Abstract
We develop joint foundations for the fields of social choice and opinion pooling using coherent sets of desirable gambles, a general uncertainty model that allows to encompass both complete and incomplete preferences. This leads on the one hand to a new perspective of traditional results of social choice (in particular Arrow’s theorem as well as sufficient conditions for the existence of an oligarchy and democracy) and on the other hand to using the same framework to analyse opinion pooling. In particular, we argue that weak Pareto (unanimity) should be given the status of a rationality requirement and use this to discuss the aggregation of experts’ opinions based on probability and (stateindependent) utility, showing some inherent limitation of this framework, with implications for statistics. The connection between our results and earlier work in the literature is also discussed.
References
Aczél, J., Wagner, C.: A characterization of weighted arithmetic means. SIAM J. Algebraic Discret. Methods 1(3), 259–260 (1980)
Arrow, K.: A difficulty in the concept of social welfare. J. Polit. Econ. 58(4), 328–346 (1950)
Arrow, K.: Social Choice and Individual Values. Wiley (1951)
Arrow, K.J., Sen, A., Suzumura, K.: Handbook of Social Choice and Welfare, vol. 2. Elsevier (2010)
Augustin, T., Coolen, F., de Cooman, G., Troffaes, M. (eds.): Introduction to Imprecise Probabilities. Wiley, New York (2014)
Barthelemy, J.P.: Arrow’s theorem: unusual domains and extended codomains. Math. Soc. Sci. 3, 79–89 (1982)
Benavoli, A., Facchini, A., Zaffalon, M.: Bernstein’s socks, polynomialtime provable coherence and entaglement. In: Proceedings of ISIPTA’2019, volume 103 of Proceedings of Machine Learning Research, pp. 23–31 (2019)
Benavoli, A., Facchini, A., Zaffalon, M.: Computational complexity and the nature of quantum mechanics. CoRR, arXiv:1902.03513 (2019)
Chevaleyre, Y., Endriss, U., Lang, J., Maudet, M.: A short introduction to computational social choice. In: van Leeuwen, J., Italiano, G.F., van der Hoek, W., Meinel, C., Sack, H., Plasil, F. (eds.) Proceedings of SOFSEM 2007, 33rd Conference on Current Trends in Theory and Practice of Computer Science, volume 4362 of Lecture Notes in Computer Science, pp 51–69. Springer (2007)
Cooke, R.M., Bedford, T.: Expert Opinion. Cambridge Program for Industry (1996)
de Cooman, G.: Belief models: an ordertheoretic investigation. Ann. Math. Artif. Intell. 45, 5–34 (2005)
de Cooman, G., Quaeghebeur, E.: Exchangeability and sets of desirable gambles. Int. J. Approx. Reason. 53, 363–395. Special issue in honour of Henry E. Kyburg Jr (2012)
Dietrich, F., List, C.: Judgement aggregation with consistency alone. Technical Report METEOR Research Memorandum No. 021, METEOR, Maastricht University School of Business and Economics. https://cris.maastrichtuniversity.nl/ws/portalfiles/portal/1415140/guid3b81b5a2eeb745958087e2f1e7df5e85ASSET1.0.pdf (2007)
Dietrich, F., List, C.: Judgment aggregation without full rationality. Soc. Choice Welf. 31, 15–39 (2008)
Dietrich, F., List, C.: Probabilistic opinion pooling generalized. Part one: General agendas. Soc. Choice Welf. 48(4), 747–786 (2017)
Dietrich, F., List, C.: Probabilistic opinion pooling generalized. Part two: The premisebased approach. Soc. Choice Welf. 48(4), 787–814 (2017)
Dubois, D., Fargier, H., Perny, P.: On the limitations of ordinal approaches to decisionmaking. In: Fensel, D., Giunchiglia, F., McGuinness, D.L., Williams, M.A. (eds.) Proceedings of the Eights International Conference on Principles and Knowledge Representation and Reasoning (KR02), Toulouse, pp. 133–146 (2002)
Farber, D.: The problematics of the Pareto principle. SSRN Electronic Journal (2003)
Feldman, A.M., Serrano, R.: Welfare Economics and Social Choice Theory. Springer Science & Business Media (2006)
Fishburn, P.: Impossibility theorems without the social completeness axiom. Econometrica 42, 695–704 (1974)
Gärdenfors, P.: A representation theorem for voting with logical consequences. Econ. Philos. 22(2), 181–190 (2006)
Gibbard, A.: Social choice and the Arrow conditions. Technical report, Harvard University. http://wwwpersonal.umich.edu/~gibbard/Gibbard1968SocialChoiceArrowConditions.pdf (1968)
Goddin, R.E., List, C.: Special majorities rationalized. Br. J. Polit. Sci. 36(2), 213–241 (2006)
Goodman, J.: Existence of compromises in simple group decisions. PhD thesis (1988)
Jain, R.: A note on the Arrow’s impossibility theorem. Econ. Ann. 60, 39–48 (2015)
Kelly, J.S.: Arrow impossibility theorems. Academic Press (1978)
Kelly, J.S.: Social Choice Theory: An Introduction. Springer (1988)
Kyburg, H.E., Pittarelli, M.: Some problems for convex Bayesians. In: Dubois, D., Wellman, M. P. (eds.) Uncertainty in Artificial Intelligence: Proceedings of the Eighth conference (UAI1992), pp 149–154. Morgan Kaufmann Publishers (1992)
Lehrer, K., Wagner, C.: Probability amalgamation and the independence issue: A reply to Laddaga. Synthese 55(3), 339–346 (1983)
Lindley, D.V., Tversky, A., Brown, R.V.: On the reconciliation of probability assessments. J. R. Stat. Soc. Ser. A 142(2), 146–162 (1979)
List, C., Pettit, P.: Aggregating sets of judgments An impossibility result. Econ. Philos. 18(1), 89–110 (2002)
MasColell, A., Sonnenschein, H.: General possibility theorems for group decisions. Rev. Econ. Stud. 39, 165–192 (1972)
McConway, K.J.: Marginalization and linear opinion pools. J. Am. Stat. Assoc. 76(374), 410–414 (1981)
Miranda, E., Zaffalon, M.: Notes on desirability and conditional lower previsions. Ann. Math. Artif. Intell. 60(3–4), 251–309 (2010)
Miranda, E., Zaffalon, M., de Cooman, G.: Conglomerable natural extension. Int. J. Approx. Reason. 53(8), 1200–1227 (2012)
Mongin, P.: Spurious unanimity and the Pareto principle. Technical Report 1(5), London School of Economics, The Centre for Philosophy of Natural and Social Science (CPNSS) (2005)
Moral, S., Del Sagrado, J.: Aggregation of imprecise probabilities. In: Aggregation and Fusion of Imperfect Information, pp. 162–188. Springer (1998)
Nash, J.F.: The bargaining problem. Econometrica 18, 155–162 (1950)
Nau, R.: The aggregation of imprecise probabilities. J. Stat. Plann. Infer. 105, 265–282 (2002)
Nehring, K., Puppe, C.: Strategyproof social choice on singlepeakeddomains Possibility, impossibility and the space between. Technical report University of California at Davies (2002)
Ng, K.C., Abramson, B.: Consensus diagnosis: a simulation study. IEEE Trans. Syst. Man Cybern. 26(3), 324–339 (1996)
Nguyen, H.T., Kosheleva, O., Kreinovich, V.: Decision making beyond arrow’s impossibility theorem, with the analysis of effects of collusion and mutual attraction. Int. J. Intell. Syst. 24, 27–47 (2009)
Pini, M.S., Rossi, F., Venable, K.B., Walsh, T.: Aggregating partially ordered preferences. J. Log. Comput. 19(3), 475–502 (2009)
Raiffa, H.: Decision Analysis. AddisonWesley (1968)
Rossi, F., Venable, K.B., Walsh, T.: A Short Introduction to Preferences: Between Artificial Intelligence and Social Choice. Synthesis Lectures on Artificial Intelligence and Machine Learning. Morgan & Claypool Publishers (2011)
Seidenfeld, T., Kadane, J.B., Schervish, M.J.: On the shared preferences of two Bayesian decision makers. J. Philos. 86(5), 225–244 (1989)
Sen, A.: Social choice theory. In: Arrow, K.J., Intriligator, M.D. (eds.) Handbook of Mathematical Economics, volume III, chapter, vol. 22, pp 1073–1181 (1986)
Stewart, R., Quintana, I.O.: Probabilistic opinion pooling with imprecise probabilities. J. Philos. Log. 47(1), 17–45 (2018)
Stone, M.: The opinion pool. Ann. Math. Stat. 32, 1339–1342 (1961)
Tideman, T.N.: Independence of clones as a criterion for voting rules. Soc. Choice Welf. 4(3), 185–206 (1987)
Van Camp, A.: Choice Functions as a Tool to Model Uncertainty. PhD thesis, University of Ghent (2018)
Van Camp, A., de Cooman, G., Miranda, E., Quaeghebeur, E.: Coherent choice functions, desirability and indifference. Fuzzy Set. Syst. 341(C), 1–36 (2018)
Walley, P.: The elicitation and aggregation of beliefs. Technical report, University of Warwick, 1982. Statistics Research Report 23
Walley, P.: Statistical Reasoning with Imprecise Probabilities. Chapman and Hall, London (1991)
Walley, P.: Towards a unified theory of imprecise probability. Int. J. Approx. Reason. 24, 125–148 (2000)
Weymark, J.A.: Arrow’s theorem with social quasiorderings. Publ. Choice 42(3), 235–246 (1984)
Williams, P.M.: Notes on conditional previsions. Technical report, School of Mathematical and Physical Science, University of Sussex, UK, 1975. Reprinted in [58]
Williams, P.M.: Notes on conditional previsions. Int. J. Approx. Reason. 44, 366–383. Revised journal version of [57] (2007)
Zaffalon, M., Miranda, E.: Probability and time. Artif. Intell. 198(1), 1–51 (2013)
Zaffalon, M., Miranda, E.: Axiomatising incomplete preferences through sets of desirable gambles. J. Artif. Intell. Res. 60, 1057–1126 (2017)
Zaffalon, M., Miranda, E.: Desirability foundations of robust rational decision making. Synthese (2018)
Acknowledgements
We acknowledge the financial support of projects PGC2018098623BI00. We would also like to thank Seamus Bradley, Teddy Seidenfeld, Franz Dietrich and the anonymous reviewers for stimulating discussion as well as for providing some relevant references.
Funding
Open Access funding provided by Università della Svizzera italiana.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Representation of a social rule Γ as a bundle of functions
A social rule Γ can be represented as a bundle of functions \(\{{\Gamma }_{f}: f \in {\mathscr{L}}\}\), where Γ_{f} identifies those profiles for which the gamble f belongs to the aggregated set
The fact that the social set of desirable gambles is coherent imposes some minimal requirements on these functions:

G1.
\({\Gamma }_{{\mathscr{L}}^{+}}=1\).

G2.
Γ_{0} = 0.

G3.
\(f,g\in {\mathscr{L}}\Rightarrow {\Gamma }_{f+g}\geq {\Gamma }_{f} \cdot {\Gamma }_{g}\).

G4.
\(f\in {\mathscr{L}},\lambda >0\Rightarrow {\Gamma }_{f}={\Gamma }_{\lambda f}\).
We can reformulate additional properties that a social rule may satisfy using this equivalent representation. In particular:

◦ Completeness is equivalent to:
$$ f\neq 0\Rightarrow {\Gamma}_{f} + {\Gamma}_{f}=1, $$and strict completeness can be expressed as
$$ \epsilon>0\Rightarrow{\Gamma}_{f}+{\Gamma}_{f+\epsilon}\geq 1. $$ 
◦ Independence of irrelevant alternatives. Let us define the function
$$ \begin{array}{@{}rcl@{}} \pi_{f}:&\mathbb{D}^{n}\rightarrow \{0,1\}^{n} \\ &[\mathcal{D}_{i}] \hookrightarrow \vec{z}, \end{array} $$where
$$\vec{z}_{i}:=\begin{cases} 1 &\text{ if } f \in\mathcal{D}_{i} \\ 0 &\text{ otherwise}. \end{cases}$$In other words, \(\pi _{f}([\mathcal {D}_{i}])=(\mathbb {I}_{\mathcal {D}_{1}}(f),\dots ,\mathbb {I}_{\mathcal {D}_{n}}(f))\). Then independence of irrelevant alternatives can be equivalently expressed as
$$ (\forall f\in\mathcal{L})\ \pi_{f}([\mathcal{D}_{i}])=\pi_{f}([\mathcal{D}^{\prime}_{i}]) \Rightarrow {\Gamma}_{f}([\mathcal{D}_{i}])={\Gamma}_{f}([\mathcal{D}^{\prime}_{i}]). $$ 
◦ Weak Pareto is equivalent to
$$ (\forall f\in\mathcal{L})\ \pi_{f}^{1}(\{\vec{1}\})\subseteq {\Gamma}_{f}^{1}(\{1\}). $$ 
◦ Given a set of individuals \(\mathcal {G}\), it is almost decisive for f if and only if
$$ \pi_{f}^{1}(\vec{1}_{\mathcal{G}},\vec{0}_{\mathcal{H}\setminus\mathcal{G}})\subseteq {\Gamma}_{f}^{1}(\{1\}), $$where \(\vec {1}_{\mathcal {G}},\vec {0}_{{\mathscr{H}}\setminus \mathcal {G}}\) is the vector \(\vec {z}\) given by
$$ z_{i}:=\begin{cases} 1 &\text{ if } i\in\mathcal{G} \\ 0 &\text{ if } i\notin\mathcal{G}, \end{cases} $$whereas \(\mathcal {G}\) is decisive for a gamble f if and only if
$$ \pi_{f}^{1}(\vec{1}_{\mathcal{G}})\subseteq {\Gamma}_{f}^{1}(\{1\}). $$
Appendix B: Results in terms of lower previsions and sets of finitely additive probabilities
In this paper, we have assumed that the beliefs of each voter are represented by means of a coherent set of desirable gambles \(\mathcal {D}_{i}\), and that the outcome of the social rule Γ on a profile \([\mathcal {D}_{i}]\) is also a coherent set of desirable gambles. The main motivation behind this choice lies in the flexibility and generality of sets of desirable gambles, that can be linked to preference relations and also encompass as particular cases coherent lower previsions or sets of probability measures.
For completeness, in this appendix we translate the results we have established in the first part of the paper to social rules defined on coherent lower previsions \({\underline {P}}\) or on closed convex sets of probability measures \({\mathscr{M}}\). We shall make use of the onetoone correspondence between these models and coherent sets of strictly desirable gambles: in the case of a coherent lower prevision \(\underline {P}\), its associated set is given by (3); given a closed convex set of probability measures \({\mathscr{M}}\), we consider the set of strictly desirable gambles associated with its lower prevision \(\underline {P}:=\min \limits {\mathscr{M}}\).
As a consequence, we shall say for instance that a social rule Γ on coherent lower previsions satisfies weak Pareto if and only if the social rule \({\Gamma }^{\prime }\) that we can determine on the associated sets of strictly desirable gambles by
satisfies weak Pareto. Similar considerations hold for the other axioms, and for social rules defined on credal sets.
Our next result summarises how the different axioms and government systems are represented in this context:
Theorem 6
Let Γ be a social rule defined on a profile \([{\underline {P}}_{i}]\) (resp., \([{\mathscr{M}}_{i}]\)) given by coherent lower previsions (resp., credal sets). Then:

1.
Γ satisfies weak Pareto \(\Leftrightarrow (\forall [{\underline {P}}_{i}] \in \mathcal {A}) {\underline {P}} \geq \min \limits _{i} {\underline {P}}_{i}, \) \(\Leftrightarrow (\forall [{\mathscr{M}}_{i}] \in \mathcal {A}) {\mathscr{M}} \subseteq \overline {\text {ch}(\cup _{i} {\mathscr{M}}_{i})}\).

2.
Γ satisfies strict completeness \(\Leftrightarrow (\forall [{\underline {P}}_{i}] \in \mathcal {A}) {\underline {P}}\) linear \(\Leftrightarrow (\forall [{\mathscr{M}}_{i}] \in \mathcal {A}) {\mathscr{M}}=1\).

3.
A set \(\mathcal {G} \subseteq {\mathscr{H}} \) is decisive \(\Leftrightarrow (\forall [{\underline {P}}_{i}] \in \mathcal {A}) {\underline {P}}\geq \min \limits _{i\in \mathcal {G}}{\underline {P}}_{i}\) \(\Leftrightarrow (\forall [{\mathscr{M}}_{i}] \in \mathcal {A}) {\mathscr{M}} \subseteq \overline {\text {ch}(\cup _{i \in \mathcal {G}} {\mathscr{M}}_{i})}\).

4.
A set \(\mathcal {G} \subseteq {\mathscr{H}} \) is an oligarchy \(\Leftrightarrow (\forall [{\underline {P}}_{i}] \in \mathcal {A}) {\underline {P}}\geq \min \limits _{i\in \mathcal {G}}{\underline {P}}_{i}\) and \({\overline {P}}\geq \max \limits _{i\in \mathcal {G}}P_{i}\).

5.
An individual \(j \in {\mathscr{H}}\) is a dictator \(\Leftrightarrow (\forall [{\underline {P}}_{i}] \in \mathcal {A}) {\underline {P}}\geq {\underline {P}}_{j}\) \(\Leftrightarrow {\mathscr{M}} \subseteq {\mathscr{M}}_{j}\).

6.
Γ satisfies anonymity \(\Leftrightarrow (\forall [{\underline {P}}_{i}],[{\underline {P}}_{\sigma (i)}] \in \mathcal {A}) {\underline {P}}={\underline {P}}_{\sigma } \Leftrightarrow (\forall [{\mathscr{M}}_{i}], [{\mathscr{M}}_{\sigma (i)}] \in \mathcal {A}) {\mathscr{M}}= {\mathscr{M}}_{\sigma }\), where σ is a permutation of \({\mathscr{H}}\) and P_{σ} (resp. \([{\mathscr{M}}_{\sigma }])\) is the coherent lower prevision (resp. credal set) given by Γ([P_{σ(i)}]) (resp. \({\Gamma }([{\mathscr{M}}_{\sigma (i)}])\)).

7.
Γ satisfies independence of irrelevant alternatives \(\Leftrightarrow (\forall [{\underline {P}}_{i}],[{\underline {P}}^{\prime }_{i}] \in \mathcal {A}) (\forall f \notin {\mathscr{L}}^{+})\):
$$ (\forall i \in \mathcal{H}) {\underline{P}}_{i}(f){\underline{P}}^{\prime}_{i}(f)>0 \Rightarrow {\underline{P}}(f){\underline{P}}^{\prime}(f)>0, $$where \({\underline {P}}^{\prime }\) denotes the coherent lower prevision that is obtained by means of \({\Gamma }([{\underline {P}}^{\prime }_{i}])\).
More generally speaking, if we consider a social rule \({\Gamma }^{\prime }\) defined on coherent sets of desirable gambles, we wonder if similar equivalences hold with respect to the social rule Γ on coherent lower previsions or credal sets that is associated with it. However, this will not be the case in general. To see just one example, consider the social rule \({\Gamma }^{\prime }\) given by \({\Gamma }^{\prime }([\mathcal {D}_{i}])=\mathcal {D}_{j\triangleright }\), for some fixed \(j\in {\mathscr{H}}\). Then \({\Gamma }^{\prime }\) will not satisfy weak Pareto in general (for instance with \(\mathcal {D}_{i}=\mathcal {D}_{j}\) for every i, and the latter being non strictly desirable), but its associated Γ will be given by \({\Gamma }([\underline {P}_{i}])=\underline {P}_{j}\geq \min \limits _{i} \underline {P}_{i}\).
As a corollary, which follows immediately from Theorem 6 and (2), we can characterise the above conditions in the particular case where the profile is precise:
Corollary 3
Let Γ be a strict complete social rule defined on a profile [P_{i}] given by linear previsions. Then:

1.
A set \(\mathcal {G}\subseteq {\mathscr{H}} \) is decisive \(\Leftrightarrow (\forall [P_{i}] \in \mathcal {A}) {\Gamma }([P_{i}]) \in \text {ch}(\{P_{i}\mid i\in \mathcal {G}\})\).

2.
A set \(\mathcal {G}\subseteq {\mathscr{H}} \) is an oligarchy \(\Leftrightarrow (\forall [P_{i}] \in \mathcal {A}), P_{j}=P_{j^{\prime }} \ \forall j,j^{\prime }\in \mathcal {G} \text { and } {\Gamma }([P_{i}])=P_{j}\).

3.
An individual \(j \in {\mathscr{H}}\) is a dictator \(\Leftrightarrow (\forall [P_{i}] \in \mathcal {A}) {\Gamma }([P_{i}])=P_{j}\).
We see then that the notion of oligarchy does not make much sense in the precise case, because it forces all the individuals that form the oligarchy to have exactly the same opinion.
Appendix C: Proofs
In the following proofs, it will be convenient at times to reformulate the condition of avoiding partial loss in the following manner:
Lemma 5
Given a set \(\mathcal {K}\) of desirable gambles, \(\text {ch}(\mathcal {K}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \Leftrightarrow 0 \notin \text {posi}(\mathcal {K} \cup {\mathscr{L}}^{+}).\) As a consequence, if any of the two conditions above hold, it follows that \(\mathcal {K}\) has a coherent (maximal) superset.
Proof
It is equivalent to prove that
To prove that this is the case, note that
and since the lefthand side belongs to \({\mathscr{L}}^{}\cup \{0\}\) and the righthand side belongs to \(\text {ch}(\mathcal {K})\), we deduce that this is equivalent to \(\text {ch}(\mathcal {K}) \cap ({\mathscr{L}}^{} \cup \{0\})\neq \emptyset \).
For the second statement, if any of these conditions hold we deduce that the set \(\text {posi}(\mathcal {K}\cup {\mathscr{L}}^{+})\) is a coherent set of desirable gambles that trivially includes \(\mathcal {K}\). Since any coherent set of desirable gambles has a coherent maximal superset [12, Corollary 4], it follows in particular that \(\mathcal {K}\) has a coherent maximal superset. □
In order to prove Lemmas 1 and 2, we also need first to establish the following auxiliary result:
Lemma 6
Assume that \(\mathcal {Z}\geq 3\). Then for every gamble f such that \(f \notin ({\mathscr{L}}^{+} \cup {\mathscr{L}}^{} \cup \{0\})\) there always exist two gambles {h, t} such that − f = f + h + t and

◦ \(\text {ch}(\{f,h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \), \(\text {ch}(\{f,h,f+h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) and \(\text {ch}(\{f,h,(f+h)\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \);

◦ \(\text {ch}(\{f+h, t\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \), \(\text {ch}(\{(f+h),t, f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) and \(\text {ch}(\{(f+h),t,f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \).
Proof
Since f is such that \(f \notin ({\mathscr{L}}^{+} \cup {\mathscr{L}}^{} \cup \{0\})\) then there exist at least two points \(z_{1},z_{2} \in \mathcal {Z}\) with z_{1} ≠ z_{2} such that f(z_{1}) = f_{1} > 0 and f(z_{2}) = f_{2} < 0. We consider a number of possibilities:

∘
If there is some z_{3} ≠ z_{1}, z_{2} such that f(z_{3}) ≥ 0, then we define, for a fixed 𝜖 > 0:
h(z) := { f_{1} > 0 if z = z_{1} −f_{2}/2 > 0 if z = z_{2} − 2f(z) − 𝜖 otherwise and t(z) := { − 3f_{1} < 0 if z = z_{1} − 3f_{2}/2 > 0 if z = z_{2} 𝜖 otherwise.
Then by construction − f = f + h + t. To see that they fulfil the conditions in the lemma, note that:

◦ \(\text {ch}(\{f,h\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \) (use z_{1});

◦ \(\text {ch}(\{f,h,f+h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \): if 0 ≥ λ_{−f}(−f) + λ_{h}h + λ_{f+h}(f + h), then it should be on the one hand λ_{−f} ≥ λ_{h} + 2λ_{f+h} because of z_{1}, and on the other 2λ_{−f} + λ_{h} ≤ λ_{f+h} because of z_{2}; these two conditions are incompatible.

◦ \(\text {ch}(\{f,h,(f+h)\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{2});
and also

◦ \(\text {ch}(\{f+h,t\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \): if λ_{f+h}(f + h) + λ_{t}t ≤ 0, it should be on the one hand λ_{f+h} ≤ 0.6 because of z_{1} and on the other λ_{f+h} ≥ 0.75 because of z_{2};

◦ \(\text {ch}(\{(f+h),t, f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{2});

◦ \(\text {ch}(\{(f+h),t,f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{3} and z_{1}).


∘
If f(z_{3}) < 0 for every z_{3} ≠ z_{1}, z_{2}, then we define:
h(z) := { −f_{1} < 0 if z = z_{1} −f_{2}/2 > 0 if z = z_{2} − 2f(z) otherwise and t(z) := { −f_{1} < 0 if z = z_{1} − 3f_{2}/2 > 0 if z = z_{2} 0 otherwise.
Then by construction − f = f + h + t. To see that they fulfil the conditions in the lemma, note that:

◦ \(\text {ch}(\{f,h\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \): if λ_{f}(f) + λ_{h}h ≤ 0, it should be on the one hand λ_{f} ≤ 0.5 because of z_{1} and on the other \(\lambda _{f}\geq \frac {2}{3}\) because of z_{3};

◦ \(\text {ch}(\{f,h,f+h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{3});

◦ \(\text {ch}(\{f,h,(f+h)\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{2});
and also

◦ \(\text {ch}(\{f+h,t\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \) (use z_{2} and z_{3});

◦ \(\text {ch}(\{(f+h),t, f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{2});

◦ \(\text {ch}(\{(f+h),t,f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \): if 0 ≥−λ_{−(f+h)}(f + h) + λ_{t}t + λ_{f}f, then it should be λ_{t} ≥ λ_{f} so that this combination is nonpositive on z_{1}, but then this implies that it is positive on z_{2}.

□
Proof of Lemma 1
Note that we can assume without loss of generality that \(\mathcal {G}\) is a proper subset of \({\mathscr{H}}\), since otherwise the thesis follows immediately from the property of weak Pareto. In addition, we can then assume that the gamble f does not belong to \({\mathscr{L}}^{+}\cup {\mathscr{L}}^{}\cup \{0\}\).
We must prove that \((\forall [\mathcal {D}_{i}])\ \cap _{i \in \mathcal {G}} \mathcal {D}_{i} \subseteq {\Gamma }([\mathcal {D}_{i}])\). Consider thus a profile \([\mathcal {D}_{i}]\) (eventually in \(\hat {\mathbb {D}}^{n}\)), and let \(g\in \cap _{i\in \mathcal {G}}\mathcal {D}_{i}\). We can assume without loss of generality that \(g\notin {\mathscr{L}}^{+}\), since in that case it trivially belongs to \({\Gamma }([\mathcal {D}_{i}])\). There are a number of possibilities:

1.
Assume first of all that g = f + h, where:
$$ \begin{array}{@{}rcl@{}} & \text{ch}(\{f,h\}) \cap (\mathcal{L}^{} \cup\{0\})= \emptyset, \end{array} $$(C.1)$$ \begin{array}{@{}rcl@{}} & \text{ch}(\{f,h,g\}) \cap (\mathcal{L}^{} \cup\{0\})= \emptyset, \end{array} $$(C.2)$$ \begin{array}{@{}rcl@{}} & \text{ch}(\{f,h,g\}) \cap (\mathcal{L}^{} \cup\{0\})= \emptyset. \end{array} $$(C.3)Hence, it holds that:

◦ By Lemma 5, if \(\text {ch}(\{f,h\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \), there exists a coherent set \(\mathcal {D}^{1}\) (possibly maximal) of desirable gambles that includes both f, h, and as a consequence also g by additivity.

◦ Since \(\text {ch}(\{f,h,g\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \), then the coherent set \(\mathcal {D}^{2}:=\text {posi}(\{f,h,g\}\cup {\mathscr{L}}^{+})\) (or a maximal set that contains it) of desirable gambles includes h and not f and g.

◦ Finally, let \(\mathcal {D}^{3}:=\text {posi}(\{f,h,g\}\cup {\mathscr{L}}^{+})\) (or a maximal set that contains it). It includes h, g but not f.
Let us now define the following profile \([\mathcal {D}^{\prime }_{i}]\):

◦ For every \(i\in \mathcal {G}\), let \(\mathcal {D}^{\prime }_{i}=\mathcal {D}^{1}\) (possibly the maximal one). Then \(f,h\in \mathcal {D}^{\prime }_{i}\), whence also \(g\in \mathcal {D}^{\prime }_{i}\) by additivity.

◦ Given \(i\in {\mathscr{H}}\setminus \mathcal {G}\), if \(g\notin \mathcal {D}_{i}\) then we let \(\mathcal {D}^{\prime }_{i}=\mathcal {D}^{2}\) (possibly the maximal one). Then \(h\in \mathcal {D}^{\prime }_{i}\) and \(f,g\notin \mathcal {D}^{\prime }_{i}\).

◦ Given \(i\in {\mathscr{H}}\setminus \mathcal {G}\), if \(g\in \mathcal {D}_{i}\) then we let \(\mathcal {D}^{\prime }_{i}=\mathcal {D}^{3}\) (possibly the maximal one). Then \(h,g\in \mathcal {D}^{\prime }_{i}\) and \(f\notin \mathcal {D}^{\prime }_{i}\).
It follows then that \(h\in \cap _{i}\mathcal {D}^{\prime }_{i}\), whence \(h\in {\Gamma }([\mathcal {D}^{\prime }_{i}])\), using the weak Pareto property; moreover, \(f\in \cap _{i\in \mathcal {G}}\mathcal {D}^{\prime }_{i}\) and \(f\notin \cup _{i\in {\mathscr{H}}\setminus \mathcal {G}}\mathcal {D}^{\prime }_{i}\). Since \(\mathcal {G}\) is almost decisive for f, we deduce that \(f\in {\Gamma }([\mathcal {D}^{\prime }_{i}])\). Since the latter is a coherent set of desirable gambles, we deduce by additivity that \(g=f+h\in {\Gamma }([\mathcal {D}^{\prime }_{i}])\). And since \(\pi _{g}([\mathcal {D}_{i}])=\pi _{g}([\mathcal {D}^{\prime }_{i}])\) by construction, we deduce from the independence of irrelevant alternatives that also \(g\in {\Gamma }([\mathcal {D}_{i}])\).
Now note that the reasoning above can be applied in particular for those \(g\gneq f\) (i.e., when \(h\in {\mathscr{L}}^{+}\)). To see that this is the case, note that when \(h\in {\mathscr{L}}^{+}\):

(C.1) \(\text {ch}(\{f,h\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \) is equivalent to {f} avoiding partial loss, which holds because \(f\notin {\mathscr{L}}^{+}\cup {\mathscr{L}}^{}\cup \{0\}\) by assumption.

(C.2) \(\text {ch}(\{f,h,g\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \) becomes equivalent to \(\text {ch}(\{f,g\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \). But if this intersection was nonempty we would deduce that for some λ_{−f} + λ_{g} = 1 it holds that 0 ≥ λ_{−f}(−f) + λ_{g}g ≥ (λ_{g} − λ_{−f})f, which contradicts the assumption that \(f\notin {\mathscr{L}}^{+}\cup {\mathscr{L}}^{}\cup \{0\}\).

(C.3) \(\text {ch}(\{f,h,g\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \) becomes equivalent to {−g} avoiding partial loss, which holds because it cannot be \(g\in {\mathscr{L}}^{}\cup \{0\}\) since we are assuming that \(g\in \cap _{i\in \mathcal {G}} \mathcal {D}_{i}\).
This allows us to deduce that
$$ \mathcal{G} \text{ is decisive for any } g\gneq f. $$(C.4)Next we establish that:
$$ \mathcal{G} \text{ almost decisive for } g \Rightarrow \mathcal{G} \text{ decisive for } g. $$(C.5)To prove this, note that we can assume without loss of generality that \(g\notin {\mathscr{L}}^{+}\cup {\mathscr{L}}^{}\cup \{0\}\). Applying Lemma 6, we can find two gambles h, t such that − g = g + h + t. Moreover, by construction in the proof of that lemma we have that \(\text {ch}(\{g,h\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \), \(\text {ch}(\{g,h, (g+h)\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \) and \(\text {ch}(\{g,h, (g+h)\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \). Applying point 1, we deduce that \(\mathcal {G}\) is decisive for g + h, and therefore also almost decisive. Moreover, we also have from the proof of the lemma that \(\text {ch}(\{g+h,t\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \), \(\text {ch}(\{(g+h),t, g\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \) and \(\text {ch}(\{(g+h),t, g\})\cap ({\mathscr{L}}^{}\cup \{0\})=\emptyset \). Applying again point 1, we deduce that \(\mathcal {G}\) is decisive for g + h + t = −g.
Thus, (C.5) holds. Applying twice this condition we deduce in particular that \(\mathcal {G}\) is decisive for − f, and also for f.


2.
Assume now that g = f + h, with \(h\in {\mathscr{L}}^{}\). Then \(g\lneq f\), or, equivalently \(f\lneq g\). Since \(\mathcal {G}\) is decisive at − f, it is in particular almost decisive at this gamble, whence by (C.4) \(\mathcal {G}\) is also almost decisive at \(g\gneq f\). Applying now (C.5), we deduce that \(\mathcal {G}\) is also decisive at g.

3.
Finally, consider a gamble \(h\notin {\mathscr{L}}^{+}\cup {\mathscr{L}}^{}\cup \{0\}\), and let g := f + h. Then we can rewrite g as g = f − h^{−} + h^{+}, where h^{+} and h^{−} are respectively the positive and the negative part function of h, i.e. \(h^{+}:=\max \limits \{0, h\}\) and \(h^{}:=\min \limits \{0, h\}\). It then follows from point 2 that \(\mathcal {G}\) is decisive on g = f − h^{−} because \(h^{}\in {\mathscr{L}}^{}\), and as a consequence it is also almost decisive on this gamble. If we now apply (C.4) we deduce that \(\mathcal {G}\) is decisive on \(g=(fh^{} )+ h^{+}\gneq fh^{}\).
□
Proof of Lemma 2
Let us prove that there is a proper subset \(\mathcal {G}^{\prime }\) of \(\mathcal {G}\) that is also decisive. For this, we consider a partition of \(\mathcal {G}\) into nonempty and disjoint subsets \(\mathcal {G}_{1}\) and \(\mathcal {G}_{2}\), and we proceed to establish that one of these two sets is also decisive.
Consider \(z_{1},z_{2}\in \mathcal {Z}\) with z_{1} ≠ z_{2}, and let us define the gambles f_{1}, f_{2}, f_{3} by:
Let z_{3} denote an element different from z_{1}, z_{2}, existing because \(\mathcal {Z}\geq 3\). These gambles satisfy the following conditions:

◦ \(\text {ch}(\{f_{1},f_{2},f_{3}\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \) (use z_{3} and z_{2});

◦ \(\text {ch}(\{f_{1},f_{2}, f_{3}\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \) (use z_{1} and z_{3});

◦ \(\text {ch}(\{f_{1},f_{3}\}) \cap ({\mathscr{L}}^{} \cup \{0\})= \emptyset \) (use z_{2}).
Then, thanks to Lemma 5 we can consider the following profile \([\hat {\mathcal {D}}_{i}] \in \hat {\mathbb {D}}^{n}\):

◦ for all \(i\in \mathcal {G}_{1}, \hat {\mathcal {D}}_{i}:= M_{1} \supseteq \text {posi}(\{f_{1},f_{2},f_{3}\}\cup {\mathscr{L}}^{+})\), with \(M_{1} \in \hat {\mathbb {D}}\);

◦ for all \(i\in \mathcal {G}_{2}, \hat {\mathcal {D}}_{i}:= M_{2} \supseteq \text {posi}(\{f_{1},f_{2},f_{3}\}\cup {\mathscr{L}}^{+})\), with \(M_{2} \in \hat {\mathbb {D}}\);

◦ for all \(i\notin \mathcal {G}, \hat {\mathcal {D}}_{i}:= M_{3} \supseteq \text {posi}(\{f_{1},f_{3}\}\cup {\mathscr{L}}^{+})\), with \(M_{3} \in \hat {\mathbb {D}}\).
Note then that \(f_{1} \notin \hat {\mathcal {D}_{i}}\) for any \(i\notin \mathcal {G}_{2}\), and that \(f_{3}\notin \hat {\mathcal {D}_{i}}\) for any \(i\notin \mathcal {G}_{1}\), because these sets of gambles are maximal. Since \(f_{2}\in \cap _{i\in \mathcal {G}} \hat {\mathcal {D}}_{i}\), it follows from decisiveness of \(\mathcal {G}\) that \(f_{2}\in {\Gamma }([\hat {\mathcal {D}}_{i}])\). Since this set is maximal, then it either includes f_{1} or − f_{1}.

1.
If \(f_{1} \in {\Gamma }([\hat {\mathcal {D}}_{i}])\), then \(f_{3} \in {\Gamma }([\hat {\mathcal {D}}_{i}])\) by additivity. Let us prove that in that case \(\mathcal {G}_{1}\) is almost decisive with respect to f_{3}.Consider any profile \([\hat {\mathcal {D}}'_{i}] \in \hat {\mathbb {D}}^{n}\), such that \(f_{3}\in \cap _{i\in \mathcal {G}_{1}}\hat {\mathcal {D}}'_{i}\) and \(f_{3}\notin \cup _{i\notin \mathcal {G}_{1}}\hat {\mathcal {D}}'_{i}\). Then \(\pi _{f_{3}}([\hat {\mathcal {D}}'_{i}])=\pi _{f_{3}}([\hat {\mathcal {D}}_{i}])\), and applying the independence of irrelevant alternatives we deduce that \(\mathcal {G}_{1}\) is almost decisive on f_{3}. Applying Lemma 1, we deduce that it is also decisive.

2.
If \(f_{1}\in {\Gamma }([\mathcal {D}_{i}])\), then let us show that \(\mathcal {G}_{2}\) is almost decisive for − f_{1}.Consider any profile \([\hat {\mathcal {D}}'_{i}] \in \hat {\mathbb {D}}^{n}\), such that \(f_{1}\in \cap _{i\in \mathcal {G}_{2}}\hat {\mathcal {D}}'_{i}\) and \(f_{1}\notin \cup _{i\notin \mathcal {G}_{2}}\hat {\mathcal {D}}'_{i}\). It follows that \(\pi _{f_{1}}([\hat {\mathcal {D}}'_{i}])=\pi _{f_{1}}([\hat {\mathcal {D}}_{i}])\), and applying the independence of irrelevant alternatives we deduce that \(f_{1}\in {\Gamma }([\hat {\mathcal {D}}'_{i}])\). Thus, \(\mathcal {G}_{2}\) is almost decisive on − f_{1}, and applying Lemma 1 we deduce that it is also decisive.
□
Proof of Theorem 1
The case of n = 1 is trivial, so let us assume n ≥ 2. By weak Pareto condition we know that \({\mathscr{H}}\) is decisive. By repeatedly applying Lemma 2, we can eventually arrive to a decisive individual, who must, thus, be a dictator. □
Proof of Proposition 1
Let j be the dictator. Since we have unlimited maximal domain, the social rule must be given by \({\Gamma }([\mathcal {D}_{i}])=\mathcal {D}_{j}\) for any profile \([\mathcal {D}_{i}]\), and as a consequence it satisfies independence of irrelevant alternatives. □
Proof of Lemma 3
Suppose \(\mathcal {G}\) and \(\mathcal {G}^{\prime }\) are both oligarchies, and assume that \(i^{*}\in \mathcal {G}^{\prime }\setminus \mathcal {G}\). Let \(f\notin {\mathscr{L}}^{+}\cup {\mathscr{L}}^{}\cup \{0\}\), and let us consider a profile (if needed made up only by maximal coherent sets of desirable gambles) \([\mathcal {D}_{i}]\) such that \(f \in \mathcal {D}_{i}\) for all \(i \in \mathcal {G}\) and \(f \in \mathcal {D}_{i}\) for all \(i \in \mathcal {G}^{\prime } \setminus \mathcal {G}\): it suffices to make
Then since \(\mathcal {G}^{\prime }\) is an oligarchy we should have \(f \notin {\Gamma }([\mathcal {D}_{i}])\), but \(\mathcal {G}\) decisive implies \(f \in {\Gamma }([\mathcal {D}_{i}])\), a contradiction. □
Proof of Theorem 2
Since by Lemma 3 there can be at most one oligarchy, we only need to establish its existence.
By the weak Pareto condition, the set \({\mathscr{H}}\) of all individuals is decisive. Let \(\mathcal {G}\) be a decisive set of minimal size, meaning that there does not exist \(\mathcal {G}^{\prime } \subset \mathcal {G}\) that is also decisive. We shall demonstrate that \(\mathcal {G}\) is an oligarchy. If \(\mathcal {G}\) contains a single member, then \(\mathcal {G}\) is trivially an oligarchy because any dictatorship is. Let us then consider the case in which \(\mathcal {G} \ge 2\). Note that we only need to prove that O2 holds, because O1 follows immediately because \(\mathcal {G}\) is decisive.
We consider first of all the case where Γ has unlimited domain.
Consider a profile \([\mathcal {D}_{i}]\), and let \(f\in \cup _{i\in \mathcal {G}}\mathcal {D}_{i}\). We must prove that \(f\notin {\Gamma }([\mathcal {D}_{i}])\). We may assume without loss of generality that \(f\notin {\mathscr{L}}^{+}\cup {\mathscr{L}}^{}\cup \{0\}\); the result otherwise is trivial. We can partition the group \(\mathcal {G}\) into the following sets:

◦ \(A:=\{ i \in \mathcal {G}  f \in \mathcal {D}_{i} \}\),

◦ \(B:=\{ i \in \mathcal {G}  f \in \mathcal {D}_{i} \}\),

◦ \(C:=\{ i \in \mathcal {G}  f \notin \mathcal {D}_{i}, f \notin \mathcal {D}_{i} \}\),
where by assumption A is nonempty, but B, C may be.
Since \(f \notin ({\mathscr{L}}^{+} \cup {\mathscr{L}}^{} \cup \{0\})\), there exist two points \(z_{1}, z_{2} \in \mathcal {Z}\) with z_{1} ≠ z_{2} such that f(z_{1}) = f_{1} > 0 and f(z_{2}) = f_{2} < 0. Let us prove that there always exist two gambles {h, g} such that f = h + g and

◦ \(\text {ch}(\{f,h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \), and \(g,g \notin \mathcal {D}^{1}:=\text {posi}(\{f,h\}\cup {\mathscr{L}}^{+})\);

◦ \(\text {ch}(\{f, h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \), and \(g \in \mathcal {D}^{2} :=\text {posi}(\{f,h\}\cup {\mathscr{L}}^{+})\);

◦ \(\text {ch}(\{g, h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \), and \(f,f \notin \mathcal {D}^{3}:=\text {posi}(\{g,h\}\cup {\mathscr{L}}^{+})\);

◦ \(\text {ch}(\{f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \), whence − g and as a consequence \(h \notin \mathcal {D}^{4}:=\text {posi}(\{f\}\cup {\mathscr{L}}^{+})\).
We consider a number of possibilities:

1.
If there is some z_{3} ≠ z_{1}, z_{2} such that f(z_{3}) > 0, then we define:
$$ h(z):=\begin{cases} 3f_{1}/2> 0 & \text{if } z=z_{1} \\ f_{2}/2<0 & \text{if } z=z_{2} \\ f(z)\epsilon & \text{otherwise} \end{cases} \quad\text{and}\quad g(z):=\begin{cases} f_{1}/2<0 & \text{if } z=z_{1} \\ f_{2}/2<0 & \text{if } z=z_{2} \\ \epsilon & \text{otherwise}, \end{cases} $$where 𝜖 > 0 is small enough for the conditions to be satisfied.
Then by construction f = h + g. To see that they fulfil the conditions, note that:

◦ \(\text {ch}(\{f,h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{1});

\(g \notin \mathcal {D}^{1}\) (use z_{1});

\(g \notin \mathcal {D}^{1}\) (use z_{3} with 0 < 𝜖 < f(z_{3}));


◦ \(\text {ch}(\{f, h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \): If λ_{−f}(−f) + λ_{h}h ≤ 0 it should be on the one hand λ_{−f} ≥ 0.6 because of z_{1} and on the other \(\lambda _{f} \le \frac {1}{3}\) because of z_{2}. Then \(g \in \mathcal {D}^{2}\) by additivity;

◦ \(\text {ch}(\{g, h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{1});

\(f \notin \mathcal {D}^{3}\): If f ≥ λ_{−g}(−g) + λ_{h}h for some nonnegative λ_{−g}, λ_{h} (with at least one positive), we have that
$$ f_{2} \geq  \lambda_{g} f_{2}/2 + \lambda_{h} f_{2}/2, $$whence
$$ 1 \geq \lambda_{g}/2  \lambda_{h}/2 $$dividing by the positive number − f_{2}. This means that
$$ \lambda_{h} \geq 2+ \lambda_{g} \geq 2. $$Now, this means that
$$ \lambda_{h} h(z_{1}) + \lambda_{g} (g(z_{1}))= \lambda_{h} 3f_{1}/2 + \lambda_{g} f_{1}/2 \geq 3 f_{1}> f_{1}, $$a contradiction;

\(f \notin \mathcal {D}^{3}\) (use z_{1});


◦ \(\text {ch}(\{f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) by definition of f;

\(g \notin \mathcal {D}^{4}\): If − g ≥ λ(−f) it should be on the one hand λ ≤ 0.5 because of z_{2} and on the other λ ≥ 𝜖/f(z_{3}) > 0.5 if we choose 0 < f(z_{3})/2 < 𝜖 < f(z_{3}), because of z_{3}.



2.
If there is some z_{3} ≠ z_{1}, z_{2} such that f(z_{3}) = 0, then we define:
$$ h(z):=\begin{cases} 3f_{1}/2> 0 & \text{if } z=z_{1} \\ f_{2}/2<0 & \text{if } z=z_{2} \\ f(z)\epsilon & \text{otherwise} \end{cases} \quad\text{and}\quad g(z):=\begin{cases} f_{1}/2<0 & \text{if } z=z_{1} \\ f_{2}/2<0 & \text{if } z=z_{2} \\ \epsilon & \text{otherwise}, \end{cases} $$where 𝜖 > 0 is small enough for the conditions to be satisfied.
Notice that h and g are the same of the previous case, so to show that they fulfil the conditions we indicate only the cases that involve the value of f(z_{3}).

◦ It only changes \(g \notin \mathcal {D}^{1}\): If − g ≥ λ_{f}f + λ_{h}h for some nonnegative λ_{f}, λ_{h} (with at least one positive), we have that
$$  \epsilon \ge \lambda_{f}(f(z_{3})) + \lambda_{h} (f(z_{3}) \epsilon)=\epsilon \lambda_{h}, $$from which follows λ_{h} ≥ 1, which does not work for z_{1} because − g(z_{1}) < h(z_{1}) and h(z_{1}),f_{1} are both positive;

◦ same as before;

◦ same as before;

◦ it only changes \(g \notin \mathcal {D}^{4}\) (use z_{3}).


3
Finally, if f(z_{3}) < 0 for every z_{3} ≠ z_{1}, z_{2}, then we define:
$$ h(z):=\begin{cases} f_{1}> 0 & \text{if } z=z_{1} \\ 3f_{2}/2<0 & \text{if } z=z_{2} \\ 0 & \text{otherwise} \end{cases} \quad \text{and}\quad g(z):=\begin{cases} 0 & \text{if } z=z_{1} \\ f_{2}/2>0 & \text{if } z=z_{2} \\ f(z) & \text{otherwise}. \end{cases} $$Then by construction f = h + g. To see that they fulfil the conditions, note that:

◦ \(\text {ch}(\{f,h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{1});

\(g \notin \mathcal {D}^{1}\) (use z_{1});

\(g \notin \mathcal {D}^{1}\) (use z_{1}) ;


◦ \(\text {ch}(\{f,h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{3} and z_{1}) and \(g \in \mathcal {D}_{2}\) by additivity;

◦ \(\text {ch}(\{g,h\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) (use z_{3} and z_{1});

\(f \notin \mathcal {D}^{3}\) (use z_{3});

\(f \notin \mathcal {D}^{3}\) (use z_{1}) ;


◦ \(\text {ch}(\{f\}) \cap ({\mathscr{L}}^{} \cup \{0\}) = \emptyset \) by definition of f;

\(g \notin \mathcal {D}^{4}\): (use z_{2}).


Let us consider now a profile \([\mathcal {D}^{\prime }_{i}]\) where \(\mathcal {D}^{\prime }_{i}=\mathcal {D}^{1}\) if i ∈ A, \(\mathcal {D}^{\prime }_{i}=\mathcal {D}^{2}\) if i ∈ B, \(\mathcal {D}^{\prime }_{i}=\mathcal {D}^{3}\) if i ∈ C and \(\mathcal {D}^{\prime }_{i}\) is either \({\mathscr{L}}^{+}\) or \(\mathcal {D}^{4}\) if \(i\notin \mathcal {G}\) (as needed below so as to include − f if necessary). We shall use these sets to establish that \(f\notin {\Gamma }([\mathcal {D}_{i}])\). We have a number of possible scenarios:

◦ If A, B, C are all nonempty and \(f\in {\Gamma }([\mathcal {D}_{i}])\), then we can consider a profile \([\mathcal {D}^{\prime }_{i}]\) as above so that \(\pi _{f}([\mathcal {D}_{i}])=\pi _{f}([\mathcal {D}^{\prime }_{i}])\). Applying independence of irrelevant alternatives, it follows that \(f\in {\Gamma }([\mathcal {D}^{\prime }_{i}])\). Since \(\mathcal {G}\) is decisive and \(h\in \cap _{i\in \mathcal {G}}\mathcal {D}^{\prime }_{i}\), we also have that \(h\in {\Gamma }([\mathcal {D}^{\prime }_{i}])\), we deduce that − g = −f + h belongs to \({\Gamma }([\mathcal {D}^{\prime }_{i}])\). By construction, \(\pi _{g}([\mathcal {D}^{\prime }_{i}])=(\vec {1}_{B\cup C})\). Applying independence of irrelevant alternatives, for any other profile \([\mathcal {D}^{\prime \prime }_{i}]\) such that \(g\in \cap _{i\in B\cup C}\mathcal {D}^{\prime \prime }_{i}\) and \(g\notin \cup _{i\notin B\cup C}\mathcal {D}^{\prime \prime }_{i}\), it holds that \(g\in {\Gamma }([\mathcal {D}^{\prime \prime }_{i}])\). But this means that B ∪ C is almost decisive with respect to g, and applying Lemma 1, it is decisive. This contradicts that \(\mathcal {G}\) is a decisive set of minimal size.

◦ If A≠∅ ≠ B and C = ∅, we reason as in the previous case, and end up concluding that B is a decisive set.

◦ If A≠∅ ≠ C and B = ∅, we reason as in the first case, and end up concluding that C is a decisive set.

◦ Finally, if A ≠ ∅ and B = C = ∅, it holds that \(f\in \cap _{i\in \mathcal {G}}\mathcal {D}_{i}\), whence, since \(\mathcal {G}\) is decisive, \(f\in {\Gamma }([\mathcal {D}_{i}])\). Since the latter is a coherent set, this means that \(f\notin {\Gamma }([\mathcal {D}_{i}])\).
We consider next the case of unlimited maximal domain. Recall that since \(\mathcal {G}\) is decisive, condition O1 is satisfied. Let us consider next condition O2. Consider a profile \([\hat {\mathcal {D}_{i}}] \in \hat {\mathbb {D}}^{n}\). Assume that \(f\in \cup _{i\in \mathcal {G}}\hat {\mathcal {D}_{i}}\). Since all the sets in the profile are maximal, we can partition again the group \(\mathcal {G}\) into the following sets:

◦ \(A:=\{ i \in \mathcal {G}  f \in \hat {\mathcal {D}_{i}} \}\),

◦ \(B:=\{ i \in \mathcal {G}  f \in \hat {\mathcal {D}_{i}} \}\),
where by assumption A is nonempty, but B may be. We need to show that \(f\notin {\Gamma }([\hat {\mathcal {D}}_{i}])\). There are two possibilities:

◦ if both A, B are nonempty and \(f\in {\Gamma }([\hat {\mathcal {D}_{i}}])\), applying independence of irrelevant alternatives, for any other profile \([\hat {\mathcal {D}^{\prime }_{i}}]\) such that \( f \in \cap _{i\in B}\hat {\mathcal {D}}'_{i}\) and \(f \notin \cup _{i\in A}\hat {\mathcal {D}}'_{i}\), it holds that \(f\in {\Gamma }([\hat {\mathcal {D}}'_{i}])\). But this means that B is almost decisive with respect to − f, and applying Lemma 1, it is decisive. This contradicts that \(\mathcal {G}\) is a decisive set of minimal size;

◦ if instead A≠∅ and B = ∅, it holds that \(f\in \cap _{i\in \mathcal {G}}\hat {\mathcal {D}_{i}}\), whence, since \(\mathcal {G}\) is decisive, \(f\in {\Gamma }([\hat {\mathcal {D}_{i}}])\). Since the latter is a coherent set, this means that \(f\notin {\Gamma }(\hat {[\mathcal {D}_{i}]})\).
□
Proof of Proposition 2
Assume exabsurdo that there exists a profile \([\mathcal {D}_{i}]\) such that \({\Gamma }([\mathcal {D}_{i}]) \supsetneq \cap _{i \in \mathcal {G}} \mathcal {D}_{i}\), and let us consider \(f\in {\Gamma }([\mathcal {D}_{i}]) \setminus \cap _{i \in \mathcal {G}} \mathcal {D}_{i}\), so that \(\emptyset \neq A=\{i\in \mathcal {G}: f \in \mathcal {D}_{i}\}\neq \mathcal {G}\). For every \(j\in \mathcal {G}\setminus A\), let \(\mathcal {D}^{\prime }_{j}\) be a maximal set of desirable gambles that includes − f, and let \(\mathcal {D}^{\prime }_{i}:=\mathcal {D}_{i}\) for every \(i\in A \cup \mathcal {G}^{c}\). Then since \(f\in \cup _{i \in \mathcal {G}} \mathcal {D}^{\prime }_{i}\), it follows from O2 that \(f\notin {\Gamma }([\mathcal {D}^{\prime }_{i}])\). But on the other hand f belongs to the same sets in the profiles \([\mathcal {D}_{i}]\) and \([\mathcal {D}^{\prime }_{i}]\), so \(f\in {\Gamma }([\mathcal {D}_{i}]), f\notin {\Gamma }([\mathcal {D}^{\prime }_{i}])\) is a contradiction with independence of irrelevant alternatives. □
Proof of Theorem 3
From Theorem 2 there exists a unique oligarchy \(\mathcal {G}\). Then we must show that if the social rule satisfies also anonymity, \(\mathcal {G}= {\mathscr{H}}\). Assume exabsurdo that \(\mathcal {G}\neq {\mathscr{H}}\), and take \(i^{*}\notin \mathcal {G}\).
By the hypothesis of unlimited (maximal) domain we can consider a profile (possibly composed only by maximal coherent sets of desirable gambles) \([\mathcal {D}_{i}]\) such that \(f \in \mathcal {D}_{i^{*}}\) and \(f \in \mathcal {D}_{j} \ \forall j\neq i^{*}\). Since \(\mathcal {G}\) is an oligarchy, it follows that \(f \in {\Gamma }([\mathcal {D}_{i}])\).
Consider now a permutation σ of \({\mathscr{H}}\) such that \(\sigma (i^{*})\neq i^{*}, \sigma (i^{*}) \in \mathcal {G} \), and let \([\mathcal {D}^{\prime }_{i}]:=[\mathcal {D}_{\sigma (i)}]\) denote the associated profile. Then there exists some \(j\in \mathcal {G}\) such that \(f\in \mathcal {D}^{\prime }_{j}\), whence, by definition of oligarchy, \(f \notin {\Gamma }([\mathcal {D}^{\prime }_{i}])\). But this means that \({\Gamma }([\mathcal {D}_{\sigma (i)}])\neq {\Gamma }([\mathcal {D}_{i}])\), meaning that anonymity is violated. □
Proof of Proposition 3
Thanks to Proposition 2 we know that for every profile (possibly composed only by maximal coherent sets of desirable gambles) \([\mathcal {D}_{i}]\), \({\Gamma }([\mathcal {D}_{i}])= \cap _{i} \mathcal {D}_{i}\). Hence it satisfies anonymity. □
Proof of Lemma 4
That \(\mathcal {E}\) is coherent follows from [35, Prop. 29].
From (6), any \(f \in \mathcal {E}\) can be written as
with \(f_{0} \in {\mathscr{L}}^{+}({\mathscr{H}} \times \mathcal {Z}) \cup \{0\}, f_{i} \in \mathcal {D}_{i}\cup \{0\}, \lambda _{i}>0, f \neq 0\). Then \(f \in \text {Marg}_{\mathcal {Z}}(\mathcal {E})\) iff \(f \in \mathcal {E}\) and
This is equivalent to
You can observe that, fixing i and j, the left term of this equation is a gamble in \(\mathcal {D}_{i}\) and the right term is a gamble in \(\mathcal {D}_{j}\), considering also that f≠ 0. Thus, \(f \in \text {Marg}_{\mathcal {Z}}(\mathcal {E})\) if and only if it depends only on \(z \in \mathcal {Z}\) and, as a function of only \(z \in \mathcal {Z}\), it belongs to \(\cap _{i \in {\mathscr{H}}} \mathcal {D}_{i}\). Hence we have the thesis. □
Proof of Theorem 4
Let us address the points of the statement in turn.

1.
The equality of the sets in (6) and Theorem 4 is well known, see for instance [35, Prop. 29].

2.
The converse implication is trivial: if such an \(\mathcal {E}^{\prime }\) exists, then \({\Gamma }([\mathcal {D}_{i}])=\text {Marg}_{\mathcal {Z}}(\mathcal {E}^{\prime })\supseteq \text {Marg}_{\mathcal {Z}}(\mathcal {E})\).
For the direct implication, consider \(\mathcal {E}^{\prime }\) as defined in (7). It includes \(\mathcal {E}\) by definition. To prove that it is coherent, note that \(\mathcal {E}^{\prime }=\text {posi}({\Gamma }([\mathcal {D}_{i}])\otimes {\mathscr{H}} \cup \cup _{i} \mathcal {D}i)\) (this is due again to the transformation referenced in point 1). It includes \({\mathscr{L}}^{+}({\mathscr{H}}\times \mathcal {Z})\) because for every \(f\in {\mathscr{L}}^{+}({\mathscr{H}}\times \mathcal {Z})\), the restriction of \(\mathbb {I}_{i}\otimes f\) on \({\mathscr{L}}(\mathcal {Z})\) belongs to \(\mathcal {D}_{i}\cup \{0\}\) for all \(i\in {\mathscr{H}}\). Thus, coherence holds if and only if \(0\notin \text {posi}({\Gamma }([\mathcal {D}_{i}])\otimes {\mathscr{H}} \cup \cup _{i} \mathcal {D}i)\).
To prove that this is the case, let us reason by contradiction. Note that the zero gamble can in principle be produced only by adding \(f_{0}\otimes {\mathscr{H}}\) with \({\sum }_{i\in {\mathscr{H}}}\mathbb {I}_{i}\otimes f_{i}\), for some \(f_{0} \in {\Gamma }([\mathcal {D}_{i}])\cup \{0\},(\forall i\in {\mathscr{H}}) f_{i}\in \mathcal {D}_{i}\cup \{0\}\), not all of them zero. In order for the sum to yield zero, f_{0} must be different from zero. And since \(f_{0}\otimes {\mathscr{H}}\) is \(\mathcal {Z}\)measurable, in order to yield zero it must hold that \(f_{1}=\dots =f_{n}=f_{0}\). This implies that \(f_{0}\in \cap _{i\in {\mathscr{H}}}\mathcal {D}_{i}\). We are showing the direct implication, so it follows by hypothesis \({\Gamma }([\mathcal {D}_{i}])\supseteq \text {Marg}_{\mathcal {Z}}(\mathcal {E})\) and Lemma 4, that \(f_{0}\in {\Gamma }([\mathcal {D}_{i}])\). But then \({\Gamma }([\mathcal {D}_{i}])\) contains both f_{0} and − f_{0}. This contradicts the coherence of \({\Gamma }([\mathcal {D}_{i}])\). Therefore the sum of two gambles in \(\mathcal {E}^{\prime }\) must be different from zero.
To conclude, we show that \({\Gamma }([\mathcal {D}_{i}])=\text {Marg}_{\mathcal {Z}}(\mathcal {E}^{\prime })\). Since the direct inclusion is trivial, we focus on the converse inclusion. Remember that \(\text {Marg}_{\mathcal {Z}}(\mathcal {E}^{\prime })=\mathcal {E}^{\prime }\cap {\mathscr{L}}_{\mathcal {Z}}({\mathscr{H}}\times \mathcal {Z})\). Let us consider a \(\mathcal {Z}\)measurable gamble \(f\in \mathcal {E}^{\prime }\): \(f=f_{0}\otimes {\mathscr{H}}+{\sum }_{i\in {\mathscr{H}}}\mathbb {I}_{i}\otimes f_{i}\). The case \(f_{1}=\dots =f_{n}=0\) is trivial, so let us assume that there is \(i\in {\mathscr{H}}\) such that f_{i}≠ 0. Since f is \(\mathcal {Z}\)measurable, this means that
$$ (\forall i,j \in \mathcal{H})(\forall z \in \mathcal{Z})\ f(i,z)= f_{0}(z)+ f_{i}(z) = f_{0}(z) + f_{j}(z) = f(j, z). $$This implies f_{j} = f_{i} for all j≠i. As a consequence, \(f_{i}\in \text {Marg}_{\mathcal {Z}}(\mathcal {E})\) and by hypothesis then \(f_{i}\in {\Gamma }([\mathcal {D}_{i}])\). It follows that \(f_{0}+f_{1}+\dots +f_{n}\in {\Gamma }([\mathcal {D}_{i}])\), whence \(f\in {\Gamma }([\mathcal {D}_{i}])\).

3.
Assume by contradiction that there is such a set \(\mathcal {E}^{\prime \prime }\): \(\mathcal {E}^{\prime }\supsetneq \mathcal {E}^{\prime \prime }\supseteq \mathcal {E}\). Since \(\mathcal {E}^{\prime }\) strictly contains \(\mathcal {E}^{\prime \prime }\), there must be some \(f_{0}\in {\Gamma }([\mathcal {D}_{i}])\) such that \(f_{0}\otimes {\mathscr{H}}\notin \mathcal {E}^{\prime \prime }\). Note that \(f_{0}\notin \cap _{i\in {\mathscr{H}}}\mathcal {D}_{i}\), otherwise it would belong to \(\mathcal {E}^{\prime \prime }\) given that \(\mathcal {E}^{\prime \prime }\supseteq \mathcal {E}\). Whence \(f_{0}\in {\Gamma }([\mathcal {D}_{i}])\setminus \cap _{i\in {\mathscr{H}}}\mathcal {D}_{i}\) and by marginalising \(\mathcal {E}^{\prime \prime }\) we obtain a set that does not contain f_{0}; therefore \({\Gamma }([\mathcal {D}_{i}])\neq \text {Marg}_{\mathcal {Z}}(\mathcal {E}^{\prime \prime })\). This is a contradiction.

4.
Remember that \(\mathcal {E}^{\prime }i=\{f\in \mathcal {E}^{\prime }:f=\mathbb {I}_{i}f\}\). Since \(f_{0}\otimes {\mathscr{H}}\), in the definition of \(\mathcal {E}^{\prime }\), is constant on the elements of \({\mathscr{H}}\), any gamble \(f=\mathbb {I}_{i}f\) must be such that f_{0} = 0. The thesis then follows immediately.

5.
Remember again that \(\text {Marg}_{{\mathscr{H}}}(\mathcal {E}^{\prime })=\mathcal {E}^{\prime }\cap {\mathscr{L}}_{{\mathscr{H}}}({\mathscr{H}}\times \mathcal {Z})\). Let us consider a \({\mathscr{H}}\)measurable gamble \(f\in \mathcal {E}^{\prime }\): \(f=f_{0}\otimes {\mathscr{H}}+{\sum }_{i\in {\mathscr{H}}}\mathbb {I}_{i}\otimes f_{i}\) for some \(f_{0} \in {\Gamma }([\mathcal {D}_{i}])\cup \{0\},(\forall i\in {\mathscr{H}}) f_{i}\in \mathcal {D}_{i}\cup \{0\}\), not all of them zero. \({\mathscr{H}}\)measurability requires that
$$ (\forall i \in \mathcal{H})(\forall z,z' \in \mathcal{Z})\ f(i,z)= f_{0}(z)+ f_{i}(z) = f_{0}(z') + f_{i}(z') = f(i, z'). $$This is valid if and only if f_{i} + f_{0} = k_{i} constant \(\forall i\in {\mathscr{H}}\). Hence we have that \(f \in \text {Marg}_{{\mathscr{H}}}(\mathcal {E}^{\prime })\) is a gamble \(f\in \mathcal {E}^{\prime }\): \(f=f_{0}\otimes {\mathscr{H}}+{\sum }_{i\in {\mathscr{H}}}\mathbb {I}_{i}\otimes f_{i}\) such that f_{i}(z) = −f_{0}(z) + k_{i} for all \(z \in \mathcal {Z}\), where k_{i} is a constant for all \(i \in {\mathscr{H}}\).
Now we can distinguish two cases:

(a) If f_{0}(z) = k for all \(z \in \mathcal {Z}\), then it must be k ≥ 0, considering that \(f_{0}\in {\Gamma }([\mathcal {D}_{i}])\cup \{0\}\). It then follows that f_{i}(z) = k_{i} − k for all \(z \in \mathcal {Z}\), for all \(i \in {\mathscr{H}}\), and it must be k_{i} − k ≥ 0, since otherwise \(f_{i} \in {\mathscr{L}}^{}(\mathcal {Z})\) and this would contradict \(f_{i} \in \mathcal {D}_{i}\cup \{0\}\). We deduce that \((\forall i \in {\mathscr{H}})(\forall z \in \mathcal {Z})\ f(i,z)=k+k_{i}k \ge 0\) and so \(f \in {\mathscr{L}}^{+}({\mathscr{H}})\). Note that f≠ 0 otherwise \(f \notin \mathcal {E}^{\prime }\).

(b) Assume next that f_{0} is not constant (and in particular nonzero). This means that it must be f_{i} = k_{i} − f_{0}≠ 0 for every \(i\in {\mathscr{H}}\), and in particular \(f_{i}\in \mathcal {D}_{i}\) for every i.
For all \(i \in {\mathscr{H}}\) we can define
$$ {\overline{P}}_{i}(f_{0}):= \inf\{ \lambda \in \mathbb{R} : \lambda  f_{0} \in \mathcal{D}_{i} \}, $$the upper prevision of f_{0} obtained from \(\mathcal {D}_{i}\). Thus, \(k_{i} \ge {\overline {P}}_{i}(f_{0})\), for all \(i \in {\mathscr{H}}\), and the equality can only hold if for some \(i \in {\mathscr{H}}\), \({\overline {P}}_{i}(f_{0})\) is a mininum, i.e. if \({\overline {P}}_{i}(f_{0})f_{0} \in \mathcal {D}_{i}\). This means that, given that f(i, z) = k_{i} for every \(i\in {\mathscr{H}},z\in \mathcal {Z}\),
$$ f(i,z) \ge {\overline{P}}_{i}(f_{0}), $$and that the inequality is strict unless \({\overline {P}}_{i}(f_{0})f_{0}\in \mathcal {D}_{i}\). Now, given the fourth statement, we know that for every \(i\in {\mathscr{H}},g\in {\mathscr{L}}(\mathcal {Z})\),
$$ {\overline{P}}_{i}(g)= {\overline{P}}(gi), $$so, defining the conditional upper prevision \({\overline {P}}(g {\mathscr{H}}):= {\sum }_{i} \mathbb {I}_{i}{\overline {P}}(gi)\), we have that \(f \ge {\overline {P}}(f_{0} {\mathscr{H}})\), and that the inequality is strict unless \({\overline {P}}(f_{0}i)f_{0} \in \mathcal {D}_{i}\).
This establishes the form of \(\text {Marg}_{{\mathscr{H}}}(\mathcal {E}^{\prime })\).

□
Proof of Corollary 1
Weak Pareto means that if a gamble f is strictly desirable for any P_{i}, \(i=1,\dots ,n\), then it should also be strictly desirable for P, or, equivalently, that \(P\geq \min \limits _{i} P_{i}\) (Theorem 6(1)). But this means that P belongs to the credal set associated with the coherent lower prevision \(\underline {P}:=\min \limits _{i} P_{i}\), and as a consequence that it is a convex combination of \(P_{1},\dots ,P_{n}\). □
Proof of Theorem 5
If π is not degenerate, then we can find j_{1} ≠ j_{2} in \({\mathscr{H}}\) such that π(j_{1}),π(j_{2}) > 0. Consider a profile \([\hat {\mathcal {D}_{i}}] \in \mathcal {A}\) such that \(\hat {\mathcal {D}_{i}}\) assigns all the mass to (s_{1}, x_{1}) if i = j_{1} and all the mass to (s_{2}, x_{2}) if i≠j_{1}, where s_{1} ≠ s_{2} and j_{1} ≠ j_{2}. Then we obtain
meaning that the linear prevision induced by \({\Gamma }([\hat {\mathcal {D}_{i}}])\) is not state independent. This is a contradiction. □
Proof of Corollary 2

1.
It follows from Theorem 2 and definition of weak dictatorship.

2.
This is a consequence of Theorem 1 and definition of strong dictatorship.

3.
It follows from Theorem 3 and definition of weak dictatorship.
□
Proof of Theorem 6
We shall establish the first equivalence; the second equivalence follows taking into account that
that the credal set associated with \(\min \limits _{i} {\underline {P}}_{i}\) is \(\overline {\text {ch}(\cup _{i} {\mathscr{M}}({\underline {P}}_{i}))}\) and that the one associated with \(\max \limits _{i} {\underline {P}}_{i}\) is \(\cap _{i} {\mathscr{M}}({\underline {P}}_{i})\), where the credal set associated with a coherent lower prevision is given by (2).
We shall denote by \(\mathcal {D}_{i}\) the coherent set of strictly desirable gambles associated with \({\underline {P}}_{i}\) by means of (3), and by \(\mathcal {D}\) the coherent set of strictly desirable gambles associated with \(\underline {P}={\Gamma }([P_{i}])\).

1.
Let us establish the direct implication. Assume that Γ satisfies weak Pareto, and as a consequence that \(\cap _{i \in {\mathscr{H}}} \mathcal {D}_{i}\subseteq {\Gamma }^{\prime }([\mathcal {D}_{i}])\), where \({\Gamma }^{\prime }\) is associated with Γ by means of (1). It follows that, for any \(f \in {\mathscr{L}}\),
$$ {\underline{P}}(f):= \sup \{ \mu \in \mathbb{R}: f \mu \in {\Gamma}^{\prime}[\mathcal{D}_{i}] \} \ge \sup \{ \mu \in \mathbb{R}: f \mu \in \cap_{i \in \mathcal{H}} \mathcal{D}_{i} \}. $$Let us prove that the righthand side is greater than or equal to \(\min \limits _{i} {\underline {P}}_{i}(f)\). Assume that \(\min \limits _{i} {\underline {P}}_{i}(f)={\underline {P}}_{j}(f)\). Then it follows from (1) that for any 𝜖 > 0, \(f\underline {P}_{j}(f)+\epsilon \) belongs to \(\mathcal {D}_{j}\), and for any \(j^{\prime }\neq j\) it holds that \(f\underline {P}_{j}(f)+\epsilon \geq f\underline {P}_{j^{\prime }}(f)+\epsilon \in \mathcal {D}_{j^{\prime }}\). As a consequence, \(f\underline {P}_{j}(f)+\epsilon \in \cap _{i} \mathcal {D}_{i}\), whence
$$ \sup \{ \mu \in \mathbb{R}: f \mu \in \cap_{i \in \mathcal{H}} \mathcal{D}_{i} \}\geq{\underline{P}}_{j}(f)\epsilon. $$Since this holds for any 𝜖 > 0, it follows that
$$ \sup \{ \mu \in \mathbb{R}: f \mu \in \cap_{i \in \mathcal{H}} \mathcal{D}_{i} \}\geq{\underline{P}}_{j}(f)=\min_{i} {\underline{P}}_{i}(f). $$We conclude that \({\underline {P}}(f)\geq \min \limits _{i} {\underline {P}}_{i}(f)\) for every gamble f.
To see the converse, given a gamble \(f\in (\cap _{i} \mathcal {D}_{i})\setminus {\mathscr{L}}^{+}\), it holds that \({\underline {P}}_{i}(f)>0\) for every i, whence \(0<\min \limits {\underline {P}}_{i}(f)\leq {\underline {P}}(f)\), and as a consequence f belongs to the set of strictly desirable gambles associated with \(\underline {P}\). Since trivially \((\cap _{i} \mathcal {D}_{i})\cap {\mathscr{L}}^{+}\subseteq {\mathscr{L}}^{+}\subseteq \mathcal {D}\), we deduce that \({\Gamma }^{\prime }\) satisfies weak Pareto and therefore so does Γ.

2.
This follows from (1) and (3) and the definition of strict completeness and maximal strict desirability.

3.
This follows applying the first statement to \(\mathcal {G}\) instead of \({\mathscr{H}}\).

4.
The first part is a consequence of the third statement. Let us now prove that
$$ \left[(\exists i \in \mathcal{G})(f \in \mathcal{D}_{i})) \Rightarrow f \notin {\Gamma}^{\prime}([\mathcal{D}_{i}])\right] \Leftrightarrow {\overline{P}}\geq \max_{i\in\mathcal{G}}{\underline{P}}_{i}. $$(C.6)To see the direct implication, consider a gamble f, and let \({\underline {P}}_{j}(f)=\max \limits _{i\in \mathcal {G}} {\underline {P}}_{i}(f)\). If \({\underline {P}}_{j}(f)>0\), then \(f\in \mathcal {D}_{j}\), whence \(f\notin {\Gamma }^{\prime }([\mathcal {D}_{i}])\) and, from (1) and the conjugacy relation \(\overline {P}(\cdot )=\underline {P}(\cdot )\), it follows that \(\overline {P}(f)\geq 0\). Moreover, it must be \(\underline {P}(f)>0\): otherwise by considering \(f^{\prime }:= f\frac {\underline {P}_{j}(f)}{2}\), we would obtain \(\underline {P}_{j}(f^{\prime })>0\) and \(\overline {P}(f^{\prime })<0\). Therefore, we conclude that \(\max \limits _{i\in \mathcal {G}} \underline {P}_{i}(f)>0 \Rightarrow \overline {P}(f)>0\). But since both \(\max \limits _{i\in \mathcal {G}} \underline {P}_{i}\) and \(\overline {P}\) satisfy constant additivity, this means that \(\overline {P}\geq \max \limits _{i\in \mathcal {G}}\underline {P}_{i}\).
Conversely, if there is a gamble f such that \(f\in \mathcal {D}_{j}\) for some \(j\in \mathcal {G}\) while \(f\in {\Gamma }^{\prime }([\mathcal {D}_{i}])\), it necessarily must be \(f\notin {\mathscr{L}}^{+}\). It then follows that \(\max \limits _{i\in \mathcal {G}}\underline {P}_{i}(f)\geq \underline {P}_{j}(f)>0\), while \(0<\underline {P}(f)=\overline {P}(f)\), meaning that \(\overline {P}(f)<0\). This is a contradiction.
As a consequence, (C.6) holds. This concludes the proof of this statement.

5.
This is a particular case of the third statement with \(\mathcal {G}=1\).

6.
Γ satisfies anonymity if and only if \({\Gamma }^{\prime }([\mathcal {D}_{i}]) = {\Gamma }^{\prime }([\mathcal {D}_{\sigma (i)}])\). It follows that, for any \(f \in {\mathscr{L}}\),
$$ {\underline{P}}(f):= \sup \{ \mu {}\in{} \mathbb{R}{}:{} f \mu \in {\Gamma}^{\prime}([\mathcal{D}_{i}]) \} {}={} \sup \{ \mu{} \in{} \mathbb{R}{}:{} f \mu \in {\Gamma}^{\prime}([\mathcal{D}_{\sigma(i)}]) \} =\!: {\underline{P}}_{\sigma}(f). $$To see the converse, let us suppose without loss of generality that there is a gamble \(f \in {\Gamma }^{\prime }([\mathcal {D}_{i}]) \cap ({\Gamma }^{\prime }([\mathcal {D}_{\sigma (i)}])^{c}\) for some permutation σ of \({\mathscr{H}}\). Then it follows that \(\underline {P}(f)>0\) and \(\underline {P}_{\sigma }(f)\le 0\) and this contradicts \(\underline {P}(f)=\underline {P}_{\sigma }(f)\).

7.
Let us consider \(f \notin {\mathscr{L}}^{+}\) and two profiles \([{\underline {P}}_{i}], [{\underline {P}}^{\prime }_{i}]\) such that \((\forall i \in {\mathscr{H}}) {\underline {P}}_{i}(f){\underline {P}}^{\prime }_{i}(f)>0\). If Γ satisfies independence of irrelevant alternatives, then \(f \in {\Gamma }^{\prime }([\mathcal {D}_{i}]) \Leftrightarrow {\Gamma }^{\prime }([\mathcal {D}^{\prime }_{i}])\). As a consequence, we have \(\underline {P}(f)>0 \Leftrightarrow \underline {P}^{\prime }(f)>0\), whence \(\underline {P}(f)\underline {P}^{\prime }(f)>0\).
Conversely, if \({\underline {P}}(f){\underline {P}}^{\prime }(f)>0\) then \(f \in {\Gamma }^{\prime }([\mathcal {D}_{i}]) \Leftrightarrow {\Gamma }^{\prime }([\mathcal {D}^{\prime }_{i}])\) by (3), whence Γ satisfies independence of irrelevant alternatives.
□
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Casanova, A., Miranda, E. & Zaffalon, M. Joint desirability foundations of social choice and opinion pooling. Ann Math Artif Intell 89, 965–1011 (2021). https://doi.org/10.1007/s10472021097337
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10472021097337
Keywords
 Arrow
 Desirability
 Social choice
 Opinion pooling
 Imprecise probabilities
 Coherence