No label defined (Q4547596)

The author presents an interesting approach for the construction of higher-order directional derivatives for real-valued functions. The notions are used for the formulation of necessary and sufficient higher-order optimality conditions for unconstrained nonsmooth optimization problems. In detail, let \(E\) be a Banach space and let \(f: E\to\mathbb{R}\) be a given (not necessarily continuous) function. The zero-order (lower) derivative of \(f\) at the point \(x^0\) in the direction \(u\) of the unit sphere \(S= \{u\in E\mid\|u\|= 1\}\) is defined by \[ f^{(0)}_- (x^0, u)= \liminf_{(t,u')\to (+0,u)} f(x^0+ tu'). \] Assuming the existence of this limit, the first-order (lower) derivative of \(f\) at \(x^0\) in this direction \(u\) is introduced by \[ f^{(1)}_-(x^0, u)= \liminf_{(t,u')\to (+0, u)}{1\over t}(f(x^0+ tu')- f^{(0)}_-(x^0, u)). \] In the same manner (assuming the existence of the former limits) the higher-order (lower) derivatives of \(f\) are given by \[ f^{(n)}_-(x^0, u)= \liminf_{(t,u')\to (+0,u)} {n!\over t^n} \Biggl(f(x^0+ tu')- \sum^{n-1}_{i=0} {t_i\over i!} f^{(i)}_- (x^0, u)\Biggr). \] In all the limits it is assumed that \(u'\in S\). Using different topologies on the unit sphere, relations to well-known directional derivatives can be pointed out. The author presents different characterizations for these directional derivatives and formulates optimality conditions in the following kind: If \(x^0\) is a local minimizer of \(f\) then (assuming the existence of the limits) for each \(u\in S\) the following conditions are satisfied: \hskip 10mm \(f^{(0)}_-(x^0, u)\geq f(x^0)\), \hskip 10mm \(f^{(0)}_-(x^0, u)= f(x^0)\Rightarrow f^{(1)}_-(x^0, u)\geq 0\), \hskip 10mm \(\vdots\) \hskip 10mm \(f^{(0)}_0(x^0, u)= f(x^0)\), \(f^{(i)}_-(x^0, u)= 0\) \((i= 1,\dots, n-1)\Rightarrow f^{(n)}_-(x^0, u)\geq 0\). If for each \(u\in S\), one of the following conditions \hskip 10mm \(f^{(0)}_-(x^0, u)> f(x^0)\), \hskip 10mm \(f^{(0)}_-(x^0, u)= f(x^0)\Rightarrow f^{(1)}_-(x^0, u)> 0\), \hskip 10mm \(\vdots\) \hskip 10mm \(f^{(0)}_-(x^0, u)= f(x^0)\), \(f^{(i)}_-(x^0, u)= 0\) \((i= 1,\dots, n-1)\Rightarrow f^{(n)}_-(x^0, u)> 0\), are satisfied (assuming the existence of the limits) then \(x^0\) is a strict local minimizer of \(f\).