Marcus’s block-permanent inequality
Minc’s list · Conjecture 9
Marcus’s block-permanent inequality
Open
- Posed by
- Marcus
- First posed
- 1965
Statement
Posed by M. Marcus in the 1965 survey of Marcus and Minc; it is Conjecture 9 in Minc’s catalogue.
Conjecture 9. Let \(A\) be an \(mk\times mk\) positive semidefinite Hermitian matrix partitioned into \(k\times k\) blocks \(A_{ij}\), \(1\le i,j\le m\), and let \(G\) be the \(m\times m\) matrix whose \((i,j)\) entry is \(\operatorname{per}A_{ij}\). Then
\[ \operatorname{per}A\ \ge\ \operatorname{per}G. \]
If the diagonal blocks \(A_{ii}\) are positive definite, equality holds if and only if \(A=A_{11}\oplus A_{22}\oplus\cdots\oplus A_{mm}\).
The equality clause needs \(k\ge2\): for \(k=1\) the matrix \(G\) is \(A\) itself. The inequality is the case of Lieb’s Conjecture 42 for the wreath-product subgroup \(S_k\wr S_m\) of \(S_{mk}\) with its trivial character.
Prior progress
- Lieb (1966) proved the case of two blocks, \(m=2\), for every \(k\), with the equality clause; Djoković (1969) gave another proof.
- Pate (1981) proved the weaker bound \(\operatorname{per}A\ge\operatorname{per}G/k!\) for all \(m\) and \(k\).
- Pate (1982) proved the inequality for real matrices with a fixed number of blocks, for all but finitely many block sizes \(k\).
- Rodtes (2024) proved Lieb’s dominance on positive semidefinite matrices whose diagonal products \(\prod_i a_{i\sigma(i)}\) are all nonnegative, which covers, for example, \(vv^*+D\) with \(D\ge0\) diagonal.
Our progress so far
- Three \(2\times2\) blocks at rank at most two. For every \(6\times6\) positive semidefinite Hermitian \(A\) of rank at most \(2\), partitioned into \(2\times2\) blocks, \[ \operatorname{per}A-\operatorname{per}G\ \ge\ \tfrac{14}{9}\,\operatorname{per}A_{11}\operatorname{per}A_{22}\operatorname{per}A_{33}, \] and the constant \(14/9\) is best possible.
- Rank reduction. If the conjecture holds for \(m-1\) blocks of size \(k\), then any counterexample with \(m\) blocks can be compressed to one of rank at most \((m-1)k\). Together with the previous result, a counterexample with three \(2\times2\) blocks would have to have rank \(3\) or \(4\).
- Near diagonal matrices. For all \(m,k\ge2\), the inequality and its equality clause hold in a whole neighbourhood of every positive definite diagonal matrix.
- Rank one plus diagonal. The conjecture holds for \(A=vv^*+D\) with \(D\ge0\) diagonal; when \(D\) is positive definite and \(k\ge2\), equality holds only for block-diagonal \(A\).
References
- M. Marcus and H. Minc, Permanents, Amer. Math. Monthly 72 (1965), 577–591. doi:10.1080/00029890.1965.11970575
- E. H. Lieb, Proofs of some conjectures on permanents, J. Math. Mech. 16 (1966), 127–134.
- D. Ž. Djoković, Simple proofs of a theorem on permanents, Glasgow Math. J. 10 (1969), 52–54.
- T. H. Pate, An extension of an inequality involving symmetric products with an application to permanents, Linear and Multilinear Algebra 10 (1981), 103–105.
- T. H. Pate, Inequalities relating groups of diagonal products in a Gram matrix, Linear and Multilinear Algebra 11 (1982), 1–17. doi:10.1080/03081088208817427
- K. Rodtes, Some remarks on permanental dominance conjecture, Adv. in Appl. Math. 160 (2024), 102758. doi:10.1016/j.aam.2024.102758
- I. M. Wanless, Lieb’s permanental dominance conjecture, preprint (2022). arXiv:2202.01867