John's decomposition of the identity in the non-convex case (Q5957340)

Let \(K_2\subset\mathbb R^n\) be a convex body whose interior contains the origin and \(K_1\subset K_2\) be a connected compact set with \(\text{vol}(K_1)>0\). A pair of points \((x,y)\) is called a contact pair for \((K_1,K_2)\) if \(x\in\partial K_1\cap\partial K_2\), \(y\in\partial K_2\) and the inner product \(\langle x,y\rangle =1\). For such a contact pair, by \(y\otimes x\) we denote the linear transformation \(y\otimes x:z\mapsto \langle z,y\rangle x\) of \(\mathbb R^n\). The main results of the paper are Theorems 3 and 4 asserting that the equality \(\text{vol}(K_1)=\max\{\text{vol}(a+A(K_1)):a\in\mathbb R^n, a+A(K_1)\subset K_2\}\) where \(A\) runs over all nondegenerate \((n\times n)\)-matrices (resp. all symmetric positive definite matrices) implies (resp. is equivalent to) the existence of positive numbers \(\lambda_1,\dots,\lambda_N\) and contact pairs \((x_1,y_1),\dots,(x_N,y_N)\) for \((K_1,K_2)\) where \(N\leq n^2+n\) (resp. \(N\leq (n^2+3n)/2\)~) such that \(\sum_{i=1}^N\lambda_iy_i=0\) and \(\sum_{i=1}^N\lambda_i(y_i\otimes x_i)=I_n\) (resp. \(\sum_{i=1}^N\lambda_i (y_i\otimes x_i+x_i\otimes y_i)=I_n\)) where \(I_n\) is the identity transformation of \(\mathbb R^n\). These theorems are applied to the estimation of the volume ratio \(vr(K_2:K_1)={\text{vol}(K_2)^{1/n}}/{\text{vol}(K_1)^{1/n}} \) where \(K_1\subset K_2\) has the maximal volume among its affine copies in \(K_2\). It is shown that \(vr(K_2:K_1)\leq n^{\frac 1{p(K_1)}}\) and \(vr(K_2:K_1)\leq n!^{\frac 1n}n^{\frac 1{p(K_1)}-1}\) if \(K_1\) and \(K_2\) are symmetric. Here \(p(K_1)=\sup\{p\in(0,1]:\exists a\in\mathbb R^n\) with \(p\)-\(\text{conv}(\text{ext}(K_1)-a)\subset K-a\}\) where \(\text{ext}(K_1)\) is the set of extreme points of \(K_1\) and \(p\)-\(\text{conv}(K)\) is the smallest \(p\)-convex set \(C\) containing \(K\) (the \(p\)-convexity of \(C\) means that \(\lambda x+\mu y\in C\) for any \(x,y\in C\) and \(\lambda,\mu\geq 0\) with \(\lambda^p+\mu^p=1\)).