• Keine Ergebnisse gefunden

6.3 First-Order Rewritings of r-Satisfiability

6.3.1 A Tuple for Testing r-Satisfiability

The first goal is to specify a fixed tuple (AR[W,BΦ], QR[W], Q¬R[W], RF[W,BΦ]) as described in Section 6.1.2: it is based on two given setsW ⊆2{p1,...,pm}such thatW ={W1, . . . , Wk} and BΦ ⊆ {B(a) |B ∈ B(O), a∈NI(Φ)}, both of constant size, and r-complete iff W is r-satisfiable w.r.t. ι and K. For testing r-completeness, we subsequently also specify the KBs regarded in these tests.

10Note that we follow the approach of [Cal+05] and introduce a relation for every basic concept, instead of only for the concept names.

6.3 First-Order Rewritings of r-Satisfiability

We define the tuple such that the four sets are minimal and satisfy the r-completeness conditions. More precisely, we focus on those parts of the conditions which we can ad-dress without considering information that is not given (e.g., the mapping ι, which depends on the data). While the definition of components such as QR[W] and Q¬R[W] is straightforward, the construction of the set RF represents a challenge. This is because the focus on data complexity imposes special limits and Condition (C6) is rather intri-cate. Regarding data complexity, it would be critical to consider the entire set RF and ABox ARF within our rewriting. Recall that ARF may contain auxiliary elements tai-lored to individual elements inNI(K), which then would need to be considered explicitly within the rewriting—this is obviously impossible. To circumvent that, we first of all discern the assertions in RF more fine-granularly, according to the kind of individual they address. In particular, the setRF is considered to be the disjoint union of three sets RF|aux,RF|Φ, andRF|o(ofor “other”), each containing only assertions on the individuals fromNauxI ,NI(Φ),andNI(K)\NI(Φ),respectively (i.e., next to the ones fromNtreeI ). For each of these sets, we can restrict our tests to parts of Condition (C6):

RF|aux is constant and hence its size is not relevant. In particular, the elements of NauxI neither occur in the ABox type, nor in the input ABoxes, and can be uniquely associated to one of the CQs in Φ. For constructing RF|aux in line with Condition (C6), it is thus enough to focus on instantiations of the latter CQs, individually.

• The construction of RF|o is more involved. The issue with the size of the corre-sponding ABox is covered later in this section. To ensure that Condition (C6) is satisfied, we first model the derivation of all relevant (rigid) basic concept asser-tions, the consequences of the TKB, in our rewriting and then use them to derive those inRF|o. Observe that the condition requires several ABoxes to be considered.

Since the elements under consideration do not occur in Φ, we can however focus on the ABox type and the input ABoxes. That is, the rigid assertions represent (the relevant part of) the ABox typeAR[W,BΦ]; the derivation resolves transitivity and thus ensures that AR[W,BΦ] is in line with Condition (C1). We define the set of these rigid assertions based on the below derivation as BR|o :=BR|o|BR(O)|.

BR|o0 :=∅,

Bj+1R|o :={B(a)|B ∈BR(O), a∈NI(K)\NI(φ),∃i.0≤in, hO,BR|oj ∪ Aii |=B(a)}.

• The setRF|Φis of constant size. For constructing it, we have to apply a derivation as above since the elements in NI(Φ) might occur in the input ABoxes. Actually, these elements may occur in all the ABoxes regarded in Condition (C6), which thus all would have to be taken into account in the derivation. However, we can obviously not apply the mapping ι to define a fixed rewriting. This is why we assume a set BΦ ⊆ {B(a) | B ∈ B(O), a ∈ NI(Φ)} to be given instead of defining it. The test ifRF|Φsatisfies Condition (C6), in dependence of BΦ, is thus postponed to the actual application and evaluation of the rewriting.

These observations lead to the following definitions:

QR[W]:={αj ∈ QΦ|W ∈ W, pjW}, Q¬R[W]:={αj ∈ QΦ|W ∈ W, pj 6∈W},

RF|aux[W]:={∃S(ay)|S∈NR(O)\NRR, ay ∈NauxI ,∃Wj ∈ W, hO,AQji |=∃S(ay)}, RF|Φ[BΦ]:={∃S(a)∈ BΦ|S ∈NR(O)\NRR},

RF|o:={∃S(a)|S∈NR(O)\NRR, a∈NI(K)\NI(Φ),

∃i.0≤in, hO,BR|o∪ Aii |=∃S(a)}.

Based on these sets, we define the corresponding auxiliary ABoxes, as in Section 6.1.2:

• AQR[W] :=Sϕ∈Q

R[W]CO(ϕ).

• AQj for j∈[1, k] is defined in Section 6.1.2; for convenience, we sometimes write AQWj instead ofAQj.

• ARF[W,B

Φ] :=ARF|aux[W] ∪ ARF|Φ[B

Φ]∪ ARF|o; ARF|aux,ARF|Φ, andARF|o represent the ABoxes corresponding to the sets RF|aux, RF|Φ, and RF|o, respectively, and are constructed in correspondence to the definition of ARF in Section 6.1.2. Note that, while the former sets contain only flexible assertions, the ABoxes contain only rigid ones.

• AR[W,BΦ] is then defined as the union of the setA+R[W,B

Φ], specified below, and all negative rigid role and basic concept assertions ¬α overNI(K) for which we have α6∈ A+R[W,B

Φ]. BΦ|R denotes the set of rigid assertions inBΦ. A+R[W,B

Φ]:=BΦ|R∪ BR|o∪ {∃R(a)|R(a, e)∈ ARF|Φ[B

Φ], a∈NI(Φ)} ∪ {R(a, b)|R∈NRR(O), a, b∈NI(K), R(a, b)∈ AQR[W] or

∃i.0≤in,hO,Aii |=R(a, b)}

As mentioned above, we cannot assume the arbitrary setBΦ to be complete and to provide the part on the elements ofNI(Φ) which follows from the TKB, especially, regarding all the assertions to be contained in AR. Since we test the sufficiency of BΦ when the rewriting is evaluated by using entailment tests, we then can however not detect rigid information missing in BΦ if it is implied by the flexible assertions. For that reason, we consider ARF|Φ[B

Φ] in the above construction and thus explicitly regard the consequences from the flexible assertions in BΦ.

Observe that, apart fromARF|o andAR[W,BΦ], which depend on the input ABoxesA, all of the ABoxes we defined are constant.

The below auxiliary lemma shows thatAR[W,BΦ]is, in a certain sense, closed regarding the data in the given ABox sequence.

Lemma 6.22 For all B ∈BR(O), R∈NRR(O) and a, b∈NI(K)\NI(Φ), we have:

B(a)∈ AR[W,BΦ] iff there is ani∈[0, n]such that hO,AR[W,BΦ]∪ Aii |=B(a),

R(a, b)∈ AR[W,BΦ] iff there is an i∈[0, n]such that hO,AR[W,BΦ]∪ Aii |=R(a, b).

6.3 First-Order Rewritings of r-Satisfiability

Proof. (⇒) This direction is trivial. (⇐) We assume that hO,AR[W,BΦ]∪ Aii |=B(a) holds for some i∈[0, n]. Observe that, if we have BR|oj =BR|oj+1 at some point, then we have BR|oj =BR|oj+l for all l ≥ 0. Hence, there must be some j0 ∈ [0,|BR(O)|] such that BjR|o0 =BR|oj0+l, for alll≥0, because every setBR|oj+1such thatBj+1R|o 6=BjR|o must contain at least one new assertion, there are only|BR(O)|relevant assertions per individual, and an assertion on a specific individual does not depend on assertions on other individuals by Lemma 2.14. GivenhO,AR[W,BΦ]∪ Aii |=B(a) and the fact that only the assertions on the individual a(possibly contained in a tuple of individuals) in AR[W,BΦ] are relevant for the entailment by Lemmas 2.14 and 2.13, we can neglect all but the second and last conjuncts in the definition of A+R[W,B

Φ]; a cannot occur in these sets. In addition to the assertions in BR|o, we thus consider those entailed by some hO,Aii,i∈[0, n]. But the basic concept assertions corresponding to those role assertions have to be contained in B1R|o and hence in also in BR|o by definition. This leads to B(a) ∈ AR[W,BΦ] by the definition of AR[W,BΦ].

We second assumehO,AR[W,BΦ]∪ Aii |=R(a, b). By Definition 2.11 and Lemma 2.13, there is some S ∈ NR such that S(a, b) ∈ AR[W,BΦ] ∪ Ai and O |= S v R. If S(a, b) ∈ Ai, then we have that hO,Aii |= R(a, b), and hence R(a, b) ∈ AR[W,BΦ] by the definition of AR[W,BΦ]. Otherwise, that definition implies that there is a j ∈[0, n]

such that hO,Aji |= S(a, b), and hence also hO,Aji |=R(a, b), which also shows that R(a, b)∈ AR[W,BΦ].

Given all the relevant ABoxes and additionally a mapping ι: [0, n] → [1, k], we can specify the KBs in the focus of the r-completeness tests in Definition 6.8 for all i∈[0, n+k], consideringAi :=∅ fori > n:

KR[W,Bi

Φ]:=hO,AiR[W,B

Φ]i, where

AiR[W,B

Φ]:=AR[W,BΦ]∪ AQR[W] ∪ AQι(i)∪ ARF[W,B

Φ]∪ Ai.

Before we continue regarding the rewriting, we show that we can useBΦ as intended.

Lemma 6.23 For allW ⊆2{p1,...,pm}such thatW ={W1, . . . , Wk}andι: [0, n]→[1, k], there is an r-complete tuple w.r.t. W andι iff there is a set BΦ such that the tuple

(AR[W,BΦ], QR[W], Q¬R[W], RF[W,BΦ]) is r-complete w.r.t. W and ι.

Proof. (⇐) This direction is trivial. (⇒) We assume (AR, QR, Q¬R, RF) to be an r-complete tuple, define BΦ as follows, and show that (AR[W,BΦ], QR[W], Q¬R[W], RF[W,BΦ]) is r-complete as well:

BΦ:={B(a)∈ ARRF|B ∈B(O), a∈NI(φ)}.

We focus on the conditions in Definition 6.8. Our tuple obviously satisfies Condi-tions (C3) and (C4) by construction.

Regarding Conditions (C1), (C2), and (C5), we describe a model ofKiR[W,B

Φ]that can be homomorphically embedded into the canonical interpretation of the consistent KB that exists for the given tuple, since it satisfies Condition (C1); we denote the latter KB by KRi. Observe that all positive assertions contained in one of the ABoxes of KiR[W,B

Φ]

must also be contained in KiR:

• AQR[W] ⊆ AQR follows from the facts that both tuples satisfy Condition (C3) and QR[W]is the minimal set satisfying that condition.

• ARF|aux[W]∪ ARF|Φ[B

Φ]∪ ARF|o ⊆ ARF is a consequence of the following observations.

By definition, every∃S(b)∈RF|aux[W]is a consequence of a KBhO,AQji,Wj ∈ W.

Since the given tuple satisfies Condition (C6) and we have ι(n+j) = j in that definition, the assertion is also contained in RF. ARF|Φ[B

Φ] ⊆ ARF follows from the construction. Each ∃S(b) ∈ RF|o follows, by definition, from BR|o together with someAi(andO). Since the given tuple satisfies Condition (C1), AR must contain all assertions in BR|o; since Condition (C6) is satisfied, the assertion is thus also contained inRF.

• All positive assertions inAR[W,BΦ]have to be positive inAR, too, by the definition of AR[W,BΦ] and the observations in the previous items. More precisely, we have BΦ|R⊆ AR,BR|o ⊆ AR, and that all positive assertions inAR[W,BΦ]\ BΦ|R\ BR|o are implied by a KB hO,ARF|Φ[B

Φ] ∪ AQR[W] ∪ Aii, i ∈ [0, n]. Since the latter assertions occur inKiR, which is consistent by assumption, the ABox typeARalso contains the latter rigid consequences.

Hence, any difference between KiR and KiR[W,B

Φ] (i.e., focusing on the assertions in KiR[W,B

Φ] and disregarding additional assertions in KRi) must be due to negative rigid assertions in AR[W,BΦ] that occur positively in AR (because AR is an ABox type) and may cause the inconsistency of KiR[W,B

Φ]. By providing a model forKiR[W,B

Φ], we show that such assertions cannot exist. Since the given tuple satisfies Condition (C1) and KiR contains all positive assertions occurring inKiR[W,B

Φ], the KB [KiR[W,B

Φ]]+, obtained from KiR[W,B

Φ] by dropping the negative assertions, is also consistent. We focus on the canonical interpretation I of that KB and show that it also satisfies KiR[W,B

Φ]. We consider negative role and basic concept assertions in KiR[W,B

Φ].

Φ], and Ai and argue with the definitions of these ABoxes.

Let S be rigid, first. We can disregard AQι(i) and ARF[W,B

Φ], since all rigid asser-tions in the former ABox are also contained in AQR[W] and because the ABoxes ARF[W,B

Φ] consists of do not contain assertions on two elements ofNI(K). Further, (rigid) role assertions that contain only elements of NI(K) are positively contained inAR[W,BΦ]if they occur inAQR[W] or are a consequence of a KBhO,Aji,j∈[0, n].

Together with the fact that S(a, b)∈ AQR[W] impliesR(a, b)∈ AQR[W], this yields that the occurrence ofS(a, b) in one of the latter ABoxes causesR(a, b)∈ AR[W,BΦ].

6.3 First-Order Rewritings of r-Satisfiability

Since AR[W,BΦ] is an ABox type (i.e., only oneR(a, b) or¬R(a, b) is contained in it), that contradicts the assumption.

If S is flexible, S(a, b) occurs in Ai or AQι(i), which implies that Wι(i) ∈ W. By the definition ofAR[W,BΦ]andAQR[W], based onQR[W]and Definition 6.2, we then get R(a, b) ∈ AR[W,BΦ] orR(a, b)∈ AQR[W], which also implies R(a, b)∈ AR[W,BΦ] and thus a contradiction to the assumption.

• Let¬B(a)∈ AR[W,BΦ]. Ifa∈NI(Φ), this implies¬B(a)∈ ARby the definitions of AR[W,BΦ] based onBΦ|R andBΦ based on AR. SinceAR is an ABox type and the given tuple satisfies Condition (C1), Lemma 2.12 yieldsI06|=B(a), assumingI0 to be the canonical interpretation of KiR. By our above observation on the positive assertions in KiR[W,B

Φ], this interpretation must also satisfy [KiR[W,B

Φ]]+. Hence, B(a) cannot be a consequence of that KB, and Lemma 2.13 yields I 6|=B(a).

For the case a6∈NI(Φ), we again proceed by contradiction and assume I |=B(a).

Lemma 2.14 then yields that there are positive assertions about ain the ABoxes AR[W,BΦ]∪ ARF|o ∪ Aj, j ∈ [0, n], that together imply B; the other ABoxes do not contain assertions on such individuals. By that lemma, we can also disregard ARF|osinceAR[W,BΦ]contains the relevant assertions. Note that assertions inARF|o only on elements ofNI(K) are always basic concept assertions. Then, Lemma 6.22 implies that we can actually focus onAR[W,BΦ]alone and obtainB(a)∈ AR[W,BΦ]. This contradicts the assumption since AR[W,BΦ] is an ABox type.

SinceI is a model of [KR[W,Bi

Φ]]+by Lemma 2.12 and we have shown that it satisfies all negative assertions inAR[W,BΦ], it is also a model of KiR[W,B

Φ]. Hence, our tuple satisfies Condition (C1).

If one of Conditions (C2) and (C5) is contradicted, then Lemma 2.13 yields that there is a homomorphism of the CQ that causes the contradiction into I. Again, the above observation that the positive assertions contained in KiR[W,B

Φ]must be contained inKiR is important. By Definition 2.11 and the semantics, every such homomorphism into I is also a homomorphism into the canonical interpretation of the positive part of KiR. This contradicts the assumption that KiR satisfies Conditions (C2) and (C5), again by Lemma 2.13.

It remains to consider Condition (C6), and we discern regarding the elements in focus. Observe that, w.r.t. the ABoxes considered in that condition, the individual names occurring inRF|aux[W] can only occur withinAQR[W]SW∈WAQW, and those in RF|o only in AR[W,BΦ]S0≤i≤nAi.

• We regard the assertions in RF|aux[W]. (⇐) For every ∃S(ay) ∈ RF|aux[W], and thusay ∈NauxI , the definition of RF|aux[W] directly yields that there is a Wj ∈ W such that hO,AQji |= ∃S(ay). This solves the claim given that Condition (C6) considersι to be such thatj=ι(n+j).

(⇒) If there is a W ∈ W such thathO,AQR[W]∪ AQWi |=∃S(ay),ay ∈NauxI , then Lemma 2.14 implies that∃S(ay) can only follow from assertions involvingay. But ay can be associated to a unique query ϕj ∈ QΦ that contains the variabley and corresponding ABox Aϕj; no other such ABoxes contains assertions on ay. This implies ϕjQR[W]. By the definition of QR[W], there is a W0 ∈ W withpjW0

and, in particular, AQ

W0 implies all assertions on ay inAQR[W]. This shows that

∃S(ay) already follows from AQ

W0 (andO), which yields∃S(ay)∈RF|aux[W].

• We regard the assertions in RF|o. (⇐) If ∃S(b) ∈ RF|o, then there is an in-dex i∈[0, n] such that hO,BR|o∪ Aii |=∃S(b). By the definition of AR[W,BΦ], we have BR|o ⊆ AR[W,BΦ], and hence obtain that hO,AR[W,BΦ]∪ Aii |=∃S(b).

(⇒) IfhO,AR[W,BΦ]∪Aii |=∃S(b), then the definition ofAR[W,BΦ]and Lemma 2.14 yield that the assertion follows from rigid assertions in BR|o and several rigid role assertions of the form R(b, a), R ∈ NR(O), each of which follows from a KB hO,Aji, j ∈ [0, n]. But, for those rigid role assertions, we have that the corresponding rigid basic concept assertions ∃R(b) are contained in BR|o by con-struction. Hence ∃S(b) follows from BR|o (andO), and is thus contained in RF|o, by the definition of that set.

• We regard the assertions inRF|Φ[BΦ]. Since the given tuple satisfies Condition (C6) and, by the definition of RF|Φ[BΦ], RF and RF|Φ[BΦ] coincide regarding NI(Φ), we have that there is an i∈[0, n] such that hO,AR∪ AQR∪ AQι(i)∪ Aii |=∃S(a) iff

∃S(a)∈RF|Φ[BΦ] for all a∈NI(Φ).

(⇒) This direction then directly follows from the above observation that all posi-tive assertions in AR[W,BΦ] occur inAR and AQR[W] ⊆ AQR.

(⇐) Since AR is an ABox type, the fact that the given tuple satisfies Condi-tion (C1) yields that all basic concept asserCondi-tions that can be derived from asser-tions in AR or AQR are also positively contained in AR; note that these ABoxes both contain only rigid assertions. Moreover, BΦ contains all these rigid basic concepts on elements of NI(Φ) which are contained in AR, by definition. Hence, Lemma 2.14 yields that ∃S(a) ∈RF|Φ[BΦ] implies that there is an i∈ [0, n] such that hO,AR[W,BΦ]∪ AQR[W]∪ AQι(i)∪ Aii |=∃S(a).

Thus, Condition (C6) is also satisfied.

In order to be able to specify rewritings, we next propose an approach to handle the critical size of the ABox ARF|o.

Introducing Prototypes

To untangle the knowledge captured in the ABox ARF|o and separate it from the in-put data as much as possible, we introduce prototypes11. Specifically, instead of the names occurring in ARF|o, we consider prototypical, fresh individual names of the form [S], a[S]S, a[S]S%, etc., meaning [S] is used instead of a concrete individual name from NI(K). We collect all these new names in the set NproI .

The ABoxA∃S for allS∈NR(O)\NRR represents a prototypical version of an ABox A∃S(b) with b∈NI(K) (see Section 6.1.2) and is obtained fromA∃S(b) as follows, where b∈NI(K) and ab%, ab%S ∈NtreeI :

• everyR(b, abS) is replaced byR(b, a[S]S),

11Not to be confused with the prototypical elements in canonical interpretations.

6.3 First-Order Rewritings of r-Satisfiability

• everyR(ab%, ab%S) is replaced byR(a[S]%, a[S]%S),

• everyB(ab%) is replaced by B(a[S]%).

For query answering regarding the adapted ABox, we propose the rewriting ·.

Definition 6.24 (·)Letprobe a unary predicate that identifies exactly the elements of NproI . For a CQ ϕ:=∃~x.ψ, define the query ϕ:=∃~x.ψψfilter with

F|o is defined as the union of the setS∃S(a)∈R

F|oA∃S and the set that, for each a∈NproI occurring in the latter set, contains the assertion pro(a). ♦ The below lemma captures the intent of the rewriting.

Lemma 6.25 Let ϕ be a CQ and A ∪ ARF|o be one of the ABoxes AiR[W,B prototypes, which exist in the domain of DB(A ∪ AR

F|o) by the definition of AR

F|o. Let R(s, t), S(t, u)ϕ;R, S ∈ NR(O); π0(s) 6∈ NproI ; and π0(t) ∈ NproI , such that the precondition of ψfilter evaluates to true under π0; otherwise, the implication clearly holds. Recall that, an ABox AiR[W,B

Φ] as considered (see the definition in the previous section) only via ARF|o refers to elements such as π(t) of NtreeI that are associated to named individuals—here,π(s)—fromNI(K)\NI(Φ) (i.e., elements which are replaced by

·). By Definition 2.23, the assumption thus impliesR(π(s), π(t)), S(π(t), π(u))∈ ARF|o. Together with the assumption thatπ0(s)6∈NproI ; the fact that·replaces all ofNtreeI , and only those, by prototypes; andπ0(s) =π(s), given by the definition ofπ0, this means that π(s)6∈NtreeI . Sinceπ(s) occurs in the ABox ARF|o, we hence haveπ(s)∈NI(K)\NI(Φ).

By the construction ofARF|o and the two role atoms contained in it, we thus must have π(u) =π(s) orπ(u)∈NtreeI . This yields that eitherπ0(u) =π0(s) orπ0(u)∈NproI by our is because atoms containing terms mapped to elements of NproI can only be satisfied by DB(A ∪ AR

F|o) through assertions inAR

F|o, by Definition 2.23 and the fact that elements ofNproI do not occur inA;AR

F|oonly contains individuals ofNproI andNI(K)\NI(Φ); and

we assume ϕto be connected. Furthermore, the elements ofNproI inAR

F|o are connected in tree structures (see the definition of ARF|o in Section 6.1.2) and, for each such struc-ture, there is at least one corresponding structure in ARF|o by the definition of AR

F|o. We hence can define π to map to the corresponding tree elements in one such structure which correspond to the ones that π0 maps to.

It remains to consider the case thatπ0maps to elements of bothNproI andNI(K)\NI(Φ).

First, we defineπ asπ0 w.r.t. all terms that are not mapped to elements ofNproI . Note that AR

F|o does not contain assertions which do not contain elements of NproI (see the definition of ARF|o). By Definition 2.23, we thus haveα∈ A for all atomsαϕwhere π0(α) does not contain elements of NproI . This particularly means that our definition of π is as required w.r.t. all those atoms.

For definingπon the terms not yet considered, we regard a termsinϕthat is mapped to an element ofNI(K)\NI(Φ) and occurs in a role atom R(s, t) together with a term t mapped to an element ofNproI ; since we assumeπ0 to map to elements of bothNproI and NI(K)\NI(Φ) and ϕ to be connected, such a role atom must exist. By Definition 2.23 and the fact that A does not contain elements of NproI , given by assumption, we have R(π0(s), π0(t))∈ AR the introduction of prototypes yielding elements of the form a[S]S does not change the fact that all assertions in AR

F|o on both π0(s) and a[S]S must stem from ∈ A∃S(π0(s)). Note that we also get π0(t) =a[S]S. We define π(t) =aπ0(s)S. Since the filter conjunct is satisfied, we have that all role atoms R0(t, u) in whicht occurs contain only termsu that are either mapped to π0(s) or to an element π0(u)∈NproI . The former kind of role atoms is satisfied by this definition of πby the above observation that all role assertions inAR

F|o that contain bothπ0(s) anda[S]S stem from A∃S(π0(s)).

We regard a role atom of the other kind. Since this case is similar to the induction case, we abstract from the given information by assuming π0(t) = a[S]%, π(t) = ab%

to be defined, and ∃S(b) ∈ RF|o. Again, by Definition 2.23 and the fact that A does not contain elements of NproI , by assumption, we have R00(t), π0(u)) ∈ AR

F|o. By the definition of AR

F|o based on ABoxes of the formA∃S0,S0 ∈NR, which do not overlap in the elements of NproI they contain, we get that R00(t), π0(u))∈ A∃S since the shape of π0(t) indicates that it is contained in that ABox. By the same argument, we then get R000(t)), π0(u)) ∈ A∃S for all R000(t), π0(u)) ∈ AR

6.3 First-Order Rewritings of r-Satisfiability

If we apply this lemma in the following, we assumeNtreeI to only contain the auxiliary elements used in ARF|aux[W] and ARF|Φ[B

Φ], meaning that we disregard the auxiliary ele-ments fromARF|o, which it contains according to the original definition. Also note that NtreeI does not contain elements from NproI .