Given a set $S$ and a partial order $\prec$ over $S$, I'm looking for a way to "efficiently" generate a list of linear extensions of $\prec$. Suppose the partial order is given by a List of pairs $\{x,y\}$ such that $x,y\in S$. For example, if $S = \{a,b,c\}$, then $\{\{a,c\},\{b,c\}\}$ defines a partial order where $a \prec c$ and $b \prec c$.

Essentially, given a list $S$ and a list of pairs $P$, I want to generate a list of permutations which respect the orders of the pairs in $P$.

linearExtensions[{a, b, c, d, e}, {{a, b}, {c, b}, {c, d}, {e, d}}]
(* {{a, c, b, e, d}, {a, c, e, b, d}, {a, c, e, d, b}, {a, e, c, b, d},
    {a, e, c, d, b}, {c, a, b, e, d}, {c, a, e, b, d}, {c, a, e, d, b},
    {c, e, a, b, d}, {c, e, a, d, b}, {c, e, d, a, b}, {e, a, c, b, d},
    {e, a, c, d, b}, {e, c, a, b, d}, {e, c, a, d, b}, {e, c, d, a, b}} *)

Since $\{a,b\}$ is in $P$, the permutation $\{b, c, a, d, e\}$ is not a linear extension because $b$ comes before $a$.

I've written two different functions which do the job, both using pattern matching, but I've been using Mathematica for less than a week and am still getting into the functional mindset. I'm interested to see how seasoned Mathematica users would tackle this problem.

My first approach was to use NestWhile (edit: this is ugly! I did not know about the Fold[] function when I wrote this):

linearExtensions[set_, po_] := 
        patterns = {___, #[[1]], ___, #[[2]], ___}& /@ po;
        First@NestWhile[{Cases[#[[1]], First[#[[2]]]], Rest[#[[2]]]}&,
            {Permutations[set], patterns}, Length[#[[2]]] > 0&]

My second approach, which turned out to be significantly slower, was to expand a set of rules from the partial order and use Select on the list of permutations.

linearExtensions[set_, po_] :=
        poQ[rule_] := And @@ (MatchQ[rule, {___, #[[1]], ___, #[[2]], ___}]& /@ po);
        Select[Permutations[set], poQ[#]&]

Note that on a list of length $n$ there can be $\Omega(n!)$ linear extensions, so by "efficient," I don't mean polynomial-time.

  • $\begingroup$ What's $\Omega$? $\endgroup$ – alancalvitti Aug 4 '12 at 20:14
  • 1
    $\begingroup$ @alancalvitti The counter part to Big O. $\Omega$ is an asymptotic lower bound $\endgroup$ – rm -rf Aug 4 '12 at 20:43
  • $\begingroup$ @R.M, same as small $o$? $\endgroup$ – alancalvitti Aug 6 '12 at 2:28
  • 1
    $\begingroup$ @alancalvitti $\Omega$ is an asymptotic lower bound. Think of $O$ as $\le$, $o$ as $<$, and $\Omega$ as $\ge$. (The analogy falls apart for some pairs of functions which are neither $O$ nor $\Omega$ of each other.) $\endgroup$ – Zach Langley Aug 6 '12 at 2:37

Could use integer linear programming. In Mathematica this can be done with Reduce[]. One way (probably not the best) to set this up is shown below. It uses an array of 0-1 variables, where a 1 in position (j,k) will indicate that the kth element of the input variables goes in position j of a particular ordering.

Caveat: I may have mixed up rows and columns.

consistentOrders[elems_, pairorders_] := Module[
  {n = Length[elems], vars, x, fvars, c1, c2, c3, c4, constraints, 
   ineqs, solns},
  ineqs[{a1_, a2_}, n_, v_] := 
   Table[Total[Take[v[[All, a1]], j]] >= 
     Total[Take[v[[All, a2]], j]], {j, 1, n - 1}];
  vars = Array[x, {n, n}];
  fvars = Flatten[vars];
  c1 = Map[0 <= # <= 1 &, fvars];
  c2 = Thread[Total[vars] == 1];
  c3 = Thread[Total[Transpose@vars] == 1];
  c4 = Flatten[
    Map[ineqs[#, n, vars] &, pairorders /. Thread[elems -> Range[n]]]];
  constraints = Join[c1, c2, c3, c4];
  solns = Reduce[constraints, fvars, Integers];
  solns = 
   solns /. {(_ == 0) :> Sequence[], aa_ == 1 :> aa, And -> List, 
     Or -> List};
  Sort[solns /. x[i_, j_] :> elems[[j]]]

--- edit ---

The first three constraint sets are fairly standard for this type of 0-1 programming. The fourth constraint subset arises as follows. The idea is that if the jth list element must precede the kth, then the 1 in column j must occur in an earlier row than the one in column k. So for every 1<=m<=n-1 (n=dimension) the sum of the first m entries in col j >= corresponding sum in col k.

--- end edit --- Example:

consistentOrders[{a, b, c, d, 
  e}, {{a, b}, {c, b}, {c, d}, {e, d}}]

(* Out[83]= {{a, c, b, e, d}, {a, c, e, b, d}, {a, c, e, d, b}, {a, e, c,
   b, d}, {a, e, c, d, b}, {c, a, b, e, d}, {c, a, e, b, d}, {c, a, e,
   d, b}, {c, e, a, b, d}, {c, e, a, d, b}, {c, e, d, a, b}, {e, a, c,
   b, d}, {e, a, c, d, b}, {e, c, a, b, d}, {e, c, a, d, b}, {e, c, d,
   a, b}} *)

Bigger example:

vars = {a, b, c, d, e, f, g, h, i, j, k, l};
porderlist = {{a, c}, {b, c}, {f, g}, {g, e}, {d, a}, {h, i}, {i, 
    d}, {g, h}, {g, i}, {h, j}, {h, k}, {k, j}, {k, c}};

Timing[ss = consistentOrders[vars, porderlist];]

(* Out[81]= {60.03, Null}

In[82]:= Length[ss]

Out[82]= 12840 *)
  • $\begingroup$ Can you explain the fourth constraint? $\endgroup$ – Zach Langley Aug 6 '12 at 4:52
  • $\begingroup$ Not easily... See edited response. $\endgroup$ – Daniel Lichtblau Aug 6 '12 at 14:27
  • $\begingroup$ Ohhh, I see. Nice! I didn't think to use integer programming for this. This is certainly the most efficient solution so far. $\endgroup$ – Zach Langley Aug 6 '12 at 14:34
  • $\begingroup$ Although it's not as concise, I'm accepting this answer because its performance trumps the others'. $\endgroup$ – Zach Langley Aug 10 '12 at 23:49

A more concise version of the first approach, using Fold[] instead of NestWhile[]:

linearExtensions[set_, po_] :=
    Fold[Cases, Permutations[set], {___, #1, ___, #2, ___} & @@@ po]
  • 1
    $\begingroup$ Or, linearExtensions[set_, po_] := Fold[Cases[#1, Riffle[#2, ___, {1, -1, 2}]] &, Permutations[set], po] $\endgroup$ – J. M. is away Aug 4 '12 at 17:40

My first modest attempt:

linearExtensions[set_List, po_?MatrixQ] :=
    Select[Permutations[set], Complement[po, Subsets[#, {Last[Dimensions[po]]}]] === {} &]

My second modest attempt:

linearExtensions[set_List, po_?MatrixQ] :=
  Select[Permutations[set], And @@ Map[Function[p, LongestCommonSequence[#, p] === p], po] &]
  • $\begingroup$ Nice! I didn't realize Subsets preserved order. Is there no nicer way to determine if a list is a subset of another than what you're doing here? $\endgroup$ – Zach Langley Aug 4 '12 at 15:13
  • $\begingroup$ If there is one, I do not know it. $\endgroup$ – J. M. is away Aug 4 '12 at 15:57
  • $\begingroup$ Interestingly, the pattern matching seems to be much faster than this. Try, for example, with linearExtensions[{a, b, c, d, e, f, g, h, i}, {{a, c}, {b, c}, {f, g}, {g, e}, {d, a}, {h, i}, {i, d}, {g, h}}]. $\endgroup$ – Zach Langley Aug 5 '12 at 14:46
  • $\begingroup$ @Zach, yes; it seems your approach is a good bit faster... anyway, I gave another possibility. $\endgroup$ – J. M. is away Aug 5 '12 at 14:47

I like @Daniel's approach using 0/1 integer programming. Instead of Reduce, however, one can achieve more optimal performance using SatisfiabilityInstances. In the code below, I've written the main function (AllTotalOrderings) for determining all total orderings of a partially ordered graph. It is then easy to encode the desired linearExtensions function as a simple call to AllTotalOrderings, for a suitably constructed graph. To further increase performance I've made use of CanonicalGraph (so as to introduce the constraints to the SAT solver in the most efficient order) and TransitiveReductionGraph (to minimize the number of constraints given to the SAT solver.) Note that I'm using the undocumented Method->"BDD" to call SatisfiabilityInstances, and BooleanConvert to preprocess the constraints into a natively-optimized Boolean Decision Diagram. The ordering is extracted from the output of the SAT solver using ArrayReshape and the AdjacencyLists property of SparseArray.



linearExtensions[set_List, po_?MatrixQ]:=AllTotalOrderings[Graph[set,DirectedEdge@@@po]];

This give dramatically improved performance. For the example in question,

With[{vars = {a, b, c, d, e, f, g, h, i, j, k, l},
porderlist = {{a, c}, {b, c}, {f, g}, {g, e}, {d, a}, {h, i}, {i, d}, {g, h}, {g, i}, {h, j}, {h, k}, {k, j}, {k, c}}},
Timing[Length[linearExtensions[vars, porderlist]]]]

(*Out[] = {1.53125, 12840}*)
  • $\begingroup$ Very nice first post! Welcome to MSE. $\endgroup$ – Carl Woll Jun 12 at 16:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.