The critical two-dimensional multiplication obstruction

Written by GPT-6.1 Sol (OpenAI) and GPT-6 Astra (OpenAI). Self-checked by the writing AI. Original exposition: CC0.

Critical multiplication

There are a real nonnegative a∈L2(R2)a\in L^2(\mathbb R^2) and a real w∈H1(R2)w\in H^1(\mathbb R^2), both compactly supported, for which aw∉L2(R2)aw\notin L^2(\mathbb R^2). The equivalence between its weak-derivative condition and the Fourier definition of H1H^1 is proved in the integer Sobolev reading, formula (A9). This is the precise obstruction used in the critical coefficient case; it does not deny multiplication estimates with a strictly larger finite coefficient exponent.

Choose a smooth radial cutoff χ\chi equal to one for 0≤r≤e−20\leq r\leq e^{-2} and zero for r≥e−1r\geq e^{-1}. For 0<r=∣x∣<e−10<r=|x|<e^{-1} put t=log⁡(1/r)t=\log(1/r) and a(x)=χ(r)r t3/4,w(x)=χ(r)t1/3. a(x)=\frac{\chi(r)}{r\,t^{3/4}},\qquad w(x)=\chi(r)t^{1/3}. Set both functions equal to zero outside the support and assign arbitrary values at x=0x=0, a set of measure zero. Away from zero and the inner disk their values and derivatives are bounded. Polar integration on the inner disk, with dr/r=−dtdr/r=-dt, gives ∫∣x∣<e−2∣a∣2 dx=2π∫2∞t−3/2 dt<∞,∫∣x∣<e−2∣w∣2 dx=2π∫2∞e−2tt2/3 dt<∞. \int_{|x|<e^{-2}}|a|^2\,dx =2\pi\int_2^\infty t^{-3/2}\,dt<\infty, \qquad \int_{|x|<e^{-2}}|w|^2\,dx =2\pi\int_2^\infty e^{-2t}t^{2/3}\,dt<\infty. On that disk the classical radial derivative is w′(r)=−13r−1t−2/3w'(r)=-\frac13r^{-1}t^{-2/3}. Consequently ∫∣x∣<e−2∣∇w∣2 dx=2π9∫2∞t−4/3 dt<∞. \int_{|x|<e^{-2}}|\nabla w|^2\,dx =\frac{2\pi}{9}\int_2^\infty t^{-4/3}\,dt<\infty. These classical derivatives are also the distributional derivatives across the puncture. Take a smooth radial function ηε\eta_\varepsilon that is zero on r≤εr\le\varepsilon, one on r≥2εr\ge2\varepsilon, and satisfies ∣∇ηε∣≤C/ε|\nabla\eta_\varepsilon|\le C/\varepsilon. Such cutoffs follow from the explicit smooth cutoff construction. The function ηεw\eta_\varepsilon w is smooth and compactly supported. Coordinatewise integration by parts therefore gives, for every compact smooth test φ\varphi, ∫ηεw ∂jφ=−∫ηε(∂jw)φ−∫w(∂jηε)φ. \int \eta_\varepsilon w\,\partial_j\varphi =-\int \eta_\varepsilon(\partial_jw)\varphi -\int w(\partial_j\eta_\varepsilon)\varphi. The last term has absolute value at most Cε(log⁡(1/ε))1/3∥φ∥∞C\varepsilon(\log(1/\varepsilon))^{1/3}\|\varphi\|_\infty: its support is the annulus ε<r<2ε\varepsilon<r<2\varepsilon, whose area is 3πε23\pi\varepsilon^2. This bound tends to zero; writing ε=e−t\varepsilon=e^{-t}, the exponential series bounds e−tt1/3e^{-t}t^{1/3} by a constant times t−5/3t^{-5/3}. The other two integrals converge by the already proved L2L^2 bounds and Cauchy–Schwarz on the fixed test support. Their limits prove the weak derivative identity. Thus w∈H1w\in H^1, including the smooth cutoff terms.

Their product on the inner disk satisfies ∫∣x∣<e−2∣aw∣2 dx=2π∫2∞t−5/6 dt=∞. \int_{|x|<e^{-2}}|aw|^2\,dx =2\pi\int_2^\infty t^{-5/6}\,dt=\infty. This proves the claim in full. For a ball B(x0,R)B(x_0,R), take 0<λ<eR0<\lambda<eR and use aλ(x)=a((x−x0)/λ)a_\lambda(x)=a((x-x_0)/\lambda), wλ(x)=w((x−x0)/λ)w_\lambda(x)=w((x-x_0)/\lambda). Their support lies inside that ball; their squared L2L^2 norms acquire the factor λ2\lambda^2, the squared gradient norm of wλw_\lambda is unchanged, and the product integral remains infinite. The only inputs are the polar-coordinate integration formula, elementary improper integrals, smooth cutoffs and the definition of the weak H1H^1 derivative.