The twentyseven package

[Tags:library, mit, program, test]

Solve 3×3×3 Rubik's cubes in the fewest possible moves. Or, if you can't wait, get close enough with the two-phase solver.

[Skip to Readme]


Versions 0.0.0
Dependencies base (>=4.8 && <5), containers (>=0.5), deepseq, directory, filepath, heap (>=1.0), monad-loops, MonadRandom, mtl (>=2.1), newtype (>=0.2), optparse-applicative, primitive (>=0.6), ref-fd (>=0.4), template-haskell, time (<1.6), transformers, twentyseven, vector (>=0.10) [details]
License MIT
Author Li-yao Xia
Category Algorithms
Home page
Uploaded Wed Mar 16 17:40:35 UTC 2016 by lyxia
Updated Tue May 17 16:01:52 UTC 2016 by lyxia to revision 1
Distributions NixOS:0.0.0
Downloads 63 total (4 in the last 30 days)
0 []
Status Docs available [build log]
Last success reported on 2016-03-16 [all 1 reports]
Hackage Matrix CI




Maintainer's Corner

For package maintainers and hackage trustees

Readme for twentyseven

Readme for twentyseven-0.0.0


Rubik's cube solver in Haskell.

Inspired by Herbert Kociemba's Cube Explorer.

The main idea is to precompute, for every configuration, the number of moves required to put certain subsets of the 27 cubies composing the 3x3 Rubik's cube in their right place and/or in the right orientation. This gives lower bounds used for an A⋆-like search in the graph of scrambled cubes.

By default, a suboptimal "two-phase" solver is used, as it runs rather quickly. It currently solves 1000 random cubes (uniformly distributed) in about one minute. The optimal solver is quite slow however, taking between five minutes and two hours to solve a random cube (18 moves in average).

The solver must precompute a certain number of lookup tables, which can be stored in files. These tables take fifteen seconds to compute and weigh 13MB for the two-phase solver, compare that to about 8 hours and 2GB for the optimal one!

You may check the produced files with the checksums in ts-tables.sha256. A compressed archive (723MB) of all precomputed tables is available in the branch fetch-tables via git-lfs. Unzip it in $HOME/.27/, or wherever (see usage below).

Usage summary

twentyseven [-p] [--strict] [-d DIR] [--optimal]
  • For the first invocation, use -p to precompute nonexistent lookup tables, otherwise an error is thrown when twentyseven tries to load them;
  • --strict loads tables immediately, otherwise they are loaded "by need" (so you can also send it a cube to solve);
  • -d DIR specifies the directory where the tables should be read and written (default: $HOME/.27/).

The input is read line by line.

Input format

A line can be one of:

  • A string of 54 characters (ignoring spaces) from a set of (almost any) 6 characters. Each character then corresponds to the color of one facelet, in the order illustrated below.

    Output: a sequence of moves to unscramble it.

    Facelets are numbered in base 9. Faces 0,1,2,3,4,5 correspond to U,L,F,R,B,D.

                00 01 02
                03 04 05
                06 07 08
      10 11 12  20 21 22  30 31 32  40 41 42
      13 14 15  23 24 25  33 34 35  43 44 45
      16 17 18  26 27 28  36 37 38  46 47 48
                50 51 52
                53 54 55
                56 57 58
  • A dot . followed by a sequence of moves to scramble the cube.

    The basic moves are given by a letter in [ULFRBD], or their lowercase counterparts. Each letter corresponds to a clockwise quarter turn of the given face (up, left, front, right, back, down). The orientation is determined when looking directly at the turning face.

    For every basic move, an optional suffix [23'] allows to specify a half turn (e.g., U2), equivalent to a sequence of two quarter turns (UU), or a counterclockwise quarter turn (e.g., U3 or U') equivalent to a sequence of three clockwise (UUU).

    Output: a description of the resulting cube if the moves are applied starting from the solved cube (in the format above, with letters ULFRBD as colors).

  • The keyword random.

    Output: a random solvable cube with uniform distribution.

  • The keyword quit (or an end-of-file) terminates the interactive session.



$ echo quit|twentyseven -p --strict



qwqwqwqwq erererere tytytytyt rerererer ytytytyty wqwqwqwqw
qwqwqwqwq erqrerere tytytytyt rerererer ytytytyty wqwqwqwqw
111121111 333313333 222232222 444454444 666646666 555565555
111111214 223222222 131333333 344444444 555555555 666666666

The output then looks like this:

$ twentyseven < examples.txt
U2 D2 L2 R2 F2 B2
Facelets [6,18,11] ("qtq") do not match any regular cubie.
U D F B L R U2 R2 F2 R2 U2 L2 B2 U' D' B2
U L B' L R2 D R U2 F U2 L2 B2 U B2 D' B2 U' R2 U L2 R2 U
U D L R F B U2 B2 L2 F2 D2 B2 R2 U' D' L2
L U' F2 U F2 U L U' L2 D F2 D' F2

Detail of current heuristics

The distance estimations are based on cosets corresponding to the following elements.


Phase 1

  • Corner Orientation × UD Slice
  • Edge Orientation × UD Slice

It is possible to store the actual distances to the goal set in phase 1 but the current speed seems good enough for now.

Phase 2

  • Corner Permutation × UD Slice Permutation (Phase 2)
  • UD Edge Permutation (Phase 2) × UD SlicePermutation (Phase 2)


  • Corner Orientation × Edge Orientation × XY Slice Permutation, for XY in {UD, LR, FB}
  • Corner Orientation × Corner Permutation