## 1 Introduction

Reachability, connectivity, and robustness in networks depend often on time.
For instance, in public transport or human contact networks, available connections or contacts are time-dependent.
To model such time-dependent aspects, one turns from static graphs to temporal graphs.
Formally, an undirected *temporal graph* is an ordered triple consisting of a set of vertices, a set of *time-edges*, and a maximal time label .
We study the problem of finding a small set of vertices in a temporal graph whose removal disconnects two designated terminals: a classic, polynomial-time solvable problem in (static) graph theory.

Herein, a vertex set is a *temporal -separator* if there is no temporal -path in .
A *temporal -path* of
length in is a sequence
, of time-edges in ,
where , , for all with , and for all .
Temporal -Separation is NP-hard [kempe2000connectivity].
In this work, we study Temporal -Separation on restricted classes of temporal graphs with the goal to identify computationally tractable cases.

So far, in the literature one basically finds two different directions concerning the definition of temporal graph classes. One direction is defining temporal graph classes through the underlying graph (that is, essentially, the graph obtained by forgetting about the time labels of the edges) [erlebach2015temporal, AxiotisF16, zschoche2017computational]. Herein, one restricts the input temporal graph to have its underlying graph being contained in some specific graph class. The other direction consist of properties expressible through temporal aspects [casteigts2012time, FlocchiniMS13, KuhnLO10, michail2016traveling]. Such properties are, for instance, each layer being a subgraph of its succeeding layer, or the temporal graph being periodic, that is, having a subsequence of layers which is repeated in the same order for some periods. In this work, we study Temporal -Separation on temporal graph classes from both directions.

### Our contributions.

We show that Temporal -Separation remains NP-complete on many restricted temporal graph classes.