A function takes an input and produces an output. Its inverse runs the process backward: given the output, it recovers the input. If f(a)=b, then f−1(b)=a.
For an inverse to exist as a function, f must be one-to-one: each output comes from exactly one input. If two different inputs gave the same output, the inverse would not know which one to return.
Trig functions fail this! Consider sinθ=21. This is true for θ=6π, but also 65π, and 6π+2π, and infinitely many more. Asking "what angle has sine 21?" has no single answer. So sine, cosine, and tangent have no inverse on their full domains.
We will use both inverse (e.g., sin−1) and arc (e.g., arcsin) notation throughout this module; both mean the same thing.
Principal Ranges
We repair the above caveat using principal ranges, i.e, an interval where the function isone-to-one. On that interval the function passes the horizontal-line test, and an inverse exists. The chosen interval is called the principal range, and its outputs are the principal values.
For sin, we restrict to [−2π,2π], where the function covers its entire range without repetition. This gives the inverse sine, written arcsin (or sin−1):
[−1,1]→[−2π,2π].
For cosine, [−2π,2π] would not work (cosine is symmetric there), so we use [0,π].
arccos:[−1,1]→[0,π].
For tangent, which covers every real value on (−2π,2π), we restrict to that open interval:
arctan:R→(−2π,2π).
Don't Get Trapped!
It is tempting to think arcsin(sinθ)=θ always. It does not. The output must land in the principal range, so if θ starts outside it, you must reduce. For example:
arcsin(sin43π)=4π,
because sin43π=22, and the angle in [−2π,2π] with that sine is 4π, not 43π. Going the other way, sin(arcsinx)=x always holds for x∈[−1,1], because that composition does not leave the range.
Evaluating Compositions
Mixed compositions like cos(arcsinx) look intimidating but are actually easy to deal with. Set θ=arcsinx, so sinθ=x with θ∈[−2π,2π]. Recall the Pythagorean identity:
sin2θ+cos2θ=1
Substitute x for sinθ:
x2+cos2θ=1
cosθ=1−x2
We take the positive root because θ∈[−2π,2π], cosine is never negative. Therefore:
cos(arcsinx)=1−x2
Similarly, for sin(arccosx):
Set θ=arccosx, so cosθ=x with θ∈[0,π]. Substitute into the identity:
sin2θ+x2=1
sinθ=1−x2
Here we take the positive root because θ∈[0,π], where sine is never negative. Therefore:
sin(arccosx)=1−x2
For tangent compositions we use tanθ=cosθsinθ. To find cos(arctanx), set θ=arctanx, so tanθ=x with θ∈(−2π,2π). Divide the Pythagorean identity by cos2θ:
tan2θ+1=cos2θ1
Substitute x for tanθ and solve:
x2+1=cos2θ1
cosθ=1+x21
The positive root holds because cosine is positive on (−2π,2π). Then sinθ=xcosθ gives the following results (you might want to memorize these!):
cos(arctanx)=1+x21,sin(arctanx)=1+x2x
sin(arctanx)=1+x2x,cos(arctanx)=1+x21.
Key Identities
Here are a couple nice identites that pop up from time to time in contests. This topic is pretty niche, but everything's worth knowing:
arcsinx+arccosx=2π,x∈[−1,1].
Proof: if θ=arcsinx then cos(2π−θ)=sinθ=x, and 2π−θ∈[0,π], so 2π−θ=arccosx.
It is also worth mentioning
arctana+arctanb=arctan(1−aba+b)+kπ,
where the correction term is
k=⎩⎨⎧01−1ab<1ab>1,a>0ab>1,a<0.
We bring in kπ as when ab>1 the sum leaves the arctan range, so the output is off by π. When ab<1 with both terms positive, you can drop k.
Telescoping with Arctangent
We use the above formula backwards to attain:
arctan(n2+n+11)=arctan(n+1)−arctan(n).
This holds because, with a=n+1 and b=−n, we get 1−aba+b=1+n(n+1)1=n2+n+11, and ab=−n(n+1)<1, as desired.
2013 AIME I · Problem 8: The domain of f(x)=arcsin(logm(nx)) is a closed interval of length 20131, where m and n are positive integers and m>1. Find the remainder when the smallest possible m+n is divided by 1000.
The whole problem hinges on the domain of arcsin. Its input must satisfy −1≤t≤1, so we need
−1≤logm(nx)≤1.
Undo the logarithm by raising each side to the power of m:
m−1≤nx≤m1.
Divide by n:
mn1≤x≤nm.
This is the domain. Its length is the right end minus the left:
nm−mn1=mnm2−1.
Set that equal to 20131:
mnm2−1=20131⟹n=m2013(m2−1).
For n to be an integer, m must divide 2013(m2−1). But m shares no factor with m2−1, so m must divide 2013=3⋅11⋅61.
Now minimize the sum. Substitute the expression for n:
m+n=m+m2013(m2−1)=2014m−m2013.
This grows as m grows, so we want the smallest divisor of 2013 with m>1, namely m=3. Then
n=32013(9−1)=32013⋅8=5368,
giving m+n=5371. The remainder when divided by 1000 is 371.