Let $\lambda_1,...,\lambda_n$ be a set of distinct eigenvalues for a linear operator $T:V \to V$. Define $$\Lambda_{\lambda_i}=\{v \in V | (\lambda_i I-T)^{n_i}(v)=0 \text{ for some } n_i\in \mathbb N\}$$ I am trying to prove that the subspaces $\Lambda_{\lambda_1},\Lambda_{\lambda_2},...,\Lambda_{\lambda_n}$ are linearly independent. My approach is to use induction on $n$.

For the base case $n=1$ the result is trivial since $\forall v\in\Lambda_{\lambda_1}$ if $v=0$ then $v=0$.

Now assume that $\Lambda_{\lambda_1},...,\Lambda_{\lambda_{r}}$ are linearly independent for some $r<n$ (Induction hypothesis). Suppose that $$v_1+ \dot{} \dot{} \dot{}+v_r+v_{r+1}=0$$ where each $v_i \in \Lambda_{\lambda_i}$ for $i=1,...,r+1.$ We can apply $(\lambda_{r+1}I-T)^{n_{r+1}}$ to both sides of the equation to get that $$(\lambda_{r+1}I-T)^{n_{r+1}}(v_1+ \dot{} \dot{} \dot{}+v_r)+(\lambda_{r+1}I-T)^{n_{r+1}}(v_{r+1})=0$$ and since $v_{r+1}\in \Lambda_{\lambda_{r+1}}$ we get that $$(\lambda_{r+1}I-T)^{n_{r+1}}(v_1+ \dot{} \dot{} \dot{}+v_r)=0$$ Now, in order to use my induction hypothesis I need to show that $v_1+ \dot{} \dot{} \dot{}+v_r=0$ and this is where I get stuck. Why must it be true at this point that $v_1+ \dot{} \dot{} \dot{}+v_r=0$?

  • $\begingroup$ What does it mean for a collection of subspaces to be linearly independent? This is not standard terminology. $\endgroup$ – littleO Jan 30 '17 at 5:37
  • $\begingroup$ @littleO By linearly independence I mean that a subspace $A$ is linearly independent with a subspace $B$ if $\forall a \in A$ and $b\in B$, $a$ and $b$ are linearly independent in $V$ $\endgroup$ – gene Jan 30 '17 at 5:39
  • $\begingroup$ Maybe that terminology is used more often than I realize. $\endgroup$ – littleO Jan 30 '17 at 5:46
  • $\begingroup$ @littleO usually we say $A \cap B = \{0\}$ $\endgroup$ – Omnomnomnom Jan 30 '17 at 6:18

We can show a more general statement: If $p_1,\ldots,p_n$ are mutually coprime polynomials, and $v_1,\ldots,v_n$ are vectors $\neq 0$ satisfying $$ p_i(T)\cdot v_i = 0\;\;\forall i\in\{1,\ldots,n\} $$ then the vectors $v_1,\ldots,v_n$ are linearly independent.

Let $k\in\{1,\ldots,n\}$. Let $\bar{p}_k=p_1\cdot\ldots\cdot p_{k-1}\cdot p_{k+1}\cdot\ldots\cdot p_n$. As $p_k$ and $\bar{p}_k$ are coprime, there are polynomials $q_k$ and $\bar{q}_k$ satisfying $$ q_kp_k+\bar{q}_k\bar{p}_k = 1 $$ Let $B_k = q_k(T)p_k(T)$ and $\bar{B}_k = \bar{q}_k(T)\bar{p}_k(T)$. As $q_kp_k+\bar{q}_k\bar{p}_k = 1$, we have $B+\bar{B} = I$.

Now we check if there is a non-trivial solution of $\alpha_1 v_1+\ldots+\alpha_n v_n=0$.

We have $$ 0 = \bar{B}_k v_i \;\;\forall i\in\{1,\ldots,k-1,k+1,\ldots,n\} $$ because $\bar{B}_k$ contains $p_i(T)$ as one of its factors, and all the factors commutate. Therefore, $$ 0 = \bar{B}_k (\alpha_1 v_1+\ldots+\alpha_{k-1}v_{k-1}+\alpha_{k+1}v_{k+1}+\ldots+\alpha_n v_n) $$ On the other hand, we have $$ 0 = \bar{B}_k (\alpha_1 v_1+\ldots+\alpha_n v_n) $$ because we assumed $\alpha_1 v_1+\ldots+\alpha_n v_n=0$. If we subtract the last two equations, we get $$ \bar{B}_k\alpha_k v_k = 0 $$ As $B_k\alpha_k v_k = 0$, we finally find $$ \alpha_k v_k = (I)\cdot\alpha_k v_k = (B_k+\bar{B}_k)\cdot\alpha_k v_k = B_k\alpha_k v_k+\bar{B}_k\alpha_k v_k = 0 $$ from which we conclude $\alpha_k = 0$


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.