1: \documentclass[a4paper]{article}

2: \usepackage{graphicx}

3: \usepackage{amsmath}

4: \linespread{1.6}

5: \begin{document}

6: \begin{center}

7: {\bf \Large

8: Heider Balance in Human Networks

9: }\\[5mm]

10: 

11: {\large

12: P. Gawro{\'n}ski and K. Ku{\l}akowski

13: }\\[3mm]

14: 

15: {\em

16: Department of Applied Computer Science,

17: Faculty of Physics and Applied Computer Science,

18: AGH University of Science and Technology\\

19: al. Mickiewicza 30, PL-30059 Krak\'ow, Poland

20: }

21: 

22: \bigskip

23: {\tt kulakowski@novell.ftj.agh.edu.pl}

24: 

25: \bigskip

26: \today

27: \end{center}

28: 

29: \begin{abstract}

30: 

31: Recently, a continuous dynamics was proposed to simulate dynamics of interpersonal relations

32: in a society represented by a fully connected graph. Final state of such a society was found

33: to be identical with the so-called Heider balance (HB), where the society is divided into two 

34: mutually hostile groups. In the continuous model, a polarization of opinions was found in HB. 

35: Here we demonstrate that the polarization occurs also in Barab\'asi-Albert networks, where the 

36: Heider balance is not necessarily present. In the second part of this work we demonstrate 

37: the results of our formalism, when applied to reference examples: the Southern women and the 

38: Zachary club.

39: \end{abstract}

40: 

41: \noindent

42: {\em PACS numbers:} 87.23.Ge 

43: 

44: \noindent

45: {\em Keywords:} numerical calculations; sociophysics

46: 

47: 

48: \section{Introduction}

49: 

50:  The Heider balance \cite{h46,hei2,hara,dor1,wt} is a final state of personal relations 

51: between members of a society, reached when these relations evolve according to 

52: some dynamical rules. The relations are assumed to be symmetric, and they can be friendly 

53: or hostile. The underlying psycho-sociological mechanism of the rules is an attempt of 

54: the society members to remove a cognitive dissonance, which we feel when two of our friends 

55: hate each other or our friend likes our enemy. As a result of the process, the society 

56: is split into two groups, with friendly relations within the groups and hostile 

57: relations between the groups. As a special case, the size of one group is zero, 

58: i.e. all hostile relations are removed. HB is the final state if each member 

59: interacts with each other; in the frames of the graph theory, where the problem is formulated, 

60: the case is represented by a fully connected graph.

61: 

62: Recently a continuous dynamics has been introduced to describe the time evolution of the 

63: relations \cite{my1}. In this approach, the relations between nodes $i$ and $j$were 

64: represented by matrix elements $r(i,j)$, which were real numbers, friendly ($r(i,j)>0$) or

65: hostile ($r(i,j)<0)$. As a consequence of the

66: continuity, we observed a polarization of opinions: the absolute values of the matrix 

67: elements $r(i,j)$ increase. Here we continue this discussion, but the condition of maximal 

68: connectivity is relaxed, as it could be unrealistic in large societies. 

69: The purpose of first part of this work is to demonstrate, that even if HB is not 

70: present, the above mentioned polarization remains true. In Section II we present 

71: new numerical results for a society of $N=100$ members, represented by Barab\'asi-Albert (BA)

72: network \cite{ab}. Although this size of considered social structure is rather small, it is 

73: sufficient to observe some characteristics which are different than those in the exponential 

74: networks. In second part (Section III) we compare the results of our equations of motion with 

75: some examples, established in the literature of the subject. The Section is closed by final 

76: conclusions.

77: 

78: 

79: \section{Calculations for Barab\'asi-Albert networks}

80: 

81: The time evolution of $r(i,j)$ is determined by the equation of motion \cite{my1}

82: 

83: \begin{equation}

84: \frac{dr(i,j)}{dt}=\Big\{1-\Big(\frac{r(i,j)}{R}\Big)^2\Big\}\sum_k r(i,k) r(k,j)

85: \end{equation}

86: where $R$ is a sociologically justified limitation on the absolute value of $r(i,j)$ \cite{my1}. 

87: Here $R=5.0$. Initial values of $r(i,j)$ are random numbers, uniformly distributed in the range 

88: $(-0.5,0.5)$.

89: The equation is solved numerically with the Runge-Kutta IV method with variable length of timestep

90: \cite{RK4}, simultaneously 

91: for all pairs $(i,j)$ of linked nodes. The method of construction of BA networks was described

92: in \cite{MK}. The connectivity parameter is selected to be $M=7$, because in this case the 

93: probability $p(M)$ of HB has a clear minimum for BA networks of $N=100$ nodes, and 

94: $p(M=7)\approx 0.5$ (see Fig. 1). This choice of $M$ is motivated by our aim to falsify the result on the 

95: polarization of opinions. This polarization was demonstrated \cite{my1} to be a consequence of HB;

96: therefore, the question here is if it appears also when HB is not present. An example of 

97: time evolution of such a network is shown in Fig. 2.

98: 

99: Our result is that the polarization is present in all investigated cases. As time 

100: increases,

101: the distribution of $r(i,j)$ gets wider and finally it reaches a stable shape, with two large peaks 

102: at $r(i,j)\approx\pm R$ and one smaller peak at the centre, where $r(i,j)\approx 0$.

103: In Fig. 3, we show a series of histograms of $r(,j)$ in subsequent times (A-E).

104: Particular networks differ quantitatively with respect to the heights of the peaks, but these 

105: differences are small. 

106: 

107: We note here that when some links are absent, the definition of HB should be somewhat relaxed,

108: because some other links, which do not enter to any triad $(i,j,k)$, will not evolve at all. 

109: Therefore we should admit that some negative relations survive within a given group. We classify a

110: final state of the graph as HB if there are no chains of friendly relations between the subgroups. 

111: On the other hand, more than two mutually hostile subgroups can appear. These facts were recognized

112: already in literature \cite{hara,wt}. Surprisingly enough, subgroups of $1<N<97$ nodes are never 

113: found in our BA networks. On the contrary, in the exponential networks groups of all sizes

114: were detected. In Figs. 4 and 5 we show diagrams for BA networks and exponential networks, respectively.

115: Each point at these diagrams marks the value of $r(i,j)$ and the size of the subgroup which

116: contains nodes $(i,j)$. Links between different subgroups are omitted. We see that for BA

117: networks (Fig.4), the lowest value of $N$ is 97. The remaining three nodes are linked with all

118: other nodes by hostile relations. 

119: 

120: \section{Examples}

121: 

122: In Ref. \cite{my1}, an example of polarization of opinions on the lustration law in Poland in 1999

123: was brought up. The presented statistical data \cite{cbos} displayed two maxima at negative and

124: positive opinions and a lower value at the centre of the plot. In our simulations performed for

125: fully connected graphs \cite{my1}, the obtained value for the center was zero. However, it is clear 

126: that in any group larger than, say, 50 persons some interpersonal relations will be absent. Taking this

127: into account, we can claim than the statistical data of \cite{cbos} should be compared to the

128: results discussed here rather than to those for a fully connected graph. Here we reproduce a 

129: peak of the histogram at its centre, on the contrary to the results in \cite{my1}. This fact

130: allows to speak on a qualitative accordance of the results of our calculations with the statistical 

131: data of \cite{cbos}.

132: 

133: Next example is the set of data of the attendance of 18 'Southern women' in local meetings in 

134: Natchez, Missouri, USA in 1935 \cite{free}. These data were used to compare 21 methods of finding 

135: social groups. The results were analysed with respect to their consensus, and ranked with consensus

136: index from 0.543 (much worse than all others) to 0.968. To apply our dynamics, we use the 

137: correlation function $<p(i,j)>-<p(i)><p(j)>$ as initial values of $r(i,j)$. Our method produced 

138: the division (1-9) against (10-18), what gives the index value 0.968. As a by-product, the method 

139: can provide the time dynamics of the relations till HB and, once HB is reached, the leadership within 

140: the cliques \cite{bl}. We should add that actually, we have no data on the possible friendship or 

141: hostility between these women, then the interpretation of these results should be done with care.

142: 

143: Last example is the set of data about a real conflict in the Zachary karate club \cite{za,bonet,gir}.

144: The input data are taken from \cite{wbpg}. All initial values of the matrix elements are reduced 

145: by a constant $\epsilon$ to evade the case of overwhelming friendship. The obtained splitting of 

146: the group is exactly as observed by Zachary: (1-8,11-14,17,18,20,22) against 

147: (9,10,15,16,19,21,23-34). These results were checked not to vary for $\epsilon$ between 1.0 

148: and 3.0. The status of all group members can be obtained with the same method as in the previous 

149: example.

150: 

151: To conclude, the essence of Eq. (1) is the nonlinear coupling between links $r(i,j)$, which produces the positive 

152: feedback between 

153: the actual values of the relations and their time evolution. We should add that the idea of such 

154: a feedback is not entirely new.

155: It is present, for example, in Boltzmann-like nonlinear master equations applied to behavioral

156: models \cite{hlb}. On the contrary, it is absent in later works on formal theory of social influence 

157: \cite{cons}. On the other hand, the theories of status \cite{bl}

158: are close to the method of transition matrix, known in non-equilibrium statistical mechanics 

159: \cite{re}.

160: 

161: 

162:  

163: \bigskip

164: \begin{thebibliography}{99}

165: \bibitem{h46} F. Heider, J. of Psychology  {\bf 21} (1946) 107.

166: \bibitem{hei2} F. Heider, {\it The Psychology of Interpersonal Relations}, J.Wiley and Sons, 

167: New York 1958.

168: \bibitem{hara} F. Harary, R. Z. Norman and D. Cartwright, {\it Structural Models: An

169:  Introduction to the Theory of Directed Graphs}, John Wiley and Sons, New York 1965.

170: \bibitem{dor1} P. Doreian and A. Mrvar, Social Networks {\bf 18} (1996) 149.

171: \bibitem{wt} Z. Wang and W. Thorngate, J. of Artificial Societies and Social Simulation, Vol 6, 

172: No 3 (2003).

173: \bibitem{my1} K. Ku{\l}akowski, P. Gawro\'nski and P. Gronek, Int. J. Mod. Phys. C (2005), 

174: in print (physics/0501073) See also (physics/0501160). 

175: \bibitem{ab} R. Albert and A.-L. Barab\'asi, Rev. Mod. Phys. {\bf 286} (2002) 47.

176: \bibitem{RK4} M. Abramowitz and I. A. Stegun (Eds.), {\it Handbook of Mathematical Functions}, Dover, 

177: New York, 1972.

178: \bibitem{MK} K. Malarz and K. Ku{\l}akowski, Physica A {\bf 345} (2005) 326 (see also cond-mat/0501545).

179: \bibitem{cbos} Report BS/152/99 of the Public Opinion Research Center, Tab. 3 (in Polish).

180: \bibitem{free} L. C. Freeman, in R. Breiger, K. Carley and P. Pattison (Eds.), {\it Dynamic Social 

181: Network Modeling and Analysis},  Washington, D.C.:The National Academies Press, 2003.

182: \bibitem{bl} Ph. Bonacich and P. Lloyd, Social Networks {\bf 23} (2001) 191.

183: \bibitem{za} W. W. Zachary, J. Anthrop. Res. {\bf 33} (1977) 452.

184: \bibitem{bonet} L. Donetti and M. Mu\~noz, J. Stat. Mech.: Theor. Exp. (2004) 10012.

185: \bibitem{gir} M. Girvan and M. E. J. Newman, Phys. Rev. E {\bf 69} (2004) 026113.

186: \bibitem{wbpg} vlado.fmf.uni-lj.si/pub/networks/data/Ucinet/UciData.htm, dataset ZACHC

187: \bibitem{hlb} D. Helbing, Physica A {\bf 196} (1993) 546.

188: \bibitem{cons} N. E. Friedkin and E. C. Johnsen, Social Networks {\bf 19} (1997) 209.

189: \bibitem{re} L. E. Reichl, {\it A Modern Course in Statistical Physics}, J. Wiley and Sons, 

190: New York 1998, p. 241.

191: \end{thebibliography}

192: \newpage

193: 

194: \begin{figure}

195: \includegraphics[angle=-90,width=.8\textwidth]{fig1.ps}

196: \caption{The number of networks per 1000 trials, where HB is not reached.}

197: \label{fig1}

198: \end{figure}

199: 

200: \begin{figure}

201: \includegraphics[angle=-90,width=.8\textwidth]{fig2.ps}

202: \caption{The number of unbalanced triangles of nearest neighbours $(i,j,k)$ against time.

203: Six successive times for Fig. 3 are marked with labels A-F.}

204: \label{fig2}

205: \end{figure}

206: 

207: \begin{figure}

208: \includegraphics[angle=270,width=7.5cm]{fig3a.ps}

209: \includegraphics[angle=270,width=7.5cm]{fig3b.ps}

210: \includegraphics[angle=270,width=7.5cm]{fig3c.ps}

211: \includegraphics[angle=270,width=7.5cm]{fig3d.ps}

212: \includegraphics[angle=270,width=7.5cm]{fig3e.ps}

213: \includegraphics[angle=270,width=7.5cm]{fig3f.ps}

214: \caption{Time evolution of the histogram of the matrix elements $r(i,j)$.}

215: \label{fig3a}

216: \end{figure}

217: 

218: 

219: 

220: 

221: \begin{figure}

222: \includegraphics[angle=-90,width=.8\textwidth]{fig4.ps}

223: \caption{Diagram of values of $r(i,j)$ and the group size for BA networks.}

224: \label{fig4}

225: \end{figure}

226: 

227: \begin{figure}

228: \includegraphics[angle=-90,width=.8\textwidth]{fig5.ps}

229: \caption{Diagram of values of $r(i,j)$ and the group size for exponential networks.}

230: \label{fig5}

231: \end{figure}

232: 

233:   

234: \end{document}

235: