[Paper Review] Permutation sorting and a game on graphs

Chad Jansen, Marion Scheepers+2 more|arXiv (Cornell University)|Nov 20, 2014

Genome Rearrangement Algorithms8 references3 citations

TL;DR

This paper introduces a two-player game on graphs based on context-directed swaps (cds) in permutation sorting, where players alternate applying cds operations until a fixed point is reached. The key contribution is proving that for permutations of size n divisible by 4, there exist cases where ONE wins even when the target set A is as small as (|P|−3)/4 + 1, demonstrating the optimality of prior bounds and revealing structural depth in cds dynamics via overlap and cycle graphs.

ABSTRACT

We introduce a game on graphs. By a theorem of Zermelo, each instance of the game on a finite graph is determined. While the general decision problem on which player has a winning strategy in a given instance of the game is unsolved, we solve the decision problem for a specific class of finite graphs. This result is then applied to a permutation sorting game to prove the optimality of a proportional bound under which TWO has a winning strategy.

Motivation & Objective

• To analyze the strategic structure of permutation sorting via context-directed swaps (cds) when permutations are not fully sortable.
• To formalize a two-player game on graphs where players alternate cds moves, with winning conditions based on reaching a target subset of fixed points.
• To determine precise conditions under which one player has a winning strategy in the cds game, particularly for permutations with full strategic piles.
• To establish the optimality of prior numerical bounds on winning strategies by constructing extremal examples.
• To introduce and analyze a new graph transformation, gcds, that models the effect of cds operations on overlap graphs.

Proposed method

• The paper defines the strategic pile P of a permutation π as the set of all possible cds fixed points reachable via cds operations.
• It introduces the overlap graph O(π), which encodes available cds moves, and defines a new graph transformation gcds that models the effect of a cds move on the graph.
• A game GCDS(G, A) is defined on finite graphs, where players alternate applying gcds transformations, and the winner is determined by whether the final fixed point lies in a target set A.
• The authors prove that for certain connected graphs G with |V| ≡ 3 mod 4 and |A| = (|V|−3)/4 + 1, player ONE has a winning strategy.
• This result is applied to permutation overlap graphs to show that for n ≡ 0 mod 4, there exist permutations π with |P| = n−1 and a set A with |A| = (|P|−3)/4 + 1 such that ONE wins in CDS(π, A).
• An end-game analysis is used to show that the bound is tight, as removing one element from A results in a win for TWO.

Experimental results

Research questions

• RQ1 For which classes of permutations and target sets A does player ONE have a winning strategy in the CDS(π, A) game?
• RQ2 Can the numerical bounds on winning strategies—such as |A| ≤ (1/4)|P| − 2 for TWO and |A| ≥ (3/4)|P| for ONE—be improved via structural analysis?
• RQ3 Is there a finite graph G and subset A such that |A|/|V| is arbitrarily small yet player ONE still has a winning strategy in GCDS(G, A)?
• RQ4 What is the complexity of determining whether a given finite graph is the move graph of some permutation under cds operations?
• RQ5 Can the structural properties of overlap graphs and gcds transformations be used to characterize all possible fixed points of cds operations?

Key findings

• For every n divisible by 4, there exists a permutation π ∈ S_n with a strategic pile P of size n−1 such that player ONE has a winning strategy in CDS(π, A) when |A| = (|P|−3)/4 + 1.
• This result shows that the bound |A| = (1/4)|P| − 2 for TWO’s winning strategy is optimal, as reducing |A| by one allows TWO to win.
• The paper constructs explicit examples where ONE wins despite A being a small fraction of the strategic pile, demonstrating that numerical advantage alone does not determine the outcome.
• The graph transformation gcds is introduced and shown to capture the effect of cds operations on the overlap graph, providing a new tool for analyzing cds dynamics.
• The authors prove that for connected graphs G with |V| ≡ 3 mod 4 and |A| = (|V|−3)/4 + 1, player ONE has a winning strategy in GCDS(G, A), and this structure is realizable as the move graph of a permutation.
• The paper establishes that the prior bounds from end-game analysis cannot be improved, as the constructed example achieves the threshold with equality.