ò2x1/2 sin(x) dx


By Anonymous on Friday, December 29, 2000 - 10:54 pm :

How about this?

ò2x1/2 sin(x) dx?
By Brad Rodgers (P1930) on Saturday, December 30, 2000 - 02:59 am :

As far as I can tell, evaluating the integral of 2x1/2 sin(x) with respect to dx is going to involve the fresnel cosine integral. In fact, solving this would actually solve the fresnel cosine integral, which I can't find a solution for anywhere, so I don't think that there is a solution. Here's what I did to show this:


y=2 cos(x)(x/(2p))1/2

Then,

dy/dx=-sin(x)2(x/2p)1/2+cos(x)/(2px)1/2

Therefore,

ò2x1/2 sin(x) dx=(2p)1/2[-cos(x)(2x/p)1/2+òcos(x)/ (2p×x)1/2 dx]

If we have a solution to

òcos(x)/(2px)1/2 dx

then we have a solution to the fresnel integral:

m=òcos(pt2/2) dt

as if we let

x=pt2/2

then

dx=pt dt

=(2px)1/2 dt

and we obtain,

dm=cos(x)/(2px)1/2 dx

so

òcos(x)/(2px)1/2 dx=òcos(pt2/2) dt

Which is the term we need to find. So, I don't think that the result can be simplified any further. In terms of the fresnel integral, the answer though is

-2x1/2 cos(x)+(2p)1/2×C((2x/p)1/2)

where C is the fresnel cosine integral.
This all of course can be simplified if there is an easy solution to the fresnel integral.

Brad


By Anonymous on Saturday, December 30, 2000 - 02:42 pm :

Thanks Brad!
I've never heard of a Fresnel Integral before? Can you explain a little detail of what it means? And when it is to be used. Maybe if you've got time, some small examples on what a Fresnel Integral is also.


By Brad Rodgers (P1930) on Saturday, December 30, 2000 - 05:10 pm :

I actually just found out about it last night while looking for a way to integrate your expression. Anyways, here's what I was able to find:


C(x) [the fresnel cosine integral] is defined as

C(x)=ò0x cos(pt2/2) dt

and

S(x) [the fresnel sine integral] is defined as

S(x)=ò0x sin(pt2/2) dt

The application my book gives here is if the error function is

erf(t)=2/(p)1/2ò0t e-y2 dy

then

(1-i)-1erf(t)=C(t)-i S(t)

where i is imaginary 1. I'm afraid I don't know a proof.
Directly above this is, "Fresnel Diffraction"; so I think it may have some applications to diffraction, but surprisingly, it says nothing about this in the book.

The only application that I can find other than the error function and possibly diffraction is soving integrals like the one you have. Since you can numerically find aproximate areas under a function, the answer to this problem is very much like a standard function. I'll try to find out more about this, sorry that I couldn't be of more help. Perhaps someone else knows more...

Brad


By Anonymous on Saturday, December 30, 2000 - 05:18 pm :

Thank you Brad!


By Sean Hartnoll (Sah40) on Saturday, December 30, 2000 - 06:53 pm :
Brad - the proof you mention is I suspect just using ei x=cos x+i sin x, maybe followed by a change of variables.

There are various series expansions for the error function. There are also so-called asymptotic expansions which give a better approximation for large x.

One famous (very useful) result that is not too difficult to prove, although not trivial (perhaps you could try) is that


ò-¥¥ e-x2 dx=
Ö
 

p
 

Sean


By Sean Hartnoll (Sah40) on Saturday, December 30, 2000 - 06:56 pm :
I should have added that the reason it is famous is that it is another mysterious link between the numbers e and p and it is useful because the curve f(x)=e-x2 is called the Gaussian or Bell or Normal curve and is used all over the place in statistics and also quantum mechanics and statistical physics.

Sean


By Brad Rodgers (P1930) on Tuesday, January 2, 2001 - 07:10 pm :

How do you come up with that result for the bell curve? I've tried a few different ways:

1. Write in terms of the error function. If the error function is equal to 1 at infinity, then the result is true. But this just leads to a harder problem unless there is an easy way to evaluate the error function.

2. Solve the indefinite integral straight forward. This appears to be impossible. Or at least very hard.

3. I had the idea of expanding it into a series, but the only way I know to expand a series is way of the Maclaurin series I get:


òe-y2dy=(y)-(y3/3)+(y5/5×2!)-(y7/7×3!)+(y9/9!×4!)-...
I have no idea how to evaluate this, or even it is able to be evaluated. You mentioned something about asyptotic expansions being accurate for larger values of y. What are these, and would they work better to solve this sort of equation?

4. My last idea was to put -y2 =ix

This gives the integral we need to evaluate as

(i1/2/2)ò[cos(x)+i×sin(x)]/x1/2
This of course involves the Fresnel Integral, which doesn't seem to be an easier integral to evaluate either.

Can you give a little hint as to what I need to do to sove this integral?

Thanks,

Brad
By Sean Hartnoll (Sah40) on Tuesday, January 2, 2001 - 11:28 pm :
Brad - It's probably best if I just give you the answer, it's probably a bit tricky to get on your own, and you may not have seen all the steps before

We want to evaluate I = ò-¥¥ e-x2 dx.

Consider

I2 = ò-¥¥ò-¥¥e-x2-y2 dx dy

This is called a multiple integral. To start with you can think of it as meaning we integrate over x and then over y. But a better way to think of it is that we are integrating the function over the whole x-y PLANE and that dx dy is an area element (wheras before dx was a length element). The trick now is to rewrite this expression in polar coordinates, defined by

x = r cosq

y = r sinq

(which imply r2 = x2 + y2)

The area element in these coordinates is

r dqdr

(this can be motivated by noting that if you integrate r from 0 to a and q from 0 to t then you get the area of a circular segment of radius a and angle t).

So the integral is now

I2 = ò0¥ò02p e-r2 r dqdr

Now because there is an extra r we can do the integral! Perhaps you would like to do it and see that we get p, which then gives the result by taking square roots.

Let me know if any of these steps don't make sense,

Sean



By Sean Hartnoll (Sah40) on Tuesday, January 2, 2001 - 11:31 pm :

Asymptotic expansions are expansions of functions in terms of other functions (such as powers) such that the difference between the function and its expansion goes to zero as the variable x goes to infinity.

Sean


By Michael Doré (Md285) on Tuesday, January 2, 2001 - 11:37 pm :

I thought that asymptotic meant that the ratio -> 1 rather than the difference -> 0. Of course these are the same in this case (I think) as the function isn't tending to 0 or infinity...


By Brad Rodgers (P1930) on Wednesday, January 3, 2001 - 03:41 pm :
I think that I understand, although a few details are a bit foggy. The main thing I don't understand is why r dqdr is the area element. In other words, why is dy dx=r dr dq? Other than this it seems to make sense to me.


Thanks,

Brad


By Sean Hartnoll (Sah40) on Thursday, January 4, 2001 - 05:09 pm :
Okay, first the point I made in the previous post is essentially why you want it to be true, if you take r dr dq to be the area element then the area of a segment with radius a and angle q¢ is

Area = ò0a ò0q¢r dr dq = q¢×a2/2 which is what it should be (note that we get the area for a circle if we put q¢ = 2p.

To see the infinitessimal case directly (draw a picture)

1) go to a point (r,q)

2) increase r slightly, by dr.

3) increase q slightly, by dq. By the formula for the length of a circular curve, we know that this gives a length r dq.

What we get is not quite a rectangle, but as we let dr and dq get arbitrarily small, then it is nearly a rectangle and it has area r dqdr.

Sean