# How could we manually approximate $\sum_{i=1}^{50} i! = 3.1035 \times 10^{64}$?

How could we manually approximate $$\sum_{i=1}^{50} i!$$ to the value $3.1035 \times 10^{64}$?

I faced this question in my aptitude test,there were four option given,I couldn't solve it during the test,in home I used Stirling's approximation with wolfram's Mathematica to identify the correct option (if I did it right),however I am interested to know if there is any way we could do this entirely manually (probably using some tricks)?

PS:By manually I mean purely and only with pencil and paper.

$1)3.1035 \times 10^{65} \quad\quad 2) 3.1035 \times 10^{64} \quad\quad 3) 3.1035\times 10^{62} \quad\quad 4) 3.3339 \times 10^{62}$

-
Can we see what the other options were? –  Henning Makholm Sep 17 '11 at 19:51
@Henning Makholm:I have added them. –  Quixotic Sep 17 '11 at 19:56

The largest term in the sum is 50 times as large as the next largest, and $49\times50$ times as large as the second largest, and so forth. So to get just within a factor of 10, computing $50!$ alone ought to suffice.

But how manually is "manually" here? Is it purely pencil-and-paper, or do you have some way to compute logs and exponentials?

(Edit: Pencil and paper!) Here's a suggestion, assuming that one is a walking encyclopedia of mathematical trivia. With the given options it is (we hope) enough to compute $\log_{10} 50!$ to a precision of about $0.5$.

Assume that we remember Stirling's formula: $$\ln n! \approx \frac{\ln(2\pi)}2+(n+\frac12)\ln n - n$$ This gives $$\log n! = \frac{\log(2\pi)}2+(n+\frac 12)\log n - \frac{n}{\ln 10}$$

Assume that we know that $\log_{10} 5\approx 0.7$ and also assume that we know that $\ln 10\approx 2.3$. I don't think it is fair to assume we remember $\log_{10}(2\pi)$, but one easily sees it must be somewhere between 0.7 and 1, so let's conservatively set $\frac{\log(2\pi)}2\approx 0.4\pm 0.1$. We then get $$\log 50! \approx 0.4 \pm 0.1 + 50.5\times1.7 - \frac{50}{2.3}$$ which is pencil-and-paper tractable and works out to $$\log 50! \approx 64.55\pm 0.1$$ This should give some confidence in the $3\times10^{64}$ option .

-
purely pencil and paper! –  Quixotic Sep 17 '11 at 19:50
Since $\pi\cong\sqrt{10}$ is a pretty good approximation, you can estimate $\log 2\pi$ fairly closely as $\log 2\sqrt{10}\cong 0.30103+0.5\cong 0.8$; the rounding errors even partly cancel. (Doesn’t everyone know $\log 2$ to $5$ places? :-)) –  Brian M. Scott Sep 17 '11 at 21:38

Given those four choices, I can make a very good guess using only information that I have in memory and pencil-and-paper calculation.

$50 \ln 50 - 49 = \int\limits_1^{50}\ln x dx < \sum\limits_{n=1}^{50}\ln n < \int\limits_1^{51}\ln x dx = 51 \ln 51 - 50$. $\ln 10 \cong 2.3$, and $\log_{10} 5 \cong 0.7$, so $\ln 50 = \log_{10} 50 \ln 10 \cong 1.7\cdot 2.3 = 3.91$. $50\cdot 3.91 - 49 = 146.5$, and $51 \cdot 3.91 - 50 = 149.41$, so $\log_{10} 50!$ ought to be between about $\dfrac{146.5}{2.3} \cong 63.7$ and about $\dfrac{149.41}{2.3} \cong 65.0$. That pretty much pins it down to the second alternative.

(Had I not independently remembered that $\ln 10 \cong 2.3$, I could have got it from remembering that $e^3 \cong 20$ and $\ln 2 \cong 0.7$.)

-
Where did that very first 49 come from? –  tzs Sep 18 '11 at 3:49
@tzs: $\int_1^a\ln x dx=[x\ln x - x]_1^a=a\ln a - a+1$. –  Brian M. Scott Sep 18 '11 at 6:07

The final answer will be dominated by the largest term. Write the series as

$$50! + 49! +\cdots + 1!$$

$$50!(1+ \frac{1}{50}+ \frac{1}{49\cdot 50}+\cdots +\frac{1}{50!})$$

The numerical values of each of the subsequent terms decreases pretty quickly. So you can truncate the sum appropriately according to the number of decimal places you require. If you take just the first three terms, i.e $$\approx 50! \times 1.0204 = 3.1034 \times 10^{64}$$

You would need to calculate 50!, for which the only way I know is to use Stirling's formula. (In this test, were you provided logarithm tables?)

-
No we are likely to be provided with only a pencil and some scratch paper. –  Quixotic Sep 18 '11 at 19:30