1

$$\lim_{x\to 0} \frac{\sin(\tan x)-\tan(\sin x)}{2 x\cos(\tan x)-2 x\cos(\sin x)+x^5}$$

I can not solve this limit without using series, and L'Hopital's Rule.

Jean Marie
  • 81,803
abdel
  • 29
  • If you don't use English properly and don't write what you have tried, I'm pretty sure that this question will be put on-hold. – Our Aug 18 '16 at 16:39
  • I don't understand. Do you mean you can't solve this using infinite series representations or l'hopital's rule? Or you aren't allowed to? – Math Aug 18 '16 at 16:40
  • with L'Hospital we get $$0$$ as the searched limit – Dr. Sonnhard Graubner Aug 18 '16 at 16:59
  • without L Hopital – abdel Aug 18 '16 at 17:04
  • please I want the solution without using series and also without using L Hopital rule – abdel Aug 18 '16 at 17:09
  • 3
    For reference: 1) Mathematica returns a result of 1/5 (not zero). 2) To get this limit using L'Hopital requires a total of seven instances of it in order to get a limit which isn't indeterminate. So that's not a great strategy here in any case... – Semiclassical Aug 18 '16 at 17:31
  • The reason it's tough, btw, is because both the numerator and denominator vanish like $x^7$ as $x\to 0$ due to cancellation of lower order terms. So finding a simple proof is likely to be hard.... – Semiclassical Aug 18 '16 at 17:40
  • This limit equals $\frac{1}{5}$. – Jan Eerland Aug 18 '16 at 17:58
  • i have found my error the searched limit is $$\frac{1}{5}$$ – Dr. Sonnhard Graubner Aug 18 '16 at 18:13
  • see http://math.stackexchange.com/q/741446/72031 – Paramanand Singh Aug 18 '16 at 18:51
  • It is not clear what you want to ask? Do you wish to evaluate the limit without L'Hospital or Taylor's series. If that is what you wish it is not possible. Using Taylor's series gives the answer easily. With L'Hospital you need a bit of effort. The denominator is easy to handle BTW. – Paramanand Singh Aug 18 '16 at 18:54
  • I think the best approach to this would be to first go back to where you proved the limit using l'Hôpital's rule (or do so if you haven't), and where you use it, try to "cut through" using the Newton quotient, since, after all, that's where the rule comes from. I'll post my own attempt if no one comes up with a better answer soon. – user361424 Aug 24 '16 at 22:34

1 Answers1

5

The beautiful thing about math is that there is no "solution with" or "solution without," only the solution, and various concordant ways of getting there. There's a technique called "eliminating cuts," mostly (this limits-without-lhopital meme notwithstanding) only useful for finding results about proofs themselves, where essentially you take a particular intermediate step, and wherever you rely on it, replace it with its proof. My first impulse was to do this with l'Hôpital's rule, first proving it with, and then using the proof to get from step to step without. I decided against this because the derivatives get very messy very quickly, and it would take seven of them to get to the point where the quotient isn't $0/0$, and the proof itself gets significantly messier if even two are needed. Taylor's theorem is essentially a consequence of l'Hôpital's rule (which, I expect, is why you don't want to use Taylor series), so an attempt to go that route without explicitly invoking Taylor series would probably get just as messy. Fortunately, however, the series of the trig functions predate Taylor's theorem by quite a long way, with Newton having found, for instance, the sine's by rearranging terms in the binomial expansion of $\sin^{-1}x = \int_0^x(1-t^2)^{-1/2}dt$. So rather than use infinite series, we'll use the same logic behind the series to come to polynomials that serve as upper and lower bounds.

So just to recap:

$$\sin^{-1}x = \int_0^x\frac{dt}{\sqrt{1-t^2}}$$ $$\tan^{-1}x = \int_0^x\frac{dt}{1+t^2}$$

These follow pretty quickly from the Cartesian equation for a semicircle and the definition of arclength as $\int_a^b\sqrt{1+(f'(x))^2}dx$ (the second isn't totally trivial, but I expect you already know these, so I won't dwell on them). You've probably noticed the absence of the arccosine. The reason for that is that both the cosines in the limit go to 1, which makes upper bounds slightly harder to establish. So let's get rid of them, rewriting the limit as

$$\lim_{x\to 0} \frac{\sin(\tan x)-\tan(\sin x)}{2x\left(\sqrt{1-\sin^2(\tan x)}-\sqrt{1-\sin^2(\sin x)}\right)+x^5}$$

Now, to break down the arctangent, we'll go right back to the very, very beginning - Elements IX.35 (the page says 36 for some reason). "If as many numbers as we please are in continued proportion, and there is subtracted from the second and last numbers equal to the first, then the excess of the second is to the first as the excess of the last is to the sum of all those before it." In modern notation (you have no idea how tempted I was to write this entire answer in the language of classical geometry), $\frac{ar-a}{a} = \frac{ar^{n+1}-a}{\sum_{k=0}^n ar^k}$, more commonly expressed

$$\sum_{k=0}^n r^k = \frac{1-r^{n+1}}{1-r}$$

Let $r = -t^2$, $n$ odd, and we've got a lower bound for $(1+t^2)^{-1}$, and as long as the value we're looking for is reasonably small (below about $17/20$ - recall Achilles and the tortoise), we can get an upper bound by doubling the last two terms. (Note that everywhere the tangent appears in the original limit it approaches zero.)

The arcsine will take a bit more finesse, but this technique can be used to put upper and lower bounds on the square of the integrand, from which a polynomial can be found whose square will be more or be less in the range in question. This time $r = +t^2$, $n$ need not be odd, and only the last term needs to be doubled provided that $t < 7/10$. Note that in both the original and the rewritten limit, the sine also always approaches zero.

So now we have these inequalities.

$$\int_0^{|x|}(1-t^2+t^4-t^6+t^8-t^{10})dt < |\tan^{-1}x| < \int_0^{|x|}(1-t^2+t^4-t^6+2t^8-2t^{10})dt$$ $$\int_0^{|x|}\left(1+\frac{t^2}{2}+\frac{3t^4}{8}+\frac{5t^6}{16}+\frac{35t^8}{128}\right)dt < \int_0^{|x|}\sqrt{1+t^2+t^4+t^6+t^8}dt < |\sin^{-1}x| < \int_0^{|x|}\sqrt{1+t^2+t^4+t^6+2t^8}dt < \int_0^{|x|}\left(1+\frac{t^2}{2}+\frac{3t^4}{8}+\frac{5t^6}{16}+\frac{99t^8}{128}\right)dt$$

The absolute value signs are needed in case $x$ is negative. You can verify for yourself that in the latter case the squares of the outermost integrands are sufficiently high and low, respectively. Evaluate the integrals; because the sine, tangent, and the outermost functions of the above inequalities are all odd functions, the absolute value signs can be eliminated by recognizing that negating x will only reverse the direction of the inequality, and since this range is symmetric, rather than explicit inequalities we'll henceforth be using an error term $\epsilon > 0$.

$$\tan^{-1}x = x-\frac{x^3}{3}+\frac{x^5}{5}-\frac{x^7}{7}\pm\left[\epsilon < 2\left|\frac{x^9}{9}-\frac{x^{11}}{11}\right|\right]$$ $$\sin^{-1}x = x+\frac{x^3}{6}+\frac{3x^5}{40}+\frac{5x^7}{112}\pm\left[\epsilon < \left|\frac{11x^9}{128}\right|\right]$$

So from this you can get polynomials to bound the sine and tangent themselves. Since both the sine and tangent are increasing at zero, the lower bounds of the functions can be found from the lower bounds of the inverse and vice versa.

For sufficiently small $x$ and $y$ (defined here as small enough that the highest-degree terms can be consolidated - remember, these are finite, so there will always be such a range),

$$y = x-\frac{x^3}{3}+\frac{x^5}{5}-\frac{x^7}{7}\pm\left[\epsilon < 2\left|\frac{x^9}{9}-\frac{x^{11}}{11}\right|\right] \implies x=y+\frac{y^3}{3}+\frac{2y^5}{15}+\frac{17y^7}{315}\pm[\epsilon<|cy^9|]$$ $$y = x+\frac{x^3}{6}+\frac{3x^5}{40}+\frac{5x^7}{112}\pm\left[\epsilon < \left|\frac{11x^9}{128}\right|\right] \implies x = y-\frac{y^3}{6}+\frac{y^5}{120}-\frac{y^7}{5040}\pm[\epsilon<|cy^9|]$$

For some $c$ independent of $x$. You can convince yourself of these with high school algebra if you really must, but I warn you it'll be tedious. So first, the numerator. This'll look familiar - it's exactly as if we'd used Taylor's theorem.

$$\sin(\tan x) = x + \left(\frac{1}{3}-\frac{1}{6}\right)x^3 + \left(\frac{2}{15}-\frac{1}{6}+\frac{1}{120}\right)x^5 + \left(\frac{17}{315}-\frac{11}{90}+\frac{1}{72}-\frac{1}{5040}\right)x^7 \pm [\epsilon<|cx^9|] = x+\frac{x^3}{6}-\frac{x^5}{40}-\frac{55x^7}{1008}\pm [\epsilon<|cx^9|]$$ $$\tan(\sin x) = x + \left(-\frac{1}{6}+\frac{1}{3}\right)x^3 + \left(\frac{1}{120}-\frac{1}{6}+\frac{2}{15}\right)x^5 + \left(-\frac{1}{5040}+\frac{13}{360}-\frac{1}{9}+\frac{17}{315}\right)x^7 \pm [\epsilon<|cx^9|] = x+\frac{x^3}{6}-\frac{x^5}{40}-\frac{107x^7}{5040} \pm [\epsilon<|cx^9|]$$ $$\sin(\tan x) - \tan(\sin x) = -\frac{x^7}{30} \pm [\epsilon<|cx^9|]$$

The denominator will be trickier. First, let's work out

$$\sin(\sin x) = x + \left(-\frac{1}{6}-\frac{1}{6}\right)x^3 + \left(\frac{1}{120}+\frac{1}{12}+\frac{1}{120}\right)x^5 + \left(-\frac{1}{5040}-\frac{13}{720}-\frac{1}{144}-\frac{1}{5040}\right)x^7 \pm [\epsilon<|cx^9|] = x - \frac{x^3}{3} + \frac{x^5}{10} - \frac{8x^7}{315} \pm [\epsilon<|cx^9|]$$

So

$$\sin^2(\sin x) = x^2 + \left(2\cdot-\frac{1}{3}\right)x^4 + \left(\frac{1}{9}+2\cdot\frac{1}{10}\right)x^6 \pm [\epsilon<|cx^8|] = x^2 - \frac{2}{3}x^4 + \frac{14}{45}x^6 \pm [\epsilon<|cx^8|]$$ $$\sin^2(\tan x) = x^2 + \left(2\cdot\frac{1}{6}\right)x^4 + \left(\frac{1}{36}+2\cdot-\frac{1}{40}\right)x^6 \pm [\epsilon<|cx^8|] = x^2 + \frac{1}{3}x^4 - \frac{1}{45}x^6 \pm [\epsilon<|cx^8|]$$

And from there

$$1 - \sin^2(\sin x) = 1 - x^2 + \frac{2}{3}x^4 - \frac{14}{45}x^6 \pm [\epsilon<|cx^8|] = \left(1 - \frac{1}{2}x^2 + \frac{5}{24}x^4 - \frac{37}{720}x^6 \pm [\epsilon<|cx^8|]\right)^2$$ $$1 - \sin^2(\tan x) = 1 - x^2 - \frac{1}{3}x^4 + \frac{1}{45}x^6 \pm [\epsilon<|cx^8|] = \left(1 - \frac{1}{2}x^2 - \frac{7}{24}x^4 - \frac{97}{720}x^6 \pm [\epsilon<|cx^8|]\right)^2$$ $$2x\left(\sqrt{1-\sin^2(\tan x)}-\sqrt{1-\sin^2(\sin x)}\right)+x^5 = x^5 + \left(2x - x^3 - \frac{7}{12}x^5 - \frac{97}{360}x^7\right) - \left(2x - x^3 + \frac{5}{12}x^5 - \frac{37}{360}x^7\right) \pm [\epsilon<|cx^9|] = x^5 - x^5 - \frac{1}{6}x^7 \pm [\epsilon<|cx^9|]$$

So you end up with

$$\lim_{x\to0}\frac{-\frac{1}{30}x^7 \pm [\epsilon<|cx^9|]}{-\frac{1}{6}x^7 \pm [\epsilon<|cx^9|]} = \frac{1}{5}$$

user361424
  • 1,719
  • There's nothing wrong mathematically with l'Hôpital's rule, only that it often obscures interesting patterns, or leads to circular arguments by students who don't fully understand where their derivatives come from. Of course, just like anywhere else, you can use the proof of l'Hôpital's rule to avoid it, but the proof of Taylor's theorem using it is so elegant I can't think why you'd want to. – user361424 Aug 31 '16 at 03:21
  • Could you provide a link to that proof? –  Aug 31 '16 at 03:30
  • https://en.wikipedia.org/wiki/Taylor%27s_theorem#Proof_for_Taylor.27s_theorem_in_one_real_variable – user361424 Aug 31 '16 at 03:33
  • Brilliant and instructive answer. – rubik Aug 31 '16 at 06:50
  • @selfawareuser: A version of Taylor's Theorem (see http://math.stackexchange.com/q/1809293/72031) is provable using L'Hospital's Rule. But the usual Taylor's theorem with Lagrange's and Cauchy's form of remainder is a consequence of Rolle's Theorem (in that sense L'Hospital's Rule is also a consequence of Rolle's Theorem). – Paramanand Singh Aug 31 '16 at 09:01