use absolute error for the termination criterion. This avoids a floating point exception

[[Imported from SVN: r5587]]

src/targetspacertrsolver.cc

@@ -40,6 +40,8 @@ setup(const AverageDistanceAssembler<TargetSpace>* assembler,
                                                                                                   energyNorm,
                                                                                                   Solver::QUIET));
 
+    innerSolver_->useRelativeError_ = false;
+
     // //////////////////////////////////////////////////////////
     //   Create obstacles
     // //////////////////////////////////////////////////////////