60-second version
A group is a discrete object: its elements multiply according to rules. A von Neumann algebra is an analytic object built from how that group acts. Connes’s rigidity conjecture proposed that, for a sufficiently rigid group, the analytic shadow might uniquely determine the original group.
OpenAI reports a disproof. The manuscript constructs infinitely many pairwise nonisomorphic property-(T) groups with the same group von Neumann algebra. The shadow can be identical even when the source structures are genuinely different.
Start with a shadow
A photograph can preserve a silhouette while losing the texture of the object that cast it. The analogy is imperfect, but it captures the reconstruction question. A group von Neumann algebra records operators generated by a group representation. It is rich—it remembers far more than a simple outline—but it may still identify two different groups.
The conjecture asked whether strong rigidity rules could prevent that ambiguity. Property (T) is one such rigidity condition: roughly, approximate invariance forces genuine invariance. The surprising result is that this strong behavior does not guarantee unique reconstruction from the algebraic shadow.
Definitions that matter
Two groups are isomorphic when there is a bijection preserving multiplication. Pairwise nonisomorphic groups are therefore structurally different even if they share some invariant. A group von Neumann algebra is generated from the group’s regular representation and completed in an analytic way; it turns algebraic multiplication into operator composition.
Property (T) is a rigidity property used to rule out almost-invariant vectors that are not truly invariant. Connes’s question concerned groups with enough rigidity that one might expect their operator-algebraic output to retain their identity. The counterexample shows that “rigid” and “uniquely recoverable” are different properties.
The research question
If two property-(T) groups produce isomorphic group von Neumann algebras, must the groups themselves be isomorphic? OpenAI’s reported result answers no and strengthens the answer by producing infinitely many distinct groups with the same algebra.
That is a classification result. It does not compare two pictures or a numerical statistic; it shows that an entire analytic object fails to be a complete invariant for a class of rigid groups.
What was known before
Rigidity phenomena have made operator algebras unusually powerful as carriers of group information. Popa’s work on finite-to-one behavior and the study of property-(T) groups supplied reasons to expect that a group algebra might remember much of its origin. The conjecture concentrated that intuition into a sharp uniqueness claim.
The prior landscape was not a blank slate. There were examples where analytic invariants did distinguish groups, and there were results showing strong constraints on possible isomorphisms. The gap was whether those constraints became complete reconstruction for the rigid class.
Why previous approaches stalled
The two structures live at different resolutions. Group isomorphism asks for a point-by-point correspondence preserving multiplication. An algebra isomorphism can reorganize operators globally, and it may hide which operator came from which original group element. Proving nonisomorphism therefore requires an invariant that survives the algebra construction but separates the source groups.
The construction also has to maintain property (T) while varying the group enough to make infinitely many nonisomorphic examples. That combination is why the result is more informative than one accidental pair.
The new result
The manuscript constructs infinitely many pairwise nonisomorphic property-(T) groups whose group von Neumann algebras are all isomorphic. This disproves Connes’s rigidity conjecture and answers a related finite-to-one question of Popa.
The diagram’s “same shadow” is literal only at the level of the stated invariant. It is not saying the groups are secretly identical. It teaches the boundary of reconstruction: an analytic representation can preserve powerful information without preserving uniqueness.
Formal result and proof architecture
The technical argument constructs a family with controlled property-(T) behavior and then proves an isomorphism between the associated group von Neumann algebras. A separate invariant distinguishes the underlying groups, establishing pairwise nonisomorphism. The result therefore has two proof obligations: common analytic output and distinct algebraic input.
The released Lean file formalizes the theorem-level dependencies and construction lemmas that can be represented in the proof assistant. The manuscript is the authoritative source for the operator-algebraic definitions and the classification argument.
What it could lead to
The counterexample gives researchers a test case for deciding which extra hypotheses might restore reconstruction. It may also help separate invariants that are merely strong from invariants that are complete for a chosen class of groups.
A broader lesson for AI-assisted mathematics is that disproving a conjecture often means finding a family with two properties that appear incompatible. The valuable work is not a random counterexample; it is the design of a controlled construction that reveals exactly which intuition failed.
What not to conclude
The result does not say von Neumann algebras are uninformative, nor that rigidity is meaningless. It says this particular analytic shadow cannot uniquely identify every property-(T) group. Other invariants and stronger assumptions may still recover information.