🔢 mathematics

A Milestone in Formalization: The Sphere Packing Problem in Dimension 8

This paper discusses the successful formal verification of Maryna Viazovska's 2016 solution to the 8-dimensional sphere packing problem using the Lean Theorem Prover, highlighting a collaborative milestone between human mathematicians and the autoformalization model "Gauss."

Original authors: Sidharth Hariharan, Christopher Birkbeck, Seewoo Lee, Ho Kiu Gareth Ma, Bhavik Mehta, Auguste Poiroux, Maryna Viazovska

Published 2026-04-28

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Sidharth Hariharan, Christopher Birkbeck, Seewoo Lee, Ho Kiu Gareth Ma, Bhavik Mehta, Auguste Poiroux, Maryna Viazovska

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Great Mathematical Puzzle: Packing the Perfect Suitcase

Imagine you are trying to pack a suitcase with as many oranges as possible. If you just throw them in, you’ll have big gaps of air between them. But if you arrange them in a very specific, rhythmic pattern, you can squeeze in more.

For centuries, mathematicians have been obsessed with this: What is the absolute best way to pack "spheres" (like oranges or marbles) into a space so that no space is wasted?

In our normal 3D world, we have a good idea of the answer. But in higher dimensions—mathematical "worlds" that we can't see but can calculate—the patterns become incredibly complex. In 2016, a mathematician named Maryna Viazovska solved this puzzle for the 8th dimension. She found a "magic" pattern (called the $E_8$ lattice) that is so perfect, it’s like finding the one perfect way to stack blocks in a universe we can't even touch.

The Problem: The "Human Error" Factor

Even though Viazovska’s math was brilliant, math is hard. Even the smartest humans can make a tiny "typo" in a calculation that is hundreds of pages long. To be 100% sure the answer is perfect, mathematicians use Formal Verification.

Think of this like a "Math Spell-Checker." Instead of a human reading a paper and saying, "Yeah, this looks right," they translate the entire proof into a super-strict computer language called Lean. Lean is like a digital judge: it doesn't care how smart you are; if your logic has even a microscopic crack, it will refuse to say "True."

The Breakthrough: The Human-AI Tag Team

This paper describes a massive milestone reached in February 2026. The team didn't just use humans to write the code; they used a specialized AI model named "Gauss."

Think of the collaboration like building a massive cathedral:

The Humans (The Architects): They designed the blueprints. They decided which "stones" (mathematical definitions) to use and where the "arches" (the main theorems) should go. They did the heavy lifting of the most complex, creative parts.
The AI, Gauss (The Master Mason): Once the blueprints were ready, Gauss went to work. It was incredibly fast, laying down thousands of "bricks" (lines of code) in just five days. It filled in the massive, tedious gaps that would have taken humans months to type out.

The "Messy" Reality of AI

The paper also admits that working with AI is a bit like hiring a very fast, very literal intern.

Gauss was amazing at finishing the job, but it wasn't "elegant." While a human mathematician tries to write a proof that is beautiful and concise (like a well-written poem), Gauss wrote a proof that was massive and repetitive (like a giant instruction manual that repeats "Turn left" every five seconds).

The team had to go back and "clean up" the AI's work—trimming the fat and making the code reusable for future mathematicians.

Why Does This Matter?

You might ask, "Who cares about packing oranges in 8 dimensions?"

While the problem itself is theoretical, the tools being built are revolutionary. By teaching AI to work alongside humans to solve the hardest problems in existence, we are building a future where:

Mathematical Truth is Absolute: We can move from "we think this is true" to "we have digitally proven this is true."
AI Becomes a Partner, Not Just a Tool: We are learning how to guide AI to do high-level reasoning, not just summarize text or generate images.

In short: We’ve successfully used a "Digital Judge" and an "AI Mason" to prove that the most beautiful pattern in the 8th dimension is, indeed, perfect.

Technical Summary: A Milestone in Formalization: The Sphere Packing Problem in Dimension 8

1. The Problem: Sphere Packing in $\mathbb{R}^8$

The sphere packing problem asks for the densest possible arrangement of non-overlapping $n$ -dimensional spheres in Euclidean space $\mathbb{R}^n$ . While the solutions for $n=1, 2,$ and $3$ are well-established, the problem in dimension 8 remained a major challenge until 2016, when Maryna Viazovska proved that the $E_8$ lattice provides the optimal packing.

Viazovska’s proof utilized the Cohn-Elkies linear programming bound, which states that if a specific type of "magic" Schwartz function $f$ exists (satisfying certain positivity and Fourier transform conditions), it can provide an upper bound on the sphere packing constant $\Delta_d$ . Viazovska constructed this function using quasimodular forms.

2. Methodology: Formalization via Lean and Autoformalization

The primary objective of this project was to transition Viazovska’s informal mathematical proof into a formally verified, machine-checked proof using the Lean Theorem Prover. The methodology involved a hybrid approach:

Human-Led Infrastructure: Human formalizers (led by Hariharan and Viazovska) built the foundational mathematical libraries required for the proof. This included defining the geometry of sphere packings, the properties of the $E_8$ lattice, and the theory of modular forms (Eisenstein series, Jacobi theta functions, etc.).
Autoformalization (The 'Gauss' Model): To bridge the gap between human-written definitions and a complete proof, the project employed 'Gauss', an autoformalization model developed by Math, Inc. Gauss was tasked with generating the bulk of the proof code based on human-provided blueprints and Lean definitions.
Metaprogramming: To handle the heavy computational requirements of complex number arithmetic and limit proofs, the team developed custom Lean tactics, such as norm_numI (for complex number normalization) and tendsto_cont (for simplifying limit proofs via continuity).

3. Key Technical Contributions

Construction of the 'Magic' Function: The team successfully formalized the construction of the function $g(x)$ as a linear combination of radial Schwartz Fourier $\pm 1$ -eigenfunctions. Because Lean's mathlib lacked advanced contour integration machinery, the team had to define specific rectangular contours to allow for the deformation of unbounded paths, enabling the proof of the function's eigenfunction properties.
Modular Form Theory: The project formalized essential identities involving the weight 2 Eisenstein series ( $E_2$ ), the discriminant form ( $\Delta$ ), and Ramanujan’s identities. This was crucial for proving the inequalities required to satisfy the Cohn-Elkies conditions.
Contour Deformation Proofs: A significant hurdle was proving that certain contours could be deformed even when they touched the real line (where integrands might be undefined). Gauss contributed proofs for these specific contour-deformation identities.
Algorithmic Improvements: The development of specialized metaprogramming tools allowed for the automated simplification of complex mathematical expressions that would otherwise be too cumbersome for manual formalization.

4. Results

Formal Verification: In February 2026, the project achieved the milestone of a complete, "sorry-free" formal verification (meaning every step is verified by the Lean kernel without using the sorry placeholder for unproven steps) that the optimal sphere packing in $\mathbb{R}^8$ is the $E_8$ lattice.
Codebase Expansion: The project expanded the formal codebase from approximately 20,000 lines to a final, refactored version of roughly 60,000 lines of Lean code.
AI Performance: The 'Gauss' model demonstrated the ability to generate complex, multi-hundred-line proofs in a matter of days, though it produced significant "code bloat" (auxiliary definitions) that required human refactoring to meet library standards.

5. Significance

This paper marks a landmark achievement at the intersection of pure mathematics and artificial intelligence.

Mathematical Rigor: It provides a permanent, machine-checked record of one of the most significant results in discrete geometry, ensuring its absolute correctness.
AI in Mathematics: It demonstrates that autoformalization models can tackle high-level, "frontier" mathematics, moving beyond trivial proofs to complex, multi-step theoretical arguments.
Future of Formalization: The project highlights a new workflow: humans design the "blueprint" and high-level infrastructure, while AI handles the "heavy lifting" of generating the granular proof steps. This collaboration suggests a future where the speed of formal verification can keep pace with the speed of mathematical discovery.