Largest perimeter of rectangle in ellipse


By Anonymous on Saturday, April 21, 2001 - 08:31 pm :

Hi,

How can I find the largest possible perimeter for a rectangle that can be drawn inside the ellipse
(x/a)2 + (y/b)2 = 1.

It's only a 2 mark question, but I just can't see how to do it at the moment.

Thanks.


[All the answers which follow assume that the sides of the rectangle are parallel to the axes. - The Editor]
By Anonymous on Saturday, April 21, 2001 - 09:22 pm :

I get 4(a2 +b2 )½ - not sure if that's right. Seems to require quite a bit of thinking compared to ordinary 2 mark questions.


By Kerwin Hui (Kwkh2) on Saturday, April 21, 2001 - 09:24 pm :
Obviously the rectangle will have centre at 0, so we need only to find the value of 4(x+y)max. Parameterise the ellipse to find out that this is equivalent to max value 4(acosq+bsinq), which is easily seen to be 4(a2+b2)1/2.

Kerwin


By Anonymous on Saturday, April 21, 2001 - 10:23 pm :

I don't actually understand how you got the 4(a2 +b2 )1/2 .

Could you please explain?


By Kerwin Hui (Kwkh2) on Saturday, April 21, 2001 - 10:55 pm :
We have acosq+bsinq = (a2+b2)1/2sin(q+ a), where tana = a/b. Since -1 £ sinx £ 1, we have the required answer.

Kerwin


By Brad Rodgers (P1930) on Sunday, April 22, 2001 - 12:09 am :

Perhaps this explanation will seem to make more sense. First, are you ok with what we are trying to maximize is y+x? (Our answer will be 4 times this though). We can use only positive values for y since this is what we are finding. We therefore want to maximize x+b(1-x2 /a2 )1/2 . We can solve this by differentiating and setting to zero, then find x, find y, and add them together, then multiply by 4, and you should be able to come up with the answer.

Hope this helps,

Brad


By Brad Rodgers (P1930) on Sunday, April 22, 2001 - 12:11 am :

Though, I should mention, Kerwin's method is perfectly fine and a good deal more elegant than this method.


By Kerwin Hui (Kwkh2) on Sunday, April 22, 2001 - 12:42 am :

A more elegant method involves the QM-AM inequality.....


By Anonymous on Sunday, April 22, 2001 - 03:34 pm :

What's that method Kerwin? QM-AM: is that something to do with the arithmetic and q***(something) means?


By Kerwin Hui (Kwkh2) on Sunday, April 22, 2001 - 06:23 pm :
It is the Quadratic Mean-Arithmetic Mean inequality, which states that
  æ
 ú
Ö

å
xi2
n
 
³ å
xi
n
with equality if and only if all the xi are equal. This inequality is easily deduced from Cauchy-Schwarz. In fact, we can prove the continuous version of this

((1-l)x2+ly2)1/2 ³ (1-l)x+ly

for all 0 £ l £ 1. Using this inequality, we get
x+y
£
(a2+b2)[(1-b2/(a2+b2))(a-2x)2+(b2/(a2+b2))(b-2y)2 ]1/2
=
(a2+b2)[(x2 a-2+y2 b-2)/(a2+b2)]1/2
=
(a2+b2)1/2
Kerwin


By Olof Sisask (P3033) on Sunday, April 22, 2001 - 06:38 pm :

What is Cauchy-Schwarz? I've seen it mentioned a couple of times on here, but never asked about it.

Regards,
Olof


By Kerwin Hui (Kwkh2) on Sunday, April 22, 2001 - 06:44 pm :
C-S states that
æ
è
å
xi yi ö
ø
2
 
£ æ
è
å
xi2 ö
ø
æ
è
å
yi2 ö
ø
with equality if and only if x and y are parallel, i.e. there exists l and m, not both zero, such that lxi=myi, all i.

Kerwin


By Michael Doré (Michael) on Wednesday, April 25, 2001 - 02:56 pm :

This is basically an extension of the fact that the magnitude of a cosine of a real angle is less than or equal to 1.

If:

x = (x1 ,x2 )
y = (y1 ,y2 )

Then (x1 y1 + x2 y2 )2 = (x .y )2 = |x |2 |y |2 cos2 (theta) < = |x |2 |y |2 = (x1 2 + x2 2 )(y1 2 + y2 2 )

where theta is the angle between x and y . This demonstrates the inequality for n = 2 and also makes it clear why they need to be parallel for equality. You can justify it in a similar way for N-D though you need to be careful about how you define angle. Alternatively, you can prove the inequality by expanding it out and collecting terms and using the fact that squares are positive.


By Kerwin Hui (Kwkh2) on Wednesday, April 25, 2001 - 10:23 pm :

Here is a previous thread where C-S was discussed about halfway down the page.

Kerwin


By Olof Sisask (P3033) on Wednesday, April 25, 2001 - 10:43 pm :

Cheers guys.

Olof