2

Im trying to express the following limit as a Riemann Sum

$$ \lim_{ n \to \infty} \sum_{i=1}^{n} \dfrac{ n^3 (2i-3) }{ (n^2- \sqrt{2n^4 + i^2n^2} + in )(2n^2-i^2)^{3/2} } $$

over the interval $[2,3]$

Clearly, we want $\dfrac{1}{n}$ out of the sum as to take $\Delta x$ in account. We can divide both sides by $\dfrac{1}{n^5}$ and we obtain the

$$ \lim_{ n \to \infty} \dfrac{1}{n}\sum_{i=1}^{n} \dfrac{ \dfrac{(2i-3)}{n} }{ (n^2- \sqrt{2n^4 + i^2n^2} + in )(2n^2-i^2)^{3/2} } \cdot \frac{1}{\dfrac{1}{n^2 \cdot n^3}}$$

Which reduces to

$$ \lim_{ n \to \infty} \dfrac{1}{n}\sum_{i=1}^{n} \dfrac{ \dfrac{(2i-3)}{n} }{ \left( 1+ \dfrac{i}{n} - \sqrt{ 2 + \dfrac{i^2}{n^2} } \right)\left(2- \left( \dfrac{i}{n} \right)^2 \right) } $$

Now, this is almost done as we want our height x-value to be $x = 2 + \dfrac{i}{n}$. However, here is where I get stuck and I ask for some help. The term $\dfrac{2i-3}{n}$ in the numerator is the only term that is problematic here. How would we handle this?

J. W. Tanner
  • 63,683
  • 4
  • 43
  • 88
James
  • 4,027
  • 3
    Note that for $n$ even, there is a term in the sum, for $i=n/2$, for which the denominator has the factor $n^2(1-\sqrt{2+1/4}+1/2)=0$. Is the limit supposed to be taken over odd $n$, or is there a typo? – conditionalMethod Nov 07 '19 at 19:01
  • @conditionalMethod Nice. It seems for odd $n$, the limit exists. – River Li Nov 17 '19 at 04:06
  • @RiverLi How do you know? – conditionalMethod Nov 17 '19 at 04:25
  • @conditionalMethod I just plotted a figure to see it. Maybe I am wrong. – River Li Nov 17 '19 at 06:01
  • @conditionalMethod For $n$ odd, I can prove that the summation is bounded. But I have not yet proved that the summation has a limit. Do you think that if $n$ is odd, or $n$ is even excluding $i=\frac{n}{2}$ in the summation, then the limit of both cases equals to the Cauchy principal value of some improper integral? – River Li Nov 18 '19 at 02:22
  • @RiverLi It is likely. That integral should be the integral below. Well, after proving that the terms that they removed below also tends to zero. – conditionalMethod Nov 18 '19 at 02:32
  • @conditionalMethod Yes, I gave a rigorous proof that the limit exists and equals to some proper integral. – River Li Nov 18 '19 at 07:32

2 Answers2

3

According to Riemann sum

$$\int_{a} ^{b} f(x) \, dx=\lim_{n\to\infty} \frac{b-a} {n} \sum_{i=1}^{n}f\left(a+i\cdot\frac {b-a} {n} \right)$$

we need to express the sum as a function of $\left(2+\frac{i}{n}\right)$ that is

$$\lim_{ n \to \infty} \frac{1}{n}\sum_{i=1}^{n} \frac{ 2\left(2+\frac{i}{n}\right)-4+\frac 3 n }{ \left( \left(2+ \frac{i}{n}\right)-1 - \sqrt{ \left(2 + \frac{i}{n}\right)^2-4\left(2+\frac in\right)+6 } \right)\left(-2+4\left(2+\frac i n\right)- \left(2+ \frac{i}{n} \right)^2 \right) }=$$

$$=\int_2^3\frac{2x-4}{(x-1-\sqrt{x^2-4x+6})(-2+4x-x^2)}dx$$

which is an improper integral since $x-1-\sqrt{x^2-4x+6}=0$ for $x=\frac 5 2$.

Refer also to

user
  • 162,563
  • The integral is improper, the term cannot be neglected. – conditionalMethod Nov 07 '19 at 18:54
  • @conditionalMethod Thanks! I'l revise that. – user Nov 07 '19 at 18:57
  • Why is the term $\dfrac{3}{n}$ neglected? – James Nov 10 '19 at 19:24
  • 1
    The integral at the end is divergent by comparison with $\frac{1}{2x-5}$. Note that it is not even one of those that diverge because its value, in the Lebesque sense, is infinite. It diverges because its positive part and its negative part both diverge. As a consequence, the integral doesn't tell anything about the limit of the special Riemann sums. The could potentially be converging similar to how Cauchy's principal value sometimes exists. – conditionalMethod Nov 17 '19 at 04:30
  • 1
    Before the integral is formed, there is the term $3/n$. One could split the sum in two $\sum\frac{1}{n}\frac{2(2+i/n)-4}{(...-\sqrt{...})(...)} + \frac{3}{n}\sum\frac{1}{n}\frac{1}{(...-\sqrt{...})(...)}$, but the integral of which $\sum\frac{1}{n}\frac{1}{(...-\sqrt{...})(...)}$ is a Riemann sum is also improper divergent. Potentially, the limit of $\sum\frac{1}{n}\frac{1}{(...-\sqrt{...})(...)}$ could be infinite such that when multiplied to $\frac{3}{n}$ the result doesn't tend to zero. – conditionalMethod Nov 17 '19 at 04:34
2

As @conditionalMethod pointed out, for $n$ even, one of the denominators is zero. If we exclude $2i-n=0$ in the summation, the limit exits and equals to some proper Riemann integral. In other words, usual Riemann integral is enough to deal with the problem, and improper integral is not necessarily introduced. Also, from @conditionalMethod's comments on @user's answer, we may need to deal with the term $\frac{3}{n}$ carefully.

Let us prove that \begin{align} &\lim_{n\to \infty} \sum_{i=1, \ 2i-n\ne 0}^n \frac{n^3(2i-3)}{(n^2-\sqrt{2n^4+i^2n^2} + in)(2n^2-i^2)^{3/2}}\\ =\ & \int_0^1 \frac{1+x+\sqrt{2+x^2}}{(2-x^2)^{3/2}} \mathrm{d}x \\ & + \int_0^1 \frac{1}{2(1-2x)} \Big(\frac{1 + (1-x) + \sqrt{2 + (1-x)^2}}{(2-(1-x)^2)^{3/2}} - \frac{1 + x + \sqrt{2 + x^2}}{(2-x^2)^{3/2}} \Big)\mathrm{d} x. \end{align} Remark: The second integral is a proper Riemann integral rather than an improper integral, since the integrand is continuous on $[0, 1]$ (it is obvious if it is rewritten to reduce $1-2x$, although the expression hence becomes complicated).

Proof: First, we have \begin{align} &\sum_{i=1, \ 2i-n\ne 0}^n \frac{n^3(2i-3)}{(n^2-\sqrt{2n^4+i^2n^2} + in)(2n^2-i^2)^{3/2}}\\ =\ & \sum_{i=1, \ 2i-n\ne 0}^n \frac{n^3(2i-n)}{(n^2-\sqrt{2n^4+i^2n^2} + in)(2n^2-i^2)^{3/2}}\\ &\qquad\qquad + \sum_{i=1, \ 2i-n\ne 0}^n \frac{n^3(n-3)}{(n^2-\sqrt{2n^4+i^2n^2} + in)(2n^2-i^2)^{3/2}}\\ =\ & \sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n+i +\sqrt{2n^2+i^2})}{(2n^2-i^2)^{3/2}} + \sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n-3)(n+i+\sqrt{2n^2+i^2})}{(2i-n)(2n^2-i^2)^{3/2}}. \end{align}

Second, we have \begin{align} &\lim_{n\to \infty} \sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n+i +\sqrt{2n^2+i^2})}{(2n^2-i^2)^{3/2}} \\ = \ & \lim_{n\to \infty} \frac{1}{n}\sum_{i=1}^n \frac{1 + \frac{i}{n} + \sqrt{2 + (\frac{i}{n})^2}}{(2 - (\frac{i}{n})^2)^{3/2}}\\ = \ & \int_0^1 \frac{1+x+\sqrt{2+x^2}}{(2-x^2)^{3/2}} \mathrm{d}x. \end{align}

Third, since (with substitution $i \rightarrow n-i$) \begin{align} &\sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n-3)(n+i+\sqrt{2n^2+i^2})}{(2i-n)(2n^2-i^2)^{3/2}}\\ =\ & \sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n-3)(n+(n-i)+\sqrt{2n^2+(n-i)^2})}{(2(n-i)-n)(2n^2-(n-i)^2)^{3/2}}\\ &\quad + \frac{(n-3)(1+\sqrt{2})}{2n^2 \sqrt{2}} + \frac{(n-3)(2+\sqrt{3})}{n^2}, \end{align} we have \begin{align} &\sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n-3)(n+i+\sqrt{2n^2+i^2})}{(2i-n)(2n^2-i^2)^{3/2}}\\ =\ & \frac{1}{2}\sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n-3)(n+i+\sqrt{2n^2+i^2})}{(2i-n)(2n^2-i^2)^{3/2}}\\ &\quad + \frac{1}{2} \sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n-3)(n+(n-i)+\sqrt{2n^2+(n-i)^2})}{(n-2i)(2n^2-(n-i)^2)^{3/2}}\\ &\quad + \frac{1}{2} \frac{(n-3)(1+\sqrt{2})}{2n^2 \sqrt{2}} + \frac{1}{2}\frac{(n-3)(2+\sqrt{3})}{n^2}\\ =\ & \sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n-3)}{2(n-2i)}\Big(\frac{n+(n-i)+\sqrt{2n^2+(n-i)^2}}{(2n^2-(n-i)^2)^{3/2}} - \frac{n+i+\sqrt{2n^2+i^2}}{(2n^2-i^2)^{3/2}}\Big)\\ &\quad + \frac{1}{2} \frac{(n-3)(1+\sqrt{2})}{2n^2 \sqrt{2}} + \frac{1}{2}\frac{(n-3)(2+\sqrt{3})}{n^2}\\ =\ & \frac{n-3}{n} \cdot \frac{1}{n} \sum_{i=1, \ 2i-n\ne 0}^n f\big(\frac{i}{n}\big) + \frac{1}{2} \frac{(n-3)(1+\sqrt{2})}{2n^2 \sqrt{2}} + \frac{1}{2}\frac{(n-3)(2+\sqrt{3})}{n^2}\\ =\ & \frac{n-3}{n} \cdot \frac{1}{n} \sum_{i=1}^n f\big(\frac{i}{n}\big) - \frac{n-3}{n^2}f\big(\frac{1}{2}\big)\cdot \mathrm{mod}(n-1,2)\\ &\quad + \frac{1}{2} \frac{(n-3)(1+\sqrt{2})}{2n^2 \sqrt{2}} + \frac{1}{2}\frac{(n-3)(2+\sqrt{3})}{n^2} \end{align} where $\mathrm{mod}(n-1,2) = 1$ is $n$ is even and $\mathrm{mod}(n-1,2) = 0$ is $n$ is odd, and $$f(x) = \frac{1}{2(1-2x)} \Big(\frac{1 + (1-x) + \sqrt{2 + (1-x)^2}}{(2-(1-x)^2)^{3/2}} - \frac{1 + x + \sqrt{2 + x^2}}{(2-x^2)^{3/2}} \Big).$$ Here we have used the fact that $f(x)$ is continuous on $[0, 1]$ with $f(\frac{1}{2})=\frac{328}{1029}\sqrt{7}$. Thus, we have $$\lim_{n\to \infty} \sum_{i=1, \ 2i-n\ne 0}^n \frac{n(n-3)(n+i+\sqrt{2n^2+i^2})}{(2i-n)(2n^2-i^2)^{3/2}} = \int_0^1 f(x) dx.$$ The desired result follows. This completes the proof.

River Li
  • 49,125
  • Cool. The value of the integral seems to be $2.5785458...$, which is also the sum for $n=10^8$. – conditionalMethod Nov 18 '19 at 11:57
  • Nice. The important thing is: That the term $\frac{3}{n}$ is neglected without explanation seems not good. – River Li Nov 18 '19 at 12:12
  • 1
    Well, I think one can ultimately prove that the $3/n$ can be removed, with the proper explanation that the sum that multiplies it is bounded. But more important is to add together the positive part and the negative parts of what would be the improper integral such that their infinite growth manages to cancel. This is what you did after the 'we have' by adding together last and first term, second and second to last, etc. – conditionalMethod Nov 18 '19 at 14:40