4
$\begingroup$

I'm going over the proof for the Final Value theorem using the Dominated Convergence theorem on Wikipedia, I don't understand how from the equation

$$sF\left(s\right)=\int_{0}^{\infty}f\left(\frac{t}{s}\right)e^{-t}dt$$

and the fact that $\left|f\left(\frac{t}{s}\right)e^{-t}\right|$ is dominated by $Me^{-t}$ they get to the limit

$$\lim_{s\searrow0}sF\left(s\right)=\int_0^{\infty}{\alpha}e^{-t}dt=\alpha$$

$\endgroup$
0

1 Answer 1

5
$\begingroup$

$\newcommand{\dd}{\mathrm{d}}\newcommand{\R}{\mathbb{R}}\newcommand{\C}{\mathbb{C}}\newcommand{\lap}{\mathscr{L}}$I went through the proof again, and I think that you're right. The variables are intertwined and the proof is not rigorous. Here's the correct statement:

Final value theorem. Suppose that the following conditions are satisfied:

  1. $f$ is continuously differentiable and both $f$ and $f'$ have a Laplace transform
  2. $f'$ is absolutely integrable, that is, $\int_0^\infty|f'(\tau)|\dd\tau$ is finite,
  3. $\lim_{t\to\infty}f(t)$ exists and is finite,

Then,

\begin{equation} \lim_{t\to \infty}f(t) = \lim_{s\to 0^+}sF(s). \end{equation}

Proof. We know that the Laplace transform of the derivative is

\begin{equation} sF(s) - f(0^+) {}={} \lap\{f'(t)\}(s) {}={} \int_{0^+}^\infty f'(\tau) e^{-s\tau}\dd\tau. \end{equation} Therefore, \begin{align} \lim_{s\to 0^+} sF(s) {}={}& f(0^+) + \lim_{s{}\to{}0^+} \int_{0^+}^{\infty}f'(\tau)e^{-s\tau}\dd\tau \notag \end{align}

We want to use the dominated convergence theorem ; Define $\phi_s(\tau) = f'(\tau)e^{-s\tau}$. We have $|\phi_s(\tau)|\leq f'(\tau)$ which is assumed to be absolutely integrable (Assumption 2). Therefore,

\begin{align} \lim_{s\to 0^+} sF(s) {}={}&f(0^+) + \int_{0^+}^{\infty} \lim_{s{}\to{}0^+} f'(\tau)e^{-s\tau}\dd\tau \notag \\ {}={}&f(0^+) + \int_{0^+}^{\infty} f'(\tau)\dd\tau \notag \\ {}={}& {f(0^+)} + \lim_{t\to\infty}f(t) - {f(0^+)}, \\ {}={}& \lim_{t\to\infty} f(t). \end{align} This completes the proof. $\Box$

Comment on the Wikipedia article (8 May, 2019): it is stated that with a change of variables, $\xi = st$, (in fact, the same symbol is used for $t$ and $\xi$), the integral becomes

$$ \int_0^\infty \lim_{s\to 0^+}f(\xi/s) e^{-\xi}\dd\xi. $$

As the OP noted, there is some confusion in that proof; in fact, $\xi$ and $s$ are not independent variables (by definition), so the limit $\lim_{s\to 0^+}f(\xi/s)$ is not equal to $\lim_{t\to\infty}f(t)$.


Update: Following @LaurentLessard's comment, here is a different proof that requires $f$ to be bounded

Final value theorem #2. Let $f$ be a function in the time domain with Laplace transform $F(s)=(\mathcal{L} f)(s)$. Suppose that the following conditions are satisfied:

  1. $f$ is bounded,
  2. $\lim_{t\to\infty}f(t)$ exists and is finite,

then, $$ \lim_{t\to \infty}f(t) = \lim_{s\to 0^+}sF(s). $$

Proof. We observe that \begin{align} sF(s) = \int_0^{\infty} f\left(\frac{y}{s}\right)e^{-y}\dd{}y. \end{align} Define the function $f_s(y) = f\left(\frac{y}{s}\right)e^{-y}$. Since $f$ is assumed to be bounded, there is an $M>0$ such that $|f(t)| \leq M$ for all $t\geq 0$, therefore, $|f_s(y)| \leq Me^{-y}$. As a result, we can apply the dominated convergence theorem: \begin{align} \lim_{s\to0^+}sF(s) {}={} & \lim_{s\to0^+}\int_0^{\infty} f\left(\frac{y}{s}\right)e^{-y}\dd{}y \notag \\ \overset{\text{dct}}{=}{} & \int_0^{\infty} \lim_{s\to0^+}f\left(\frac{y}{s}\right)e^{-y}\dd{}y {}={} \lim_{t\to\infty}f(t). \end{align} This completes the proof. $\Box$

$\endgroup$
6
  • $\begingroup$ How did you make the step $\int_0^{\infty}\lim_{s\to 0^+} f\left(\frac{\xi}{s}\right)e^{-\xi}d\xi = \lim_{s\to 0^+} f\left(\frac{\xi}{s}\right)\int_0^{\infty}\lim_{s\to 0^+}$ if the limit depends on $\xi$? $\endgroup$ Commented May 8, 2019 at 16:16
  • $\begingroup$ @SIMEL Thank you for accepting my answer. Let me rewrite it in a more clear way. I'll also add the statement of FVT for completeness. $\endgroup$ Commented May 8, 2019 at 16:45
  • $\begingroup$ I think Assumption 3 is stronger than it needs to be. It appears the only place Assumption 3 is used in the proof is in the first step; to ensure that the formula for the Laplace transform of the derivative holds true (because proving it involves integration by parts, and we need $f(t) e^{-st} \to 0$ as $t\to\infty$ when $s>0$). But this will hold under weaker conditions, such as $f$ being bounded, or even growing polynomially. Am I missing something? $\endgroup$ Commented May 20, 2022 at 5:59
  • $\begingroup$ @LaurentLessard are you referring to assumptions 1 and 2? Indeed, they are stronger and FVT can be proven using the assumptions that $f$ is bounded and the limit $\lim_{t\to\infty}f(t)$ exists and is finite. The boundedness assumption can be relaxed further as you said. The proof uses the dominated convergence theorem. I will add another answer in the weekend with the proof. $\endgroup$ Commented May 21, 2022 at 1:49
  • $\begingroup$ @Pantelis I was talking about assumption 3. I have seen other versions of FVT where very few assumptions are needed --- see e.g. the "elementary proof" in the FVT wiki article, which only assumes $f$ is measurable and bounded, and the limit $\lim_{t\to\infty} f(t)$ exists and is finite; no need for differentiability, existence of Laplace, or assumption 2 about $f'\in L_1$. I was actually hoping for a version of FVT where assumption 3 is not needed; where you can actually prove that this limit must exist and be equal to $\lim_{s\to 0^+} s F(s)$. Kinda like a converse FVT, I suppose. $\endgroup$ Commented May 22, 2022 at 10:23

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.