In my Advanced ODE's course we are learning about diagonalization of a coupled system of ODE's and using this to find the solution to the system such that the original system is:

$x'(t) = Ax$

So we find the eigen values and corresponding eigen vectors of matrix A and create a matrix P such that $P=[v_1 , v_2, ..., v_n]$ where the $v_i$'s are the eigen vectors as columns of matrix $P$, corresponding to eigen value $\lambda_i$.

Then we find matrix $B$ such that $B = PAP^{-1}$ and then B is diagonalized, with the previously found eigen values of $A$ on the diagonal and easily shows the new uncoupled system for, $y'(t) = By$, by change of variables, that can then be easily solved.

This will then result in a solution matrix, we call it $ E(t)$, such that: $y(t) = E(t)y(0)$ which can then be used to find the general solution to the original coupled system of ODE's, $x'(t) = Ax$ such that $x(t) = PE(t)P^{-1}x(0)$.

My problem with understanding this method is why go through all the extra work to find the solution to the original system of ODE's by finding matrix B and using it to first solve the system $y'(t) = By$ to then finally solve the system $x'(t)=Ax$ when you have already found the eigen vectors and eigen values of $A$, and can use then to write the general solution, for example as:

$x(t) = c_1e^{\lambda_1t}[v_1] + c_2e^{\lambda_2t}[v_2]$

The diagonalization method just seems like extra unnecessary steps that I don't understand why we would take them?


The equation $x'(t) = A x(t)$ has solutions $$ x(t) = e^{t A}\, x(0) \, , $$ where $e^{t A}$ is the matrix exponential of $t A$. If $A$ is diagonalizable, i.e. $A = P B P^{-1}$ where $B = \text{diag}(\beta_i)$ is a diagonal matrix, then the matrix exponential can be computed according to $$ e^{t A} = P e^{t B} P^{-1}\, , \qquad\text{where}\qquad e^{t B} = \text{diag}(e^{\beta_i t}) = E(t) \, . $$ Otherwise, it may be more difficult to compute the matrix exponential $e^{t A}$.

  • $\begingroup$ Here you wrote $A$ as being diagonalizable, $A=UDU^{-1}$, but in the method I am using we create a whole new matrix $B$ that is diagonalized BY original matrix $A$, which is not necessarily originally already a diagonal matrix. Is this still the same thing? $\endgroup$ – TunaBooties Aug 31 '17 at 14:11
  • $\begingroup$ @JustynaTafoya If I understood well, the matrix $B$ is diagonal since $\dot{y} = B y $ is uncoupled. Thus, it is the same as the diagonal matrix which I previously denoted by $D$. I have updated the notation to be consistent with the OP. $\endgroup$ – Harry49 Aug 31 '17 at 15:35

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.