<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&#8212;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>