An extension of Alexandrov's theorem on second derivatives of convex functions (Q643434)

The author proves the following extension of Alexandrov's theorem on the a.e.-existence of second derivatives of convex functions. One considers a sequence \((f_k)\) of functions in \(C^2(U),\;U\subset \mathbb R^n\) open, which converges in \(L^1_{\text{loc}}(U)\) to a function \(f\). If all the minors of the Hessians of the functions \(f_k\) are uniformly locally bounded, then: (1) the family \(\{f_k\}\) of functions is locally uniformly bounded; (2) \(f_k\to f\) pointwise a.e. in \(U\); (3) both \(\limsup_kf_k\) and \(\liminf_kf_k\) are twice differentiable a.e. on \(U\). Here, locally bounded means bounded on every compact contained in \(U\). A sequence \((f_k)\) satisfying the above hypotheses is called a strong approximation to \(f\). Since any convex function can be locally uniformly approximated by smooth convex functions and every such approximation is a strong approximation, it follows that Alexandrov's theorem is a particular case of the above result. But, as the author points out, the class of strongly approximable functions is much larger, including, for instance, the functions in the Sobolev space \(W^{2,n}(U)\). The strongly approximable functions belong to the class of Monge-Ampère functions, considered earlier by the author [``Monge-Ampère functions'', Part I, Indiana Univ. Math. J. 38, No.~3, 745--771 (1989; Zbl 0668.49010); Part II, ibid., 773--789 (1989; Zbl 0697.49012)], an open question being whether or not the above result extends to this class of functions.