Special orthogonal splittings of \(L_1^{2k}\) (Q1885599)

The space \(L_p^n\) is the vector space \(\mathbb R^n\) with norm \(\| x\| _p := (n^{-1}\sum| x_i| ^p)^{1/p}\) (rather than the usual \(\ell_p\) norm). A {Kashin splitting} [\textit{B. Kashin}, Izv. Akad. Nauk SSSR Ser. Mat. 41, 334--351 (1977; Zbl 0354.46021)] of \(L_1^{2k}\) is a decomposition into two \(k\)-dimensional orthogonal (with respect to the usual inner product) subspaces \(E\), \(E^{\perp}\) on each of which the \(L_1^k\) and \(L_2^k\) norms are {universally} equivalent; i.e., there is a constant \(C\), independent of \(k\), such that for all \(x\) in \(E\), \(E^{\perp}\) one has \(\| x\| _1 \leq \| x\| _2 \leq C\| x\| _1\). Kashin gave a probabilistic argument to show that such splittings exist. One would like a more explicit construction. Here it is shown that, with high probability, a random \(k \times k\) matrix \(B\) with \(\pm 1\) entries together with \(\sqrt{k}I\) form a \(k \times 2k\) matrix \(A:=[\sqrt{k}I,B]\) whose rows span a space \(E\) which gives a Kashin splitting. The author notes that \(\sqrt{k}\) may be replaced by \(\lfloor \sqrt{k} \rfloor\) to obtain a formal strengthening of the recent result of \textit{G. W. Anderson} [Isr. J. Math. 138, 139--156 (2003; Zbl 1057.46012)].