Fancy physics predicts the equality $${n+3\choose 3} =\frac{n+2}{2}\;\sum_{k=1}^{n+1}\csc^2\frac{k\pi}{n+2}$$ which I can check (numerically and symbolically) for small $n$, but cannot prove for every $n$.

Does someone see an elegant way of doing this? Expressing $\csc$ in terms of exponentials allows one to see this as a sum involving roots of unity, but I don't see how to proceed.

  • 6
    $\begingroup$ Why not write it as $${n+1\choose 3} =\frac{n}{2}\sum_{i=1}^{n-1}\frac1{\sin^2(i\pi/n)}?$$ Surely it must come from looking at the coefficients of a Chebyshev polynomial. $\endgroup$ – Lord Shark the Unknown Jun 1 at 18:02
  • 2
    $\begingroup$ Physicists like their context, Shark. :) $\endgroup$ – The Count Jun 1 at 18:04
  • 2
    $\begingroup$ @LordSharktheUnknown, or $$\sum_{k=1}^{n-1}\frac{1}{\sin^2(k\pi/n)} = \frac{n^2-1}{3}$$, or indeed $$H\left(\left\{1 - \cos\left(\frac{2\pi k}n\right) : 1 \le k < n \right\}\right) = \frac{6}{n+1}$$ where $H$ denotes the harmonic mean. $\endgroup$ – Peter Taylor Jun 1 at 22:03

Using parts of this question and answers $$\sum_{k=1}^{n+1}\frac{1}{\sin^2\left(\frac{k\pi}{n+2}\right)}= \sum_{k=1}^{n+1}\left(1+\cot^2\left(\frac{k\pi}{n+2}\right)\right)=\\ n+1+\sum_{k=1}^{n+1}\left(\cot^2\left(\frac{k\pi}{n+2}\right)\right)=n+1+\frac{(n+1)n}{3}=\frac{(n+3)(n+1)}{3}$$

Note: Jack's answer, with a link to wikipedia, touches this question/answers as well.

  • 1
    $\begingroup$ Beautiful! Nice and short. $\endgroup$ – Herri herri Jun 3 at 10:28

Proof using linear algebra :

Initial expression

$${m+3\choose 3} =\frac{m+2}{2}\sum_{k=1}^{m+1}\sin\left(\tfrac{k\pi}{m+2}\right)^{-2},\tag{1}$$

can be transformed, by expressing ${m+3\choose 3}=\dfrac16(m+3)(m+2)(m+1)$ into :

$$\dfrac13(m+3)(m+1) =\sum_{k=1}^{m+1}\sin\left(\tfrac{k\pi}{m+2}\right)^{-2},\tag{1'}$$

Let us separate cases $m$ even and $m$ odd.

First case : $m:=2n$.

In this case, expression (1') is equivalent (due to the property $\sin(\pi-x)=\sin(x)$) to :

$$\dfrac13(2n+3)(2n+1) =1+2\sum_{k=1}^{n}\sin\left(\tfrac{k\pi}{2(n+1)}\right)^{-2},\tag{1''}$$

which is itself equivalent to :

$$\sum_{k=1}^n \dfrac{1}{4 \sin\left(\tfrac{k\pi}{2(n+1)}\right)^2}= \dfrac{n(n+2)}{6}\tag{2}$$

Consider the $n \times n$ tridiagonal matrix

$$D_n=\begin{pmatrix}2&-1&0&0&0&0&\cdots\\ -1&2&-1&0&0&0&\cdots\\ 0&-1&2&-1&0&0&\cdots\\ 0&0&-1&2&-1&0&\cdots\\ 0&0&0&-1&2&-1&\cdots\\ 0&0&0&0&-1&2&\cdots\\ \vdots&\vdots&\vdots&\vdots&\vdots&\vdots& \end{pmatrix}\tag{3}$$

The eigenvalues of $D_n$ are (see Eigenvalues of tridiagonal symmetric matrix with diagonal entries 2 and subdiagonal entries 1) :

$$\lambda_k:=4 \sin\left(\tfrac{k\pi}{2(n+1)}\right)^2, \ \ k=1,2,\cdots n\tag{4}$$

Remark : matrix $D_n$ is the classical discretized version of the second derivative (with a minus sign) : see the "Neumann case" in https://en.wikipedia.org/wiki/Eigenvalues_and_eigenvectors_of_the_second_derivative.

Let us now consider $E_n:=D_n^{-1}.$

For example :

$$E_6:=D_6^{-1}=\dfrac17\begin{pmatrix} 6&5&4&3&2&1\\ 5&10&8&6&4&2\\ 4&8&12&9&6&3\\ 3&6&9&12&8&4\\ 2&4&6&8&10&5\\ 1&2&3&4&5&6 \end{pmatrix}\tag{5}$$

In these matrices, we just need the general fact that the diagonal entries of $D_n$ obey the following formula :


(see Remark 1 below for a proof of (6))

Therefore :

$$trace(E)=\tfrac{1}{n+1}(\sum_{k=1}^n k(n+1-k)) \tag{7}$$

$$=\tfrac{1}{n+1}((n+1)\tfrac{n(n+1)}{2}-\tfrac{n(n+1)(2n+1)}{6})=\tfrac{n(n+2)}{6} \tag{8} $$

i.e., the RHS of (2).


$$trace(E)=\sum \dfrac{1}{\lambda_k}$$ gives the LHS of (2), establishing this relationship.

Remarks :

1) Proof of formula (6). We need a first result which is rather easy to establish by recurrence : $\det(D_n)=n+1$. The cofactor associated with $E_{kk}$ is obtained by cancelling the $k$th row and $k$th column of $A_n$, giving a diagonal-by-block determinant with blocks $A_{k-1}$ and $A_{n-k)}$, whose determinants are resp. $k$ and $n-k+1$. This minor is to be divided by the determinant of $A_n,$ i.e., $n+1.$

2) In (8), we have used the classical formulas for the sum of the $n$ first integers, resp. the squares of the $n$ first integers.

3) A slightly different version of matrix $D$ and its remarkable inverse can be found as second example in the following question : Looking for examples of Discrete / Continuous complementary approaches This is a typical case of an ill-conditionned matrix where a small perturbation (on the bottom right coefficient) gives a very different inverse.

4) Matrix $D_n$ is connected to Chebyshev polynomials of the 2nd kind $U_n$ ; see the answer https://math.stackexchange.com/q/1770607

Second case : $m=2n-1$

Relationship (1') (to be established) is :

$$\dfrac16 n(n+1) =\sum_{k=1}^{n}\left(4\sin\left(\tfrac{k\pi}{2n+1}\right)\right)^{-2},\tag{9}$$

Let us consider now the $n \times n$ tridiagonal matrix :


(please note the bottom right exceptional entry $3$).

It is not difficult to prove that $\Delta_n$ has the following eigenvalues :

$$\mu_k=4 \sin^2\left(\frac{k\pi}{2n+1}\right),\qquad k=1\dots n$$

One can prove as well that $\det \Delta_n=2n+1$.

The inverse $\Gamma_n:=\Delta_n^{-1}$ is a structured matrix whose diagonal elements are :


(proof using the same arguments as those used in Remark 1 above.)

Using (10), the trace of $\Gamma$ is

$$\sum_{k=1}^n k-2\dfrac{1}{2n+1}\sum_{k=1}^n k^2=\dfrac{n(n+1)}{6}$$

which is the LHS of (9), completing the proof.

  • $\begingroup$ Interesting! Thanks for your answer. $\endgroup$ – Herri herri Jun 3 at 10:28
  • 1
    $\begingroup$ I have now a proof covering the two cases (odd and even $m$). $\endgroup$ – Jean Marie Jun 3 at 21:14

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.