= Solution
There is a necessary distinction between a failure bound averaged over the unknown state, a worst-case bound, and a separate bound for each candidate state. The last interpretation is false: a forger can discard the inputs and always prepare $|\psi_0\rangle^{\otimes N}$. If the actual state is $|\psi_0\rangle$, every authenticity test passes with <probability> one. \b[The meaningful uniform statement concerns worst-case failure, or average failure under any fixed prior giving both states positive probability.] The PDF does not specify that prior; the argument below gives an explicit bound for either interpretation.
Put $c=|\langle\psi_0|\psi_1\rangle|\in(0,1)$. Any unconditional strategy producing $N$ output <qubits>, including ancillas, adaptive <quantum measurements> and correlated outputs, is a <quantum channel> $\mathcal E$. Its inputs and ideal outputs are the <density operators>
$$
\sigma_i=\big(|\psi_i\rangle\langle\psi_i|\big)^{\otimes M},\qquad P_i=\big(|\psi_i\rangle\langle\psi_i|\big)^{\otimes N},\qquad \rho_i=\mathcal E(\sigma_i).
$$
The <projective measurements> on the different banknotes commute, and their all-pass effect is exactly $P_i$. Therefore the <probability> of at least one rejection, conditional on input $i$, is
$$
e_i=1-\operatorname{Tr}(P_i\rho_i).
$$
This remains true for entangled outputs; no independence of their individual test outcomes is assumed.
Use <trace distance> $D(\rho,\sigma)=\tfrac12\|\rho-\sigma\|_1$. The <pure-state trace distance and fidelity identity> gives
$$
D(\sigma_0,\sigma_1)=\sqrt{1-c^{2M}},\qquad D(P_0,P_1)=\sqrt{1-c^{2N}}.
$$
For completeness, two pure-state projectors differ by a <matrix> with nonzero <eigenvalues> $\pm\sqrt{1-|\langle u|v\rangle|^2}$, which proves this identity. Also the <pure-target upper bound on trace distance> gives $D(\rho_i,P_i)\leq\sqrt{e_i}$. To see the mixed-state step, write $\rho_i=\sum_jq_j|u_j\rangle\langle u_j|$ and use <convexity> of the <trace norm> followed by the <Cauchy-Schwarz inequality>:
$$
D(\rho_i,P_i)\leq\sum_jq_j\sqrt{1-\langle u_j|P_i|u_j\rangle}\leq\sqrt{1-\operatorname{Tr}(P_i\rho_i)}.
$$
Finally, <trace-distance contraction under quantum channels> gives $D(\rho_0,\rho_1)\leq D(\sigma_0,\sigma_1)$. This follows from $D(\rho,\sigma)=\max_{0\leq E\leq I}\operatorname{Tr}[E(\rho-\sigma)]$: the adjoint channel is positive and unital, so its image of any effect $E$ is still an effect.
The <triangle inequality> now yields
$$
\begin{aligned}
\sqrt{1-c^{2N}}&\leq D(P_0,\rho_0)+D(\rho_0,\rho_1)+D(\rho_1,P_1)\\
&\leq\sqrt{e_0}+\sqrt{1-c^{2M}}+\sqrt{e_1}.
\end{aligned}
$$
Thus the <cloning failure bound from trace-distance contraction> is
$$
\sqrt{e_0}+\sqrt{e_1}\geq\Delta,\qquad \Delta=\sqrt{1-c^{2N}}-\sqrt{1-c^{2M}}>0,
$$
and in particular $\max(e_0,e_1)\geq\Delta^2/4$. For positive priors $\pi_0+\pi_1=1$, another <Cauchy-Schwarz inequality> gives
$$
(\sqrt{e_0}+\sqrt{e_1})^2\leq(\pi_0e_0+\pi_1e_1)(\pi_0^{-1}+\pi_1^{-1}),
$$
so the average rejection <probability> satisfies
$$
\boxed{P_{\mathrm{fail}}=\pi_0e_0+\pi_1e_1\geq\pi_0\pi_1\Delta^2>p_0:=\tfrac12\pi_0\pi_1\Delta^2>0.}
$$
For equal priors, one may take $p_0=\Delta^2/8$; the same choice is a strict lower bound on worst-case rejection. This proves a quantitative version of the <no-cloning theorem> for the bank's all-pass test. It concerns an unconditional attempt: heralded success cannot be postselected while omitting failed attempts from the accounting.
Back to article page