The border rank
of a tensor
over the complex numbers
is the least integer
such that
is a limit of tensors of tensor
decomposition rank at most
. Equivalently,
lies in the closure of the set of tensors
expressible as a sum of at most
simple tensors. Border rank can
be smaller than tensor decomposition rank
(Lovitz 2026); neither is the index-counting meaning of tensor
rank.
Lovitz (2026) reported an explicit family of tensors with
integer entries and
The entries are computable in polynomial time in . Lovitz (2026) also showed that linear
tensor flattening methods cannot establish a
border-rank lower bound exceeding
, with that ceiling attained when
is odd.
Lovitz (2026) states that GPT-6 Astra obtained proofs that were subsequently verified and simplified.