16

The inequality $$\boxed{e^3 > 20}$$ is occasionally useful, including in the answer I wrote for this question that comes from a GRE subject exam.

This bound is relatively tight: $$e^3 = 20.08553\!\ldots ,$$ a relative error of $< \frac{1}{200}$, which means establishing the inequality might be a little delicate. In a comment under the linked answer, TheSimpliFire posed the following natural question:

What is an efficient way to prove the inequality $e^3 > 20$ by hand?

(I would have guessed that this had been asked before, but neither the internal search nor searchonmath turned up any duplicates.)

A naive method is to use the series truncation $$e = \sum_{k = 0}^\infty \frac{1}{k!} > 1 + 1 + \frac{1}{2} + \frac{1}{6} + \frac{1}{24} + \frac{1}{120} = \frac{120 + 120 + 60 + 20 + 5 + 1}{160} = \frac{163}{60} .$$ Then, it suffices to prove that $\left(\frac{163}{60}\right)^3 > 20$, which is equivalent to $4\,330\,747 > 4\,320\,000$. This last step could even be outsourced to an enthusiastic primary school student, but it involves cubing a three-digit prime and so is slightly tedious.

One might try to refine this method by looking for rationals that are easier to cube, but the only rational numbers satisfying $\sqrt[3]{20} < q < e$ with denominator $< 60$ are $\frac{106}{39}, \frac{125}{46}, \frac{144}{53}$. It's again straightforward to show that the cube of any of these $> 20$, but doing so is no faster than cubing $\frac{163}{60}$ and one then has the additional burden of showing the number is $< e$.

One could also search for integrals analogous to the classic Dalzell integrals for the difference $e^3 - 20$ (or to the difference corresponding to some other inequality equivalent thereto), by which I mean evidently positive definite integrals equal to that difference.

For example, some experimentation yields the definite integral \begin{align} &\int_1^2 - \frac{(x - 1) (2 - x) p(x) \,dx}{20 x (x^2 + 1)} \\ &\qquad = \int_1^2 \left(-\frac{1}{2} x^3 + \frac{63}{20} x^2 - \frac{153}{20} x + 9 - \frac{3}{x} - \frac{2 x}{x^2 + 1} \right) dx \\ &\qquad = 3 - \log 20 , \end{align} where $p(x) = 10 x^4 - 33 x^3 + 44 x^2 - 45 x + 30$. Computing gives that all of the coefficients of $p(x + 1)$ are positive, so $p$ is strictly positive for $x \geq 1$, and thus the integrand is strictly positive on $(1, 2)$. So, the integral is positive, that is, $3 > \log 20$, which is equivalent via exponentiation to $e^3 > 20$. This is again elementary, but not terribly fast.

Remark Incidentally this latter method lets us extract cheap but relatively sharp rational bounds on $\log 20$: Since $2 < x (x^2 + 1) < 10$ on the interval of integration, our integral is bounded by polynomial integrals: $$\int_1^2 - \frac{(x - 1) (2 - x) p(x) \,dx}{20 \cdot 10} < \int_1^2 - \frac{(x - 1) (2 - x) p(x) \,dx}{20 x (x^2 + 1)} < \int_1^2 - \frac{(x - 1) (2 - x) p(x) \,dx}{20 \cdot 2} .$$ Integrating gives $$\frac{163}{84000} < 3 - \log 20 < \frac{163}{16800},$$ and rearranging gives the bounds $$2.99027\!\ldots = \frac{251185}{84000} < \log 20 < \frac{251837}{84000} = 2.99805\!\ldots .$$

Travis Willse
  • 99,363
  • 1
    What about calculating enough terms of the series $e^3 = \sum_{k=0}^\infty \frac{3^k}{k!}$ to exceed 20? Though that does seem to require up to $k=8$. – Daniel Schepler Aug 26 '19 at 17:43
  • 3
    I think you mean $(\frac{163}{60})^3 \gt 20$ – Ross Millikan Aug 26 '19 at 17:43
  • 1
    The silliest way is to compute $(2.718)^2 \cdot 2.71$! – dcolazin Aug 26 '19 at 17:49
  • Yes, thanks, Ross, fixed! – Travis Willse Aug 26 '19 at 17:52
  • 1
    @dcolazin That certainly works, of course, but aesthetically I'd prefer methods that use definitions rather than decimal approximations. Of course one can recover $e > 2.718$ as an intermediate step using, for example, the power series for $e^x$ at $x = 1$, but doing so already requires $7$ terms, which is more than the $6$-term calculation I'm looking to improve on. For that matter, $2.718 \cdot 2.718 \cdot 2.71$ certainly requires more single-digit multiplications and carries than $163 \cdot 163 \cdot 163$. – Travis Willse Aug 26 '19 at 18:10
  • @Travis oh, you are so right! – dcolazin Aug 27 '19 at 07:37

5 Answers5

8

Similar to your last proof, I found a positive function whose integral is $e^3-20$.

For $$f(x)=\frac{1}{186}(x-1)^2(x-2)^4e^x\ge0$$

we have $$\int_{0}^{3}f(x)dx=e^3-20$$

rrainbow
  • 96
6

If you know your powers of $3$ well, you know $2.7^3=19.683$. Since $e>2.718=2.7\left(1+\frac{2}{300}\right)$,$$e^3>19.683\left(1+\frac{2}{100}\right)=19.683+0.39366>20.$$

J.G.
  • 115,835
1

$$1+3+\frac92+\frac92+\frac{27}8+\frac{81}{40}+\frac{81}{80}+\frac{243}{560}+\frac{729}{4480}\\ 13+3.375+2.025+1.025+0.433928\cdots+0.162723\cdots=20.021651$$

isn't so difficult. Only the last two term require a "true" division.

1

An extended comment.

Not really a proof, but an interesting consequence:

$$\log 20=4 \log 2+\log \left(1+\frac{1}{4}\right)<3$$

$$\log 2< \frac34 -\frac14 \log \left(1+\frac{1}{4}\right) $$

$$\log 2< \frac34 -\frac14 \left(\frac{1}{4}-\frac{1}{32}\right) $$

$$\log 2< \frac34 -\frac1{18} $$

The error here is approximately $0.0013$.

That said, there's a lot of inequalities for logarithms, especially for $\log 2$ already known. This can be used to prove the OP.

Yuriy S
  • 31,474
  • 1
    @Travis, at least $\log 2 < \frac{5^2}{6^2}$ is easy to remember. However, the original inequality doesn't follow from this, because $8^{36/25}<20$. – Yuriy S Aug 26 '19 at 19:11
0

How about you simply use the Taylor series for $e^x$? $$e^x = \sum_{r=0}^\infty \frac {x^r}{r!} \\ e^3 \gt 1+3/1+9/2+9/2+27/8+ 81/40+81/80+243/560+(243/560)*3/8 + ((243*3)/(560*8))*1/3 = 1+3+9+3.375+2.025+1.0125+0.4339..+0.1627...+0.0542... = 20.0633 \gt 20 $$ Basically add up the first ten terms of the Taylor series of e^x with x=3.

Green05
  • 157
  • 5