I have this constraint in an optimization problem:
$$\frac{a}{x_1} + \frac{b}{x_2} + \frac{c}{x_3} = d$$
where $x_i$ are the variables. Other than this my problem is linear. Is it possible to transform this constraint into a linear form?
I have this constraint in an optimization problem:
$$\frac{a}{x_1} + \frac{b}{x_2} + \frac{c}{x_3} = d$$
where $x_i$ are the variables. Other than this my problem is linear. Is it possible to transform this constraint into a linear form?
No.
Non-linear equality constraints are always non-convex, and as you have other linear constraints, you will probably not be able to make any clever variable change, as that will turn your other constraints non-linear and most likely non-convex instead.
However, depending on the sign of your coefficients, and the objective function, it could be that your problem can be written as a convex problem at least. If your objective is deceasing in $x$ (i.e. smaller $x$ better) and $a$, $b$ and $c$ are positive, you can replace the inequality with $\leq$, as it will be tight at optimality. With that, you have a convex program (and if you want, you can write the fractional constraint using a second-order cone reformulation if you want, and thus solve it using second-order cone programming which is a slight generalization of linear programming, otherwise just use a standard nonlinear solver)
a/x1 + b/x2 + c/x3 <= d
How do I transform this into a second-order cone problem?
– Jakob Loegstrup Dec 21 '17 at 14:32