Mathematical programs with complementarity constraints and a non-Lipschitz objective: optimality and approximation (Q2220667)

The authors consider the complementarity problem (1.3) \(F(x):=f(x)+\Vert Dx\Vert^p_p\) subject to \(G(x)\ge 0\), \(H(x)\ge 0\), \(G(x)^T,H(x)= 0\) where \(f:\mathbb{R}^n\rightarrow \mathbb{R},G,H:\mathbb{R}^n\rightarrow \mathbb{R}^m\) are continuously differentiable and \(D\in \mathbb{R}^{n\times m}\) with a non Lipschitzian part of the objective because of \(p\in (0,1)\). They consider modified problems \(\left( \mathcal{P}_{\epsilon,\sigma}\right) \) with Lipschitzian relaxations \[F_\epsilon(x)=f(x)+\sum_{i=1}^n\left( \vert D_i^Tx\vert+\epsilon \right)^p\] for the objective and the Scholtes relaxation \(G(x)^T\,H(x)\le \sigma\) [\textit{S. Scholtes}, SIAM J. Optim. 11, No. 4, 918--936 (2001; Zbl 1010.90086)] for the complementarity constraint where the parameter sequences of \(\epsilon,\sigma >0\) tends to zero. Each cluster point of associated solution sequences is a solution of problem (1.3). Using the subdifferential calculus of \textit{R. T. Rockafellar} and \textit{R. J. B. Wets} [Variational analysis. Berlin: Springer (1998; Zbl 0888.49001)] and modified regularity conditions for MPCC, with \(D\) as additional summand, they derive necessary optimality conditions (multiplier rules) of first order for problem (1.3) and under some modified LICQ and continuous twice differentiability of \(f,G,H\) the KKT conditions as well as some weak necessary second order conditions for the modified problems \(\left( \mathcal{P}_{\epsilon,\sigma}\right) \). They propose a simple Algorithm 4.1. Under the above modified LICQ accumulation points of approximate solutions are C-stationary points of problem (1.3) (multipliers of inactive \(G,H\) are zero) [\textit{H. Scheel} and \textit{S. Scholtes}, Math. Oper. Res. 25, No. 1, 1--22 (2000; Zbl 1073.90557)]. If additionally for all sufficiently small \(\epsilon,\sigma>0\) the second order condition is satisfied then the accumulation points of approximate solutions are M-stationary points of problem (1.3) (addditionally nonnegative multipliers of common active \(G,H\)), beyond that S-stationarity is given under upper level strict complementarity (ULSC). It seems to be the first successful attempt to develop a theory and suitable approximations for solving non Lipschitzian complementarity problems. Both is successfully applied to a sparse LCP and to a second best road pricing problem. The paper can be well studied. All results are exactly cited or proven in detail.