# EMFLAG

n}\newq\\{\tt !PXEM `n`}\warnq}&\parbox[t]{8.6cm}{
requests \ASReml use
Expectation-Maximization (EM) rather than Average Information (AI) updates
when the AI updates would make a US structure non-positive definite.
This only applies to US structures and is still under development.
When the !GP is associated with a US structure, \ASReml checks whether
the updated matrix is positive definite (PD).
If not, it replaces the AI update
with a single round of EM. If the non PD characteristic is transitory, then
the EM update is only used as necessary. If the converged solution
would be non PD, there will be a EM update each iteration even
though !EM is omitted.
Faster convergence
is achieved for G structures in this situation if the !EM
or !PXEM qualifiers are used.
!EMFLAG 1 EM in US term involving gamma(EMLEV)\\
!EMFLAG 2 PXEM\\
!EMFLAG 3 EM on all US\\
!EMFLAG 4 PXEM on all US\\
!EMFLAG 5 Score EM only\\
!EMFLAG 6 Single local PXEM\\
!EMFLAG 7 EM Score + 1 EM local\\
!EMFLAG 8 EMSCORE + 1 PXEM local
}\vspace{.25cm}\\
%\parbox[t]{4cm}{\tt !EQO {\em n}}&\parbox[t]{8.6cm}{ controls the ordering of equations during the inversion process. This qualifier is not useful to the normal user and is presently only for testin
%\parbox[t]{4cm}{\tt !EXTRA {\em n}}&\parbox[t]{8.6cm}{ forces \ASReml
% to perform ` n` more iterations after it has satisfied the
% convergence criterion. !MAXIT overrides {\tt !EXTRA}.}\vspace{.25cm}\\
\parbox[t]{4cm}{{\tt !EQORDER} `o` \newq}&\parbox[t]{8.6cm}{
modifies the algorithm used for choosing the order for solving the
mixed model equations. A new algorithm devised for release 2 is now the default and is formally selected by !EQORDER 3. The algorithm used
for release 1 is essentially that selected by !EQORDER 1.
The new order is generally superior. !EQORDER -1 instructs
\ASReml to process the equations in the order they are specified
in the model. Generally this will make a job much slower, if it can run at all.
It is useful if the model has a suitable order as in the IBD model
% can be set up is disastrous but in one particular
%case is advantageous. It is the case where the model is specified
\\\tab Y $\sim$ mu !r !\{ giv(id) id !\}\\
giv(id) invokes a dense inverse of an IBD matrix
and id has a sparse structured inverse of an additive relationship matrix. While {\tt !EQORDER 3} generates a more sparse solution,
!EQORDER -1 runs faster.}\vspace{.25cm}\\
**Return to start**