Endpoint estimates for one-dimensional Fourier multipliers of bounded 2-variation: do $V_2$ multipliers map $H^1$ and $L\log^{1/2}L$ into $L^{1,\infty}$?

A Fourier multiplier operator $T_m f=(m\hat f)^{\vee}$ is of class $V_s$ if its dyadic frequency components have uniformly bounded $s$-variation on the support interval of a fixed bump. Tao and Wright (2001) proved the sharp near-$L^1$ endpoint theory for the Marcinkiewicz class ($s=1$) and for the $R_2$ class of Coifman, Rubio de Francia, and Semmes, and extended it to $V_s$ for $1\le s<2$; for the critical index $s=2$ they conjectured, but could not prove, that $V_2$ multipliers map $H^1(\mathbb{R})$ into $L^{1,\infty}(\mathbb{R})$ and locally map $L\log^{1/2}L$ into $L^{1,\infty}$, both conjecturally sharp; they left undetermined the endpoint behavior of $V_s$ for $s>2$ (the interpolation lines $|1/2-1/p|=1/s$) and, at the critical index $s=2$, the sharp $L^p$ blow-up rate as $p\to1^+$. Later work has resolved neighboring axes (higher lacunary order, multiparameter, weighted and sparse bounds, noncommutative and operator-valued settings), but every variation-based endpoint result still stops at indices strictly below $2$, so the $s\ge2$ endpoint problem, with the $V_2$ conjecture at its core, remains open.

Background

For a bounded function $m$ on $\mathbb{R}$, the associated Fourier multiplier operator is $T_m f=(m\hat f)^{\vee}$, where $\hat f$ is the Fourier transform and $^{\vee}$ its inverse. Classical multiplier theorems give conditions on $m$ under which $T_m$ is bounded on $L^p(\mathbb{R})$, $1<p<\infty$; the delicate question is the behavior of $T_m$ near $L^1$. The relevant endpoint function spaces are: the real Hardy space $H^1(\mathbb{R})$, the atomic subspace of $L^1$ whose atoms are $L^\infty$-functions supported on intervals with mean zero and $L^\infty$-norm controlled by the inverse interval length; the Lorentz spaces $L^{1,q}$, which refine $L^1$ (for a function $g$, $\|g\|_{L^{1,q}}\approx\big(\int_0^\infty(\lambda\,|\{|g|>\lambda\}|)^{q}\frac{d\lambda}{\lambda}\big)^{1/q}$, with $q=\infty$ the weak-$L^1$ space $L^{1,\infty}$); and the Orlicz spaces $L\log^r L$, defined by the integrability condition $\int |f|\,(\log(2+|f|))^r<\infty$. One says that $T_m$ locally maps $X$ to $L^{1,q}$ if $\|T_m f\|_{L^{1,q}(K)}\le C_K\|f\|_{X(K)}$ for every compact set $K$; locality is needed because translation-invariant operators do not act globally on $L^1$-type spaces. Extrapolation theory links these endpoint bounds to quantitative $L^p$ theory: by Yano's extrapolation theorem and Tao's converse extrapolation theorem for translation-invariant operators, a local $L\log^r L\to L^{1,\infty}$ bound corresponds to $L^p\to L^p$ operator norms growing like a power of $(p-1)^{-1}$ as $p\to1^+$, so the sharp Orlicz exponent and the sharp $L^p$ blow-up rate are two faces of the same question.

Fix an even bump function $\phi$ adapted to the interval $[1/2,4]$ and equal to $1$ on $[3/4,3]$, as in Tao and Wright. The $j$-th frequency component of a symbol $m$ is $m_j(\xi)=\phi(\xi)\,m(2^j\xi)$, so that $m$ is reassembled from the pieces $m_j(2^{-j}\cdot)$, each supported on a dyadic frequency band. A Marcinkiewicz multiplier is one whose components $m_j$ have uniformly bounded variation (total $V_1$-norm) on $[1/2,4]$; the Marcinkiewicz multiplier theorem gives $L^p$ boundedness for $1<p<\infty$. Coifman, Rubio de Francia, and Semmes introduced the larger class $R_2$: each $m_j$ lies in the atomic space generated by sums $\sum_I c_I\,\mathbf{1}_I$ over disjoint subintervals of $[1/2,4]$ with square-summable coefficients, and they proved $R_2$ multipliers are bounded on every $L^p$, $1<p<\infty$. Tao and Wright proved the sharp endpoint theory for both classes: Marcinkiewicz and $R_2$ multipliers map $H^1$ to $L^{1,\infty}$ and locally map $L\log^r L$ to $L^{1,q}$ exactly when $r\ge 1/2+1/q$, with $L^p$ operator norm growth $\max\{p,p'\}^{3/2}$, all best possible; the smoother $R_{1/2,2}^2$ class (components built from smooth bumps instead of interval indicators) maps $H^1$ to $L^{1,2}$ with growth $\max\{p,p'\}$.

For $1\le s<\infty$ the $s$-variation of a function $F$ on an interval is $\sup\big(\sum_{i=0}^{N-1}|F(a_{i+1})-F(a_i)|^s\big)^{1/s}$, the supremum over all partitions $a=a_0<\cdots<a_N=b$; a multiplier is of class $V_s$ if $\|m_j\|_{V_s}$ is uniformly bounded in $j$. The Marcinkiewicz class is $V_1$. Coifman-Rubio de Francia-Semmes showed $V_s\subset R_2$ for $s<2$, so Tao and Wright's Corollary 8.1 transfers the full sharp endpoint theory to $V_s$ for $1\le s<2$; for $s>2$ they showed (by interpolation, with earlier work of Hirschman) that $V_s$ multipliers are bounded on $L^p$ exactly in the range $|1/2-1/p|<1/s$, sharp up to endpoints by Hirschman's classical oscillatory multipliers $m(\xi)=e^{i|\xi|^\alpha}(1+|\xi|^2)^{-\beta/2}$, and $V_2$ multipliers are bounded on all $L^p$, $1<p<\infty$, but with only a non-sharp quantitative bound $O((p-1)^{-C})$ on the norm growth. The inclusion $V_s\subset R_2$ holds only for $s<2$, and no known argument extends it, or the endpoint theory it implies, to the critical index $s=2$: bounded $2$-variation controls square-summable increments over arbitrary partitions, a condition on how fast the components may move, while the $R_2$ atomic condition controls square-summable level values over disjoint intervals, and no argument is known that recovers the atomic structure, or the endpoint estimates it implies, from bounded $2$-variation alone. Tao and Wright state that they were unable to prove the conjectured $V_2$ estimates with their techniques, and they single out as a natural model case the subclass whose components are uniformly Hölder-continuous of order $1/2$, into which any function of bounded $2$-variation can be transformed by a change of variables, per Coifman-Rubio de Francia-Semmes. This is why their Section 8 leaves the $s\ge2$ endpoint behavior open.

Problem Statement

Determine the sharp near-$L^1$ endpoint mapping behavior of one-dimensional Fourier multiplier operators of class $V_s$ for variation indices $s\ge2$, with the critical case $s=2$ at the core. As in Tao and Wright (2001), for a bounded symbol $m$ let $m_j(\xi)=\phi(\xi)m(2^j\xi)$ be its frequency components ($\phi$ a fixed even bump adapted to $[1/2,4]$ and equal to $1$ on $[3/4,3]$), and call $T_m$ a $V_s$ multiplier if the $s$-variation of $m_j$ on $[1/2,4]$ is bounded uniformly in $j$. The problem has one objective, settled for $s<2$ and open for $s\ge2$, whose parts are logically tied by interpolation and extrapolation rather than independent. A complete solution must settle both bounds that (i) conjoins — the $H^1\to L^{1,\infty}$ bound and the local $L\log^{1/2}L\to L^{1,\infty}$ bound; a counterexample to either refutes (i) while leaving the other bound to be settled. If (i) is answered affirmatively, the $s>2$ endpoint Lorentz lines of (ii) — $1/p=1/s+1/2$ and its mirror $1/p'=1/2-1/s$ — follow by the cited complex interpolation and symmetry, and the growth-rate/Orlicz statement (iii) follows by extrapolation, so none of them is a separate demand; if (i) is refuted, those two $s>2$ endpoint lines of (ii) must still be determined, and any bound of (i) left unsettled by the refutation — in particular, when the local $L\log^{1/2}L$ bound survives, the equivalent sharp-exponent/blow-up statement (iii) — must be settled as well:

(i) The $V_2$ endpoint conjecture: prove or disprove that every $V_2$ multiplier maps $H^1(\mathbb{R})$ into $L^{1,\infty}(\mathbb{R})$ and locally maps $L\log^{1/2}L$ into $L^{1,\infty}$, with operator norm controlled by the uniform $2$-variation bound and $\|m\|_\infty$. Both exponents are known to be best possible, since the sharpness counterexamples of Tao and Wright are already Marcinkiewicz (hence $V_2$) multipliers; moreover, by Yano extrapolation and Tao's converse extrapolation theorem for translation-invariant operators, the local Orlicz part of the conjecture is equivalent to asking whether the $L^p\to L^p$ operator norms of $V_2$ multipliers grow at the sharp rate $O(\max\{p,p'\}^{3/2})$ as $p\to1^+$, the rate known for the $R_2$ class.

(ii) The endpoint lines for $s>2$: for $V_s$ multipliers with $s>2$, $L^p$ boundedness holds exactly for $|1/2-1/p|<1/s$ and is sharp up to endpoints; determine whether the endpoint Lorentz refinements hold, in particular whether $V_s$ multipliers map $L^p$ to $L^{p,p'}$ on the line $1/p=1/s+1/2$ — which, as observed by Tao and Wright, would follow from a positive answer to (i) by complex interpolation — and, by the operator-adjoint symmetry $m(\xi)\mapsto\overline{m(-\xi)}$ (equivalently, the dual of the $1/p=1/s+1/2$ line), whether they map the Lorentz space $L^{p',p}$ into $L^{p'}$ at the mirror endpoint $1/p'=1/2-1/s$ on the $p'>2$ side.

(iii) The sharp $L^p$ blow-up exponent as $p\to1^+$ for $V_2$, equivalently the sharp Orlicz exponent in the local $L\log^r L\to L^{1,\infty}$ bound: this is equivalent to the Orlicz part of (i) by extrapolation; only the non-sharp bound $O((p-1)^{-C})$ for some constant $C$ is currently known. (For $s>2$ the $L^p$ boundedness range $|1/2-1/p|<1/s$ of item (ii) stays away from $p=1$, so a blow-up question as $p\to1^+$ arises only at the critical index $s=2$.)

A negative resolution of (i) consists of a $V_2$ multiplier (components of uniformly bounded $2$-variation) that fails $H^1\to L^{1,\infty}$ or the local $L\log^{1/2}L\to L^{1,\infty}$ bound; note that such a symbol must still be bounded on every $L^p$, $1<p<\infty$, by the Coifman-Rubio de Francia-Semmes theorem. Tao and Wright single out as a natural model case the subclass whose components are uniformly Hölder-continuous of order $1/2$, into which any bounded-$2$-variation function can be transformed by a change of variables; this model case is a source-discussed special subclass and does not by itself resolve (i).

References

1. T. Tao and J. Wright, Endpoint multiplier theorems of Marcinkiewicz type, Rev. Mat. Iberoamericana 17 (2001), no. 3, 521-558. DOI 10.4171/rmi/303 (arXiv:math/0001050). URL: https://doi.org/10.4171/rmi/303
2. R. R. Coifman, J. L. Rubio de Francia, and S. Semmes, Multiplicateurs de Fourier de $L^p(\mathbb{R})$ et estimations quadratiques, C. R. Acad. Sci. Paris Sér. I Math. 306 (1988), no. 8, 351-354. Zbl 0643.42010. URL: https://zbmath.org/4048330
3. I. I. Hirschman Jr., On multiplier transformations, Duke Math. J. 26 (1959), no. 2, 221-242. DOI 10.1215/s0012-7094-59-02623-7. URL: https://doi.org/10.1215/s0012-7094-59-02623-7
4. Q. Xu, Fourier multipliers for $L^p(\mathbb{R}^n)$ via $q$-variation, Pacific J. Math. 176 (1996), no. 1, 287-296. DOI 10.2140/pjm.1996.176.287. URL: https://doi.org/10.2140/pjm.1996.176.287
5. T. Tao, A converse extrapolation theorem for translation-invariant operators, J. Funct. Anal. 180 (2001), no. 1, 1-10. DOI 10.1006/jfan.2000.3646 (arXiv:math/9912001). URL: https://doi.org/10.1006/jfan.2000.3646
6. S. Yano, Notes on Fourier analysis. XXIX. An extrapolation theorem, J. Math. Soc. Japan 3 (1951), no. 2, 296-305. DOI 10.2969/jmsj/00320296. URL: https://doi.org/10.2969/jmsj/00320296
7. S. V. Kislyakov, Weak type (1,1) in a refinement of the Marcinkiewicz theorem, Funct. Anal. Appl. 43 (2009), no. 3, 236-238. DOI 10.1007/s10688-009-0030-1. URL: https://doi.org/10.1007/s10688-009-0030-1
8. O. Bakas, V. Ciccone, I. Parissis, and M. Vitturi, Endpoint estimates for higher order Marcinkiewicz multipliers, Rev. Mat. Iberoamericana 41 (2025), no. 4, 1271-1304. DOI 10.4171/rmi/1518 (arXiv:2401.06083). URL: https://doi.org/10.4171/rmi/1518
9. O. Bakas, V. Ciccone, F. Di Plinio, M. Fraccaroli, I. Parissis, and M. Vitturi, Singular multipliers on multiscale Zygmund sets, preprint (2024). arXiv:2406.17521. URL: https://arxiv.org/abs/2406.17521
10. O. Bakas and I. Parissis, Endpoint estimates for multiparameter multipliers of Marcinkiewicz type, preprint (2026). arXiv:2607.21153. URL: https://arxiv.org/abs/2607.21153
11. A. K. Lerner, Quantitative weighted estimates for the Littlewood-Paley square function and Marcinkiewicz multipliers, Math. Res. Lett. 26 (2019), no. 2, 537-556. DOI 10.4310/mrl.2019.v26.n2.a7 (arXiv:1803.06981). URL: https://doi.org/10.4310/mrl.2019.v26.n2.a7
12. W. Chen, A. Culiuc, F. Di Plinio, M. Lacey, and Y. Ou, Endpoint sparse bounds for Walsh-Fourier multipliers of Marcinkiewicz type, preprint (2018). arXiv:1805.06060. URL: https://arxiv.org/abs/1805.06060
13. L. Grafakos, M. Mastyło, and L. Slavíková, A sharp variant of the Marcinkiewicz theorem with multipliers in Sobolev spaces of Lorentz type, J. Funct. Anal. 282 (2022), no. 3, 109295. DOI 10.1016/j.jfa.2021.109295 (arXiv:2008.11490). URL: https://doi.org/10.1016/j.jfa.2021.109295
14. F. Sukochev, F. Yang, D. Zanin, and D. Zhou, Optimal distribution estimates for commutators and Marcinkiewicz multipliers, Math. Ann. 393 (2025), no. 3-4, 3513-3560. DOI 10.1007/s00208-025-03306-3. URL: https://doi.org/10.1007/s00208-025-03306-3
15. C. Deng, E. Lorist, and M. Veraar, Operator-valued Fourier multipliers of bounded $s$-variation, preprint (2026). arXiv:2601.04803. URL: https://arxiv.org/abs/2601.04803