I am designing an autonomous agent whose behaviour depends on a function of two variables, say $f(x,y)$. The variables $x$, $y$ and $f$ are all in $[0,1]$, and I know that:

$$f(0,y)=f(x,0)=0$$ $$f(1,y)=f(x,1)=1$$

(The values $f(0,1)$ and $f(1,0)$ are not defined).

I am looking for $f$ that is smooth in $(0,1)\times(0,1)$.

After some playing around, I found the following function:


If am not mistaken, this function is continuous in $(0,1)\times(0,1)$, but it is not smooth because of the absolute-value function.

My questions are:

  • Are there smooth functions that satisfy the above conditions?
  • Is there a more systematic way to find such functions, other than just guessing and checking?

To answer your general question, there is a way to find such functions without just guessing and checking...sort of. One way is to say "I'd like to find a function that satisfies some differential equation, like Laplace's equation, on the interior of this square, but takes on these values on the boundary." That's a classic problem in heat-transfer, electrostatics, and many other fields, and is much studied.

One of the great ways to solve such a problem is to find a map from your domain to some other domain, solve the problem there, and then map back. If your map happens to be complex-analytic (the complex variables analog of "differentiable"), then it will carry solutions of Laplace's equation from one domain back to the other.

So if we can find a map that takes two sides of your square to the $v = 0$ line in the $u+ i v$ plane, and the other two side to the $v = 1$ line, then transforming the function "$v$" back to your square will give a solution.

The only problem is that building the analytic transform from your problem to an easier problem is not always easy. On the good side, many smart people have catalogued a bunch of approaches -- search for "conformal mapping" to get an idea of how they're done.

I don't have my complex variables book handy here, so I can't find a good map doing what I suggested for the square, but perhaps I can look it up tomorrow. If you look in Churchill's Complex Analysis, solving problems like this is in one of the last sections of the book.

There's one more approach: Letting $C$ denote the contour that's the boundary of the square, you can define $$ f(z) = \int_C \frac{h(u)}{z-u} du $$ where $h$ is your function (the $+1/0$ function you described). (Warning: I might be off by a minus-sign, or have simply mis-remembered the formula entirely!). The function $f$ will then be a solution of Laplace's equation subject to the given boundary conditions. It's just possible that the integral I've written down is evaluatable by mere mortals, which gives you a solution.


$$ f(x,y) ~:=~ \frac{xy}{xy+ (1-x)(1-y)}.$$


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.