I've found this proof of the independence of eigenspaces in a textbook and I can't reproduce it myself.

Theorem. Let $\varphi: L \rightarrow L$ be a linear operator on a linear space L. If numbers ${\lambda}_{1},...,{\lambda}_{k}$ are different, then the subspaces $\ker(\varphi - {\lambda}_{i}\varepsilon)$ are independent.


  1. Suppose that nonzero vectors ${x}_{i} \in \ker(\varphi - {\lambda}_{i}\varepsilon)$ satisfy ${x}_{1} + ... + {x}_{k}=0$.
  2. Let us apply the operators $(\varphi - {\lambda}_{i}\varepsilon)$ to the identity above for $i = 2, 3,...,k$.
  3. Then we get $({\lambda}_{1}-{\lambda}_{2})\cdot({\lambda}_{1}-{\lambda}_{3})\cdot...\cdot({\lambda}_{1}-{\lambda}_{k}){x}_{1}$
  4. Since the numbers ${\lambda}_{i}$ are different from the identity above we get ${x}_{1} = 0$. That is a contradiction, hence the subspaces are independent.

Well, first of all I don't quite get what the second step means. Does it mean a composition of all these operators is applied to the equation, i.e. $(\varphi - {\lambda}_{k}\varepsilon)\circ...\circ(\varphi - {\lambda}_{2}\varepsilon)({x}_{1}+...+{x}_{k})=0$? If so, how do we get ${\lambda}_{1}$ in the third step if we don't apply $(\varphi - {\lambda}_{1}\varepsilon)$? Thank you in advance.


There's something not really precise in point 1, which should be

Suppose that $x_1+x_2+\dots+x_k=0$ where at least one of the vectors $x_1,x_2,\dots,x_k$ is nonzero; without loss of generality we can assume $x_1\ne0$.

Now follow the hint; if we apply $\varphi-\lambda_2\varepsilon$ we get $$ \varphi(x_1)-\lambda_2x_1+ \varphi(x_2)-\lambda_2x_2+ \varphi(x_3)-\lambda_2x_3+ \dots+ \varphi(x_k)-\lambda_2x_k =0 $$ that, in view of the properties of the vectors, becomes $$ \lambda_1x_1-\lambda_2x_1+ \lambda_3x_3-\lambda_2x_3+ \dots+ \lambda_kx_k-\lambda_2x_k =0 $$ or $$ (\lambda_1-\lambda_2)x_1+(\lambda_3-\lambda_2)x_3+\dots+ (\lambda_k-\lambda_2)x_k=0 $$ Upon applying $\varphi-\lambda_3\varepsilon$, the term with $x_3$ vanishes and the term with $x_1$ is multiplied by $\lambda_1-\lambda_3$.

Proceeding in the same way, we're indeed left with $$ (\lambda_1-\lambda_k)\dots(\lambda_2-\lambda_1)x_1=0 $$ and this implies $x_1=0$, which is a contradiction.

I prefer a different kind of proof, by induction on $k$. The result is obvious for $k=1$, so assume we know it for $k-1$ eigenvalues. Suppose $$ x_1+\dots+x_{k-1}+x_k=0 $$ with $x_i\in\ker(\varphi-\lambda_i\varepsilon)$ for $i=1,2,\dots,k$. Apply $\varphi$ to the relation, getting $$ \lambda_1x_1+\dots+\lambda_{k-1}x_{k-1}+\lambda_kx_k=0 $$ and also multiply the relation by $\lambda_k$: $$ \lambda_kx_1+\dots+\lambda_kx_{k-1}+\lambda_kx_k=0 $$ Now subtract getting $$ (\lambda_1-\lambda_k)x_1+\dots+(\lambda_{k-1}-\lambda_k)x_{k-1}=0 $$ By the induction hypothesis, $$ (\lambda_1-\lambda_k)x_1=0,\quad \dots,\quad (\lambda_{k-1}-\lambda_k)x_{k-1}=0 $$ and since $\lambda_i-\lambda_k\ne0$ for $i=1,2,\dots,k-1$, we get $$ x_1=0,\quad \dots,\quad x_{k-1}=0 $$ which also implies $x_k=0$.

The proof is direct, as you see.

  • $\begingroup$ Thank you for this detailed explanation. $\endgroup$ – Eli Korvigo Jun 13 '15 at 21:00

Yes the second step is what you think. Then to understand step 3) just keep in mind that $x_1$ was assumed to be in $\ker(\varphi - \lambda_1 \epsilon)$ i.e. $\varphi (x_1) = \lambda_1 x_1$.

  • $\begingroup$ Thank you for your answer. Your final note helped me sort the idea in my mind. I accepted the answer by @egreg, because he provided a detailed explanation. $\endgroup$ – Eli Korvigo Jun 13 '15 at 20:59
  • $\begingroup$ Indeed @egreg answer put in evidence something not really precise so you did well in accepting his answer. There is also a nice proof of the independence by using the Vandermonde matrix en.wikipedia.org/wiki/Vandermonde_matrix $\endgroup$ – Holonomia Jun 13 '15 at 21:06

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.