The multilinear restriction conjecture

For n1n\geq 1, let UiRnU_i\subset\mathbb{R}^n be open bounded neighborhoods and let Σi:UiRn+1\Sigma_i:U_i\to\mathbb{R}^{n+1} be smooth parametrizations of nn-dimensional submanifolds, with associated extension operators

Eifi(x)=UieixΣi(ξ)fi(ξ)dξ.\mathcal E_i f_i(x)=\int_{U_i}e^{ix\cdot\Sigma_i(\xi)}f_i(\xi)\,d\xi.

Assume that, for every multi-index α\alpha, the derivatives satisfy αΣiL(Ui)α1\|\partial^\alpha\Sigma_i\|_{L^\infty(U_i)}\lesssim_\alpha 1, and that there is ν>0\nu>0 such that the unit normals Ni(ζi)N_i(\zeta_i) obey

det(N1(ζ1),,Nn+1(ζn+1))ν\left|\det\bigl(N_1(\zeta_1),\ldots,N_{n+1}(\zeta_{n+1})\bigr)\right|\geq\nu

for all ζiΣi(Ui)\zeta_i\in\Sigma_i(U_i).

Multilinear restriction conjecture. Under these assumptions, there is a constant CC, depending on finitely many derivatives of the Σi\Sigma_i, the sets UiU_i, ν\nu, and nn, such that

i=1n+1EifiL2/n(Rn+1)Ci=1n+1fiL2(Ui).\left\|\prod_{i=1}^{n+1}\mathcal E_i f_i\right\|_{L^{2/n}(\mathbb{R}^{n+1})}\leq C\prod_{i=1}^{n+1}\|f_i\|_{L^2(U_i)}.

This is the endpoint multilinear restriction estimate, an almost optimal form of the restriction problem. The corresponding estimate with an arbitrarily small loss is known, while obtaining the displayed estimate with no loss remains an open problem.

Equivalent formulations 4

Other statements of this same problem, merged from separate entries. Each is equivalent to the statement above — proving any one settles them all.

  1. Multilinear restriction conjecture

    Let UjU_j be compact neighbourhoods of the origin in Rd1\mathbb{R}^{d-1}, let Σj:UjRd\Sigma_j:U_j\to\mathbb{R}^d be C1,βC^{1,\beta} parametrisations of codimension-one submanifolds SjS_j, and let Ej\mathcal{E}_j be the associated Fourier extension operators

    Ejg(ξ)=Ujg(x)eiξ,Σj(x)dx.\mathcal{E}_jg(\xi)=\int_{U_j}g(x)e^{i\langle \xi,\Sigma_j(x)\rangle}\,\mathrm{d}x.

    Suppose that the submanifolds S1,,SdS_1,\ldots,S_d are transversal in a neighbourhood of the origin, q2dd1q\geq\frac{2d}{d-1}, and pd1dqp'\leq\frac{d-1}{d}q. Multilinear restriction conjecture. There exists a constant CC such that

    j=1dEjgjLq/d(Rd)Cj=1dgjLp(Uj)\left\|\prod_{j=1}^{d}\mathcal{E}_{j}g_{j}\right\|_{L^{q/d}(\mathbb{R}^d)}\leq C\prod_{j=1}^{d}\|g_{j}\|_{L^p(U_j)}

    for all g1,,gdg_1,\ldots,g_d supported in a sufficiently small neighbourhood of the origin. This conjecture is part of the multilinear restriction theory for the Fourier transform; the cited work of Bennett, Carbery and Wright and Bennett, Carbery and Tao concerns this family of estimates. The supplied material does not establish whether the stated range is resolved, so its status remains open.

    source: Jonathan Bennett and Neal Bez, “Some nonlinear Brascamp-Lieb inequalities and applications to harmonic analysis”, arXiv:0906.2064 (2010).

  2. The multilinear restriction conjecture

    Let 2kd2\leq k\leq d. A kk-tuple S1,,SkS_1,\ldots,S_k of smooth codimension-one submanifolds of Rd\mathbb{R}^d is transversal if the wedge product of any choice of unit normal vectors has a positive uniform lower bound. Let dσjd\sigma_j be the induced measures and let pp' be the Hölder conjugate of pp. Multilinear restriction conjecture. If S1,,SkS_1,\ldots,S_k are transversal with everywhere positive principal curvatures, 1q<d12d\frac{1}{q}<\frac{d-1}{2d}, 1qd+k2d+k1p\frac{1}{q}\leq\frac{d+k-2}{d+k}\frac{1}{p'}, and 1qdkd+k1p+k1k+d\frac{1}{q}\leq\frac{d-k}{d+k}\frac{1}{p'}+\frac{k-1}{k+d}, then

    j=1kfjdσj^Lq/k(Rd)j=1kfjLp(dσj).\left\|\prod_{j=1}^{k}\widehat{f_jd\sigma_j}\right\|_{L^{q/k}(\mathbb{R}^d)}\lesssim\prod_{j=1}^{k}\|f_j\|_{L^p(d\sigma_j)}.

    This generalizes the bilinear restriction perspective and is intended to clarify the roles of curvature and transversality.

    source: Jonathan Bennett, “Aspects of Multilinear Harmonic Analysis Related to Transversality”, arXiv:1405.5369 (2014).

  3. The multilinear restriction conjecture

    Let {Uj:j[d]}\{U_j:j\in[d]\} be a collection of C2C^2 hypersurfaces in Rd\mathbb{R}^d, with the normal to UjU_j at every point within 1100\frac{1}{100} of the xjx_j-axis. Let Ej\mathcal E_j be the corresponding extension operators. For ε>0\varepsilon>0, q2dd1q\geq\frac{2d}{d-1}, and pq(d1)dp'\leq\frac{q(d-1)}{d}, the multilinear restriction conjecture asserts that

    j=1dEjgjLq/d(B(0,R))j=1dgjLp(Uj)\left\|\prod_{j=1}^d\mathcal E_jg_j\right\|_{L^{q/d}(B(0,R))}\lesssim\prod_{j=1}^d\|g_j\|_{L^p(U_j)}

    for all gjLp(Uj)g_j\in L^p(U_j) and all R1R\geq1. It is known away from the endpoint and at the endpoint with RεR^\varepsilon losses; the supplied counterexample shows only that the Mizohata–Takeuchi route cannot prove the lossless endpoint estimate, not that this conjecture itself is refuted.

    source: Hannah Cairo, “A Counterexample to the Mizohata-Takeuchi Conjecture”, arXiv:2502.06137 (2025).

  4. The multilinear restriction conjecture

    For each 1jd1\leq j\leq d, let UjU_j be a parameter domain, let Σj:UjRd\Sigma_j:U_j\to\mathbb{R}^d be smooth, and let Ej\mathcal{E}_j be the associated extension operator. Define

    Yj(x)=k=1d1xkΣj(x).Y_j(x)=\bigwedge_{k=1}^{d-1}\frac{\partial}{\partial x_k}\Sigma_j(x).

    Assume the transversality condition det(Y1(x(1)),,Yd(x(d)))ν\det(Y_1(x^{(1)}),\ldots,Y_d(x^{(d)}))\geq\nu for all x(j)Ujx^{(j)}\in U_j, and the smoothness condition ΣjC2(Uj)A\|\Sigma_j\|_{C^2(U_j)}\leq A for all jj. Multilinear restriction conjecture. If q2dd1q\geq\frac{2d}{d-1} and pd1dqp'\leq\frac{d-1}{d}q, then there exists a constant CC, depending only on AA, ν\nu, dd, and U1,,UdU_1,\ldots,U_d, such that

    j=1dEjgjLq/d(Rd)Cj=1dgjLp(Uj)\left\|\prod_{j=1}^d\mathcal{E}_jg_j\right\|_{L^{q/d}(\mathbb{R}^d)}\leq C\prod_{j=1}^d\|g_j\|_{L^p(U_j)}

    for all gjLp(Uj)g_j\in L^p(U_j). This is the multilinear analogue of the restriction conjecture and is designed to retain only the transversality needed among the normals; its full asserted range is open in general.

    source: Jonathan Bennett, Anthony Carbery and Terence Tao, “On the Multilinear Restriction and Kakeya conjectures”, arXiv:math/0509262 (2005).

Sources & referencesView supporting material

Primary source

Ioan Bejenaru, “The multilinear restriction estimate: a short proof and a refinement”, arXiv:1601.03336 (2016).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.