Skip to main content
Log in

On the core and nucleolus of directed acyclic graph games

  • Full Length Paper
  • Series A
  • Published:
Mathematical Programming Submit manuscript

Abstract

We introduce directed acyclic graph (DAG) games, a generalization of standard tree games, to study cost sharing on networks. This structure has not been previously analyzed from a cooperative game theoretic perspective. Every monotonic and subadditive cost game—including monotonic minimum cost spanning tree games—can be modeled as a DAG-game. We provide an efficiently verifiable condition satisfied by a large class of directed acyclic graphs that is sufficient for the balancedness of the associated DAG-game. We introduce a network canonization process and prove various structural results for the core of canonized DAG-games. In particular, we characterize classes of coalitions that have a constant payoff in the core. In addition, we identify a subset of the coalitions that is sufficient to determine the core. This result also guarantees that the nucleolus can be found in polynomial time for a large class of DAG-games.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8

Similar content being viewed by others

Notes

  1. The proof of Faigle et al. [7] uses reduction to the minimum cover problem and relies heavily on the undirectedness of the edges of the graph. Thus, it does not necessarily indicate that the same difficulties exists in DAG-games. The difference between monotonic and ordinary MCST games can also be of importance here.

  2. Since we analyze the game from a cooperative perspective, we do not formalize the bargaining process that takes place between the players. However, such processes are certainly present. Consider the case of Example 2. Without the shortcut of \(\mathbf {g}'\) player 6 may be forced to pay as much as 10 in the core, while with the shortcut present his core payoff is bounded from above by 7.

  3. For a detailed example of the decomposition process see [25].

  4. Note that the simplifications of Lemmata 9 and 10 alter the Shapley-value (and many other solution concept) of the game. This also sheds a light why the Shapley-value is not a good solution concept in case of DAG-games. It is easy to generate an example where the Shapley-value lies outside the core. Alternatively one could argue that the core conditions are too demanding. It is unrealistic that the residents of free nodes—like player 3 in this example—do not contribute to the construction costs.

  5. If we think about the DAG as a partially ordered set then width is equivalent to the cardinality of the maximum antichain the poset has.

References

  1. Bergantiños, G., Vidal-Puga, J.J.: A fair rule in minimum cost spanning tree problems. J. Econ. Theory 137, 326–352 (2007)

    Article  MathSciNet  MATH  Google Scholar 

  2. Bird, C.: On cost allocation for a spanning tree: a game theoretic approach. Networks 6, 335–350 (1976)

    Article  MathSciNet  MATH  Google Scholar 

  3. Bjørndal, E., Koster, M., Tijs, S.: Weighted allocation rules for standard fixed tree games. Math. Methods Oper. Res. 59(2), 249–270 (2004)

    Article  MathSciNet  MATH  Google Scholar 

  4. Bogomolnaia, A., Moulin, H.: Sharing a minimal cost spanning tree: beyond the folk solution. Games Econ. Behav. 69, 238–248 (2010)

    Article  MathSciNet  MATH  Google Scholar 

  5. Deng, X., Fang, Q., Sun, X.: Finding nucleolus of flow game. J. Comb. Optim. 18(1), 64–86 (2009)

    Article  MathSciNet  MATH  Google Scholar 

  6. Faigle, U., Kern, W., Hochstättler, W., Fekete, S.: On the complexity of testing membership in the core of min-cost spanning tree games. Int. J. Game Theory 26(3), 361–366 (1997)

    Article  MathSciNet  MATH  Google Scholar 

  7. Faigle, U., Kern, W., Kuipers, J.: Computing the nucleolus of min-cost spanning tree games is np-hard. Int. J. Game Theory 27, 443–450 (1998)

    Article  MathSciNet  MATH  Google Scholar 

  8. Granot, D., Granot, F.: Computational complexity of a cost allocation approach to a fixed cost spanning forest problem. Math. Oper. Res. 17(4), 765–780 (1992)

    Article  MathSciNet  MATH  Google Scholar 

  9. Granot, D., Huberman, G.: Minimum cost spanning tree games. Math. Program. 21, 1–18 (1981)

    Article  MathSciNet  MATH  Google Scholar 

  10. Granot, D., Huberman, G.: On the core and nucleolus of minimum cost spanning tree games. Math. Program. 29(3), 323–347 (1984)

    Article  MathSciNet  MATH  Google Scholar 

  11. Granot, D., Maschler, M.: Spanning network games. Int. J. Game Theory 27, 467–500 (1998)

    Article  MathSciNet  MATH  Google Scholar 

  12. Granot, D., Granot, F., Zhu, W.R.: Characterization sets for the nucleolus. Int. J. Game Theory 27(3), 359–374 (1998)

    Article  MathSciNet  MATH  Google Scholar 

  13. Huberman, G.: The nucleolus and essential coalitions. In: Bensoussan, A., Lions, J.L. (eds.) Analysis and Optimization of Systems, Lecture Notes in Control and Information Sciences, vol. 28, pp. 416–422. Elsevier B.V., Amsterdam (1980)

    Google Scholar 

  14. Hwang, F.K., Richards, D.S., Winter, P.: The Steiner Tree Problem, Annals of Discrete Mathematics, vol. 53. Elsevier, North-Holland (1992)

  15. Kuipers, J.: Minimum cost forest games. Int. J. Game Theory 26, 367–377 (1997)

    Article  MathSciNet  MATH  Google Scholar 

  16. Maschler, M., Peleg, B., Shapley, L.: Geometric properties of the kernel, nucleolus and related solution concepts. Math. Oper. Res. 4, 303–338 (1979)

    Article  MathSciNet  MATH  Google Scholar 

  17. Maschler, M., Potters, J., Reijnierse, H.: The nucleolus of a standard tree game revisited: a study of its monotonicity and computational properties. Int. J. Game Theory 39(1–2), 89–104 (2010)

    Article  MathSciNet  MATH  Google Scholar 

  18. Megiddo, N.: Computational complexity of the game theory approach to cost allocation for a tree. Math. Oper. Res. 3(3), 189–196 (1978)

    Article  MathSciNet  MATH  Google Scholar 

  19. Potters, J.A.M., Sudhölter, P.: Airport problems and consistent allocation rules. Math. Soc. Sci. 38, 83–102 (1999)

    Article  MathSciNet  MATH  Google Scholar 

  20. Reijnierse, H., Potters, J.A.M.: The \({\cal B}\)-nucleolus of tu-games. Games Econ. Behav. 24, 77–96 (1998)

    Article  MathSciNet  MATH  Google Scholar 

  21. Rosenthal, E.C.: The minimum cost spanning forest game. Econ. Lett. 23, 355–357 (1987)

    Article  MathSciNet  MATH  Google Scholar 

  22. Schmeidler, D.: The nucleolus of a characteristic function game. SIAM J. Appl. Math. 17, 1163–1170 (1969)

    Article  MathSciNet  MATH  Google Scholar 

  23. Skorin-Kapov, D., Skorin-Kapov, J.: A note on steiner tree games. Networks 59(2), 215–225 (2012)

    Article  MathSciNet  MATH  Google Scholar 

  24. Solymosi, T., Sziklai, B.: Characterization sets for the nucleolus in balanced games. Oper. Res. Lett. 44, 520–524 (2016)

    Article  MathSciNet  Google Scholar 

  25. Sziklai, B.: On the computation of the nucleolus of cooperative transferable utility games. Ph.d. thesis, Eötös Loránd University, Budapest (2015)

  26. Sziklai B, Fleiner T, Solymosi T (2014) On the core of directed acyclic graph games. IEHAS Discussion Papers MT-DP 2014/18

  27. Trudeau, C.: A new stable and more responsive cost sharing solution for minimum cost spanning tree problems. Games Econ. Behav. 75, 402–412 (2012)

    Article  MathSciNet  MATH  Google Scholar 

  28. van den Nouweland, A., Tijs, S., Maschler, M.: Monotonic games are spanning network games. Int. J. Game Theory 21(4), 419–427 (1993)

    Article  MathSciNet  MATH  Google Scholar 

Download references

Acknowledgments

The authors would like to thank Prof. Tamás Király of ELTE Operations Research Department and the anonymous referees for their valuable comments. Research was funded by OTKA Grants K101224 and K108383, by the Hungarian Academy of Sciences under its Momentum Programme (LD-004/2010) and by the MTA-ELTE Egerváry Research Group. This research was also partially supported by Pallas Athene Domus Scientiae Foundation. The views expressed are those of the authors’ and do not necessarily reflect the official opinion of Pallas Athene Domus Scientiae Foundation.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Balázs Sziklai.

Appendix

Appendix

Theorem 2

Let N be a player set and \({\hat{c}}:2^N \rightarrow \mathbb {R}\) a monotonic, subadditive cost function. There exists a DAG-network \(\mathcal {D}=(G(V,A),\delta )\) and a residency mapping \(\mathcal {R}:N\rightarrow V\) such that

$$\begin{aligned} c_{(\mathcal {D},\mathcal {R})}(S)={\hat{c}}(S) \quad \forall S \subseteq N. \end{aligned}$$

Proof

We will use the characteristic DAG-representation of \({\hat{c}}\). We have to show that the cost of the cheapest trunk of any coalition S equals to \({\hat{c}}(S)\). This is trivially true for the singleton coalitions. Each player—as a singleton—will use its direct connection to the root as any other route could be only more expensive due to the monotonicity of \({\hat{c}}\). Now let S be an arbitrary non-singleton coalition. We may suppose that \(\delta (a_S)>0\), otherwise both \({\hat{c}}(S)\) and \(c_{(\mathcal {D},\mathcal {R})}(S)\) are trivially zero. Let \(T'\) be the trunk whose arc set consists of \(a_S\) and the zero arcs \(\{a^i_S \ | \ i \in S\}\) and whose node set consists of the nodes spanning these arcs. We will prove \(T' \in T_S\). Let \(T \in T_S\) and suppose that \(\{a_{S_1},\dots ,a_{S_k}\} \) are the non-zero arcs of T. If \(S=S_{\ell }\) for some \(\ell \in \{1,\dots ,k\}\), then by deleting the \(a_{S_j}\), \(j\ne \ell \) arcs and the zero arcs that enter \(\mathbf {p}_{S_j}\), \(j\ne \ell \) we can obtain \(T'\). This would imply \(T' \in T_S\) as the cost could only decrease by deleting these arcs. If \(S\subset S_{\ell }\) for some \(\ell \in \{1,\dots ,k\}\), then again we can obtain a weakly cheaper trunk by connecting the members of S via \(a_S\) and deleting all the other \(a_{S_{\ell }}\) arcs. Thus \(S_{\ell } \subset S\) for all \(\ell \in \{1,\dots ,k\}\). Although a player may be connected to more than one \(\mathbf {p}_{S_{\ell }}\) node, in real he only needs one of the \(a_{S_{\ell }}\) arcs to reach the root. Let us assign the players of S to one of the \(a_{S_{\ell }}\) arcs. Let us denote by \(S^{\ell } \subset S\) those players of S that were assigned to \(S_{\ell }\). Note that the \(\{S^{\ell } \ | \ {\ell }\in \{1,\dots ,k\}, S^{\ell }\ne \emptyset \}\) coalitions comprise a partition of S. If \(S^{\ell }=\emptyset \) for some \(\ell \in \{1,\dots ,k\}\) then \(a_{S_{\ell }}\) and the entering zero arcs of \(\mathbf {p}_{S_{\ell }}\) can be deleted. If \(\emptyset \ne S^{\ell } \subset S_{\ell }\) then the players of \(S^{\ell }\) can be reassigned to \(a_{S^{\ell }}\), that is, we can delete \(a_{S_{\ell }}\) and the entering zero arcs of \(\mathbf {p}_{S_{\ell }}\) and construct \(a_{S^{\ell }}\) and the the entering zero arcs of \(\mathbf {p}_{S^{\ell }}\) instead. Due to the monotonicity, the cost can only decrease this way, while all the players of S are still able to reach the root. Let us perform this transformation for all the \(S_{\ell }\) coalitions. We obtain a trunk \(\hat{T} \in T_S\) such that

$$\begin{aligned} A(\hat{T})=\{a_{S^{\ell }} \ | \ \ell \in \{1,\dots ,k\}, S^{\ell }\ne \emptyset \}\cup \{a^i_{S^{\ell }} \ | \ \ell \in \{1,\dots ,k\},S^{\ell }\ne \emptyset ,i \in S\}. \end{aligned}$$

By the subadditivity of \({{\hat{c}}}\)

$$\begin{aligned} {\hat{c}}(S)&\le \sum _{\ell \in \{1,\dots ,k\},S^{\ell }\ne \emptyset }{\hat{c}}(S^{\ell }) \\ \delta (a_S)&\le \sum _{\ell \in \{1,\dots ,k\},S^{\ell }\ne \emptyset }\delta (a_{S^{\ell }}) \end{aligned}$$

Thus, \(T'\) is as least as cheap as \(\hat{T}\), from which \(T' \in T_S\) follows. \(\square \)

Lemma 1

The node set of every trunk that corresponds to a coalition \(S\subset N\) can be obtained by deleting some branches from V. The removed branches can be chosen in such way that each of them originates from a passage. Formally for any \(S \subset N\) there exists \(Q_1,\dots , Q_k \subset V\) and \(\mathbf {p}_1,\dots , \mathbf {p}_k \in V\) such that

$$\begin{aligned} V(T_S)=V{\setminus }\cup _{j=1}^kB_{\mathbf {p}_j}^{Q_j}, \end{aligned}$$

where \(\mathbf {p}_j\) is a passage for all \(j \in \{1,2,\dots ,k\}\).

Proof

Any trunk T has a representation where V(T) is obtained by removing branches from V (any single node is a branch in itself if we trim all its children). The only thing we need to prove is that these branches can be picked in such way that each of them originates from a passage. Let \(\{\mathbf {p}_1,\dots ,\mathbf {p}_{k}\} \subset V{\setminus } V(T_S)\) denote those passages that connect to \(V(T_S)\) from the outside, i.e. for which \(\pi (\mathbf {p}_j)\in V(T_S)\) for all \(j=1,\dots ,k\). Due to the definition of \(T_S\) there exists at least one such passage. Note that any entering zero arc is included in the trunk of S even if no player of S resides there, due to the definition of \(T_S\). If we remove all the \(B_{\mathbf {p}_1},\dots ,B_{\mathbf {p}_{k}}\) branches from V it can happen that we removed some nodes in \(V(T_S)\) as well i.e. \(V{\setminus } (\cup _{j=1}^{k}B_{\mathbf {p}_j}) \subset V(T_S)\). In order to retain all the nodes of \(V(T_S)\) we trim the \(B_{\mathbf {p}_j}\) branches where they intersect with \(V(T_S)\). Let \(Q_j = V(T_S) \cap B_{\mathbf {p}_j}\) then \(B_{\mathbf {p}_j}^{Q_j}\) is a proper branch for any j and \(V(T_S) =V{\setminus } (\cup _{j=1}^{k}B_{\mathbf {p}_j}^{Q_j})\). \(\square \)

Lemma 4

If \(B_{\mathbf {p}}^Q\) is a building block, then \(x(N(B_{\mathbf {p}}^Q))= \tau (B_{\mathbf {p}}^Q,N)\) for any core allocation x.

Proof

Since \(\pi (\mathbf {p})\) is free, it is a junction and \(x(N(\pi (\mathbf {p})))=0\). We know from Theorem 3 that \(exc(N(V{\setminus } B_{\pi (\mathbf {p})}),x)=0\) for any core allocation x. It follows that \(exc(N((V{\setminus } B_{\pi (\mathbf {p})}) \cup \{\pi (\mathbf {p})\}),x)=0\) is also true. With a similar argument as in Lemma 3 it can be shown that \(x(N(B_{\mathbf {p}}^Q))\le \tau (B_{\mathbf {p}}^Q,N)\).

Each node of Q has (at least) two semi-arc-disjoint paths that leads to the root. As \(B_{\mathbf {p}}^Q\) does not contain a free node one of these paths for each node by-passes \(B_{\mathbf {p}}^Q\). We prove this by contradiction. Let \(\mathbf {q} \in Q\) an arbitrary free node. Suppose there exists two semi-arc-disjoint paths in \(T_N\), \(P_1\) and \(P_2\) that leads from \(\mathbf {q}\) to the root and crosses \(B_{\mathbf {p}}^Q\). Let \(\mathbf {q}_1 \in B_{\mathbf {p}}^Q \cap V(P_1)\) be such that there exist no other \(\mathbf {q}' \in B_{\mathbf {p}}^Q \cap V(P_1)\) such that \(\mathbf {q}' \prec \mathbf {q}_1\). Similarly let \(\mathbf {q}_2\) be the node closest to the root that is an element of both \(B_{\mathbf {p}}^Q\) and \(P_2\). As \(\mathbf {q}_1\) and \(\mathbf {q}_2\) lie on semi-arc-disjoint paths, one of them—say \(\mathbf {q}_1\)—is not \(\mathbf {p}\). Thus the \(P_1\) path leaves the \(B_{\mathbf {p}}^Q\) node set at \(\mathbf {q}_1\) on a zero-arc. There leads a path in \(T_N\) from \(\mathbf {q}_1\) to \(\pi (\mathbf {p})\) through \(B_{\mathbf {p}}^{Q}\) that is arc-disjoint of \(P_1\). As \(\pi (\mathbf {p})\) is free there leads two semi-arc-disjoint paths \(P_3\) and \(P_4\) from \(\pi (\mathbf {p})\) to the root. Without loss of generality we may assume that \(P_1\) intersects with \(P_3\) first (or at the same time as it intersects with \(P_4\)). Let us denote this node by \(\mathbf {q}^*\). Note that if \(\mathbf {q}^*\) is a common node of \(P_3\) and \(P_4\) it is a junction, otherwise the two paths would not be semi-arc-disjoint. Let \(P_A\) be the path that starts from \(\mathbf {q}_1\), follows \(P_1\) till \(\mathbf {q}^*\), then reaches the root following \(P_3\). Let \(P_B\) be the path that originates at \(\mathbf {q}_1\),reaches \(\pi (\mathbf {p})\) using only \(T_N\)-arcs and nodes from \(B_{\mathbf {p}}^Q\), and goes to the root following \(P_4\). By construction \(P_A\) and \(P_B\) are semi-arc-disjoint, thus \(\mathbf {q}_1\) is free, which contradicts the assumption that \(B_{\mathbf {p}}^Q\) is a building block.

It follows that there exists a path in \(T_N\) for every \(\mathbf {q} \in Q\) that leads to the root, that does not pass through any node of \(B_{\mathbf {p}}^Q\). A straightforward consequence is that \(B_{\mathbf {p}}^Q\) is a proper branch and every node in \(V {\setminus } B_{\mathbf {p}}^Q\) can reach the root by using only \(T_N\)-arcs. Note that there is no zero-arc that leaves \(B_{\mathbf {p}}^Q\) and enters in \(V{\setminus } B_{\mathbf {p}}^Q\), otherwise \(B_{\mathbf {p}}^Q\) would contain a free node. Thus the node set \(V {\setminus } B_{\mathbf {p}}^Q\) corresponds to a trunk, namely to \(T_{N(V {\setminus } B_{\mathbf {p}}^Q)}\). Finally, for any core allocation x

$$\begin{aligned} c_{\mathcal {D}}(N)-x(N)&=[c_{\mathcal {D}}(N(V{\setminus } B_{\mathbf {p}}^Q))-x(N(V{\setminus } B_{\mathbf {p}}^Q))]+[\tau (B_{\mathbf {p}}^Q,N)-x(N(B_{\mathbf {p}}^Q))] \\ 0&=[exc(N(V{\setminus } B_{\mathbf {p}}^Q)),x)]+[\tau (B_{\mathbf {p}}^Q,N)-x(N(B_{\mathbf {p}}^Q))] \end{aligned}$$

Both expressions in the square brackets are non-negative, thus \(x(N(B_{\mathbf {p}}^Q))= \tau (B_{\mathbf {p}}^Q,N)\). \(\square \)

Lemma 5

Let \(\cup _{j=1}^k B_{\mathbf {p}_j}^{F_j}\) be a union of branches such that \(\mathbf {p}_j\) is a passage, \(\pi (\mathbf {p}_j) \in F\) and \(F_j \subset F\) for \(j=1,\dots ,k\). Then \(\cup _{j=1}^k B_{\mathbf {p}_j}^{F_j}\) can be decomposed into a disjoint union of building blocks and free nodes.

Proof

The proof proceeds by induction on the number of nodes. If \(\cup _{j=1}^k B_{\mathbf {p}_j}^{F_j}\) consist of a single node, then \(k=1\) and \(B_{\mathbf {p}_1}^{F_1}\) must be a building block. Now suppose the lemma is true for node sets with less than l nodes and let \(|\cup _{j=1}^k B_{\mathbf {p}_j}^{F_j}|=l\). Let \(B_{\mathbf {p}_1}^Q\) be a branch where \(Q=B_{\mathbf {p}_1} \cap F\) and let \(B_{\mathbf {p}_1}^{Q'}\) be the standard form of this branch. Note that \(B_{\mathbf {p}_1}^{Q'}\) is a building block and it is a subset of \(B_{\mathbf {p}_1}^{F_1}\). Let us delete \(B_{\mathbf {p}_1}^{Q'}\) from \(B_{\mathbf {p}_1}^{F_1}\). If \(Q'\cap B_{\mathbf {p}_1}^{F_1}\) is not empty we delete those nodes too (these are free as all the nodes of \(Q'\) are free). If some descendant of a node in \(Q'\) is a junction then it is free therefore it can be deleted too. If we deleted all the free nodes in this way and there are still some nodes in \(B_{\mathbf {p}_1}^{F_1}\) then those must be passages. Let us denote these by \(\mathbf {p}_1',\dots ,\mathbf {p}_K'\). Note that \(\pi (\mathbf {p}_1'),\dots ,\pi (\mathbf {p}_K') \in F\). Hence the remaining nodes can be written as \(\cup _{i=1}^K B_{\mathbf {p}_i'}^{F_1}\cup _{j=2}^k B_{\mathbf {p}_j}^{F_j}\). By reindexing \(\mathbf {p}_i'\) we are done as \(|\cup _{i=1}^K B_{\mathbf {p}_i'}^{F_1}\cup _{j=2}^k B_{\mathbf {p}_j}^{F_j}|<l\). \(\square \)

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Sziklai, B., Fleiner, T. & Solymosi, T. On the core and nucleolus of directed acyclic graph games. Math. Program. 163, 243–271 (2017). https://doi.org/10.1007/s10107-016-1062-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10107-016-1062-y

Keywords

Navigation