Mixed linear geometric programming with cvx

1

Is it possible to solve Mixed linear geometric programming with cvx through the use of the special GP mode?
If yes, how? (a short example would help a lot).
If not, is it planned?

2

Mixed integer geometric programming problems can be entered in CVX, and MOSEK called to (attempt to) solve them. Unfortunately, geometric programs require use of CVX’s successive approximation method, which does not work correctly with integer variables.

So the bottom line is that such problems can be entered in CVX and passed to MOSEK, but they wiill not be solved correctly. To enter such a problem, enter the same way you would enter the problem if all variables were continuous, except declare your integer and binary variables as such. But you may very well get the wrong answer.

3

I’m afraid they cannot be solved in CVX, no. I can offer no guidance about whether or not this will ever be possible.

4

The question was about Mixed linear geometric programming not Mixed integer geometric programming, as discussed on page 100 of “A tutorial on geometric programming”:

e.g. with constraints like:
0.5x^2y^2+1.2y^0.5z<=3a+0.5b-2*c+10

5

Ah, thanks for the clarification. It’s possible that you could do this manually by doing something like this:

variables a, b, c, x_l, y_l, z_l
x = exp(x_l)
y = exp(y_l)
z = exp(z_l)
6

Thank you for your help,
Danny