I took $\frac{1}{x-a_{1}} + \frac{1}{x-a_{2}} + ... +\frac{1}{x-a_{n-1}}$, tried to write it with a common denominator, I thought it would yield something useful in the numerator, the denominator would just simplify to $\frac{x^n -1}{x-1}$, but the numerator doesn't give anything useful when evaluated, it keeps going in circles.
-
1Trying some values in WolframAlpha seems to yield $\sum_{k=0}^{n-1} \frac{1}{2-\exp(2ik\pi/n)}=\frac{n2^{n-1}}{2^n-1}$. I'll try to prove it and help you if I can. – Kolakoski54 Sep 25 '24 at 14:38
-
@Kolakoski54 I think a proof by induction could work – Amogh Sep 25 '24 at 14:48
-
1You can modify the numerous approaches that calculate $ \sum \frac{1}{ x_i1 - 1 } =-n/2 $ – Calvin Lin Sep 26 '24 at 00:12
-
FYI I added an approach to show how we can push through your idea without going around in circles. – Calvin Lin Sep 26 '24 at 05:53
2 Answers
To solve this question, we use $\textbf{log }$
Consider $f(x)=\log \prod_{i=1}^{n-1} (x-a_i)$ The problem is to find $f'(2)$
Now $a_i's$ are roots of $x^n-1$. so $x^n-1=(x-a_1)\cdots(x-a_{n-1})(x-1)$
so $\frac{x^n-1}{x-1}=\prod_i (x-a_i)$
Thus $$f(x)=\log \frac{x^n-1}{x-1}$$
From this we can easily find $f'(2)$.
- 11,989
-
Side note: Slightly easier to set $g(x) = \log (x^n -1 ) $ and find $g'(x) = \frac{nx^{n-1}}{x^n-1}$ by chain rule, then subtract off $ 1/(2 - a_n)$. – Calvin Lin Sep 26 '24 at 00:02
Standard approach via root transformation:
Let's tweak it slightly to find $\sum_{i=1}^n \frac{1}{ 2 - a_i }$ with $a_n = 1$, then subtract $\frac{ 1}{ 2 - a_n } = 1 $ later.
The roots of $ x^n - 1$ are $\{ a_i \}_{i=1}^n$.
Let $ y = \frac{1}{2-x}$, then $ x = 2 - \frac{1}{y} = \frac{ 2y - 1 } {y}$.
So the roots of $ ( \frac{2y-1}{y} )^n - 1 = 0 \Rightarrow (2y-1)^n - y^n = 0 $ are $ \{ \frac{1}{2-a_i } \}_{i=1}^{n}$.
By Vieta, the sum of roots is $ \frac{ (-1) \text{ coefficient of } y^{n-1} } { \text{coefficient of } y^n } = \frac{ - {n\choose 1 } 2^{n-1}(-1) } { 2^n - 1 } = \frac{ n2^{n-1} } { 2^n-1 } $ (per Kolakoski54's comment).
Finally, subtract 1 and we get $ \frac{ (n-2)2^{n-1} + 1 } { 2^n - 1 } $.
Bonus: Can you use this approach to find $\sum_{i=1}^{n-1} \frac{ 1}{ 2-a_i} $ directly with my tweak of adding $ \frac { 1 } { 2 - a_n } $?
How do we modify $ \frac{ x^n - 1 } { x - 1 } $ to get at the desired expression in terms of $y$?
We can generalize the above approach to show that for $ k ^n \neq 1$, $ \sum_{i=1}^{n-1} \frac{1}{ k - a_i } = \frac{nk^{n-1} } {k^n - 1 } - \frac{ 1 } { k - 1 } $. So, can we push through OP's idea?
Alternative solution via OP's approach:
Again, we add back the $a_n$ term and look at
$$ \sum_{i=1}^n \frac{1}{k - a_i } = \frac{ \sum_{i=1}^n \prod_{j\neq i } ( k - x_j) } { k^n - 1 }. $$
What do we get in the numerator when we expand it? Let's track the coefficients:
- $k^{n-1} $ term: $\prod_{j\neq i } ( k - x_j)$ will contribute $1$. Summing this across $i$, we get $ \sum_i 1 = n $.
- $k^{n-2} $ term: $\prod_{j\neq i } ( k - x_j)$ will contribute $ \sum_{j\neq i } - x_j $. Summing this across $i$, we get $ (n-1 ) \sum x_i = 0$.
- $k^{n-3} $ term: $\prod_{j\neq i } ( k - x_j)$ will contribute $ \sum_{j\neq i, l \neq i } (-x_j)(- x_l) $. Summing this across $i$, we get $ { n-1 \choose 2 } \sum_{j\neq l} x_j x_l = 0 $.
- $ \vdots$
- $k^0$ term: $\prod_{j\neq i } ( k - x_j)$ will contribute $ \prod_{j \neq i } (-x_j) $. Summing this across $i$, we get $ \sum_{i} (-1)^{n-1} \prod_{j\neq i } x_j = 0 $.
Hence, the numerator is indeed just $ nk^{n-1}$. Thus, we arrive at the desired identity:
$$ \sum_{i=1}^n \frac{1}{k - a_i } = \frac{ \sum_i \prod_{j\neq i } ( k - x_j) } { k^n - 1 } = \frac{ nk^{n-1 } } { k^n - 1 }. $$
Bonus: Use this approach and carefully track the coefficients to directly show that
$$\sum_{i=1}^{n-1} \frac{ 1}{ k-a_i} = \frac{ (n-1)k^{n-2} + (n-2)k^{n-3} + \ldots + 2k + 1 } { \frac{ k^n-1 } { k-1} } \\ = \frac{(n-1)k^{n-1} - k^{n-2} - k^{n-3} - \ldots - k^1 - 1 } { k^n - 1 } = \frac{ nk^{n-1 } } { k^n -1 } - \frac{ 1}{ k - 1 } . $$
- 77,541
-
+1 Nice to see another, natural approach. I've only seen logarithmic derivatives used for such types of exercises (and arguably they are much more elegant, but not easy for someone unfamiliar with calculus) – Sahaj Sep 30 '24 at 16:14