TNNMG: Use UMFPack for the linear coarse grid problems again

Moving to CHOLMOD was a mistake:  After all, the brittle-fracture
increment functional is non-convex, and hence second derivatives
can be indefinite.

src/brittle-fracture.cc

@@ -39,7 +39,7 @@
 #include <dune/fracture-phasefields/elementquantityassembler.hh>
 
 #include <dune/solvers/solvers/loopsolver.hh>
-#include <dune/solvers/solvers/cholmodsolver.hh>
+#include <dune/solvers/solvers/umfpacksolver.hh>
 #include <dune/solvers/norms/energynorm.hh>
 #include <dune/solvers/iterationsteps/multigridstep.hh>
 #include <dune/solvers/iterationsteps/truncatedblockgsstep.hh>
@@ -464,7 +464,9 @@ int main (int argc, char *argv[]) try
                                           baseEnergyNorm,
                                           Solver::QUIET);
 #else
-  Solvers::CholmodSolver<Matrix,Vector> linearBaseSolver;
+  // The matrix of the linear correction problem is symmetric,
+  // but possibly indefinite.
+  Solvers::UMFPackSolver<Matrix,Vector> linearBaseSolver;
 #endif
 
   std::shared_ptr<LinearIterationStep<Matrix,Vector> > linearIterationStep;

src/ductile-fracture.cc

@@ -577,6 +577,8 @@ int main (int argc, char *argv[]) try
                                           baseEnergyNorm,
                                           Solver::QUIET);
 #else
+  // The matrix of the linear correction problem is symmetric,
+  // but possibly indefinite.
   Solvers::UMFPackSolver<Matrix,Vector> linearBaseSolver;
 #endif