So as far as I know to be riemann integrable there should be an value (lets say e>0 ) that $U(f,P)-L(f,P) < e$. However my teacher showed that below function is not riemann integrable by this method.


$$f(x) = \begin{cases} 1 & x\in[0,1]\cap \Bbb Q , \\ 0 & otherwise .\end{cases}$$

What she did was, $U(f,P) = 1(1-0)$ and $L(f,P) = 0(1-0)$ Since both are not equal this is not riemann integrable.


What she did was clear to me. So I tried to prove this beliw one using the same logic. But I don't know how to start it. If someone can please tell me what should I do to prove whether this is reimann integrable or not.


$$f(x) = \begin{cases} 1 & x\in[0,1]\setminus\ \frac{1}{2}\, \\ 0 & x=\frac{1}{2} .\end{cases}$$

  • $\begingroup$ There are actually several different ways in which people define "the" Riemann integral. It looks like you might be using something like this one: math.ucdavis.edu/~hunter/m125b/ch1.pdf -- but maybe you are using some other variation. To prove the second integral to your satisfaction, it would be much better if we knew exactly what your definition is. $\endgroup$ – David K Oct 29 '16 at 20:00
  • $\begingroup$ Hmm, based on your responses to answers I think maybe the definition I pointed to is not the one you use, but evidently neither do you use the definition at mathworld.wolfram.com/RiemannIntegral.html. Perhaps you use a hybrid of those two definitions, with the $L$ and $U$ sums but with a uniform mesh? Regardless, if at least one of the answers resolves your doubts then (perhaps after waiting a couple of hours to make sure no new information changes your mind) you should accept the one you like best. $\endgroup$ – David K Oct 29 '16 at 20:08

Let $P$ be a partition and let $a$ be the length of the sub-intervals which includes $1/2$.

Note that: if $1/2$ is an end point, then there exist two such sub-intervals and if it is not, then there exists one sub-interval.

Since each sub-interval has image value $1$, we have the supremum on each interval is $1$. Therefore $$U(f,P)=1.$$ Since the length of interval having the $1/2$ is $a$, we have $$L(f,P)=1-a.$$

Now $$U(f)=1$$ and $$L(f)=\sup\{ 1-a \; | \; a \in (0,1]\}=1.$$ Therefore, it is Riemann integrable and the integral value is $1$.

  • $\begingroup$ Can you tell me what you implied by " length of the interval having 1/2 is a" ? How did you get that? $\endgroup$ – User9125 Oct 29 '16 at 19:36
  • $\begingroup$ I just called it "$a$". Do you know what partition, sub -interval and lenght of sub-interval are? $\endgroup$ – student forever Oct 29 '16 at 19:38
  • $\begingroup$ I know partition and sub interval. But never heard about taking length to do calculations. Please enlighten me. Also the way you got L(f,P) as 1-a . Please $\endgroup$ – User9125 Oct 29 '16 at 19:40
  • $\begingroup$ This is the usual calculation of Riemann integral. BECAUSE, infimum values get zero on the intervals having $1/2$ and the length we talked as before is $a$. $\endgroup$ – student forever Oct 29 '16 at 19:43
  • $\begingroup$ Thank you. But my problem is how you got that 1-a in riemann sum. Can you be more elaborate? $\endgroup$ – User9125 Oct 29 '16 at 19:46

It's a good idea to actually draw a few Riemann approximations, in this case.

Suppose $n$ is "large", and let's cut $[0, 1]$ into $n$ equal subintervals. Then:

  • The upper Riemann sum will have every single rectangle of height $1$ (why?), so will guess that the area is $(1\cdot {1\over n})\cdot n=1$.

  • The lower Riemann sum will have all but one rectangle of height $1$, and one will be of height $0$ (why?). So the upper Riemann sum will be $(1\cdot{1\over n})\cdot (n-1)+ 0\cdot{1\over n}={n-1\over n}$.

Now, do these Riemann sums look like they're converging to the same value as $n$ goes to infinity? What does that suggest about whether $f$ is Riemann integrable?

The distinction between your $f$ and the function your teacher showed you (the Dirichlet function) is that your $f$ only has one "bad" point ($f$ is discontinuous at $x={1\over 2}$), whereas the Dirichlet function has lots (every point is a bad point).

  • $\begingroup$ I can't understand how you get the lower riemann sum. Please explain it in the answer if you can. Thank you. $\endgroup$ – User9125 Oct 29 '16 at 19:15
  • $\begingroup$ @User9125 If I'm doing the lower Riemann sum, the height of the rectangle with base $(x_i, x_{i+1})$ is the minimum value $f$ achieves on that interval. (Actually, the infimum value - $f$ need not have a minimum - but here it does.) If I chop $[0, 1]$ into a bunch of pieces, one of them contains $x={1\over 2}$ - what can you say about the height of the corresponding rectangle, in the lower Riemann sum? $\endgroup$ – Noah Schweber Oct 29 '16 at 19:16
  • $\begingroup$ Yeah. As I know it is $mi(xi - x(i-1))$. I'm not sure how you get $(1\cdot{1\over n})\cdot (n-1)+ 0\cdot{1\over n}={n-1\over n}$. specially this \cdot. Sorry I have poor knowledge about this $\endgroup$ – User9125 Oct 29 '16 at 19:18
  • 1
    $\begingroup$ @User9125 Well, there are $n$ rectangles total, and each one has width ${1\over n}$. One of the rectangles (the one whose base contains $x={1\over 2}$) has height $0$, the rest have height $1$ (do you understand why?). So we have $n-1$ rectangles of area $1\cdot {1\over n}$, and one rectangle of area $0\cdot {1\over n}$ - that is, total area $(1\cdot {1\over n})\cdot (n-1)+(0\cdot {1\over n})\cdot 1$. (I left out the "$\cdot 1$" in my answer.) $\endgroup$ – Noah Schweber Oct 29 '16 at 19:20
  • $\begingroup$ Oh my GOD! Thanks alot. I noa can understand. as you have asked when n goes to infinity it equls to 1. So this integrable exists? Right? $\endgroup$ – User9125 Oct 29 '16 at 19:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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