<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</head>
<body text="#000000" bgcolor="#FFFFFF">
Hi Marco,<br>
<br>
looks as if some propagators do their job and reduce some of the
variables domains before the LP is resolved. By executing "display
statistics" you get an overview which propagator plugin is
responsible for domain reductions, perhaps the root reduced cost
propagator.<br>
<br>
De rien,<br>
Gregor<br>
<br>
<div class="moz-cite-prefix">Am 17.10.19 um 12:55 schrieb Marco
Lübbecke:<br>
</div>
<blockquote type="cite"
cite="mid:CAM5ELArnARXw+G9R9_3Ru7g6VnXgJKQyS=iGsqRj1gJLMS22mg@mail.gmail.com">
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
<div dir="ltr">
<div>Wisdom of the crowd:</div>
<div><br>
</div>
<div>I use SCIP 6.0.1 to compute the root node dual bound of a
MILP (say, miplib2010/mine-166-5.mps.gz), without any cuts
added, so I</div>
<div><br>
</div>
<div>set separating emphasis off</div>
<div>set limits nodes 1</div>
<div><br>
</div>
<div>To be on the safe side, I also disable conflict analysis,
however, I don't know whether this is necessary</div>
<div><br>
</div>
<div>set conflict enable FALSE</div>
<div><br>
</div>
<div>I expected that this would solve me the LP relaxation of
the presolved instance, which is not true.</div>
<div><br>
</div>
<div><br>
</div>
<div> time | node | left |LP iter|LP it/n| mem |mdpt |frac
|vars |cons |cols |rows |cuts |confs|strbr| dualbound |
primalbound | gap <br>
V 1.0s| 1 | 0 | 0 | - | 40M| 0 | - | 712
|6725 | 712 |6725 | 0 | 0 | 0 |-1.008549e+09
|-1.691274e+08 | 496.33%<br>
1.2s| 1 | 0 | 1839 | - | 40M| 0 | 481 | 712
|5947 | 712 |6725 | 0 | 0 | 0 |-7.233357e+08
|-1.691274e+08 | 327.69%<br>
u 1.5s| 1 | 0 | 4840 | - | 40M| 0 | - | 712
|5947 | 712 |6725 | 0 | 0 | 0 |-7.233357e+08
|-4.463478e+08 | 62.06%<br>
1.7s| 1 | 0 | 8009 | - | 41M| 0 | 480 | 712
|5502 | 712 |6725 | 0 | 0 | 0 |-7.233034e+08
|-4.463478e+08 | 62.05%<br>
2.8s| 1 | 0 | 8009 | - | 41M| 0 | - | 712
|5502 | 712 |6725 | 0 | 0 | 24 |-6.589159e+08
|-4.463478e+08 | 47.62%</div>
<div><br>
</div>
<div><br>
</div>
<div>I realize that, even though there will not be any
branching, branching appears to be "prepared" already and the
information gathered from the strong branching LPs is used in
the dual bound computation. So I</div>
<div><br>
</div>
<div>set branching random priority 100000000</div>
<div><br>
</div>
<div>My personal opinion is that the choice of branching rule
should *not* influence the root node dual bound, but it does:</div>
<div><br>
</div>
<div> time | node | left |LP iter|LP it/n| mem |mdpt |frac
|vars |cons |cols |rows |cuts |confs|strbr| dualbound |
primalbound | gap <br>
V 1.1s| 1 | 0 | 0 | - | 39M| 0 | - | 712
|6725 | 712 |6725 | 0 | 0 | 0 |-1.008549e+09
|-1.691274e+08 | 496.33%<br>
1.2s| 1 | 0 | 1839 | - | 39M| 0 | 481 | 712
|5947 | 712 |6725 | 0 | 0 | 0 |-7.233357e+08
|-1.691274e+08 | 327.69%<br>
u 1.5s| 1 | 0 | 4840 | - | 39M| 0 | - | 712
|5947 | 712 |6725 | 0 | 0 | 0 |-7.233357e+08
|-4.463478e+08 | 62.06%<br>
1.7s| 1 | 0 | 8009 | - | 39M| 0 | 480 | 712
|5502 | 712 |6725 | 0 | 0 | 0 |-7.233034e+08
|-4.463478e+08 | 62.05%<br>
1.7s| 1 | 2 | 8009 | - | 39M| 0 | 480 | 712
|5502 | 712 |6725 | 0 | 0 | 0 |-7.233034e+08
|-4.463478e+08 | 62.05%</div>
<div><br>
</div>
<div><br>
</div>
<div>OK, I accept that, but you see that last tiny improvement
in the dual bound that makes me suspicious: what happens in
addition (primal heuristic?) that impacts the root node dual
bound that I am not aware of?<br>
</div>
<div><br>
</div>
<div><br>
</div>
<div>Merci beaucoup</div>
<div>Marco</div>
<div><br>
</div>
<div><br>
</div>
<div><br>
</div>
<div><br>
</div>
</div>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<pre class="moz-quote-pre" wrap="">_______________________________________________
Scip mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Scip@zib.de">Scip@zib.de</a>
<a class="moz-txt-link-freetext" href="https://listserv.zib.de/mailman/listinfo/scip">https://listserv.zib.de/mailman/listinfo/scip</a>
</pre>
</blockquote>
<br>
</body>
</html>