It is well-known that arithmetic and geometric mean are strongly related via logarithmic transformation, i.e. if we take arithmetic mean of logarithmic-transformed values we get the same as if we take logarithm of their geometric mean. With formula: $$\frac{\ln(x_1)+\cdots +\ln(x_n)}{n}=\ln \big( (x_1\cdots x_n)^{(1/n)} \big)$$ for all positive numbers $x_1, \ldots ,x_n$.

It is also known that both geometric and arithmetic mean belong to the same family of power mean (generalized mean) $$M_{\alpha}(x_1, \ldots ,x_n) = \Big( \frac{x_1^\alpha + \cdots + x_n^\alpha}{n} \Big)^{\frac{1}{\alpha}} $$ with $\alpha =1$ representing arithmetic mean and $\alpha=0$ (in limit) representing geometric mean.

Using this notation we can say that $M_1\circ \ln = \ln \circ M_0$ where $\circ$ represents the composition of functions.

My question is generalization of the case above. Let's say that we chose two real numbers $\alpha$ and $\beta$ and we are interested in function $f$ for which $M_\alpha\circ f = f \circ M_\beta$. Does such transformation function exist, is it unique (up to some constraint)?


It seems that $f_{\beta,\alpha}(x)=x^{\beta/\alpha}$ do the trick for $\alpha<>0$.

$$M_\alpha\circ f_{\beta,\alpha} = \Big( \frac{(x_1^{{\beta/\alpha})^\alpha}+ \cdots + (x_n^{\beta/\alpha})^\alpha}{n} \Big)^{\frac{1}{\alpha}} = \Big( \frac{x_1^\beta + \cdots + x_n^\beta}{n} \Big)^{\frac{1}{\alpha}} $$ and $$f_{\beta,\alpha} \circ M_\beta= \Big(\Big( \frac{x_1^\beta+ \cdots + x_n^\beta}{n} \Big)^{\frac{1}{\beta}}\Big)^{\beta/\alpha}= \Big( \frac{x_1^\beta+ \cdots + x_n^\beta}{n} \Big)^{\frac{1}{\alpha}}$$


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.