Some integrands are a product of two very different functions — xex, xcosx, logx, xsin−1x — where substitution gets you nowhere. Integration by parts is the tool for these. It comes straight from reversing the product rule for differentiation.
Starting from dxd(uv)=uv′+u′v and integrating both sides gives the working formula:
∫udxdvdx=uv−∫vdxdudx.
In words: integral of (first × derivative-of-second) = first × integral-of-second − integral of (derivative-of-first × integral-of-second).
Choosing u: the ILATE rule
The whole game is picking which factor is u (to differentiate) and which is dv (to integrate). Pick u by ILATE — the first type that appears:
(i) uses dv=1−x2xdx; (ii) substitutes t=x2 then applies Bernoulli; (iii)-(iv) use the double-angle identities tan−11−u22u=2tan−1u and sin−11+u22u=2tan−1u. …
Part (i) is a by-parts step where dv is chosen as the piece with a ready antiderivative; part (ii) needs a substitution before Bernoulli's formula applies; parts (iii)-(iv) simplify first via inverse-trig double-angle identities, then integrate tan−1 by parts.
Part (i): 1−x2xsin−1x. Take u=sin−1x and dv=1−x2xdx; since dxd[−1−x2]=1−x2x, v=−1−x2, and du=1−x2dx.