The X axis is conflicts, Y axis is some measure of sovling. Every datapoint corresponds to a restart. You may zoom in by clicking on an interesting point and dragging the cursor along the X axis. Double-click to unzoom. Blue vertical lines indicate the positions of simplification sessions. Between the blue lines are search sessions. The angle of the "time" graph indicates conflicts/second. Simplification sessions are not detailed. However, time jumps during simplifcaition, and the solver behaviour changes afterwards.
| Abbreviation | Explanation |
|---|---|
| red. | reducible, also called learnt |
| irred. | irreducible, also called non-learnt |
| confl | conflict reached by the solver |
| learnt | clause learnt during 1UIP conflict analysis |
| trail depth | depth of serach i.e. the number of variables set when the solver reached a conflict |
| brach depth | the number of branches made before conflict was encountered |
| trail depth delta | the number of variables we jumped back when doing conflict-directed backjumping |
| branch depth delta | the number of branches jumped back during conflict-directed backjumping |
| propagations/decision | number of variables set due to the propagation of a decision (note that there is always at least one, the variable itself) |
| vars replaced | the number or variables replaced due to equivalent literal simplfication |
| polarity flipped | polarities of variables are saved and then used if branching is needed, but if propagation takes place, they are sometimes flipped |
| std dev | standard deviation, the square root of variance |
| confl by | the clause that caused the conflict |
| agility | See here. |
| glue | the number of different decision levels of the literals found in newly learnt clauses. See here |
| conflict after conflict % | How often does it happen that a conflict , after backtracking and propagating immeediately (i.e. without branching) leads to a conflict. This is displayed because it's extremely high percentage relative to what most would expect. Thanks to Said Jabbour for this. |
This webpage shows the partial solving of a SAT instance. I was amazed by Edward Tufte's work (hence the subtitle) and this came out of it.
Copyright Mate Soos, 2016. Licensed under CC BY-NC-SA 2.5