[guided]Let
\begin{align*}
\theta=[a_0;a_1,a_2,\dots]
\end{align*}
be the simple continued fraction expansion of the irrational number $\theta$, and let its convergents be
\begin{align*}
\frac{p_n}{q_n},
\qquad
q_n>0.
\end{align*}
For every $n\geq2$, define
\begin{align*}
\alpha_{n+1}:=[a_{n+1};a_{n+2},a_{n+3},\dots],
\qquad
y_n:=\frac{q_n}{q_{n-1}}.
\end{align*}
Then
\begin{align*}
\alpha_{n+1}>1,
\qquad
y_n>1.
\end{align*}
The complete-quotient formula and the determinant identity give
\begin{align*}
\theta=\frac{\alpha_{n+1}p_n+p_{n-1}}{\alpha_{n+1}q_n+q_{n-1}},
\qquad
|p_nq_{n-1}-p_{n-1}q_n|=1.
\end{align*}
Subtracting the two adjacent convergents gives
\begin{align*}
q_n^2\left|\theta-\frac{p_n}{q_n}\right|
=
\frac{1}{\alpha_{n+1}+1/y_n}
\end{align*}
and
\begin{align*}
q_{n-1}^2\left|\theta-\frac{p_{n-1}}{q_{n-1}}\right|
=
\frac{1}{y_n+1/\alpha_{n+1}}.
\end{align*}
The mediant
\begin{align*}
\frac{p_n+p_{n-1}}{q_n+q_{n-1}}
\end{align*}
is reduced because any common divisor of its numerator and denominator divides the determinant
\begin{align*}
p_{n-1}q_n-p_nq_{n-1}.
\end{align*}
Subtracting this mediant from the complete-quotient formula gives
\begin{align*}
(q_n+q_{n-1})^2
\left|\theta-\frac{p_n+p_{n-1}}{q_n+q_{n-1}}\right|
=
\frac{(\alpha_{n+1}-1)(y_n+1)}{\alpha_{n+1}y_n+1}.
\end{align*}
We now use Borel's algebraic lemma. For $x>1$ and $y>1$, at least one of
\begin{align*}
x+\frac{1}{y},
\qquad
y+\frac{1}{x},
\qquad
\frac{xy+1}{(x-1)(y+1)}
\end{align*}
is at least $\sqrt{5}$, and the case where the maximum is exactly $\sqrt{5}$ forces
\begin{align*}
x=y=\varphi:=\frac{1+\sqrt{5}}{2}.
\end{align*}
To prove the lemma, put
\begin{align*}
r:=\sqrt{5}.
\end{align*}
If the first two quantities are at most $r$, then
\begin{align*}
x\leq r-\frac{1}{y},
\qquad
y+\frac{1}{y}\leq r.
\end{align*}
For $y>1$, this gives
\begin{align*}
1<y\leq\varphi.
\end{align*}
The map
\begin{align*}
C:(1,\infty)\times(1,\infty)\to\mathbb R,
\qquad
C(x,y):=\frac{xy+1}{(x-1)(y+1)}
\end{align*}
is decreasing in $x$, since
\begin{align*}
\frac{\partial C}{\partial x}<0.
\end{align*}
Hence the third Borel quantity is minimized at
\begin{align*}
x=r-\frac{1}{y}.
\end{align*}
At that value, the inequality
\begin{align*}
C\left(r-\frac{1}{y},y\right)\geq r
\end{align*}
is equivalent to
\begin{align*}
(r-2)(\varphi-y)(y+\varphi^2)\geq0,
\end{align*}
which holds on $1<y\leq\varphi$. Equality forces $y=\varphi$ and then $x=\varphi$.
Apply the lemma with
\begin{align*}
x=\alpha_{n+1},
\qquad
y=y_n.
\end{align*}
The equality case is impossible because $y_n$ is rational and $\varphi$ is irrational. Therefore one of the three reciprocal constants is strictly larger than $\sqrt{5}$. The corresponding one of the three reduced rational numbers
\begin{align*}
\frac{p_n}{q_n},
\qquad
\frac{p_{n-1}}{q_{n-1}},
\qquad
\frac{p_n+p_{n-1}}{q_n+q_{n-1}}
\end{align*}
satisfies
\begin{align*}
\left|\theta-\frac{p}{q}\right|
<
\frac{1}{\sqrt{5}q^2}.
\end{align*}
As $n$ tends to infinity, all possible denominators in this list tend to infinity, so the construction supplies infinitely many distinct rational approximations.
For sharpness, take
\begin{align*}
\varphi=\frac{1+\sqrt{5}}{2}=[1;1,1,1,\dots].
\end{align*}
Let its convergents be
\begin{align*}
\frac{P_n}{Q_n}.
\end{align*}
The exact error formula gives
\begin{align*}
Q_n^2\left|\varphi-\frac{P_n}{Q_n}\right|
=
\frac{1}{\varphi+\frac{Q_{n-1}}{Q_n}}.
\end{align*}
The denominator ratios converge as
\begin{align*}
\frac{Q_n}{Q_{n-1}}\longrightarrow\varphi,
\end{align*}
so
\begin{align*}
Q_n^2\left|\varphi-\frac{P_n}{Q_n}\right|
\longrightarrow
\frac{1}{\varphi+1/\varphi}
=
\frac{1}{\sqrt{5}}.
\end{align*}
Let
\begin{align*}
C<\frac{1}{\sqrt{5}}.
\end{align*}
Then all sufficiently late convergents satisfy
\begin{align*}
\left|\varphi-\frac{P_n}{Q_n}\right|
\geq
\frac{C}{Q_n^2}.
\end{align*}
If a reduced rational number satisfied
\begin{align*}
\left|\varphi-\frac{p}{q}\right|
<
\frac{C}{q^2},
\end{align*}
then, since
\begin{align*}
C<\frac{1}{2},
\end{align*}
Legendre's continued-fraction criterion would imply that this rational number is a convergent of $\varphi$. Only finitely many early convergents are not excluded by the preceding lower bound. Thus every constant smaller than $1/\sqrt{5}$ gives only finitely many approximations to $\varphi$, so no smaller universal constant can replace the Hurwitz constant.[/guided]