3
$\begingroup$

Let $J_n$ be an $n\times n$ matrix all of whose entries are $1$, and $I_n$ be the identity matrix. Define $$K_n = J_n-I_n$$

For $n=1$ to $5$ my (usually unreliable!) hand calculations suggest that

$$\det K_n = (-1)^{n-1}(n-1)$$

Question

(a) are these values correct? (b) is the generalization to any positive integer $n$ valid? (c) if (b) is true, how can the result be demonstrated?

My only idea so far is to use the product of eigenvalues:

$$\det K_n = \prod_{j=1}^n \lambda_j$$

($-1$ is an eigenvalue for all $n$)

Any assistance much appreciated.

(note: a similar question concerning skew-symmetric matrices Determinant of a special skew-symmetric matrix may contribute some relevant ideas. or perhaps Determinant of a matrix with $t$ in all off-diagonal entries. has greater relevance)

$\endgroup$
2
$\begingroup$

This is the adjacency matrix of the complete graph $K_n$. The eigenvalues are $\lambda_1 = (n - 1)$ and $\lambda_2 = \dots = \lambda_n = -1$.

The eigenvector for $\lambda_1$ is the vector $\pmb{1}$ consisting of all $1$s.

An eigenbasis for $\lambda_2,\dots,\lambda_n$ is given by $e_1 - e_j$. Note that

$$ (J_n - I_n)(e_1 - e_n) = \pmb{1}\pmb{1}^T(e_1 - e_j) - (e_1 - e_j) = - (e_1 - e_j). $$

This is because $J_n = \pmb{1}\pmb{1}^T$ and $\pmb{1}^Te_i = 1$ for all basis vectors $e_i$.


The fact that $n - 1$ is an eigenvalue reflects the fact that $K_n$ is an $(n - 1)$-regular graph. There may also be a combinatorial interpretation for the other eigenvalues but I don't know enough algebraic graph theory to say.

$\endgroup$
  • $\begingroup$ thank you. very nice insight re the complete graph. $\endgroup$ – David Holden Jul 23 '17 at 3:34
3
$\begingroup$

Using Sylvester's determinant identity,

$$\begin{array}{rl} \det ( 1_n 1_n^\top - \mathrm I_n ) &= \det ( -( \mathrm I_n - 1_n 1_n^\top ) )\\ &= (-1)^n \cdot \det(\mathrm I_n - 1_n 1_n^\top)\\ &= (-1)^n \cdot \det ( 1 - 1_n^\top 1_n )\\ &= (-1)^n \cdot (1-n)\\ &= (-1)^{n-1} \cdot (n-1)\end{array}$$

$\endgroup$
  • 1
    $\begingroup$ really nice answer, thanks, and very educational for me. i hadn't encountered the determinant identity before, so this has clarified the meaning of the Schur complement - a concept which i encountered in reading recently without understanding its purport. have already accepted another answer, but yours is a very nice application of a powerful result to my rather simple question. thanks again! $\endgroup$ – David Holden Jul 23 '17 at 14:33
2
$\begingroup$

Finding the eigenvalues of a matrix $A$ that has $a$ along its main diagonal and $b$ everywhere else comes up fairly often on this site, but I’ll repeat the calculation here. Write the matrix as $A=b\mathbb1+(a-b)I$, where $\mathbb 1$ is the $n\times n$ matrix with every entry equal to $1$. Suppose $v$ is an eigenvector of $A$. We then have $b\mathbb1v=(\lambda-a+b)v$, so $v$ is also an eigenvector of $b\mathbb1$. Moreover, the eigenvalues of this matrix and those of $A$ differ by $a-b$. Now, for $b\ne0$, $b\mathbb1$ has rank 1, so its eigenvalues are $0$ with multiplicity $n-1$ and, using the fact that the trace of a matrix is equal to the sum of its eigenvalues, $nb$ with multiplicity 1. (You could also find the non-zero eigenvalue by noticing that all of the row sums are identical.) This means that the eigenvalues of $A$ are $a+(n-1)b$ and $a-b$, giving $$\det A=(a+(n-1)b)(a-b)^{n-1}.$$ For your matrices, $a=0$ and $b=1$, so $\det{K_n}=(-1)^{n-1}(n-1)$, just as you’ve calculated.

$\endgroup$
  • $\begingroup$ thanks very much! another very enlightening answer, and very nicely explained. each of the three answers given here has helped me to expand my rather limited understanding of linear algebra. $\endgroup$ – David Holden Jul 25 '17 at 3:07

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.