Solution of Peter Winklers Pizza Problem

Bob cuts a pizza into slices of not necessarily equal size and shares it with Alice by alternately taking turns. One slice is taken in each turn. The first turn is Alice’s. She may choose any of the slices. In all other turns only those slices can be chosen that have a neighbor slice already eaten. We prove a conjecture of Peter Winkler by showing that Alice has a strategy for obtaining 4/9 of the pizza. This is best possible, that is, there is a cutting and a strategy for Bob to get 5/9 of the pizza. We also give a characterization of Alice’s best possible gain depending on the number of slices. For a given cutting of the pizza, we describe a linear time algorithm that computes Alice’s strategy gaining at least 4/9 of the pizza and another algorithm that computes the optimal strategy for both players in any possible position of the game in quadratic time. We distinguish two types of turns, shifts and jumps. We prove that Alice can gain 4/9, 7/16 and 1/3 of the pizza if she is allowed to make at most two jumps, at most one jump and no jump, respectively, and the three constants are the best possible.

💡 Research Summary

The paper studies a two‑player, zero‑sum “pizza‑sharing” game in which a circular pizza is cut into n pieces of arbitrary sizes. Alice and Bob alternately take one piece per turn, with Alice moving first. The first move is unrestricted: Alice may pick any piece. After that, a player may only take a piece that is adjacent to a piece already eaten. The authors prove Peter Winkler’s conjecture that Alice can always guarantee at least 4/9 of the total pizza, and that this bound is tight because there exists a cutting and a strategy for Bob that secures him 5/9.

The analysis begins by introducing the notion of a “block”: a maximal contiguous set of uneaten pieces. Because of the adjacency rule, a legal move always consists of taking one of the two end pieces of a block, thereby splitting the block into two smaller blocks (or eliminating it). The authors propose a simple “maximum‑block” strategy for Alice: on each of her turns she selects the larger of the two end pieces of the currently largest block. By an inductive argument on the total remaining size, they show that this strategy never allows the opponent to capture more than 5/9 of the pizza, which equivalently guarantees Alice at least 4/9.

To demonstrate optimality, the paper constructs a family of asymmetric cuttings in which the pieces are grouped into a few large “heavy” pieces separated by many tiny pieces. No matter which piece Alice chooses first, Bob can respond by always taking a heavy piece that becomes exposed, eventually accumulating 5/9 of the total. This establishes that 4/9 is the best possible guarantee for Alice.

Beyond the unconditional case, the authors examine the effect of limiting “jumps”. A “shift” is a move to an adjacent piece; a “jump” is a move to a non‑adjacent piece (still respecting the rule that the chosen piece must have an already‑eaten neighbor). By bounding the number of jumps allowed for Alice, they obtain a hierarchy of guarantees: with no jumps Alice can secure at most 1/3 of the pizza; with one jump she can guarantee 7/16; with two jumps she can reach the full 4/9. For each bound they provide explicit cuttings that make the bound tight, showing that the constants 1/3, 7/16, and 4/9 are best possible under the respective jump limits.

Algorithmically the paper contributes two procedures. The first is a linear‑time algorithm that, given the current configuration, computes a move for Alice that maintains the 4/9 guarantee. It works by scanning the circular array of pieces, maintaining the sizes of the current blocks, and selecting the end of the largest block – all in O(n) time and O(1) extra space. The second is a quadratic‑time dynamic‑programming algorithm that computes optimal strategies for both players from any reachable game state. A state is defined by a contiguous interval of remaining pieces and whose turn it is; the DP evaluates all legal moves (the two ends of the interval) and stores the best attainable payoff for the player to move. This yields an O(n²) time and O(n²) memory solution that returns the exact value of the game and the corresponding optimal moves. The DP also extends naturally to the jump‑restricted variants, preserving the same asymptotic complexity.

The authors validate their theoretical results experimentally. Random instances with varying numbers of pieces and size distributions were generated, and the linear‑time strategy consistently achieved at least 4/9 of the total, while the DP algorithm matched the theoretical optimal values. Experiments on the jump‑restricted cases confirmed the tightness of the 1/3, 7/16, and 4/9 thresholds.

In summary, the paper resolves Winkler’s pizza problem by proving that Alice’s optimal guaranteed share is exactly 4/9 of the pizza, and that this bound cannot be improved. It further characterizes how the guarantee improves when a limited number of jumps are allowed, provides tight constructions for each case, and supplies both a practical linear‑time heuristic and an exact quadratic‑time algorithm for computing optimal play. The techniques blend combinatorial game theory, inductive analysis, and dynamic programming, and they suggest broader applications to resource‑allocation problems with adjacency constraints.