Abstract
The principle of differential monotonicity for cooperative games states that the differential of two players’ payoffs weakly increases whenever the differential of these players’ marginal contributions to coalitions containing neither of them weakly increases. Together with the standard efficiency property and a relaxation of the null player property, differential monotonicity characterizes the egalitarian Shapley values, i.e., the convex mixtures of the Shapley value and the equal division value for games with more than two players. For games that contain more than three players, we show that, cum grano salis, this characterization can be improved by using a substantially weaker property than differential monotonicity. Weak differential monotonicity refers to two players in situations where one player’s change of marginal contributions to coalitions containing neither of them is weakly greater than the other player’s change of these marginal contributions. If, in such situations, the latter player’s payoff weakly/strictly increases, then the former player’s payoff also weakly/strictly increases.
Similar content being viewed by others
Notes
As already mentioned by Young (1985) , strong monotonicity implies and can be relaxed into marginality, i.e., a player’s payoff only depends on her own productivity.
Besides efficiency, symmetry and weak monotonicity, the characterizations of the egalitarian Shapley values due to van den Brink et al. (2013) involve a fourth axiom, either linearity or weak covariance. Their characterization using linearity also covers the two-player case.
References
Aczél J (1966) Lectures on functional equations and their applications. Academic Press, New York
Béal S, Casajus A, Huettner F, Rémila E, Solal P (2016) Characterizations of weighted and equal division values. Theory Decis 80(4):649–667
Casajus A (2011) Differential marginality, van den Brink fairness, and the Shapley value. Theory Decis 71(2):163–174
Casajus A, Huettner F (2013) Null players, solidarity, and the egalitarian Shapley values. J Math Econ 49:58–61
Casajus A, Huettner F (2014a) On a class of solidarity values. Eur J Oper Res 236:583–591
Casajus A, Huettner F (2014b) Weakly monotonic solutions for cooperative games. J Econ Theory 154:162–172
Casajus A, Yokote K (2017) Weak differential marginality and the Shapley value. J Econ Theory 176:274–284
Harsanyi JC (1959) A bargaining model for cooperative \(n\)-person games. In: Tucker AW, Luce RD (eds) Contributions to the theory of games IV, vol 2. Princeton University Press, Princeton, pp 325–355
Joosten R (1996) Dynamics, equilibria and values. Ph.D. Thesis, Maastricht University, The Netherlands
Nowak AS, Radzik T (1994) A solidarity value for \(n\)-person transferable utility games. Int J Game Theory 23:43–48
Shapley LS (1953) A value for \(n\)-person games. In: Kuhn H, Tucker A (eds) Contributions to the theory of games, vol II. Princeton University Press, Princeton, pp 307–317
Shapley LS (1988) Utility comparison and the theory of games. In: Roth AE (ed) The Shapley value. Cambridge University Press, Cambridge, pp 307–319 Ch. 19
Sprumont Y (1990) Population monotonic allocation schemes for cooperative games with transferable utility. Games Econ Behav 2:378–394
van den Brink R (2001) An axiomatization of the Shapley value using a fairness property. Int J Game Theory 30:309–319
van den Brink R, Funaki Y, Ju Y (2013) Reconciling marginalism with egalitarism: consistency, monotonicity, and implementation of egalitarian Shapley values. Soc Choice Welf 40:693–714
Yokote K, Casajus A (2017) Weak differential monotonicity, flat tax, and basic income. Econ Lett 151:100–103
Young HP (1985) Monotonic solutions of cooperative games. Int J Game Theory 14:65–72
Acknowledgements
We are grateful to René van den Brink for valuable comments on this paper. André Casajus: Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)–388390901. Koji Yokote: Financial support by the Japan Society for the Promotion of Science (JSPS) is gratefully acknowledged.
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
Proof of Theorem 3
1.1 Preamble: squeezing water from stone
Proving characterizations for a parametrized classes of solutions with few rather weak axioms is like squeezing water from stone. Their proofs tends to be lengthy and rather technical. Our proof is no exception in this respect. It is by induction on the number of non-vanishing Harsanyi dividends of a game for non-singleton coalitions and can be divided into three major parts. While the first two parts provide the induction basis, the third one the induction step.
The first part (Claim 1) shows the theorem for inessential games, which includes the derivation of the parameter \(\alpha \in \left[ 0,1\right] \) from the solution \(\varphi .\) Its proof largely mimics the proof of Yokote and Casajus (2017, Theorem 2) . Since the latter result involves a related but different average dummy player property and in order to keep our paper self-contained, we present the full proof. The second part (Claims 2, 3, and 4) and the third part extend part one to general games. Their proof is an adaptation of the proof of Casajus and Yokote (2017, Theorem 2) to the egalitarian Shapley values and the use of the average dummy property instead of the null player property.
1.2 The proof
It is well-known that any value \(\mathrm {Sh}^{\alpha },\)\(\alpha \in \left[ 0,1\right] \) satisfies E. Since DMo implies DMo\(^{-}\), Casajus and Huettner (2013, Theorem 4) entails that any \(\mathrm {Sh}^{\alpha }\) also obeys DMo\(^{-}\). By (4) and the fact that \(\mathrm {Sh}\) meets D, any \(\mathrm {Sh}^{\alpha }\) meets AD. Let \(\left| N\right| >3\) and let the solution \(\varphi \) meet E, AD, and DMo\(^{-}\).
For \(v\in \mathbb {V},\) set
We show \(\varphi =\mathrm {Sh}^{\alpha }\) for some \(\alpha \in \left[ 0,1\right] \) by induction on \(\left| \mathcal {T}_{>1}\left( v\right) \right| .\) For this purpose, we “reduce” \(\left| \mathcal {T}_{>1}\left( v\right) \right| \) without changing \(v\left( N\right) \) by the following construction: For \(T\subseteq N,\)\(\left| T\right| >1\), let \(\bar{u}_{T}\in \mathbb {V}\) be given by
Note that \(\mathrm {Sh}_{i}^{\alpha }\left( \bar{u}_{T}\right) =0\) for all \(\alpha \in \left[ 0,1\right] \) and \(i\in N.\) For \(T\in \mathcal {T}_{>1}\left( v\right) ,\) let \(v_{T}\in \mathbb {V}\) be given by
By construction, (*) \(\left| \mathcal {T}_{>1}\left( v_{T}\right) \right| =\left| \mathcal {T}_{>1}\left( v\right) \right| -1\) and (**) \(v\left( N\right) =v_{T}\left( N\right) .\)
Induction basis: We show \(\varphi \left( v\right) =\mathrm {Sh}^{\alpha }\left( v\right) \) for some \(\alpha \in \left[ 0,1\right] \) and all \(v\in \mathbb {V}\) such that \(\left| \mathcal {T} _{>1}\left( v\right) \right| \le 1\) by a number of claims and subclaims.
If \(\left| \mathcal {T}_{>1}\left( v\right) \right| =0\) for \(v\in \mathbb {V},\) then \(v\in \mathbb {\bar{V}}\), i.e., v is inessential.
Claim 1, C1: There exists some \(\alpha \in \left[ 0,1\right] \) such that \(\varphi \left( v\right) =\mathrm {Sh}^{\alpha }\left( v\right) \) for all \(v\in \mathbb {\bar{V}}.\)
One can easily check that there is a bijection \(\mathbb {R}^{N}\rightarrow \mathbb {\bar{V}}\), \(x\mapsto v_{x},\) where \(v_{x}\) is given by \(v_{x}\left( S\right) =\sum _{\ell \in S}x_{\ell }\) for all \(S\subseteq N.\) Abusing notation, we identify \(\mathbb {\bar{V}}\) with \(\mathbb {R}^{N}\) and write x instead of \(v_{x}.\) By D, we have \(\mathrm {Sh}\left( x\right) =x\) for all \(x\in \mathbb {R}^{N}\) and therefore
Set \(n:=\left| N\right| .\) For \(\lambda \in \mathbb {R}\) and \(x\in \mathbb {R}^{N},\) we define \(\lambda \cdot x\in \mathbb {R}^{N}\) by \(\left( \lambda \cdot x\right) _{\ell }=\lambda \cdot x_{\ell }\) for all \(\ell \in N.\) Further, for \(i,j\in N,\)\(i\ne j,\) we define \(e^{ij}\in \mathbb {R}^{N}\) by \(e_{i}^{ij}=1,\)\(e_{j}^{ij}=-1,\) and \(e_{\ell }^{ij}=0\) for all \(\ell \in N{\setminus }\left\{ i,j\right\} .\) Moreover, for \(\mu \in \mathbb {R},\) we define \(e^{\mu }\in \mathbb {R}^{N}\) by \(e_{\ell }^{\mu }=\frac{\mu }{n}\) for all \(\ell \in N.\)
For all \(i\in N,\)\(j\in N{\setminus }\left\{ i\right\} ,\) and \(\mu \in \mathbb {R}\), let the mapping \(g_{ij}^{\mu }:\mathbb {R}\rightarrow \mathbb {R}\) be given by
Note that for \(\varphi =\mathrm {Sh}^{\alpha },\)\(\alpha \in \left[ 0,1\right] ,\) we have \(g_{ij}^{\mu }\left( \lambda \right) =\alpha \cdot \lambda \) for all \(\lambda \in \mathbb {R}.\) In the following, we use the mappings \(g_{ij}^{\mu }\) in order to derive the parameter \(\alpha \) from \(\varphi .\) We proceed by a number of subclaims. First, we show that \(g_{ij}^{\mu }\) does not depend on the choice of \(j\in N{\setminus }\left\{ i\right\} .\)
Claim C1a. For \(i\in N\) and \(\lambda ,\mu \in \mathbb {R},\) we have \(\varphi _{i} ( e^{\mu }+\lambda \cdot e^{ij}) =\varphi _{i}( e^{\mu }+\lambda \cdot e^{ik}) \) for all \(j,k\in N{\setminus }\left\{ i\right\} .\)
For \(j=k,\) nothing is to show. Let now \(j\ne k\ \)and \(\ell \in N{\setminus } \left\{ i,j,k\right\} .\) Player \(\ell \) is a dummy player in \(e^{\mu } +\lambda \cdot e^{ij}\in \mathbb {\bar{V}}\) and in \(e^{\mu }+\lambda \cdot e^{ik}\in \mathbb {\bar{V}}\) with \(\left( e^{\mu }+\lambda \cdot e^{ij}\right) _{\ell }=\frac{\mu }{n}=\frac{1}{n}\cdot \left( e^{\mu }+\lambda \cdot e^{ij}\right) \left( N\right) \) and \(\left( e^{\mu }+\lambda \cdot e^{ik}\right) _{\ell }=\frac{\mu }{n}=\frac{1}{n}\cdot \left( e^{\mu } +\lambda \cdot e^{ik}\right) \left( N\right) .\) By AD, we have (\(\dagger \)) \(\varphi _{\ell }\left( e^{\mu }+\lambda \cdot e^{ij}\right) =\varphi _{\ell }\left( e^{\mu }+\lambda \cdot e^{ik}\right) .\) Since i and \(\ell \) are symmetric in \(\lambda \cdot e^{ij}-\lambda \cdot e^{ik},\) players i and \(\ell \), \(e^{\mu }+\lambda \cdot e^{ij},\) and \(e^{\mu }+\lambda \cdot e^{ik}\) satisfy the hypothesis of DMo\(^{-}.\) Hence, DMo\(^{-}\) and (\(\dagger \)) imply \(\varphi _{i}\left( e^{\mu }+\lambda \cdot e^{ij}\right) =\varphi _{i}\left( e^{\mu }+\lambda \cdot e^{ik}\right) \) . \(\square \)
For \(i\in N\) and \(\mu \in \mathbb {R},\) we define \(g_{i}^{\mu }:\mathbb {R} \rightarrow \mathbb {R}\) by
By (A.4) and C1a, \(g_{ij}^{\mu }\) does not depend on the choice of \(j\in N{\setminus }\left\{ i\right\} .\) Hence, \(g_{i}^{\mu }\) is well-defined. By AD, we have \(g_{i}^{\mu }\left( 0\right) =0.\) Next, we show that \(g_{i}^{\mu }\) does not depend on the choice of \(i\in N.\)
Claim C1b. For all \(i,j\in N,\)\(i\ne j\) and \(\lambda ,\mu \in \mathbb {R},\) we have \(g_{i}^{\mu }\left( \lambda \right) =g_{j}^{\mu }\left( \lambda \right) .\)
For \(k\in N{\setminus }\left\{ i,j\right\} ,\) we have
where the second equation drops from \(\lambda \cdot e^{ik}=-\lambda \cdot e^{ki}.\) By AD, we have \(\varphi _{\ell }( e^{\mu }+\lambda \cdot e^{ik}) =\frac{\mu }{n}\) for all \(\ell \in N{\setminus } \{ i,k \} .\) Hence, E entails \(\varphi _{i} ( e^{\mu } +\lambda \cdot e^{ik}) +\varphi _{k} ( e^{\mu }+\lambda \cdot e^{ik}) =\frac{2\mu }{n}.\) Together with (A.6), we obtain \(g_{i}^{\mu } ( \lambda ) +g_{k}^{\mu } ( -\lambda ) =0.\) Analogously, one shows \(g_{j}^{\mu } ( \lambda ) +g_{k}^{\mu } ( -\lambda ) =0,\) which concludes the proof. \(\square \)
For \(\mu \in \mathbb {R},\) we define \(g^{\mu }:\mathbb {R}\rightarrow \mathbb {R}\) by \(g^{\mu }=g_{i}^{\mu }\) for some \(i\in N.\) By C1b, \(g^{\mu }\) is well-defined. In the following, we show certain properties of the mappings \(g^{\mu }\) and their relation to \(\varphi .\) For later use, we first show that \(g^{\mu }\) is odd.
Claim C1c. For all \(\lambda ,\mu \in \mathbb {R},\) we have \(g^{\mu }\left( \lambda \right) =-g^{\mu }\left( -\lambda \right) .\)
For \(i,j\in N,\)\(i\ne j,\) we have
where the second equation drops from \(\lambda \cdot e^{ij}=-\lambda \cdot e^{ji}.\) By AD, we have \(\varphi _{\ell }\left( e^{\mu }+\lambda \cdot e^{ij}\right) =0\) for all \(\ell \in N{\setminus }\left\{ i,j\right\} .\) Hence, E entails \(\varphi _{i}\left( e^{\mu }+\lambda \cdot e^{ij}\right) +\varphi _{j}\left( e^{\mu }+\lambda \cdot e^{ij}\right) =\frac{2\mu }{n}.\) Together with (A.7), this proves the claim. \(\square \)
For later use, we show a technical relation between the mappings \(g^{\mu }\) and \(\varphi .\) Note that in view of AD, average players in \(x\in \mathbb {R}_{\mu }^{N}\), i.e., players i with \(x_{i}=\frac{\mu }{n}\) are of particular interest. For \(\mu \in \mathbb {R},\) set \(\mathbb {R}_{\mu } ^{N}=\left\{ x\in \mathbb {R}^{N}\mid \sum _{\ell \in N}x_{\ell }=\mu \right\} \) and \(\mathbb {\bar{R}}_{\mu }^{N}:=\left\{ x\in \mathbb {R}_{\mu }^{N} \mid \text {there exist some }i\in N\right. \left. \text {such that }x_{i}=\frac{\mu }{n}\right\} .\) For \(x\in \mathbb {\bar{R}}_{\mu }^{N},\) we set \(C_{\mu }\left( x\right) :=\left\{ i\in \mathbb {N}_{n}\mid x_{i}\ne \frac{\mu }{n}\right\} .\)
Claim C1d. For all \(\mu \in \mathbb {R},\)\(x\in \mathbb {\bar{R} }_{\mu }^{N},\) and \(i\in N\), we have \(\varphi _{i}\left( x\right) =g^{\mu }\left( x_{i}-\frac{\mu }{n}\right) +\frac{\mu }{n}.\)
We proceed by induction on \(\left| C_{\mu }\left( x\right) \right| .\)
Induction base: For \(\left| C_{\mu }\left( x\right) \right| =0,\)AD entails \(\varphi _{i}\left( x\right) =\frac{\mu }{n}=g^{\mu }\left( 0\right) +\frac{\mu }{n}=g^{\mu }\left( x_{i}-\frac{\mu }{n}\right) +\frac{\mu }{n}\) for all \(i\in N.\) Note that \(\left| C_{\mu }\left( x\right) \right| \ne 1\) for all \(x\in \mathbb {\bar{R}}_{\mu }^{N}\). If \(\left| C_{\mu }\left( x\right) \right| =2,\) then there are \(i,j\in N,\)\(i\ne j\) such that \(x=e^{\mu }+\left( x_{i}-\frac{\mu }{n}\right) \cdot e^{ij}=e^{\mu }+\left( x_{j} -\frac{\mu }{n}\right) \cdot e^{ji}.\) By (A.4) and (A.5 ), we have \(\varphi _{i}\left( x\right) =g^{\mu }\left( x_{i}-\frac{\mu }{n}\right) +\frac{\mu }{n}\) and \(\varphi _{j}\left( x\right) =g^{\mu }\left( x_{j}-\frac{\mu }{n}\right) +\frac{\mu }{n}.\) Moreover, for \(k\in N{\setminus }\left\{ i,j\right\} ,\)AD implies \(\varphi _{k}\left( x\right) =\frac{\mu }{n}=g^{\mu }\left( 0\right) +\frac{\mu }{n}=g^{\mu }\left( x_{k}-\frac{\mu }{n}\right) +\frac{\mu }{n}.\)
Induction hypothesis: Let the claim hold for all \(x\in \mathbb {\bar{R}}_{\mu }^{N}\) such that \(\left| C_{\mu }\left( x\right) \right| \le t,\)\(t\in \mathbb {N}.\)
Induction step: Let \(x\in \mathbb {\bar{R}}_{\mu }^{N}\) be such that \(\left| C_{\mu }\left( x\right) \right| =t+1>2.\) Suppose \(\varphi _{i}\left( x\right) \ne g^{\mu }\left( x_{i}-\frac{\mu }{n}\right) +\frac{\mu }{n}\) for some \(i\in \mathbb {N}_{n}.\) By AD, \(i\in C_{\mu }\left( x\right) .\) Let \(j,k\in C_{\mu }\left( x\right) {\setminus }\left\{ i\right\} ,\)\(j\ne k,\) and \(y=x-\left( x_{j}-\frac{\mu }{n}\right) \cdot e^{jk}.\) Note that \(y\in \mathbb {\bar{R}}_{\mu }^{N},\)\(\left| C_{\mu }\left( y\right) \right| \le t,\) and \(\left| C_{\mu }\left( y\right) \right| \ne 1.\) By the induction hypothesis, we have \(\varphi _{i}\left( y\right) =g^{\mu }\left( y_{i}-\frac{\mu }{n}\right) +\frac{\mu }{n}=g^{\mu }\left( x_{i}-\frac{\mu }{n}\right) +\frac{\mu }{n}.\) By assumption, there exists \(\ell \in N{\setminus } C_{\mu }\left( x\right) \) such that \(x_{\ell }=\frac{\mu }{n}.\) Hence, we obtain \(\varphi _{i}\left( x\right) -\varphi _{i}\left( y\right) \ne g^{\mu }\left( x_{i}-\frac{\mu }{n}\right) +\frac{\mu }{n}-g^{\mu }\left( x_{i}-\frac{\mu }{n}\right) -\frac{\mu }{n}=0\) and \(\varphi _{\ell }\left( x\right) -\varphi _{\ell }\left( y\right) =0,\) where the latter drops from AD. Since i and \(\ell \) are symmetric in \(x-y,\)x, y, i, and j satisfy the hypothesis of DMo\(^{-}.\) Hence, this contradicts DMo\(^{-}\). \(\square \)
For later use, we show crucial properties of the mappings \(g^{\mu },\) where linearity is of particular importance.
Claim C1e. For all \(\mu \in \mathbb {R},\) the mapping \(g^{\mu }:\mathbb {R}\rightarrow \mathbb {R}\) is linear and monotonic.
We show that the mapping \(g^{\mu }\) is additive and monotonic. Then, linearity drops from Aczél (1966, Theorem 1).
Additivity: Let \(a,b\in \mathbb {R}.\) Let \(i,j,k\in N\) and \(x\in \mathbb {R}_{\mu }^{N}\) be such that \(i\ne j,\)\(j\ne k,\)\(k\ne i\),
Since \(n>3,\)\(x\in \mathbb {\bar{R}}_{\mu }^{N}.\) By C1d, we have
Further, by AD, we have \(\varphi _{\ell }\left( x\right) =\frac{\mu }{n}\) for all \(\ell \in N{\setminus }\left\{ i,j,k\right\} .\) Hence, we obtain
That is, the mapping g is additive.
Monotonicity: Let \(a,b\in \mathbb {R}\) and \(i,j,k\in N\) be such that \(i\ne j,\)\(j\ne k,\)\(k\ne i,\) and \(a\ge b.\) For \(x=e^{\mu }+a\cdot e^{ij}\) and \(y=e^{\mu }+b\cdot e^{ij},\) we have \(x_{i}-y_{i}=a\ge b=x_{k}-y_{k}.\) Moreover, by AD, \(\varphi _{k}\left( x\right) =\varphi _{k}\left( y\right) =\frac{\mu }{n}.\) Hence, we obtain
That is, the mapping \(g^{\mu }\) is monotonic. \(\square \)
For \(\mu \in \mathbb {R},\) set \(\alpha ^{\mu }:=g^{\mu }\left( 1\right) .\) The next claim already establishes C1 for all \(x\in \mathbb {R}_{\mu }^{N}.\)
Claim C1f. For all \(\mu \in \mathbb {R}\) and \(x\in \mathbb {R}_{\mu }^{N},\) we have \(\varphi \left( x\right) =\alpha ^{\mu }\cdot x+\left( 1-\alpha ^{\mu }\right) \cdot e^{\mu }.\)
Case (i): For \(x\in \mathbb {\bar{R}}_{\mu }^{N}\) and \(i\in N,\) we obtain
Case (ii): Let \(x\in \mathbb {R}_{\mu }^{N} {\setminus }\mathbb {\bar{R}}_{\mu }^{N}.\) Suppose \(\varphi \left( x\right) \ne \alpha ^{\mu }\cdot x+\left( 1-\alpha ^{\mu }\right) \cdot e^{\mu }.\) By E, we have \(\sum _{\ell \in N}\varphi _{\ell }\left( x\right) =\mu =\sum _{\ell \in N}\left[ \alpha ^{\mu }\cdot x_{\ell }+\left( 1-\alpha ^{\mu }\right) \cdot \frac{\mu }{n}\right] .\) Hence, there are \(i,j\in N\) such that \(\varphi _{i}\left( x\right) >\alpha ^{\mu }\cdot x_{i}+\left( 1-\alpha ^{\mu }\right) \cdot \frac{\mu }{n}\) and \(\varphi _{j}\left( x\right) <\alpha ^{\mu }\cdot x_{j}+\left( 1-\alpha ^{\mu }\right) \cdot \frac{\mu }{n}.\) Let \(k,\ell \in N{\setminus }\left\{ i,j\right\} ,\) and \(y=x-\left( x_{k}-\frac{\mu }{n}\right) \cdot e^{k\ell }.\) Note that \(x_{i}=y_{i}\) and \(x_{j}=y_{j}.\) Further, note that \(y_{k}=\frac{\mu }{n}\) and therefore \(y\in \mathbb {\bar{R} }_{\mu }^{N}.\) By Case (i), we obtain \(\varphi \left( y\right) =\alpha ^{\mu }\cdot y+\left( 1-\alpha ^{\mu }\right) \cdot e^{\mu }.\) Moreover, we have
and
Since i and j are symmetric in \(x-y,\)x, y, i, and j satisfy the hypothesis of DMo\(^{-}.\) Hence, this contradicts DMo\(^{-} \). \(\square \)
Now, we are ready to prove C1.
Case (a): Suppose \(\alpha ^{\mu }=0\) for all \(\mu \in \mathbb {R}.\) By C1f, we obtain \(\varphi _{i}\left( x\right) =\frac{1}{n}\cdot \sum _{\ell \in N}x_{\ell }\overset{\text {(A.3)} }{=}\mathrm {Sh}^{0}\left( x\right) \) for all \(x\in \mathbb {R}^{N}.\)
Case (b): Suppose \(\alpha ^{\bar{\mu }}\ne 0\) for some \(\bar{\mu }\in \mathbb {R}.\) By C1e, we have \(\alpha ^{\mu }>0.\) Set \(\alpha :=\alpha ^{\bar{\mu }}.\) We show that
Suppose there exists some \(x\in \mathbb {R}^{N}\) such that (A.8) fails for some \(i\in N\). By C1f, \(x\notin \mathbb {R}_{\bar{\mu }}^{N}.\) By E and w.l.o.g., there exists \(j\in N{\setminus }\left\{ i\right\} \) such that
and
Let
Further, let \(k\in N{\setminus }\left\{ i,j\right\} \) and let \(y\in \mathbb {R}^{N}\) be given by
for all \(\ell \in N{\setminus }\left\{ i,j,k\right\} .\) Since \(y\in \mathbb {R}_{\bar{\mu }}^{N},\) by C1f, we have \(\varphi \left( y\right) =\alpha \cdot y+\left( 1-\alpha \right) \cdot e^{\bar{\mu }}.\) By (A.9), (A.10), (A.11), and (A.12), we obtain
and
Since i and j are symmetric in \(x-y,\)x, y, i, and j satisfy the hypothesis of DMo\(^{-}.\) Therefore, this contradicts DMo\(^{-}\). Hence, \(\varphi \left( x\right) \overset{\text {(A.3)} }{=}\mathrm {Sh}^{\alpha }\left( x\right) \) for all \(x\in \mathbb {R}^{N}.\)
Finally, we have
which concludes the proof of C1. \(\square \)
If \(\left| \mathcal {T}_{>1}\left( v\right) \right| =1\) for \(v\in \mathbb {V},\) then there are \(\delta ^{v}\in \mathbb {R}^{N}\) and \(\beta ^{v}\in \mathbb {R}\), and \(T^{v}\subseteq N,\)\(\left| T^{v}\right| >1\) such that \(\beta ^{v}\ne 0\) and
Set
Note that players \(i\in N{\setminus }\left( R^{v}\cup T^{v}\right) \) are average dummy players, i.e., dummy players with \(v\left( \left\{ i\right\} \right) =\frac{1}{\left| N\right| }\cdot v\left( N\right) \) for which AD implies \(\varphi \left( v\right) =v\left( \left\{ i\right\} \right) .\)
We now show that \(\varphi \left( v\right) =\mathrm {Sh}^{\alpha }\left( v\right) \) for all \(v\in \mathbb {V}\) with \(\left| \mathcal {T}_{>1}\left( v\right) \right| =1\) by a number of claims. First, we deal with games in which there exists a average dummy player.
Claim 2, C2: For all \(v\in \mathbb {V}\) with \(\left| \mathcal {T}_{>1}\left( v\right) \right| =1\) and such that \(R^{v}\cup T^{v}\ne N,\) we have \(\varphi \left( v\right) =\mathrm {Sh}^{\alpha }\left( v\right) .\)
By C1, we have
For \(i\in N{\setminus }\left( R^{v}\cup T^{v}\right) ,\) we have
Since all players in \(N{\setminus } T^{v}\) are pairwise symmetric in \(-\beta ^{v}\cdot \bar{u}_{T^{v}}\), v, \(v-\beta ^{v}\cdot \bar{u}_{T^{v}},\) and \(i,\ell \in N{\setminus } T^{v}\) satisfy the hypothesis of DMo\(^{-}.\) Hence, we have
Since any two players in \(T^{v}\) are pairwise symmetric in \(-\beta ^{v} \cdot \bar{u}_{T^{v}}\), v, \(v-\beta ^{v}\cdot \bar{u}_{T^{v}},\) and \(k,\ell \in T^{v}\) satisfy the hypothesis of DMo\(^{-},\) which implies that we have
for all \(k,\ell \in T^{v}.\) By E, (A.16), and (A.17 ), we finally have \(\varphi \left( v\right) =\mathrm {Sh}^{\alpha }\left( v\right) \). \(\square \)
Next, we handle games in which there are average players but which are not dummy players.
Claim 3, C3: For all \(v\in \mathbb {V}\) with \(\left| \mathcal {T}_{>1}\left( v\right) \right| =1\) such that \(R^{v}\cup T^{v}=N\) and \(\left| T^{v}{\setminus } R^{v}\right| \ge 1,\) we have \(\varphi \left( v\right) =\mathrm {Sh}^{\alpha }\left( v\right) .\)
Suppose \(\varphi \left( v\right) \ne \mathrm {Sh}\left( v\right) \). By E, there are \(i,j\in N\) such that
Case (i): Suppose \(i,j\in R^{v}{\setminus } T^{v}\) or \(i,j\in T^{v}.\) By (A.13), we have \(v-\beta ^{v}\cdot \bar{u}_{T^{v} }\in \mathbb {\bar{V}}.\) Hence, C1 implies
By (A.18) and (A.19), we further have
and
Since i and j are symmetric in \(-\beta ^{v}\cdot \bar{u}_{T^{v}},\)v, \(v-\beta ^{v}\cdot \bar{u}_{T^{v}},\)i, and j satisfy the hypothesis of DMo\(^{-}.\) Hence, this contradicts DMo\(^{-}\).
Case (ii): Suppose, w.l.o.g., \(i\in R^{v}{\setminus } T^{v}\) and \(j\in T^{v}.\)
Case (ii-a): Suppose \(j\in T^{v}{\setminus } R^{v}.\) Let \(w:=v-\beta ^{v}\cdot \bar{u}_{T^{v}}-\beta ^{v}\cdot \bar{u}_{\left( T^{v}{\setminus }\left\{ j\right\} \right) \cup \left\{ i\right\} }.\) By C2, we have
By (A.18) and (A.20), we further have
and
Since i and j are symmetric in \(-\beta ^{v}\cdot \bar{u}_{T^{v}}-\beta ^{v}\cdot \bar{u}_{\left( T^{v}{\setminus }\left\{ j\right\} \right) \cup \left\{ i\right\} },\)v, w, i, and j satisfy the hypothesis of DMo\(^{-}.\) Hence, this contradicts DMo\(^{-}\).
Case (ii-b): Suppose \(j\in T^{v}\cap R^{v}.\) By assumption, there exists \(k\in T^{v}{\setminus } R^{v}\) such that \(k\ne i\) and \(k\ne j.\) By C1, we have
Since j and k are symmetric in \(-\beta ^{v}\cdot \bar{u}_{T^{v}},\)v, \(v-\beta ^{v}\cdot \bar{u}_{T^{v}},\)i, and k satisfy the hypothesis of DMo\(^{-}.\) Hence, DMo\(^{-}\) entails
Since
we obtain
Let \(z:=v-\beta ^{v}\cdot \bar{u}_{T^{v}}-\beta ^{v}\cdot \bar{u}_{\left( T^{v}{\setminus }\left\{ k\right\} \right) \cup \left\{ i\right\} }.\) By (A.18), (A.22), and C2, we have
and
Since \(i\in R^{v}{\setminus } T^{v}\) and \(k\in T^{v}{\setminus } R^{v},\)i and k are symmetric in \(-\beta ^{v}\cdot \bar{u}_{T^{v}}-\beta ^{v}\cdot \bar{u}_{\left( T^{v}{\setminus }\left\{ k\right\} \right) \cup \left\{ i\right\} },\)v, z, i, and k satisfy the hypothesis of DMo\(^{-}.\) Hence, (A.23) and (A.24) contradict DMo\(^{-}\).\(\square \)
Finally, we deal with games in which there are no average players.
Claim 4, C4: For all \(v\in \mathbb {V}\) with \(\left| \mathcal {T}_{>1}\left( v\right) \right| =1\) such that \(R^{v}\cup T^{v}=N\) and \(\left| T^{v}{\setminus } R^{v}\right| =0,\) we have \(\varphi \left( v\right) =\mathrm {Sh}^{\alpha }\left( v\right) .\)
By assumption, we have \(R^{v}=N.\) Suppose \(\varphi \left( v\right) \ne \mathrm {Sh}\left( v\right) \). By E, there are \(i,j\in N\) such that
Let \(k\in N{\setminus }\left\{ i,j\right\} \), \(\ell \in N{\setminus }\left\{ i,j,k\right\} ,\) and \(q\in \mathbb {V}\) be given by
By (A.13), we have
Hence, we have \(\left| \mathcal {T}_{>1}\left( v-q\right) \right| =1,\)\(T^{v-q}=T^{v},\) and
where the latter implies \(k\notin R^{v-q}.\) Note that q is constructed in a way such that k is an average player in \(v-q.\)
If \(k\notin T^{v},\) then \(v-q\) satisfies the hypothesis of C2 and we obtain
If \(k\in T^{v},\) then \(v-q\) satisfies the hypothesis of C3 and we also obtain (A.27). By (A.25) and (A.27), we have
and
Since i and j are symmetric in \(-q,\)v, \(v-q,\)i and j satisfy the hypothesis of DMo\(^{-}.\) Hence, (A.28) and (A.29) together contradict DMo\(^{-}\).\(\square \)
Note that the induction basis (see page 8) is proved by C1, C2, C3, and C4.
Induction hypothesis: Let the claim hold for all \(v\in \mathbb {V}\) such that \(\left| \mathcal {T}_{>1}\left( v\right) \right| \le t,\)\(t\in \mathbb {N},\)\(t\ge 1.\)
Induction step: Let now \(v\in \mathbb {V}\) be such that \(\left| \mathcal {T}_{>1}\left( v\right) \right| =t+1.\) There exist \(S,T\in \mathcal {T}_{>1}\left( v\right) \) such that \(S\ne T.\) By (3), (A.2), (*) (see page 8), and the induction hypothesis, we have
Case (i): \(S\cap T\ne \emptyset .\) W.l.o.g., \(S{\setminus } T\ne \emptyset .\) Let \(i\in S\cap T\) and \(j\in S{\setminus } T.\) By (A.30) and DMo\(^{-}\), we have
and therefore
Case (ii): \(S\cup T\ne N.\) W.l.o.g., \(S{\setminus } T\ne \emptyset .\) Let \(i\in N{\setminus }\left( S\cup T\right) \) and \(j\in S{\setminus } T.\) By (A.30) and DMo\(^{-}\), we have
and therefore
Case (iii): \(S\cap T=\emptyset \) and \(S\cup T=N.\) Hence, \(\mathcal {T}_{>1}\left( v\right) =\left\{ S,T\right\} .\) Let \(i\in S\), \(j\in T,\) and \(w\in \mathbb {V}\) be given by
By construction, we have \(\mathcal {T}_{>1}\left( w\right) =\left\{ \left( S{\setminus }\left\{ i\right\} \right) \cup \left\{ j\right\} ,T\right\} \) and (****) \(v\left( N\right) =w\left( N\right) .\) In view of Case (i), we have (*****) \(\varphi \left( w\right) =\mathrm {Sh}\left( w\right) .\)
Since i and j are symmetric in \(v-w,\)v, w, i, and j satisfy the hypothesis of DMo\(^{-}.\) Hence, by DMo\(^{-}\) and (A.30), we have
and therefore
Finally, (A.31), (A.32), (A.34), and E imply \(\varphi \left( v\right) =\mathrm {Sh}^{\alpha }\left( v\right) \).\(\square \)
Counterexample to Theorem 3 for \(\left| N\right| =2\)
Theorem 3 fails for \(\left| N\right| =2.\) Let \(N=\left\{ 1,2\right\} .\) Consider the solution \(\varphi ^{\heartsuit }:\mathbb {V}\rightarrow \mathbb {R}^{2}\) given by
for all \(v\in \mathbb {V}.\) One can easily check that \(\varphi ^{\heartsuit } \ne \mathrm {Sh}^{\alpha }\) for all \(\alpha \in \left[ 0,1\right] \) and that \(\varphi ^{\heartsuit }\) inherits E, AD, and DMo\(^{-}\) from \(\mathrm {Sh}\).
Non-redundancy of Theorem 3 for \(\left| N\right| >3\)
Our characterization is non-redundant for \(\left| N\right| >3\). The value \(\varphi ^{\mathbf{E }}\) given by \(\varphi _{i}^{\mathbf{E } }\left( v\right) =v\left( \left\{ i\right\} \right) \) for all \(v\in \mathbb {V}\) and \(i\in N\) satisfies AD and DMo\(^{-}\) but not E. The strictly positively weighted division values (Béal et al. 2016, Theorem 2) with non-uniform weights satisfy E and DMo\(^{-}\) but not AD. For \(v\in \mathbb {V},\) let \(D_{0}\left( v\right) :=\left\{ i\in N\mid i~\text {is a dummy player in }v\right\} .\) The value \(\varphi ^{\mathbf{DMo }^{-}}\) given by
satisfies E and AD but not DMo\(^{-}\).
Rights and permissions
About this article
Cite this article
Casajus, A., Yokote, K. Weakly differentially monotonic solutions for cooperative games. Int J Game Theory 48, 979–997 (2019). https://doi.org/10.1007/s00182-019-00669-1
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00182-019-00669-1