Jacobian Conjecture

The Jacobian Conjecture is a question from mathematics that remains unsolved to this day. It concerns whether a certain kind of computational rule can always be uniquely reversed. Such a rule takes in several numbers and returns just as many numbers, with only additions and multiplications allowed. For such rules there is a test that indicates whether the rule only ever slightly stretches or compresses at every point, but never collapses anywhere. The conjecture states: if this test yields the same result everywhere, then the rule can be uniquely undone globally. The German mathematician Ott-Heinrich Keller formulated it in 1939; it is named after a computational tool devised by Carl Gustav Jacob Jacobi.

A trap many have fallen into

The Jacobian Conjecture is famous for looking much easier than it actually is. One can explain it to a second-semester student in five minutes. That is precisely what makes it dangerous. Over the decades, dozens of purported proofs have been published, and all of them contained errors.

The mathematician Shreeram Abhyankar therefore called it a conjecture that reveals whether someone knows their limits. Fields Medalist Stephen Smale included it in his 1998 list of the most important open problems for the 21st century. There it stands as Problem Number 16, alongside questions on which entire fields of research hinge.

It is also important because it connects a surprising number of areas. It is linked to questions from geometry, to the theory of computational operations, and even to the question of how to have mathematics automatically checked by computer programs. Several other open problems have turned out to be equivalent to the Jacobian Conjecture. Whoever solves one of them automatically solves all of them.

The test with the determinant

The test mentioned above is called the Jacobian determinant. It measures how strongly a computational rule distorts the neighborhood of a point. If the value at a point is zero, something collapses there, and the rule cannot be reversed nearby. If the value is a fixed nonzero number everywhere, then it is at least locally, i.e. in small regions, always reversible.

The leap from local to global is the crux of the problem. Imagine a map that renders every small section of the Earth cleanly and without distortion. Nevertheless, it could still place two far-apart locations onto the same point. That is precisely what the conjecture rules out, at least for rules built from additions and multiplications. It has not been proven, however.

Some partial results are known. For rules with only a single input number, the statement is provably true. The general case could also be reduced to a manageable form in which only third powers occur. A typical mistake, incidentally, is to test the conjecture with real numbers: there it is simply false, since it concerns complex numbers, i.e. an extended number system.

Why the term appears in AI news

You don’t encounter the Jacobian Conjecture in everyday life. Its namesake, the Jacobian determinant, on the other hand, constantly: it is embedded in the training of every neural network, i.e. those programs that learn from example data. There it describes how a small change to the settings affects the outcome. Computer graphics and robotics also compute with it.

The conjecture itself today appears mainly in reports about proof assistants. These are programs that check mathematical proofs step by step for gaps. As AI systems increasingly propose proofs on their own, old open problems are being used as stress tests. The Jacobian Conjecture is particularly well suited for this because it can be stated briefly yet is extremely hard.

No imminent breakthrough is expected. It is more realistic that such tools will expose flawed proof attempts faster than human reviewers can. For a conjecture with this track record of errors, that is already a noticeable gain.