Difference between majorization and weak majorization.

70 Views Asked by At

I am trying to understand the concept of weak majorization and majorization. Following the definitions in the book of Marshall et. al I have formulated the following reasoning. I suppose I am wrong somewhere, or something in the definitions is not clear to me yet. Can you please help me out here?

I know that

(1) if $x\geq y$, for $x,y \in \mathbb{R}^n_+$, one has $y\prec^w x$ and $y\prec_w x$, that is $y$ is both weakly supermajorized and weakly submajorized by $x$;

(2) $y\prec^w x$ and $y\prec_w x$ if and only if $y \prec x$ (i.e. $y$ is majorized by $x$)

(3) if $f:\mathbb{R}^n_+\rightarrow \mathbb{R}_+$ is Schur concave, then $y \prec x$ implies $f(x)\leq f(y)$.

Now, suppose $f:\mathbb{R}^n_+\rightarrow \mathbb{R}_+$ is continuous, strictly increasing and Schur concave. Let $x,y \in \mathbb{R}^n_+$ be such that $x>y$ (that is $x_i\geq y_i$ for all $i=1,...,n$ with at least one strict inequality), then we must have $f(x)>f(y)$.

However, by (1), $y\prec^w x$ and $y\prec_w x$ and, by (2) $y\prec x$. Since $f$ is Schur concave, $y\prec x$ implies $f(x)\leq f(y)$, which contradicts the previous inequality.

I suspect the problem comes from (2) because $\prec$ is defined for vectors such that $\sum_i x_i=\sum_i y_i$, while the weak majorization does not require this restriction. (1) and (2) are respectively euqations (15a) and (14) page 13 of the book.

1

There are 1 best solutions below

0
On BEST ANSWER

Actually the mistake lies within the first point you mentioned, and it is a typo in the book of Marshall et al. Their equation (15a) on page 13 should read:

$x\leq y$ (that is, $x_i\leq y_i$, $i=1,\ldots,n$) $\Rightarrow$ $x\prec_wy$ and ${\bf y\prec^wx }$

One readily verifies this by plugging $x_i\leq y_i$ (hence $x_{[i]}\leq y_{[i]}$ and $x_{(i)}\leq y_{(i)}$) into the definition of $\prec_w$ and $\prec^w$, respectively.