Inequalities for the \(q\)-permanent (Q1319998)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Inequalities for the \(q\)-permanent |
scientific article |
Statements
Inequalities for the \(q\)-permanent (English)
0 references
23 October 1994
0 references
Let \(I(\sigma)\) be the length (= number of inversions) of a permutation \(\sigma\). Inequalities for the matrix function \(\text{per}_ q (A) = \sum_{\sigma \in S_ n} q^{I (\sigma)} \prod_{i = 1}^ n\) \(a_{i \sigma (i)}\) are considered for \(-1 \leq q \leq 1\); determinant \((q= -1)\) and permanent \((q=1)\) inequalities follow as special cases. Define two \(n! \times n!\)-matrices as follows: \((\Pi(A))_{\sigma \tau} = d_{\tau \sigma^{-1}} (A) = \prod_{i=1}^ n\) \(a_{ \sigma (i) \tau (i)}\), (i.e. the Schur power matrix and \((\Gamma_ q)_{\sigma \tau} = q^{I(\tau \sigma^{-1})}\) respectively. Compose them by the entrywise Hadamard product to obtain \(\Pi_ q (A) = \Pi (A) \circ \Gamma_ q\). With 1 being the column vector of ones, one obtains the fundamental representation \(\text{per}_ q A = {1 \over n!} \langle \Pi_ q (A)1,1\rangle\). From this first a Cauchy-type inequality is derived: For \(B,C\) \(n \times m\), \(| \text{per}_ q (BC^*) |^ 2 \leq | \text{per}_ q (CC^*) |\) \(| \text{per}_ q (BB^*) |\). Using \(\Gamma_ q \geq 0\) [cf. \textit{M. Bozejko} and \textit{R. Speicher}, Commun. Math. Phys. 137, No. 3, 519-531 (1991; Zbl 0722.60033)] and hereditarity of positivity under Hadamard products (Schur), \(A \geq 0\) implies \(\text{per}_ q (A) \geq 0\). The authors derive in fact better lower bounds -- a Gram type inequality: If \(A=(B_ 1, \dots, B_ r) (B_ 1,\dots, B_ r)^* \geq 0\) \((B_ i\) \(n \times 1\) (columns)) then \(\text{per}_ q (A) \geq \sum_{\mathbf k} {1 \over k_ 1!! \cdots k_ r!!} | \text{per}_ q (B{\mathbf k}) |^ 2\). Here the \(B({\mathbf k})\) are the \(n \times n\) matrices formed from \(B\) containing column \(i\) \(k_ i\) times in natural order, \(k!!= (1 + q) (1 + q + q^ 2) \cdots (1+\cdots + q^{k-1})\); and the \({\mathbf k} = (k_ 1, \dots k_ r)\) range over all such \(r\)-tuples summing to \(n\). The authors close with three open conjectures having some credibility.
0 references
determinant inequality
0 references
permanent inequality
0 references
matrix function
0 references
Hadamard product
0 references
Cauchy-type inequality
0 references
Gram type inequality
0 references