The Bapat-Sunder eigenvalue conjecture asserts that the permanent of a positive semidefinite matrix
is the largest eigenvalue of the matrix
where
is obtained by deleting row
and column
of
(Bapat and Sunder 1986, Drury 2018). Every row sum of
equals
, so the permanent is always an eigenvalue
with the all-ones eigenvector.
Drury (2018) disproved the conjecture for complex Hermitian matrices using a rank-2 example of order 8 and proposed retaining the assertion
for real positive semidefinite matrices.
Chalmers (2026) disproved the real assertion with a rank-4 integer Gram
matrix of order 32 and an integer vector satisfying
The certificate is verified by exact integer arithmetic. Direct sums with an identity matrix give counterexamples of every order at least 32, and sufficiently small positive multiples of the identity matrix can be added to obtain positive definite matrices. The minimum order of a real counterexample remains unknown.
The same construction also yields real positive semidefinite matrices and
,
with
entrywise nonnegative, for which
, where
is the Hadamard product (Chalmers 2026). Chalmers
(2026) reports that GPT-5.6 Sol and Claude Fable 5 assisted with implementing the
search for the counterexample.