^{\mathbb {N}}\right) ^{m-k}\) by the formula
Then, for fixed \(y^{(> k)}=\left( y^{(k+1)},\ldots ,y^{(m)}\right) \in \left( \Omega ^{\mathbb {N}}\right) ^{m-k}\),
Here, \(i_k\) plays the role of j and (2.48) is an application of Theorem 2.9 to functions \(\left| \varphi _j\right| ^2\). Integrating the resulting inequality with respect to \(y^{(>k)}\), we get
which by induction from \(k=m\) to \(k=1\) proves (2.44). \(\square \)
3 Boundedness of \(P_m\) on \(L^p\left( \Omega ^{\mathbb {N}}\right) \)
The main motivation for this part is the following theorem, proved by Bourgain with \(c_p\lesssim \frac{{\hat{p}}^{\frac{5}{2}}}{\ln {\hat{p}}}\) and by Kwapień with \(c_p\lesssim \frac{{\hat{p}}}{\ln {\hat{p}}}\), where \({\hat{p}}=\max \left( p,\frac{p}{p-1}\right) \).
Theorem 3.1
[4, 14] \(P_m\) is bounded on \(L^p\left( \Omega ^{\mathbb {N}}\right) \) for \(1<p<\infty \), with norm \(\lesssim c_p^m\).
We will present a proof that yields and .
Proof
Without loss of generality, we may assume that we are working in \(L^p\left( \Omega ^{[1,N]},{\mathcal {F}}^{\otimes N}\right) \). Indeed, by (2.33) and (2.37), \(P_m\) preserves \(L^2\left( \Omega ^{\mathbb {N}},{\mathcal {F}}_{[1,N]}\right) \), which can be canonically identified with \(L^2\left( \Omega ^{[1,N]},{\mathcal {F}}^{\otimes N}\right) \). Since the sequence \(\left( L^2\left( \Omega ^{\mathbb {N}},{\mathcal {F}}_{[1,N]}\right) :N\in {\mathbb {N}}\right) \) is increasing and its sum is dense in \(L^p\left( \Omega ^{\mathbb {N}},{\mathcal {F}}^{\otimes {\mathbb {N}}}\right) \), all we need to prove is
\(P_0={\mathbb {E}}\) is bounded. The \(L^p\) boundedness of \(P_1\) is essentially a known result [5], but we provide a proof for the sake of completeness. Let \(\left( {\mathcal {F}}_k:k\in [0,N]\right) \) be the natural filtration and \(\left( {\mathcal {F}}^*_k\right) _{k=0}^N\) be the natural reversed filtration, i.e. \({\mathcal {F}}^*_k={\mathcal {F}}_{[k,N]}\). By (2.32) and (2.37) we see that
By mutual orthogonality of \(P_A\)’s
Applying Theorem 2.3, (3.3) and Theorem 2.4, we obtain
We will now proceed by induction. Suppose that (3.1) is satisfied with \(m-1\) in the place of m. Let \(N=mn\) and define an operator \(Q_m\) acting on \(L^p\left( \Omega ^{[1,N]}\right) \) by
Utilising (2.37) we get
By (2.31),
Putting the last four equations together, we get
However, by (3.8) and the induction hypothesis,
Let \(a_n\approx b_n\) denote \(\lim _{n\rightarrow \infty }\frac{a_n}{b_n}=1\). By the Stirling formula,
Thus
Finally, by (3.15), (3.18) and (3.23),
\(\square \)
We provide a short proof of a fact taken from [5] that Theorem 3.1 can not be, extended to \(p=1\) or \(\infty \), which motivates the next section.
Proposition 3.2
If \(\Omega \) is not a single atom, then \(P_m\) for \(m\ge 1\) is not bounded on \(L^1\left( \Omega ^\infty \right) \) or \(L^\infty \left( \Omega ^\infty \right) \).
Proof
It is enough to consider \(L^1\left( \Omega ^\infty \right) \), because \(P_m\)’s are self-adjoint. Let \(f\in L^2\left( \Omega \right) \) be such that \({\mathbb {E}}f=1\), \(f\ge 0\) and \(\mu \left( \mathrm {supp}\,f\right) <1\). Then \({\mathbb {E}}|f-1|^2>0\). For \(F_n=f^{\otimes n}\in L^2\left( \Omega ^n\right) \) we have
which is not dominated by \(\left\| F_n\right\| _{L^1\left( \Omega ^n\right) }= \left\| f\right\| _{L^1\left( \Omega \right) }^n=1\). To prove the unboundedness of \(P_m\) for \(m>1\), we simply notice that
\(\square \)
4 Boundedness of \(P_m\) on \(H^1 \left( {\mathbb {D}}^{\mathbb {N}}\right) \)
The projection \(P_m\) can be described even more explicitly in the case \(\Omega ={\mathbb {T}}\). Indeed, if \(n\in {\mathbb {Z}}^{\oplus {\mathbb {N}}}\) is supported on the set A, then
Thus
and
In particular, \(P_m\) preserves the space \(H^2\left( {\mathbb {D}}^{\mathbb {N}}\right) = L^2\left( {\mathbb {T}}^{\mathbb {N}}\right) \cap H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) \).
In order to adapt the proof of Theorem 3.1 to the \(H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) \) case, we will need a replacement for the argument proving that \(P_1\) is bounded. The role of the combination of Burkholder–Gundy and Doob inequalities will be played by the following theorem, which can be found in [3].
Theorem 4.1
(Bourgain) For \(f\in H^1_{\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \), there is an equivalence of norms
where \(\left( {\mathcal {F}}_n\right) _{n=0}^\infty \) is the natural filtration on \({\mathbb {T}}^{\mathbb {N}}\).
For later use, we note the Hilbert space valued extension.
Corollary 4.2
Let B be a Hilbert space. For \(f\in H^1_{\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}},B\right) \), there is an equivalence of norms
where \(\left( {\mathcal {F}}_n\right) _{n=0}^\infty \) is the natural filtration on \({\mathbb {T}}^{\mathbb {N}}\).
Proof
Theorem 4.1 gives a map
which is an isomorphism onto the subspace of \(L^1\left( {\mathbb {T}}^{\mathbb {N}},\ell ^2\right) \) consisting of functions f such that \(f_k\) is a k-th martingale difference and is analytic in the k-th variable, defined by
Thus, applying Lemma 2.8 with \(I_1\) being a singleton, \(I_2={\mathbb {N}}\), T as above (and then the same for \(T^{-1}\)) we get
\(\square \)
The role of the Stein martingale inequality will be played by the following simple observation.
Corollary 4.3
For any sequence \(\left( f_n:n\in {\mathbb {N}}\right) \) adapted to the natural filtration on \(\Omega ^{\mathbb {N}}\),
Proof
Let \({\tilde{f}}_n\) be a sequence of functions on \(\Omega ^{\mathbb {N}}\times \Omega ^{\mathbb {N}}\) defined by
Applying Theorem 2.9 and conditional expectation with respect to the second of two sets of variables,
\(\square \)
By conditioning with respect to the first set of variables, we obtain the inequality
due to Lepingle [17].
Theorem 4.4
For any \(\Omega \), \(P_1\) is bounded on \(H^1\left[ \left( {\mathcal {F}}_n\right) _{n=0}^\infty \right] \).
Proof
We proceed as in the proof of Theorem 3.1. First, we reduce the problem to the \(\Omega ^{[1,N]}\) realm. Then we notice that
which by Corollary 4.3 yields
\(\square \)
Theorem 4.5
\(P_m\) is bounded on \(H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) \) with norm \(\le \frac{1}{\mathrm {e}}c_1^m\), where
Proof
The case \(m=0\) is trivial, \(m=1\) follows directly from Theorems 4.1 and 4.4. The induction step is identical to the proof of Theorem 3.1, up to changing \(L^p\left( \Omega ^I\right) \) to \(H^1\left( {\mathbb {D}}^I\right) \). Alternatively, we can prove the same in a single step. Set
It is easily seen that for each set B of cardinality m, \(P_B\) appears \(m!\left( {\begin{array}{c}(n-1)m\\ n-1,\ldots ,n-1\end{array}}\right) \) times in the sum. Therefore
and since
we get
\(\square \)
It has to be noted that our proofs of Theorems 3.1 and 4.5 extend naturally to a vector valued case, respectively UMD and AUMD valued. Indeed, Bourgain’s proof of Theorem 2.4, as presented in [20], extends to the UMD valued version, while Theorem 4.1 is just the statement that a one-dimensional space has the AUMD property. In both cases, the induction follows without change. There is also a second direction in which we can generalize. Namely, by looking carefully at the proof of Theorem 4.1, one can see that the only place in which analyticity plays a role is the \(H^1=H^2\cdot H^2\) theorem, which is true for \(H^1\) on any compact and connected group with ordered dual [22], which means that we can replace \({\mathbb {T}}\) with any such group.
Given that Kwapień’s constant \(c_p\) in Theorem 3.1 has the best known asymptotics as a function of p for \(m=1\), one can ask about the dependence of and on m.
Proposition 4.6
The inequalities
for nontrivial \(\Omega \) and
where \(L^p_0\) and \(H^1_0\) stand for functions of mean 0, are true. Also,
Proof
Let \(f\in L^p\left( \Omega ^n\right) \) be of mean 0. Then \(f^{\otimes m}\in L^p\left( \Omega ^{mn}\right) \) and
Indeed, we have \(f=\sum _{|A|\ge 1}P_A f\) because \(P_{\emptyset }f={\mathbb {E}}f=0\), hence
The only way to get a summand in \(U_m\) is to have \(\left| A_i\right| =1\) for all i and the sum of such summands is the right hand side of (4.32). Taking an f which is close to attaining the norm of \(P_1\) on a respective space proves (4.29) and (4.30).
In order to see (4.31), assume for the sake of contradiction that \(P_1\) is a contraction on \(H^1_0\left( {\mathbb {D}}^2\right) \). We will test it on functions of the form \({\mathbb {T}}^2 \ni (w,z)\mapsto F(z)+w+azw \in {\mathbb {C}}\), where \(F\in H^1_0\left( {\mathbb {D}}\right) \) and a is a scalar. It is easy to see that
for \(\alpha , \beta \in {\mathbb {C}}\). Hence, from the inequality
we get
Since any nonnegative function can be approximated by the modulus of an \(H^1_0\left( {\mathbb {D}}\right) \) function, (4.35) is true for any nonnegative F. In particular, the left hand side attains a local minimum at \(a=0\), so by \(|u+v|=|u|+\mathrm {Re}\,\frac{u{\overline{v}}}{|u|}+o(v)\) we infer that
Now let
for \(r\ge 0\). This is a continuous function, whose values lie on some curve \(\gamma \) connecting 0 and 1 (because \(\phi (0)=1\) and \(\lim _{r\rightarrow \infty }\phi (r)=0\)). The condition (4.37) can be rewritten as
Since F was allowed to be any positive function, \(\phi (F)\) can be any function with values in \(\gamma \), making (4.39) obviously false. \(\square \)
5 Martingale Hardy spaces
5.1 Double indexed martingales
Above we noticed that the boundedness of \(P_1\) on \(H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) \) follows from the boundedness of \(P_1\) on a bigger space \(H^1\left[ \left( {\mathcal {F}}_n\right) \right] \). It is tempting to find an abstract martingale inequality responsible for the boundedness of \(P_m\) on \(H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) \). We can do this for \(m=2\).
By the natural double-indexed filtration on \(\Omega ^{\mathbb {N}}\) we will mean the family \(\left( {\mathcal {F}}_{[a,b]}:a\le b\right) \) (note that the inclusion order in the first index is reversed). Let \(\Delta _n={\mathbb {E}}_n-{\mathbb {E}}_{n-1}\) be the martinagle differences with respect to \(\left( {\mathcal {F}}_n\right) \) and \(\Delta ^*_n={\mathbb {E}}^*_n-{\mathbb {E}}^*_{n+1}\) be the martingale differences with repsect to \(\left( {\mathcal {F}}^*_n\right) \), where \({\mathcal {F}}^*_n={\mathcal {F}}_{[n,\infty )}\). We define the martingale differences with respect to \(\left( {\mathcal {F}}_{[a,b]}\right) \) by
and an \(H^1\) norm for this filtration by
The definition of double martingale differences coincides with what is considered in [25].
Corollary 5.1
For \(f\in H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) \), there is an equivalence of norms
where \(\left( {\mathcal {F}}_{[a,b]}\right) _{a\le b}\) is the natural double-indexed filtration on \({\mathbb {T}}^{\mathbb {N}}\).
Proof
For any \(\pm 1\)-valued sequence \(\left( \varepsilon _n:n\in {\mathbb {N}}\right) \), we define operators \(S_\varepsilon \) and \(S^*_\varepsilon \) by
By Theorem 4.1, \(S_\varepsilon \) is an isomorphism from \(H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) \) to itself, uniformly in \(\varepsilon \). By reversing the order of variables, the same can be said about \(S^*_\varepsilon \). Thus for any \(\varepsilon , \varepsilon '\),
By averaging the last quantity over all choices of \(\varepsilon , \varepsilon '\) and applying the Khintchine-Kahane inequality twice, we get the desired inequalities. \(\square \)
Theorem 5.2
\(P_2\) is bounded on \(H^1\left[ \left( {\mathcal {F}}_{a,b}\right) _{a\le b}\right] \), for any \(\Omega \).
Proof
As usual, we reduce the problem to the \(\Omega ^{[1,N]}\) version. By (3.2),
Thus
for \(a<b\). We can assume that \(P_0 f=P_1 f=0\) (i.e. \({\mathbb {E}}f=0\) and \(\Delta _{[a,a]}f=0\) for all a), because \(U_{\le 1}\), being the image of \({\mathbb {E}}+ \sum _a \Delta _{[a,a]}\) is trivially complemented in the underlying norm and \(P_2\) is 0 on \(U_{\le 1}\). By applying Corollary 2.10,
as desired.\(\square \)
5.2 Multiple indexed martingales
We will make an attempt at generalizing the above for multiple indexed martingales. Suppose there is a family \(\left( T_i,\partial T_i\right) _{i\in {\mathcal {I}}}\) of pairs of finite subsets of some set X (finite or not) indexed by some set \({\mathcal {I}}\), such that \(\partial T_i\subseteq T_i\) (\(\partial T_i\) is not a boundary in a topological sense - we use this notation for resemblance with the case where \(T_i\) are intervals and \(\partial T_i\) are their endpoints). We would like to define operators \(\Delta _i\) on \(L^2\left( \Omega ^X\right) \) by the formula
where \(T_i'\) stands for the complement of \(T_i\) in X. This is supposed to mimic the standard martingale differences when \(X={\mathbb {N}}\), \({\mathcal {I}}={\mathbb {N}}\), \(T_i=[0,i]\), \(\partial T_i=\{i\}\) and double martingale differences when \({\mathcal {I}}=\left\{ \left( a,b\right) :a\le b\right\} \), \(T_{a,b}=[a,b]\), \(\partial T_{a,b}=\{a,b\}\). The natural condition
is guaranteed by
Indeed,
Hence
and each \(P_A\) appears in the above sum exactly once if and only if the condition (5.19) is satisfied. For a family \(\left( T_i,\partial T_i\right) _{i\in {\mathcal {I}}}\) we may define a norm by the formula
and ask the following:
-
Is it true that
$$\begin{aligned} \Vert f\Vert _{H^1\left[ \left( T_i,\partial T_i\right) _{i\in {\mathcal {I}}}\right] }\simeq \Vert f\Vert _{H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) } \end{aligned}$$(5.26)for \(f\in \Vert f\Vert _{H^1\left( {\mathbb {D}}^{\mathbb {N}}\right) }\)?
-
If yes, is there any interesting example of a set \({\mathbb {N}}^{\oplus \infty }\subset \Gamma \subset {\mathbb {Z}}^{\oplus \infty }\) such that (5.26) is true for \(f\in L^1\left( {\mathbb {T}}^\infty \right) \) with \(\mathrm {supp}\,{\widehat{f}}\subset \Gamma \)?
-
For which, if any, m is \(P_m\) bounded on \(H^1\left[ \left( T_i,\partial T_i\right) _{i\in {\mathcal {I}}}\right] \)?
We are able to answer them in the case when
For a finite set \(B\subset {\mathbb {N}}\), the unique \(A\in {\mathcal {I}}\) such that \(\partial T_A\subseteq B\subseteq T_A\), which we will denote by \(\partial B\), is
Theorem 5.3
Let \(m\ge 1\) be fixed and \(\left( T_i,\partial T_i\right) _{i\in {\mathcal {I}}}\) be defined by (5.27), (5.28). Then
for \(f\in H^1_{m\,\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \), where \({\mathbb {T}}\) is used as \(\Omega \). Moreover, for \(m'\in {\mathbb {N}}\) and nontrivial \(\Omega \), the following are equivalent.
-
(i)
\(m'\le m\)
-
(ii)
\(P_{m'}\) is bounded on \(H^1\left[ \left( T_i,\partial T_i\right) _{i\in {\mathcal {I}}}\right] \)
-
(iii)
\(P_{m'}\) is bounded on \(H^1_{m\,\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \).
Proof
For \(A,B\subset {\mathbb {N}}\), we will write \(T^{(m)}_A\), \(\partial T^{(m)}_A\), \(\Delta ^{(m)}_A\), \(\partial ^{(m)}B\) to indicate the value of m we are currently using. For brevity we will denote \(\left( T^{(m)}_A,\partial T^{(m)}_A\right) _{|A|\le m}\) by \({\mathcal {T}}_m\). For \(|A|<m\), we have \(\Delta ^{(m)}_A=\left( \mathrm {id}-{\mathbb {E}}\right) ^{\otimes A}\otimes {\mathbb {E}}^{{\mathbb {N}}{\setminus } A}= P_A\). In particular, \(\Delta ^{(m)}_{\emptyset }= {\mathbb {E}}\). Therefore, by definition of the \(H^1\left[ {\mathcal {T}}_m\right] \) norm and Corollary 2.11,
Here, we identify an increasing sequence with the set of its elements, write \(\mathrm {d}y^{(1,\ldots ,m)}\) to denote \(\mathrm {d}y^{(1)}\ldots \mathrm {d}y^{(m)}\) and treat \(\Delta ^{(m)}_if\) as a function on \(\Omega ^{\left[ 1,i_1-1\right] }\times \Omega ^{\left\{ i_1,\ldots ,i_m\right\} }\). From this expression, we immediately see the implication \((i)\implies (ii)\). Indeed, for \(m'<m\),
which trivializes the inequality \(\Vert f\Vert _{H^1\left[ {\mathcal {T}}_m\right] } > rsim \Vert P_{m'}f\Vert _{H^1\left[ {\mathcal {T}}_m\right] }\). For \(m'=m\), we notice that
and the desired inequality follows from
The implication \((ii)\implies (iii)\) follows from (5.30), which we will prove by induction with respect to m. For \(m=1\) this is just Theorem 4.1. Suppose it is true for some m and let \(f\in H^1_{m+1\,\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \). In particular, \(f\in H^1_{m\,\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \). By (5.30), which is now the induction hypothesis, and (5.31),
The last two summands are as they are in the desired expression for \(m+1\) instead of m and we only have to deal with the first. For any \(i_1<\cdots <i_m\) and \(t\in {\mathbb {T}}^{\mathbb {N}}\) we have
Thus, treating \(\Delta ^{(m)}_i f\) as a function on \({\mathbb {T}}^{\left[ 1,i_1-1\right] }\times {\mathbb {T}}^{\left\{ i_1,\ldots ,i_m\right\} }\),
Let y be fixed and \(\Delta ^{(1)}_k\), where \(k\in {\mathbb {N}}\cup \{\emptyset \}\), act with respect to the variable \(x\in {\mathbb {T}}^{\mathbb {N}}\) (so, technically, \(\Delta ^{(1)}_k\) stands for \(\Delta ^{(1)}_k\otimes \mathrm {id}\)). Then
and
For \(k<i_1\),
By (5.40), \(\Delta ^{(m)}_{i_1,\ldots ,i_m}f\left( x_{<i_1},y_1,\ldots ,y_m\right) \) is in \(H^1_{\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \) with respect to x. Therefore, applying Corollary 4.2 to the vector valued function \(x\mapsto \left( \Delta ^{(m)}_{i}f\left( x_{<i_1},y^{(1)}_{i_1},\ldots ,y^{(m)}_{i_m}\right) \right) _{i_1<\ldots <i_m}\) with fixed \(y^{(1,\ldots ,m)}\), plugging in (5.41), (5.42), (5.47) and using Corollary 2.11, we get
Integrating the resulting equivalence with respect to \(y^{(1,\ldots ,m)}\) and plugging into (5.38), we verify that \(\Vert f\Vert _{L^1\left( {\mathbb {T}}^{\mathbb {N}}\right) }\simeq H^1\left[ {\mathcal {T}}_{m+1}\right] \), which finishes the proof of (5.30).
In order to see that \((iii)\implies (i)\), let us take \(m'>m\). For any \(g\in L^1\left( {\mathbb {T}}^n\right) \), the function \(G\in L^1\left( {\mathbb {T}}^{\mathbb {N}}\right) \) defined by
is in \(H^1_{m\,\mathrm {last}}\left( {\mathbb {T}}^{\mathbb {N}}\right) \). But
so
which by Proposition 3.2 can be arbitrarily big.\(\square \)
It is worth noting that by repeating the above proof of the equivalence between \(H^1_{m\text { last}}\) norm and \(H^1\left[ {\mathcal {T}}_m\right] \), one can obtain
where \(H^p\left[ {\mathcal {T}}_m\right] \) is defined in a natural way. Moreover, by iterating the \(\Vert f\Vert _{H^1}\ge \Vert f\Vert _{L^1}\) inequality for linearly ordered martingales,
References
Bohr, H.: Über die Bedeutung der Potenzreihen unendlich vieler Variabeln in der Theorie der Dirichletschen Reihen, Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse, pp. 441–488 (1913)
Bonami, A.: Ensembles \(\Lambda (p)\) dans le dual de \(D^\infty \). Ann. Inst. Fourier (Grenoble) 18(fasc. 2), 193–204 (1969) (1968)
Bourgain, J.: Embedding \(L^1\) in \(L^1/H^1\). Trans. Am. Math. Soc. 278(2), 689–702 (1983)
Bourgain, J.: Walsh subspaces of \(L^p\)-product spaces, Séminaire Analyse fonctionnelle (dit Maurey-Schwartz), pp. 1–14 (1979–1980)
Bryc, W., Kwapień, S.: On the conditional expectations with respect to a sequence of independent -fields. Z. Wahrsch. Verw. Gebiete 46, 221–225 (1979) (MR 0516742)
Burkholder, D.L.: Sharp inequalities for martingales and stochastic integrals. Colloq. Paul Lévy, Astéristique 157–158, 75–94 (1988)
Burkholder, D.L., Gundy, R.F.: Extrapolation and interpolation of quasi-linear operators on martingales. Acta Math. 124, 249–304 (1970)
Davis, B.: On the integrability of the martingales square function. Isr. J. Math. 8, 187 (1970)
Garsia, A.: Martingale inequalities. Math. Lecture Notes, Benjamin, New York (1973)
Gundy, R., Stein, E.M.: \(H^p\)-theory for the polydisk. Proc. Nat. Acad. Sci. USA 76, 1026–1029 (1979)
Hoeffding, W.: A class of statistics with asymptotically normal distributions. Ann. Math. Stat. 19, 293–325 (1948)
Kiener, K.: A problem concerning \(\Lambda (p)\) sets of the Walsh-Paley-system. Math. Balk. 4, 331–333 (1974) [Papers presented at the Fifth Balkan Mathematical Congress (Belgrade, 1974)]
Kiener, K.: Über Produkte von quadratisch integrierbaren Funktionen endlicher Vielfalt, Dissertation, Universität Innsbruck (1969)
Kwapień, S.: On Hoeffding decomposition in \(L_p\). Ill. J. Math. 54(3), 1205–1211 (2010) (Fall 2010)
Kwapień, S., Woyczyński, W.: Random series and stochastic integrals: single and multiple. Birkhäuser, Basel (1992)
Lee, A.J.: \(U\)-Statistics. Marcel Dekker Inc., New York (1990)
Lepingle, D.: Une inégalite de martingales. In: Séminaire de Probabilités XII, Lecture Notes in Math., vol. 649, pp. 134–137. Springer, Berlin (1978)
Marcinkiewicz, J., Zygmund, A.: Quelques inégalités pour les opérations linéaires. Fund. Math. 32(1), 115–121 (1939)
Müller, P.F.X.: A decomposition for Hardy martingales. Indiana Univ. Math. J. 61, 1802–1816 (2012)
Müller, P.F.X.: Isomorphisms between \(H^1\) spaces. Springer Science & Business Media, New York (2005)
Queffélec, H., Queffélec, M.: Diophantine approximation and Dirichlet series, Harish-Chandra Research Institute Lecture Notes, vol. 2. Hindustan Book Agency, New Delhi (2013)
Rudin, W.: Fourier Analysis on Groups, Interscience Tracts in Pure and Applied Math., No. 12. Wiley, New York (1962)
Rzeszut, M.: Higher order independent sums in product \(L^1\) and \(H^1\) spaces, doctoral dissertation, Kent State University (2018)
Stein, E.M.: Topics in harmonic analysis related to the littlewood-Paley theory, Chap 4. Princeton University Press, (1970)
Weisz, F.: Martingale Hardy spaces and their applications in Fourier analysis. Lecture Notes in Mathematics, vol. 1568. Springer, Berlin (1994)
Zinn, J.: Comparison of martingale difference sequences in Probability in Banach Spaces V. Lecture Notes in Math., vol. 1153, pp. 453–457 (1985)
Acknowledgements
The results of this paper are taken from my doctoral thesis [23]. I am grateful to my advisors: Fedor Nazarov and Michał Wojciechowski for their mentorship and support, especially their unending willingness to discuss my research.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
We present two proofs of Theorem 2.9 different from the original one by Zinn.
Let us recall the non-linear telescoping lemma due to Bourgain and Müller.
Lemma 6.1
[3, 19] Let \(\lambda _1,\ldots ,\lambda _n\), \(\varphi _1,\ldots ,\varphi _n\) be nonnegative random variables such that
Then
Corollary 6.2
Let \(X_1,\ldots ,X_n\) be independent and set
Then
Proof
The right inequality of (6.4) follows directly from Lemma 6.1, since \(\lambda _k\) is an increasing sequence of constants. To prove the other inequality, we see that conditioning with respect to \(\sigma \left( X_{k+1},\ldots ,X_n\right) \) gives
thus by induction
which for \(k=n\) is the desired inequality.\(\square \)
Proof of Theorem 2.9
In order to prove the \( > rsim \) inequality in (2.38), we merely perform a slight modification of the proof of Lepingle inequality presented in [3]. Let us denote \(\Omega ^N\times \Omega ^N\ni (x,y)\mapsto f_n\left( x_1,\ldots ,x_{n-1},y_n\right) \) by \(\tilde{f_n}\). By tensoring \(\left( f_n\right) \) against the Rademacher sequence, we may assume that it is a martingale difference sequence. Then the left hand side equals \(\Vert F\Vert _{H^1}\), where \(F=\sum _{n=1}^\infty f_n\) and \(f_n=\Delta _n F\). By Theorem 2.6 it is enough to check the boundedness of the right hand side in the case when F is an atom, because we have an a priori bound for finite sums. Let \(F=u-{\mathbb {E}}_{j-1}u\), where u satisfies (2.17). Then
By \(A\in {\mathcal {F}}_j\), the support of \({\mathbb {E}}_k u \) for \(k\ge j\) is contained in A as well, because
Thus for \(k>j\) we have \(\mathrm {supp}\,f_k\subset \mathrm {supp}\,{\mathbb {E}}_k u\cup \mathrm {supp}\,{\mathbb {E}}_{k-1}u\subset A\). Consequently
because if \(\left( x,y\right) \in \mathrm {supp}\,{\tilde{f}}_k\), then \(\left\{ \left( x_1,\ldots ,x_{k-1},y_k\right) \right\} \times \Omega ^{\left\{ k+1,\ldots \right\} }\subset \mathrm {supp}\,f_k\subset A\), which by \(A\in {\mathcal {F}}_j\) implies \(x\in A\). By and (6.8) we have
Combining (2.17), (6.8), (6.12), (6.13) with the inequality
and the fact that the projections \(\Delta _k\) are mutually orthogonal, we obtain
We will prove the \(\lesssim \) inequality in (2.38) now. It is clear that it is enough to prove it with only finitely many of \(f_k\) nonzero. We define the sequence \(\left( \lambda _k(x)\right) _{k=1}^n\) of functions in \(L^1\left( \Omega ^\infty \right) \) inductively by
For any fixed \(x\in \Omega ^\infty \), this sequence coincides with the sequence defined by (6.3) applied to the independent random variables \(f_k\left( x_1,\ldots ,x_{k-1},\cdot \right) \in L^1(\Omega )\), so the left part of (6.4) yields the pointwise inequality
By induction it is obvious that \(\lambda _k\) is \({\mathcal {F}}_{k-1}\)-measurable. Thus
In particular, \(\lambda _k\) verify the condition (6.1) with respect to \(f_k\) and are pointwise increasing, so Lemma 6.1 gives
Integrating (6.22) with respect to x and applying (6.25) we obtain
\(\square \)
Proof
(Yet another proof of Theorem 2.9) Without loss of generality, we may assume that \(\Omega \) has a structure of a compact abelian group with Haar measure, e.g. by embedding \(\Omega \) in \({\mathbb {T}}\). Just like previously, we also may assume that \(f_k\) is a k-th martingale difference and notice that the left hand side is just \(\left\| \sum _k f_k\right\| _{H^1}\). For \(\xi \in \Omega ^N\) we define an operator \(T_\xi \) by
Since \(\Delta _k T_\xi f\) is just a translation of \(\Delta _k f\),
For \(k>n\), by translation in the variable \(y_k\),
Therefore
We have \(T_\xi ^*=T_{-\xi }\), because
By Theorem 2.7, \(T_{\xi }\) are uniformly bounded on martingale \(H^1\), so
and thus
Ultimately, by translating \(\xi \) for fixed x,
\(\square \)
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Rzeszut, M., Wojciechowski, M. Hoeffding decomposition in \(H^1\) spaces. Math. Z. 298, 1113–1141 (2021). https://doi.org/10.1007/s00209-020-02627-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00209-020-02627-0