I'm assuming you're referring to the problems I label "tricky trig".
If you have something like integral(1/sqrt(9-x^2))dx, you're kinda screwed. u-substitution won't work. integration by parts (u dv) doesn't work either.
However, look at the values that x can take on... x has to be between -3 and 3.
Now, for the tricky trig part:
You have your pythagorean identities:
sin^2 + cos^2 =1
tan^2 + 1 = sec^2
watch what happens to the square root if I make this tricky trig substitution: let x = 3sinu
Shouldn't be a problem, because the value of sin of an angle vary between -1 and 1. 3 times these values makes it between -3 and 3. Thus, we're not "stepping out of bounds, running up the sideline, and stepping back in bounds." Such a weird substitution ought to be allowed. Seems weird, but watch...
x^2 becomes 9sin^2(u)
and 9 - 9sin^2(u) can be factored into
9(1-sin^2(u)) and, heyyy, 1-sin^2(u) is the same as cos^2(u) --- it's a pythagorean trig identity.
So, suddenly our ugly square root has transformed into
sqrt(9 cos^2(u) ) that's the square root of a perfect square.
= 3cos(u)
Now, when you're making a substitution like this, you'll have to remember to substitute for dx as well.
Since x = 3sinu
dx = 3cosu du
And, if there are other x's in the integral, substitute for them as well.
--------------------
if you run into (a+x^2) inside a square root, take advantage of another identity:
1+tan^2 = sec^2
So, for x you'll substitute sqrt(a) tan u
then, when substituting for x^2, you'll have (a + a tan^2(u) )
factor out the a, substitute for 1+tan^2, and you're all set to proceed.
(also substitute for dx and any other x's.)
----
There's a 3rd case: x^2 - a
you'll take advantage of sec^2 - 1 = tan^2
in a similar fashion.
Note, you don't necessarily have to have square roots present to take advantage of these substitutions. i.e. if you let x = sqrt(a) * tan u, the value of the right side can be anything (if the domain of x also allows it to take on any values and isn't restricted by a square root.)
Helpful?