From d59385db9d30e5c45e525c8e9e2cc9eac89522db Mon Sep 17 00:00:00 2001 From: Alexander Voigt Date: Tue, 14 Oct 2014 16:14:41 +0200 Subject: [PATCH] address referee 1 comment 4a: diagonalization precision --- doc/flexiblesusy-paper.tex | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/doc/flexiblesusy-paper.tex b/doc/flexiblesusy-paper.tex index e71b9714fd..71d5894b2b 100644 --- a/doc/flexiblesusy-paper.tex +++ b/doc/flexiblesusy-paper.tex @@ -1467,7 +1467,15 @@ \subsection{Tree-level spectrum} matrices with less than four rows and columns, and the LAPACK routines \code{zgesvd}, \code{dgesvd} for larger matrices. For the other types of diagonalization, \code{Eigen::SelfAdjointEigenSolver} from Eigen is -used regardless of the matrix size. +used regardless of the matrix size. Note, that \fs uses double +precision floating point data types with $15$ significant digits to +store the mass matrices and the mass eigenvalues. In case a particle +multiplet contains a very split mass hierarchy, where the mass +difference between the smallest and the largest mass in the multiplet +is of the order or greater than 10 orders of magnitude, double +precision data types are no longer sufficient. In this case we +recommend to either split the multiplet into sub-multiplets with +smaller mass hierarchies, or integrate out the heavy states. \fs uses the following conventions for the diagonalization: A mass matrix $M^2$ for real scalar fields $\phi_i$ is diagonalized with an