High-Power Partials

Substitution without substitution

When we see

1x2(x2+k) dxint rac{1}{x^2(x^2+k)} dx

We might be tempted to split it with partial fractions into

px+qx2+rx+sx2+k dxint rac{p}{x} + rac{q}{x^2} + rac{rx+s}{x^2+k} dx

But if we solve for this, we’ll find that p=0p=0 and r=0r=0. In fact, pp and qq will always be 00 regardless of what value kk takes.

Something strange is going on here…

Abstraction

Recall that partial fractions is not an integration technique, but an algebraic manipulation, just like trigonometric identities or logarithm laws. Rewriting the integrand with partial fractions produces a mathematically identical expression.

This means we can perform a substitution, but without needing to worry about changing the dxdx or bounds of integration. It’s not a calculus substitution which changes the nature of the expression, but only a visually abstraction to make stuff simpler to manage.

So going back to that integral, let’s just pluck out the integrand and consider that on its own:

1x2(x2+k) rac{1}{x^2(x^2+k)}

We write x2x^2 as tt. Drop the calculus cannons, there’s no dxdx here to replace – we’re just rewriting an expression with another variable for simplicity, like in solving disguised quadratics.

1t(t+k) rac{1}{t(t+k)}

Now it’s evident that we can separate this with partial fractions:

pt+qt+k rac{p}{t} + rac{q}{t+k}

Maybe you’re thinking, “Wait, but tt is x2x^2, what about the lower powers of xx?” But tt is just a variable. Forget xx ever existed; this relation holds true for all values of tt, and that’s all that matters.

So now we replace tt with x2x^2 to undo our substitution. The relation still holds for all values of xx.

px2+qx2+k rac{p}{x^2} + rac{q}{x^2+k}

This is exactly the same result that we derived before, but we simplified things a lot by using tt instead of x2x^2.

Application

This can feel a little unintuitive at first, especially if you’ve been integrating for a while – we’ve been trained to consider derivatives when substituting. But this is a more classic kind of ‘symbolic’ substitution than the involved process of substitution in calculus.

It’s important to recognise where this works. It certaiy does not mean you don’t ever need to consider lower powers of xx. This:

1x3(x2+1) rac{1}{x^3(x^2+1)}

still decomposes like

=px+qx2+rx3+sx+kx2+1= rac{p}{x} + rac{q}{x^2} + rac{r}{x^3} + rac{sx+k}{x^2+1}

Whereas with this:

1x3(x3+1) rac{1}{x^3(x^3+1)}

we can skip those lower powers by writing x3x^3 as tt.

1t(t+1)=pt+qt+1 rac{1}{t(t+1)} = rac{p}{t} + rac{q}{t+1}

It seems a bit obvious to state “you must be able to write every term in terms of integer powers of the new variable tt”, but that’s essentially what it is. It’s only when the powers of xx have a common factor1 that you can rewrite those powers of xx as lower powers of tt.

Here’s another example. We’ll split

1x8(x4+2) rac{1}{x^8(x^4+2)}

We spot that all terms of xx can be expressed as a power of x4x^4, so writing x4x^4 as tt we obtain

1t2(t+2) rac{1}{t^2(t+2)}

Decomposing with partial fractions gives

pt+qt2+rt+st+2 rac{p}{t} + rac{q}{t^2} + rac{rt+s}{t+2}

And after re-substitution we end up with

px4+qx8+rx4+sx4+2 rac{p}{x^4} + rac{q}{x^8} + rac{rx^4+s}{x^4+2}

Augmentation

So, remembering that partial fractions is purely a symbolic manipulation, guess what – we can do it for any expression!

Take a fraction with trig:

1(sinx)(sinx+1) rac{1} { left(sin{x} ight) left(sin{x} + 1 ight) }

Now partial fractions here would be crazy2 but we’ll just do it to demonstrate the method.

Write sinx=t\sin{x} = t, so we have:

1t(t+1) rac{1}{t(t+1)}

Writing as partial fractions gives:

1t(t+1)=pt+qt+1 rac{1}{t(t+1)} = rac{p}{t} + rac{q}{t+1}

And solving gives:

=1t1t+1= rac{1}{t} - rac{1}{t+1}

Putting the trig back, we now find

1(sinx)(sinx+1)=1sinx1sinx+1 rac{1}{ left(sin{x} ight) left(sin{x} + 1 ight) } = rac{1}{sin{x}} - rac{1}{sin{x}+1}

Both of these remaining integrals are quotable, though not necessarily any easier than our original integral. But look, we did partial fractions with trig! Isn’t that cool.

It wouldn’t have worked with both sin(x)\sin(x) and cos(x)\cos(x), of course, since we need every expression of xx to be the same or have a common factor. Although, speaking of trigonometric partial fractions, maybe we could…


  1. The common factor can’t just be 11, since that would be a pointless substitution of x=tx = t
  2. A much nicer method would be to multiply through by (1sinx)(1-\sin{x}).