I was wondering under what general conditions and how it's possible to transform a weighted arithmetic mean to a weighted geometric mean.

Assume there are positive weights $a_1,\dots,a_n$ that sum up to $1$ (i.e. $a_1+...+a_n=1$ and $a_i\geq0$ $\forall i\in\{1,\dots,n\}$) such that: $$y=\sum_{1\leq i\leq n}a_i\cdot x_i$$ Under what conditions and how can positive weights $g_1,\dots,g_n$ that also sum up to $1$ be calculated such that: $$y=\prod_{1\leq i\leq n}x_i^{g_i}$$

Addendum: The above formulation of the question has to be specified a little bit further: The famous theorem on the (in)equality of weighted arithmetic and geometric means states: Given positive weights $w_1,\dots,w_n$ that sum up to one it holds that: For all positive $x_1,\dots,x_n$:

$$\sum\limits_{i=1}^nw_i\cdot x_i\geq\prod\limits_{i=1}^nx_i^{w_i}$$

... with equality iff $x_1=\dots=x_n$. Note that the weights are fixed here.

Now, I am interested in a result on varying the weights, especially I am interested whether a statement of the following form holds:

$\forall x_1\in(0,1],\dots,x_n\in(0,1]$
$\forall a_1\in[0,1],\dots,a_n\in[0,1]$ ($a_1+\dots+a_n=1$)
$\exists g_1\in[0,1],\dots,g_n\in[0,1]$ ($g_1+\dots+g_n=1$)
such that the equivalence holds -- for short:

$$\forall x\forall a\exists g\sum\limits_{i=1}^na_i\cdot x_i=\prod\limits_{i=1}^nx_i^{g_i}~~\text{where for the $x_i$s,$a_i$s,$g_i$s hold the restrictions as stated above}$$

And if so, is there a way to calculate solutions for the $g_i$s with Mathlab, Maple, Wolfram etc.? If not, which restrictions on the $a_i$s and/or the $x_i$s are necessary?

I'd be very happy for any hints.

Thanks a lot, Christian


Take the logarithm of the second identity. You have

$$\log y=\sum_{i=1}^n g_i\log x_i=\log\sum_{i=1}^n a_ix_i.$$

With the condition

$$\sum_{i=1}^n g_i=1$$

this is a linear system of $2$ equations in $n$ unknowns, which is $n-2$ times indeterminate.

  • $\begingroup$ Dear Yves, thanks a lot for your reply. I have specified my request a little bit -- do you have also any hints regarding calculating the weights with software like WolframAlpha etc.? $\endgroup$
    – Christian
    Sep 11 '17 at 10:23
  • $\begingroup$ @Christian: set $n-2$ parameters $g_i$ arbitrarily and solve for the two remaining ones. $\endgroup$
    – user65203
    Sep 11 '17 at 10:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.