# Leibniz rule derivation

How is Leibniz Integral rule derived?

$$\frac {\mathrm{d}}{\mathrm{d}x}\left(\int_{a(x)}^{b(x)}f(x, t) \,\mathrm{d}t\right)= f(x,b(x))\frac{\mathrm{d}}{\mathrm{d}x}b(x)- f(x, a(x))\dfrac{\mathrm{d}}{\mathrm{d}x}a(x)+ \displaystyle\int_{a(x)}^{b(x)}\dfrac{\partial f(x,t)}{\partial x} \,\mathrm{d}t.$$

Also, what is the intuition behind this formula?

• I guess I derived a special case of it here math.stackexchange.com/a/3788429/577710? Aug 12, 2020 at 13:14
• This is also exercise 1. in chapter 8.11 of Dieudonné, Foundations of modern Analysis, second edition (1972). Oct 24, 2021 at 23:09

First consider the simplest case where $$a(x)=a$$ and $$b(x)=b$$ for all $$x$$. Then the Leibniz formula becomes $$\frac{d}{dx}\left(\int_a^bf(x,t)dt\right)=\int_a^b\frac{\partial }{\partial x}f(x,t)dx$$ i.e. it is reduced to moving the derivative inside the integral. In this special case, the formula may be proven using the uniform bound on $$\frac{\partial}{\partial x}f(x,t)$$ which is amongst the hypotheses of Leibniz's rule.

Another thing to notice is that by the fundamental theorem of calculus, if we differentiate with respect to the extrema of integration, we have $$\frac{d}{db}\int_a^bf(x,t)dt=f(x,b),\qquad \frac{d}{da}\int_a^bf(x,t)dt=-f(x,a)$$

In the general case, I like to see it as a consequence of the chain rule (i.e. differentiation of a composition of multivariate functions). Suppose $$f(x,t)$$ is defined for $$x\in [\alpha,\beta]$$, and let $$I:=a([\alpha,\beta])$$, $$J:=b([\alpha,\beta])$$, so that $$f(x,t)$$ is defined for all $$t\in I\cup J$$. consider the map \begin{align*}F: [\alpha,\beta]\times I \times J &\to \mathbb{R}\\ (x,a,b)&\mapsto\int_a^bf(x,t)dt \end{align*} as well as the curve \begin{align*}\gamma: [\alpha,\beta]&\mapsto [\alpha,\beta]\times I\times J\\ x&\mapsto (x,a(x),b(x)) \end{align*} Which (by assumption) is differentiable, with derivative given by $$\gamma'(x)=(1,a'(x),b'(x))$$ Finally, using the chain rule, as well as the special cases considered at the beginning: \begin{align*}&\frac{d}{dx}\left(\int_{a(x)}^{b(x)}f(x,t)dt\right)=\frac{d}{dx}(F\circ \gamma)(x)={\nabla F}(\gamma(x))\cdot\gamma'(x)= \\ &=\frac{\partial F}{\partial x}(\gamma(x))+a'(x)\frac{\partial F }{\partial a}(\gamma(x))+b'(x)\frac{\partial F}{\partial b}(\gamma(x))=\\ &=\int_{a(x)}^{b(x)}\frac{\partial f(x,t)}{\partial x}dt-f(x,a(x))a'(x)+f(x,b(x))b'(x) \end{align*} As desired.

• Hi friends, Is it possible to use the same rule to derivate determine $u_x$ and $u_y$, where $$u(x,y)=\int_0^x \int_0^y f(t)dtds?$$ I suppose it is possible but I'm having problems on doing it! Thanks Dec 14, 2021 at 21:46

We have a function $\Phi$of three variables, namely $$\Phi(u,v,w):=\int_u^v f(w,t)\>dt\ ,$$ with the necessary continuity assumptions when $(u,v,w)$ range in some three-dimensional domain $\Omega$. With these givens the function$$g(x):=\Phi\bigl(a(x),b(x),x\bigr)$$ is defined, and we are told to compute its derivative $g'(x)$. By the chain rule we have $$g'(x)=\Phi_{.1}\bigl(a(x),b(x),x\bigr)a'(x)+\Phi_{.2}\bigl(a(x),b(x),x\bigr)b'(x)+\Phi_{.3}\bigl(a(x),b(x),x\bigr)x'(x)\ .$$ Since $$\Phi_{.1}\bigl(u,v,w\bigr)=-f(w,u),\quad \Phi_{.2}\bigl(u,v,w\bigr)=f(w,v),\quad \Phi_{.3}\bigl(u,v,w\bigr)=\int_u^v f_{.1}(w,t)\>dt\ ,$$ whereby only the third part really needs some work, we arrive at the stated formula by plugging everything in.

It remains to prove the "vanilla" Leibniz rule $${d\over dw}\int_u^v f(w,t)\>dt=\int_u^v f_{.1}(w,t)\>dt\ .$$ For this we need that under suitable continuity assumptions the convergence $$\lim_{h\to0}{f(w+h,t)-f(w,t)\over h}=f_{.1}(w,t)$$ is uniform in $t\in[u,v]$.

• Isn't the vanilla Leibniz rule proved with the dominated convergence theorem (DCT)? See the section of https://en.wikipedia.org/wiki/Leibniz_integral_rule that talks about the DCT. I am trying to understand what's the relationship between the DCT and the Leibniz integral rule.
– user168764
Jul 24, 2020 at 14:09
• @nbro: I'm sure there are various proofs. I don't know whether the OP wanted the weakest assumptions for the formula. Jul 24, 2020 at 14:53

\begin{align} f(u,v) &= \frac{\partial}{\partial v} F(u,v) \\ g(u,v) &= \frac{\partial}{\partial u} F(u,v) \\ \lambda (u,v) &= \frac{\partial}{\partial v} g(u,v) \\ &= \frac{\partial^2}{\partial v \, \partial u} F(u,v) \\ &= \frac{\partial^2}{\partial u \, \partial v} F(u,v) \\ &= \frac{\partial}{\partial u} f(u,v) \\ \int f(u,v) \, dv &= F(u,v) \\ \frac{d}{dx} \int f(u,v) \, dv &= u'(x) \frac{\partial}{\partial u} F(u,v)+ v'(x) \frac{\partial}{\partial v} F(u,v) \\ &= u'(x) g(u,v)+v'(x)f(u,v) \\ \frac{d}{dx} \int_{a(x)}^{b(x)} f(x,t) \, dt &= b'(x)f[x,b(x)]-a'(x)f[x,a(x)]+g[x,b(x)]-g[x,a(x)] \\ &= b'(x)f[x,b(x)]-a'(x)f[x,a(x)]+\int_{a(x)}^{b(x)} \lambda (x,t) \, dt \\ &= b'(x)f[x,b(x)]-a'(x)f[x,a(x)]+ \int_{a(x)}^{b(x)} \frac{\partial}{\partial x} f(x,t) \, dt \\ \end{align}