This was asked in Math Stackexchange here but generated no comments or answers. I have slightly edited the original question with the comment in the fourth paragraph and the explicit matrix example at the end.

I have a very concrete question. First an example: If one considers the additive group $Z/(n)$ which is cyclic, the corresponding group characters are the rows of the discrete fourier transform (DFT)matrix $D$.

Take any $n\times n$ circulant matrix $C$. Obviously this matrix is generated by the action of the cyclic group on the first row. Its eigenvectors are the columns of $D$ (which is symmetric) and the matrix of eigenvectors diagonalizes any circulant matrix. Corresponding to each root of unity $\omega$ of order $n,$ its consecutive powers make up the eigenvector and the inner product of the first row of $C$ is clearly the corresponding eigenvalue.

Now comes the question. For simplicity, assume that $n$ is a prime. Consider the multiplicative group of $Z/(n)$ which has $n-1$ elements. The Dirichlet characters of this group form an $(n-1)\times(n-1)$ matrix $D'$. Now take a ``multiplicative matrix'', $M$ where first one specifies the first row $(a_1,\ldots,a_{n-1})$ and then obtains the $k^{th}$ row as $(a_{k},a_{2k},\ldots,a_{(n-1)k})$ where the indices are computed modulo $n.

In fact, if $n$ is assumed to be prime, I'd expect that the eigenvectors to be obtained by a coordinate permutation of the eigenvectors which are the columns of the DFT matrix since both groups in question are cyclic.

I'd have assumed that such a matrix would have exactly the Dirichlet characters, of which there are enough when $n$ is a prime, as its eigenvectors. A few basic computations on Mathematica seemed to contradict this hope. For example taking $n=7,$ and the first row as $(1,1,1,0,0,0)$ yielding the matrix $$ \left( \begin{array}{cccccc} 1 &1 &1 & 0& 0& 0\\ 0 &1 &0 & 1& 0& 1\\ 0 &1 &1 & 0& 0& 1\\ 1 &0 &0 & 1& 1& 0\\ 1 &0 &1 & 0& 1& 0\\ 0 &0 &0 & 1& 1& 1 \end{array} \right) $$ the expected eigenvectors materialized. However, taking the first row as $(1,1,0,0,0,0)$ yielding the matrix $$ \left( \begin{array}{cccccc} 1 &1 &0 & 0& 0& 0\\ 0 &1 &0 & 1& 0& 0\\ 0 &0 &1 & 0& 0& 1\\ 1 &0 &0 & 1& 0& 0\\ 0 &0 &1 & 0& 1& 0\\ 0 &0 &0 & 0& 1& 1 \end{array} \right) $$ did not.

Why is this the case? References, pointers welcome.


I believe the eigenvectors are the ones you guessed, but in your second example, the dimensions of some of the eigenspaces are larger than one. I would guess that Mathematica chose a basis for those eigenspaces different than the eigenvectors coming from the multiplicative Dirichlet characters. By the way, if you view the second example as the adjacency matrix of a directed graph, the graph is two disjoint directed 3-cycles (with loops), and from this point of view the eigenvectors might be more transparent.


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.