2
$\begingroup$

Let $\Delta_n$ be the closed unit simplex in $\mathbb R^n$. For any $a,b \in \Delta_n$, define the differential equation: $$ a'(u) = b-a(u) \quad\quad\quad a(0) = a $$

How does one go about solving this system of ODEs? Solving the case where $n=1$ is simple, does this carry over to the $n$ dimensional case?

I've looked at using eigen values but is there a simpler way when dealing with a large choice of $n$ ? Any hints appreciated

Attempt for the Matrix Differential Equation: $$ \frac{dx}{dt} = Ax + f(t), \quad\quad\quad x(0) = x_0 $$ where $A$ is a $n\times n$ matrix, $x\in \mathbb R^n$

the solution is given by:

$$ x(t) = e^{tA} \int_0^te^{-sA}f(s)ds + e^{tA}x_0 $$

Using the fact that for a diagonal matrix : $D=\text{diag}(a_1,...,a_n)$, $e^D = \text{diag}(e^{a_1},....e^{a_n})$

then for my problem:

$$ \frac{da}{dt} = \text{diag}(-1,...-1)a(u) + b $$ then \begin{align*} a(t) &= \text{diag}(e^{-t},...,e^{-t}) \int_0^t \text{diag}(e^{-s},...,e^{-s})~ b~ ds + \text{diag}(e^{-t},...,e^{-t}) a\\ &=\text{diag}(e^{-t},...,e^{-t})\text{diag}(1-e^{-t},...1-e^{-t}) b +\text{diag}(e^{-t},...e^{-t}) a\\ &=\text{diag}(e^{-t}-e^{-2t},...e^{-t}-e^{-2t}) b + \text{diag}(e^{-t},...e^{-t}) a\\ \end{align*} $$ =\begin{pmatrix} e^{-t}(a_1(0)+b_1) -b_1e^{-2t} \\ \vdots\\ \vdots \\ e^{-t}(a_n(0)+b_n) -b_ne^{-2t} \\ \end{pmatrix} $$

Is this correct?

$\endgroup$
2
+50
$\begingroup$

Well, the answer is already there, I would just like to add a typo correction. The answer via your method also yields the correct solution. \begin{align*} a(t) &= \text{diag}(e^{-t},...,e^{-t}) \int_0^t \text{diag}(e^{\color{blue}{+}t},...,e^{\color{blue}{+}t})~ b~ ds + \text{diag}(e^{-t},...,e^{-t}) a\\ &=\text{diag}(e^{-t},...,e^{-t})\text{diag}(\color{blue}{e^{+t}-1},...,\color{blue}{e^{+t}-1}) b +\text{diag}(e^{-t},...e^{-t}) a\\ &=b+ + \text{diag}(e^{-t},...e^{-t}) (a-b)\\ \end{align*}

Which is the result you can find above, as Christian said, the problem was when you plugged the $A$ in your recipe.

Hope this helps

$\endgroup$
  • $\begingroup$ isn't the matrix integral just integrating each element of the matrix? $\int_0^t e^{-s}ds = -e^{-s} \vert^t_0 = 1-e^{-t} $? $\endgroup$ – dimebucker Jul 12 '15 at 15:59
  • $\begingroup$ Yes, I had a typo there, sorry $\endgroup$ – Conrado Costa Jul 12 '15 at 16:13
2
$\begingroup$

(You have $e^{-sA}$ not converted correctly when you plugged your special $A$ into the general recipe.)

Your system $$x'(t)=b-x(t),\qquad x(0)=a,$$ separates without any further ado into the $n$ IVPs $$x_k'(t)=b_k-x_k(t),\qquad x_k(0)=a_k\qquad(1\leq k\leq n)\ .\tag{1}$$ The solutions of $(1)$ are given by $$x_k(t)=(a_k-b_k)e^{-t}+b_k\qquad(1\leq k\leq n)\ ,$$ and we can combine them to the vector solution $$x(t)=e^{-t} (a-b) +b\ .$$

$\endgroup$
  • $\begingroup$ shouldn't the matrix method give the same solution? $\endgroup$ – dimebucker Jul 12 '15 at 9:48

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.