Functions that cannot be integrated

  • Thread starter trdiayw
  • Start date
  • Tags
    Functions
In summary: I think it's called a " Simpson's Rule ". But it's more involved than that. In summary, you can study functions that are "non-integrable" by approximating them or doing a numerical integration. However, there are certain conditions that must be met for the function to be considered "elementary".
  • #1
trdiayw
1
0
Hi all,

I am in Calculus II now, and after studying several techniques to integrate functions I started wondering about functions that either cannot be integrated or are so time consuming and complex to integrate that it becomes impractical.

How do we study such functions? Can we still gather information about the function?

Thanks,

Trdiayw
 
Physics news on Phys.org
  • #2
If you are referring to functions that, theoretically, have an integral but it's not in any simple form, there are two options: approximate the function by something you can handle or do a numerical integration.

I'm not sure what you mean by "gather information". The integral seldom gives a much information about a function as the derivative.
 
  • #3
what is your definition of a function that "can be integrated"?

e.g. any bounded function on a bounded interval, whose discontinuities can be overed by a sequence of intervals of toital length less than any arbitrarily given positive number, has a riemann integral.


are you talking about the problem of finding "elementary" antiderivatives of "elementary" functions? if so you need to define "elementary".
 
  • #4
Any function that is differentiable between a and b can be integrated between a and b by taking the area under the curve. This area can be approximated by calculating the riemann sum of an infinite number of rectangles, using a limit. If you're suggesting the possibility of attaining the integrated equation in exact form, I can't help you out there. There are advanced integration methods that cover most situations that you would ever encounter, as far as I know. But there must be some that are inplausable to calculate.

[tex]\frac{1}{\sqrt{1-x^{2}}}[/tex]

For example. The integral of that is an inverse sine wave. However, I'm not sure if you can solve that without using the process that was used to determine the derivative of an inverse sine wave. In a different situation, you wouldn't know the derivative of the function, in order to determine the proof for finding the integral of the derivative. I could guess that this makes some equations "impossible" to prove.
 
Last edited:
  • #5
Sane said:
[tex]\frac{1}{\sqrt{1-x^{2}}}[/tex]

For example. The integral of that is an inverse sine wave. However, I'm not sure if you can solve that without using the process that was used to determine the derivative of an inverse sine wave.
I can solve this integral by letting x = sin(t), with [tex]t \in \left[ - \frac{\pi}{2} ; \ \frac{\pi}{2} \right][/tex]
[tex]\int \frac{dx}{\sqrt{1 - x ^ 2}} = \int \frac{\cos t dt}{\sqrt{1 - \sin ^ 2 t}} = \int \frac{\cos t dt}{|\cos t|}} = \int \frac{\cos t dt}{\cos t}[/tex]
Since cos(t) is positive for: [tex]t \in \left[ - \frac{\pi}{2} ; \ \frac{\pi}{2} \right][/tex]
[tex]\int \frac{dx}{\sqrt{1 - x ^ 2}} = ... = \int dt = t + C = \arcsin (x) + C[/tex] (Q.E.D)
:smile:
 
  • #6
I think you missed my point. I was saying you couldn't solve that if you didn't know its origins were a trigonmetric wave in the first place...
 
  • #7
Sane said:
I think you missed my point. I was saying you couldn't solve that if you didn't know its origins were a trigonmetric wave in the first place...

it is proven that only certain functions have elementary antiderivative.
http://www.claymath.org/programs/outreach/academy/LectureNotes05/Conrad.pdf"

if we know they have elementary antiderivative, we just need to play with trig, inverse trig, log, exp, polynomial, and intergration by parts.
 
Last edited by a moderator:
  • #8
[tex]\int e^{-x^2} dx[/tex]

This function is essentially the be all and end all of "non-integrable" functions. Though at this point, the numerical integral of this function can be computed in many cases faster than a good number of "known" functions .
 
  • #9
Even if you have something that's not "integrable", as long as it's differentiable you can find the taylor's series and integrate it term by term. It's not pretty, but it works
 
  • #10
ObsessiveMathsFreak said:
[tex]\int e^{-x^2} dx[/tex]

This function is essentially the be all and end all of "non-integrable" functions. Though at this point, the numerical integral of this function can be computed in many cases faster than a good number of "known" functions .

How is this different from

[tex]\int \mathrm{sin} x dx? [/tex]
 
  • #11
ObsessiveMathsFreak said:
[tex]\int e^{-x^2} dx[/tex]

This function is essentially the be all and end all of "non-integrable" functions. Though at this point, the numerical integral of this function can be computed in many cases faster than a good number of "known" functions .
Isn't that integral really similar to the normal distribution?

Substitute x^2 for k^2/2 => x = 2^-.5*k

Then you have the normal distribution without the usual normalization constant of 1/(2 pi)^.5,

So that integral should be equal to... pi^-.5. Assuming I did my math correctly. (if the integral is taken from negative infinity to infinity). To find smaller, arbitrary intervals you could use all the familiar stats tricks with normal densities.

IIRC, the way to actually calculate that out was to expand it to a double integral and then use a trig-substitution.
 
  • #12
Just a side note:

[tex]\int e^{-x^2} dx[/tex]

when integrated from +infinity to -infinity gives you Root Pi, if my memory serves me well. O and yes, it is quite a good approximation to the normal distribution, as is 1/(1+x^2).
 
Last edited:
  • #13
Office_Shredder said:
Even if you have something that's not "integrable", as long as it's differentiable you can find the taylor's series and integrate it term by term. It's not pretty, but it works

As long as it is infinitely differentiable, you can find its Taylor's series. Of course, it might not be equal to that Taylor's series so the integral you get might not be correct. In order to find an integral as an infinite series by integrating the Taylor's series term by term, you would have to have an analytic function.

Since analytic functions are very "nice" and the only requirement for integrability is that the function be bounded and have discontinuities only on a set of measure 0, jumping form integrable to analytic leaves out "almost all" integrable functions!
 
  • #14
i think u can integrate the function by expanding it over infinite terms
 
  • #15
Karthikthe said:
i think u can integrate the function by expanding it over infinite terms
What function are you talking about? And what do you mean by "expanding it over infinite terms"? Use the Taylor's series? That had already been said.
 
  • #16
Gib Z said:
Just a side note:

[tex]\int e^{-x^2} dx[/tex]

when integrated from +infinity to -infinity gives you Root Pi, if my memory serves me well. O and yes, it is quite a good approximation to the normal distribution, as is 1/(1+x^2).
[tex]f(x)= \frac{1}{2\sqrt{\pi}}e^{-x^2}[/tex]
is more than a "good approximation"! It is the normal distribution. The integral you gives, from [itex]-\infty[/itex] to [itex]\infty[/itex] is [itex]2\sqrt{\pi}[/itex], making the "area under the curve" for the normal distribution 1 as it should be. The integral from 0 to [itex]\infty[/itex] is [itex]\sqrt{\pi}[/itex].

And [itex]1/(1+x^2)[/itex] is a "not so good" approximation.
 
  • #17
I feel any function which shows you a trigonometric function on the numerator and something like x or x^2 in the denominator is non-integrable.( Hey, I know special cases may get them canceled). I am saying this only coz there happen to be infinite terms in expansion of a negative or fractional index
 
  • #18
On a bounded interval, any function whose discontinuities have measure zero can be Riemann integrated but there is no general technique. Take for instance, the characteristic function of the Cantor set on the unit interval or a 1 dimensional continuous Brownian path on the unit interval.

On an unbounded domain an integral may be infinite but this really is still integrable.

Some function's integrals however do not have well defined limits on unbounded domains. These are non-integrable.

Functions that have discontinuities of positive measure can not be Riemann integrated, for instance the characteristic function of the rational numbers on the unit interval. However, a more general notion of integral, the Lebesque integral, includes this function and many other that can not be Riemann integrated.

Much of mathematics is figuring out ways to integrate functions.
 
  • #19
Is there an algorithm that takes as input a function f(x) and numbers a and b (they can be infinite) that can decide if

[tex]\int_{a}^{b}f(x) dx[/tex]

has a closed form expression? So, if you take f(x)=exp(-x^2) and a = 0 and b = 1, the algorithm would return: "no", while if you take b to be infinity, it would say: "yes". Of course, the algorithm depends on a precise definition of "closed form expression". If you allow expressions involving the error function, then it would return a different output in this case.
 
  • #20
Count Iblis said:
Is there an algorithm that takes as input a function f(x) and numbers a and b (they can be infinite) that can decide if

[tex]\int_{a}^{b}f(x) dx[/tex]

has a closed form expression? So, if you take f(x)=exp(-x^2) and a = 0 and b = 1, the algorithm would return: "no", while if you take b to be infinity, it would say: "yes". Of course, the algorithm depends on a precise definition of "closed form expression". If you allow expressions involving the error function, then it would return a different output in this case.

I think that you have to be precise about what "closed form expression means". One can always define a new function as an integral and give it a name. Then does that integral suddenly have a closed form solution?
 
  • #21
wofsy said:
I think that you have to be precise about what "closed form expression means". One can always define a new function as an integral and give it a name. Then does that integral suddenly have a closed form solution?

Agreed. E.g. like erf(x). I find this discussion kinda meaningless because of this fact. I don't think of cos(x), or even just the identity function x -> x, as more important than any other function, e.g. erf(x).

Torquil
 
  • #22
Well, you indeed do have to specify a finite list that defines the elementary functions. The question is then given f(x), a, b, and the list, what is the algorithm that decides that the value of the integral has a finite formal expression in terms of the functions in the list. Any number appearing in the expression must be a rational numbers or . E.g. pi can be expressed as 4arctan(1), or e = exp(1), etc.
 
  • #23
Note one can always define an integral of a given function as the definition of the new function. See for example the error function, and also think of the natural logarithm being define as simply the integral of 1/x:

[tex]\ln(x)\equiv \int_1^x \frac{dz}{z}[/tex]

Also look at the elliptic functions for examples. Basically the integral is the solution to the equation F'(x) = f(x).

Defining the new function as the solution isn't all that strange. Consider that we didn't have a way to write the solution to [tex]x^2 = 2[/tex] so we invent the notation [tex]\sqrt{2}[/tex] to represent the positive solution. Similarly we "invent" a new function name to represent the solution to various integration problems.

Now there is a bigger question concerning functions for which no integral is defined period. Part of the subject of real analysis in higher mathematics deals with extending the definition of integration so that more classes of functions are integrable. Thus you'll see references to Riemann-Stieltjes integration and Lebesgue integration.

This, by the way, is why you start with the Riemann sums and such to first define the integral of a function as a limit of Riemann sums and then once it is defined, invoke the Fundamental Theorem of Calculus, to calculate certain integrals.

Generally any bounded piecewise continuous function (with finite pieces) will be Riemann integrable, there will be a function whose derivatives take on the values of the given function. It's out there but may not be named by any existing notation.
 
  • #24
Count Iblis said:
Is there an algorithm that takes as input a function f(x) and numbers a and b (they can be infinite) that can decide if

[tex]\int_{a}^{b}f(x) dx[/tex]

has a closed form expression?
Yes, but for indefinite integrals: Risch algorithm.
 
  • #25
HallsofIvy said:
[tex]f(x)= \frac{1}{2\sqrt{\pi}}e^{-x^2}[/tex]
is more than a "good approximation"! It is the normal distribution. The integral you gives, from [itex]-\infty[/itex] to [itex]\infty[/itex] is [itex]2\sqrt{\pi}[/itex], making the "area under the curve" for the normal distribution 1 as it should be. The integral from 0 to [itex]\infty[/itex] is [itex]\sqrt{\pi}[/itex].

And [itex]1/(1+x^2)[/itex] is a "not so good" approximation.

actually it's [tex]\frac{1}{2\sqrt{\pi}}e^{-x^2/2}[/tex]

and the integral of [tex]e^{-x^2}[/tex] is [itex]\sqrt{\pi}[/itex],
 

Related to Functions that cannot be integrated

1. What are "functions that cannot be integrated"?

Functions that cannot be integrated are mathematical functions that do not have a definite integral. This means that there is no single function that can represent the area under the curve of the function. In other words, these functions cannot be solved for using traditional integration methods.

2. Why can't some functions be integrated?

There are several reasons why some functions cannot be integrated. One reason is that the function may be too complex and cannot be expressed in terms of elementary functions such as polynomials, trigonometric functions, and exponential functions. Another reason is that the function may be discontinuous or have sharp corners, making it impossible to find a continuous function that represents it.

3. Can we approximate the integral of a function that cannot be integrated?

Yes, we can approximate the integral of a function that cannot be integrated using numerical methods such as the trapezoidal rule, Simpson's rule, or Monte Carlo integration. These methods use small intervals to approximate the area under the curve and provide a close estimation of the integral. However, these methods may not always provide an exact solution.

4. Are there any real-life applications for functions that cannot be integrated?

Yes, there are many real-life applications for functions that cannot be integrated. Some examples include calculating the area under a curve in physics, determining the optimal route for a vehicle in traffic engineering, and analyzing the spread of diseases in epidemiology. In these cases, numerical methods are often used to approximate the integral of the function.

5. Is there ongoing research on functions that cannot be integrated?

Yes, there is ongoing research on functions that cannot be integrated. Mathematicians and scientists are constantly exploring new methods and techniques to solve or approximate these functions. Some areas of research include the use of computer algorithms, advanced numerical methods, and the development of new mathematical theories to better understand these functions.

Similar threads

  • Calculus
Replies
5
Views
2K
Replies
1
Views
3K
Replies
3
Views
1K
Replies
14
Views
1K
Replies
36
Views
4K
Replies
2
Views
937
  • Calculus
Replies
6
Views
1K
Replies
12
Views
2K
Replies
8
Views
368
Back
Top