I am working in data science and I have to deal with the following problem for which I would like to find a simplification:

We call a function almost positive if $f(x_1,y_1)f(x_2,y_2)-f(x_1,y_2)f(x_2,y_1) \ge 0$ for all $0< x_1\le x_2 < \infty$ and $0 < y_1\le y_2 < \infty.$

I would like to know: Are there any sufficient and necessary criteria for a function $f$ to be almost positive?

Background: The problem is that I often have a positive smooth function $f$ which I need to check for almost positivity. Those functions $f$ are usually cumbersome expressions such that checking

$f(x_1,y_1)f(x_2,y_2)-f(x_1,y_2)f(x_2,y_1) \ge 0$ is almost impossible analytically, because one has to compare infinitely many variables which each other and unless one can simplify the expression in a clever way, checking this condition is hopeless.

I am therefore asking whether there is an equivalent criterion to the almost positivity condition which I can check in a more direct way? Ideally there would exist an "intrinsic" criterion for functions $f$ which implies this property.

If there is nothing equivalent to almost positivity, perhaps there exist rather general sufficient conditions which imply almost positivity?

  • $\begingroup$ What is an example of such a function? Have you computed the local conditions? $\endgroup$ – Igor Rivin Sep 1 '18 at 22:44
  • $\begingroup$ Also, can you compute some level sets, especially where the function f has value zero? Gerhard "Maybe Special Cases Can Help" Paseman, 2018.09.01. $\endgroup$ – Gerhard Paseman Sep 1 '18 at 22:49

You say your function is smooth, so letting $x_1 = x, y_1 = y, x_2 = x + \Delta x, y_2 = y+\Delta y,$ we get in the limit as the deltas go to zero, if we ignore the second order terms, then

$$ \dfrac{\partial f}{\partial x} \dfrac{\partial f} {\partial x} d x d y < 0.$$ This indicates that we cannot ignore the second order terms, and when the smoke clears, we seem to get

$$f\dfrac{\partial^2 f}{\partial x \partial y} - \dfrac{\partial f}{\partial x} \dfrac{\partial f} {\partial x} \geq 0,$$ which is nonlinear and hyperbolic. Changing it to an equation may or may not be enlightening.


To follow up @igor answer, the equation can be written as: $$ \dfrac{\partial^2 \ln f}{\partial x \partial y} \geq 0. $$ Moreover, the inequality is not only necessary, bu sufficient, since: $$ \ln \left( \frac{f(x_2,y_2)f(x_1,y_1)}{f(x_2,y_1)f(x_1,y_2)} \right) = \int_{x_1}^{x_2} \int_{y_1}^{y_2} \dfrac{\partial^2 \ln f}{\partial x \partial y} dy \ dx. $$

  • $\begingroup$ The logarithm of a positive expression $$\frac{f(x_2,y_2)f(x_1,y_1)}{f(x_2,y_1)f(x_1,y_2)} $$ may be negative, is not so? $\endgroup$ – user64494 Sep 2 '18 at 9:20
  • 1
    $\begingroup$ It may be negative. The logarithm of such an expression (or any number) is positive if and only if the expression is greater than or equal to 1. $\endgroup$ – VictorZurkowski Sep 2 '18 at 9:24
  • $\begingroup$ Looks like an easily checkable condition... $\endgroup$ – Igor Rivin Sep 7 '18 at 19:49

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.