Yes - this is correct. It is like the standard two-phase revised-simplex method.<br><br><br><div class="gmail_quote"><span dir="ltr"></span><br><blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">
Is the method for achieving master problem feasibility described<br>
anywhere? From what I can tell, looking at the code and the master<br>
problems being generated, there's a two phase approach where:<br>
- during the first phase DIP minimizes infeasibility until enough<br>
columns have been generated to ensure feasibility.<br>
- this is done by adding positive and negative slack to all constraints.<br></blockquote><div><br>For 'E' it uses 2 slacks. For 'L' or 'G' it only needs one.<br><br><br><br> <br></div><blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">
- once feasibility is achieved the algorithm switches to the real<br>
restricted master problem without the "soft" constraints.<br></blockquote><div><br><br>Yes, once the Phase 1 objective is 0 (minimizing the slacks), it fixes all the slacks to 0 (essentially removing them) and sets the real (Phase 2) objective. Since we do full branch-price-and-cut, we might fall back into Phase1 after cuts or branching - so we keep the slacks around and unfix them when needed.<br>
<br><br><br></div></div>