Bilinear Smoothing Conjecture

• Bilinear Smoothing Conjecture is a hypothesis stating that bilinear operators, especially those with dispersive or negative order characteristics, improve the regularity and integrability of function products beyond individual inputs.
• It leverages techniques such as Littlewood–Paley decompositions, dyadic frequency partitioning, and paraproduct expansions to extend classical linear smoothing results to nonlinear settings.
• The conjecture has wide-ranging implications across harmonic analysis, PDE theory, multiplier operators, and convex optimization, while highlighting open challenges in achieving a unified smoothing theorem.

The Bilinear Smoothing Conjecture posits that certain bilinear operators, notably those with negative order or dispersive structure, exhibit a smoothing effect that strictly improves the regularity or integrability of products of functions relative to their individual input regularities. This conjecture arises in harmonic analysis, PDE theory, and convex optimization contexts, and serves as a nonlinear analog of foundational smoothing results for linear operators.

1. Classical Formulations: Wave and Dispersive PDEs

In the context of dispersive PDEs, the Bilinear Smoothing Conjecture addresses phenomena such as the wave equation propagator $U(t)f(x) = e^{it\sqrt{-\Delta}}f(x)$ and the Airy evolution $e^{−t∂_x^3}u_0(x)$. Classical local smoothing theory for the linear flow asserts a fractional derivative gain on space-time averages, e.g., for Airy,

$$\|D^{1/4} e^{-t \partial_x^3} u_0\|_{L_x^\infty L_t^2} \lesssim \|u_0\|_{L^2}$$

[Kwon–Roy, (Kwon et al., 2010)]. Bilinear smoothing strengthens this principle: given two waves $u = U(t)f$, $v = U(t)g$ with angularly/frequency-separated $\widehat{f}, \widehat{g}$, the conjecture states

$$\|U(t)f \cdot U(t)g\|_{L_{t,x}^p} \lesssim \|f\|_{L^2} \|g\|_{L^2}$$

for $p$ above a critical threshold, e.g., $p > \frac{2n}{n-1}$ for $n$-dimensional wave propagation (Gao et al., 2019). In the Airy (KdV) setting, exact gains (of the form $(M/N)^\theta$ for frequency-separated initial data) have been established, confirming that genuine regularity improvement occurs in bilinear interactions (Kwon et al., 2010).

2. Smoothing for Bilinear Operators: Multiplier and Calderón–Zygmund Theory

The conjecture generalizes to bilinear multiplier and pseudodifferential operators, often modeled as

$$T_v(f, g)(x) = \iint_{\mathbb{R}^{2n}} m_v(\xi, \eta) \widehat{f}(\xi) \widehat{g}(\eta) e^{ix \cdot (\xi + \eta)} d\xi d\eta$$

where $m_v(\xi, \eta)$ is a symbol of order $-v$, typically satisfying either pointwise derivative decay or Sobolev-type regularity (Hart et al., 2017). The Bilinear Smoothing Conjecture asserts that the output Sobolev regularity satisfies

$$\|T_v(f,g)\|_{\dot{W}^{s,p}} \lesssim \|f\|_{\dot{W}^{s-v, p_1}} \|g\|_{L^{p_2}} + \|f\|_{L^{p_1}} \|g\|_{\dot{W}^{s-v, p_2}}$$

with $s$ either an even integer or exceeding certain dimensional thresholds. This result is realized for fractional integrals $I_v(f,g)$, (negative-order) bilinear multipliers, and bilinear pseudodifferential operators under mild regularity (Hart et al., 2017).

3. Bilinear Smoothing for Maximal and Averaging Operators

Recent advances extend the conjecture to bilinear maximal and averaging operators over hypersurfaces, notably with fractal dilation sets. For a hypersurface $S \subset \mathbb{R}^{2d}$ and dilation set $E \subset [1,2]$, consider the bilinear maximal averaging operator

$$\mathcal{A}_{m,E}(f,g)(x) = \sup_{t \in E}|T_{m_t}(f,g)(x)|$$

where $m$ is a multiplier with decay $$|\partial^\alpha m(\xi, \eta)| \lesssim (1+|\xi|+|\eta|)^{-a}$$ (Borges et al., 2023). The main smoothing theorem asserts, for $s_1 + s_2 < (2a-d-\beta)/2$ where $\beta$ is the Minkowski dimension of $E$, that

$$\|\mathcal{A}_{m,E}(f,g)\|_{L^2} \lesssim \|f\|_{H^{-s_1}} \|g\|_{H^{-s_2}}$$

This establishes a precise link between geometric decay, fractal dimension, and smoothing index. Additional implications include sharp $L^p$ improving and weighted sparse bounds for multi-scale maximal functions.

4. Local Smoothing for Bilinear Fourier Integral Operators

The conjecture has been rigorously formulated for bilinear FIOs. Let $T_a^{\varphi_1, \varphi_2}(f,g)$ denote a bilinear FIO of total order $m < 0$ with cinematic curvature. The Bilinear Smoothing Conjecture asserts, for $p_i \geq \overline{p}_d$ (dimension-dependent), and symbol order parameters $m_i, \sigma_i < 1/p_i$:

$$\|T_a^{\varphi_1, \varphi_2}(f,g)\|_{L^p(\mathbb{R}^d_x \times I_t)} \lesssim \|f\|_{L^{p_1}_{m_1+(d-1)(1/2-1/p_1)-\sigma_1}} \|g\|_{L^{p_2}_{m_2+(d-1)(1/2-1/p_2)-\sigma_2}}$$

Exact sharpness holds in $d=2$ and all odd $d$, with partial results in higher dimensions (Cardona, 22 Jan 2026). The proof structure leverages paraproduct decompositions, frequency separation, square-function and maximal-function lemmas, and reductions to linear smoothing estimates.

5. Smoothing Criteria and Geometric Conditions

Recent work provides sharp multiplier criteria for smoothing inequalities of fiber-wise bilinear operators, including the triangular Hilbert transform along non-flat analytic curves. The smoothing estimate,

$$\|T_m(f,g)\|_{L^1} \leq C \lambda^{-c_{2,2}} \|f\|_{L^2} \|g\|_{L^2}$$

requires multiplier difference-operator conditions: $L^2 \otimes L^2$ boundedness and “half-derivative” decay in translation parameters. The underlying geometric structures (curvature, o-minimality, fractal dimensions) inherit directly into the smoothing index and maximal bounds (Hsu et al., 2024).

6. Applications in Game Theory: Bilinear Saddle-Point Smoothing

In convex optimization, the conjecture governs the acceleration of bilinear smoothing techniques for extensive-form zero-sum games (sequence-form saddle-point problems). The excessive gap technique (EGT) yields $O(1/k)$ convergence, with error bounds depending on the condition number $\frac{D_x D_y}{\sigma_x \sigma_y}$, where $D_x$ is the diameter and $\sigma_x$ the strong convexity parameter of the prox function. Optimized prox constructions can reduce the complexity to $O(M_Q \sqrt{\ln \Sigma}/k)$ (Habara et al., 2023). Empirical centering heuristics yield further acceleration, but rigorous improvements over $O(1/k)$ rates are open.

7. Methodological Frameworks and Open Problems

Techniques across settings employ Littlewood–Paley decompositions, dyadic frequency partitioning, paraproduct expansions, Fourier multiplier theory, square-function and Kakeya-type estimates, and sparse domination via continuity estimates. Endpoint cases, variable coefficients with minimal regularity, higher-order multilinear smoothing gains, and sharp geometric factorization remain the central open directions. Full conjecture verification across Calderón–Zygmund kernels and abstract FIOs in high dimensions requires new analytic inputs, possibly connecting molecular decompositions and mixed Hardy-space embeddings (Hart et al., 2017, Gao et al., 2019, Borges et al., 2023).

---

The Bilinear Smoothing Conjecture thus specifies the optimal regularity gains possible from bilinear, dispersive, or averaging operators as a function of their analytic and geometric parameters. It incorporates and extends some of the deepest principles in harmonic analysis and nonlinear PDE theory, with broad implications for maximal inequalities, product Sobolev embeddings, and convex optimization in saddle-point problems. While major special cases are resolved, a truly unified and general smoothing theorem for bilinear operators remains an open frontier in mathematical analysis.