Let $f:X\rightarrow Y$ be a function where $|X| = n$ and $|Y| = 3$.
I think it is $n(n-1)(n-2)$ since for the first element in $Y$, I pick out of $n$ possible elements that map to this first element and so on.
Is this the correct line of thinking?
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It only takes a minute to sign up.
Sign up to join this communityLet $f:X\rightarrow Y$ be a function where $|X| = n$ and $|Y| = 3$.
I think it is $n(n-1)(n-2)$ since for the first element in $Y$, I pick out of $n$ possible elements that map to this first element and so on.
Is this the correct line of thinking?
Okay, so you write down $Y = \{y_1, y_2, y_3\}$ and you pick out an element $x_1$ of $X$ and you say that $f(x_1) = y_1$ and you pick out a second element, $x_2$ and you say that $f(x_2) = y_2$, and you pick a third and say $f(x_3) = y_3$. That's fine, you've defined $f$ on three values but what about the rest of $X$? How do you define $f$ on $X - \{x_1,x_2,x_3\}$? What if you have two elements of $X$, say $x_1$ and $x_1'$ such that $f(x_1) = y_1$ and $f(x_1') = y_1$, which one are you picking? How do you distinguish between them?
Trevor Gunn has explained why your approach leads to an incorrect answer.
Observe that if we did not have the restriction that $f: X \to Y$ is surjective, we would have three choices for the image of each of the $n$ elements of $X$, so there are $3^n$ functions $f: X \to Y$. From these, we must exclude those in which there are fewer than three elements in the range.
There are $\binom{3}{k}$ ways to exclude $k$ of the $3$ elements in the codomain from the range and $(3 - k)^n$ ways to map the $n$ elements of $X$ to the remaining $3 - k$ elements in the codomain. Hence, by the Inclusion-Exclusion Principle, the number of surjective functions $f:X \to Y$ is $$\sum_{k = 0}^{3} (-1)^k\binom{3}{k}(3 - k)^n = \binom{3}{0}3^n - \binom{3}{1}2^n + \binom{3}{2}1^n - \binom{3}{3}0^n = 3^n - 3 \cdot 2^n + 3$$
$\sum _{i=1}^{n-2}\sum_{j=1}^{n-i-1} \binom{n}{i}\binom{n-i}{j}$ where $i$ of the $n$ elements of $X$ are mapped to $1$, $j$ of the remaining $n-i$ elements are mapped to $2$ and the remaining $n-i-j$ elements are mapped to $3$ and where $ Y= \{1, 2, 3\}$
Observe that this answer is equivalent to the answer by @N. F. Taussig as follows. $\begin {align} \sum _{i=1}^{n-2}\sum_{j=1}^{n-i-1} \binom{n}{i}\binom{n-i}{j} &= \sum _{i=1}^{n-2}\binom{n}{i}\sum_{j=1}^{n-i-1} \binom{n-i}{j}\\ &= \sum _{i=1}^{n-2}\binom{n}{i}(2^{n-i}-2)\\ &= \sum _{i=1}^{n-2}\binom{n}{i}2^{n-i}-2\sum _{i=1}^{n-2}\binom{n}{i}\\ &= \sum _{i=0}^{n}\binom{n}{i}2^{n-i}-(1+2n+2^n)-2[\sum _{i=0}^{n}\binom{n}{i}-(1+n+1)]\\ &= 3^n - 3\times2^n +3 \end{align}$
One way of thinking about it, which ties into other areas which you may have studied or may study in the future, is that if $f:X\to Y$ is surjective then the inverse $f^{-1}$ creates a set partition of $X$ into $|Y|$ non-empty subsets, or an equivalence relation on $X$ with $|Y|$ equivalence classes.
In particular, the number of partitions of a set of $n$ into $k$ non-empty subsets is important enough to be named: a Stirling number of the second kind, commonly denoted ${n \brace k}$.
As Stephen Meskin pointed out in a comment, this isn't quite the full story: the correspondence between the parts of each partition and the elements of $|Y|$ allows for an arbitrary permutation, so the final answer is $|Y|! {|X| \brace |Y|}$
Further reading: