<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=ISO-8859-1">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<font face="Droid Sans">I need to perform<font face="Droid Sans"> a
seque<font face="Droid Sans">nce <font face="Droid Sans">of
optimization </font></font></font> calc<font face="Droid
Sans">ulations<font face="Droid Sans"> in which the <font
face="Droid Sans">initial guess of the i-th <font
face="Droid Sans">calculation</font></font></font></font></font><font
face="Droid Sans"><font face="Droid Sans"><font face="Droid Sans"><font
face="Droid Sans"></font></font> is the result of the i-1-th
calculation.<font face="Droid Sans"> <br>
<br>
The objective function remains the same on these
calculations—only the <font face="Droid Sans">equa<font
face="Droid Sans">lity </font></font>constraints suffer
some slight modifications.<font face="Droid Sans"> Therefore,
the <font face="Droid Sans">previous result is a excel<font
face="Droid Sans">lent initial guess for the next
calculation.<br>
<br>
</font></font></font></font></font></font><font
face="Droid Sans"><font face="Droid Sans"><font face="Droid Sans"><font
face="Droid Sans"><font face="Droid Sans"><font face="Droid
Sans"><font face="Droid Sans">I would like to know how I
can provide not only the primal va<font face="Droid
Sans">riables <i>x</i>, but <font face="Droid
Sans">also the Lagran<font face="Droid Sans">g<font
face="Droid Sans">e<font face="Droid Sans">
multipliers <i>y</i> and <i>z</i><font
face="Droid Sans">? I have observed that
providing only a good <i>x</i> is not
enough<font face="Droid Sans">, since
several iterations are required to
determine <i>y</i> and <i>z</i><font
face="Droid Sans">.<br>
<br>
<font face="Droid Sans">I believe
there is a catch also as to the <font
face="Droid Sans">initial value of
the barrier parameter <i>mu</i></font></font><font
face="Droid Sans">. I would be
appreciate for <font face="Droid
Sans">a</font>ny advice</font> on
this too.<br>
<br>
<font face="Droid Sans">One <font
face="Droid Sans">important
remark: on <font face="Droid
Sans">these ca<font face="Droid
Sans">lculations some <font
face="Droid Sans">variables
will be very near to the
boundary. This is what I
expect for those variables
that should in fact be on
the boundary, and the <font
face="Droid Sans">small
inaccuracy <font
face="Droid Sans">is
fine for me</font></font>.</font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font></font>
Therefore, the calculation should be able to start from an initial
guess very close to the boundary.<br>
<br>
Thanks,<br>
Allan<br>
</body>
</html>