The means ¯r(a) and (a) are of the form
¯ϕ(a)=ϕ−1(∑qϕ(a)) ,
where ϕ(x) is one of the functions: xr and logx and ϕ−1(x) the inverse function. In a nutshell, for ϕ(x)=x,logx and xr , ¯ϕ reduces to , and ¯r respectively.
§
In order that
¯ψ(a)=¯χ(a)
for all a andq , it is necessay and sufficient that
χ=αψ+β,
where α and β are contants and α≠0 .
Since −ϕ is a linear function of ϕ , and −ϕ increases if ϕ decreases, we may always suppose, if we please, that the ϕ involved in ¯ϕ(x) is an increasing function.
§
Suppose that ϕ(x) is continuous in the open interval (0,+∞) , and that
¯ϕ(ka)=k¯ϕ(a)
for all positive a ,q , and k . Then¯ϕ(a) is ¯r(a) .
§
log¯rr(a)=rlog¯r(a) is a convex function of r .
If ϕ(x) is continuous, and there is at least one point of every chord of the curve y=ϕ(x) , besides the end points of the chord, which lies above or on the curve, then every point of every chord lies above or one the curve, so that ϕ(x) is convex.
§
If ψ and χ are continuous and strictly montonic, and χ is increasing, then a necessary and sufficient condition that ¯ψ≤¯χ for all a and
q is that ϕ=χψ−1 should be convex.
§
If ϕ(x,y) is convex and continuous, then
ϕ(∑qx,∑qy)≤∑qϕ(x,y)
§
If F and
G are continuous and strictly monotonic, then a necessary and sufficient condition that (ab) should be comparable with ¯F(a)¯G(b) is that F−1(x)G−1(y) should be a concave or convex function of the two variables x andy ; in the first case(ab)≤¯F(a)¯G(b), in the second the reverse inequality.