Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I want to find the value of $\vec{p}$, $p_s$, $p_t$ each of which is a function of the form $f:\mathbb{R}^2 \to \mathbb{R}$ that maximize the following function :

$$\begin{align} \int_\mathbb{R^2} \{ p_s &- \alpha(x,y)|\vec{p}| - \beta(x,y)(p_s - C_s) \\ &- \gamma(x,y)(p_t - C_t) + \lambda(x,y)(\vec{\nabla}\cdot\vec{p} - p_s + p_t)\} \end{align} $$

Each of $\alpha, \beta, \gamma$ is constrained to be non negative, $\lambda$ is unconstrained. $C_s$ and $C_t$ are given functions of the form $f:\mathbb{R}^2 \to \mathbb{R}$.

Edit :

Also we want the sup to be finite and therefore can put constraints on $\alpha, \beta, \gamma, \lambda$ accordingly.

The problem is a convex optimization problem.

I am hoping to minimize it by using a derivative approach : minimize function for a given variable by finding a derivative with respect to the variable and equating it to zero. I just dont know how to do that here.

share|improve this question
    
@amWhy thanks for editing, I am still learning to use latex properly. –  AnkurVijay Jun 4 '11 at 1:27
    
I think the idea should be to find partial derivatives with respect to the variables and then solve. –  AnkurVijay Jun 4 '11 at 1:32
    
Since $p_s$, $p_t$, $p_1$, and $p_2$ are real-valued functions, this is a calculus of variations problem. I have therefore added the "calculus-of-variations" tag. –  Jim Belk Jun 4 '11 at 1:42
    
Thanks for tagging it correctly –  AnkurVijay Jun 4 '11 at 1:43
    
@AnkurVj: no problem; actually, you did a fine job with your formatting, given your learning LaTeX...I just fine-tuned it, but the problem was fully comprehensible and well presented as you posted it. –  amWhy Jun 4 '11 at 2:34

1 Answer 1

up vote 3 down vote accepted

There is no maximum unless $\beta+\lambda\equiv1$ and $\gamma\equiv\lambda$. The terms containing $p_s$ add up to

$$\iint_{\mathbb R_2}p_s(1-\beta-\lambda)\mathrm dx\mathrm dy\;,$$

so you can make the integral arbitrarily large by choosing $p_s$ arbitrarily large (positive or negative depending on the sign of $1-\beta-\lambda$). Likewise, the terms containing $p_t$ add up to

$$\iint_{\mathbb R_2}p_t(\lambda-\gamma)\mathrm dx\mathrm dy\;.$$

This fixes $\beta$ and $\gamma$ in terms of $\lambda$, and then $p_s$ and $p_t$ no longer occur in the integral. We can drop the constant terms containing $\beta$ and $\gamma$, leaving us with

$$I=\iint_{\mathbb R^2}\left[- \alpha|\vec{p}|+ \lambda(\vec{\nabla}\cdot\vec{p})\right]\mathrm dx\mathrm dy\;.$$

Let's restrict ourselves to solutions $\vec{p}$ that decay sufficiently rapidly at infinity that we can integrate by parts and omit the boundary term:

$$I=\iint_{\mathbb R^2}\left[- \alpha|\vec{p}|- (\vec{\nabla}\lambda)\cdot\vec{p}\right]\mathrm dx\mathrm dy\;.$$

So we want to find $\vec{p}$ such that its magnitude (weighted by $\alpha$) is small but its component along $-\vec{\nabla}\lambda$ is large. This will work best if we choose $\vec{p}$ to always point along $-\vec{\nabla}\lambda$, i.e. $\vec{p}=-\mu\vec{\nabla}\lambda$, with as yet undetermined $\mu(x,y)$. With $\vec{g}:=\vec{\nabla}\lambda$, this gives

$$ \begin{eqnarray} I &=& \iint_{\mathbb R^2}\left[- \alpha|-\mu\vec{g}|- \vec{g}\cdot(-\mu\vec{g})\right]\mathrm dx\mathrm dy \\ &=& \iint_{\mathbb R^2}\left[- \alpha|\mu||\vec{g}|+\mu|\vec{g}|^2\right]\mathrm dx\mathrm dy \\ &=& \iint_{\mathbb R^2}|\vec{g}|\left[- \alpha|\mu|+\mu|\vec{g}|\right]\mathrm dx\mathrm dy \;. \end{eqnarray}$$

Thus, we must have $\alpha\ge|\vec{\nabla}\lambda|$ everywhere; else we could make the integral arbitrarily large by choosing $\vec{p}$ to vanish wherever that inequality holds and to make the integral arbitrarily large where it doesn't.

Given that inequality, the integral is forced to be non-positive, and thus it is maximized for $\vec{p}\equiv0$.

The first conclusion, $\alpha\ge|\vec{\nabla}\lambda|$, holds independent of the assumption about the behaviour at infinity, since we can choose $\vec{p}$ to have that behaviour if we want. The second conclusion, $\vec{p}\equiv0$, however, holds only under that assumption, since the boundary term might compensate a negative value of the integral. I'm not sure how to treat the general case, without decay at infinity.

[Edit:] Quite generally speaking, the parts of the integral that depend on $\vec{p}$ all scale linearly if you scale $\vec{p}$ by a positive scale factor, so there can't be any maximum other than $\vec{p}=0$ -- all you can do by imposing constraints on $\alpha$, $\beta$, $\gamma$ and $\lambda$ is to ensure that $\vec{p}=0$ is indeed a maximum.

share|improve this answer
    
This is what i landed up with too, making the above substitutions eliminates $p_s$ and $p_t$ from the equation. Now I just need to find $\vec{p}$ that maximizes the rest of the equation. Am I on the right track ? could you help me get a bit more further .. $ \vec{p}= p_1 i + p_2 j $ –  AnkurVijay Jun 4 '11 at 5:39
    
From the question, it seemed that $\beta$, $\gamma$ and $\lambda$ were given, so I don't understand what you mean by "making the above substitutions". If $\beta$, $\gamma$ and $\lambda$ are also unknown, then the problem is wildly underdetermined. –  joriki Jun 4 '11 at 5:43
    
Basically, we are allowed to put restrictions on $\beta, \gamma, \lambda$ so that the equation has a finite sup. That is we can assume such conditions on these variables so that the sup is finite –  AnkurVijay Jun 4 '11 at 5:46
    
@AnkurVj: That's a very important part of the problem that you should state in the question. –  joriki Jun 4 '11 at 5:49
    
Okay, should I edit the question now to reflect the same ? –  AnkurVijay Jun 4 '11 at 5:50

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.