Re: [lmi] An experimental change for withdrawal solves

[Greg Chicares]

[resending due to apparent ISP problem]

On 2009-02-02 03:06Z, Greg Chicares wrote:
[...]
> That's why we need to test this work carefully. Solves are subtle.

Try the $0 solve with a product that sets 'LapseIgnoresSurrChg',
which can be found thus:

$cd /opt/lmi/data
$grep -A2 LapseIgnoresSurrChg *.dbt |sed -e '/scalar\|name\|--/d' -e'/1$/!d'

With HEAD prior to 20090206T0153Z, I saw a maximal withdrawal at
the beginning of the solve period, instead of a level withdrawal
throughout it. The objective function was exactly zero on the
second iteration (with a billion-dollar withdrawal); iteration
ceases as soon as a zero of the objective function is found.