Cool Stars 14 - Submitted Abstract # 104
This version created on 05 October 2006

A new Zeeman Doppler Imaging code utilizing a fast Stokes-Profile
synthesis


Markus Kopf
Thorsten A. Carroll
Klaus G. Strassmeier

We present a new Zeeman Doppler Imaging (ZDI) code "iMap" which is
designed to reconstruct the temperature and magnetic field
distribution of solar-type stars.  The synthesis (forward) module of
our code is based either on a conventional numerical solution of the
polarized radiative transfer equation (RTE) or on a novel approach
with Artificial Neural Networks (ANN's) (see below).  The inversion
module implements a least square method which uses a regularized
conjugate gradient method to reconstruct the surface temperature and
magnetic field.  In addition to our new ZDI code we present a novel
method for a fast LTE spectrum synthesis in magnetic stellar
atmospheres suitable for Doppler Imaging and Zeeman Doppler Imaging.
To demonstrate the potential of this method, we present a number of
test inversions of artificial stars.  Our method is based on a
Principal Component Analysis (PCA) and on Multi Layer Perceptrons
(MLP's), a common type of ANN's, which provide a tool to approximate
the polarized radiative transfer.  At first, local Stokes spectra are
decomposed into their respective eigenspectra via PCA.  A set of
MLP's, is then trained to calculate the reduced eigenspectrum of
Stokes profiles as a function of line of sight, effective temperature
and the magnetic field configuration.  The accuracy of the PCA-MLP
synthesis is evaluated by comparison with a conventional numerical
solution of the polarized RTE.  The precision and the impressive speed
(factor 1000) of the PCA-MLP spectrum synthesis offers a viable and
promising alternative for a rapid calculation of local and disk
integrated Stokes spectra.

----------------------------------