Fix calculation of S matrix inverse (#156)

Author: robertodr

CHANGELOG.md

@@ -39,6 +39,9 @@
   are handled in the API functions.
 - Properly enforce `const`-correctness of the `Meddle` object and of its usage
   in the context API.
+- [Cholesky decomposition](http://eigen.tuxfamily.org/dox/classEigen_1_1LLT.html) is used
+  in CPCMSolver to get the inverse of the S matrix. The robust Cholesky (LDLT)
+  previously used is broken with the latest version of the Intel compilers.
 
 ## [Version 1.2.0-rc1] - 2018-03-02
 

src/solver/CPCMSolver.cpp

@@ -76,7 +76,7 @@ Eigen::VectorXd CPCMSolver::computeCharge_impl(const Eigen::VectorXd & potential
   int nrBlocks = blockS_.size();
   int irrDim = fullDim / nrBlocks;
   charge.segment(irrep * irrDim, irrDim) =
-      -blockS_[irrep].ldlt().solve(potential.segment(irrep * irrDim, irrDim));
+      -blockS_[irrep].llt().solve(potential.segment(irrep * irrDim, irrDim));
 
   return charge;
 }