Is a strictly Increasing, midpoint convex, symmetric probability function, iff it is midpoint concave/linear?

206 Views Asked by At

Is A, Strictly increasing , Symmetric Probability function with symmetric inverse probability,Midpoint Convex, iff it is midpoint concave?

Is this derivation (A)-rough proof (see below) valid?

And thus does F satisfy Jensen's equation if F is midpoint convex?. And what are the names of the properties in $(5.1)$ and $(5.2) $ ,a bi-conditional and inverse version of symmetry in nd $(6)$ the. $(5.1)$ and $(5.2) $ a form of dd-ness; leftright reflection symmetry

: 1.$$F:[0,1]\to[0,1]$$ 2. Strictly Monotone Increasing and in-jective. 3.$$F(0)=0, \,F(1)=1,\, F(0.5)=0.5$$

4.$$F\text{ is}\text{ Midpoint Convex},\forall x,y\in\text{dom}(F)=[0,1] F(\frac{x+y}{2})\leq(\frac{F(x)}{2}+\frac{F(y)}{2})$$ *5.Satisfies, the the bi-conditional symmetry properties 5.1.(1)$$\forall(x,(1-x))\in\text{Dom}(F)=[0,1]:F(x)+F(1-x)=1$$ : 5.2.(2)$$\forall(p,(1-p))\in\text{Im}(F)\subseteq [0,1]:F^{-1}(p)+F^{-1}(1-p)=1$$. $$(x,x_2)\in\text{dom}(F);\, [x_1+x_2=1][\leftrightarrow[F(x_1) + F(x_2)=F(1)=1]$$ $$(p_1,p_2)\in \text{codom}(F);\,[p_1+p_2=1]\leftrightarrow [F^{-1}(p_1) + F^{-1}(p_2)=F^{-1}(1)=1]$$ 6. $$∀(x_1,x_2)∈Dom(F)=[0,1]:[x_1+x_2]>1↔[F(x_1)+F(x_2)]>1$$ $$∀(x_1,x_2)∈Dom(F)=[0,1]:[x_1+x_2]<1↔[F(x_1)+F(x_2)]<1 $$∀ $$\forall(p_1,p_2)∈Im(F);[p_1+p_2]>1↔[F^{−1}(p_1)+F^{−1}(p_2)]>1$$ $$\forall(p_1,p_2)∈Im(F);[p_1+p_2]>1↔[F^{−1}(p_1)+F^{−1}(p_2)]

Does Midpoint Convexity(5), given symmetry conditions (5.1,5.2), andF injective (2), and real valued, (1), with boundary constraints (3) entail that F is Midpoint concavity and thus a Jensen function?

(A)Rough Proof $F:[0,1] \to [0,1]$; (A1)$F$ is in-jective and strictly monotonic increasing (A2) with $F(0)=0$, $F(0.5)=0.5$ $F(1)=1$

Premises $(1)$ Midpoint convexity $\forall(x,y)\in \text{dom}(F)=[0,1]:F(\frac{x+y}{2})\leq\frac{F(x)}{2}+\frac{F(y)}{2}$ $(1a)$ Midpoint convexity $F$ at all points and thus midpoint convex @$\frac{(1-y)+(1-x)}{2})$ this follows from $(1)$ $$ F(\frac{(1-y)+(1-x)}{2})\leq\frac{F(1-x)}{2}+\frac{F(1-y)}{2}$$ $(2)$ Symmetry:$\forall(x_1,x2)\in dom(F)=[0,1]: $(x_1+x_2)=1 \leftrightarrow F(x_1)+F(x_2)=1$

$(2a)$ Inverse symmetry: $\forall(p_1,p2) \in Im(F)=[0,1]: $(p_1+p_2)=1 \leftrightarrow F^{-1}(p_1)+F^{-1}(p_2)=1$

Notice that there is always one pair of domain values in the system for each mid point domain point, $m\in \text{dom}(F)$ , $m=\frac{x}{2} +\frac{y}{2}, \,\text{between}\, \frac{x}{2}\,{\&} \frac{y}{2},\text{is s.t}\, (1-m)\in\text{Dom}(F) \,\text{lies halfway between}\,\, \frac{1-x}{2}\in\text{dom}(F)\, {\&} \frac{1-y}{2}\in\text{dom}(F)$.

$$1-m=1-\frac{x+y}{2}= 1-(\frac{x}{2} + \frac{y}{2})$$

(1)$$=(1-\frac{x}{2}-\frac{y}{2})=\frac{1-x}{2}+\frac{1-y}{2}$$

(2) $$F(1-m)= F(1-\frac{x+y}{2})=F(\frac{2-(x+y)}{2})$$ $$=F(\frac{1 +1 -x-x}{2})=F(\frac{(1-x)+(1-y)}{2})$$

(2.1) From and (2) and (1) $$\frac{F(1-m)}{2} \leq \frac{F(1-x)}{2}+\frac{F(1-y)}{2}$$

By Midpoint Convexity and (2.1)

(3) $$F(\frac{(1-y)+(1-x)}{2})\leq\frac{F(1-x)}{2}+\frac{F(1-y)}{2}$$

Thus(2.1)

(3) $$F(1 - \frac{x+y}{2})\,\leq \, \frac{F(1-x)}{2}+\frac{F(1-y)}{2}$$

$$\text{And Midpoint convexity}$$ $$F(\frac{x+y}{2}) \leq \frac{F(x)}{2}+\frac{F(y)}{2}$$

(4)By the symmetry equations (1) an (2)**

(4.1)$$F(x)+F(1-x)=1$$ and (4.2)$$F^{-1}(p)+F^{-1}(1-p)=1$$

(4.3), $$F(t)+F(1-t)=1\, \Rightarrow\,F(1-t)=1-F(t)$$

From (4)$$\text{for any arbitrary t}\,\in \text{dom}(F)=[0,1];\,\text{there exists, one, and and only one}$$ $$ t_2=(1-t_1)\,t_2\in\, \text{dom(F)}=[0,1]\, [t_1+t_2=1]\quad\text{s.t;}$$

$$\forall (t_1, t_2)\in \text{dom(F)};\,[t_1+t_2=1]\, \Leftrightarrow\,[F(t_1)+F(t_2)=1]$$

(5)Thus letting $t=\frac{x+y}{2}$ in (4.3)

(5.1)$$F(t) +F(1-t)=1\rightarrow [(\frac{x+y}{2})+(1-\frac{x+y}{2})=1 \rightarrow F(\frac{x+y}{2})+ F(1-\frac{x+y}{2})=1$$

$$F(1-t)=F(1-\frac{x+y}{2})\,\text{which by (5) gives}$$

(5.2)$$F(1-t)=1-F(t)=1-F(\frac{x+y}{2})=F(1-\frac{x+y}{2})$$

So by (5.2)and by midpoint convexity** and (5.3)

(5.4) $$1-F(\frac{x+y}{2})= F(1-\frac{x+y}{2})= F(\frac{1-x}{2}+\frac{1-y}{2}) \leq F(\frac{1-x}{2}+F(\frac{1-y}{2}) $$

we derive (6):

(6) $$1-F(\frac{x+y}{2})\leq \frac{F(1-x)}{2}+\frac{F(1-y)}{2}$$

(7)By Symmetry again; however

$$1-F(x)=F(1-x)\,{\&}\,1-F(y)=F(1-y)$$

Thus from (6) and (7) we derive (8)

(8) $$1-F(\frac{x+y}{2})\leq \frac{F(1-x)}{2}+\frac{F(1-y)}{2}=\frac{1-F(x)}{2}+\frac{1-F(y)}{2}=1-[\frac{F(x)}{2}+\frac{F(y)}{2}]$$

and this entails (9)

(9) $$1-F(\frac{x+y}{2})\leq 1-[\frac{F(x)}{2}+\frac{F(y)}{2}]$$

Thus subtracting one from both sides we get (10)

(10)$$$-F(\frac{x+y}{2})\leq -[\frac{F(x)}{2}+\frac{F(y)}{2}]$$

but (10) implies (11)

(11)$$F(\frac{x+y}{2})\geq\frac{F(x)}{2}+\frac{F(y)}{2}$$

$$\text{but this is midpoint concavity}$$

$$\text{however, given F is midpoint convex}$$

$$\text{but by midpoint convexity}$$

(12)$$F(\frac{x+y}{2})\leq\frac{F(x)}{2}+\frac{F(y)}{2}$$

$$\text{then, at}\quad\frac{x+y}{2}\in\text{dom}(F)$$ $$\text{ given (11), and (12)}\,\text{F is midpoint convex}\,\text{and F is midpoint concave}$$

(13) $$\{[F(\frac{x+y}{2})\leq(\frac{F(x)}{2}+\frac{F(y)}{2})] \,\land\,[ F(\frac{x+y}{2})\geq(\frac{F(x)}{2}+\frac{F(y)}{2})] \}$$ $$(13)\Leftrightarrow F(\frac{x+y}{2})=(\frac{F(x)}{2}+\frac{F(y)}{2})$$

And from (13) see that, (C) Jensen's equation holds.

A similar argument is used to derive jensen's equation @ $1-m$ using $(1)$ $\in\text{Dom}(F)$.

(1)$$\text{ Use midpoint convexity at}\quad m=\frac{x+y}{2}\quad F(m)=F(\frac{x+y}{2})$$

$$F(\frac{x+y}{2})\leq(\frac{F(x)}{2}+\frac{F(y)}{2})\rightarrow 1-F(\frac{x+y}{2}) \geq 1-[\frac{F(x)}{2}+\frac{F(y)}{2}]$$

which by symmetry $F(x)+F(1-x)=1$ induces midpoint concavity @ $1-m=\frac{(1-x)+(1-y)}{2}\in \text{dom}(F)$, which by assumption is midpoint convex as well

as $F$ is midpoint convex at all points, and thus jensens equation

$$ \frac{(1-x)+(1-y)}{2}=1- \frac{x+y}{2} \rightarrow \frac{x+y}{2}+\frac{(1-x)+(1-y)}{2}=1 \rightarrow F(\frac{x+y}{2})+F(\frac{(1-x)+(1-y)}{2})=1 \rightarrow 1-F(\frac{x+y}{2}) = F(1- \frac{x+y}{2})=F( \frac{(1-x)+(1-y)}{2})$$

(2.1)$$F(\frac{(1-x)+(1-y)}{2})=1-F(\frac{x+y}{2}) \geq 1-[\frac {F(x)}{2}+\frac{F(y)}{2}]= \frac{1-F(x)}{2} + \frac{1-F(y)}{2}=\frac{F(1-x)}{2}+\frac{F(1-x)}{2}$$ using symmetry $1-F(y)=F(1-y)$ etc

but,$F$ is midpoint convex (everywhere;by assumption) therefore is midpoint convex at $1-m$ (as used in the previous derivation; ie by(4)) .

Which results in Jensen's equality (equation) at $1-m$ as well(3):

(3) $$F(\frac{(1-x)+(1-y)}{2})=F(\frac{1-x}{2})+F(\frac{1-x}{2})$$

See my post question for the additional properties that F has

satisfies.

References

1.Ger, Roman, Almost additive functions on semigroups and a functional equation, Publ. Math. 26, 219-228 (1979). ZBL0444.39004.

.

5.[AczD] J. Acz´el, J. Dhombres, Functional equations in several variables.With applications to mathematics, information theory and to the natural and social sciences. Encyclopedia of Math. and its App., 31, CUP, 1989

2

There are 2 best solutions below

0
On

One can derive continuity in multiple just by strict monotone continuity and the fact that midpoint so that its continuous at all points except @1 without symmetry due to the fact @0 one uses midpoint convexity and the fact that $F(0)=0$ to get that continuity @1: $$F(\frac{1}{2} x) \leq \frac{1}{2} F(x) \rightarrow F(\frac{1}{2^{n}}x) \leq \frac{1}{2^{n}} F(x) $$

by setting x=1 F(1)=1

$$ 0 \leq F(\frac{1}{2^{n}}) \leq \frac{1}{2^{n}} $$,

$$0 \leq F(\frac{1}{2^{n}} \times x) \leq [\frac{1}{2^{n}}\times F(x)] \rightarrow [0 \leq F(\frac{1}{2^{n}}\times 1)=F(\frac{1}{2^{n}})\leq [\frac{1}{2^{n}}\times F(1)] = \frac{1}{2^{n}} \times 1= \frac{1}{2^{n}}] $$ so that $n\to \infty$, or as $x\to 0$ from above, $F(x)\to 0$

And due to strict monotonicity (I presume one does not need to know much more about the limit from below at @0) and setting a particular fixed pt $F(1)=1, x=1$

, vanishes

Note however that $F(\frac{1}{2^{n}}) =\frac{1}{2^{n}})$ however due to symmetry

In any case to derive continuity @1 one uses the symmetry and that $$(A)[F(1-\frac{1}{2^{n}}x)+F(\frac{1}{2^{n}}x)=1 ]\rightarrow[F(1-\frac{1}{2^{n}}x)= 1- F(\frac{1}{2^{n}}x) $$

and that $$(B)F(\frac{1}{2^{n}}x) \leq \frac{1}{2^{n}}F(x) \rightarrow [1- F(\frac{1}{2^{n}}x)] \geq [1-\frac{1}{2^{n}}F(x)]$$

Using $(A)$ and $(B)$

$$ F(1-\frac{1}{2^{n}} x)=1-F(\frac{1}{2^{n}}x) \geq [1-\frac{1}{2^{n}}F(x)]$$ $$\leftrightarrow$$

Which reduces to: using $F(1)=1$ at the boundary and monotonicity

$$1 \geq F(1-\frac{1}{2^{n}} x) \geq 1-\frac{1}{2^{n}}F(x)$$

using $F(1)=1$;$x=1$

$$1 \geq F(1-\frac{1}{2^{n}}\times1)=1-F(\frac{1}{2^{n}}) \geq 1-\frac{1}{2^{n}} \times 1=1-\frac{1}{2^{n}}$$ so that $n\to \infty$, or as $x\to 1$ from below, $F(x)\to 1$

Although this is an equality as well. Technically In fact it appears that I can get all of the rationals, $F(r)=r$ (the dyadics and most non-dyadics) out just using midpoint convexity at $0$ and at $1$ as a result of symmetry and use an argument that agrees with F(x) on a dense set. This would limit the use of midpoint convexity, and make greater use of my symmetry equations I believe. It might require more use of my inverse symmetry.

$0$although it would be perhaps continuous at $1$ and $0$ and and strictly monotonic

0
On

I am not sure if just midpoint convexity at zero would entail anything; due to using the fact that a mid-convex strictly increasing function has a mid-concave inverse function, and use my inverse symmetrical relation to jensens equation on both;

I don't think it generalizes mid-star convexity at $o$, $F(x)\leq F(\frac{1}{2})$ would generalize to star convexity at $0$ or even dyadic star convexity at $0$ which would ensure that $F(x)=x$ for dyadic rationals before continuity is applied.

Unless continuity at $0$ and @$1$ given my symmetry equations, and midstar convexity at $0$ and strict monotonic increasing-ness, entails continuity, given the symmetry and three fixed points, continuity at $0$ and $1$ and strict mono-tonicity, and this entails star convexity, or dyadic star convexity? I do not think that it would