3
$\begingroup$

Apologies if this question is already known - I have three population ($x(t),y(t)$ and $z(t)$ evalutated at discrete time steps. If we define the sum of these as $T(t) = x(t) + y(t) + z(t)$, you can write

$x(t+1) = ax(t)T(t)$

$y(t+1) = by(t)T(t)$

$z(t+1) = cz(t)T(t)$

where $a$, $b$ and $c$ are constants. For now, $T(t)$ is a constant. If so, one can solve this as a recursive series. However, for a variety of reasons we'd prefer to make this a series of ODEs if possible. Naively, from Taylor's theorem, we know that a function can be written as a sum of the form

$x(t + \delta t) = x(t) + \delta t x'(t) + .....$

If we let $\delta t = 1$, and truncate the series after the first derivative, then we might write these as a system of ODEs after re-arrangement to get..

$x'(t) \approx x(t)(aT(t) - 1)$

$y'(t) \approx y(t)(bT(t) - 1)$

$z'(t) \approx z(t)(cT(t) - 1)$

Solving this seems to give answers close to the exact solution (see image attached, these are acceptably close for our purposes) but I'd like to quantify the error if I can - I considered using a version of Lagrange Remainder formula, but since the form here is cobbled together from a discrete series I don't think there's any way of doing this?

The other question I have involves our motivation for switching to ODEs - it would be great if we could consider situations where $T(t)$ might not be constant, and there may be growth or decay with time, for example if a growth term $g$ was added to $x$ so that

$x'(t) \approx x(t)(aT(t) - 1) + g$ .

There might also be situations when $a,b,$ and $c$ could be themselves functions of time. Is it acceptable to make the simplification here, and what is the best way to quantify the error that arises from such approximation?

Example ODE versus discrete solutions

0 Answers 0