Strong-type endpoint obstructions for bilinear spherical maximal operators

Choose f\in C_c^\infty(\mathbb R^d) nonnegative, supported in B(0,1), with f\ge1 on B(0,1/2). Choose \eta\in C^\infty([0,\infty)) with 0\le\eta\le1, \eta=0 on [0,3] and \eta=1 on [4,\infty).

Suppose first 1\le q<\infty. Select \alpha with

and define the smooth radial function, set equal to zero near the origin,

Since \alpha q>1,

For q=\infty, set g\equiv1 and \alpha=0.

For k\ge6, write t=2^k and consider the disjoint shells and integration balls

When x\in D_k and y\in Y_{t,x}, we have |x-ty|\le1/2 and |y|\le3/5+1/(2t)<5/8. Hence f(x-ty)\ge1 and \sqrt{1-|y|^2}>3/4. For every \omega\in S^{d-1}, the triangle inequalities imply

Since t\ge64, the lower endpoint is greater than 4. The cutoff in (?) is therefore identically one throughout this region. In the case q<\infty, monotonicity of the radial tail gives

The same inequality holds with q=\infty, \alpha=0. The weight in (?) is at least (39/64)^{(d-2)/2} on Y_{t,x}, whose volume equals |B(0,1)|(2t)^{-d}. Restricting the integral to that ball thus gives

The shells are pairwise disjoint and satisfy |D_k|=|B(0,1)|((3/5)^d-(1/2)^d)2^{kd}. Consequently

by (Proof); for q=\infty the same calculation uses r=1 and \alpha=0. Finally, the averages are symmetric in their inputs and \mathcal M_{\mathrm{full}}\ge\mathcal M_{\mathrm{lac}} pointwise.

Take the preceding f, and let \chi_N\in C_c^\infty(\mathbb R^d) be a radial cutoff with 0\le\chi_N\le1, equal to one on 4\le|u|\le 2^{N+1} and supported in 2<|u|<2^{N+2}. Set g_N(u)=\chi_N(u)|u|^{-d/q}, where |u|^{-d/\infty}=1. For finite q, polar coordinates yield \|g_N\|_q^q\lesssim\int_2^{2^{N+2}}d\rho/\rho\lesssim N; for q=\infty its supremum norm is at most one. For 6\le k\le N, (?) lies in the region where \chi_N=1. The lower bound (?) holds with \alpha=0, so the rth power of the output quasi-norm is at least c(N-5). Dividing the resulting estimate by \|f\|_1\|g_N\|_q gives (N-5)^{1/r}/N^{1/q}\gtrsim N, since 1/r-1/q=1.

For large |u| the radial function g_\alpha is decreasing and its decreasing rearrangement satisfies g_\alpha^*(v)\asymp v^{-1/q}(\log v)^{-\alpha} as v\to\infty. Its boundedness and vanishing near the origin handle bounded v. This proves (3) by direct integration.

For any \alpha\ge0, the geometric argument leading to (?) remains valid, independently of whether g_\alpha belongs to L^q. Put H=\mathcal M_{\mathrm{lac}}(f,g_\alpha) and write m_k=|D_k|=c_d'2^{kd}, with c_d'>0. By (?) and the definition of rearrangement, H^*(m_k/2)\gtrsim 2^{-kd/r}k^{-\alpha}. Fix a,b with 2^{-d}b<a<b<1/2. The intervals I_k=[am_k,bm_k] are disjoint, and monotonicity gives H^*(v)\gtrsim2^{-kd/r}k^{-\alpha} for v\in I_k. Therefore

which diverges under the stated condition.

When s\le t, (?)–(?), followed by the embedding L^{r,s}\hookrightarrow L^{r,t}, prove both estimates. Conversely, if t<s<\infty, choose 1/s<\alpha\le1/t. Lemma 3 provides an input g_\alpha\in L^{q,s} for which the lacunary output fails to belong to L^{r,t}. The same construction works for s=\infty and finite t upon choosing 0\le\alpha\le1/t. Because \mathcal M_{\mathrm{full}}\ge\mathcal M_{\mathrm{lac}} on these nonnegative inputs, both bounds fail.

Disintegration with respect to the first d coordinates gives

where y'=(y_2,\ldots,y_m), \omega=(\omega_2,\ldots,\omega_m), and

The normalization follows by the same beta integral as (?). Fix f_1=f from Section ?. Set P=\sum_{j=2}^m1/q_j. For each finite q_j, choose \alpha_j>1/q_j so that

and let f_j(u)=\eta(|u|)|u|^{-d/q_j}(\log|u|)^{-\alpha_j}, extended by zero near the origin. If q_j=\infty take f_j\equiv1. Condition (Proof) is possible by choosing positive excesses whose sum is smaller than one; all these inputs belong to their stated spaces by (?).

Fix x\in D_k, t=2^k, and restrict y_1 to Y_{t,x} in (Proof). The marginal weight is bounded from below and this ball has measure comparable to t^{-d}. On the residual sphere impose

Sign changes \omega_j\mapsto-\omega_j preserve residual surface measure; the boundaries have zero measure and so \sigma(E_x)=2^{-(m-1)}. Writing b=\sqrt{1-|y_1|^2}\le1, for every \omega\in E_x and j\ge2 we have

In particular all finite-q_j cutoffs equal one on this region and f_j(x-tb\omega_j)\gtrsim t^{-d/q_j}k^{-\alpha_j}; the same bound with zero exponents holds for constant inputs. Hence

Integrating its rth power over disjoint shells gives \sum_{k\ge6}k^{-Ar}=\infty by (Proof). Symmetry permits the L^1 entry in any position, and domination gives the full maximal conclusion.