8
$\begingroup$

Someone asked me for a formula for the sum of the harmonic progression. So I did some calculations and gave him an approximate formula:

$$\int_1^n\frac{dx}{x} = \frac{y_1 + y_2}{2} + \frac{y_2 + y_3}{2} + \cdots +\frac{y_{n-1} + y_n}{2}$$ where $y_i$ is $i$th term of the HP $$\ln(n) = \frac{y_1}{2} + y_2 +y_3 + \cdots +\frac{y_n}{2}$$

so

$$\sum_{i=1}^n y_i = \ln(n) + \frac{y_1 + y_n}{2}$$

e.g. $1+1/2+\cdots+1/10 = 2.8525 $

actual result $= 2.9289$

My question is, how to correct this formula?

|cite|improve this question
$\endgroup$
15
$\begingroup$

The partial sums of the harmonic series are called "harmonic numbers." The difference between the nth harmonic number and ln(n) tends to a limit as n increases, and that limit is called Euler's constant or gamma.

There's a great book about all this called Gamma: Exploring Euler's Constant.

|cite|improve this answer
$\endgroup$
6
$\begingroup$

A good quick approximation: $\log(n + 0.5) + \gamma$.

A yet more accurate approximation: $\log(n)+\gamma+1/2n + 1/12n^2-1/120n^4+1/252n^6-1/240n^8+1/132n^{10}-691/32760n^{12}+\cdots$, taking as many terms as is convenient.

Error analysis:

The first formula above has maximum error of $1/24n^2$. uday's first answer has error of about 0.0024 ln n. Mine is a better approximation for all $n\neq6$.

The first formula is also better than uday's revised formula for all $n\ge46$. The expected error in uday's answer is about 0.000016.

|cite|improve this answer
$\endgroup$
1
$\begingroup$

Hey guys.... There is no correct simple general formula for sum to n terms of the series 1+1/2+1/3+1/4+ ............. + 1/n

but the following formula will be a good approximation for sum to n-terms of the above series when n>5

S = log(n+0.5) + 0.5772 + 0.04021/(n*n + 0.8848)

Deviation from the actual value fluctuates but remains relatively low..

so i guess this may be a good approximation

|cite|improve this answer
$\endgroup$
  • $\begingroup$ it would be interesting to see how you found these numbers if they are not just experimentally determined. $\endgroup$ – anon Oct 13 '10 at 17:55
  • 1
    $\begingroup$ @muad: To get close to the best approximation with constants in those places, 1.841 should be about $\exp(\gamma)$, 0.9973 should be about 1, and 0.6184 should be about 1/2. $\endgroup$ – Charles Oct 13 '10 at 18:19
  • 1
    $\begingroup$ @muad: Hi... i have taken values of n from 1 to 1000000 and found the values of S. Using curvefitting tool in matlab i came to that formula... $\endgroup$ – uday kiran Oct 20 '10 at 11:21
  • 1
    $\begingroup$ @charls: Hi... i totally agree with you but for smaller values of n,log(n+0.5)+γ has significant amount of error.. so for about n=30 my formula will give a better approximation... $\endgroup$ – uday kiran Oct 20 '10 at 11:25
  • $\begingroup$ @uday kiran: Your new formula is worse than my simple formula for all n > 45. $\endgroup$ – Charles Oct 24 '10 at 23:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.