Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $T: \mathbb{R}^3 \rightarrow \mathbb{R}^3 $ be the linear operator

$$ A = \begin{pmatrix} 1 & 0 & 0 \\ 1 & 2 & 0 \\ 0 & 0 & 3 \end{pmatrix} $$

The problem I am trying to understand is the following.

True or False? If $W$ is a $T$-invariant subspace of $\mathbb{R}^3$ $\exists T$-invariant subspace $W'$ of $\mathbb{R}^3$ such that $W \oplus W' = \mathbb{R}^3$. I think the answer is true and I will list my ideas below but I think there must be an easier way to do approach this.

Since $A$ has distinct eigenvalues $1,2,3$ we see the minimal polynomial is $m_A(x) = (x-1)(x-2)(x-3)$

and therefore using the fundamental structure theorem for PID's we have $\mathbb{R}^3 = \mathbb{R}/(x-1)\oplus\mathbb{R}/(x-2)\oplus \mathbb{R}/(x-3) = \mathbb{R}\oplus\mathbb{R}\oplus\mathbb{R} $

From this calculation does it follow that $W' = \mathbb{R}\oplus\mathbb{R}$ is the invariant subspace we are looking for?

Is there a way to do this problem without using the fundamental structure theorem for modules over a PID?

share|cite|improve this question
I think you can use Lagrange’s Interpolation Polynomial. – Pierre-Yves Gaillard Aug 2 '11 at 7:28
Your decomposition of $\mathbf{R}^3$ into a direct sum of 3 pairwise non-isomorphic simple $\mathbf{R}[T]$-submodules shows that there are several $T$-invariant subspaces. Some of them are 1-dimensional, some are 2-dimensional. I do think that you are nearly there. Have you covered semisimple modules in your course? – Jyrki Lahtonen Aug 2 '11 at 7:54
In the case of torsion modules, the fundamental structure theorem for modules over a PID reduces to the Chinese Remainder Theorem. – Pierre-Yves Gaillard Aug 2 '11 at 8:27
Yes. But I would call them $\mathbf{R}_1, \mathbf{R}_2$ and $\mathbf{R}_3$ because, A) they are distinct spaces with clear identities, B) they are non-isomorphic as $\mathbf{R}[T]$-modules. So there are 3 different things you seem to call $\mathbf{R}$, and 3 different things that you call $\mathbf{R}\oplus\mathbf{R}$. Altogether 6 (non-trivial) choices for $W$. Finding the corresponding $W'$ in each case is straightforward, but in a first (or second) course on modules over PIDs to get full marks you need to show that you are aware of this. – Jyrki Lahtonen Aug 2 '11 at 9:15
For example, if two of the 1-dimensional submodules were isomorphic, there would be infinitely many submodules, and you need to do a little bit of work (not much, but some) to show that a complementary space can always be found. Also, if the space would not split into a direct sum of simple modules things like Dennis Gulko's example appear. – Jyrki Lahtonen Aug 2 '11 at 9:18
up vote 3 down vote accepted

I am not used to your notation, but I think I understand what you mean. If a linear transformation $T$ on an $n$-dimensional vectors space $V$ has $n$ distinct eigenvalues, say $\{\lambda_1, \lambda_2,\ldots, \lambda_n \}$, then $V$ has a basis consisting of eigenvectors of $T$. Most proofs of this fact make (at least) implicit use of the fact that the $n$-polynomials $ \{p_i(x): 1 \leq i \leq n \}$, where $p_i(x)= \prod_{j \neq i}(x- \lambda_j)$ have greatest common divisor one. Probably the way to disguise this use as far as possible is to say that the minimum polynomial divides the characteristic polynomial, while whenever $\lambda$ is an eigenvalue of a linear transformation, $x - \lambda$ is a factor of its minimum polynomial. Hence when there are $n$ distinct eigenvalues, as here, the minimum polynomial must be $\prod_{i=1}^{n}(x-\lambda_i)$. This means that, with the earlier notation, $p_i(T) \neq 0$ for $1 \leq i \leq n.$ On the other hand, $(T- \lambda_i)p_i(T)= [0]$ for each $i$. Thus for each $i$, $p_i(T)V$ is a non-zero subspace of $V$ consisting of eigenvectors for $T$ associated to eigenvalue $\lambda_i$. Since eigenvectors associated to different eigenvalues are linearly independent, this produces a basis of eigenvectors of $T$. Let $V_i$ be the $1$-dimensional $\lambda_i$-eigenspace for $T$ on $V$. Then we have $V = \oplus_{i=1}^{n}V_i$.

If $W$ is a $T$-invariant subspace of $V$, then for some subset $I$ of $\{1,2,\ldots, n \}$, the minimum polynomial of $T$ on $W$ is $\prod_{i \in I}(x-\lambda_i)$ (it must divide the minimum polynomial on $V$. Then for each $i \in I$, arguing as above, $W$ must contain the $\lambda_i$-eigenspace of $T$, as $W$ does contain an eigector with that eigenvalue, but the eigenspace in $V$ is $1$-dimensional. Hence $W = \oplus_{i \in I}V_i$. Thus $V = W \oplus W'$, where $W' = \oplus_{i \in I'} V_i$ and $I' = \{1,2,\ldots, n\} \backslash I$.

share|cite|improve this answer

The answer is true for the given matrix $A$, since it is semi-simple (diagonalizable), as Geoff explained in his answer, but this is not true in general: for example, if you take $$B = \begin{pmatrix} 1 & 0 & 0 \\ 1 & 1 & 0 \\ 0 & 0 & 1 \end{pmatrix}$$ Then $W=Span\{(0,1,0),(0,0,1)\}$ is $B$-invariant, but there is no $B$-invariant subspace $W'$ such that $W\oplus W'=\mathbb{R}^3$.

share|cite|improve this answer

Let $a$ be an endomorphism of a finite dimensional vector space $V$ over a field $K$. Assume that the eigenvalues of $a$ are in $K$. Let $\lambda$ be such an eigenvalue, let $f\in K[X]$ be the minimal polynomial, and let $d$ its degree. Then there is a unique polynomial $g_\lambda$ of degree $ < d$ such that $g_\lambda(a)$ is the projector onto the generalized $\lambda$ eigenspace. Moreover $g_\lambda$ is determined by the congruences $$g_\lambda\equiv\delta_{\lambda,\mu}\ \bmod\ (X-\mu)^{m(\mu)}$$ for all eigenvalues $\mu$, where $m(\mu)$ is the multiplicity of $\mu$ as a root of $f$, and where $\delta$ is the Kronecker symbol. If $K$ is of characteristic $0$, these congruences can be solved by Taylor's Formula.

EDIT. We have $$g_\lambda=T_\lambda\left(\frac{(X-\lambda)^{m(\lambda)}}{f}\right)\frac{f}{(X-\lambda)^{m(\lambda)}}\quad,$$ where $T_\lambda$ means "degree $ < m(\lambda)$ Taylor polynomial at $X=\lambda$".

In particular, if, as in your case, all the $m(\lambda)$ are equal to one, we get $$g_\lambda=\prod_{\mu\not=\lambda}\ \frac{X-\mu}{\lambda-\mu}\quad.$$

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.