[proofplan]
We compare two descriptions of the same correspondence on the analytic moduli space $Y_0(N)(\mathbb{C})$. First we write an explicit set of representatives for the double coset and identify their action on $\tau \in \mathfrak{H}$. Then we classify the $p+1$ cyclic subgroups of $E_\tau[p]$ and show that the quotients by these subgroups are represented by exactly those same transformed points. The condition $p \nmid N$ ensures that the level-$N$ subgroup survives quotienting by every order-$p$ subgroup.
[/proofplan]
custom_env
admin
[step:Decompose the double coset into $p+1$ analytic branches]Let
\begin{align*}
\alpha_a &:=
\begin{pmatrix}1&a\\0&p\end{pmatrix}
\quad \text{for } a \in \{0,\dots,p-1\},
&
\alpha_\infty &:=
\begin{pmatrix}p&0\\0&1\end{pmatrix}.
\end{align*}
Since $p \nmid N$, the standard double-coset decomposition is
\begin{align*}
\Gamma_0(N)
\begin{pmatrix}1&0\\0&p\end{pmatrix}
\Gamma_0(N)
=
\coprod_{a=0}^{p-1}\Gamma_0(N)\alpha_a
\;\coprod\;
\Gamma_0(N)\alpha_\infty.
\end{align*}
Thus the associated analytic correspondence sends the point $[\tau] \in \Gamma_0(N)\backslash \mathfrak{H}$ to the unordered multiset
\begin{align*}
\left\{
\left[\frac{\tau+a}{p}\right] : a=0,\dots,p-1
\right\}
\cup
\{[p\tau]\}.
\end{align*}[/step]
custom_env
admin
[guided]We need an explicit double-coset decomposition because a correspondence on the quotient $\Gamma_0(N)\backslash \mathfrak{H}$ is computed by choosing right coset representatives. For the matrix
\begin{align*}
\begin{pmatrix}1&0\\0&p\end{pmatrix},
\end{align*}
one may choose the representatives
\begin{align*}
\alpha_a =
\begin{pmatrix}1&a\\0&p\end{pmatrix}
\quad (a=0,\dots,p-1),
\qquad
\alpha_\infty =
\begin{pmatrix}p&0\\0&1\end{pmatrix}.
\end{align*}
The hypothesis $p \nmid N$ is exactly the condition ensuring that these representatives are valid for $\Gamma_0(N)$ without introducing extra congruence classes in the lower-left entry.
The fractional linear action of a matrix
\begin{align*}
\begin{pmatrix}r&s\\t&u\end{pmatrix}
\end{align*}
on $\mathfrak{H}$ is
\begin{align*}
\tau \longmapsto \frac{r\tau+s}{t\tau+u}.
\end{align*}
Therefore
\begin{align*}
\alpha_a \tau = \frac{\tau+a}{p},
\qquad
\alpha_\infty \tau = p\tau.
\end{align*}
Hence the double-coset correspondence sends $[\tau]$ to the unordered multiset
\begin{align*}
\left\{
\left[\frac{\tau+a}{p}\right] : a=0,\dots,p-1
\right\}
\cup
\{[p\tau]\}.
\end{align*}[/guided]
custom_env
admin
[step:Classify the cyclic order-$p$ subgroups of $E_\tau$]Let
\begin{align*}
\Lambda_\tau := \mathbb{Z}\tau+\mathbb{Z}
\end{align*}
and
\begin{align*}
E_\tau := \mathbb{C}/\Lambda_\tau.
\end{align*}
The $p$-torsion group is
\begin{align*}
E_\tau[p]
=
\frac{\frac{1}{p}\Lambda_\tau}{\Lambda_\tau}
=
\left\{
\frac{m\tau+n}{p}+\Lambda_\tau : m,n \in \mathbb{Z}
\right\}.
\end{align*}
Its cyclic subgroups of order $p$ are precisely
\begin{align*}
H_a &:=
\left\langle \frac{\tau+a}{p}+\Lambda_\tau \right\rangle
\quad \text{for } a=0,\dots,p-1,
\\
H_\infty &:=
\left\langle \frac{1}{p}+\Lambda_\tau \right\rangle.
\end{align*}
Indeed, $E_\tau[p]$ is a two-dimensional [vector space](/page/Vector%20Space) over $\mathbb{F}_p$, and its cyclic order-$p$ subgroups are its one-dimensional $\mathbb{F}_p$-subspaces. The displayed list gives the $p$ finite slopes and the vertical slope, hence all $p+1$ such subspaces.[/step]
custom_env
admin
[guided]The quotient isogenies of degree $p$ out of $E_\tau$ are indexed by cyclic subgroups of $E_\tau[p]$ of order $p$. Since
\begin{align*}
E_\tau = \mathbb{C}/\Lambda_\tau,
\qquad
\Lambda_\tau = \mathbb{Z}\tau+\mathbb{Z},
\end{align*}
a point $z+\Lambda_\tau$ is killed by multiplication by $p$ exactly when $pz \in \Lambda_\tau$. Therefore
\begin{align*}
E_\tau[p]
=
\frac{\frac{1}{p}\Lambda_\tau}{\Lambda_\tau}
=
\left\{
\frac{m\tau+n}{p}+\Lambda_\tau : m,n \in \mathbb{Z}
\right\}.
\end{align*}
The two generators $\tau/p+\Lambda_\tau$ and $1/p+\Lambda_\tau$ identify $E_\tau[p]$ with $\mathbb{F}_p^2$.
The one-dimensional subspaces of $\mathbb{F}_p^2$ are the $p$ finite-slope lines and the vertical line. Translating that description back to $E_\tau[p]$, these are
\begin{align*}
H_a =
\left\langle \frac{\tau+a}{p}+\Lambda_\tau \right\rangle
\quad \text{for } a=0,\dots,p-1,
\end{align*}
and
\begin{align*}
H_\infty =
\left\langle \frac{1}{p}+\Lambda_\tau \right\rangle.
\end{align*}
This accounts for exactly $p+1$ cyclic order-$p$ subgroups.[/guided]
custom_env
admin
[step:Identify each quotient elliptic curve with the corresponding transformed lattice]For $a \in \{0,\dots,p-1\}$, define the lattice
\begin{align*}
\Lambda_a := \mathbb{Z}\frac{\tau+a}{p}+\mathbb{Z}.
\end{align*}
Since $\Lambda_\tau \subset \Lambda_a$ and $\Lambda_a/\Lambda_\tau = H_a$, the quotient map
\begin{align*}
q_a: E_\tau \to \mathbb{C}/\Lambda_a,
\qquad
z+\Lambda_\tau \mapsto z+\Lambda_a
\end{align*}
has kernel $H_a$. Hence
\begin{align*}
E_\tau/H_a \cong \mathbb{C}/\Lambda_a = E_{(\tau+a)/p}.
\end{align*}
Similarly, define
\begin{align*}
\Lambda_\infty := \mathbb{Z}\tau+\mathbb{Z}\frac{1}{p}.
\end{align*}
Then $\Lambda_\tau \subset \Lambda_\infty$ and $\Lambda_\infty/\Lambda_\tau = H_\infty$, so
\begin{align*}
E_\tau/H_\infty \cong \mathbb{C}/\Lambda_\infty.
\end{align*}
Multiplication by $p$ gives an isomorphism
\begin{align*}
\mathbb{C}/\Lambda_\infty \longrightarrow \mathbb{C}/(p\Lambda_\infty)
=
\mathbb{C}/(\mathbb{Z}p\tau+\mathbb{Z})
=
E_{p\tau}.
\end{align*}
Therefore the quotient curves obtained from the $p+1$ cyclic subgroups are represented by
\begin{align*}
\frac{\tau+a}{p}
\quad (a=0,\dots,p-1),
\qquad
p\tau.
\end{align*}[/step]
custom_env
admin
[guided]We now compare the subgroup classification with the analytic representatives from the double coset. For $a=0,\dots,p-1$, set
\begin{align*}
\Lambda_a := \mathbb{Z}\frac{\tau+a}{p}+\mathbb{Z}.
\end{align*}
This is a lattice containing $\Lambda_\tau$, because $\tau = p\cdot \frac{\tau+a}{p}-a$ and $1 \in \Lambda_a$. The [quotient group](/page/Quotient%20Group) $\Lambda_a/\Lambda_\tau$ is generated by the class of $(\tau+a)/p$, so
\begin{align*}
\Lambda_a/\Lambda_\tau = H_a.
\end{align*}
The natural quotient map
\begin{align*}
q_a: E_\tau \to \mathbb{C}/\Lambda_a,
\qquad
z+\Lambda_\tau \mapsto z+\Lambda_a
\end{align*}
is a holomorphic group homomorphism whose kernel is exactly $\Lambda_a/\Lambda_\tau = H_a$. Hence
\begin{align*}
E_\tau/H_a \cong \mathbb{C}/\Lambda_a = E_{(\tau+a)/p}.
\end{align*}
For the remaining subgroup, define
\begin{align*}
\Lambda_\infty := \mathbb{Z}\tau+\mathbb{Z}\frac{1}{p}.
\end{align*}
Again $\Lambda_\tau \subset \Lambda_\infty$, and the quotient $\Lambda_\infty/\Lambda_\tau$ is generated by $1/p+\Lambda_\tau$, so it equals $H_\infty$. Thus
\begin{align*}
E_\tau/H_\infty \cong \mathbb{C}/\Lambda_\infty.
\end{align*}
This lattice is not written in the standard form $\mathbb{Z}\tau'+\mathbb{Z}$, so we rescale it. Multiplication by $p$ is a complex-linear isomorphism
\begin{align*}
\mathbb{C}/\Lambda_\infty \longrightarrow \mathbb{C}/(p\Lambda_\infty).
\end{align*}
Since
\begin{align*}
p\Lambda_\infty
=
\mathbb{Z}p\tau+\mathbb{Z},
\end{align*}
we get
\begin{align*}
E_\tau/H_\infty \cong E_{p\tau}.
\end{align*}
Thus the quotient isogenies give exactly the analytic points
\begin{align*}
\left[\frac{\tau+a}{p}\right]
\quad (a=0,\dots,p-1),
\qquad
[p\tau].
\end{align*}[/guided]
custom_env
admin
[step:Check that the level-$N$ subgroup is transported correctly]Let
\begin{align*}
C_\tau :=
\left\langle \frac{1}{N}+\Lambda_\tau \right\rangle
\subset E_\tau
\end{align*}
be the cyclic subgroup of order $N$. For every cyclic subgroup $H \subset E_\tau[p]$ of order $p$, we have
\begin{align*}
C_\tau \cap H = \{0\},
\end{align*}
because every element of $C_\tau$ has order dividing $N$, every element of $H$ has order dividing $p$, and $\gcd(N,p)=1$. Therefore the image
\begin{align*}
(C_\tau+H)/H \subset E_\tau/H
\end{align*}
is cyclic of order $N$. Under the lattice identifications above, this is exactly the level subgroup attached to the transformed point of $Y_0(N)(\mathbb{C})$.[/step]
custom_env
admin
[guided]The modular curve $Y_0(N)$ does not parametrize only elliptic curves; it parametrizes pairs $(E,C)$ where $C \subset E$ is cyclic of order $N$. Thus, after quotienting by an order-$p$ subgroup $H$, we must verify that the level structure remains a cyclic subgroup of order $N$.
Here
\begin{align*}
C_\tau =
\left\langle \frac{1}{N}+\Lambda_\tau \right\rangle
\end{align*}
has order $N$. If $x \in C_\tau \cap H$, then the order of $x$ divides $N$ because $x \in C_\tau$, and also divides $p$ because $x \in H$. Since $p \nmid N$, we have $\gcd(N,p)=1$, so the only possible order of $x$ is $1$. Hence
\begin{align*}
C_\tau \cap H = \{0\}.
\end{align*}
The quotient map $E_\tau \to E_\tau/H$ is therefore injective on $C_\tau$, and its image
\begin{align*}
(C_\tau+H)/H
\end{align*}
is again cyclic of order $N$. Under the identifications
\begin{align*}
E_\tau/H_a \cong E_{(\tau+a)/p},
\qquad
E_\tau/H_\infty \cong E_{p\tau},
\end{align*}
this transported subgroup is precisely the cyclic subgroup represented by the corresponding point of $Y_0(N)(\mathbb{C})$.[/guided]
custom_env
admin
[step:Compare the two correspondences on the open modular curve]
The moduli definition of $T_p$ sends
\begin{align*}
(E_\tau,C_\tau)
\end{align*}
to the unordered multiset
\begin{align*}
\left\{
(E_\tau/H_a,(C_\tau+H_a)/H_a) : a=0,\dots,p-1
\right\}
\cup
\left\{
(E_\tau/H_\infty,(C_\tau+H_\infty)/H_\infty)
\right\}.
\end{align*}
By the preceding identifications, this multiset is represented analytically by
\begin{align*}
\left\{
\left[\frac{\tau+a}{p}\right] : a=0,\dots,p-1
\right\}
\cup
\{[p\tau]\},
\end{align*}
which is exactly the multiset produced by the double-coset correspondence. Hence the Hecke correspondence $T_p$ agrees on $Y_0(N)(\mathbb{C})$ with the correspondence attached to
\begin{align*}
\Gamma_0(N)
\begin{pmatrix}1&0\\0&p\end{pmatrix}
\Gamma_0(N).
\end{align*}
Since $Y_0(N)(\mathbb{C})$ is the open analytic curve inside $X_0(N)(\mathbb{C})$, this proves the asserted agreement on the open curve.
[/step]