Skip to main content
Log in

Dynamic Network Formation with Ordered Partitioning and Incomplete Information

  • Published:
Dynamic Games and Applications Aims and scope Submit manuscript

Abstract

We introduce a dynamic network formation model with incomplete information and asymmetric players who are partitioned into ordered groups. At each stage, a pair of players is randomly selected to update the link connecting them. The feature of the model is that the cost of a link depends on the distance between the labels of the groups players belong to. As a result, information about the other player’s label is significant in the decision to form a link. Assuming incomplete information, players initially do not know each other’s labels, but learn them upon being connected. Comparing the topologies of stable networks for cases of complete and incomplete information indicates that the presence of the latter can have deep implications. Furthermore, our simulation experiments are focused on three issues: the stabilization time, the heterogeneity index, and the impact of the group size gap.

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
Fig. 9
Fig. 10
Fig. 11
Fig. 12
Fig. 13
Fig. 14
Fig. 15
Fig. 16
Fig. 17
Fig. 18

Similar content being viewed by others

Data Availability

Not applicable.

Notes

  1. We thank the Reviewer for this interesting suggestion.

  2. We produce 100 simulations. Parameters used in these simulations are completely identical to control for factors other than players’ knowledge. The parameters are \(m=3\), \(f(0)=\frac{1}{2}\), \(f(1)=\frac{8}{5}\), \(f(2)=3\), and \(E_i=0.95\) for \(i\in N\). Moreover, we assume that players are equally distributed in three groups.

  3. Here we mean the first-order stochastic domination principle by the following. Random variable A first-order stochastic dominates B if \(F_A(x) \le F_B(x)\) for all x, with strict inequality at some x. In our model, the stochastic dominance of stabilization time under incomplete information indicates the shift of its values to larger values in comparison with the case of complete information.

  4. We use 100 simulations to find the average result. Parameters used for these simulations are the same avoiding the impact of other factors on the results. The parameters are as follows: \(m=3\), \(f(0)=\frac{1}{2}, f(1)=\frac{8}{5}, f(2)=\frac{13}{5}\), and \(E_i=0.95\) for any \(i\in N\). Players are equally distributed in three groups. Moreover, in counting the number of links and heterogeneous links, we keep the agents symmetric, i.e., they only differ by their group labels, so that the result reflects the number of links connecting players in various groups rather than the number of permutations.

  5. In each experiment, we use 100 simulations to find the average result. Parameters used in these simulations are the same avoiding the impact of other factors. They are: \(f(0)=\frac{3}{5}, f(1)=\frac{9}{5}\), and \(E_i=0.95\) for any \(i\in N\).

References

  1. Arnold TB, Emerson JW (2011) Nonparametric goodness-of-fit tests for discrete null distributions. R J 3(2):34–39

    Article  Google Scholar 

  2. Aumann RJ, Myerson RB (1988) Endogenous formation of links between players and coalitions: an application of the Shapley value. In: Roth A (ed) The Shapley value. Cambridge Univ. Press, Cambridge, pp 175–191

    Chapter  Google Scholar 

  3. Bala V, Goyal S (2000) A noncooperative model of network formation. Econometrica 68:1181–1229

    Article  MathSciNet  Google Scholar 

  4. Castaldo M, Catalano C, Como G, Fagnani F (2020) On a centrality maximization game. IFAC-Pap 53(2):2844–2849

    Google Scholar 

  5. Cominetti R, Quattropani M, Scarsini M (2022) The buck-passing game. Math Oper Res 47(3):1731–1756

    Article  MathSciNet  Google Scholar 

  6. Como G, Fagnani F, Proskurnikov AV (2022) Reaching optimal distributed estimation through myopic self-confidence adaptation. IFAC-Pap 55(30):442–447

    Google Scholar 

  7. Currarini S, Jackson MO, Pin P (2009) An economic model of friendship: homophily, minorities, and segregation. Econometrica 77(4):1003–1045

    Article  MathSciNet  Google Scholar 

  8. Currarini S, Jackson MO, Pin P (2010) Identifying the roles of race-based choice and chance in high school friendship network formation. Proc Natl Acad Sci 107(11):4857–4861

    Article  CAS  PubMed  PubMed Central  ADS  Google Scholar 

  9. Currarini S, Mengel F (2016) Identity, homophily and in-group bias. Eur Econ Rev 90:40–55

    Article  Google Scholar 

  10. de Kerchove C, Ninove L, Van Dooren P (2008) Maximizing pageRank via outlinks. Linear Algebra Appl 429(5–6):1254–1276

    Article  MathSciNet  Google Scholar 

  11. De Martí J, Zenou Y (2015) Network games with incomplete information. J Math Econ 61:221–240

    Article  MathSciNet  Google Scholar 

  12. De Martí J, Zenou Y (2017) Segregation in friendship networks. Scand J Econ 119:656–708

    Article  Google Scholar 

  13. Feri F (2007) Stochastic stability in networks with decay. J Econ Theory 135(1):442–457

    Article  MathSciNet  Google Scholar 

  14. Galeotti A, Goyal S, Jackson MO, Vega-Redondo F, Yariv L (2010) Network games. Rev Econ Stud 77(1):218–244

    Article  MathSciNet  Google Scholar 

  15. Galeotti A, Goyal S, Kamphorst J (2006) Network formation with heterogeneous players. Games Econ Behav 54(2):353–372

    Article  MathSciNet  Google Scholar 

  16. Gallo E (2012) Small world networks with segregation patterns and brokers. Rev Netw Econ 11(3):7

    Article  Google Scholar 

  17. Haller H, Sarangi S (2005) Nash networks with heterogeneous links. Math Soc Sci 50(2):181–201

    Article  MathSciNet  Google Scholar 

  18. Hellmann T (2021) Pairwise stable networks in homogeneous societies with weak link externalities. Eur J Oper Res 291(3):1164–1179

    Article  MathSciNet  Google Scholar 

  19. Jackson MO, Rogers BW (2005) The economics of small worlds. J Eur Econ Assoc 3(2–3):617–627

    Article  Google Scholar 

  20. Jackson MO, Wolinsky A (1996) A strategic model of social and economic networks. J Econ Theory 71:44–74

    Article  MathSciNet  Google Scholar 

  21. Jackson MO, Watts A (2002) On the formation of interaction networks in social coordination games. Games Econ Behav 41(2):265–291

    Article  MathSciNet  Google Scholar 

  22. Janssen RHP, Monsuur H (2012) Stable network topologies using the notion of covering. Eur J Oper Res 218(3):755–763

    Article  MathSciNet  Google Scholar 

  23. Johnson C, Gilles RP (2000) Spatial social networks. Rev Econ Des 5:273–300

    Google Scholar 

  24. Laakasuo M, Rotkirch A, van Duijn M, Berg V, Jokela M, David-Barrett T, Miettinen A, Pearce E, Dunbar R (2020) Homophily in personality enhances group success among real-life friends. Front Psychol 11:710

    Article  PubMed  PubMed Central  Google Scholar 

  25. Leung MP (2015) Two-step estimation of network-formation models with incomplete information. J Econom 188(1):182–195

    Article  MathSciNet  Google Scholar 

  26. Olaizola N, Valenciano F (2014) Asymmetric flow networks. Eur J Oper Res 237(2):566–579

    Article  MathSciNet  Google Scholar 

  27. Song Y, van der Schaar M (2015) Dynamic network formation with incomplete information. Rev Econ Des 59(2):301–331

    MathSciNet  Google Scholar 

  28. Sun P, Parilina E (2022) Impact of utilities on the structures of stable networks with ordered group partitioning. Dyn Games Appl 12(4):1131–1162

    Article  MathSciNet  CAS  Google Scholar 

  29. Tarbush B, Teytelboym A (2017) Social groups and social network formation. Games Econ Behav 103:286–312

    Article  MathSciNet  Google Scholar 

  30. Watts A (2001) A dynamic model of network formation. Games Econ Behav 34(2):331–341

    Article  MathSciNet  Google Scholar 

Download references

Funding

This research was supported by National Natural Science Foundation of China (No. 72171126).

Author information

Authors and Affiliations

Authors

Contributions

The authors equally contributed to this work.

Corresponding author

Correspondence to Elena Parilina.

Ethics declarations

Conflict of interest

Ping Sun declares that she has no conflict of interest. Elena Parilina declares that she has no conflict of interest.

Ethical approval

This article does not contain any studies with human participants performed by any of the authors.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

This research was supported by National Natural Science Foundation of China (No. 72171126).

Appendices

Appendix A. Proof of Theorem 1

Proof

 

  1. 1.

    When \(E_i>1\) holds for any \(i\in N\), as shown in Proposition 1, no pair would form a link in the entire process. Therefore, the empty network is the unique one that emerges and achieves stability.

  2. 2.

    We first prove for \(G_{C}\subseteq G_{IC}\).

    Let \(E_i\le 1\) be true for any \(i\in N\) and \(g\in G_{C}\). If \(g=\varnothing \), there must exist a pair of players \(i, j\in N\) such that \(f(|\pi (i)-\pi (j)|)>1\). Consider selection path \(\gamma _2=((i, j), (i, j))\) under incomplete information. It is evident that link ij will be formed at stage 1 and consequently severed at stage 2, which implies that \(g\in G_{IC}\). If \(g\ne \varnothing \), consider selection path \(\gamma _{t_1}\) such that \(g=g_C(\gamma _{t_1})\) and network g emerges for the first time at stage \(t_1\). Let construct another selection path \({\bar{\gamma }}_{t_2}\) based on \(\gamma _{t_1}\) by removing any pair (ij) at certain stages if players i and j do not form a link ij when it does not exist or do not sever it when it exists in the network at these stages. It is obvious that \(g=g_C({\bar{\gamma }}_{t_2})\) and \(t_2\le t_1\). Now we aim to demonstrate that with \({\bar{\gamma }}_{t_2}\) the formation process gives identical link formation and severance results under complete and incomplete information cases.

    Let pair \((i_1, j_1)\) be selected at stage 1 of path \({\bar{\gamma }}_{t_2}\). By construction of \({\bar{\gamma }}_{t_2}\), it is clear that link \(i_1j_1\) is built under complete information. Since \(E_{i_1}, E_{j_1}\le 1\), \(i_1j_1\) is also established under incomplete information, indicating that \(g_C({\bar{\gamma }}_{1})=g_{IC}({\bar{\gamma }}_{1})=\{i_1j_1\}\). Suppose \(g_C({\bar{\gamma }}_{k})=g_{IC}({\bar{\gamma }}_{k})\) holds true for \(k<t_2\). We demonstrate that \(g_C({\bar{\gamma }}_{k+1})=g_{IC}({\bar{\gamma }}_{k+1})\). Assume that players \(i_{k+1}\) and \(j_{k+1}\) are selected at stage \(k+1\), and three possible cases are discussed, respectively:

    1. (a)

      \(i_{k+1}j_{k+1}\in g_C({\bar{\gamma }}_{k})=g_{IC}({\bar{\gamma }}_{k})\). Then players \(i_{k+1}\) and \(j_{k+1}\) know each other’s exact group when they choose their actions at stage \(k+1\). As a result, each of them will make the same action under complete and incomplete information cases.

    2. (b)

      \(i_{k+1}j_{k+1}\notin g_C({\bar{\gamma }}_{k})=g_{IC}({\bar{\gamma }}_{k})\) and players \(i_{k+1}\) and \(j_{k+1}\) have never been connected before stage \(k+1\), i.e., their mutual beliefs remain at the prior. Taking into account selection path \({\bar{\gamma }}_{t_2}\) and assuming that \(E_{i_{k+1}}, E_{j_{k+1}}\le 1\), players \(i_{k+1}\) and \(j_{k+1}\) will form a link under both complete and incomplete information cases.

    3. (c)

      \(i_{k+1}j_{k+1}\notin g_C({\bar{\gamma }}_{k})=g_{IC}({\bar{\gamma }}_{k})\) and players \(i_{k+1}\) and \(j_{k+1}\) have already been connected before stage \(k+1\). Then, both players choose their actions knowing the exact payoff. Since such a link will be formed under complete information, it will also be established by these two players under incomplete information case.

    Above all, we conclude that \(g=g_C({\bar{\gamma }}_{t_2})=g_{IC}({\bar{\gamma }}_{t_2})\), implying \(g\in G_{IC}\).

    To prove for \(S_{C}^*\subseteq S_{IC}^*\), we first clarify that any network emerging under complete information can also appear under incomplete information by the analysis given above. Thus, it is sufficient to show that for any \(g\in S_C^*\) there exists a selection path under incomplete information after the first appearance of g that makes it stable. We demonstrate it by construction.

    If \(g=\varnothing \), consider selection path \(\gamma _{\frac{n(n-1)}{2}}\) such that each pair of players is selected twice consecutively. Since \(g\in S_C^*\), we know that under incomplete information each pair first forms a link and then severs it at the next stage. Finally, at stage \(\frac{n(n-1)}{2}+1\), information is complete and empty network g becomes stable, i.e., \(g\in S_{IC}^*\).

    If \(g\ne \varnothing \), denote the number of components (including isolated players) in network g by q(g). Under incomplete information let the subsequent selection path after the first appearance of g be such that at first \(q(g)(q(g)-1)\) stages, two players from various components are selected twice consecutively, and every two components are involved in the process. As \(g\in S_C^*\), even if the two players may form a link at the former stage misled by the prior beliefs, the link will be severed when they are selected at the latter stage. Alternatively, if two players have been connected before, the link would not be even formed at two stages when they are consecutively selected. In either case, players know each other’s group as well as the groups of players in the counter party’s component. Respectively, after \(q(g)(q(g)-1)\) stages, information is completely revealed. Again, when \(g\in S_C^*\), no link will be modified given any subsequent selection path. Thus, \(g\in S_{IC}^*\), which completes the proof. \(\square \)

Appendix B. Proof of Lemma 2

Suppose \(g\in S^*_C\), and it directly follows from the utility function form (no decay of benefit) that g must be minimal. Now we prove the connectivity of network g. Since \(b+\frac{a}{4}(m-\frac{1}{m})\le 1\), we obtain that \(f(0)=b<1\). Thus for any two players ij such that \(\pi (i)=\pi (j)\), i and j must be connected in g. We assume that there exist two players i and k where \(|\pi (i)-\pi (k)|=1\), i and k are not connected in g, then three possible cases should be discussed:

  1. 1.

    If \(m=3\), then \(b+\frac{a}{4}(m-\frac{1}{m})\le 1\) is equivalent to \(b+\frac{2a}{3}\le 1\). Moreover, \(f(0)=b>0\), so \(a<\frac{3}{2}\). Therefore, we obtain

    $$\begin{aligned} f(1)=a+b\le 1+\frac{a}{3}<\frac{3}{2}<M. \end{aligned}$$

    As a result, for player i,

    $$\begin{aligned} \Pi _i(g+ik)-\Pi _i(g)\ge M-f(1)>0, \end{aligned}$$

    and the same inequality can be obtained for player k.

  2. 2.

    If \(m=4\), then \(b+\frac{a}{4}(m-\frac{1}{m})\le 1\) is equivalent to \(b+\frac{15a}{16}\le 1\). Moreover, \(f(0)=b>0\), thus, \(a<\frac{16}{15}\). So we obtain

    $$\begin{aligned} f(1)=a+b\le 1+\frac{a}{16}<\frac{16}{15}<M. \end{aligned}$$

    As a result, for player i,

    $$\begin{aligned} \Pi _i(g+ik)-\Pi _i(g)\ge M-f(1)>0, \end{aligned}$$

    and the same inequality can be proved for player k.

  3. 3.

    If \(m\ge 5\), then with \(b+\frac{a}{4}(m-\frac{1}{m})\le 1\) and \(a>0, b>0\), we prove that \(f(1)=a+b\le 1<M\). As a result, for player i,

    $$\begin{aligned} \Pi _i(g+ik)-\Pi _i(g)\ge M-f(1)>0, \end{aligned}$$

    and the same inequality can be found for player k.

In any case from the list above, if pair (ik) is selected, link ik must be established, which contradicts that \(g\in S_C^*\). Therefore, any two players from the nearby groups must be connected in g. Hence, g is minimally connected.

Appendix C. Proof of Theorem 3

We consider two cases:

  1. 1.

    \(\varnothing \in S_C^*\). Then, \(f(0)=b>1\); thus, for any player \(i\in N\), \(E_i>1\). As a result, we obtain that \(S_C^*=S_{IC}^*=\{\varnothing \}\).

  2. 2.

    \(\varnothing \notin S_C^*\). This implies that \(f(0)=b\le 1\). From condition (6), we obtain that \(b+\frac{a}{4}(m-\frac{1}{m})\le 1\), then from Lemma 2 it follows that g is minimally connected. Moreover, \(\max \limits _{i\in N} E_i=b+\frac{a}{4}(m-\frac{1}{m})+\frac{a}{4}(m+\frac{1}{m}-2)<2\). Consider a network \(g\in S^{*}_{C}\), such that g emerges and becomes stable after selection path \(\gamma \) under complete information. Now we are going to prove that there exists a selection path \({\bar{\gamma }}\), after which network g appears and becomes stable under incomplete information.

    Let \(i_f\) be the first neighbor of player \(i\in N\) in the selection path \(\gamma \) under complete information. We first prove that for any player \(i\in N_{\frac{m+1}{2}}\), \(E_{i_f}\le 1\). We prove this by contradiction. Suppose there exists player \(i\in N_{\frac{m+1}{2}}\) such that \(E_{i_f}>1\). Let \(c_1=|\pi (i_f)-\frac{m+1}{2}|\). Calculating the expected unit cost \(E_{i_f}=b+\frac{a}{4}(m-\frac{1}{m})+\frac{c_{1}^2a}{m}>1\) and combining with condition (6), we will get that \(f(c_1)>1\), contradicting that \(i_f\) is the first neighbor of player i under complete information. Thus, we prove that \(E_{i_f}\le 1\) for any player \(i\in N_{\frac{m+1}{2}}\).

    Any link \(ij\in g\) corresponds to one of three cases: (i) \(E_i, E_j\le 1\), (ii) \(E_i>1, E_j\le 1\), or (iii) \(E_i>1, E_j>1\). We will construct the selection path \({\bar{\gamma }}\) for the emergence of links corresponding to these three cases following the steps:

  3. Step 1:

    We randomly choose a player \(d\in N_{\frac{m+1}{2}}\), and let pair \((d, d_f)\) be selected in path \({\bar{\gamma }}\). From the proof given above, we know that \(E_{d_f}\le 1\), thus link \(dd_f\in g\) will be formed. Let \(F=\{i\mid E_i>1, E_{i_f}>1\}\) be the set of players, whose expected costs of forming a link are larger than 1 and for whose first neighbor in path \(\gamma \) under complete information, the expected costs of forming a link are also larger than 1. Let \(B_k=\{i\in F\mid i_f=k\}\), \(k\in N\), be the set of such players from set F, for whom the first neighbor in the formed network under complete information is player k. For any \(B_k\ne \varnothing \), let pairs \((d, i^{k}_1), \ldots , (d, i^{k}_{|B_k|}), (d, k)\) be selected in a successive order in path \({\bar{\gamma }}\), where \(\{i^{k}_1, \ldots , i^{k}_{|B_k|}\}=B_k\). Because \(E_d\le 1\) and \(\max \limits _{i\in N} E_i< 2\), all links dh, where \(h\in B_k\cup \{k\}\), are formed. Since for each player \(h\in B_k\cup \{k\}\), \(E_h>1\), i.e., \( \frac{a}{4}(m-\frac{1}{m})+b+\frac{(\pi (h)-\pi (d))^2a}{m}>1\), then from (6) we have \(f(|\pi (h)-\pi (d)|)>1\). Thus when pairs \((d, i^{k}_1), \ldots , (d, i^{k}_{|B_k|}), (d, k)\) are selected again, all links dh, \(h\in B_k\cup \{k\}\) are deleted by player d. Next for each player \(i\in F\), pair \((i, i_f)\) is chosen, i.e., \((i_1, (i_1)_{f}), \ldots , (i_{|F|}, (i_{|F|})_f)\) are consequently selected in path \({\bar{\gamma }}\). Each link \(ii_f\), where \(i\in F\), will be formed since players i and \(i_f\) know the labels of each other when they have been connected via player d and \(f(|\pi (i)-\pi (i_f)|)\le 1\).

    Remark on Step 1: With respect to the selected path of pairs constructed in Step 1, there are two exceptions: (i) If there exists set \(B_k\ne \varnothing \), such that \(dk\in g\) or \(di\in g\) for some \(i\in B_k\), then pair (dk) (pair (di)) is not selected in the second round to save a link dk (link di) at the end of Step 1. (ii) If there exists set \(B_k\ne \varnothing \), such that \(kd_f\in g\) or \(id_f\in g\) for some \(i\in B_k\), then pair (dk) (pair (di)) is replaced by pair \((d_f, k)\) (pair \((d_f, i)\)) in our selection path. It is also selected once in the first round to keep link \(kd_f\) (link \(id_f\)) at the end of Step 1, but not link dk (link di).

    Moreover, there is a special rule for Step 1: if there exists set \(B_k\) satisfying one of the above exceptions (given in Remark on Step 1), then the pairs, in which one of the players belongs to \(B_k\), i.e., pairs such that \((d, i^{k}_1), \ldots , (d, i^{k}_{|B_k|}), (d, k)\) should be selected after the pairs described in Step 1, which do not satisfy these exceptions.

    We move on to Step 2 to proceed with construction of path \({\bar{\gamma }}\).

  4. Step 2:

    Let \(A=\{i, E_i\le 1\mid \text {for any } j\in N_i(g), E_j>1\}\), that is the set of players, whose expected costs are not larger than 1, but all neighbors of these players have expected costs larger than 1. Denote by \({\bar{c}}=\min \limits _{\begin{array}{c} f(c)>1\\ c=1, \ldots , (m-1)/2 \end{array}}c\). Obviously, for any player \(i\in A\), \(|\pi (i)-\pi (i_f)|<{\bar{c}}\) while \(|\pi (i_f)-\frac{m+1}{2}|\ge {\bar{c}}\). As a result, for \(i\in A\), either \(\frac{m+1}{2}<\pi (i)<\pi (i_f)\), or \(\pi (i_f)<\pi (i)<\frac{m+1}{2}\) is satisfied. For any player \(k\in N\), denote \(A_k=\{i\in A, i_f=k\}\). For a certain set \(A_k\ne \varnothing \), without loss of generality, we assume that \(\frac{m+1}{2}<\pi (i)<\pi (k)\) for any \(i\in A_k\), then some player \(a_k\ne d_f\) satisfying \(E_{a_k}\le 1\), and such that \(\pi (a_k)<\frac{m+1}{2}\) and \(|\pi (a_k)-\pi (i)|\ge {\bar{c}}\), can be found. Till now (and after Step 1), the players \(a_k\) and \(i\in A_k\) have never been chosen. Therefore, when pairs \((a_k, i^{k}_1), \ldots , (a_k, i^{k}_{|A_k|}), (a_k, k)\), where \(\{i^{k}_1, \ldots , i^{k}_{|A_k|}\}= A_k\), are selected consequently with two rounds, then all links of type \(ia_k\), where \(i\in A_k\), as well as of type \(ka_k\) will be definitely added in the first round of sequential selection and then be deleted by player \(a_k\) in the second round of selection.

    During such a process, players \(i\in A_k\) and k recognize the labels of each other. Then, if we select pairs \((i^{k}_1, k), \ldots , (i^{k}_{|A_k|}, k)\) consequently, all these links will be formed. The case when \(\pi (i_f)<\pi (i)<\frac{m+1}{2}\) is satisfied for any \(i\in A_k\) can be similarly discussed and addressed. For each set \(A_k\ne \varnothing \), \(k\in N\), the pairs are selected in the same way as we have designed above.

    Remark on Step 2: For the selected pairs in Step 2, there is an exception: if there exists set \(A_k\ne \varnothing \) such that \(ka_k\in g\), then pair \((a_k, k)\) is not selected in the second round to save a link \(ka_k\) in the network after Step 2.

    Moreover, there is a special rule for Step 2: if there exists set \(A_k\) satisfying the exception above, then the pairs with the players from \(A_k\), that are \((a_k, i^{k}_1), \ldots , (a_k, i^{k}_{|A_k|}), (a_k, k)\), are selected after the pairs from the sets which do not satisfy such an exception.

  5. Step 3:

    Denote the network structure formed after Step 2 by \({\bar{g}}\) and let \(D=\{ij\in g{\setminus } {\bar{g}}\mid E_i\le 1~ \text {and}~ E_j\le 1\}\). The set \(D\ne \varnothing \) since for any player \(i\in N_{\frac{m+1}{2}}\) we have \(E_{i_f}\le 1\). All pairs (ij), \(ij\in D\) are chosen one by one, then all links belonging to set D can be established since no pair of players has been ever connected and expected costs of both players are not larger than 1.

  6. Step 4:

    Denote the network structure formed after Step 3 by \(g^{'}\). If \(g\setminus g^{'}=\varnothing \), then we finish constructing a selection path \({\bar{\gamma }}\). If \(g{\setminus } g^{'}\ne \varnothing \), then we proceed by considering any links \(ij\in g{\setminus } g^{'}\) such that \(E_i>1\) and \(E_j\le 1\). It is easy to check that players i and j have never been connected via other players through Steps 1–3. This indicates that they do not recognize the labels of each other.

    We describe the procedure of further construction of the selection path depending on the neighbors of players i and j in network \(g^{'}\). Our steps are described as follows:

    1. (a)

      \(n_i(g^{'})>0\) and \(n_j(g^{'})>0\). Choose pair (ij), then link ij is formed.

    2. (b)

      \(n_i(g^{'})>0\) and \(n_j(g^{'})=0\). Since and link \(pq\in g\) such that \(E_p\le 1\) and \(E_q\le 1\), has been formed at the previous steps, then for any \(k\in N_j(g)\), we may only have \(E_k>1\), i.e., \(j\in A\). As \(n_j(g^{'})\ge 1\) for any \(j\in A\), hence the situation when \(n_j(g^{'})=0\) never happens.

    3. (c)

      \(n_i(g^{'})=0\) and \(n_j(g^{'})>0\). Choose pair (ij), and then link ij will be formed.

  7. Step 5:

    Denote the network structure formed after Step 4 by \(g^{''}\), i.e., \(g^{''}=g^{'}\cup \{ij\mid ij\in g, E_i>1, E_j\le 1\}\). In this step we consider the links \(ij\in g\setminus g^{'}\), for which \(E_i>1\) and \(E_j >1\). We can easily show that through Steps 1-4, players i and j have never been connected via other players, thus they do not know the labels of each other.

    Our next steps are described as follows (depending on the neighbors of players i and j in network \(g^{''}\)):

    1. (a)

      \(n_i(g^{''})>0\) and \(n_j(g^{''})>0\). Select a pair (ij), and link ij will be established.

    2. (b)

      \(n_i(g^{''})\cdot n_j(g^{''})=0\). We prove that this situation cannot happen. Suppose \(n_j(g^{''})=0\), since for any \(kl\in g\), such that \(E_k>1\) and \(E_l\le 1\), we have \(kl\in g^{''}\). Thus we may obtain that \(E_{j_f}>1\) implying that \(j\in F\). Recall that each link \(kk_f\), such that \(k\in F\), has been already established, contradicting that \(n_j(g^{''})=0\).

Finally, the construction of path \({\bar{\gamma }}\) is finished. All links in network g have been built, i.e., g appears after \({\bar{\gamma }}\) under incomplete information. Since g is minimally connected and stable under complete information, thus, g is also stable under incomplete information. We conclude that any \(g\in S_{C}^*\) belongs to the set \(S_{IC}^*\).

Appendix D. Proof of Proposition 6

  1. 1.

    Since \(f(0)\le 1\), then \(E_i=\frac{M-1}{2M-1}f(0)+\frac{M}{2M-1}f(1)\le 1\) for each player \(i\in N\) because \(f(1)\le 2-\frac{1}{M}-(1-\frac{1}{M})f(0)\). Thus, for any two players ij such that \(\pi (i)=\pi (j)\), no matter whether they know the labels of each other when they are selected while not connected, link ij must be formed. Thus, in any network \(g\in S_{IC}^*\), players in the same group are connected. Next, consider a pair ik such that \(\pi (i)\ne \pi (k)\). Suppose they are not connected in \(g\in S_{IC}^*\). When they are selected after the appearance of g, link ik is formed since \(E_i=E_k\le 1\) if they do not know the labels of each other, or \(\Pi _i(g+ik)-\Pi _i(g)=\Pi _k(g+ik)-\Pi _k(g)=M-f(1)>0\) if they know each other.

  2. 2.

    We prove the second item by contradiction. Assume that there exists network \(g\in S_{SIC}^{*}\) such that g is not connected and g appears after selection path \(\gamma \). Two possible cases should be discussed:

    1. (a)

      There are two players i and j who are not connected in g, and \(\pi (i)=\pi (j)\). Without loss of generality, we assume that \(\pi (i)=\pi (j)=1\). There are several possible situations for players i and j:

      1. i.

        \(n_i(g)=n_j(g)=0\). In this case, players i and j have no idea about the labels of each other, but the belief of at least one of them (say i) has been updated to be higher than the initial one. The belief of player i can be increased only if another player from the same group has been connected with her before. Then, player i is either indirectly connected with such a player or his neighbor in network g due to \(f(0)\le 1\), contradicting \(n_i(g)=0\).

      2. ii.

        \(n_i(g)\ge 1\), \(n_j(g)=0\). In this case, players i and j again have no idea about the labels of each other. Moreover, in the component \(C_i(g)\), there is at least one player \(k\in N_2\). Otherwise, there exists player \(l\in N_2\) such that \(n_l(g)\ge 1\), and link il will be built if they are selected. It is a contradiction with the fact that g is stable.

        There exists exactly one player, say player q, such that \(n_q(g)=0\). If such a player does not exist, then either players i and j know each other (there are only two parts in the network, \(C_i(g)\) and isolated j), or link iq is formed when i and q are selected (\(n_q(g)\ge 1\)). Otherwise, if \(\pi (q)=1\), then player i must know the labels of j and q and make the connection when they are selected. If \(\pi (q)=2\), then link ij must be created when i and j are selected since \(E_i^{\gamma }=\frac{1}{2}\cdot f(0)+\frac{1}{2}\cdot f(1)=1-\frac{1-f(0)}{2\,M}<1\).

    2. (b)

      There are two players i and j which are not connected in g, where \(\pi (i)=1, \pi (j)=2\). Following (a) above, player i (j) is connected with each player in the same group with her. Since \(f(1)< M\), when i and j are selected, the link ij is formed, contradicting that g is stable.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Sun, P., Parilina, E. Dynamic Network Formation with Ordered Partitioning and Incomplete Information. Dyn Games Appl (2024). https://doi.org/10.1007/s13235-024-00552-z

Download citation

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1007/s13235-024-00552-z

Keywords

Navigation