# Numerical Integration of sin(1/x)

• Urkel
In summary, The conversation discusses the use of Numerical Recipes to integrate an integral involving sin(x)/x^2 and the suggestion to use a high-order series expansion or a spectral method for better convergence. The use of finite difference methods is preferred, but the current methods fail to converge at smaller values. A suggestion is made to approximate the function with spikes for smaller values to improve convergence.
Urkel
Hi everyone,
I am writing a simple code using Numerical Recipes (that bible of numerical method) to

integrate using trapezoid rule the following integral
int_pi/2_inf {sin(x)/x^2} dx
I first make variable change y = 1/x to change limit of integration so that now the integral

becomes int_0_2/pi sin(1/y) dy
I ever took analysis and this function has very special properties because of its extremely rapid

oscillation near x=0. Also, it has a special name, I forgot what it is. Anybody knows what

function sin(1/x) is commonly called in analysis?
Because of this rapid oscillation, one must use much finer sampling density when employing

trapezid rule to get accurate answer.What do you think further simple analytic transformation

that I can make to improve the convergence of this numerical integration using trapezoid rule?

Suggestions would be highly appreciated.

Urkel

Urkel said:
Hi everyone,
I am writing a simple code using Numerical Recipes (that bible of numerical method) to

integrate using trapezoid rule the following integral
int_pi/2_inf {sin(x)/x^2} dx
I first make variable change y = 1/x to change limit of integration so that now the integral

becomes int_0_2/pi sin(1/y) dy
I ever took analysis and this function has very special properties because of its extremely rapid

oscillation near x=0. Also, it has a special name, I forgot what it is. Anybody knows what

function sin(1/x) is commonly called in analysis?
Because of this rapid oscillation, one must use much finer sampling density when employing

trapezid rule to get accurate answer.What do you think further simple analytic transformation

that I can make to improve the convergence of this numerical integration using trapezoid rule?

Suggestions would be highly appreciated.

Urkel

Yes, it's related to the Cosine Integral.

It's not an easy integral to do. Essentially, your best bet is to use high-order series expansions of the solution.

Another route is to go with a spectral method like Fourier or Chebyshev functions. Just make sure that you verify convergence as the number of basis functions increases. I'm fairly sure you won't get anywhere fast with typical finite difference integration.

Last edited:
Thanks for the advice but I particulalry intend to work with finite difference methods.
I use 3 methods at once; trapezoid, Romberg and Gaussian quadrature to integrate out int_pi/2_inf sin(x)/x^2 dx followed first by defining y = 1/x and transforming the integral to int_0_2/pi sin(1/y) dy.
All the methods fail in doing the integral in the latter form when I set the integration limits to something like 0.001 for lower limit and 2/pi for the upper one. When I set the lower to, say, 0.5, all three methods give exactly the same answer so at least I know my codes for 3 algorithms are correct. So, what should I do, what further analytic transformation I need to do, if any, to make the integral converges (the code works) when I set lower limit back to small value close to 0?

Regards

As x gets smaller and smaller, each of the "bumps" on the curve basically looks like a little spike, with zeros at 1/(n*Pi) and 1/((n+1)*Pi), and maximum 1 at 1/((n+1/2)*Pi).

You could do an integration scheme where you approximate the function with spikes for x<a (for small 'a' determined by error estimates), and with better approximation for x>=a.

Here is an image:
http://img132.imageshack.us/img132/259/spikyintegrationpk1.png

Then you can just sum the alternating series rather than actually evaluating points.

This can be made completely rigorous in terms of getting within an error tolerance, since you also have the bound on the derivative on that interval, which is the derivative on the left hand side. For x in [1/nPi, 1/(n+1)Pi], you have |f'(x)| < |f'(1/(n+1)Pi)| = |cos(1/(n+1)Pi)/((n+1)Pi)^2|. Then you can use this to bound the error in approximation of each spike with the error estimate for polynomial interpolation.

Last edited by a moderator:

## 1. What is numerical integration of sin(1/x)?

Numerical integration of sin(1/x) is a method used to approximate the area under the curve of the function sin(1/x) for a given interval. It involves dividing the interval into smaller subintervals and using numerical techniques, such as the trapezoidal rule or Simpson's rule, to estimate the area.

## 2. Why is numerical integration necessary for sin(1/x)?

Numerical integration is necessary for sin(1/x) because the function cannot be integrated analytically. This means that there is no closed-form solution for the definite integral of sin(1/x), and numerical methods must be used to approximate the value.

## 3. How does the choice of integration method affect the accuracy of the approximation?

The choice of integration method can have a significant impact on the accuracy of the approximation. Generally, more advanced methods, such as Simpson's rule, will yield more accurate results compared to simpler methods, such as the trapezoidal rule. However, the accuracy also depends on the number of subintervals used and the smoothness of the function.

## 4. What are some real-world applications of numerical integration of sin(1/x)?

Numerical integration of sin(1/x) has many applications in physics, engineering, and other fields. It can be used to calculate the electric field around a charged wire, the force on a charged particle in a magnetic field, or the potential energy of a particle in a gravitational field.

## 5. Are there any limitations or challenges when using numerical integration for sin(1/x)?

One limitation of using numerical integration for sin(1/x) is that it can be computationally intensive, especially for functions with large or infinite intervals. Additionally, the accuracy of the approximation may decrease if the function has sharp changes or singularities within the interval. It is also important to choose an appropriate number of subintervals to balance accuracy and computation time.

• Calculus
Replies
2
Views
261
• Calculus
Replies
5
Views
1K
• Calculus
Replies
3
Views
1K
• Calculus
Replies
4
Views
1K
• Calculus and Beyond Homework Help
Replies
5
Views
663
• Calculus
Replies
3
Views
1K
• MATLAB, Maple, Mathematica, LaTeX
Replies
13
Views
2K
• Calculus
Replies
4
Views
2K
• Calculus
Replies
5
Views
3K
• Calculus
Replies
2
Views
2K