I would like a better understanding of the set of pairs $(f_1,f_2)$ of functions $[0,1] \times [0,1] \to [0,1]$ which satisfy the following conditions:

  1. For all $y \in [0,1]$, $f_1(x,y) \geq f_1(x',y) \Leftrightarrow x>x'$. For all $x \in [0,1]$, $f_2(x,y) \geq f_2(x,y') \Leftrightarrow y>y'$.

  2. $f_1(x,y)+f_2(x,y) = 1$, for all $(x,y) \in [0,1] \times [0,1]$

  3. $f_1(x,y)=f_2(y,x)$ for all $(x,y) \in [0,1] \times [0,1]$

  4. Due to monotonicity (Condition 1), the functions $f_1$ and $f_2$ are integrable. Now define $$p_1(x,0) = xf_1(x,0) - \int_0^x f_1(t,0) dt,$$ $$p_2(0,y) = yf_2(0,y) - \int_0^y f_2(0,t) dt,$$ $$p_1(x,y) = p_1(x,0) + xf_1(x,y) - \int_0^x f_1(t,y) dt,$$ $$p_2(x,y) = p_2(0,y) + yf_2(x,y) - \int_0^y f_2(x,t) dt.$$

Then for all $(x,y) \in [0,1] \times [0,1]$, $p_1(x,y)+p_2(x,y)=0$.

Example: Let $\alpha \geq 0$. For $x \geq y$, let $$f_1(x,y) = (1/2)+(1/2)(x^\alpha-y^\alpha)$$ $$f_2(x,y) = (1/2)-(1/2)(x^\alpha-y^\alpha)$$ For $x < y$, let $$f_1(x,y) = (1/2)-(1/2)(y^\alpha-x^\alpha)$$ $$f_2(x,y) = (1/2)+(1/2)(y^\alpha-x^\alpha)$$ Then $(f_1,f_2)$ satisfy the conditions.

  • 2
    $\begingroup$ What is the motivation for this question? $\endgroup$ – Noah Schweber Jun 23 '14 at 17:16
  • $\begingroup$ It is related to "mechanism design", a sub-field of game theory. The set of all solutions provides the domain of all mechanisms which satisfy certain criteria. A "best" needs to be chosen from this "domain". $\endgroup$ – user54297 Jun 23 '14 at 17:38
  • $\begingroup$ @S. Carnahan: Thanks for editing the text. $\endgroup$ – user54297 Jun 24 '14 at 11:56

This is not as clear as it might be. I give at the end a more general set of solutions to what I think are your intended conditions. I wonder if these are all the solutions, however I am not sure that I fully understand what you want.

Three issues (in decreasing order of importance)

  1. I am guessing that (maybe) you wanted to have $$p_1(x,y) = p_1(x,0) + yf_1(x,y) - \int_0^y f_1(x,u) du$$ This is valid when you put $y=0$ whereas what you currently have written would give the generally invalid $$p_1(x,0) = p_1(x,0) + xf_1(x,0) - \int_0^x f_1(t,0) dt.$$

  2. Your example at the end has cases but there is no difference between them since $$\frac{1}{2}+\frac{1}{2}(x^\alpha-y^\alpha)=\frac{1}{2}-\frac{1}{2}(y^\alpha-x^\alpha)$$ (similarly for $f_2$). Nothing wrong with that (if that was your intention), but confusing.

  3. In your first condition you need to use $\ge$ or $\gt$ in both places or (else change $ \Leftrightarrow$ to just $\Leftarrow$). As I say, least important. But do you want $f_1(x,y)$ strictly increasing in $x$ or merely non-decreasing?

Here are a few suggestions:

Since your conditions give $f_1(x,x)=\frac{1}{2}$ and $f_2(x,y)=1-f_1(x,y)$, define $f_1(x,y)=\frac{1}{2}+g(x,y)$ and $f_2(x,y)=\frac{1}{2}-g(x,y).$ We should note that $g$ is a function $[0,1] \times [0,1] \to [-\frac{1}{2},\frac{1}{2}]$.

Then condition 2 is automatic, condition 3 says $g(y,x)=-g(x,y)$ and the $f_1$ part of condition 1 (corrected) says that for fixed $v$ and $u \gt u'$ we have $$g(u,v) \gt g(u',v).$$ Then, using condition 3, we then have $$-g(v,u) \gt -g(v,u')$$ meaning that the $f_2$ part is automatic and also that $$g(v,u) \lt g(v,u').$$

Your could rewrite your definition $$p_1(x,0) = xf_1(x,0) - \int_0^x f_1(t,0) dt=\int_0^x \big(f_1(x,0)-f_1(t,0)\big) dt$$ so $$p_1(x,0)=\int_0^x \big(g(x,0)-g(t,0)\big) dt.$$ Then with my suggest revision one would have in general

$$p_1(x,y)=\int_0^x \big(g(x,0)-g(t,0)\big) dt+\int_0^y \big(g(x,y)-g(x,u)\big) du$$

$$p_2(x,y)=\int_0^y -\big(g(0,y)-g(0,u)\big) du+\int_0^x -\big(g(x,y)-g(t,y)\big) dt$$ $$=\int_0^y \big(g(y,0)-g(u,0)\big) du+\int_0^x \big(g(y,x)-g(y,t)\big) dt$$

I'm not so sure about this last part but, if it is correct, then I suppose generalizing your example to $$g(x,y)=q(x)-q(y)$$ where $q:[0,1] \to [0,\frac{1}{2}]$ is increasing would meet your conditions as

$$p_1(x,y)=\int_0^x \big(q(x)-q(t)\big) dt+\int_0^y \big(q(u)-q(y)\big) du=xq(x)-yq(y)-\int_x^yq(v)dv$$

$$p_2(x,y)=\int_0^y \big(q(y)-q(u)\big) du+\int_0^x \big(q(t)-q(x)\big) dt=yq(y)-xq(x)+\int_x^yq(v)dv.$$

  • $\begingroup$ Thanks a lot for your feed back. Yes, there was a mistake. But, elsewhere: it should have been p_1(0,y) instead of p_1(x,0) $\endgroup$ – user54297 Jun 25 '14 at 21:44
  • $\begingroup$ Thanks a lot for your feed back. (1) Yes, there was a mistake. But, elsewhere. It should have been p_1(0,y) instead of p_1(x,0); (2) Sorry for the confusion; (3) I meant it to be non-decreasing, so the "iff" should be valid? Your suggested procedure is neat, and thanks for the suggestion. I think it will help me proceed. I am pretty sure that f_1 has to be concave in x. But can it have kinks, and if so then how many? [I have problems typing out math characters here and I wish I could explain matters better]. Thanks once again. $\endgroup$ – user54297 Jun 25 '14 at 21:57
  • $\begingroup$ Oh, and p_1(0,y) is simply some function which is constant in x. $\endgroup$ – user54297 Jun 25 '14 at 22:35
  • $\begingroup$ I am having problems writing math symbols. For example the concavity issue makes no sense above. Can I post pdf. s? $\endgroup$ – user54297 Jun 25 '14 at 23:49

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.