New estimates for bounded oscillation operators

Fix an arbitrary point x_0\in X and let \ensuremath{\mathscr A}\subset \ensuremath{\mathscr B} be the family of balls containing x_0. By B2) condition we can write X=\cup_{A\in \ensuremath{\mathscr A}}A. Then there exists a B\in \ensuremath{\mathscr A} such that \mu(B)>\frac{1}{2} \sup_{A\in \ensuremath{\mathscr A}}\mu(A) and using B4) condition we obtain

Thus X=[B]\in \ensuremath{\mathscr B}.

Fix an arbitrary x_0\in X and let \ensuremath{\mathscr A}\subset \ensuremath{\mathscr B} be the family of balls containing the point x_0. Choose a sequence \eta_n\nearrow\eta=\sup_{A\in \ensuremath{\mathscr A}}\mu(A), where \eta can also be infinity. Let us see by induction that there is an increasing sequence of balls G_n\in \ensuremath{\mathscr A} such that \mu(G_n)> \eta_n and G_n^*\subset G_{n+1}. The base of induction is trivial. Suppose we have already chosen the first balls G_k, k=1,2,\ldots, l. There is a ball A\in \ensuremath{\mathscr A} so that \mu(A)>\eta_{l+1}. Let A_l be the biggest in measure among two balls A and G_l. By property B4) we have A\cup G_l\subset A_l^* and there exists a ball G_{l+1}\supset A_l^{**}. This implies \mu(G_{l+1})\ge \mu(A)>\eta_{l+1} and G_{l+1}\supset G_l^*, completing the induction. Let us see that G_n is our desired sequence of balls. Indeed, let B be an arbitrary ball. By B2) property there is a ball A containing x_0, such that A\cap B\neq \varnothing. Then by property B4) we may find a ball C\supset A\cup B. For some n we will have \mu(C)\le 2\mu(G_n) and so once again using B4), we get B\subset C\subset G_{n}^*\subset G_{n+1}.

The boundedness of E implies E\subset B for some B\in \ensuremath{\mathscr B}. If there is a ball G\in{\mathscr G\,} so that G\cap B\neq \varnothing, \mu(G)> \mu(B), then by two balls relation we will have E\subset B\subset G^{*}. Thus our desired sequence can be formed by a single element G. Hence we can suppose that every G\in{\mathscr G\,} satisfies G\cap B\neq \varnothing, \mu(G)\le \mu(B) and again by two balls relation G\subset B^*. Therefore, \bigcup_{G\in{\mathscr G\,}}G\subset B^{*}. Choose G_1\in {\mathscr G\,}, satisfying \mu(G_1)> \frac{1}{2}\sup_{G\in{\mathscr G\,}}\mu(G). Then, suppose by induction we have already chosen elements G_1,\ldots,G_k from {\mathscr G\,}. Choose G_{k+1}\in {\mathscr G\,} disjoint with the balls G_1,\ldots,G_k such that

If for some n we will not be able to determine G_{n+1} the process will stop and we will get a finite sequence G_1,G_2,\ldots, G_n. Otherwise our sequence will be infinite. We shall consider the infinite case of the sequence (the finite case can be done similarly). Since the balls G_n are pairwise disjoint and G_n\subset B^{*}, we have \mu(G_n)\to 0. Choose an arbitrary G\in{\mathscr G\,} with G\neq G_k, k=1,2,\ldots and let m be the smallest integer satisfying \mu(G)\ge 2\mu(G_{m+1}). So we have G \cap G_j\neq\varnothing for some 1\le j\le m, since otherwise by (Proof), G had to be chosen instead of G_{m+1}. Besides, we have \mu(G)< 2\mu(G_{j}) because of the minimality property of m, and so by two balls relation G\subset G_{j}^{*}. Since G\in {\mathscr G\,} was chosen arbitrarily, we get E \subset\bigcup_{G\in {\mathscr G\,}} G\subset \bigcup_k G_k^{*}.

Applying Lemma 2.2 one can check that it is enough to establish the density property for the bounded measurable sets. Suppose to the contrary there exist a bounded measurable set E\subset X together its subset F\subset E (maybe non-measurable) with an outer measure

such that

where 0<\alpha<1. According to the definition of the outer measure one can find a measurable set \bar F such that

By B3)-condition there is a sequence of balls B_k, k=1,2,\ldots, such that

We discuss two kind of balls B_k, satisfying either B_k\cap F=\varnothing or B_k\cap F\neq\varnothing. For the first collection we have

which implies

Thus, combining also (Proof) we obtain

For the balls B_k\cap F\neq\varnothing, using (Proof) and (Proof) we have

Since E and so \bar F are bounded, applying Lemma 2.3 and (Proof), one can find a subsequence of pairwise disjoint balls \tilde B_k, k=1,2,\ldots, such that

Thus, from B4)-condition, (Proof), (Proof) and (Proof), we obtain

Since \varepsilon can be arbitrarily small, we get \mu(F)=0 and so a contradiction.

Suppose G_1,G_2,\ldots, G_N are some elements of {\mathscr G\,} with hull-balls satisfying \mu([G_1])\ge \max_j\mu([G_i]). Consider the ball

Using (2.5), one can check that for both cases in (Proof) we have G_k\subset [G_k]\subset B^{**}, k=1,\ldots, N. Thus, since G_k are pairwise disjoint, from (2.5) we obtain

which completes the proof of lemma.

Let B\in \ensuremath{\mathscr B}, B^{*}\neq X. Suppose to the contrary there is no ball B' satisfying (2.6) and denote

According to the assumption every A\in\ensuremath{\mathscr A} should satisfy the bound \mu(A)<2\mu(B). This and B4) imply \bigcup_{A\in\ensuremath{\mathscr A}}A\subset B^*\subset [B]\in \ensuremath{\mathscr B}, where \mu([B])\le \ensuremath{\mathcal K}\mu(B). Thus, we have [B]\in \ensuremath{\mathscr A} and so \bigcup_{A\in\ensuremath{\mathscr A}}A= B^*=[B]. On the other hand, according to the doubling condition, there is a ball A'\supsetneq [B] such that \mu(A')\le\eta\cdot \mu([B])\le \eta\ensuremath{\mathcal K}\mu(B). The latter implies A'\in\ensuremath{\mathscr A} and therefore A'\subset [B] that gives a contradiction. Lemma is proved.

The case of \mu(B)\le 2\mu([A]) is trivial, we may simply choose A_0=A and A_1=B. So we can suppose \mu(B)> 2\mu([A]). Applying Lemma 2.6 we may find a sequence of balls A=B_0, B_1,\ldots, B_{n-1} such that [B_k]\subset B_{k+1} and 2\mu([B_k])\le \mu(B_{k+1})\le \eta \ensuremath{\mathcal K} \mu([B_k]), \mu(B)\le\mu(B_{n-1})\lesssim \mu(B). Obviously, we may define our desired sequence by A_0=A, A_n=[B] and A_k=[B_k], k=1,2,\ldots,n-1.

First consider a general ball-basis. Let D be the density points set of F (see Definition 2.1). Thus for any x\in D the family of balls

is nonempty and therefore there is a ball B_x\in {\mathscr {C}}_x such that \mu(B_x)>\frac{1}{2}\sup_{C\in {\mathscr {C}}_x}\mu(C). From the two ball relation it follows that \cup_{C\in {\mathscr {C}}_x}C\subset B_x^*. Thus, if a ball G' satisfies G'\supsetneq B_x^*, then x\in G'\notin {\mathscr {C}} and so

Since \{B_x:\,x\in E\cap D\} is a ball-covering for E\cap D, applying Lemma 2.3, we find sequence of pairwise disjoint balls \{B_{x_k}\} such that

Choose a hull-ball G_k\supset B_{x_k}^* such that \mu(G_k)\le \ensuremath{\mathcal K}\mu(B_{x_k}). We claim that the family {\mathscr G\,}(E)=\{G_k\} satisfies the conditions of lemma. Indeed, (3.1) follows from the observation in (Proof). By density Lemma 2.4 we have \mu(F\setminus D)=0 and combining also (Proof), we obtain (3.1). Then, since B_x\in {\mathscr {C}}_x, it follows that

that gives (3.1). This completes the proof in the case of non-doubling ball-basis. Now let \ensuremath{\mathscr B} is doubling. Applying the above procedure first we get {\mathscr G\,}(E), then replace each G\in {\mathscr G\,}(E) by a ball \tilde G\supsetneq G satisfying \mu(\tilde G)\le \eta\mu(G). Clearly, all the conditions of Lemma 3.1 will be satisfied for the new collection as well, besides we will have (3.1) for G'=G.

Without loss of generality we can suppose that

since otherwise one can replace every G\in \ensuremath{\mathscr A} by G\cap \left(\cup_{A\in \ensuremath{\mathscr A}}E_A\right). The family \bar \ensuremath{\mathscr A} will be constructed inductively as follows. Since \ensuremath{\mathscr A} is countable we can write \ensuremath{\mathscr A}=\{A_k,\, k=1,2,\ldots\}, using an arbitrary numeration as well as keeping the same children relation between the elements. We will inductively change the terms of \ensuremath{\mathscr A} as follows. At the n-th stage of induction every element A\in \ensuremath{\mathscr A} will be replaced by A\setminus A_n\cap E_{A_n} provided A_n\notin \{A\}\cup {\mathfrak {Gen}}(A). Finishing the induction procedure we will finally get the desired sequence \bar \ensuremath{\mathscr A}=\{\bar A_k,\, k=1,2,\ldots\} as a limiting version of the sequence \ensuremath{\mathscr A}_n. Let us check the claims of the lemma. 1) To prove that \bar A_k\cap E_{A_k} are pairwise disjoint let \bar A_n and \bar A_j be two different elements of \bar \ensuremath{\mathscr A}. Without loss of generality we can suppose that A_n\notin {\mathfrak {Gen}}(A_j). So one can check that after the n-th stage of induction A_n\cap E_{A_n} and A_j\cap E_{A_j} became disjoint, staying disjoint during the next stages of induction. 2) For (3.2) one just need to check that the union \cup_{A\in \ensuremath{\mathscr A}}A\cap E_A is not changed after any stage of induction. 3) To prove relation (3.2) observe that for two sets G\in {\mathfrak {Ch}}(A) the relation G\subset A that we have at the beginning, will be satisfied after any stage of induction. Indeed, if we are at the n-th stage and A_n\in \{A\}\cup {\mathfrak {Gen}}(A), then A is not changed by the n-th stage action. If A_n\notin \{A\}\cup {\mathfrak {Gen}}(A), then both G and A are reduced similarly. So in both cases we will still have G\subset A. 4) To show (3.2) we need to see that if A and B are not in a parental relationship, then \bar A\cap \bar B=\varnothing. By (3.2) and (Proof) it is enough to show that

for any n=1,2,\ldots. It is clear that each A_n is not in the generation of either A or B. Thus the set \bar A_n\cap E_{A_n} has been removed at least from one of the sets A or B at the n-th stage of the induction, which yields (Proof). With this we complete the proof of lemma.

Consider the set E=\{ x\in X:\, \ensuremath{\mathcal M} f(x)>\lambda\}, which can be non-measurable. For any x\in E there exists a ball B(x)\subset X such that

which implies

We have E=\cup_{x\in E} B(x). Given G\in\ensuremath{\mathscr B} consider the collection of balls \{B(x):\, x\in E\cap G\}. Applying Lemma 2.3, we find a sequence of pairwise disjoint balls \{B_k\} taken from this collection such that E \cap G\subset \cup_kB_k^*=Q(G). Applying the inequality \sum_kx_k^d\le (\sum_kx_k)^d for positive numbers x_k and for d=\varrho/\rho\ge 1, we get

According to Lemma 2.2 there is a sequence of balls G_1\subset G_2\subset \ldots such that X=\cup_{k}G_k, so we conclude

and so (4.1).

Let B be a \gamma-ball so there are balls B_1\subset B\subset B_2 such that \mu(B_2)\le \gamma \mu(B_1). Thus for f\in L^r(X,\ensuremath{\mathbb U}) we can write

for some ball \bar B_1\supset B_1. If \mu(\bar B_1)\le \mu(B_2), then we have \bar B_1\subset B_2^* and therefore

Otherwise, if \mu(\bar B_1)\ge \mu(B_2), then we will have B\subset B_2\subset \bar B_1^{*} and then

Thus we obtain (4.1).

Let f\in L^r(X,\ensuremath{\mathbb U}), A\subset B\subset C and x\in B. We have

that completes the proof.

First consider the case when A=B and so \Delta(A,B)=0. Applying (?) with

we get

and so we find a point x_0\in B such that

According to T1)-condition and Lemma 4.1, for any x\in B we also have

By the definition of \langle f \rangle^*_{B} there is a ball G\supset B such that

If \mu(G)\le \mu(C), then we have G\subset C^{*} and therefore

If \mu(G)> \mu(C), then C^*\subset G^{**}. Hence we get

The combination of (Proof), (Proof) and (Proof) imply the inequality

which together with (Proof) and (Proof) gives

The latter implies

that is (4.3) if A=B. Now let A be arbitrary. Thus, for f\in L^r(X,\ensuremath{\mathbb U}) and x\in A we have

which is the full version of (4.3).

Let balls B\supsetneq A satisfy the conditions of lemma. Using T2)-condition, we find a ball C\supsetneq A such that

If \mu(C)>\mu(B), then we have B\subset C^{*}. Thus, applying Lemmas 4.2 and 4.3, we obtain

In the case \mu(C)\le \mu(B) we have C\subset B^{*}, and since C\supsetneq A, we obtain

Thus, once again applying Lemmas 4.2, 4.3 and (Proof), we conclude

This completes the proof of lemma.

Let a pair (G,A) of balls G\supsetneq A satisfy T2) condition, that is \Delta(A,G)\le \ensuremath{\mathcal L}_2. If for a hull-ball [G] we have \mu([G])\le \mu(B), then G^*\subset [G]\subset B^* and one can choose \tilde A=G in (4.6). In the case of \mu([G])>\mu(B) we have B\subset [G]^{*}. By Lemma 4.3 the pair (A, [G]^{*}) also satisfies T2)-condition, and according to Lemma 4.2, so we have for (A,B). Hence we can choose \tilde A=B. One can check that for both cases we will have (4.6). The second part of the lemma may be proved similarly.

Denote by E=E_\lambda the set on the left of (4.2), which can also be non-measurable. By the definition of T^* for any x\in E there is a ball B(x)\in\ensuremath{\mathscr B} such that

One can check that E\subset \cup_{x\in E}B(x) and

Given ball G apply Lemma 2.3, we find a sequence x_k\in E such that the balls \{B_k=B(x_k)\} are pairwise disjoint and

Since T satisfies T1)-condition, by (Proof) we obtain

Thus, we conclude from (Proof) that

For a \beta>0 we define measurable sets

Using T0)-condition, the measure of the complement of \tilde B_k is estimated by

Using also (Proof), for an appropriate constant \beta\sim \ensuremath{\mathcal L}_0 we have

If x\in \tilde B_k, then, using sublinearity of T together with relations (Proof) and (Proof), we obtain

where the last inequality can be satisfied for \delta= 1/2(\ensuremath{\mathcal L}_1+\beta)\sim (\ensuremath{\mathcal L}_1+ \ensuremath{\mathcal L}_0)^{-1}. Hence we conclude \cup_k\tilde B_k\subset \{\|Tf(x)\|_\ensuremath{\mathbb V}>\lambda/2\} and taking into account also (Proof) and (Proof), we get

Since G is arbitrary, applying Lemma 2.2 we complete the proof.

Weak type inequality (?) or condition (4.3) easily follows from Theorems 4.1 and 4.2. One can also check that for any balls A,B satisfying A\subset B we have

Thus, if balls A and B satisfy T2)-condition for the operator T, then the same conditions hold also for T^* with \ensuremath{\mathcal L}_2(T^*)=\ensuremath{\mathcal L}_2(T). To prove (4.3), let B\in \ensuremath{\mathscr B} and f\in L^r(X,\ensuremath{\mathbb U}) satisfy

Choose arbitrary points x,x'\in B and let estimate |T^*f(x)- T^*f(x')|. If T^*f(x)= T^*f(x'), then the estimation is trivial. So we can suppose that T^*f(x)> T^*f(x'). Using the definition of T^*f(x), we find a ball A\ni x such that

First suppose that \mu(A^*)>\mu(B). So we can say that x'\in B\subset A^{**}\subset A', where A' is a ball satisfying \mu(A')\le \ensuremath{\mathcal K}^2\mu(A). By definition of T^* we can write

which together with (Proof) yields

Besides, since x\in A, from (4.1) (or Lemma 4.3) and (Proof) it follows that

Thus, applying the T1)-condition for T, from (Proof) and (Proof) we conclude

that implies (4.3). Now suppose \mu(A^*)\le \mu(B) and therefore A^*\subset B^*. Then by (Proof) we get f\cdot \ensuremath{\textbf 1}_{X\setminus A^{*}}=f\cdot \ensuremath{\textbf 1}_{X\setminus B^{*}}. Thus, using the same argument, we can write

that again gives (4.3).

Let A\in\ensuremath{\mathscr B} be an arbitrary ball and {\rm supp\,} f\subset X\setminus A^{*}. Take arbitrary points x,x'\in A and suppose that Tf(x)\ge Tf(x'). According to the definition of T, for any \delta> 0 there exists an index \alpha_0 such that Tf(x)\le \|T_{\alpha_0}f(x)\|_\ensuremath{\mathbb V}+\delta. On the other hand for the same \alpha_0 we have Tf(x')\ge \|T_{\alpha_0} f(x')\|_\ensuremath{\mathbb V}. Thus, applying T1) property of T_{\alpha_0}, we obtain

Since \delta>0 can be arbitrarily small, we get (4.4). (4.4) follows from Lemma 4.5, according which we can choose a ball B independent of the operators T_\alpha such that

for any \alpha. This implies (4.4), completing the proof of theorem.

We need to check T2)-condition. Let A be an arbitrary ball and B satisfy conditions (1.2). Then applying (4.1), we get the bound

which implies (4.5).

Given 0<\beta<1. For any ball A\in \ensuremath{\mathscr B} we may fix a set E_{A}\subset [A] such that

For the sets F_A=[A]\setminus E_A we have \mu(F_A)\le\alpha \mu([A])\lesssim\alpha \mu(A), where \alpha=1-\beta. Thus, applying Proposition 3.1, we find a sparse tree-collection of balls \ensuremath{\mathcal S}, satisfying a1)-a4). Since our ball-basis is doubling, we may also have (?) for G'=G. Besides, for a small enough \alpha inequality (?) may imply \mu([G])<\mu(A) and so we get [G]\subset A^*\subset [A] whenever G\in {\mathfrak {Ch}}(A). Hence we can write

Moreover, for an admissible \alpha, inequality (?) also implies

According to a2) we have

We also claim for a smaller enough \alpha that

Indeed, by (Proof) and (Proof) we have \mu(G\cap E_A)>\mu(G)/2. Thus, using also (Proof), we get

and so (Proof) follows. According to (Proof) the tree-collection of measurable sets \ensuremath{\mathscr A}=\{[A]:\, A\in \ensuremath{\mathcal S}\} satisfies (3.2). Thus, applying Proposition 3.2 and (Proof), we find a martingale family of measurable sets \bar \ensuremath{\mathscr A}\Subset \ensuremath{\mathscr A} such that

Thus for a.e. x\in A_0 we have x\in \bar A \cap E_A for some A\in \ensuremath{\mathcal S}. Fix a unique chain sequence of balls A_0, A_1,\ldots ,A_n=A such that A_{j+1}\in {\mathfrak {Ch}}(A_j), j=0,1,\ldots,n-1. According to the martingale property of \bar \ensuremath{\mathscr A}, the relation x\in \bar A=\bar A_n implies

Using (Proof), we find points \xi_j\in E_{A_{j-1}}\cap E_{A_{j}}, j=1,2,\ldots, n-1. Set also \xi_n=x. Since \xi_{j},\xi_{j+1}\in E_{A_j}, from (Proof) it follows that

In addition, if \beta>1/2 from (Proof) it follows that E_{A_0}\subset M_f(A_0)(see Definition 6.1). Then, since \xi_1\in E_{A_0}, by Lemma 6.1 we can write

Applying, (Proof), (Proof) and (Proof), we obtain

which implies (6.1). Relation (6.1) immediately follows from (Proof) and (Proof) implies (6.1). This completes the proof.

According to Lemma 2.7, we may find a sequence of balls A=A_0\subset A_1\subset \ldots\subset A_n=[B] satisfying such that \mu(A_{k+1})\lesssim \mu(A_k). One can easily check that n\lesssim \log \left(1+\mu(B)/\mu(A)\right). By Lemma 7.1 we have

Thus we obtain

First apply Corollary 8.3 for the identical operator T(f)=f, which is obviously a restricted type of {\rm BO} operator with parameter r=1. We have \ensuremath{\mathcal M}_\#f(x)\le \|f\|_{{\rm BMO }} for all x\in X, where we suppose r=1 in the definition of \ensuremath{\mathcal M}_\#. Thus, applying (8.3), we obtain the John-Nirenberg inequality

where by a standard argument m can be also replaced by f_B. Returning to the general case of r\ge 1 from (Proof) it follows that

Now let T be the operator in Corollary 8.3. From (8.3) and (Proof) we obtain

then by the same standard argument we get \|T(f)\|_{\rm BMO }\lesssim \|f\|_{\rm BMO }.

Sparse domination (1.1) implies boundedness of T on L^p for any r<p<\infty. Namely we can write

Take a function f\in L^\infty (X,\ensuremath{\mathbb U}) and for B\in \ensuremath{\mathscr B} denote

Strong-subadditivity of T implies

Hence, applying T1)-condition and (Proof), we get

Let f\in L^r(\ensuremath{\mathbb R}^d) and B=B(x_0,R)\subset \ensuremath{\mathbb R}^d be an Euclidean ball of center x_0 and radius R. One can check that B^*=B(x_0,R^*), where R^*=(1+2\cdot 2^{1/d})R\in [R,5R]. If any point x\in B=B(x_0,R), using (?) and (9.1), we get

where \sup is taken over all the balls A\supset B. Thus we get R4).

The case of \alpha=0 for the maximal operator \ensuremath{\mathcal M}_\alpha was proved in [Kar3] and the general case may be proved similarly. So let us consider the operator (?). We need to prove only conditions T0) and T1), since the Euclidean ball-basis is doubling. It is well-known that I_\alpha is of weak (1,q) type with q satisfying q^{-1}=\rho= 1-\alpha/n. This implies T0) condition with parameters (9.3). Then suppose B=B(x_0,R) and x,x'\in B, y\in B^*, are arbitrary points. One can check that

Thus for f\in L^1(\ensuremath{\mathbb R}^d) we obtain

which implies T1) condition.

Suppose to the contrary that d(x,A)< \mu(B), which means there is a ball A'\supset A such that x\in A' and \mu(A')<\mu(B). Then we get A'\subset B^*, which is in contradiction with x\notin B^*.

The proof of the left hand side of the inequality is straightforward. To prove the right hand side inequality, we will need the following standard inequalities

where f\in L^r(X,\ensuremath{\mathbb U}) and B is an arbitrary ball. For any x\in B there exists a ball B(x)\ni x such that

Applying Lemma 2.3, we find a sequence of pairwise disjoint balls \{B_k\}\subset \{B(x):\, x\in B\} such that \cup_kB_k^*\supset B. If some B_k satisfies \mu(B_k)>\mu(B), then we have B\subset B_k^* and, using (Proof), we get

If \mu(B_k)\le \mu(B) for every k, then \cup_kB_k\subset B^*. Therefore by (Proof), (Proof) and the pairwise disjointness of B_k we obtain

which gives the right inequality in (10.2).

Applying the doubling condition we find a sequence of balls

For x\in B_{n}^*\setminus B_{n-1}^* by Lemma 10.1 we have

Thus, using (?), we obtain

Then by Lemma 7.2 we have

Thus we obtain

Applying this inequality with B=A we get

then, combining (Proof) and (Proof), we obtain