2

I am trying to find the Maclaurin series of $\sqrt[a]{x+1}$ where $a > 1$.

My approach so far is:

$$\sqrt[a]{x+1} = c_0 + c_1x + c_2x^2 + c_3x^3 ...$$ $$Let \; x \rightarrow 0$$ $$c_0 = 1$$ $$\sqrt[a]{x+1} - 1 = c_1x + c_2x^2 + c_3x^3 + c_4x^4...$$ $$\frac {\sqrt[a]{x+1} - 1} x = c_1 + c_2x^1 + c_3x^2 + c_4x^3...$$ $$Let \; x \rightarrow 0$$ $$\lim_{x \rightarrow 0}{\frac {\sqrt[a]{x+1} - 1} x} = c_1$$

But then I get stuck at solving this limit, as my goal is avoid derivatives and L'Hopital's rule. If I can find the Maclaurin series of $\sqrt[a]{x+1}$, I'm hoping I can find Binomial Coefficients (or rows of Pascal's Triangle) for all real powers (I originally said fractional powers in the question, my mistake).

Is there a way to solve this limit without L'Hoptial (e.g. squeeze theorem) or is there an easier approach to finding Binomial Coefficients of fractional power without derivatives?

P.S. This is my first time writing in LaTeX, so I appologize if any formatting is off

  • Maybe the proof here is useful? https://en.wikipedia.org/wiki/Binomial_series – Erik Tellgren Aug 07 '24 at 19:38
  • I'm having a hard time understanding the proof. For now, I'm trying to focus on binomial coefficients for real numbers, and I'm afraid that that proof in the wiki may rely on lemma that includes derivatives. My fear is probably not based on much (mostly because I don't understand what it's saying), so clarification would be appeciated – MathEnthusiastThatsBadAtMath Aug 07 '24 at 19:50
  • https://math.stackexchange.com/questions/1789815/how-to-solve-lim-x-rightarrow-0-frac1xa-1x – Sine of the Time Aug 07 '24 at 20:04
  • I was looking at the link (to the other stack exchange forum) that you sent and it doesn't really explain the Generalized Binominal Theorem. In the replies, the author says they'd just prove the Generalized Binomial Theorem with L'Hopitals, which doesn't help. Perhaps I've misinterpreted that forum --- if so, let me know. – MathEnthusiastThatsBadAtMath Aug 07 '24 at 20:15
  • This is extremely tedious to do without derivatives. Why do you want to avoid derivatives? – Qiaochu Yuan Aug 07 '24 at 22:35

5 Answers5

2

Suppose that $\frac{1}{a} = \frac{p}{q} > 0$ is a rational number, i.e. $p,q>0$ are integers. Suppose that

$\sqrt[a]{1+x} = (1+x)^{p/q} = c_0 + c_1 x + c_2 x^2 + \ldots$

Raising both sides to the power $q$ yields

$(1+x)^p = (c_0 + c_1 x + \ldots)^q$

and by truncating to the relevant order and using the ordinary binomial theorem we find

$1 + \binom{p}{1} x + \binom{p}{2} x^2 + \ldots = c_0^q + c_0^{q-1} c_1 \binom{q}{1} x + c_0^{q-2} c_1^2 \binom{q}{2} + c_0^{q-1} c_2 \binom{q}{1} x^2 + \ldots$

where the terms indicated by '$\ldots$' are all cubic or higher. Hence, by comparing terms, we obtain

$c_0 = 1$,

$c_1 = p/q$, and

$c_2 = \frac{1}{q} \binom{p}{2} - \frac{1}{q} c_1^2 \binom{q}{2} = \frac{p}{2q} \left(\frac{p}{q} - 1\right)$.

One can in principle go to higher orders, but in practice it is increasingly tedious.

Finally, to say something about the case when $\frac{1}{a}$ is irrational, note that $g(a;x) = (1+x)^{1/a}$ is a continuous function of $a$. Hence, the expression derived above for rational $1/a>0$ must also apply for irrational $a$, just with $1/a$ substituted for the ratio $p/q$.

  • Why the choice $a/2$? The OP is considering $1/a$. – Gary Aug 07 '24 at 23:15
  • Ugh, a lapse on my part. Should have been $1/a = p/q$ to connect to the question. As it is now, my $a/2$ is MathEnthusiast's $1/a$. – Erik Tellgren Aug 07 '24 at 23:20
  • 1
    @Gary now edited to make it $1/a=p/q$ – Erik Tellgren Aug 08 '24 at 08:01
  • I like the work you did for when $\frac 1 a$ is a rational number, but I would like to make sure I'm understanding the last part about when $\frac 1 a$ is irrational. When you say that g(a,x) is a continuous function of $a$, are you implying that bounding the limit with the squeeze theorem (using two rational functions) is the most convinient way to evaluate each of the limits for the coefficients? – MathEnthusiastThatsBadAtMath Aug 08 '24 at 08:58
  • @MathEnthusiastThatsBadAtMath There are probably different ways of expressing it. I had in mind that an irrational $1/a$ can be approximated arbitrarily close by a rational number $p/q$ and therefore $(1+x)^{1/a}$ can also be approximated arbitrarily close by $(1+x)^{p/q}$. You can also make arbitrarily close lower and upper approximations $p_1/q_1 < 1/a < p_2/q_2$, and $(1+x)^{p_1/q_1} < (1+a)^{1/a} < (1+x)^{p_2/q_2}$, if that's easier to follow. – Erik Tellgren Aug 10 '24 at 09:28
1

We'll do $n=2.$ $$\begin{align} \sqrt{1+x}-1&=\sqrt{1+x}-(1+x/2)+x/2 \\&=-\frac{-x^2}{\sqrt{1+x}+1+x/2} +\frac{x}2. \end{align}\tag1$$

The second step you are using $a-b=\frac{a^2-b^2}{a+b},$ with $a=\sqrt{1+x},b=1+x/2.$

Dividing the right side of $(1)$ by $x$ and letting $x\to 0,$ you get a limit of $\frac12.$

For general $n,$ we use $$a-b=\frac{a^n-b^n}{a^{n-1}+a^{n-2}b+\cdots+b^{n-1}}.$$

But each individual step after this is harder and harder, and this assumes you know the value $c_1$ already.

The next term you want $c_2$ such that $(1+x/n+c_2x^2)^n$ to start with $1+x+0x^2+\cdots.$ That, in this case, means $nc_2+ \frac1{n^2}\binom{n}2 =0.$

Thomas Andrews
  • 186,215
  • For a = $\frac 3 2$, would this approach still work? I was looking at the part where you said "For general $n$ we use..." and I was curious if that formula would hold for non-integer $n$ (assuming $n$ is positive) – MathEnthusiastThatsBadAtMath Aug 07 '24 at 20:05
  • What do you mean? We use $a=\sqrt[n]{1+x}$ and $b=1+x/n$ for the general $n.$ Why would we use a constant? @MathEnthusiastThatsBadAtMath For later terms, it gets harder – Thomas Andrews Aug 07 '24 at 20:33
  • I mean if we were to do $\sqrt[\frac 3 2]{1+x}$, cause then the statement about $a-b = \frac {a^n - b^n} {a^{n-1} + a^{a-2}b + ... + b^{n-1}}$ wouldn't be true – MathEnthusiastThatsBadAtMath Aug 07 '24 at 20:38
  • It will be harder. But we don't write $\sqrt[n]{1+x}$ for $n$ not a. positive integer. We just write $(1+x)^{1/n}.$ So I assumed that was what you wanted. @MathEnthusiastThatsBadAtMath – Thomas Andrews Aug 07 '24 at 20:42
  • But if $n=3/2,$ we can instead expand $\sqrt[3]{1+2x+x^2}.$ But for $n=\sqrt 2,$ it will be harder. – Thomas Andrews Aug 07 '24 at 20:44
  • So for something like $n = \sqrt{2}$, would we use a fraction (with a lot of digits in the numerator and denominator) to approximate $\sqrt{2}$ and then follow the same methodology? Or is there something else that's more "mathematically pure" – MathEnthusiastThatsBadAtMath Aug 07 '24 at 20:48
  • Ultimately, it is an artificial condition to say you don't want to use derivatives. All the terms are derivatives. All the terms of the series, for any non-zero real $n,$ can be gotten if you know the derivative of $f(x)=x^a$ where $a$ is a real number. Then you get $f^{(k)}(1)=a(a-1)\cdots(a-k+1).$ – Thomas Andrews Aug 07 '24 at 20:48
  • Where's 44? I was actually hoping to use a Generalized Binomial Theorem to prove the integral power rule for real powers only using Riemann sums (as to mimic what mathematicians first tried), so I was hoping to avoid derivatives and the FTC – MathEnthusiastThatsBadAtMath Aug 07 '24 at 20:51
  • If you know the generalized binomial theorem, that is proved using derivatives. The generalized binomial theorem makes this a trivial problem, so I assumed you didn't know it. Also, what do you mean by "Where's 44?" @MathEnthusiastThatsBadAtMath – Thomas Andrews Aug 07 '24 at 20:54
  • I was hoping to find a version of the Generalized Binomial Theorem without derivatives so I could calculate integrals of $x^n, n \epsilon \mathbb{R}$ without derivatives. I thought I saw a "44" in your last comment and I didn't know why, but it's not there anymore – MathEnthusiastThatsBadAtMath Aug 07 '24 at 20:57
0

Let us show that $$\lim_{x \to 0} \frac{(x+1)^a - 1}{x} = a$$ without using L'Hospital.

If $a \geq 1$ use Bernoulli's inequality for real exponent. Let us concentrate on case where $x > 0$. (Case x < 0 is symmetric).

We have $\frac{(x+1)^a - 1}{x} \geq \frac{1 + ax - 1}{x} = a$ and, on the other hand, use famous inequality $(1+x)^a \leq 1 + ax + \frac{a(a-1)}{2}x^2$ (which holds for all $x > 0$ and $a\geq1$) to obtain that $\frac{(x+1)^a - 1}{x} \leq a + \frac{a(a-1)}{2}x \to a, x \to 0+$.

If $0 < a < 1$. Analogously, consider only $x > 0$.

In this case we use Bernoulli's inequality for real exponent $0<a \leq 1$ and obtain $\frac{(x+1)^a - 1}{x} \leq \frac{1 + ax - 1}{x} = a$

And use inequality $(1+x)^a \geq 1 + ax + \frac{a(a-1)}{2}x^2$ (which holds for all $x > 0$ and $0<a\leq1$) to get $\frac{(x+1)^a - 1}{x} \geq a + \frac{a(a-1)}{2}x \to a, x \to 0+$

  • I was looking at the wiki page for Bernoulli's inequality, and it says to generalize the inequality for decimal number $a$ that are between 0 and 1, we need to compare derivatives. I'm trying not to solve this with derivatives, so if I missed something let me know. Wiki link: https://en.wikipedia.org/wiki/Bernoulli%27s_inequality – MathEnthusiastThatsBadAtMath Aug 07 '24 at 19:56
  • Yes, you are right. All the inequalities I used are in fact can be proven using derivatives(and it is the easiest way to prove them). I did not understand your question properly :(

    I suppose these inequalities may be proven without derivatives, I will think about it.

    – Oleksandr Liubimov Aug 07 '24 at 19:59
0

Let $y:=x+1$ and $z:=y^{p/q}$ for conciseness.

$$\frac{z-1}{x}=\frac{z-1}{x}\frac{z^{q-1}+z^{q-2}+\cdots 1}{z^{q-1}+z^{q-2}+\cdots 1}=\frac{z^q-1}{x(z^{q-1}+z^{q-2}+\cdots 1)}=\frac{y^p-1}{x(z^{q-1}+z^{q-2}+\cdots 1)}=\frac{x(y^{p-1}+y^{p-2}+\cdots 1)}{x(z^{q-1}+z^{q-2}+\cdots 1)}.$$

Now if you let $x$ vanish, $y$ and $z$ tend to $1$ and the ratio to $\dfrac pq$.

  • I'm having a hard time understanding the jump from $\frac {y^p-1} {z(z^{q-1}+z^{q-2}+...1)}$ to $\frac {x(y^{p-1}+y^{p-2}+...1)} {z(z^{q-1}+z^{q-2}+...1)}$. After that step, I understand the logic of canceling $x$ and counting the amount of time 1 is added to itself in the numerator and denominator (which is $p$ and $q$ respectively).

    Unrelated, but when I tried finding the integral of a polynomial without the FTC, I used the logic your presented in https://www.codeproject.com/Tips/792255/Faulhaber-made-easy (Thank you for that). I was hoping that by proving the...

    – MathEnthusiastThatsBadAtMath Aug 08 '24 at 09:08
  • ...maclaurin series of $\sqrt[a]{x+1}$ I could somehow prove the integral power rule for all non-negative powers without derivatives or the FTC. – MathEnthusiastThatsBadAtMath Aug 08 '24 at 09:11
  • @MathEnthusiastThatsBadAtMath: $x$ is just $y-1$ and the factorization of the numerator is standard. –  Aug 08 '24 at 09:43
  • @MathEnthusiastThatsBadAtMath: Faulhaber is a nice problem, there are several ways to address it. Unfortunately, it does not generalize easily to non-integer powers. You have to resort to the Euler–Maclaurin summation. –  Aug 08 '24 at 09:50
0

By $y=\exp(\frac1a\ln(x+1))$ and well-known Maclaurin series, we have $$y=\sum_{m=0}^\infty\frac1{a^mm!}\left(\sum_{n=1}^\infty\frac{(-1)^{n+1}x^n}{n}\right)^m =1+\frac1a\sum_{n=1}^\infty\frac{(-1)^{n+1}x^n}n+\frac1{2a^2} \left(\sum_{n=1}^\infty\frac{(-1)^{n+1}x^n}n\right)^2+... $$

Bob Dobbs
  • 15,712
  • While the series of $e^x$ is pretty easily found without derivatives (by using the limit definition of $e$), I don't know if there's a way to find the maclaurin series of ln(x +1) without the use of derivatives (which I would like to avoid) – MathEnthusiastThatsBadAtMath Aug 08 '24 at 18:46
  • https://mathworld.wolfram.com/SeriesReversion.html @MathEnthusiastThatsBadAtMath – Bob Dobbs Aug 08 '24 at 19:14
  • Cool, I looked at it and I think I will fiddle around with it a little bit. This seems to be a pretty good approach because finding the maclaruin series of $e^x$ is easy for all real numbers using the limit definiion of $e$. Once I explore it I'll probably give this answer the green checkmark – MathEnthusiastThatsBadAtMath Aug 10 '24 at 06:45
  • @MathEnthusiastThatsBadAtMath But, binomial expansion is cooler. Link is given in the comments. – Bob Dobbs Aug 10 '24 at 12:04