2
$\begingroup$

Let $A$ be a non-negative (all entries $\geq 0$) square matrix. Is it always true that

$$ (a_{11}+a_{12}+a_{21}+a_{22})^2\geq 4a_0a_2 $$

where

$a_{ij}$ is the permanent of a matrix obtained by deleting $i$-th row and $j$-th column from $A$.

$a_0=\rm{perm}(A)$

$a_2$ is the permanent of $A$ without the first two rows and columns?

$\endgroup$
3
  • $\begingroup$ Its not true for example for $$\begin{pmatrix} \phantom{-}1 & -1 & \phantom{-}0\\-1 & \phantom{-}1 & \phantom{-}0\\\phantom{-}0 & \phantom{-}0 & \phantom{-}1 \end{pmatrix}.$$ We have $LS=(1-1-1+1)^2=0\ngeq 4\cdot 2 \cdot 1=RS$. $\endgroup$
    – user100927
    Commented Aug 11, 2017 at 10:52
  • $\begingroup$ @user100927 all entries non-negative. Actually I would be happy if it wasn't true... $\endgroup$ Commented Aug 11, 2017 at 10:59
  • $\begingroup$ oh, sorry I didn't read the question carefully enough. $\endgroup$
    – user100927
    Commented Aug 11, 2017 at 11:03

1 Answer 1

5
$\begingroup$

Notice that $$ (a_{11}+a_{22}+a_{12}+a_{21})^2\geq (a_{11}+a_{22})^2+(a_{12}+a_{21})^2\geq 4a_{11}a_{22}+4a_{12}a_{21}. $$ Thus, it suffices to show $$ a_{11}a_{22}+a_{12}a_{21}\geq a_0a_2. $$ Expand each permanent and perform multiplication to obtain the sums of products of $2n-2$ entries on both sides.. We will provide an injection from summands on the right to those on the left; this clearly yields the required inequality.

Visualize the $(i,j)$th entry of $A$ as an edge $(r_i,c_j)$ of a bipartite graph with parts $R=\{r_1,\dots,r_n\}$, $C=\{c_1,\dots,c_n\}$. Each summand on the right corresponds to a graph where the degrees of $r_1,r_2,c_1,c_2$ are $1$, the other degrees are $2$. Its edges are colored in red (corresponding to factors from $a_0$) and blue (from $a_2$).

This graph contains several cycles and two paths, which are either $r_1\to c_1$ and $r_2\to c_2$, or $r_1\to c_2$ and $r_2\to c_1$ (the starting and ending edges of each path are red, so it has an odd number of edges). Repainting the path from $r_1$, we obtain a graph correspoding to a summand from one of the products on the left hand part (in $a_{1i}a_{2j}$, we assume that blue edges correspond to the factors from $a_{2j}$). This is clearly an injection, since the inverse map is provided by the same repainting. So we are done.

$\endgroup$
2
  • $\begingroup$ This is wonderful! Do you know if your technique of using two colored subgraphs to analyze products of subpermanents is explored somewhere in more detail? Or did you invent it as an ad hoc argument? --- Eventually I've proved this inequality using a brute force adaptation of a paper by Nijenhuis "On Permanents and the Zeroes of Rook Polynomials" (1975). This makes me appreciate such a neat argument even more. $\endgroup$ Commented Sep 21, 2017 at 10:36
  • $\begingroup$ Not for subpermanents; but, after realizing that we need such an involution, that was a natural (for me) idea to visualize things... $\endgroup$ Commented Sep 21, 2017 at 15:41

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .