You already know that the last two sets of connectives are functionally complete! $\{\neg, \to\} = \{\to, \neg\}$, for example.
The only remaining part of the puzzle is {∧,∨}. This set of connectives is not functionally complete. To show this, it's enough to show that some truth function can't be expressed using only these two connectives.
First, a few conventions and definitions. Assume that the symbols of the language $\mathscr{L}$ are the connectives, parentheses and propositional variables $p_0, p_1, \dots, p_n, \dots$, and that formulas are built up in the usual way. The variables of a formula $\varphi$ are all the $p_i$ occurring in $\varphi$.
Let $\mathsf{TV} = \{0,1\}$ be the set of truth values. Given an integer $n$, call the members of $\mathsf{TV}^n$ *valuations: a valuation $v \in \mathsf{TV}^n$ can be thought of as assigning the value $v_i$ to the $i$th variable $p_i$. $n$ is large enough for a formula $\varphi$ if for every variable $p_i$ occurring in $\varphi$, $i < n$. A valuation $v$ is long enough for $\varphi$ if its length is large enough.
The value of a formula under a valuation that's long enough for it is defined by induction on the construction of formulas:
- $Val(p_i, v) = v_i$
- $Val((\varphi \wedge \psi), v) = Val(\varphi, v) \cdot Val(\psi, v)$
- $Val((\varphi \vee \psi), v) = \max(Val(\varphi, v), Val(\psi, v))$
Finally, an $n$-ary truth function is a function $\mathsf{TV}^n \to \mathsf{TV}$, i.e. a mapping from $n$-ary valuations to truth values.
Every formula $\varphi$ induces truth functions $f_{\varphi} \colon \mathsf{TV}^n \to \mathsf{TV}$ for every large enough $n$:
- $f_{p_i}(v) = v_i$
- $f_{(\varphi \wedge \psi)}(v) = f_{\varphi}(v) \cdot f_{\psi}(v)$
- $f_{(\varphi \vee \psi)}(v) = \max(f_{\varphi}(v), f_{\psi}(v))$
To show that $\{\wedge, \vee\}$ is not functionally complete, show that:
For any formula $\varphi$ of $\mathscr{L}$, if $v = (1)_{i<n}$ is a constantly-$1$ valuation that's long enough for $\varphi$, then $f_{\varphi}(v) = 1$.
The proof is by induction on the construction of formulas, and follows easily from the inductive definition of $f_{\varphi}$. From this, the result is immediate:
If $g \colon \mathsf{TV}^n \to \mathsf{TV}$ is a truth function taking the value $0$ on the constantly-$1$ valuation $(1)_{i<n}$, then there is no formula $\varphi$ of $\mathscr{L}$ such that $g = f_{\varphi}$. That is, $g$ is not represented by any formula of $\mathscr{L}$.