A matrix

Geometry Level 5

Let θ = 2 π 1996 \theta = \dfrac{2\pi}{1996} . Evaluate the determinant of the 1996 × 1996 1996\times1996 matrix I + A I + A , where I I is the 1996 × 1996 1996\times1996 identity matrix and A = ( a j k ) A = (a_{jk}) has entries a j k = c o s ( j θ + k θ ) a_{jk} = cos(j\theta+k\theta) for all j , k j, k .


The answer is -996003.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

1 solution

Mark Hennings
Aug 8, 2019

Suppose that N 3 N \ge 3 and consider the N × N N \times N matrix A A with components a j k = cos ( 2 π ( j + k ) N ) 1 j , k N a_{jk} \; = \; \cos\left(\tfrac{2\pi(j+k)}{N}\right) \hspace{2cm} 1 \le j,k \le N If we define X r = cos ( 2 π ( r + 2 ) N ) r Z X_r \; = \; \cos\left(\tfrac{2\pi(r+2)}{N} \right) \hspace{2cm} r \in \mathbb{Z} then X r + N = X r X_{r+N} = X_r for all r r , and we have a j k = X j + k 2 1 j , k N a_{jk} \; = \; X_{j+k-2} \hspace{2cm} 1 \le j,k \le N Let us define ζ = e 2 π i N \zeta = e^{\frac{2\pi i}{N}} and define the vector v ( k ) C N \mathbf{v}(k) \in \mathbb{C}^N by v ( k ) j = ζ k ( j 1 ) 1 j N \mathbf{v}(k)_j \; = \; \zeta^{k(j-1)} \hspace{2cm} 1 \le j \le N It is well-known that { v ( k ) : 1 k N } \big\{\mathbf{v}(k) \,:\, 1 \le k \le N\big\} is an orthogonal basis for C N \mathbf{C}^N . If we define Λ j = k = 1 N X k ζ k j 1 j N \Lambda_j \; = \; \sum_{k=1}^N X_k \zeta^{kj} \hspace{2cm} 1 \le j \le N then we can show that A v ( k ) = { Λ k v ( N k ) 1 k N 1 Λ N v ( N ) k = N A\mathbf{v}(k) \; = \; \left\{ \begin{array}{lcl} \Lambda_k\mathbf{v}(N-k) & \hspace{1cm} & 1 \le k \le N-1 \\ \Lambda_N\mathbf{v}(N) & & k = N \end{array} \right. Thus C N \mathbf{C}^N can be split into A A -invariant subspaces:

  • one-dimensional subspaces spanned by v ( N ) \mathbf{v}(N) and (if N N is even) v ( 1 2 N ) \mathbf{v}(\tfrac12N) ,
  • two-dimensional subspaces spanned by { v ( k ) , v ( N k ) } \{\mathbf{v}(k),\mathbf{v}(N-k)\} for any 1 k < 1 2 N 1 \le k < \tfrac12N . Since the X k X_k are real, we see that Λ N k \Lambda_{N-k} is the complex conjugate of Λ k \Lambda_k , and so either Λ k \Lambda_k and Λ N k \Lambda_{N-k} are both nonzero or both zero, and so this vector space has a basis of eigenvectors if A A ,

and hence there exists a basis for eigenvectors for A A , and hence for I + A I+A . Thus the eigenvalues of I + A I+A are

  • 1 + Λ N 1 + \Lambda_N and (if N N is even) 1 + Λ 1 2 N 1 + \Lambda_{\frac12N} ,

  • 1 + Λ k Λ N k 1 + \sqrt{\Lambda_k\Lambda_{N-k}} and 1 Λ k Λ N k 1 - \sqrt{\Lambda_k\Lambda_{N-k}} for all 1 k < 1 2 N 1 \le k < \tfrac12N .

Thus we deduce that d e t ( I + A ) = { ( 1 + Λ N ) 1 k < 1 2 N ( 1 Λ k Λ N k ) N o d d ( 1 + Λ N ) ( 1 + Λ 1 2 N ) 1 k < 1 2 N ( 1 Λ k Λ N k ) N e v e n \mathrm{det}\,(I+A) \; = \; \left\{ \begin{array}{lcl} \displaystyle(1 + \Lambda_N)\prod_{1 \le k < \frac12N}\big(1 - \Lambda_k\Lambda_{N-k}\big) & \hspace{1.5cm} & N\; \mathrm{odd} \\ \displaystyle (1 + \Lambda_N)(1 + \Lambda_{\frac12N})\prod_{1 \le k < \frac12N}\big(1 - \Lambda_k\Lambda_{N-k}\big) & & N\;\mathrm{even} \end{array}\right. We now calculate that Λ k = j = 1 N X j ζ k j = 1 2 j = 1 N ( ζ j + 2 + ζ j 2 ) ζ k j = 1 2 ζ 2 j = 1 N ζ ( k + 1 ) j + 1 2 ζ 2 j = 1 N ζ ( k 1 ) j = { 1 2 N ζ 2 k = 1 1 2 N ζ 2 k = N 1 0 o . w . \begin{aligned} \Lambda_k & = \; \sum_{j=1}^N X_j \zeta^{kj} \; = \; \tfrac12\sum_{j=1}^N \big(\zeta^{j+2} + \zeta^{-j-2}\big)\zeta^{kj} \; = \; \tfrac12\zeta^2\sum_{j=1}^N \zeta^{(k+1)j} + \tfrac12\zeta^{-2}\sum_{j=1}^N \zeta^{(k-1)j} \\ & = \; \left\{ \begin{array}{lcl} \tfrac12N\zeta^{-2} & \hspace{1cm} & k = 1 \\ \tfrac12N\zeta^2 & & k = N-1 \\ 0 & & \mathrm{o.w.} \end{array}\right. \end{aligned} from which we deduce that d e t ( I + A ) = 1 1 4 N 2 N 3 \mathrm{det}(I+A) \; = \; 1 - \tfrac14N^2 \hspace{2cm} N \ge 3 irrespective of whether N N is even or odd. With N = 1996 N = 1996 we obtain the answer is 996003 \boxed{-996003} .

nice solution. we can directly use the result about circulant matrices. A is a circulant matrix. we also know the eigenvalues of a circulant matrix. ones we know the eigenvalues of A we know the eigenvalues of I+A. Determinant of I+A is equal to the product of the eigenvalues of I+A.

Srikanth Tupurani - 1 year, 10 months ago

Log in to reply

Very true, but this is not a circulant matrix, which is why the eigenvalue/eigenvector system is more complicated. If A A were circulant, then the v ( k ) \mathbf{v}(k) would be the eigenvectors. However A A is not circulant. However P A PA is a circulant matrix, where P P is the matrix with 1 1 s on the top-right/bottom-left diagonal and 0 0 s everywhere else. This is why A v ( k ) A\mathbf{v}(k) is a multiple of v ( k ) \mathbf{v}(-k) , and not simply of v ( k ) \mathbf{v}(k) .

Mark Hennings - 1 year, 10 months ago

@Mark Hennings, have you ever written the Putnam Exam? I'm just curious because you seem so good at math. Also, btw, I'm not posting any more of these problems because I'm not really gaining anything from posting them.

Donald Wang - 1 year, 7 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...