1: \begin{abstract}

2: The aim of this paper is twofold. First, we show that a certain concatenation of a proximity operator with an affine operator

3: is again a proximity operator on a suitable Hilbert space. Second, we use our findings to establish so-called

4: proximal neural networks (PNNs) and stable tight frame proximal neural networks.

5: 

6: Let $\HH$ and $\KK$ be real Hilbert spaces, $b \in \KK$ and 

7: $T \in \mathcal{B} (\HH,\KK)$ a linear operator with closed range and Moore-Penrose inverse $T^\dagger$.

8: Based on the well-known characterization of proximity operators by Moreau, 

9: we prove that for any proximity operator $\Prox \colon \KK \to \KK$ the operator 

10: $T^\dagger \, \Prox ( T \cdot + b)$ is a proximity operator on $\HH$ equipped with a suitable norm.

11: In particular, it follows for the frequently applied soft shrinkage operator 

12: $\Prox = S_{\lambda}\colon \ell_2 \rightarrow \ell_2$ and any frame analysis operator $T\colon \HH \to \ell_2$ that the frame shrinkage operator $T^\dagger\, S_\lambda\, T$ is a proximity operator on a suitable Hilbert space.

13: 

14: The concatenation of proximity operators on $\mathbb R^d$ equipped with different norms establishes 

15: a  PNN.

16: If the network arises from tight frame analysis or synthesis operators, then it forms an averaged operator.

17: In particular, it has Lipschitz constant 1 and belongs to the class of

18: so-called Lip\-schitz networks, which were

19: recently applied to defend against adversarial attacks.

20: Moreover, due to its aver\-aging property, PNNs  can be used within so-called Plug-and-Play algorithms

21: with convergence guarantee.

22: In case of Parseval frames, we call the networks Parseval proximal neural networks (PPNNs). 

23: Then, the involved linear operators

24: are in a Stiefel manifold and corresponding minimization methods can be applied for training of such networks.

25: Finally, some proof-of-the concept examples demonstrate the performance of PPNNs.

26: \end{abstract}