0
$\begingroup$

Hey Folks

I've encountered an optimization problem which has a linear programming structure but it's coefficients are nonlinear function of another variable. here is the problem: $$\max _{{p_i},\theta} \sum_{i=1}^N (\omega_i p_i e^{-1.2\times (\frac{\theta-\theta_i}{\theta_{3dB}})^2}-\eta\times p_i) $$

$$s.t. \sum_{i=1}^N p_i = P$$ $$\forall i \qquad p_i\ge0 \qquad and \qquad 0\le\theta\le 90^\circ$$ $$\forall i \qquad \omega_i\ge0 \qquad are \; known \; real \; positive \; numbers$$ $$\forall i \qquad 0\le\theta_i\le 90^\circ \qquad are \; known \; real \; positive \; numbers \; (degrees)$$ $$0\le\eta\le 1 \qquad and \qquad \theta_{3dB}=6^\circ \qquad and \qquad P\ge0 \qquad are \; known\; constants$$ $$e \qquad is \; neper \; exponential \; function$$

I'm interested in finding global maximum of the above function.

I think closed form solution is not possible however an algorithm that guarantees to approach global maximum is highly probable to exist.

Thanks a lot in Advance and Best Regards

$\endgroup$
3
  • 1
    $\begingroup$ It is probably not a convex problem. However I could solve the problem easily with the global solver Baron. I used some random data. I assumed "neper exponential" is just the standard exp(). $\endgroup$ Commented Jan 4, 2016 at 11:47
  • $\begingroup$ Indeed, it is not convex. $\endgroup$ Commented Jan 4, 2016 at 14:22
  • $\begingroup$ But the approach you describe in the comment below, in which you fix $\theta$ and solve an LP, seems reasonable. $\endgroup$ Commented Jan 4, 2016 at 14:30

1 Answer 1

Reset to default
0
$\begingroup$

I assume that the neper exponential function is increasing in its argument. Take the $j\in {1,\ldots,N}$ of the largest $\omega_j$. Set $\theta_j=\theta$ and $p_j=P$ and $p_i=0$ for $i$ not equal to $j$. $\theta_i$ for $i$ other than $j$ can be anything satisfying the constraint.

$\endgroup$
1
  • 1
    $\begingroup$ thanks for the answer. but there is a problem in doing that.I've wrote a MATLAB Program which do one dimensional exhaustive search over variable theta and for each theta solve a linear program with CVX and fmincon both.sometime the best strategy is not that. for example consider two weights that are a little less than the chosen weight but these two ways are very close together( and by very close i mean less than theta3dB for example theta3dB/3). in this case best strategy for theta is somewhere between those two and then allocating p to those close weights. $\endgroup$ Commented Jan 3, 2016 at 10:05

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.