Random polarizations (Q1935194)

Polarization is a method that may reduce geometric inequalities to combinatorial identities. It consists of constructing random sequences of rearrangements of a function \(f\), \(S_{W_1\dots W_n}f\), where each random variable \(W_i\) determines a reflection, that converge to the symmetric decreasing rearrangement \(f^*\), given by the lower semicontinuous function equimeasurable with \(f\) which is radially decreasing about \(o\). The authors consider the Hausdorff distance between two sets on a space-form \(\mathbb{X}\) of dimension \(d\) and with a fixed origin \(o\). The polarization of \(f\) with respect to a reflexion \(\sigma\) of \(\mathbb{X}\) that does not fix the origin is defined by the two-point symmetrization \(Sf(x)\) as the maximum (resp., minimum) of \(\{f(x), f(\sigma(x))\}\) for \(x\in H^+\) (resp., \(x\in H^-\)), where \(H^{\pm}\) is the upper (resp., lower) half-space defined by the origin. The definition is extended to sets \(A\subset \mathbb{X}\) by taking the indicator functions. A random polarization \(S_W\) is given by a Borel probability measure \(\mu\) on the parameter space \([0, l)\times \mathbb{S}^{d-1}\) defined by polar coordinates of \(\mathbb{X}\), and that also determines the distribution of the random variable \(W=(R,U)\). The authors assume \(\mu(R=0)=0\), while, for a random Steiner-symmetrization, \(\mu(R=0)=1\) is assumed. As a Steiner symmetrization, the polarization is monotone, equimeasurable, \(L_p\)-contractive for \(p\geq 1\), and preserves or improves the modulus of continuity of each function \(f\). In the first main theorem, the authors prove for a given sequence of polarizations \(\{S_{W_1\dots W_n}\}_{n\geq 1}\) defined by a sequence of independent random variables \(\{W_i\}_{i\geq 1}\) that, if \(\sum_{i=1}^{\infty}\operatorname{P}(d( \sigma_{W_i}a_i,b_i)<\rho)=\infty\) for any \(\rho>0\) and bounded sequences \(\{a_i\}, \{b_i\}\) of elements of \(\mathbb{X}\) satisfying \(d(b_i,o),\geq d(a_i,o)+2\rho\), then the sequence of polarizations converges almost surely, that is, \[ \operatorname{P}\left(\lim_{n\to \infty} \|S_{W_1\dots W_n}f-f^*\|_{\infty}=0\; \forall f \in C_c^+(\mathbb{X})\right)=1, \] where \(\operatorname{P}\) is the product measure \(\operatorname{P}(W_1\in A_1,\dotsc, W_n\in A_n)=\Pi_{i=1}^{n}\mu_i(A_i)\). From the author's abstract: ``The proof of convergence hinges on an estimate for the expected distance from the limit that yields a bound on the rate of convergence.'' In the second main theorem, they consider the special case of i.i.d. sequences, that is, the independent random variables \(W_i\) are identically distributed according to a probability measure \(\mu\) with \(\mu(R=0)=0\), and consider \(G :=\{u \in \mathbb{S}^{d-1}: (0,u)\in \mathrm{support} \mu\}\). The authors prove that, if each orbit \(\mathcal{O}_{G,x}=\{\tau_{u_n}\ldots \tau_{u_1} x: u_k\in G \}\) is dense in the \((d-1)\)-dimensional sphere for each \(x\in \mathbb{S}^{d-1}\), where \(\tau_{u_k}\) is the map that fixes \(H^+_u\) and values \(\sigma_u\) on \(H_u^-\), then \(S_{W_1\dots W_n}\) converges to the symmetric decreasing rearrangement and \[ \operatorname{P}\left(\lim_{n\to \infty} \|S_{W_1\dots W_n}f-f^*\|_{p}=0\; \forall f \in L_p^+(\mathbb{X})\right)=1 \] holds for all \( 1\leq p<\infty\). With these conditions, the distribution does not need to be uniform, a necessary condition used in some recent results to guarantee convergence. These two results imply conclusions for Steiner symmetrizations when \(\mathbb{X}=\mathbb{R}^d\). From the author's abstract: ``We find bounds on the rate of convergence of Steiner symmetrizations that require no convexity assumptions, and show that full rotational symmetry can be achieved by randomly alternating Steiner symmetrizations in a finite number of directions that satisfy an explicit non-degeneracy condition.'' Furthermore, the authors obtain some bounds on the rate of convergence and construct examples where convergence fails.