| The
aim of the Multilevel Minimization algorithm is to find the global
minimizer of the potential function used to describe a protein. For this
purpose several coarser representations, the levels, of the protein are
constructed. One minimization step of the algorithm consists of local
minimizations on every level as described below. Each step is then
accepted or rejected according to a metropolis probability. The
resulting global minimization algorithm is very efficient in finding low
lying minimizers of the potential function. The multilevel structure
speeds up the convergence of the method by reducing the number of needed
steps for the local minimizations as well as for the global Monte-Carlo
steps. |
 |
 |
This is the initital
configuration of a small sample protein (trp-cage). This all atom
representation is reduced to a coarser representation by merging some
atoms to pseudo atoms.
 |
 |
The resulting structure of
the next coarser level is transferred to the all atom representation and
again minimized. The result represents a single step of the global
minimization algorithm.
 |
 |
This coarsening procedure
is repeated once.
|
 |
The minimized structure of
the coarsest level is expanded to the next finer level. Here, local
minimization leads to a refinement of the configuration.
 |
 |
On the coarsest level the
protein is represented by only a few pseudo atoms.
 |
 |
Local Minimization on this
level already leads to a packed configuration.
|