gpt 6 astra ultra attempt on Math-361

Iwaniec Conjecture for the Beurling-Ahlfors Transform Norm

Definitions and mathematical statement

For suitable complex-valued functions on ${\mathbb{C}}\cong{\mathbb{R}}^2$, the Beurling--Ahlfors transform is the principal-value singular integral \[ Bf(z)=-\frac1\pi\operatorname{p.v.}\int_{{\mathbb{C}}} \frac{f(w)}{(z-w)^2}{\,\mathrm{d}} A(w). \] Its bounded extension to $L^p({\mathbb{C}})$, $1<p<\infty$, has norm $\|B\|_p=\sup_{f\ne0}\|Bf\|_p/\|f\|_p$. Put $p^*=\max\{p,p/(p-1)\}$. The conjecture is \[ \|B\|_p=p^*-1\quad\text{for every }1<p<\infty. \] The kernel normalization fixes the numerical constant. The functions are complex-valued and the norm is the full planar operator norm, not a restriction to radial or real-valued test functions. A nonsharp bound proportional to $p^*-1$ does not determine the requested value.

Formulation and concept references: [S1].

Short English statement

Is the exact norm of the Beurling--Ahlfors transform on every complex-valued $L^p$ space equal to $p^*-1$?

Research attempt

Scope and source audit.

The normalization in the question is essential. We work with complex functions, planar Lebesgue measure and the stated kernel. Section 6 of the inspected 2026 version of [S1] still presents the sharp upper bound as a problem. Its martingale results are not identified with a proof of the conjecture. The elementary argument below proves the entire proposed lower bound, proves equality at $p=2$, and reduces the missing upper bound to a precise inequality for complex derivatives. All these statements concern the full operator; restricting a test family supplies a lower bound, not an upper bound on that operator.

Set \[ \partial=\tfrac12(\partial_x-i\partial_y),\qquad \bar\partial=\tfrac12(\partial_x+i\partial_y). \] Use the Fourier convention $\widehat f(\xi)=\int f(x)e^{-ix\cdot\xi}\,dx$. The distributional fundamental solution $\bar\partial(1/(\pi z))=\delta_0$ and its $\partial$ derivative give the multiplier \[ m(\xi)=\frac{\xi_1-i\xi_2}{\xi_1+i\xi_2},\qquad \xi\ne0. \tag{357.1} \] For completeness, the fundamental-solution identity follows by integrating against a smooth test function outside a circle of radius $\varepsilon$ and applying the complex Green formula. The outer boundary term vanishes for a compactly supported test function and the inner boundary integral tends to its value at zero. Differentiating $1/(\pi z)$ away from zero gives $-1/(\pi z^2)$, with the circular principal value. Thus $357.1$ uses exactly the kernel in the question.

The Hilbert-space case and the differential identity.

Since $|m(\xi)|=1$ almost everywhere, Plancherel gives \[ \|Bf\|_2=\|f\|_2. \tag{357.2} \] In particular the sharp constant is one at $p=2$. This argument does not extend to general $p$: bounded modulus of a Fourier multiplier does not identify its $L^p$ norm unless $p=2$.

The same multiplier computation gives, for every $u\in C_c^\infty(\mathbb C;\mathbb C)$, \[ B(\bar\partial u)=\partial u. \tag{357.3} \] There is no uncontrolled additive holomorphic function in this identity: it is an equality of Fourier transforms, including equality as distributions at zero frequency. Both derivatives are ordinary compactly supported functions. Consequently any such $u$ gives the rigorous test \[ \|B\|_p\geq\frac{\|\partial u\|_p}{\|\bar\partial u\|_p} \quad\hbox{when }\bar\partial u\ne0. \tag{357.4} \]

A cutoff power construction attaining the lower threshold.

Fix $p\in(1,\infty)$ and put $\alpha=2/p\in(0,2)$. First consider $v(z)=z|z|^{-\alpha}$ away from zero. Direct differentiation, using $\partial |z|^2=\bar z$, gives \[ \partial v=(1-\alpha/2)|z|^{-\alpha},\qquad \bar\partial v=-\frac\alpha2\frac{z^2}{|z|^2}|z|^{-\alpha}. \tag{357.5} \] Neither derivative belongs to $L^p$ on the entire punctured plane: the radial integral at the chosen exponent is logarithmic. We therefore must cut off both ends rather than use $v$ as an inadmissible extremizer.

Choose smooth real radial cutoff functions $\chi_0,\chi_1$ such that $\chi_0(r)=0$ for $r\leq1$, $\chi_0(r)=1$ for $r\geq2$, $\chi_1(r)=1$ for $r\leq1$, and $\chi_1(r)=0$ for $r\geq2$. For $0<\varepsilon<1/4$, define \[ u_\varepsilon(z)=\chi_0(|z|/\varepsilon)\chi_1(|z|) z|z|^{-\alpha}. \] This is a smooth compactly supported complex function, identically zero near the apparent singularity. On $2\varepsilon<|z|<1$ its derivatives are exactly $357.5$. Therefore the contributions from this annulus are \[ \begin{split} \int|\partial u_\varepsilon|^p &=2\pi(1-\alpha/2)^p\log(1/(2\varepsilon))+O_p(1),\\ \int|\bar\partial u_\varepsilon|^p &=2\pi(\alpha/2)^p\log(1/(2\varepsilon))+O_p(1). \end{split} \tag{357.6} \] Here the bounded errors require verification. On the outer transition $1\leq r\leq2$, every derivative is uniformly bounded. On the inner transition $\varepsilon\leq r\leq2\varepsilon$, the product rule bounds each derivative by $C_p\varepsilon^{-\alpha}$, while its area is $O(\varepsilon^2)$. Its $p$-th power integral is bounded because $\alpha p=2$. These are all remaining regions. Thus the errors in $357.6$ really are bounded independently of $\varepsilon$.

It follows from $357.4$--$357.6$ that \[ \|B\|_p\geq\frac{1-\alpha/2}{\alpha/2}=p-1. \tag{357.7} \] Apply the same construction with $\bar z|z|^{-\alpha}$, or simply conjugate $u_\varepsilon$. Complex conjugation exchanges the moduli of $\partial u$ and $\bar\partial u$. It therefore gives the second lower bound \[ \|B\|_p\geq\frac1{p-1}. \tag{357.8} \] Taking the larger of $357.7$ and $357.8$ proves \[ \boxed{\ \|B\|_p\geq p^*-1\quad(1<p<\infty).\ } \tag{357.9} \] This proof also explains the change of the relevant test at $p=2$: one orientation of the complex power is useful above two, and its conjugate is useful below two. A calculation with only one orientation would miss half of the proposed lower bound.

Duality and why it does not provide the upper bound.

Write $Cf=\bar f$. Since $m(-\xi)=m(\xi)$, Fourier transformation shows $CBC=B^*$ on $L^2$, where the latter has multiplier $\bar m$. These identities extend to the bounded $L^p$ realizations by density. Conjugation is an isometry. If $q=p/(p-1)$, Banach-space duality gives \[ \|B\|_p=\|B^*\|_q=\|B\|_q. \tag{357.10} \] Thus proving the conjectured upper bound for $p\geq2$ would settle the remaining range. This equality is a reduction, not an estimate: it only relates two currently unknown norms. The lower-bound tests already respect it.

There is also a useful identity at the level of Jacobians. If $u=u_1+iu_2$, direct expansion gives \[ |\partial u|^2-|\bar\partial u|^2 =\partial_xu_1\partial_yu_2-\partial_yu_1\partial_xu_2. \tag{357.11} \] For a compactly supported smooth $u$ the integral of the right-hand side is zero by integration by parts, recovering equality of the two $L^2$ derivative norms. A tempting argument multiplies $357.11$ by a power of $|Du|$ and tries the same integration. It fails because the new weight has nonzero derivatives. There is no longer a null Lagrangian with zero integral. The logarithmic tests make clear that any replacement must permit derivative-norm ratios as large as $p^*-1$.

An exact analytic form of the missing estimate.

For fixed $p$, the conjectured upper bound would imply \[ \int_{\mathbb C}|\partial u|^p\,dA \leq(p^*-1)^p\int_{\mathbb C}|\bar\partial u|^p\,dA \quad(u\in C_c^\infty(\mathbb C;\mathbb C)). \tag{357.12} \] Conversely, this family of derivative inequalities is sufficient for the full operator upper bound. Here is the density point needed in that assertion. The subspace $\{\bar\partial u:u\in C_c^\infty\}$ is dense in $L^p$. Indeed, if a continuous dual functional annihilates it, its representing $L^q$ function solves one of the complex Cauchy--Riemann equations in distributions. Mollification gives an entire holomorphic or antiholomorphic function with bounded $L^q$ norm. The mean-value inequality on a disc of radius $R$ bounds its value at any fixed point by $C R^{-2/q}$ times that norm. Letting $R$ tend to infinity makes the mollified function zero; then mollifiers tending to the identity make the original dual function zero. Hahn--Banach proves density. The possible conjugation in the dual pairing only exchanges the two Cauchy--Riemann equations and does not change this argument.

Now apply $357.12$ and $357.3$ on this dense subspace and use the already known bounded extension of $B$ on $L^p$. Passage to the $L^p$ limit yields the desired bound for every input. Thus $357.12$ is a precise first missing statement; it is not enough to prove it for radial $u$ or for a single angular mode.

Why mode computations and discretization do not finish it.

Rotations decompose $L^2$ orthogonally into angular Fourier modes, and $357.1$ shifts the angular behavior in a particularly simple way. For $p\ne2$, however, \[ \|f_1+f_2\|_p^p\ne\|f_1\|_p^p+\|f_2\|_p^p \] in general, even for different angular frequencies. Bounds proved on each mode therefore cannot be summed with constant one. Cancellations in the input and reinforcement in the output are precisely what a full-space upper estimate must control. An operator matrix truncated in angular and radial frequency likewise has a restricted test space; its computed norm is not an upper bound on the infinite-dimensional operator without an independent uniform remainder estimate.

The associated diagnostic checks the derived power-coefficient ratios with exact rational arithmetic for selected $p$, and checks $|m(\xi)|=1$ numerically on nonzero integer frequencies. It does not approximate a proof of $357.12$. The annular cutoff estimate is analytic and uniform, so the lower bound is not inferred from numerical fitting.

Outcome.

The lower threshold is rigorously necessary at every $p$, and it is the exact norm at $p=2$. Duality reduces the open upper estimate to one side of two, and density identifies its full differential form. The first gap is establishing $357.12$ for arbitrary complex compactly supported maps. Neither the $L^2$ cancellation, the power tests nor an angular-mode restriction supplies that estimate. The full claim remains unresolved by this attempt.

Sources

• S1 Rodrigo Banuelos, Cotlar martingale transforms and related singular integrals, arXiv:2604.09365v2 $2026$. https://arxiv.org/html/2604.09365v2#S6. Formulation source.