Rank-one optimal solution conjecture for dissipative bilinear control semidefinite programs
Rank-one optimal solution conjecture for dissipative bilinear control semidefinite programs
Let be the system matrix in the semidefinite program associated with the dissipative bilinear control system, and let denote its symmetric part. An optimal solution is a positive semidefinite matrix satisfying the semidefinite-program constraints.
Rank-one optimal solution conjecture. If
then the semidefinite program has an optimal solution of rank .
The preceding proposition proves this conclusion under the additional hypothesis that is -diagonal, with the rank bound matching that value of . The conjecture asserts that dissipativity alone suffices for a rank-one optimal solution; the source reports strong numerical evidence but does not provide a proof.
Progress summary
A posted calculation claims to disprove the conjecture with a strictly dissipative four-dimensional example having no rank-one optimum, but the claim has not been independently verified.
The 2005 manuscript formulates the conjecture that strict dissipativity, , guarantees an optimal solution of rank . It reports strong numerical evidence but gives no proof.
Known results
- For and system matrices, an optimal solution of rank at most exists (2005 manuscript).
- If is -diagonal, an optimal solution has rank at most (2005 manuscript).
- Under , an optimum exists with rank at most (2005 manuscript).
Posted attempt
A complete counterexample is posted: it claims an irreducible matrix with whose semidefinite program has a unique optimal solution of rank , certified by an explicit dual matrix. If correct, this disproves the conjecture even with strictly positive constrained initial data. The calculation has not been independently verified.
Current status (as of August 2026): The special cases and general existence bound are settled, while the unrestricted conjecture has an unverified claimed counterexample rather than a confirmed resolution.
Sources
Sources & referencesView supporting material
Primary source
Dionisis Stefanatos and Navin Khaneja, “Semidefinite Programming and Reachable Sets of Dissipative Bilinear Control Systems”, arXiv:math/0504308 (2005).
Solutions 1
Sign in to submit a solution.
Consider the source semidefinite program
where and, for a real matrix , the symmetric matrices are defined by
with all remaining entries zero.
Take
Every off-diagonal entry is nonzero, so is irreducible. Moreover,
Define
Since and , we have , so has rank two. Explicitly,
Its four constraint values are
Thus is feasible for strictly positive initial data
To certify global optimality, let
and
Direct rational calculation gives
In particular, . Every feasible satisfies
Equality holds at , so the exact optimum is .
Finally, if is any optimal feasible matrix, then
Since , it vanishes, and hence . Therefore
so for a symmetric matrix . The three feasibility equations have coefficient matrix
Thus they uniquely force . Consequently is the unique optimum, and
There is no rank-one optimum. The conjecture is therefore false even for an irreducible strictly dissipative system with strictly positive constrained initial data.