Source-linked AI summary
Constructing circuit codes by permuting initial sequences
Ed Wynn
TL;DR
Finding long d-coils is the snake-in-the-box problem. This paper presents two search methods; its constrained searches can compete with more general searches and produce new records in several cases.
Problem
The paper addresses the search for long d-coils, known as the snake-in-the-box problem.
Source: §I · p. 1Method
The paper presents two methods, including one that builds a circuit code from multiple permuted copies of an initial sequence.
Source: §VIII · p. 7Results
The constrained searches can be competitive with more general searches and produced new records in several cases.
Source: §VIII · p. 7Takeaways & Limitations
The paper also presents a method for detecting symmetries.
Source: §VIII · p. 7
Abstract
from arXiv · showhide
Two new constructions are presented for coils and snakes in the hypercube. Improvements are made on the best known results for snake-in-the-box coils of dimensions 9, 10 and 11, and for some other circuit codes of dimensions between 8 and 13. In the first construction, circuit codes are generated from permuted copies of an initial transition sequence; the multiple copies constrain the search, so that long codes can be found relatively efficiently. In the second construction, two lower-dimensional paths are joined together with only one or two changes in the highest dimension; this requires a search for a permutation of the second sequence to fit around the first. It is possible to investigate sequences of vertices of the hypercube, including circuit codes, by connecting the corresponding vertices in an extended graph related to the hypercube. As an example of this, invertible circuit codes are briefly discussed.
I. INTRODUCTION
Long d-coils are the subject of the snake-in-the-box problem. This paper presents two new circuit-code constructions and reports results from them.
I. INTRODUCTION
- Long d-coils are sought in the snake-in-the-box problem.The paper notes that d-coils have also been called snakes.Source: §I · p. 1
- A circuit code is represented by a cyclic transition sequence, with each entry specifying the coordinate changed at that step.The sequence wraps around modulo its length.Source: §I · p. 1
- The paper presents two new constructions of circuit codes and reports results from them.Source: §I · p. 1
II. PERMUTED CIRCUIT CODES
A permuted circuit code repeats an initial transition sequence after applying a coordinate permutation to its copies. The construction uses those linked copies to constrain a backtracking search for long circuit codes.
II. PERMUTED CIRCUIT CODES
- Duplicate skeletons from the same permutation are rejected by comparing them up to isomorphism.Source: §II · p. 2
- The algorithm combines a permutation and initial leap to form a returning skeleton, then backtracks over an initial sequence while testing the spread condition.Proposed transitions induce changes in each permuted sequence.Source: §II · pp. 2–3
- Permuted copies of proposed transitions impose extra constraints, allowing a short initial sequence to produce a long coil.Source: §II · p. 3
- A six-dimensional example forms a permuted coil of length 24 from an initial sequence and its compatible permuted copies.The example uses permutation (345201) and initial leap (0,1,1,0,0,0).Source: §II · p. 3
III. EXAMPLES OF PERMUTED CIRCUIT CODES
The examples include longer d-coils for dimensions 10 and 11 than the previously known lengths, alongside results for spread-3 codes and natural coils.
III. EXAMPLES OF PERMUTED CIRCUIT CODES
- The previous section’s construction produced d-coils of lengths 348 and 640, longer than the previously known longest lengths of 344 and 630.Source: §III · p. 3
- For spread-3 circuit codes in dimensions 10 and 11, the construction produced lengths 100 and 160, exceeding previous lengths 86 and 154.Source: §III · p. 3
- The reported search for 8-coils of this type, starting at x0, produced longest results of length 94, compared with 96 for the longest known general 8-coil.Source: §III · p. 4
- Special cases with a truncated final permuted sequence generally yielded shorter results than full permuted repetitions with similar parameters.Source: §III · p. 4
IV. COMPUTATION TIMES
The searches found long coils in dimensions 10 and 11, including a length-640 11-coil. The authors also describe increasing search difficulty beyond dimension 12 and limits to comparing reported computation times.
- Large leap periods can require only a short initial sequence for a long coil, while new vertices must avoid occupied vertices under the required spread condition.Source: §IV · p. 4
- Length 320 was found for a 10-coil with leap period P = 16; searching periods P > 12 took 80 minutes.Source: §IV · p. 4
- Length 576 was the longest reported 11-coil for leap period P = 24; searching periods P > 22 took 1320 minutes.Source: §IV · p. 4
- A 6.9-week search found its first length-638 result after about 2 days, while a restricted search at P = 20 found a length-640 11-coil.Source: §IV · p. 4
- An exhaustive search for a length-48 7-coil took 3480 minutes, and the longest 8-coil exhaustive search was described as prohibitive with current methods.The paper notes that like-for-like comparison with an earlier exhaustive search is difficult.Source: §IV · p. 4
- For d > 12, large leap periods are quickly unsuccessful and short periods take prohibitively long to search exhaustively; the construction is not expected to have a broad useful range without modification.Source: §IV · p. 4
V. USE OF GRAPHS TO COMPARE AND INVESTIGATE SEQUENCES OF HYPERCUBE VERTICES
The section uses an extended-graph representation to compare hypercube vertex sequences and study their symmetries, including invertible coils. It also reports invertible sequences and notes that all but one have vertex-fixed inversions.
- The extended graph encodes hypercube vertices through coordinate nodes and represents sequences by links between original vertices, preserving hypercube automorphisms.Sequences equivalent up to hypercube symmetries are isomorphic in the extended graph; canonical forms can be compared using NAUTY.Source: §V · p. 5
- Although generally slower than brute-force methods for simple paths and circuits, the extended-graph method also handles chains or skeletons of disconnected vertices.Source: §V · p. 5
- A d-coil is invertible when its two oriented circuits are isomorphic; in an isomorphism, one pair of vertices or changes must remain fixed, distinguishing vertex-fixed from change-fixed inversions.Source: §V · p. 6
- Of the 4 maximal 6-coils (N = 26), only 1 is invertible, and d-coils appear to be rare.Source: §V · p. 6
- All but one of the referenced invertible sequences have vertex-fixed inversions.Source: §V · p. 6
VI. CONSTRUCTION USING LOWER-DIMENSIONAL SNAKES
The construction combines two lower-dimensional snakes by permuting one change sequence to make the paths fit. The reported combinations improved the known lengths, though the approach has limits for larger dimensions.
- Given two d-snakes, the method generally permutes one change sequence and appends its changes one by one to seek a valid combination.Source: §VI · p. 6
- When assigning permutations, the search backtracks if the new snake violates the required spread condition, trying another permutation or an earlier assignment.Source: §VI · p. 7
- The construction produced a length-188 coil and a length-190 snake, exceeding the previously known lengths of 180 and 188, respectively.Source: §VI · p. 7
- The authors do not expect this construction to be widely useful for large d.Source: §VI · p. 7
VII. RESULTS FROM DIRECT SEARCHES
Direct backtracking searches found circuit codes longer than the previous known records for three dimension-and-spread settings. The search rejects partial sequences with a lower-renumberable subsequence, but is not exhaustive for snakes.
VII. RESULTS FROM DIRECT SEARCHES
- Backtracking rejected partial circuit-code sequences when a forward or backward subsequence could be renumbered below the starting subsequence.Every subsequence was tested at every step.Source: §VII · p. 7
- The snake search was not exhaustive, and more efficient search strategies may be available.The approach avoids repeatedly finding circuits from multiple starting points.Source: §VII · p. 7
- Lengths 58, 58, and 50 were found for d = 9, spread 3; d = 12, spread 5; and d = 13, spread 6, exceeding the respective previous records of 56, 56, and 48.The earlier records are cited as,, and, respectively.Source: §VII · p. 7
VIII. CONCLUSIONS
The conclusions describe two constrained search methods for circuit codes and a graph-based method for detecting symmetries. The constrained searches can be competitive with general searches and have produced new records in several cases.
VIII. CONCLUSIONS
- Both search methods constrain the search to reduce its combinatorial explosion.One assembles a sequence from permuted copies of a shorter sequence; the other combines two lower-order sequences using only one pair of changes in the highest coordinate.Source: §VIII · p. 7
- The constrained searches can be competitive with general searches and have produced new records in several cases.The paper presents those cases among its results.Source: §VIII · p. 7
- A graph-based method detects circuit-code symmetries by adding vertices that represent nodes related to, but not initially linked with, hypercube nodes.The method brings about symmetries equivalent to hypercube symmetries through additional vertices.Source: §VIII · p. 7
DETAILS OF CIRCUIT CODES
This section presents explicit transition-sequence listings for circuit codes, including reported sequence lengths and a constraint on the distribution of coordinate changes.
DETAILS OF CIRCUIT CODES
- One listed sequence has fewer than one quarter of its changes in a single coordinate.The passage then introduces the full sequence.Source: DETAILS OF CIRCUIT CODES · p. 8
- The listed transition sequences include N = 94 and N = 188.The entries are preceded by transition-sequence strings.Source: DETAILS OF CIRCUIT CODES · p. 8
- The cited transition-sequence entry gives the string 4501430540 and reports N = 190, preserving that supported count without extending it to other entries.Source: DETAILS OF CIRCUIT CODES · p. 9