A quasi-arithmetic mean is a mean obtained by applying a common continuous strictly increasing function to inputs in a nondegenerate
real interval
, taking their arithmetic
mean, and applying the inverse function
. Its weighted form is
where the weights satisfy and
(Aczél 1948). Equal weights
give the unweighted form.
For positive inputs, the choices
,
, and
give the arithmetic
mean, geometric mean, and harmonic
mean, respectively.
A function is reflexive when
and bisymmetric when evaluating an
array first by rows and then by columns gives the same
result as evaluating first by columns and then by rows. For
, this identity is
Kiss and Okamura (2026) proved that, for every fixed integer , reflexivity, bisymmetry, and strict increase
in each variable characterize weighted quasi-arithmetic means. Continuity need not
be assumed. The positive weights are uniquely determined,
and
is unique up to a positive affine transformation.
Strict increase cannot be replaced by nondecrease, even for symmetric bisymmetric
strict means.
For a family of means of all finite arities, strong decomposability means that selecting any nonempty set of coordinate indices, preserving their order, and replacing the selected coordinates by repeated copies of their own mean leaves the overall mean unchanged. Kiss and Okamura (2026) also proved that a reflexive, strongly decomposable family strictly increasing in each variable has a common continuous generator and equal weights at every arity. The authors credit ChatGPT-6 Astra with the fixed-point uniqueness argument used in their proof.