Just now, there is big news in the mathematical world again.
In September, the successful AI formal verification of the "Poincaré Conjecture" had already shaken the mathematical community.
Just now, a top academic team led by Ayush Khaitan, in collaboration with the NVIDIA Humanfia team, has successfully completed the full Lean formalization of the Thurston Geometrization Conjecture!
Hamilton-Perelman's "century-proof" spans hundreds of pages, taking even top human scholars years to fully comprehend.
Following the capture of the famous Poincaré Conjecture this September, it has now, with AI assistance, been transformed into lines of code.
4.7 million lines of code, completed in just about two weeks.
This is another monument in the history of mathematics, and more importantly, a sign that AI-assisted high-level geometric analysis has entered an "era of rapid advancement."
AI-assisted formalized mathematics has officially entered the "uncharted territory" of humanity's most profound high-level geometric analysis.
Astonishing! Poincaré Has Been Generalized to the Complete Geometrization Conjecture
If someone told you that a team wrote 4.7 million lines of flawless modern mathematical code in two weeks, you would surely think they were crazy.
But this is exactly what just happened.
Recently, Stanford University assistant professor of mathematics Jared Duker Lichtman exclaimed:
Truly an astonishing piece of work.
The proof of Poincaré has been generalized to the complete Geometrization Conjecture, and all of this was done entirely in Lean!

The results were announced by Princeton mathematics and AI scholar Ayush Khaitan. In his tweet, he introduced this incredibly strong cross-disciplinary team.
Look at this hardcore team roster:
Bennett Chow: Disciple of Shing-Tung Yau, professor in the Department of Mathematics at UC San Diego, an absolute authority and leading figure in the field of Ricci flow.
Ziyang Qin and Yuan Liao: The absolute leaders of the project's formalization work.
Ayush Khaitan: Princeton University scholar, the core driving force of the project.
NVIDIA Humanfia team (Jui-Hui Chung & Ligeng Zhu): Representing the cutting-edge forces of the world's strongest AI computing giant, NVIDIA.
Through the collaboration of this super team, a miracle was born.
Ayush revealed that the proof consists of approximately 4.7 million lines of code.
Building upon the previous work of about 2 million lines of code by Chow, Liao, and Qin, an additional 2.7 million lines of code were written in just about two weeks!
Currently, this great project has been fully open-sourced on GitHub.
https://github.com/qinz1yang/differential-geometry
Dimensionality Reduction Strike: From "Poincaré" to the "Full Geometrization Conjecture," What Exactly Have They Proved?
Why is the Stanford mathematician so excited?
First, we need to understand exactly how difficult the Thurston Geometrization Conjecture is.
Everyone has heard of the "Poincaré Conjecture."
It asks: If all closed curves in a three-dimensional space can be contracted to a point, must that space be equivalent to a three-dimensional sphere?
Between 2002 and 2003, the Russian mathematical genius Grigori Perelman emerged, proving the Poincaré Conjecture in three preprint papers. He subsequently declined the million-dollar prize, disappearing into obscurity.

But the Poincaré Conjecture is actually just the "tip of the iceberg" of a grand blueprint. This blueprint is called the "Thurston Geometrization Conjecture."
In other words, if the Poincaré Conjecture teaches you how to recognize a "sphere," the Thurston Geometrization Conjecture is akin to giving the entire three-dimensional universe a "periodic table of elements."
It states that any three-dimensional manifold can ultimately be precisely cut into several pieces. More importantly, each piece must belong to one of the 8 standard geometric structures!

Back then, Perelman precisely proved the Thurston Geometrization Conjecture, thereby "incidentally" proving the Poincaré Conjecture.
The masterstroke he used is called "Ricci Flow with Surgery."
It's like using a "geometric iron" (the Ricci flow equation) to smooth out a distorted space. When encountering dead knots (singularities) that cannot be smoothed, a "surgical" procedure is performed to cut and seal them, then continue smoothing.

This theory is extremely profound. Even the world's top several dozen mathematicians took two to three years to fully verify Perelman's proof.
And today, this group of great scholars and engineers has embedded this pinnacle of human intelligence theory, in its original form, into the computer's mind!
The complete Lean proof of the Thurston Geometrization Conjecture signifies that machines now possess the capability to handle advanced differential geometry, topology, and partial differential equations.
Machines have officially entered the "uncharted territory" of high-level geometry.
Delving into the GitHub Repository: What Exactly Do These 4.7 Million Lines of Code Contain?
If you open this GitHub repository named differential-geometry, you will be hit with a strong sense of "hardcore impact."

Link: https://github.com/qinz1yang/differential-geometry
This is a "Digital Great Wall of Modern Differential Geometry" built with machine language.
In Lean, this interactive theorem prover, the computer doesn't understand what "obviously follows" means; it only recognizes the most fundamental logical axioms. To make the machine understand the Thurston Geometrization Conjecture, the team had to start from scratch, defining the universe's rules in the code.
Judging from the technical details in the repository, the team has robustly filled in a large amount of missing foundational infrastructure in Lean's official math library (Mathlib).
They reconstructed Riemannian geometry from the ground up.
The code defines extremely complex concepts like Manifolds, Tangent Bundles, Tensor Fields, and Riemannian Metrics.
The team used rigorous generic programming to finally make the machine "understand" what a curved space is.
In traditional mathematical papers, Ricci flow is just a partial differential equation

.
But in the GitHub repository, the team needed tens of thousands of lines of code to precisely define "time-dependent metric evolution," "local computation of the Ricci curvature tensor," and the existence boundaries for solutions of incredibly complex partial differential equations.
The most difficult part of Perelman's proof is undoubtedly the "Surgery Theory."
How to implement the "cutting" and "stitching" of an abstract space in code?
The team introduced complex topological connectivity and geometric decomposition algorithms into the codebase, transforming the geometric handling of singularities into logical trees that computers can execute and verify step-by-step.
Two Weeks of Frenzied Output: NVIDIA Joins the Fray, LLMs Take Over the "Code Production Line"
The most thrilling part is this data: two weeks, nearly 2.7 million new lines of code, totaling 4.7 million lines.
In the software engineering world, a 2.7-million-line project would be enough to keep a team of dozens of senior developers working overtime for a whole year.
In Lean, a language with extremely high logical density where each line of code requires lengthy compilation and verification time, writing this volume in two weeks is absolutely impossible by human effort alone!
So how did they do it?
The answer is clear: The NVIDIA Humanfia team brought about a "dimensionality reduction strike" with large language models (LLMs) and automated reasoning.
NVIDIA not only provides the world's strongest computing power; they are also exploring how to have AI directly assist the most cutting-edge fundamental scientific research.
Although the official details of all AI-generated code haven't been disclosed, this "industrialized" proof pipeline can be inferred from the explosive increase in code volume in such a short time.
Top mathematicians like Chow, Liao, and Qin are responsible for drawing the "tactical maps," breaking down the colossal Geometrization Conjecture into thousands of lemmas and sub-goals.
The NVIDIA team utilizes LLMs and automated scripts to generate massive amounts of low-level proof code based on the context provided by humans, filling in tedious algebraic derivations and exhaustive boundary condition enumerations.
Finally, all code is fed into the Lean compiler for ruthless "logical quality control." If it doesn't pass, it's sent back for rewriting until the green light turns on.
This is not only a major event for the mathematical community but also an absolute focal point in the history of AI evolution!
Current large language models still have "hallucinations" in complex, long-chain logical reasoning. The ultimate weapon to solve the logical deficiencies of large models is precisely mathematical formalization.
If AI can skillfully generate Lean code to prove something as pinnacle of human intelligence as the "Thurston Geometrization Conjecture," then the distance to AI "proposing new conjectures and providing proofs" is really just one step away.
Paving a Highway for the Mathematical Universe, a New Era Has Begun
The complete formalization of the Thurston Geometrization Conjecture is far more than just a high-difficulty "code translation." What it leaves for the world is a valuable digital legacy.
Through the differential-geometry project on GitHub, the team has left humanity with a set of differential geometry libraries verified by "absolute truth."
This means that in the future, when mathematicians worldwide want to study multi-dimensional manifolds, the geometric structure of black holes, or even the evolution of spacetime in a computer, they can directly call upon the libraries they have written!
This is a great project of "predecessors plant the trees, later generations enjoy the shade."
From Poincaré writing down that famous conjecture over a hundred years ago, to Thurston outlining the grand blueprint of the three-dimensional universe.
From Perelman calculating in his apartment in St. Petersburg, to today where scholars like Ayush Khaitan, Ziyang Qin, Yuan Liao, Bennett Chow, and others, in collaboration with NVIDIA, use 4.7 million lines of code to make the machine completely "comprehend"...
Humanity's pursuit of truth is like a relay race.
AI is not taking mathematicians' jobs; rather, it's giving them a "starship" to explore the universe.
The pace at which humanity solves difficult problems will burst forth with unprecedented acceleration.
References:
https://x.com/ayushkhaitan343/status/2108654050203840528?s=20
This article is from WeChat Official Account "Xin Zhi Yuan," author: David Aeneas





