By January 2020, Papadimitriou had been thinking about the pigeonhole principle for 30 years. So he was surprised when a playful conversation with a frequent collaborator led them to a simple twist on the principle that they’d never considered: What if there are fewer pigeons than holes? In that case, any arrangement of pigeons must leave some empty holes. Again, it seems obvious. But does inverting the pigeonhole principle have any interesting mathematical consequences?
It may sound as though this “empty-pigeonhole” principle is just the original one by another name. But it’s not, and its subtly different character has made it a new and fruitful tool for classifying computational problems.
To understand the empty-pigeonhole principle, let’s go back to the bank-card example, transposed from a football stadium to a concert hall with 3,000 seats—a smaller number than the total possible four-digit PINs. The empty-pigeonhole principle dictates that some possible PINs aren’t represented at all. If you want to find one of these missing PINs, though, there doesn’t seem to be any better way than simply asking each person their PIN. So far, the empty-pigeonhole principle is just like its more famous counterpart.
The difference lies in the difficulty of checking solutions. Imagine that someone says they’ve found two specific people in the football stadium who have the same PIN. In this case, corresponding to the original pigeonhole scenario, there’s a simple way to verify that claim: Just check with the two people in question. But in the concert hall case, imagine that someone asserts that no person has a PIN of 5926. Here, it’s impossible to verify without asking everyone in the audience what their PIN is. That makes the empty-pigeonhole principle much more vexing for complexity theorists.
Two months after Papadimitriou began thinking about the empty-pigeonhole principle, he brought it up in a conversation with a prospective graduate student. He remembers it vividly, because it turned out to be his last in-person conversation with anyone before the Covid-19 lockdowns. Cooped up at home over the following months, he wrestled with the problem’s implications for complexity theory. Eventually he and his colleagues published a paper about search problems that are guaranteed to have solutions because of the empty-pigeonhole principle. They were especially interested in problems where pigeonholes are abundant—that is, where they far outnumber pigeons. In keeping with a tradition of unwieldy acronyms in complexity theory, they dubbed this class of problems APEPP, for “abundant polynomial empty-pigeonhole principle.”
One of the problems in this class was inspired by a famous 70-year-old proof by the pioneering computer scientist Claude Shannon. Shannon proved that most computational problems must be inherently hard to solve, using an argument that relied on the empty-pigeonhole principle (though he didn’t call it that). Yet for decades, computer scientists have tried and failed to prove that specific problems are truly hard. Like missing bank-card PINs, hard problems must be out there, even if we can’t identify them.
Historically, researchers haven’t thought about the process of looking for hard problems as a search problem that could itself be analyzed mathematically. Papadimitriou’s approach, which grouped that process with other search problems connected to the empty-pigeonhole principle, had a self-referential flavor characteristic of much recent work in complexity theory—it offered a new way to reason about the difficulty of proving computational difficulty.
