Integration by parts
Integration by parts is the product rule used backwards1. We check it with ∫xe−xdx.
The formula
∫fg′dx=fg−∫f′gdx It comes from integrating both sides of the product rule (fg)′=f′g+fg′ and rearranging. One chooses f and g′ so that the integral left on the right is simpler than the original.
Working it out
Taking f=x and g′=e−x gives f′=1 and g=−e−x, so the result is as follows.
∫xe−xdx=−xe−x+∫e−xdx=−(x+1)e−x+C Differentiating to check gives −e−x+(x+1)e−x=xe−x, returning the original.
Making the choice
| Choice | Integral left over | Outcome |
|---|
| f=x, g′=e−x | ∫e−xdx | simpler |
| f=e−x, g′=x | ∫2x2e−xdx | the degree goes up |
The knack is to put the factor that gets simpler under differentiation into f.
It works over an infinite interval
Since −(x+1)e−x→0 as x→∞, the following holds.
∫0∞xe−xdx=0−(−1)=1 That says the mean of the exponential distribution is 1.
Reading in a factor of 1
For ∫logxdx, taking f=logx and g′=1 gives xlogx−∫1dx=xlogx−x. Even when nothing looks like a product, reading in a factor of 1 lets integration by parts apply.
The type that comes back after two rounds
In ∫exsinxdx, two rounds of integration by parts bring the original integral back on the right. Solving the result as an equation gives the following.
∫exsinxdx=2ex(sinx−cosx)+C It looks like going in circles, yet it is a rare type that solves itself.
The humped curve on the graph is y=xe−x, the curve rising from below toward 0 is the antiderivative y=−(x+1)e−x, and the large dots are the local maximum and (0,−1).
- Integration by parts, Wikipedia