There are many variations of "differentiating under the integral sign" theorem; here is one:

If $U$ is an open subset of $\mathbb{R}^n$ and $f:U \times [a,b] \rightarrow \mathbb{R}$ is continuous with continuous partial derivatives $\partial_1 f, \dots \partial_n f$ then the function $$ \phi(x) = \int^b_a f(x,t)dt $$ is continuously differentiable and $$ \partial_i \phi (x) = \int^b_a \partial_i f(x,t)dt $$

Can anyone suggest a textbook that provides a proof of this version of the theorem?

  • $\begingroup$ What are the tools you are allowed to use? For example, you can try a proof with dominated convergence theorem. $\endgroup$ – Davide Giraudo Jun 25 '12 at 16:03
  • $\begingroup$ I am familiar with the statement of that theorem, but don't know it's proof. I would like a reference that proves a more self-contained proof, inasumuch as one is possible... $\endgroup$ – ItsNotObvious Jun 25 '12 at 16:05
  • $\begingroup$ IMHO, you'll do faster by proving it yourself. Anyway, try looking at this Wikipedia page: en.wikipedia.org/wiki/Differentiation_under_the_integral_sign (which I find a bit too lengthy, though). $\endgroup$ – Giuseppe Negro Jun 25 '12 at 16:47

Isn't the proof sort of "follow your nose"? Let $\Delta x$ be nonzero, consider

$$ \phi(x+\Delta x)-\phi(x) = \int^{b}_{a}f(x+\Delta x,t)-f(x,t)\,\mathrm{d}t$$

Then construct the quotient

$$ \frac{\phi(x+\Delta x)-\phi(x)}{\Delta x} = \frac{\int^{b}_{a}f(x+\Delta x,t)-f(x,t)\,\mathrm{d}t}{\Delta x} $$

But because we do not integrate over $x$, we treat $x$ like a constant. So we can rewrite the integral as

$$ \frac{\phi(x+\Delta x)-\phi(x)}{\Delta x} = \int^{b}_{a}\frac{f(x+\Delta x,t)-f(x,t)}{\Delta x}\,\mathrm{d}t $$

Taking the limit as $\Delta x\to0$ gives us

$$ \frac{\mathrm{d}\phi(x)}{\mathrm{d} x} = \int^{b}_{a}\frac{\partial f(x,t)}{\partial x}\,\mathrm{d}t $$

precisely as desired? [Edit: We can take the limit under the integral sign, as Giuseppe Negro points out, if the function $f(x,t)$ is continuously differentiable in $x$.]

Addendum: Why, oh why, do we need $f(x,t)$ to be continuously differentiable in $x$?

Why can we take this limit? Well, there's a number of different arguments.

One is the Dominated convergence theorem, which states if we have a sequence of functions $f_{n}(t)\to F(t)$ which is "dominated" by some function $g(t)$, meaning $$ |f_{n}(t)|\leq g(t)\quad\mbox{for any }t $$ then we have $$ \lim_{n\to\infty}\int|f_{n}(t)-F(t)|\,\mathrm{d}t=0 $$ which implies $$ \lim_{n\to\infty}\int f_{n}(t)\,\mathrm{d}t=\int F(t)\,\mathrm{d}t. $$ Take $F(t)=\partial f(x,t)/\partial x$ and $f_{n}(t)$ to be $$ f_{n}(t) = \frac{f(x + \varepsilon_{n},t)-f(x,t)}{\varepsilon_{n}} $$ using any sequence $\varepsilon_{n}\to 0$.

Addendum 2: A second different way begins with the observation $$ \int^{b}_{a}\int^{x}_{0}\frac{\partial f(y,t)}{\partial y}\,\mathrm{d}y\,\mathrm{d}t = \phi(x)-\phi(0)$$ by the fundamental theorem of calculus. Fubini's theorem lets us switch the order of integration $$ \int^{x}_{0}\int^{b}_{a}\frac{\partial f(y,t)}{\partial y}\,\mathrm{d}t\,\mathrm{d}y = \phi(x)-\phi(0)$$ Then we can use Leibniz's rule differentiating both sides with respect to $x$. This gives us the desired result $$ \int^{b}_{a}\frac{\partial f(x,t)}{\partial x}\,\mathrm{d}t = \phi'(x).$$

Recall Leibniz's rule states if $G(x) = \int^{x}_{0}g(y)\,\mathrm{d}y$ then $$ G'(x) = g(x). $$ We can prove this quickly by $$ \frac{G(x+\Delta x)-G(x)}{\Delta x} = \frac{1}{\Delta x}\int^{x+\Delta x}_{x} g(y)\,\mathrm{d}y$$ and taking $\Delta x$ to be "sufficiently small", we can approximate the Riemann sum as $$ \int^{x+\Delta x}_{x} g(y)\,\mathrm{d}y\approx g(c)\Delta x$$ where $x\leq c\leq x+\Delta x$. Plugging this back in gives us $$ \frac{G(x+\Delta x)-G(x)}{\Delta x} = \frac{1}{\Delta x}\left(g(c)\Delta x\right) = g(c)$$ Taking $\Delta x\to 0$ gives us $c\to x$, and $$ G'(x) = g(x)$$ as desired.

  • 3
    $\begingroup$ You need to justify the interchange of limit and integral in the last step. This can be done by exploiting the hypothesis of continuous differentiability of $f$. $\endgroup$ – Giuseppe Negro Jun 25 '12 at 16:45
  • $\begingroup$ Good catch, @GiuseppeNegro, I'll update my post! $\endgroup$ – Alex Nelson Jun 25 '12 at 16:52
  • 2
    $\begingroup$ Would you mind including the details as to why the fact that $f(x,t)$ is continuously differentiable implies you can swap the limit and the integral? My own feeling is that this is the real heart of the question. $\endgroup$ – Jason DeVito Jun 25 '12 at 16:58
  • 5
    $\begingroup$ @AlexNelson It's unclear how you used the assumption of continuity to come up with your dominating function $g$ in your first proof. You may want to use mean value theorem together with the fact that continuous functions on compact sets are bounded. $\endgroup$ – Mark Jun 24 '14 at 6:36
  • 2
    $\begingroup$ Also, I agree with Mark. In your DCT part, what is the dominating function $g$? This answer is incomplete without that information. $\endgroup$ – layman May 3 '15 at 15:44

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.