Abstract
We generalize, by a progressive procedure, the notions of conjunction and disjunction of two conditional events to the case of n conditional events. In our coherence-based approach, conjunctions and disjunctions are suitable conditional random quantities. We define the notion of negation, by verifying De Morgan’s Laws. We also show that conjunction and disjunction satisfy the associative and commutative properties, and a monotonicity property. Then, we give some results on coherence of prevision assessments for some families of compounded conditionals; in particular we examine the Fréchet-Hoeffding bounds. Moreover, we study the reverse probabilistic inference from the conjunction \(\mathcal {C}_{n + 1}\) of n + 1 conditional events to the family \(\{\mathcal {C}_{n},E_{n + 1}|H_{n + 1}\}\). We consider the relation with the notion of quasi-conjunction and we examine in detail the coherence of the prevision assessments related with the conjunction of three conditional events. Based on conjunction, we also give a characterization of p-consistency and of p-entailment, with applications to several inference rules in probabilistic nonmonotonic reasoning. Finally, we examine some non p-valid inference rules; then, we illustrate by an example two methods which allow to suitably modify non p-valid inference rules in order to get inferences which are p-valid.
Similar content being viewed by others
References
Adams EW (1965) The logic of conditionals. Inquiry 8:166–197
Adams EW (1975) The logic of conditionals. Reidel, Dordrecht
Benferhat S, Dubois D, Prade H (1997) Nonmonotonic reasoning, conditional objects and possibility theory. Artif Intell 92:259–276
Biazzo V, Gilio A (2000) A generalization of the fundamental theorem of de Finetti for imprecise conditional probability assessments. Int J Approx Reason 24(2–3):251–272
Biazzo V, Gilio A, Lukasiewicz T, Sanfilippo G (2005) Probabilistic logic under coherence: complexity and algorithms. Ann Math Artif Intell 45(1–2):35–81
Biazzo V, Gilio A, Sanfilippo G (2008) Generalized coherence and connection property of imprecise conditional previsions. In: Proceedings of the IPMU 2008, Malaga, Spain, June 22–27, pp 907–914
Biazzo V, Gilio A, Sanfilippo G (2012) Coherent conditional previsions and proper scoring rules. In: Advances in computational intelligence. IPMU 2012, volume 300 of CCIS. Springer, Berlin, pp 146–156
Boole G (1857) On the Application of the theory of probabilities to the question of the combination of testimonies or judgments. Trans R Soc Edinb 21(4):597–653
Calabrese P (1987) An algebraic synthesis of the foundations of logic and probability. Inf Sci 42(3):187–237
Capotorti A, Lad F, Sanfilippo G (2007) Reassessing accuracy rates of median decisions. Am Stat 61 (2):132–138
Ciucci D, Dubois D (2012) Relationships between connectives in three-valued logics. In: Advances on computational intelligence, volume 297 of CCIS. Springer, Berlin, pp 633–642
Coletti G, Scozzafava R (1999) Conditioning and inference in intelligent systems. Soft Comput 3(3):118–130
Coletti G, Scozzafava R (2002) Probabilistic logic in a coherent setting. Kluwer, Dordrecht
Coletti G, Scozzafava R, Vantaggi B (2013) Coherent conditional probability, fuzzy inclusion and default rules. In: Yager RR, Abbasov A M, Reformat M Z, Shahbazova S N (eds) Soft computing: state of the art theory and novel applications. Springer, Heidelberg, pp 193–208
Coletti G, Scozzafava R, Vantaggi B (2015) Possibilistic and probabilistic logic under coherence: default reasoning and System P. Math Slovaca 65(4):863–890
de Finetti B (1936) La logique de la probabilité. In: Actes du Congrès International de Philosophie Scientifique, Paris, 1935. Hermann et C.ie, Paris, pp IV 1–IV 9
Edgington D (1995) On conditionals. Mind 104:235–329
Gilio A (2002) Probabilistic reasoning under coherence in System P. Ann Math Artif Intell 34:5–34
Gilio A (2012) Generalizing inference rules in a coherence-based probabilistic default reasoning. Int J Approx Reason 53(3):413–434
Gilio A, Sanfilippo G (2010) Quasi conjunction and p-entailment in nonmonotonic reasoning. In: Combining soft computing and statistical methods in data analysis, volume 77 of Advances in intelligent and soft computing. Springer, Berlin, pp 321–328
Gilio A, Sanfilippo G (2011) Coherent conditional probabilities and proper scoring rules. In: Proceedings of ISIPTA’11. Innsbruck, pp 189–198
Gilio A, Sanfilippo G (2013) Conditional random quantities and iterated conditioning in the setting of coherence. In: van der Gaag L C (ed) Symbolic and Quantitative Approaches to Reasoning with Uncertainty, ECSQARU 2013, volume 7958 of LNCS. Springer, Berlin, pp 218–229
Gilio A, Sanfilippo G (2013) Conjunction, disjunction and iterated conditioning of conditional events. In: Synergies of soft computing and statistics for intelligent data analysis, volume 190 of AISC. Springer, Berlin, pp 399–407
Gilio A, Sanfilippo G (2013) Probabilistic entailment in the setting of coherence: the role of quasi conjunction and inclusion relation. Int J Approx Reason 54(4):513–525
Gilio A, Sanfilippo G (2013) Quasi conjunction, quasi disjunction, t-norms and t-conorms: probabilistic aspects. Inf Sci 245:146– 167
Gilio A, Sanfilippo G (2014) Conditional random quantities and compounds of conditionals. Stud Log 102 (4):709–729
Gilio A, Sanfilippo G (2017) Conjunction and disjunction among conditional events. In: Benferhat S, Tabia K, Ali M (eds) IEA/AIE 2017, Part II, volume 10351 of LNCS. Springer, Cham, pp 85–96
Gilio A, Pfeifer N, Sanfilippo G (2016) Transitivity in coherence-based probability logic. J Appl Log 14:46–64
Gilio A, Over D, Pfeifer N, Sanfilippo G (2017) Centering and compound conditionals under coherence. In: Soft methods for data science, volume 456 of AISC. Springer, Berlin, pp 253–260
Gilio A, Pfeifer N, Sanfilippo G Probabilistic entailment and iterated conditionals, submitted. Available on arXiv:1804.06187v1
Goodman IR, Nguyen HT (1988) Conditional objects and the modeling of uncertainties. In: Gupta M M, Yamakawa T (eds) Fuzzy computing. North-Holland, pp 119–138
Goodman IR, Nguyen HT, Walker EA (1991) Conditional inference and logic for intelligent systems: a theory of measure-free conditioning. North-Holland, Amsterdam
Hailperin T (1996) Sentential probability logic. Origins, development, current status, and technical applications. Lehigh University Press, Bethlehem
Kaufmann S (2009) Conditionals right and left: probabilities for the whole family. J Philos Log 38:1–53
Lad F (1996) Operational subjective statistical methods: a mathematical, philosophical, and historical introduction. Wiley, New York
Lad F, Sanfilippo G, Agró G (2012) Completing the logarithmic scoring rule for assessing probability distributions. AIP Conf Proc 1490(1):13–30
Lad F, Sanfilippo G, Agró G (2015) Extropy: complementary dual of entropy. Stat Sci 30(1):40–58
McGee V (1989) Conditional probabilities and compounds of conditionals. Philos Rev 98(4):485–541
Milne P (1997) Bruno de Finetti and the logic of conditional events. Br J Philos Sci 48(2):195–232
Petturiti D, Vantaggi B (2017) Envelopes of conditional probabilities extending a strategy and a prior probability. Int J Approx Reason 81:160–182
Sanfilippo G, Pfeifer N, Gilio A (2017) Generalized probabilistic modus ponens. In: Antonucci A, Cholvy L, Papini O (eds) Symbolic and Quantitative Approaches to Reasoning with Uncertainty, ECSQARU 2017, volume 10369 of LNCS. Springer, Berlin, pp 480–490
Sanfilippo G, Pfeifer N, Over DE, Gilio A (2018) Probabilistic inferences from conjoined to iterated conditionals. Int J Approx Reason 93(Supplement C):103–118
Verheij B (2017) Proof with and without probabilities. Artif Intell Law 25(1):127–154
Acknowledgements
We thank the anonymous referees for their useful criticisms and suggestions.
Author information
Authors and Affiliations
Corresponding author
Additional information
Both authors contributed equally to this work.
This paper is a substantially extended version of [27].
Angelo Gilio is now retired.
Appendix A
Appendix A
Proof of Theorem 5
We observe that \((ii)\) follows by \((i)\), by replacing \(\mathcal {F}\) by \(\overline {\mathcal {F}}\); indeed, by \((i)\) it holds that \(\overline {\mathcal {D}(\overline {\mathcal {F}})}=\mathcal {C}(\overline {\overline {\mathcal {F}}})=\mathcal {C}(\mathcal {F})\). Then, it is enough to proof the assertion \((i)\). We will prove the assertion by induction.
-
Step 1:
\(n = 1, \mathcal {F}=\{E_{1}|H_{1}\}\).We have \(\overline {\mathcal {D}(\mathcal {F})}=\overline {E_{1}|H_{1}}= 1-E_{1}|H_{1}=\overline {E}_{1}|H_{1}=\mathcal {C}(\overline {\mathcal {F}})\).
Thus the assertion holds when \(n = 1\).
-
Step 2:
\(n = 2\), \(\mathcal {F}=\{E_{1}|H_{1},E_{2}|H_{2}\}\).
We set
$$\begin{array}{@{}rcl@{}} &&P(E_{1}|H_{1})=x, P(E_{2}|H_{2})=y, \mathbb{P}[(E_{1}|H_{1})\\ &&\vee (E_{2}|H_{2})]=w, \mathbb{P}[(\overline{E}_{1}|H_{1}) \wedge (\overline{E}_{2}|H_{2})]=t. \end{array} $$We observe that the family \(\{ E_{1}H_{1}\vee E_{2}H_{2}, \overline {E}_{1}H\overline {E}_{2}H_{2}, \overline {H}_{1}\overline {E}_{2}H_{2}, \overline {E}_{1}H_{1}\overline {H}_{2}, \overline {H}_{1}\overline {H}_{2}\}\) is a partition of the sure event \({\Omega }\). Moreover, by Definitions 3 and 4 we have
$$\begin{array}{@{}rcl@{}} \overline{\mathcal{D}(\mathcal{F})}&=&1-(E_{1}|H_{1})\vee (E_{2}|H_{2})\\ &=&\left\{\begin{array}{ll} 0, &\text{ if}~E_{1}H_{1}\vee E_{2}H_{2}~\text{is true,}\\ 1, &\text{ if}~\overline{E}_{1}H_{1}\overline{E}_{2}H_{2}~\text{is true,}\\ 1-x, &\text{ if}~\overline{H}_{1}\overline{E}_{2}H_{2}~\text{is true,}\\ 1-y, &\text{ if}~\overline{E}_{1}H_{1}\overline{H}_{2}~\text{is true,}\\ 1-w, &\text{ if}~\overline{H}_{1}\overline{H}_{2}~\text{is true}. \end{array} \right. \end{array} $$(39)and
$$\begin{array}{@{}rcl@{}} \mathcal{C}(\overline{\mathcal{F}})&=&(\overline{E}_{1}|H_{1})\wedge (\overline{E}_{2}|H_{2})\\ &=& \left\{\begin{array}{ll} 0, &\text{ if}~E_{1}H_{1}\vee E_{2}H_{2}~\text{is true,}\\ 1, &\text{ if}~\overline{E}_{1}H\overline{E}_{2}H_{2}~\text{is true,}\\ 1-x, &\text{ if}~\overline{H}_{1}\overline{E}_{2}H_{2}~\text{is true,}\\ 1-y, &\text{ if}~\overline{E}_{1}H_{1}\overline{H}_{2}~\text{is true,}\\ t, &\text{ if}~\overline{H}_{1}\overline{H}_{2}~\text{is true}. \end{array} \right. \end{array} $$(40)We observe that \(\overline {\mathcal {D}(\mathcal {F})}\) and \( {\mathcal {C}(\overline {\mathcal {F}})}\) coincide when \(H_{1}\vee H_{2}\) is true. Thus, by Theorem 3, \(\mathbb {P}(\overline {\mathcal {D}(\mathcal {F})})=\mathbb {P}(\mathcal {C}(\overline {\mathcal {F}}))\) and hence 1 − w = t. Therefore \(\overline {\mathcal {D}(\mathcal {F})}\) still coincides with \( \mathcal {C}(\overline {\mathcal {F}})\) when \(H_{1}\vee H_{2}\) is false, so that \(\overline {\mathcal {D}(\mathcal {F})}= \mathcal {C}(\overline {\mathcal {F}})\).
-
Step 3:
\(\mathcal {F}=\{E_{1}|H_{1},E_{2}|H_{2},\ldots , E_{n}|H_{n}\}\).
(Inductive Hypothesis) Let us assume that for any (strict) subset \(S\subset \{1,\ldots ,n\}\), by defining \(\mathcal {F}_{S}=\{E_{i}|H_{i}, i\in S\}\), it holds that \(\overline {{\mathcal {D}(\mathcal {F}_{S})}}=\mathcal {C}(\overline {\mathcal {F}_{S}})\). Now we will prove that \(\overline {{\mathcal {D}(\mathcal {F}_{S})}}=\mathcal {C}(\overline {\mathcal {F}_{S}})\) when \(S=\{1,\ldots ,n\}\), in which case \(\mathcal {F}_{S}=\mathcal {F}\). By Definition 7 we have
$$\begin{array}{@{}rcl@{}} \overline{\mathcal{D}(\mathcal{F})}\!&=&\!{\sum}_{h = 0}^{m} w_{h}C_{h}, \text{where}\\ w_{h}&=& \left\{ \begin{array}{llll} \!0, &\!\text{ if } S_{h}^{\prime}\neq \emptyset,\\ \!1, &\!\text{ if } S_{h}^{\prime\prime}=\{1,2,\ldots,n\}, \\ \!1 - y_{S^{\prime\prime\prime}_{h}},\! &\!\text{ if } S_{h}^{\prime}=\emptyset \text{ and } S_{h}^{\prime\prime\prime} \!\neq\emptyset. \end{array} \right.\\ \end{array} $$(41)We continue to use the subsets \(S_{h}^{\prime }, S_{h}^{\prime \prime }\), \(S_{h}^{\prime \prime \prime }\) as defined in formula (9) also with the family \(\overline {\mathcal {F}}\); moreover we set \(t_{S}=\mathbb {P}[\bigwedge _{i\in S} (\overline {E}_{i}|H_{i})]=\mathbb {P}[{\mathcal {C}(\overline {\mathcal {F}}_{S})}]\). Based on Definition 5, we have
$$\begin{array}{@{}rcl@{}} \mathcal{C}(\overline{\mathcal{F}})&=&{\sum}_{h = 0}^{m} z_{h}C_{h}, \text{where}\\ z_{h}&=&\left\{ \begin{array}{llll} 0, &\text{ if } S_{h}^{\prime}\neq \emptyset,\\ 1, &\text{ if } S_{h}^{\prime\prime}=\{1,2,\ldots,n\}, \\ t_{S^{\prime\prime\prime}_{h}}, &\text{ if } S_{h}^{\prime}=\emptyset \text{ and } S_{h}^{\prime\prime\prime} \neq\emptyset. \end{array} \right.\\ \end{array} $$(42)Then, \(\overline {\mathcal {D}(\mathcal {F})}-\mathcal {C}(\overline {\mathcal {F}})={\sum }_{h = 0}^{m} (w_{h}-z_{h})C_{h}\), where
$$\begin{array}{@{}rcl@{}} &&{}w_{h}-z_{h}\\ &&{\kern2pt}= \left\{ \begin{array}{llll} 0, &\text{ if } S_{h}^{\prime}\neq \emptyset,\\ 0, &\text{ if } S_{h}^{\prime\prime}=\{1,2,\ldots,n\}, \\ 1-y_{S^{\prime\prime\prime}_{h}}-t_{S^{\prime\prime\prime}_{h}}, &\text{ if } S_{h}^{\prime}=\emptyset \text{ and } S_{h}^{\prime\prime\prime} \neq\emptyset. \end{array} \right.\\ \end{array} $$(43)By the inductive hypothesis, it holds that \(1-y_{S^{\prime \prime \prime }_{h}}=\mathbb {P}[\overline {{\mathcal {D}(\mathcal {F}_{{S^{\prime \prime \prime }_{h}}})}}]=\mathbb {P}[\mathcal {C}(\overline {\mathcal {F}}_{{S^{\prime \prime \prime }_{h}}})]=t_{S^{\prime \prime \prime }_{h}}\) for h = 1,…,m, because \(S^{\prime \prime \prime }_{h}\subset \{1,2,\ldots ,n\}\). Then, \(\overline {\mathcal {D}(\mathcal {F})}-\mathcal {C}(\overline {\mathcal {F}})={\sum }_{h = 0}^{m} (w_{h}-z_{h})C_{h}\), where
$$ w_{h}-z_{h}= \left\{ \begin{array}{llll} 0, & h = 1,\ldots,m,\\ 1-y_{S^{\prime\prime\prime}_{0}}-t_{S^{\prime\prime\prime}_{0}}, & h = 0. \end{array} \right. $$(44)By recalling that \(S^{\prime \prime \prime }_{0}=\{1,2,\ldots ,n\}\), \(\overline {\mathcal {D}(\mathcal {F})}\) and \( {\mathcal {C}(\overline {\mathcal {F}})}\) coincide when \(H_{1}\vee H_{2} \vee {\cdots } \vee H_{n}\) is true. Thus, by Theorem 3, \(\mathbb {P}[\overline {\mathcal {D}(\mathcal {F})}]=\mathbb {P}[\mathcal {C}(\overline {\mathcal {F}}))]\), that is \(1-y_{S^{\prime \prime \prime }_{0}}=t_{S^{\prime \prime \prime }_{0}}\). Therefore \(\overline {\mathcal {D}(\mathcal {F})}\) still coincides with \( \mathcal {C}(\overline {\mathcal {F}})\) when \(H_{1}\vee H_{2} \vee {\cdots } \vee H_{n}\) is false, so that \(\overline {\mathcal {D}(\mathcal {F})}= \mathcal {C}(\overline {\mathcal {F}})\).
□
Proof of Theorem 6
-
(i)
Assume that, for every \((\mu ,\nu )\in {\Pi }\), the values of \(X|H\) and \(Y|K\) associated with the constituent \(C_{h}\) are such that \(X|H\leq Y|K\), for each \(C_{h}\) contained in H ∨ K; then for each given coherent assessment \((\mu ,\nu )\), by choosing \(s_{1}= 1, s_{2}=-1\) in the random gain, we have
$$\begin{array}{@{}rcl@{}} G\!&=&\!H(X - \mu)-K(Y - \nu) = (X|H - \mu) - (Y|K - \nu)\\ \!&=&\! (X|H - Y|K) + (\nu - \mu) . \end{array} $$Then, by the hypothesis, \(\mathcal {G}_{H\vee K}\leq (\nu - \mu )\) and by coherence \(0=\mathbb {P}(\mathcal {G}_{H\vee K}) \leq \nu - \mu \). Then \(\mu \leq \nu \), \( \forall (\mu ,\nu ) \in {\Pi }\).
-
(ii)
By hypothesis, it holds that \((XH+\mu H^{c})(H\vee K)\leq (YK+\nu K^{c})(H\vee K)\); moreover, from condition (i), \(\mu \leq \nu \) for every \((\mu ,\nu )\in {\Pi }\); then
$$\begin{array}{@{}rcl@{}} X|H\! &=&\! XH+\mu \overline{H}=(XH+\mu \overline{H})(H\vee K)\\ &&\!+(XH + \mu \overline{H})\overline{H} \overline{K} = (XH + \mu \overline{H})(H\!\vee\! K)\\ &&\!+\mu \overline{H} \overline{K}\leqslant (YK + \nu \overline{K})(H\vee K) + \nu \overline{HK}\\ &&\!=(YK + \nu \overline{K})(H\vee K) + (YK + \nu \overline{K})\overline{HK}\\ &&\!= YK + \nu \overline{K} = Y|K . \end{array} $$Vice versa, \(X|H\leq Y|K\) trivially implies \(X|H\leq Y|K\) when \(H\vee K\) is true.
□
Proof of Theorem 7
We distinguish three cases: \((a)\) the value of \(\mathcal {C}_{n}\) is 0, with some \(E_{i}|H_{i}\) false, \(i\leq n\); \((b)\) the value of \(\mathcal {C}_{n}\) is 1, with \(E_{i}|H_{i}\) true, \(i = 1,\ldots ,n\); \((c)\) the value of \(\mathcal {C}_{n}\) is \(\mathbb {P}[\bigwedge _{i\in S} (E_{i}|H_{i})]=\mathbb {P}(\mathcal {C}_{S})=x_{S}\), for some subset \(S\subseteq \{1,2,\ldots ,n\}\).
-
Case (a). It holds that \(\mathcal {C}_{n + 1}= 0=\mathcal {C}_{n}\).
-
Case (b). The value of \(\mathcal {C}_{n + 1}\) is 1, or 0, or \(x_{n + 1}\), according to whether \(E_{n + 1}|H_{n + 1}\) is true, or false, or void; thus \(\mathcal {C}_{n + 1}\leq \mathcal {C}_{n}\).
-
Case (c). We distinguish three cases: \((i)\)En+ 1|Hn+ 1 is true; \((ii)\)En+ 1|Hn+ 1 is false; \((iii)\)En+ 1|Hn+ 1 is void. In the case \((i)\) the value of \(\mathcal {C}_{n + 1}\) is \(x_{S}\), thus \(\mathcal {C}_{n + 1}=\mathcal {C}_{n}\). In the case \((ii)\) the value of \(\mathcal {C}_{n + 1}\) is 0, thus \(\mathcal {C}_{n + 1}\leq \mathcal {C}_{n}\). In the case \((iii)\) the value of \(\mathcal {C}_{n + 1}\) is \(x_{S\cup \{n + 1\}}=\mathbb {P}[\bigwedge _{i\in S\cup \{n + 1\}}(E_{i}|H_{i})]\); then, in order to prove that \(\mathcal {C}_{n + 1}\leq \mathcal {C}_{n}\), we need to prove that \(x_{S\cup \{n + 1\}}\leq x_{S}\). We proceed by induction on the cardinality of S, denoted by s. Let be \(s = 1\), with \(\mathcal {C}_{S}=E_{i}|H_{i}\), for some \(i\in \{1,\ldots ,n\}\). We note that \(x_{S}=\mathbb {P}(E_{i}|H_{i})=x_{i}\), \(x_{S\cup \{n + 1\}}=\mathbb {P}((E_{i}|H_{i})\wedge (E_{n + 1}|H_{n + 1}))=x_{\{i,n + 1\}}\) and by Theorem 4 it holds that \(x_{S\cup \{n + 1\}}=x_{\{i,n + 1\}}\leq x_{i}=x_{S}\). Now, let be s ≥ 2 and \(x_{S\cup \{n + 1\}}\leq x_{S}\) for every \(s< n\), so that, based on Definition 5, \(\mathcal {C}_{n + 1}\leq \mathcal {C}_{n}\) when S is a strict subset of {1, 2,…,n}. If \(S=\{1,2,\ldots ,n\}\), as \(E_{i}|H_{i}\) is void for all \(i = 1,\ldots ,n + 1\), it holds that \(\mathcal {C}_{n}=\mathbb {P}(\mathcal {C}_{n})=x_{\{1,\ldots ,n\}}\) and \(\mathcal {C}_{n + 1}=x_{\{1,\ldots ,n + 1\}}=\mathbb {P}(\mathcal {C}_{n + 1})\) and, in order to prove that \(\mathcal {C}_{n + 1}\leq \mathcal {C}_{n}\), it remains to prove that \(\mathbb {P}(\mathcal {C}_{n + 1})\leq \mathbb {P}(\mathcal {C}_{n})\). By applying Theorem 6, with \(X|H=\mathcal {C}_{n + 1}=Z_{n + 1}|(H_{1}\vee {\cdots } \vee H_{n + 1})\) and \(Y|K=\mathcal {C}_{n}=Z_{n}|(H_{1}\vee {\cdots } \vee H_{n})\), as \(\mathcal {C}_{n + 1}\leq \mathcal {C}_{n}\) when H1 ∨⋯ ∨ Hn+ 1 is true (i.e., \(s<n\) ), it follows that \(\mathbb {P}(\mathcal {C}_{n + 1})\leq \mathbb {P}(\mathcal {C}_{n})\); therefore \(\mathcal {C}_{n + 1}\leq \mathcal {C}_{n}\).
□
Proof
of Theorem 9.
Case \((i)\). We proceed by induction. The property is satisfied for \(n = 1\); indeed, if \(\mathcal {C}_{1}=E_{1}|H_{1}\in \{1,0,x_{1}\}\), where \(x_{1}=\mathbb {P}(E_{1}|H_{1}) \in [0,1]\), then \(\mathcal {C}_{1}\in [0,1]\). Let us assume that the property holds for \(k<n\), that is \(\mathcal {C}_{k}\in [0,1]\), for every \(k<n\). Based on Definition 5 we distinguish three cases: (a) the value of \(\mathcal {C}_{n}\) is 0; \((b)\) the value of \(\mathcal {C}_{n}\) is 1; \((c)\) the value of \(\mathcal {C}_{n}\) is \(\mathbb {P}[\bigwedge _{i\in S} (E_{i}|H_{i})]=x_{S}\), for some subset \(S\subseteq \{1,2,\ldots ,n\}\). In the cases \((a)\) and \((b)\), \(\mathcal {C}_{n}\in [0,1]\). In the case \((c)\), if \(S=\{i_{1},\ldots ,i_{k}\}\subset \{1,2,\ldots ,n\}\), then \(\mathcal {C}_{n} \in [0,1]\), because \(x_{S}=\mathbb {P}(\bigwedge _{j = 1}^{k} (E_{i_{j}}|H_{i_{j}}))\) is a possible value of \(\mathcal {C}_{k}=\bigwedge _{j = 1}^{k} (E_{i_{j}}|H_{i_{j}})\), with \(k<n\). Finally, if \(S=\{1,2,\ldots ,n\}\) (that is the conditioning events \(H_{1},\ldots , H_{n}\) are all false), then \(\mathcal {C}_{n}=\mathbb {P}(\mathcal {C}_{n})\) and \(\mathbb {P}(\mathcal {C}_{n})\in [0,1]\) because the values of \(\mathcal {C}_{n}\) restricted to H1 ∨⋯ ∨ Hn all belong to \([0,1]\). Therefore \(\mathcal {C}_{n}\in [0,1]\). By a similar reasoning, based on Definition 7 we can prove that \(\mathcal {D}_{n}\in [0,1]\). □
Proof of Theorem 10
Let \(C_{0},\ldots , C_{m}\), with \(m = 3^{n}-1\) be the constituents associated with \(\mathcal {F}_{n + 1}=\{E_{1}|H_{1},\ldots , E_{n + 1}|H_{n + 1}\}\), where \(C_{0}=\overline {H}_{1}{\cdots } \overline {H}_{n + 1}\). With each \(C_{h}\), \(h = 1,\ldots ,m\), we associate the point \(Q_{h}=(q_{h1},q_{h2},q_{h3})\), which represents the value of the random vector \((\mathcal {C}_{n},E_{n + 1}|H_{n + 1},\mathcal {C}_{n + 1})\) when \(C_{h}\) is true, where \(q_{h1}\) is the value of \(\mathcal {C}_{n}\), \(q_{h2}\) is the value of \(E_{n + 1}|H_{n + 1}\), and \(q_{h3}\) is the value of \(\mathcal {C}_{n + 1}\). With \(C_{0}\) it is associated the point \(Q_{0}=(\mu _{n},x_{n + 1},\mu _{n + 1})=\mathcal {M}\). We observe that the set of points \(\{Q_{h}, h = 1,\ldots ,m\}\) contains in particular the points
which are respectively associated with the following constituents or logical disjunction of constituents
Based on Remark 7, we need to prove that the set of coherent assessments \({\Pi }\) on \(\{\mathcal {C}_{n}, E_{n + 1}|H_{n + 1}, \mathcal {C}_{n + 1}\}\) coincides with the convex hull \(\mathcal {I}\) of Q1,Q2,Q3,Q4. We recall that coherence of \((\mu _{n},x_{n + 1},\mu _{n + 1})\) implies coherence of all the sub-assessments on the associated subfamilies of \(\{\mathcal {C}_{n}, E_{n + 1}|H_{n + 1}, \mathcal {C}_{n + 1}\}\). The coherence of the single assessments \(\mu _{n}\) on \(\mathcal {C}_{n}\), or \(x_{n + 1}\) on \(E_{n + 1}|H_{n + 1}\), or \(\mu _{n + 1}\) on \(\mathcal {C}_{n + 1}\), simply amounts to conditions
respectively. Then, by the hypothesis of logical independence, the sub-assessment \((\mu _{n},x_{n + 1})\) is coherent, for every \((\mu _{n},x_{n + 1}) \in [0,1]^{2}\). By Remark 6, the coherence of the sub-assessments \((\mu _{n},\mu _{n + 1})\) and \((x_{n + 1},\mu _{n + 1})\) amounts to the conditions \(0\leq \mu _{n + 1}\leq \mu _{n} \leq 1\) and 0 ≤ μn+ 1 ≤ xn+ 1 ≤ 1. Finally, assuming that the above conditions are satisfied, to prove coherence of \((\mu _{n},x_{n + 1},\mu _{n + 1})\), by Theorem 1, it is enough to show that the point \((\mu _{n},x_{n + 1},\mu _{n + 1})\) belongs to the convex hull of the points \(Q_{1},\ldots ,Q_{m}\). Moreover, in order \(\mathcal {M}\) belongs to the convex hull of Q1,…,Qm the following system \(({\Sigma })\) must solvable
We show that the convex hull of the points \(Q_{1},\ldots ,Q_{m}\) coincides with the convex hull \(\mathcal {I}\) of the points \(Q_{1},Q_{2},Q_{3},Q_{4}\), described in Remark 7, because all the other points \(Q_{5},\ldots ,Q_{m}\), are linear convex combinations of \(Q_{1},Q_{2},Q_{3},Q_{4}\), that is \(Q_{h}\in \mathcal {I}\) for each \(h = 5,\ldots ,m\).
We examine the following different cases which depend on the logical value of \(E_{n + 1}|H_{n + 1}\): \(a)\)En+ 1|Hn+ 1 is true; \(b)\)En+ 1|Hn+ 1 is false; \(c)\)En+ 1|Hn+ 1 is void.
-
a)
In this case
$$\begin{array}{@{}rcl@{}} Q_{h}&=&(q_{h1},1,q_{h1})= q_{h1}(1,1,1)+(1-q_{h1})(0,1,0)\\ &=&q_{h1} Q_{1}+(1-q_{h1})Q_{3}. \end{array} $$ -
b)
In this case
$$\begin{array}{@{}rcl@{}} Q_{h}&=&(q_{h1},0,0)= q_{h1}(1,0,0)+(1-q_{h1})(0,0,0)\\ &=&q_{h1} Q_{2}+(1-q_{h1})Q_{4}. \end{array} $$ -
c)
In this case \(Q_{h}=(q_{h1},x_{n + 1},q_{h3})\) and we distinguish the following subcases: \((i)\)\(\bigwedge _{i = 1}^{n} E_{i}H_{i}\) true, so that \(Q_{h}=(1,x_{n + 1},x_{n + 1})\); \((ii)\)\(\bigvee _{i = 1}^{n} \overline {E}_{i}H_{i}\) true, so that \(Q_{h}=(0,x_{n + 1},0)\); \((iii)\)Ei|Hi void, for every \(i \in S\) and \(E_{i}|H_{i}\) true for every i ∈{1, 2,…,n}∖ S,. for some \(\emptyset \neq S \subset \{1,2,\ldots ,n\}\), so that \(Q_{h}=(x_{S},x_{n + 1},x_{S\cup \{n + 1\}})\). In subcase \((i)\) it holds that
$$\begin{array}{@{}rcl@{}} Q_{h}\!&=&\!({\kern-.5pt}1{\kern-.5pt},{\kern-.5pt}x_{n + 1}{\kern-.5pt},{\kern-.5pt}x_{n + 1{\kern-.5pt}}) = x_{n{\kern-.5pt}+{\kern-.5pt}1}({\kern-.5pt}1{\kern-.5pt},{\kern-.5pt}1{\kern-.5pt},{\kern-.5pt}1{\kern-.5pt}) + ({\kern-.5pt}1 - x_{n + 1})({\kern-.5pt}1{\kern-.5pt},{\kern-.5pt}0{\kern-.5pt},{\kern-.5pt}0{\kern-.5pt})\\ \!&=&\! x_{n + 1} Q_{1}+(1-x_{n + 1})Q_{2}. \end{array} $$In subcase \((ii)\) it holds that
$$\begin{array}{@{}rcl@{}} Q_{h}&=&({\kern-.5pt}0{\kern-.5pt},{\kern-.5pt}x_{n + 1}{\kern-.5pt},{\kern-.5pt}0)= x_{n + 1}({\kern-.5pt}0{\kern-.5pt},{\kern-.5pt}1{\kern-.5pt},{\kern-.5pt}0{\kern-.5pt})+(1\!-x_{n + 1})({\kern-.5pt}0{\kern-.5pt},{\kern-.5pt}0{\kern-.5pt},0)\\ &=& x_{n + 1} Q_{3}+(1-x_{n + 1})Q_{4}. \end{array} $$In subcase \((iii)\), it can be verified by a finite iterative procedure that the point \(Q_{h}=(x_{S},x_{n + 1},x_{S\cup \{n + 1\}})\in \mathcal {I}\). We examine the different cases on the cardinality s of S. We recall that \(\bigwedge _{i\in S}(E_{i}|H_{i})\) is denoted by \(\mathcal {C}_{S}\).
-
Step 1.
\(s = 1\). Without loss of generality we assume \(S=\{1\}\), so that \(Q_{h}=(x_{S},x_{n + 1},x_{S\cup \{n + 1\}})=(x_{1},x_{n + 1},x_{\{1,n + 1\}})\), where \(x_{1}=P(E_{1}|H_{1})\), \(x_{\{1,n + 1\}}=\mathbb {P}[(E_{1}|H_{1})\wedge (E_{n + 1}|H_{n + 1})]\). By Theorem 4 it holds that \(\max \{x_{S}+x_{n + 1}-1,0\} \leq x_{S\cup \{n + 1\}} \leq \min \{x_{S},x_{n + 1}\}\), with (xS,xn+ 1) ∈ [0, 1]2. In other words, \(Q_{h}=(x_{S},x_{n + 1},x_{S\cup \{n + 1\}})\in \mathcal {I}\). The reasoning is the same for \(S=\{i\}\), \(i = 2,\ldots ,n\).
-
Step 2.
\(s = 2\). Without loss of generality we assume \(S=\{1,2\}\), so that \(x_{S}=\mathbb {P}[(E_{1}|H_{1})\wedge (E_{2}|H_{2})]\), \(x_{S\cup \{n + 1\}}=\mathbb {P}[\mathcal {C}_{S\cup \{n + 1\}}]=\mathbb {P}[(E_{1}|H_{1})\wedge (E_{2}|H_{2})\wedge (E_{n + 1}|H_{n + 1})]\). We denote by \(C_{0}^{*},C_{1}^{*}, \ldots ,C_{m^{*}}^{*}\), the constituents associated with \(\{E_{i}|H_{i}, i \in S\cup \{n + 1\}\}\), where \(C_{0}^{*}=\bigwedge _{i\in S\cup \{n + 1\}}\overline {H}_{i}\). Moreover, with \(C_{h}^{*}\), \(h = 0,1, \ldots ,m^{*},\) we associate the point \(Q_{h}^{*}=(q_{h1}^{*},q_{h2}^{*},q_{h3}^{*})\) which represents the value of the random vector \(\{\mathcal {C}_{S},E_{n + 1}|H_{n + 1},\mathcal {C}_{S\cup \{n + 1\}}\}\) when \(C_{h}^{*}\) is true. We observe that \(Q_{0}^{*}=(x_{S},x_{n + 1},x_{S\cup \{n + 1\}})\) and that \(Q_{1},Q_{2},Q_{3},Q_{4}\) still belongs to the set of points \(\{Q_{h}^{*}, h = 1,\ldots ,m^{*}\}\). In order that the assessment (xS,xn+ 1,xS∪{n+ 1}) on \(\{\mathcal {C}_{S},E_{n + 1}|H_{n + 1},\mathcal {C}_{S\cup \{n + 1\}}\}\) be coherent, the point \(Q_{0}^{*}=(x_{S},x_{n + 1},x_{S\cup \{n + 1\}})\) must belong to the convex hull of points \(Q_{1}^{*}, Q_{2}^{*}, \ldots , Q_{m}^{*}\). We show that for each point \(Q_{h}^{*}\neq Q_{i}\), \(i = 1,2,3,4\), it holds that \(Q_{h}^{*}\in \mathcal {I}\). By repeating the previous reasoning we only need to analyze the subcase \((iii)\) of case \(c)\). We have to show that, for every nonempty subset \(S^{\prime } \subset S\), the point \(Q_{h}^{*}=(x_{S^{\prime }},x_{n + 1},x_{S^{\prime }\cup \{n + 1\}})\) belongs to the convex hull \(\mathcal {I}\) of \(Q_{1},\ldots ,Q_{4}\). As \(S=\{1,2\}\), it holds that \(S^{\prime }=\{1\}\), or \(S^{\prime }=\{2\}\), so that \(Q_{h}^{*}=(x_{S^{\prime }},x_{n + 1},x_{S^{\prime }\cup \{n + 1\}})=(x_{1},x_{n + 1},x_{\{1,n + 1\}})\), or \(Q_{h}^{*}=(x_{2},x_{n + 1},x_{\{2,n + 1\}})\). By Step 1, in both cases \(Q_{h}^{*} \in \mathcal {I}\). Thus \(Q_{h}=(x_{S},x_{n + 1},x_{S\cup \{n + 1\}}) \in \mathcal {I}\). In other words, \(\max \{x_{S}+x_{n + 1}-1,0\} \leq x_{S\cup \{n + 1\}} \leq \min \{x_{S},x_{n + 1}\}\), with \((x_{S},x_{n + 1}) \in [0,1]^{2}\). The reasoning is the same for every \(S=\{i,j\} \subset \{1,2,\ldots ,n\}\).
$$............................................................................... $$Step\(k + 1\). \(s=k + 1\), \(2<k + 1<n\). By induction, assume that \((x_{S^{\prime }},x_{n + 1},x_{S^{\prime }\cup \{n + 1\}}) \in \mathcal {I}\) for every \(S^{\prime }=\{i_{1},i_{2},\ldots ,i_{k}\} \subset \{1,2,\ldots ,n\}\). Then, by the previous reasoning, it follows that \(Q_{h}=(x_{S},x_{n + 1},x_{S\cup \{n + 1\}}) \in \mathcal {I}\) for every \(S=\{i_{1},i_{2},\ldots ,i_{k + 1}\}\). In other words, \(\max \{x_{S}+x_{n + 1}-1,0\} \leq x_{S\cup \{n + 1\}} \leq \min \{x_{S},x_{n + 1}\}\), with \((x_{S},x_{n + 1}) \in [0,1]^{2}\), for every \(S=\{i_{1},i_{2},\ldots ,i_{k + 1}\}\).
Thus, by this iterative procedure, also in the subcase \((iii)\) of case \(c)\) it holds that \(Q_{h}\in \mathcal {I}\). Then, \(Q_{h}\in \mathcal {I}\), \(h = 5,\ldots ,m\). Finally, the condition (45) is equivalent to \(\mathcal {M} \in \mathcal {I}\), so that the assessment \(\mathcal {M}\) is coherent if and only if
□
Proof of Theorem 15
The computation of the set \({\Pi }\) is based on Section 2.2. The constituents \(C_{h}\)’s and the points \(Q_{h}\)’s associated with \((\mathcal {F},\mathcal {M})\) are illustrated in Table 1.
We recall that \(Q_{h}=(q_{h1},\ldots ,q_{h7})\) represents the value associated with \(C_{h}\) of the random vector (E1|H1,E2|H2,E3|H3, (E1|H1) ∧ (E2|H2), \((E_{1}|H_{1})\wedge (E_{3}|H_{3}), (E_{2}|H_{2})\wedge (E_{3}|H_{3})\), \({(E_{1}|H_{1})\wedge (E_{2}|H_{2})\wedge (E_{3}|H_{3})})\), \(h = 1,\ldots ,26\). With \(C_{0}=\overline {H_{1}}\overline {H_{2}}\overline {H_{3}}\) it is associated \(Q_{0}=\mathcal {M}\). Denoting by \(\mathcal {I}\) the convex hull generated by \(Q_{1},Q_{2}, \ldots ,Q_{26}\), the coherence of the prevision assessment \(\mathcal {M}\) on \(\mathcal {F}\) requires that the condition \(\mathcal {M}\in \mathcal {I}\) be satisfied; this amounts to the solvability of the following system
We observe that
Thus, \(\mathcal {I}\) coincides with the convex hull of the points \(Q_{1}, Q_{2}, Q_{4}, Q_{5}, Q_{10}, Q_{11}, Q_{13}, Q_{14}\). For the sake of simplicity, we set: \(Q_{1}^{\prime }=Q_{1}, Q_{2}^{\prime }=Q_{2}\), \(Q_{3}^{\prime }=Q_{4}\), \(Q_{4}^{\prime }=Q_{5}\), \(Q_{5}^{\prime }=Q_{10}\), \(Q_{6}^{\prime }=Q_{11}, Q_{7}^{\prime }=Q_{13}\), \(Q_{8}^{\prime }=Q_{14}\). Then, the condition \(\mathcal {M}\in \mathcal {I}\) amounts to the solvability of the following system
that is
System \(({\Sigma }^{\prime })\) can be written as
As it can be verified, by non-negativity of \(\lambda ^{\prime }_{1},\ldots ,\lambda ^{\prime }_{8}\) it follows that \(({\Sigma }^{\prime })\) is solvable (with a unique solution) if and only if
or, in a more explicit way, if and only if the following conditions are satisfied
Notice that the conditions in (47) coincide with that ones in (18). Moreover, assuming \(({\Sigma }^{\prime })\) solvable, with the solution \((\lambda _{1}^{\prime },\ldots ,\lambda _{8}^{\prime })\), we associate the vector \((\lambda _{1},\lambda _{2},\ldots ,\lambda _{26})\), with \(\lambda _{1}=\lambda _{1}^{\prime }, \lambda _{2}=\lambda _{2}^{\prime }, \lambda _{4}=\lambda _{3}^{\prime }, \lambda _{5}=\lambda _{4}^{\prime }, \lambda _{10}=\lambda _{5}^{\prime }, \lambda _{11}=\lambda _{6}^{\prime }, \lambda _{13}=\lambda _{7}^{\prime }, \lambda _{14}=\lambda _{8}^{\prime }, \lambda _{h}= 0, h\notin \{1,2,4,5,10,11,13,14\}, \) which is a solution of (Σ). Moreover, defining \(\mathcal {J}=\{ 1,2,4,5,10,11,13,14\}\), it holds that \(\bigvee _{h\in \mathcal {J}}C_{h}= H_{1}\wedge H_{2}\wedge H_{3}\). Therefore, \({\sum }_{h\in \mathcal {J}}\lambda _{h} ={\sum }_{h: C_{h} \subseteq H_{1}H_{2}H_{3}} \lambda _{h} = 1\) and hence \({\sum }_{h: C_{h} \subseteq H_{i}}\lambda _{h}= 1\), \(i = 1,2,3\), \({\sum }_{h: C_{h} \subseteq H_{i}\vee H_{j}}\lambda _{h}= 1\), \(i\neq j\), \({\sum }_{h: C_{h} \subseteq H_{1}\vee H_{2}\vee H_{3}}\lambda _{h}=~1\); thus, by (2), \(I_{0} = \emptyset \). Then, by Theorem 2, the solvability of (Σ) is also sufficient for the coherence of \(\mathcal {M}\). Finally, \({\Pi }\) is the set of conditional prevision assessments \((x_{1},x_{2},x_{3},x_{12},x_{13},_{23},x_{123})\) which satisfy the conditions in (18). □
Proof of Theorem 16
Notice that, \((E_{i}|H)\wedge (E_{j}|H)=(E_{i}E_{j})|H\), for every \(\{i,j\}\subset \{1,2,3\}\), and \((E_{1}|H)\wedge (E_{2}|H)\wedge (E_{3}|H)=(E_{1}E_{2}E_{3})|H\). Then \(\mathcal {F}=\{E_{1}|H,E_{2}|H,E_{3}|H, (E_{1}E_{2})|H, (E_{1}E_{3})|H, (E_{2}E_{3})|H\), \((E_{1}E_{2}E_{3})|H\}\). The computation of the set \({\Pi }\) is based on Section 2.2. The constituents \(C_{h}\)’s and the points \(Q_{h}\)’s associated with \((\mathcal {F},\mathcal {M})\) are illustrated in Table 2.
We recall that \(Q_{h}=(q_{h1},\ldots ,q_{h7})\) represents the value associated with \(C_{h}\) of the random vector \( (E_{1}|H,E_{2}|H,E_{3}|H, (E_{1}E_{2})|H\), (E1E3)|H, (E2E3)|H, \((E_{1}E_{2}E_{3})|H)\), \(h = 1,\ldots ,8\). With \(C_{0}=\overline {H}\) it is associated \(Q_{0}=\mathcal {M}\). Denoting by \(\mathcal {I}\) the convex hull generated by Q1,Q2,…,Q8, as all the conditioning events coincide with H the assessment \(\mathcal {M}\) on \(\mathcal {F}\) is coherent if and only if \(\mathcal {M}\in \mathcal {I}\); that is, if and only if the following system is solvable
The points \(Q_{1},Q_{2}, \ldots ,Q_{8}\) coincide with the points \(Q_{1}^{\prime },Q_{2}^{\prime }, \ldots ,Q_{8}^{\prime }\) in the proof of Theorem 15, respectively. Then, system (48) coincides with system \(({\Sigma })^{\prime }\) in the proof of Theorem 15. Therefore, it is solvable if and only if the conditions in (18) are satisfied. In other words, the set \({\Pi }\) of all coherent assessments \(\mathcal {M}\) on \(\mathcal {F}\) coincides with the set of points (x1,x2,x3,x12,x13,x23,x123) which satisfy the conditions in (18). □
Proof of of Theorem 18
In order to prove the theorem it is enough to prove the following implications: a) \((i) \Rightarrow (ii)\); b) \((ii) \Rightarrow (iii)\); c) \((iii) \Rightarrow (i)\).
-
a)
\((i) \Rightarrow (ii)\). We recall that \(\mathcal {F}\) p-entails \(E_{n + 1}|H_{n + 1}\) if and only if either \(H_{n + 1} \subseteq E_{n + 1}\), or there exists a nonempty \(\mathcal {F}_{{\Gamma }} \subseteq \mathcal {F}\), where \({\Gamma } \subseteq \{1,\ldots ,n\}\), such that \(QC(\mathcal {F}_{{\Gamma }})\) implies \(E_{n + 1}|H_{n + 1}\) (see, e.g. [24, Theorem 6]). Let us first consider the case where \(H_{n + 1} \subseteq E_{n + 1}\). In this case \(P(E_{n + 1}|H_{n + 1}) = 1\) and \(E_{n + 1}|H_{n + 1} =H_{n + 1}+\overline {H}_{n + 1}= 1\). We have \(\mathcal {C}_{n + 1}=\mathcal {C}_{n} \wedge (E_{n + 1}|H_{n + 1})\), with En+ 1|Hn+ 1 = 1. We distinguish two cases: \((\alpha )~H_{n + 1}\) is true; \((\beta )~H_{n + 1}\) is false. In case \((\alpha )\), by Definition 5 and Remark 4, as \(E_{n + 1}|H_{n + 1}\) is true it follows that the values of \(\mathcal {C}_{n + 1}\) and of \(\mathcal {C}_{n}\) coincide. In case \((\beta )\), let \(C_{0},\ldots ,C_{m}\) be the constituents associated with \(\mathcal {F}\), where \(C_{0}=\overline {H}_{1}{\cdots } \overline {H}_{n}\). Then, the constituents \(C_{0}^{\prime },\ldots ,C_{m}^{\prime }\) associated with \(\mathcal {F}\cup \{E_{n + 1}|H_{n + 1}\}\) and contained in \(\overline {H}_{n + 1}\) are \(C_{0}^{\prime }=C_{0}\overline {H}_{n + 1},\ldots ,C_{m}^{\prime }=C_{m}\overline {H}_{n + 1}\). For each constituent \(C_{h}^{\prime }\), \(h = 1,\ldots ,m\), by formula (10) the corresponding value of \(\mathcal {C}_{n}\) is \(z_{h}\in \{1,0,x_{S_{h}^{\prime \prime \prime }}\}\). We denote by \(z_{h}^{\prime }\) the value of \(\mathcal {C}_{n + 1}\) associated with \(z_{h}\) and we recall that \(C_{h}^{\prime }\subseteq \overline {H}_{n + 1}\), \(h = 0,1,\ldots ,m\). For each index h, if \(z_{h}= 1\), then \(z^{\prime }_{h}= 1\); if \(z_{h}= 0\), then \(z^{\prime }_{h}= 0\); if \(z_{h}=x_{S_{h}^{\prime \prime \prime }}\), then \(z_{h}^{\prime }=x_{S_{h}^{\prime \prime \prime }\cup \{n + 1\}}\). We set \(P(E_{n + 1}|H_{n + 1})=x_{n + 1}\); in our case \(x_{n + 1}= 1\). Moreover, by Theorem 10
$$\begin{array}{@{}rcl@{}} \max\!\left\{x_{S_{h}^{\prime\prime\prime}} + x_{n + 1} - 1,\!0\right\}\!\leq\! x_{S_{h}^{\prime\prime\prime}\cup\{n + 1\}}\!\leq\! \min\!\left\{x_{S_{h}^{\prime\prime\prime}},\!x_{n + 1}{\kern-.5pt}\right\}\!; \end{array} $$therefore \( x_{S_{h}^{\prime \prime \prime }\cup \{n + 1\}}=x_{S_{h}^{\prime \prime \prime }}\). Then, the values of \(\mathcal {C}_{n + 1}\) and of \(\mathcal {C}_{n}\) coincide for every \(C^{\prime }_{h}\). Thus, \(\mathcal {C}_{n + 1}=\mathcal {C}_{n}\) when \(H_{n + 1}\subseteq E_{n + 1}\).
We consider now the case where there exists \(\mathcal {F}_{{\Gamma }}\subseteq \mathcal {F}\), \(\mathcal {F}_{{\Gamma }}\neq \emptyset \), such that \(QC(\mathcal {F}_{{\Gamma }})\subseteq E_{n + 1}|H_{n + 1}\). First of all we prove that \(\mathcal {C}(\mathcal {F}_{{\Gamma }}\cup \{E_{n + 1}|H_{n + 1}\} )=\mathcal {C}(\mathcal {F}_{{\Gamma }})\). For the sake of simplicity, we set \(\mathcal {C}(\mathcal {F}_{{\Gamma }})=\mathcal {C}_{{\Gamma }}\) and \(\mathcal {C}(\mathcal {F}_{{\Gamma }}\cup \{E_{n + 1}|H_{n + 1}\} )=\mathcal {C}_{{\Gamma }\cup \{n + 1\}}\).
If the value of \(\mathcal {C}_{{\Gamma }}\) is 1 (because all the conditional events in \(\mathcal {F}_{{\Gamma }}\) are true), then \(QC(\mathcal {F}_{{\Gamma }})\) is true and hence \(E_{n + 1}|H_{n + 1}\) is also true; thus \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}= 1\), so that \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}=\mathcal {C}_{{\Gamma }}\).
If the value of \(\mathcal {C}_{{\Gamma }}\) is 0 (because some conditional event in \(\mathcal {F}_{{\Gamma }}\) is false), then \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}\) is 0 too, so that \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}=\mathcal {C}_{{\Gamma }}\).
If \(\mathcal {C}_{{\Gamma }}\) is \(x_{S}\) for some nonempty subset \(\mathcal {S}\subset {\Gamma }\) (that is, all the conditional events in \(\mathcal {F}_{S}\) are void and the other ones in \(\mathcal {F}_{{\Gamma } \setminus S}\) are true), then \(QC(\mathcal {F}_{{\Gamma }})\) is true and and hence \(E_{n + 1}|H_{n + 1}\) is also true; thus \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}=x_{S}\), so that \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}=\mathcal {C}_{{\Gamma }}\).
If \(\mathcal {C}_{{\Gamma }}\) is \(x_{{\Gamma }}\) because all the conditional events in \(\mathcal {F}_{{\Gamma }}\) are void, then \(QC(\mathcal {F}_{{\Gamma }})\) is void and for \(E_{n + 1}|H_{n + 1}\) there are two cases: 1) \(E_{n + 1}|H_{n + 1}\) true; 2) \(E_{n + 1}|H_{n + 1}\) void. In case 1), by also recalling Remark 4, it holds that \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}=x_{{\Gamma }}\) so that \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}=\mathcal {C}_{{\Gamma }}\).
In case 2) it holds that \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}} =x_{{\Gamma } \cup \{n + 1\}}\), where \(x_{{\Gamma }\cup \{n + 1\}}=\mathbb {P}(\mathcal {C}_{{\Gamma }\cup \{n + 1\}})\). Now, we observe that the random quantities \(\mathcal {C}_{{\Gamma }}\) and \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}\) coincide conditionally on \(\bigvee _{i\in {\Gamma }\cup \{n + 1\}} H_{i}\) being true; then by Theorem 3 it holds that \(\mathbb {P}(\mathcal {C}_{{\Gamma }})=\mathbb {P}(\mathcal {C}_{{\Gamma }\cup \{n + 1\}})\), that is \(x_{{\Gamma }}=x_{{\Gamma }\cup \{n + 1\}}\); thus \(\mathcal {C}_{{\Gamma }\cup \{n + 1\}}=\mathcal {C}_{{\Gamma }}\).
Finally, denoting by Γc the set \(\{1,\ldots ,n\}\setminus {\Gamma }\), by the associative property of conjunction we obtain
$$\begin{array}{@{}rcl@{}} \mathcal{C}_{n + 1}&=&\mathcal{C}_{n}\wedge E_{n + 1}|H_{n + 1}=\mathcal{C}_{{\Gamma}^{c}}\wedge \mathcal{C}_{{\Gamma}} \wedge E_{n + 1}|H_{n + 1}\\ &=&\mathcal{C}_{{\Gamma}^{c}}\wedge \mathcal{C}_{{\Gamma}}=\mathcal{C}_{n}. \end{array} $$ -
b)
\((ii) \Rightarrow (iii)\). By monotonicity property of conjunction it holds that \(\mathcal {C}_{n + 1}\leq E_{n + 1}|H_{n + 1}\). Then, by assuming \(\mathcal {C}_{n}=\mathcal {C}_{n + 1}\), it follows \(\mathcal {C}_{n}\leq E_{n + 1}|H_{n + 1}\).
-
c)
\((iii) \Rightarrow (i)\). Let us assume that \(\mathcal {C}_{n}\leq E_{n + 1}|H_{n + 1}\), so that \(\mathbb {P}(\mathcal {C}_{n})\leq P(E_{n + 1}|H_{n + 1})\). Moreover, by assuming that \(P(E_{i}|H_{i})= 1\), \(i = 1,\ldots , n\), from (15) it follows \(\mathbb {P}(\mathcal {C}_{n})= 1\) and hence \(P(E_{n + 1}|H_{n + 1})= 1\), that is \(\mathcal {F}\) p-entails \(E_{n + 1}|H_{n + 1}\).
□
Rights and permissions
About this article
Cite this article
Gilio, A., Sanfilippo, G. Generalized logical operations among conditional events. Appl Intell 49, 79–102 (2019). https://doi.org/10.1007/s10489-018-1229-8
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10489-018-1229-8