Temporal Description Logics over Finite Traces Alessandro Artale, Andrea Mazzullo, Ana Ozaki KRDB Research Centre, Free University of Bozen-Bolzano, Italy surname@inf.unibz.it Abstract. Linear temporal logic interpreted over finite traces (LTLf ) has been used as a formalism for temporal specification in automated planning, process modelling and (runtime) verification. In this paper, we lift some results from the propositional case to a decidable fragment of first-order temporal logic obtained by combining the description logic ALC with LTLf , denoted as TfU ALC. We show that the satisfiability problem in TfU ALC is ExpSpace-complete, as already known for the infinite trace case, while it decreases to NExpTime when we consider finite traces with a fixed number of instants, and even to ExpTime when additionally restricting to globally interpreted TBoxes. Moreover, we investigate two model-theoretical notions for TfU ALC: we show that, differently from the infinite trace case, it enjoys the bounded model property; we also discuss the notion of insensitivity to infiniteness, providing characterisations and sufficient conditions to preserve it. 1 Introduction Since the introduction of linear temporal logic (LTL) by Pnueli [33], several (propositional and first-order) LTL-based formalisms have been developed for applications such as automated planning [11, 12, 15–17], process modelling [1, 31] and (runtime) verification of programs [14, 24, 32–35]. Related research in knowledge representation has focussed on decidable fragments of first-order LTL [28], many of them obtained by combining LTL with description logics (DLs), and for this reason called temporal description logics (TDLs) [3, 4, 30, 36]. In this latter setting, the complexity of satisfiability problems can range from ExpSpace-complete, for the full-fledged combination of the description logic ALC with LTL [23, 27], down to ExpTime- or PSpace-complete, by restricting the application of temporal operators [10, 23], or even in NP or NLogSpace, by weakening both the temporal and the description logic components, for some TDLs used in temporal conceptual data modelling [5, 6]. Besides the usual LTL semantics defined over the natural numbers, rececently attention has been devoted to study LTL semantics based on finite traces, which are temporal structures having only a finite number of time points, with the result- ing logic sometimes known as linear temporal logic over finite traces (LTLf ) [13, 18–22]. Indeed, for several of the already mentioned applications, the finiteness of the time dimension is a fairly natural restriction, since the processes involved are usually required to terminate within a finite amount of time. Motivated by similar applications, some recent work has considered TDLs in the context of runtime verification [8] and business process modelling [2]. Nonetheless, these attempts are still based on the infinite trace semantics for temporal operators (sometimes allowed only over axioms [8]). On the way to overcome such limitations, in this work we combine DLs with LTLf . In particular, we consider the logic TfU ALC defined as the DL ALC extended with temporal operators on both concepts and formulas (but without temporalised or rigid roles), and interpreted over models with a finite time dimension. We show that the complexity of the satisfiability problem in TfU ALC is ExpSpace-complete, as for the infinite case, and lowers down to NExpTime if we restrict to traces with a bound k on the number of instants, and even to ExpTime for a fragment with only globally interpreted TBoxes. We then consider two model-theoretical notions. Firstly, we show that TfU ALC has the bounded model property, with the bound being double exponential in the size of the formula. We then study the notion of a formula being insensitive to infiniteness, considered in [19, 8, 13, 14] for dealing with the possibility to preserve the satisfiability of formulas from the finite to the infinite case and thus to reuse TDL algorithms developed for the infinite case. This paper is organised as follows. In Section 2, we introduce TU ALC and we provide the semantics both over infinite and over finite traces. In Section 3 we study the complexity of the satisfiability problem for TfU ALC formulas, while Section 4 refines these results for traces with fixed k instants. Section 5 investigates model-theoretical notions of TfU ALC formulas. Finally, in Section 6, we conclude with some open problems and possible directions for future work. 2 Temporal Description Logics In the following, we first define the temporal language TU ALC as a temporal extension of the description logic ALC. The language TU ALC is obtained by extending ALC with the binary temporal operator until (U). We then define the semantics based on traces, which are temporal structures having either a finite or an infinite number of time points. We denote by TfU ALC the language TU ALC interpreted over finite traces. We further add the upperscript k and write TkU ALC to indicate that the formulas are interpreted over finite traces with at most k ∈ N, k > 0, time points. Syntax. Let NC , NR and NI be countably infinite and pairwise disjoint sets of concept names, role names, and individual names, respectively. A TU ALC concept is an expression of the form: C, D ::= A | ¬C | C u D | ∃R.C | C U D, where A ∈ NC and R ∈ NR . For instance, the expression A U (B u ∃R.¬A) is a TU ALC concept. A TU ALC axiom is either a concept inclusion (CI) of the form C v D, or an assertion of the form A(a) or R(a, b), where C, D are TU ALC concepts, A ∈ NC , R ∈ NR , and a, b ∈ NI . TU ALC formulas are of the form: ϕ, ψ ::= α | ¬ϕ | ϕ ∧ ψ | ϕ U ψ, where α is a TU ALC axiom. Semantics. A (TDL) interpretation is a structure I = (∆I , (In )n∈I ), where I is an interval of the form [0, ∞) or [0, l], with l ∈ N, and each In is a classical DL interpretation with domain ∆I : we have AIn ⊆ ∆I , RIn ⊆ ∆I ×∆I , and aIi = aIj ∈ ∆I for all a ∈ NI and i, j ∈ N, i.e., the interpretation of individual names is fixed (in the following denoted as aI ). We refer to a temporal interpretation I = (∆I , (In )n∈I ) as a finite trace if I = [0, l], and as an infinite trace if I = [0, ∞). The stipulation that all time points share the same domain ∆I is called the constant domain assumption (meaning that objects are not created or destroyed over time), and it is the most general choice in the sense that increasing, decreasing, and varying domains can all be reduced to it [23]. We now define the semantics of TU ALC considering I = [0, ∞). The semantics of TfU ALC is defined in the same way except that I is of the form [0, l], for some l ∈ N. For a fixed k ∈ N, k > 0, we define the semantics of TkU ALC considering I = [0, l] with l ≤ k − 1. The interpretation of concepts at instant n of interpretation I = (∆I , (In )n∈I ) is defined as follows: (¬C)In = ∆I \ C In , (C u D)In = C In ∩ DIn , (∃R.C)In = {d ∈ ∆I | ∃d0 ∈ C In : (d, d0 ) ∈ RIn }, (C U D)In = {d ∈ ∆I | ∃m ∈ I, m > n : d ∈ DIm ∧ ∀i ∈ (n, m) : d ∈ C Ii }. We say that a concept C is satisfied in I if C I0 6= ∅. Satisfaction of a formula ϕ in I at time point n ∈ I (written I, n |= ϕ) is inductively defined as follows: I, n |= C v D iff C In ⊆ DIn , I, n |= ¬φ iff not I, n |= φ, I, n |= A(a) iff aI ∈ AIn , I, n |= φ ∧ ψ iff I, n |= φ and I, n |= ψ, I, n |= R(a, b) iff (aI , bI ) ∈ RIn , I, n |= φ U ψ iff ∃m ∈ I, m > n : I, m |= ψ, and ∀i ∈ (n, m) : I, i |= φ We say that ϕ is satisfied in I, writing I |= ϕ, if I, 0 |= ϕ. A formula ϕ logically implies a formula ψ if every I that satisfies ϕ satisfies also ψ, and we write ϕ |= ψ. Moreover, ϕ is said to be TfU ALC satisfiable (or satisfiable over finite traces) if it is satisfied in some finite trace. Given a concept C and a formula ϕ, we say that C is satisfiable over finite traces with respect to ϕ if there is a finite trace I such that ϕ is satisfied in I and C is satisfied in I, in symbols ϕ 6|= C v ⊥. Concept satisfiability w.r.t. ϕ can be reduced to formula satisfiability by checking whether the following formula is satisfiable: ϕ ∧ C(a), for a fresh a ∈ NI . We will use the following standard equivalences for concepts: ⊥ ≡ A u ¬A, > ≡ ¬⊥; ∀R.C ≡ ¬∃R.¬C; (C t D) ≡ ¬(¬C u ¬D); #C ≡ #1 C ≡ ⊥ U C; #n C ≡ # #n−1 C, with n > 1; 3C ≡ > U C; 2C ≡ ¬3¬C. Similar abbreviations also hold for formulas. We often write F = (∆F , (Fn )n∈[0,l] ) to indicate that a trace is finite and I = (∆I , (In )n∈I ) when it is infinite. 3 Satisfiability over Finite Traces In the following we show how to reduce the formula satisfiability problem over finite traces to the same problem over infinite traces. Similar to the encoding proposed in [20] for propositional LTL, we introduce a fresh concept name E to capture the finiteness of the temporal dimension. The fundamental properties regarding the end of time can be described as follows. (i) There is a least one instant before the end of time (E is initially false). (ii) The end of time comes for all objects (E is unavoidable). (iii) The end of time comes at the same time for every object (E is synchronous). (iv) The end of time never goes away (E is persistent). We axiomatise the above end of time properties: ψf1 = > v ¬E, ψf2 = (> v ¬E) U (> v E), ψf3 = 2(E v #E). Intuitively, ψf1 captures Point (i), ψf2 captures Points (ii), (iii), and ψf3 captures Point (iv). In the following, we use ψf = ψf1 ∧ ψf2 ∧ ψf3 . We now show that satisfiability in TfU ALC is ExpSpace-complete. We start by characterising, in Lemma 1, models satisfying the end of time formula, ψf . Our characterisation uses the following notion of an extension of a finite trace. Let F = (∆F , (Fn )n∈[0,l] ) and I = (∆I , (In )n∈[0,∞) ) be, respectively, a finite and an infinite trace s.t. ∆F = ∆I (with a small abuse, we will write just ∆) and aF = aI , for all a ∈ NI . We denote by F · I = (∆F·I , (F · In )n∈[0,∞) ) the extension of F with I defined as the infinite trace with ∆F·I = ∆, aF ·I = aF , for all a ∈ NI , and for n ∈ N: ( ( F ·In X Fn , if n ∈ [0, l] F ·In ∅, if n ∈ [0, l] X = , E = , X In−(l+1) , if n ∈ [l + 1, ∞) ∆, if n ∈ [l + 1, ∞) where X ∈ (NC \ {E}) ∪ NR and E is the concept name encoding the end of time. Clearly, the interpretation of E characterises the satisfiability of ψf . Lemma 1. For every infinite trace I, I |= ψf iff I = F · I0 , for some finite trace F and some infinite trace I0 . We now introduce a translation ·† of TfU ALC concepts and formulas into TU ALC concepts and formulas, respectively, such that, together with the end of time formula, ψf , captures those concepts and formulas satisfiable over finite traces. More formally, a concept C (formula ϕ) of TfU ALC is satisfiable if and only if its translation C † (ϕ† ) is satisfiable in a TU ALC model that also satisfies the formula ψf . Our translation ·† is defined as: (α)† 7→ α (¬β)† 7→ ¬β † (C u D)† 7→ C † u D† (φ ∧ ψ)† 7→ φ† ∧ ψ † (∃R.C)† 7→ ∃R.C † (C v D)† 7→ C † v D† (C U D)† 7→ C † U (D† u ¬E) (φ U ψ)† 7→ φ† U (ψ † ∧ ¬(> v E)) where α can be an assertion or a concept name A ∈ NC , β can be a concept C or a formula ϕ, and R ∈ NR . Lemma 2 states the correctness of ·† . Lemma 2. Let F be a finite trace, and let F · I be an extension of F. Then, for every TfU ALC formula ϕ, we have: F |= ϕ iff F · I |= ϕ† . From the previous lemmas, we obtain a reduction of the TfU ALC satisfiability problem to the corresponding problem for TU ALC. Theorem 3. Let ϕ be a TfU ALC formula. Then ϕ is satisfiable iff ϕ† ∧ ψf is a satisfiable TU ALC formula. Since it is known that the TU ALC formula satisfiability problem is ExpSpace- complete [23, Theorem 14.15], the above theorem gives us an ExpSpace upper bound for checking satisfiability in TfU ALC. The lower bound can be proved using similar ideas as those used to prove hardness of TU ALC. We are now ready to state the main theorem of this section. Theorem 4. Satisfiability in TfU ALC is ExpSpace-complete. 4 Traces with at Most k Time Points In this section we study the satisfiability in TkU ALC, where we restrict the problem to traces with at most k time points, with k given in binary, as part of the input. We provide for this case a lower NExpTime complexity result which further reduces to ExpTime when just global CIs are allowed. To prove our upper bound for the satisfiability problem in this setting we use quasimodels, which have been used to prove the satisfiability of various TDLs [5, 9, 23, 25, 26, 36]. In particular, we show that there is a model with at most k time points iff there is a quasimodel with a sequence of quasistates of length at most k. Then, our upper bound is obtained by guessing an exponential size sequence of sets of types which serves as a certificate for the existence of a quasimodel (and therefore a model) for the input formula. Let ϕ be a TkU ALC formula. Assume w.l.o.g. that ϕ does not contain abbre- viations (i.e., it only contains the logical connectives ¬, u, ∧, the existential quantifier ∃, and the temporal operator U). Denote by clc (ϕ) and clf (ϕ) the closures under single negation of the sets of all concepts and formulas occurring in ϕ, respectively. Also, let ind(ϕ) be the set of individuals occurring in ϕ. A concept type for ϕ is any subset t of clc (ϕ) ∪ ind(ϕ) such that: T1 ¬C ∈ t iff C 6∈ t, for all ¬C ∈ clc (ϕ); T2 C u D ∈ t iff C, D ∈ t, for all C u D ∈ clc (ϕ); and T3 t contains at most one individual name in ind(ϕ). Similarly, we define formula types t ⊆ clf (ϕ) for ϕ with the conditions: T10 ¬φ ∈ t iff φ 6∈ t, for all ¬φ ∈ clf (ϕ); and T20 φ ∧ ψ ∈ t iff φ, ψ ∈ t, for all φ ∧ ψ ∈ clf (ϕ). We may omit ‘for ϕ’ if there is no risk of confusion. A concept type describes one domain element at a single time point, while a formula type expresses constraints on all domain elements. If a ∈ t ∩ ind(ϕ), then t describes an named element. We may write ta to indicate this and call it a named type. The next notion captures how sets of types need to be constrained so that the DL dimension is respected. We say that a pair of concept types (t, t0 ) is R-compatible if {¬F | ¬∃R.F ∈ t} ⊆ t0 . A quasistate for ϕ is a set S of concept or formula types for ϕ such that: Q1 S contains exactly one formula type tS ; Q2 S contains exactly one named type ta for each a ∈ ind(ϕ); Q3 for all C v D ∈ clf (ϕ), we have C v D ∈ tS iff C ∈ t implies D ∈ t for all concept types t ∈ S; Q4 for all A(a) ∈ clf (ϕ), we have A(a) ∈ tS iff A ∈ ta Q5 t ∈ S and ∃R.D ∈ t implies there is t0 ∈ S such that D ∈ t0 and (t, t0 ) is R-compatible; Q6 for all R(a, b) ∈ clf (ϕ), we have R(a, b) ∈ tS iff (ta , tb ) is R-compatible. A (concept/formula) run segment for ϕ is a finite sequence σ = σ(0) . . . σ(n) composed exclusively of concept or formula types, respectively, such that: R1 for all a ∈ ind(ϕ) and all i ∈ (0, n], we have a ∈ σ(0) iff a ∈ σ(i); R2 for all α U β ∈ cl∗ (ϕ) and all i ∈ [0, n], we have α U β ∈ σ(i) iff there is j ∈ (i, n] such that β ∈ σ(j) and α ∈ σ(m) for all m ∈ (i, j), where cl∗ is either clc or clf (as appropriate), and R1 does not apply to formula run segments. Intuitively, a concept run segment describes the temporal dimension of a single domain element, whereas a formula run segment describes constraints on the whole DL interpretation. Finally, a quasimodel for ϕ is a pair (S, R), with S a finite sequence of quasistates S(0)S(1) . . . S(n) and R a non-empty set of run segments such that: M1 ϕ ∈ tS where tS is the formula type in S(0); M2 for every σ ∈ R and every i ∈ [0, n], σ(i) ∈ S(i); and, conversely, for every t ∈ S(i), there is σ ∈ R with σ(i) = t. By M2 and the definition of a quasistate for ϕ, R always contains exactly one formula run segment and one named run segment for each a ∈ ind(ϕ). Every quasimodel for ϕ describes an interpretation satisfying ϕ and, conversely, every such interpretation can be abstracted into a quasimodel for ϕ. We formalise this notion for finite traces with the following lemma. Lemma 5. There is a finite trace satisfying ϕ with at most k time points iff there is a quasimodel for ϕ with a sequence of quasistates of length at most k. We devise a non-deterministic algorithm to check satisfiability of a TkU ALC formula ϕ in NExpTime. First we compute in exponential time w.r.t. |ϕ| the sets of all concept and formula types for ϕ, i.e., satisfying conditions T1-T3 and conditions T10 -T20 , respectively. We then guess a sequence S(0), . . . , S(n) of sets of (concept/formula) types for ϕ of length n ≤ k; and for each type at position i in this sequence we also guess a sequence of types of length n. Denote by R the set of such sequences of types. We now check (1) whether each S(i) satisfies conditions Q1-Q6; (2) whether each sequence in R satisfies conditions R1-R2 (where R1 only applies for concept run segments); and (3) whether ϕ is in the formula type in S(0) (condition M1) and each sequence in R satisfies condition M2. All these conditions can be checked in exponential time w.r.t. |ϕ| and |k|. The algorithm returns ‘satisfiable’ iff all conditions are satisfied. Since the conditions exactly match the definition of a quasimodel for ϕ, their satisfaction implies that (S, R) is a quasimodel for ϕ. Hardness can be proved by adapting Lemma 6.2 by Baader et al. [10], for ALC-LTL with rigid concepts. We are now ready to state our main result in this section. Theorem 6. Satisfiability in TkU ALC is NExpTime-complete. We end this section considering the satisfiability problem in TkU ALC restricted to global CIs, defined as the fragment of TkU ALC in which formulas can only be of the form T ∧ 2(T ) ∧ φ, where T is a conjunction of CIs and φ does not contain CIs. We write TkU (g)ALC to denote this fragment interpreted over finite traces with at most k time points. We establish that satisfiability in TkU (g)ALC is in ExpTime, which is tight since satisfiability in ALC is already ExpTime-hard [7]. Our upper bound is based on type elimination [23, 26, 30]. Theorem 7. Satisfiability in TkU (g)ALC is ExpTime-complete. We leave satisfiability in TfU ALC restricted to global CIs, analogously defined as the fragment TfU (g)ALC of TfU ALC with only formulas of the form T ∧2(T )∧φ, as an open problem. It is known that the complexity of the satisfiability problem in this fragment over infinite traces is ExpTime-complete [9, 30]1 . Though, the end of time formula ψf is not in TfU (g)ALC. We cannot use the same strategy of defining a translation for the semantics based on infinite traces, as we did in Section 3. The upper bound in [30] is based on type elimination. The main difficulty of devising a type elimination procedure for TfU (g)ALC is that the number of time points is not fixed and the argument in [30] showing that there is a quasimodel iff there is a quasimodel (S, R) such that S(i + i) ⊆ S(i), for all i ≥ 0, is not applicable to finite traces. A type with a concept equivalent to ¬ # > can only be in the last quasistate of the quasimodel. So it is not clear whether one can show that if there is a quasimodel then there is a quasimodel with an exponential sequence of quasistates, as in TkU (g)ALC (Theorem 7). 5 Model-Theoretical Properties In this section we discuss some model-theoretical properties of TfU ALC. In particular, we show that formulas of TfU ALC enjoy the bounded model property: if there is a finite trace (which can be arbitrarily large) satisfying a formula ϕ 1 In [30, Theorem 3] the authors show that satisfiability in LTLALC w.r.t. TBoxes with constant domains is ExpTime-complete. An argument in [9, Theorem 1] extends this result to TU ALC formulas, which include assertions. then there is a finite trace with at most double exponentially many time points w.r.t. the size of ϕ. Moreover, the bound on the number of time points implies a bound on the size of the domain, in the DL dimension of the model. Finally, we discuss insensitivity to infiniteness, a model-theoretical property introduced by De Giacomo et al. [19]. We lift to TfU ALC a characterisation for propositional LTL [19]. We then discuss a limitation of this notion and propose a new related notion of insensitivity to infiniteness. 5.1 Bounded Model Property We first show that if there is a finite trace which satisfies a TfU ALC formula ϕ then there is a finite trace with at most k time points, with k double exponentially large w.r.t. the size of ϕ. This bound is obtained by (1) showing that if there is a quasimodel for ϕ then there is a quasimodel for ϕ where there is no repetition of quasistates, except for the last quasistate; and (2) applying Lemma 5 from Section 4, which correlates the length of a sequence of quasistates with the number of time points in a finite trace. Since the number of formula types is f bounded by tpf (ϕ) = 2|cl (ϕ)| and the number of concept types is bounded by c c tpc (ϕ) = |ind(ϕ)| · 2|cl ϕ| , there are at most tpf (ϕ) · 2tp (ϕ) quasistates for ϕ: each quasistate contains one formula type and a subset of the set of concept types. The next lemma formalises Point (1) above and can be proved with an argument similar as that of Lemma 11.27 by Gabbay et al. [23]. Lemma 8. There is a quasimodel for ϕ iff there is a quasimodel for ϕ of the c form (S, R) where S is a sequence of length at most tpf (ϕ) · 2tp (ϕ) + 1. Lemmas 5 and 8 directly imply that TfU ALC enjoys the bounded model property. We formalise this result with the following theorem. Theorem 9. Satisfiability of a formula ϕ in TfU ALC implies satisfiability of ϕ c in TkU ALC with k bounded by tpf (ϕ) · 2tp (ϕ) . We now show that if a formula is satisfiable in TkU ALC then there is a model where the size of the domain is exponential in k. Since satisfiability in TfU ALC implies satisfiability in TkU ALC for some k > 0, the formula: #A(a) ∧ 2(A v 2(¬A u ∃R.A)) which only admits models with an infinite domain [30] (recall that in this paper we use the irreflexive until) is unsatisfiable over finite traces. We formalise the bounded domain property with the following theorem. Theorem 10. Satisfiability of a formula ϕ in TkU ALC implies the existence of a model with domain size bounded by |tpc (ϕ)|k . 5.2 Insensitivity to Infiniteness In this subsection, we investigate the notion of insensitivity to infiniteness, introduced by De Giacomo et al. [19] for propositional LTL formulas. This property is meant to capture those LTLf formulas that can be equivalently interpreted over infinite traces, provided that, from a certain instant, these traces satisfy an end event forever, falsifying all other atomic propositions. Firstly, we lift this notion of insensitivity to our TDL setting, providing a characterisation analogous to the propositional one [19]. Secondly, we propose a different definition for insensitivity, closer to other notions from the the literature on runtime verification [8, 13, 14] and motivated by the following idea: a TfU ALC formula should be considered as insensitive whenever being satisfiable on a finite trace implies that it is satisfiable on every infinite trace extending the finite one, provided that it satisfies the end of time formulas. We introduce the following definitions. Let Σ be a finite subset of NC ∪ NR . Assume w.l.o.g. that TfU ALC concepts and formulas we mention in this subsection have concept and role names in Σ and that Σ contains the end concept E. Given an infinite trace I, the Σ-reduct of I is the infinite trace I|Σ coinciding I with I on Σ and such that X |Σn = ∅, for X 6∈ Σ and n ∈ [0, ∞) [29]. Let F = (∆ , (Fn )n∈[0,l] ) be a finite trace, and let E = (∆E , (En )n∈[0,∞) ) be the F infinite trace such that ∆E = ∆F (for simplicity, we may write just ∆), aE = aF for all a ∈ NI , and for all A ∈ NC \ {E}, R ∈ NR , AEn = ∅ and REn = ∅ while E En = ∆, where n ∈ [0, ∞). The extension of F with E (see its definition in Section “Satisfiability over Finite Traces”), F · E, will be denoted as the end extension of F. Recalling the definition of ψf , we define, χf = ψf ∧ χf1 ∧ χf2 , where l l χf1 = 2(E v ¬A), χf2 = 2(E v ¬∃R.>). A∈Σ\{E} R∈Σ The following two lemmas correspond to Theorems 2 and 3 in [19], respectively. In particular, Lemma 11 characterises infinite traces of the form F · E, for a finite trace F. Lemma 11. For every infinite trace I, I |= χf iff I|Σ = F · E, for some finite trace F. Lemma 12, which specialises Lemma 2, shows the equivalence between TfU ALC satisfiability of a formula ϕ, and satisfiability of its TU ALC translation ϕ† (see Section ‘Satisfiability over Finite Traces’) on end extensions of finite traces. Lemma 12. Let F be a finite trace and let ϕ be a TfU ALC formula. We have that: F |= ϕ iff F · E |= ϕ† . We are now ready to state the first definition of insensitivity to infiniteness. Given a TfU ALC formula ϕ, we say that ϕ is insensitive to infiniteness if: I1 for every finite trace F, F |= ϕ iff F · E |= ϕ. The next characterisation result, obtained from Lemmas 11 and 12, corresponds to Theorem 4 in [19]. Theorem 13. A TfU ALC formula ϕ is insensitive to infiniteness I1 iff the following TU ALC logical implication holds: χf |= ϕ ↔ ϕ† . Thus, insensitive I1 formulas allow us to blur the distinction between finite and infinite traces, provided that the infinite ones are as described in Lemma 11. We now analyse some features of this property. Firstly, notice that TfU ALC for- mulas without any occurrence of temporal operators, are insensitive I1. Moreover, given a TfU ALC formula ϕ, this property is preserved under boolean operators [19, Theorem 5]: if ϕ is insensitive I1, ¬ϕ is insensitive as well, and similarly for conjunction. It is also possible to capture several standard temporal patterns derived from the declarative process modelling language declare [1, 19]. For instance, given a TfU ALC assertion α, we have that 3α is insensitive I1. On the other hand, a formula of the form 2α (which is not included among the standard declare patterns) is not I1. However, we have also that 3¬α is not insensitive I1, while 2¬α is, due to the condition on the interpretation of concept and role names after the end of time. Therefore, in general, if a TfU ALC formula ϕ is I1, neither 3ϕ nor 2ϕ are. The same holds for formulas of the form ϕ U ψ. In order to overcome this temporal asymmetry, and so to obtain a notion of insensitivity to infiniteness that is preserved by the application of the until operator, we propose the following definition. Given a TfU ALC formula ϕ, we say that ϕ is insensitive to infiniteness if: I2 for every finite trace F and every extension F · I, if F |= ϕ, then F · I |= ϕ. We will use similar notational conventions as for I1. Notice that this definition matches the first condition of the impartiality maxim introduced by Bauer et al. [14]. It is possible to provide an analogous characterisation of insensitive I2 formulas, as we did above for the I1 case. Theorem 14. A TfU ALC formula ϕ is insensitive to infiniteness I2 iff the following TU ALC logical implication holds: ψf |= ϕ† → ϕ. As a consequence of Theorems 13 and 14, checking I1 or I2 can be done in ExpSpace using the same algorithms developed for TDLs on infinite traces [30]. As before, TfU ALC formulas without any occurrence of temporal operators are clearly insensitive I2. However, being insensitive I2 is not preserved under negation. For instance, we have that 3α (with α a TfU ALC assertion) is insensi- tive I2, but its negation, ¬3α ≡ 2¬α, is not insensitive I2. The theorem below states that the property of being insensitive I2 is preserved under conjunction and the until operator. Thus, differently from Theorem 5 in [19], we are now able to provide a recursive sufficient condition of insensitivity also in the case of temporal operators. Theorem 15. Let ϕ and ψ be insensitive I2 TfU ALC formulas. Then ϕ ∧ ψ and ϕ U ψ are insensitive I2. In particular, as a consequence of the above statement, both 3ϕ and #ϕ are insensitive I2, if ϕ is, thus overcoming the temporal asymmetry discussed for insensitivity I1. On the other hand, formulas of the form 2ϕ should not be considered, in general, as insensitive I2, since the infinite interpretation of such formulas, when lifted from a finite trace, depends on how the finite trace is Formula I1 I2 ϕ (ϕ no temporal operators) Y Y ¬ϕ (ϕ insensitive) Y N ϕ ∧ ψ (ϕ, ψ insensitive) Y Y ϕ U ψ (ϕ, ψ insensitive) N Y 2ϕ (ϕ insensitive) N N Table 1. Comparison between I1 and I2 extended. Notice, however, that concepts and formulas equivalent to (respec- tively) 2> and 2(> v >), where > and > v > are clearly insensitive I2, are insensitive I2 as well. Table 1 summarizes differences and similarities between I1 and I2. Finally, notice that both I1 and I2 definitions can be easily adapted to finite traces with fixed length k > 0. This allows us to compare the cases of nested next operators. For example, we have that #n ¬A(a) is I1 iff n < k. On the other hand, #n ¬A(a) turns out to be insensitive I2 for both n < k and n ≥ k, since in the latter case the I2 condition is vacuously satisfied. 6 Conclusion We presented TfU ALC, a TDL obtained by combining ALC with LTLf . Firstly, we have focussed on the complexity of the TfU ALC formula satisfiability problem. Despite a result of ExpSpace-completeness for the general case, coinciding with the complexity of the same problem over infinite traces, we have shown how to improve it to NExpTime, when considering finite traces with a fixed number k of instants, and even to ExpTime, if additionally we consider the language restricted with only global CIs, namely TkU (g)ALC. We leave the complexity of satisfiability in TfU (g)ALC as an open problem. In order to further refine these results, we plan to consider temporal DL-Lite logics interpreted over finite traces. Concerning the section on model-theoretical properties, we have shown that the logic TfU ALC has the bounded model property. Moreover, we have studied the notion of insensitivity to infiniteness, both as found in the literature (I1), and as re-defined here (I2). We have shown their characterisations, and sufficient conditions to preserve I2. As future work, we plan to refine our results, so to cover the case of CIs with temporal operators applied on concepts, that are nonetheless insensitive I2 (such as > v 3A). A characterisation of insensitive CIs could in turn be based on a suitably defined notion of insensitivity for concepts. We leave this task for future work, together with an in depth discussion of the interaction between temporal operators (including additional ones, as weak next , release R, or weak until W) and insensitivity. References 1. van der Aalst, W.M.P., Pesic, M.: Decserflow: Towards a truly declarative service flow language. In: Web Services and Formal Methods, Third International Workshop, WS-FM 2006 Vienna, Austria, September 8-9, 2006, Proceedings. pp. 1–23 (2006) 2. van der Aalst, W., Artale, A., Montali, M., Tritini, S.: Object-centric behavioral constraints: Integrating data and declarative process modelling. In: CEUR Workshop Proceedings. vol. 1879 (2017) 3. Artale, A., Franconi, E.: A survey of temporal extensions of description logics. Annals of Mathematics and Artificial Intelligence 30(1-4), 171–210 (2000) 4. Artale, A., Franconi, E.: Temporal description logics. In: Handbook of Temporal Reasoning in Artificial Intelligence, pp. 375–388. Elsevier (2005) 5. Artale, A., Kontchakov, R., Lutz, C., Wolter, F., Zakharyaschev, M.: Temporal- ising tractable description logics. In: 14th International Symposium on Temporal Representation and Reasoning (TIME 2007), Alicante, Spain, 28-30 June 2007. pp. 11–22 (2007) 6. Artale, A., Kontchakov, R., Ryzhikov, V., Zakharyaschev, M.: A cookbook for temporal conceptual data modelling with description logics. ACM Transactions on Computational Logic (TOCL) 15(3), 25 (2014) 7. Baader, F., Calvanese, D., McGuinness, D.L., Nardi, D., Patel-Schneider, P.F. (eds.): The Description Logic Handbook: Theory, Implementation, and Applications. Cambridge University Press (2003) 8. Baader, F., Bauer, A., Lippmann, M.: Runtime verification using a temporal description logic. In: Frontiers of Combining Systems, 7th International Symposium, FroCoS 2009, Trento, Italy, September 16-18, 2009. Proceedings. pp. 149–164 (2009) 9. Baader, F., Borgwardt, S., Koopmann, P., Ozaki, A., Thost, V.: Metric temporal description logics with interval-rigid names. In: Frontiers of Combining Systems - 11th International Symposium, FroCoS 2017, Brasília, Brazil, September 27-29, 2017, Proceedings. pp. 60–76 (2017) 10. Baader, F., Ghilardi, S., Lutz, C.: LTL over description logic axioms. ACM Trans. Comput. Log. 13(3), 21:1–21:32 (2012) 11. Bacchus, F., Kabanza, F.: Using temporal logics to express search control knowledge for planning. Artif. Intell. 116(1-2), 123–191 (2000) 12. Baier, J.A., McIlraith, S.A.: Planning with first-order temporally extended goals using heuristic search. In: Proceedings, The Twenty-First National Conference on Artificial Intelligence and the Eighteenth Innovative Applications of Artificial Intelligence Conference, July 16-20, 2006, Boston, Massachusetts, USA. pp. 788–795 (2006) 13. Bauer, A., Haslum, P.: LTL goal specifications revisited. In: Proc. of the 19th European Conference on Artificial Intelligence, ECAI’10. pp. 881–886 (2010) 14. Bauer, A., Leucker, M., Schallhart, C.: Comparing LTL semantics for runtime verification. J. Log. Comput. 20(3), 651–674 (2010) 15. Calvanese, D., De Giacomo, G., Vardi, M.Y.: Reasoning about actions and planning in LTL action theories. In: Proc. of the 8th International Conference on Principles and Knowledge Representation and Reasoning (KR-02). pp. 593–602 (2002) 16. Camacho, A., Triantafillou, E., Muise, C.J., Baier, J.A., McIlraith, S.A.: Non- deterministic planning with temporally extended goals: LTL over finite and infinite traces. In: Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence, February 4-9, 2017, San Francisco, California, USA. pp. 3716–3724 (2017) 17. Cerrito, S., Mayer, M.C.: Bounded model search in linear temporal logic and its application to planning. In: Automated Reasoning with Analytic Tableaux and Related Methods, International Conference, TABLEAUX ’98, Oisterwijk, The Netherlands, Proceedings, May 5-8, 1998. pp. 124–140 (1998) 18. Cerrito, S., Mayer, M.C., Praud, S.: First order linear temporal logic over finite time structures. In: Logic Programming and Automated Reasoning, 6th International Conference, LPAR’99, Tbilisi, Georgia, Proceedings, September 6-10, 1999. pp. 62–76 (1999) 19. De Giacomo, G., De Masellis, R., Montali, M.: Reasoning on LTL on finite traces: Insensitivity to infiniteness. In: Proceedings of the Twenty-Eighth AAAI Conference on Artificial Intelligence, Québec City, Québec, Canada, July 27 -31, 2014. pp. 1027–1033 (2014) 20. De Giacomo, G., Vardi, M.Y.: Linear temporal logic and linear dynamic logic on finite traces. In: IJCAI, Proceedings of the 23rd International Joint Conference on Artificial Intelligence. pp. 854–860 (2013) 21. De Giacomo, G., Vardi, M.Y.: Synthesis for LTL and LDL on finite traces. In: Proceedings of the Twenty-Fourth International Joint Conference on Artificial Intelligence, IJCAI. pp. 1558–1564 (2015) 22. Fionda, V., Greco, G.: The complexity of LTL on finite traces: Hard and easy fragments. In: Proc. of the Thirtieth AAAI Con- ference on Artificial Intelligence. pp. 971–977. AAAI Press (2016), http://www.aaai.org/ocs/index.php/AAAI/AAAI16/paper/view/12250 23. Gabbay, D.M., Kurucz, A., Wolter, F., Zakharyaschev, M.: Many-dimensional Modal Logics: Theory and Applications, Studies in Logic and The Foundations of Mathematics, vol. 148. North Holland Publishing Company (2003) 24. Giannakopoulou, D., Havelund, K.: Automata-based verification of temporal prop- erties on running programs. In: 16th IEEE International Conference on Automated Software Engineering (ASE 2001), 26-29 November 2001, Coronado Island, San Diego, CA, USA. pp. 412–416 (2001) 25. Gutiérrez-Basulto, V., Jung, J.C., Lutz, C.: Complexity of branching temporal description logics. In: Proc. ECAI (2012) 26. Gutiérrez-Basulto, V., Jung, J.C., Ozaki, A.: On metric temporal description logics. In: ECAI 2016 - 22nd European Conference on Artificial Intelligence, 29 August-2 September 2016, The Hague, The Netherlands - Including Prestigious Applications of Artificial Intelligence (PAIS 2016). pp. 837–845 (2016) 27. Hodkinson, I.M., Kontchakov, R., Kurucz, A., Wolter, F., Zakharyaschev, M.: On the computational complexity of decidable fragments of first-order linear temporal logics. In: 10th International Symposium on Temporal Representation and Reasoning / 4th International Conference on Temporal Logic (TIME-ICTL 2003), Cairns, Queensland, Australia, 8-10 July 2003. pp. 91–98 (2003) 28. Hodkinson, I.M., Wolter, F., Zakharyaschev, M.: Decidable fragment of first-order temporal logics. Ann. Pure Appl. Logic 106(1-3), 85–134 (2000) 29. Konev, B., Lutz, C., Wolter, F., Zakharyaschev, M.: Conservative rewritability of description logic tboxes. In: Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence, IJCAI. pp. 1153–1159 (2016) 30. Lutz, C., Wolter, F., Zakharyaschev, M.: Temporal description logics: A survey. In: 15th International Symposium on Temporal Representation and Reasoning, TIME 2008, Université du Québec à Montréal, Canada, 16-18 June 2008. pp. 3–14 (2008) 31. Maggi, F.M., Montali, M., Westergaard, M., van der Aalst, W.M.P.: Monitoring business constraints with linear temporal logic: An approach based on colored automata. In: Rinderle-Ma, S., Toumani, F., Wolf, K. (eds.) Proc. of Business Process Management (BPM’11). pp. 132–147. Springer (2011) 32. Manna, Z., Pnueli, A.: Temporal verification of reactive systems - safety. Springer (1995) 33. Pnueli, A.: The temporal logic of programs. In: 18th Annual Symposium on Foun- dations of Computer Science, Providence, Rhode Island, USA, 31 October - 1 November 1977. pp. 46–57 (1977) 34. Reger, G., Rydeheard, D.E.: From first-order temporal logic to parametric trace slicing. In: Runtime Verification - 6th International Conference, RV 2015 Vienna, Austria, September 22-25, 2015. Proceedings. pp. 216–232 (2015) 35. Vardi, M.Y.: An automata-theoretic approach to linear temporal logic. In: Proc. of Logics for Concurrency - Structure versus Automata, 8th Banff Higher Order Workshop. pp. 238–266 (1995) 36. Wolter, F., Zakharyaschev, M.: Temporalizing description logics. In: Proceedings of FroCoS’98. pp. 104–109 (1998)