I'm trying to prove that every symmetric matrix is diagonalizable. I know that there are already many answers regarding this question, but I just want to check out whether my approach is correct or not. After some searching, I found some clues to my problem (Here). Based on Tunococ's answer, I want to prove the following : For every symmetric matrix (with real entries), there exists an orthogonal matrix B and a diagonal matrix D such that $$ B^{T}AB = D $$ Let's use induction on orders of $A$. For $1 \times 1$ matrix (which is symmetric), this is trivial. Suppose that this result is true for $k \times k$ symmetric matrix ($k\geq1$). Let $A$ be a $(k+1) \times (k+1)$ symmetric matrix. Let $\lambda$ be any eigenvalue of $A$ and $v$ be an associated eigenvector. Since an eigenvector multiplied by any nonzero scalar is still an eigenvector, we can assume that $\left \| v \right \|=1$. By Gram-Schmidt process, we can obtain an orthonormal basis $\{v, w_2, \ldots, w_{k+1}\}$ of $\mathbb{R}^{k+1}$. Let C be an orthogonal matrix defined by $$ C = \begin{bmatrix} | & | & \ldots & |\\ v & w_2 & \ldots & w_{k+1} \\ | & | & \ldots & | \end{bmatrix} $$
Then, after some computation, we obtain $$C^{T}AC =\begin{bmatrix} \lambda & O_{1 \times k} \\ O_{k \times 1} & F \end{bmatrix}.$$ Here, $F$ is a $k \times k$ symmetric matrix. By the induction hypothesis, there exists an orthogonal matrix $G$ and a diagonal matrix $H$ such that $G^{T}FG=H$. Let $$J =\begin{bmatrix} 1 & O_{1 \times k} \\ O_{k \times 1} & G \end{bmatrix}.$$ Since $G$ is an orthogonal matrix, $J$ is also an orthogonal matrix. By computation, we obtain $$(CJ)^{T}A(CJ)=\begin{bmatrix} \lambda & O_{1 \times k} \\ O_{k \times 1} & H \end{bmatrix}.$$ Note that since $C$ and $J$ are orthogonal matrices, their product $CJ$ is also an orthogonal matrix.

Is this argument correct?


No, you need to explain why you can find an eigenpair $(\lambda, v)$.

  • $\begingroup$ Umm.... Since characteristic equation is polynomial equation of order (k+1), equation always has roots.... Choose any root(eigenvalue), and thus we can find an associated eigenvector.... Also, since A is symmetric, roots are always real. $\endgroup$ – Kim Aug 14 at 15:58
  • $\begingroup$ No, $1 + x^2$ has no real root. $\endgroup$ – user7440 Aug 14 at 16:20
  • $\begingroup$ When A is symmetric, if there is an eigenvalue, then it is real. (using $x^* A x = (Ax)^* x$ ). $\endgroup$ – user7440 Aug 14 at 16:21
  • $\begingroup$ I mean all the roots of the characteristic polynomial of a symmetric matrix are real numbers... So.... I think $1+x^2$ is not our candidate.... $\endgroup$ – Kim Aug 14 at 16:29
  • $\begingroup$ Look here control.ucsd.edu/mauricio/courses/mae280a/lecture11.pdf (Proof of bullet 2). $\endgroup$ – user7440 Aug 14 at 16:38

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.