7
$\begingroup$
  1. What can we tell about $\det(X)$ if we know $\det(I + X)$? Will it give some kind of bound for $\det(X)$?
  2. In general, if we know the determinant of matrix $A + X$, where $A$ is a constant matrix, how much can we say about $\det(X)$?

Thank you for the attention.

$\endgroup$
4
  • 9
    $\begingroup$ This is essentially what you ask: "Let $p(x)\in\mathbb{C}[x]$ be a monic polynomial of degree $n$. I know the value of $p(t)$ for some $t\in\mathbb{C}$. What do we know about $p(0)$?" The answer is nothing. For any $t\neq 0$ and any possible value of $p(t)$, $p(0)$ can take any value, provided that $n>1$. $\endgroup$ Jul 6, 2016 at 14:59
  • $\begingroup$ @Batominovski, thank you for the reply. How is det(A + X), det(X) related to the polynomial of p(t) , p(0) in your answer ? $\endgroup$
    – david
    Jul 6, 2016 at 15:25
  • 1
    $\begingroup$ @david Take $p(x):=\det(x\,I-X)$ (i.e., the characteristic polynomial of $X$), and you will see. $\endgroup$ Jul 6, 2016 at 15:48
  • 1
    $\begingroup$ @david The situation is rather specific to the case where the shift is by a multiple of $I$. In this case the function $p(\lambda)=\operatorname{det}(A+\lambda I)$ is a polynomial. In general it can be any monic polynomial of degree $n$ (there is a so-called "companion matrix" for any monic polynomial, which has that polynomial as its characteristic polynomial). You're saying that you know $p(1)$ and want to know $p(0)$, but this cannot happen. $\endgroup$
    – Ian
    Jul 6, 2016 at 15:49

4 Answers 4

1
$\begingroup$

The determinant is the product of the eigenvalues $$\det({\bf A}) = \prod_k \lambda_k({\bf A})$$

Then it is well known that all eigenvalues increase by 1 if adding $\bf I$ to any matrix:

$$\det({\bf A+I}) = \prod_k (\lambda_k({\bf A})+1)$$

Maybe you can rewrite this product into something you are comfortable working with. I do not think this extends in any nice way to adding $\bf X\neq I$.

$\endgroup$
1
$\begingroup$

Let $\{\lambda_1,\dots,\lambda_n\}$ be the set of eigenvalues of $X$. Then the set of eigevalues of $I+X$ is $\{1+\lambda_1,\dots,1+\lambda_n\}$. Determinant of any matrix $X$ is equal to product of all eigenvalues of $X$, thus \begin{align} \det(X) = \prod_{i=1}^n\lambda_i &&\det(I+X) = \prod_{i=1}^n(1+\lambda_i) \tag{1} \end{align} From (1) it is clear that (in general) nothing can be said about $\det(I+X)$.

If all eigenvalues of $X$ are real and positive (i.e. $X$ is positive definite), then $\det(I+X) > \det(X)$.

$\endgroup$
1
$\begingroup$

If we are given an invertible matrix $X \in \mathbb R^{n \times n}$ and the following determinant

$$\beta := \det (X + \alpha 1_n 1_n^T)$$

then

$$\beta = \det (X + \alpha 1_n 1_n^T) = \det (X) \cdot \underbrace{\det (I_n + \alpha X^{-1} 1_n 1_n^T)}_{= 1 + \alpha 1_n^T X^{-1} 1_n} = (1 + \alpha 1_n^T X^{-1} 1_n) \cdot \det (X)$$

where we used Weinstein-Aronszajn determinant identity. Thus,

$$\det (X) = \frac{\beta}{1 + \alpha 1_n^T X^{-1} 1_n}$$

$\endgroup$
7
  • $\begingroup$ The OP wants $\det(X+I)$, not $\det(X+1_n1_n^T)$. $\endgroup$
    – egreg
    Jul 6, 2016 at 23:05
  • $\begingroup$ @egreg Please read the 2nd paragraph. $\endgroup$ Jul 6, 2016 at 23:06
  • $\begingroup$ “In general…” doesn't mean “for a carefully selected case”. ;-) $\endgroup$
    – egreg
    Jul 6, 2016 at 23:07
  • 1
    $\begingroup$ @egreg The OP wrote "constant matrix". The case is not carefully selected, it's what he asked for. $\endgroup$ Jul 6, 2016 at 23:09
  • $\begingroup$ @RodrigodeAzevedo, thank you for the reply. I am not familiar with the symbol you used. What is a1n, 1Tn in your expression ? $\endgroup$
    – david
    Jul 7, 2016 at 1:30
0
$\begingroup$

Nothing To document that "nothing if $d>1$" where $d$ is the dimension is the answer (as said above) :

Claim: For any $\lambda\in \mathbb R$ there exists a matrix $X$ such that $\det X = \lambda$, while $\det(I+X)=0$.

As a corollary (take $\lambda = \pm n\in\mathbb N$ in the first claim),

Corollary: there exists a sequence $X_n$ such that $\det X_n \to \pm \infty$ while $\det(I+X_n)=0$.

This tells you that nothing general can be said.

Proof. If $d$ is odd take $X$ to be be diagonal matrix with $X_{11}=\lambda$ and $X_{kk}=1$ for $2\leq k\leq d-1$ and $X_{dd}=-1$. If $d$ is even and greater or equal to $4$, $X_{kk}=1$ for $2\leq k\leq d-2$ and $X_{d-1,d-1}= X_{dd}=-1$. if$d=2$, take $X=\begin{pmatrix} -1 & \lambda \\ -1 & -1\end{pmatrix}$.

Something

On the other hand, it could be that you meant something else than what you asked. For example, what is true is Hadamard's inequality: $$ \| \det(A+X) -\det(A) \| \leq C(d) \|X\| \max(\|A+X\|^{d-1},\|A\|^{d-1}) $$

$\endgroup$

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.