What is the Jacobian conjecture and why is its counterexample a significant development in mathematics?

What is the Jacobian conjecture?

The Jacobian conjecture asks: if you take a polynomial map from an (n)-dimensional complex space to itself and the derivative matrix (the Jacobian) has a nonzero constant determinant, must the map have a polynomial inverse that works the other way too? In plainer terms, if the “slope” never goes to zero and never changes, must the function be a nice, reversible one? [5][6]

The conjecture was formally stated by Ott-Heinrich Keller in 1939 for integer-coefficient polynomials, although the two‑variable version can be traced back to the 1880s. [2][34][11]

The counterexample that disproved it

On July 19 2026, Levent Alpöge, a mathematician working at the AI company Anthropic, casually announced on X that he had found a counterexample. [1][20][55] He had used Anthropic’s large language model Claude Fable 5 to search through an enormous space of possible polynomial mappings. [18][23][35]

The counterexample itself is an explicit polynomial map in three dimensions (from ℂ³ to ℂ³) with a constant Jacobian determinant of –2. Crucially, it sends several different input points to the same output point (a three‑point fiber), so the map is not invertible — directly contradicting the conjecture for three or more variables. [17][21][49] The construction is short and easy for other mathematicians to check, though exactly how Alpöge and the AI arrived at it remains unclear. [3][26]

Why it matters so much

• A nearly 90‑year‑old puzzle partially solved. The conjecture stood for 87 years (and its two‑dimensional form even longer). Many top mathematicians, including well‑known names like Segre and Gröbner, published failed proofs — showing how stubborn the problem was. [2][39] A counterexample in the still‑open two‑variable case would need degree at least 125, so the three‑dimensional breakthrough was the first major dent in the armor. [8]
• Now disproved in higher dimensions. The counterexample shows the conjecture is false for every dimension (n \ge 3). The one‑variable case is trivially true, and the two‑variable case — the simplest non‑trivial one — remains wide open. [33][10][68] So the discovery doesn’t close the book entirely, but it reshapes the whole field.
• AI as a mathematical copilot. This is a landmark in AI‑assisted mathematics. The counterexample not only broke the conjecture but also gave insight into why earlier proof strategies failed. [4][24] It suggests that AI can explore huge combinatorial forests to spot objects that humans missed, though what this means for the future roles of human mathematicians is still up in the air. [24][53]
• Domino effect on other conjectures. The falsification immediately pulled down related conjectures — the image conjecture and the vanishing conjecture — in some finite dimension. [36]
• A beautiful object in its own right. Mathematicians have called the three‑dimensional counterexample a “remarkable piece of mathematics” whose full geometric meaning is still being unpacked. [9]

So the discovery is significant not just because it overturned a famous conjecture, but also because it demonstrated a new way of doing mathematics with AI and opened fresh research questions, leaving the ultimate two‑variable case still unsolved.