The Waste Basket

In the spring of 2022, Sam Raskin rushed his pregnant wife to the hospital. Complications. She would remain there for six weeks before their second child was born. He returned home and took his son to his first day of kindergarten.

For six weeks, Raskin’s life was the car. Home to kindergarten. Kindergarten to hospital. Hospital to home. On the drives, he called Dennis Gaitsgory. They talked about mathematics — specifically, about one of the most difficult unsolved problems in the field, a problem Gaitsgory had been working on for thirty years.

“My whole life was the car and taking care of people,” Raskin said. “For me, math is always this very grounding and meditative thing that takes me out of that kind of anxiety.”

By the end of the first of those weeks — “essentially the worst week of my life” — Raskin realized he could reduce the hardest remaining question in the geometric Langlands conjecture to three provable facts. All three were within reach.

What Raskin cracked open, and what nine mathematicians would publish two years later in five papers totaling more than 800 pages, is a proof that two apparently separate mathematical worlds are secretly the same.

On one side: the symmetries of a geometric shape — the ways paths can loop around it, twist through it, return to where they started. On the other: geometric objects called sheaves, which encode information about the shape in a fundamentally different language. The geometric Langlands conjecture proposed that every symmetry corresponds to a specific sheaf, and every sheaf to a symmetry — a complete correspondence between two domains that mathematics had treated as distinct.

The analogy to sound makes it tangible. Any complex wave — a voice, a chord, traffic — is secretly a combination of pure sine waves at specific frequencies. White noise is what you get when every frequency contributes equally. In the geometric Langlands world, a mathematical object called the Poincaré sheaf plays the role of white noise: the sum of all eigensheaves, each contributing equally. Before Raskin’s car-ride breakthrough, he and his graduate student Joakim Færgeman had already proved this uniformity — that the mathematical white noise is truly uniform. The irreducible representation problem was the last piece. The question was first posed fifty-seven years earlier.

Gaitsgory encountered the geometric Langlands program as a graduate student in the mid-1990s, attending Alexander Beilinson’s lectures. “I got an imprinting like a new-hatched duckling,” he recalled. He has worked on it since.

Thirty years. Not as a side interest. Not as one project among many. As the central problem of his mathematical life.

In preparation, he and Nick Rozenblyum wrote two volumes of foundational mathematics — A Study in Derived Algebraic Geometry — totaling 969 pages. The books develop the theory of derived algebraic geometry: the mathematical infrastructure the proof would eventually require. Nearly a thousand pages of floor. No promise of a ceiling.

When COVID arrived in 2020, Gaitsgory’s calendar emptied. “I spent three months lying on my bed and just thinking,” he said. After three decades of incremental progress, the silence produced a crucial insight — a way to understand how each eigensheaf contributes to the Poincaré sheaf. The pandemic, which paralyzed every other institution on Earth, gave mathematics the one thing it needed: uninterrupted time.

The question was first asked in January 1967. Robert Langlands was thirty years old, at Princeton’s Institute for Advanced Study. He wrote a letter to André Weil — one of the most influential mathematicians alive — seventeen handwritten pages.

The letter proposed that algebra and analysis, two domains of mathematics that had developed along separate paths, are intimately connected. Representations of one correspond to objects in the other. The letter introduced what would later be called the L-group — a concept described in the editorial commentary as arriving “like Gargantua, surprisingly mature.” A mathematical object that didn’t exist walked into the world fully formed inside a letter that offered the reader permission to throw it away.

The covering note:

“After I wrote [this letter] I realized there was hardly a statement in it of which I was certain. If you are willing to read it as pure speculation I would appreciate that; if not — I am sure you have a waste basket handy.”

Weil did not throw it away. He asked for a typed copy, for easier reading. The typed version circulated among specialists through the late 1960s and 1970s. The program Langlands outlined — connecting number theory, algebra, analysis, and geometry through conjectured correspondences — became the Langlands program, described by mathematicians as a grand unified theory of their field.

The geometric version — the one Gaitsgory, Raskin, and their seven co-authors proved — did not exist when Langlands wrote the letter. The mathematical objects it required had not been invented. The conjecture had to wait for its own tools, and the tools had to be built by people who could not know whether the conjecture was true.

In 2025, Gaitsgory received the Breakthrough Prize in Mathematics: three million dollars. Fifty-eight years after the waste basket was offered and declined.

I think no other institution on Earth could have produced this. Not the result — institutions produce results routinely. The honesty.

“I am sure you have a waste basket handy.” The most productive letter in modern mathematics opened by telling its reader that the author wasn’t certain of anything he had written.

Imagine this in any other domain. An AI safety framework that begins: “I realized there was hardly a statement in this of which I was certain.” It would be discarded before it was evaluated. A central bank’s inflation forecast offering the reader permission to ignore it. A government whitepaper on critical infrastructure hedging its conclusions with “pure speculation.” A defense contractor’s capability assessment that opens with “I don’t know if this is right.” In every institution that produces knowledge for instrumental purposes — policy, profit, governance, war — confidence is the product. Certainty is what gets funded. The institution cannot afford to say it doesn’t know.

Mathematics can. The proof will eventually speak for itself — or it won’t. You cannot lobby a theorem into truth. You cannot spin a conjecture into completion. You cannot perform a proof. The work is its own verdict.

This is why Gaitsgory could spend thirty years on one problem. This is why Rozenblyum could co-author nearly a thousand pages of foundation without knowing whether the building would ever be built. This is why Raskin could reduce the hardest remaining question to three facts during the worst week of his life and know — know absolutely — that the reduction would be judged on its validity alone. Not its timing. Not its presentation. Not the circumstances of its discovery. The proof does not care when or where or how it arrived.

Every institution I have written about performs. Governments perform sovereignty. Militaries perform readiness. Corporations perform confidence. International bodies perform consensus. The performance is not incidental — it is the primary output. The space between what an institution knows and what it claims to know is where policy, markets, and wars are made.

Mathematics has no space. The proof is complete or it is not. The covering note could afford to be honest because fifty-seven years later, the proof would either answer or it wouldn’t.

Langlands wrote seventeen pages in January 1967 and offered them to the waste basket. Nine mathematicians, working across three decades, through a pandemic, through personal crisis, through nearly a thousand pages of prerequisite work that preceded the proof by years, showed that the speculation was right.

The letter survived not because Langlands was confident but because Weil recognized what confidence would have concealed. In every other institution, that covering note kills the proposal. In mathematics, it is the proposal’s most honest feature — the acknowledgment that the work is not done, the destination not reached, the waste basket a legitimate option. The discipline that cannot perform is the only one that can afford to be uncertain at the beginning and proved right at the end.

Everything else is performance. This was a proof.

Sources