PaperlayerView exact revision on arXiv

Source-linked AI summary

The role of frustration in collective decision-making dynamical processes on multiagent signed networks

Angela Fontan, Claudio Altafini

arXiv:2105.11396v1math.OCcs.SIcs.SYeess.SYphysics.soc-ph

TL;DR

The paper asks how frustration in signed social networks affects collective decisions. It analyzes a nonlinear opinion-dynamics model using social effort as a bifurcation parameter, finding that more frustration raises the effort needed to reach alternative decision states.

  • Problem

    The paper investigates how frustration in structurally unbalanced signed networks affects the bifurcations that determine community decision states.

    Source: §I · pp. 1–2

  • Method

    The authors analyze a nonlinear interconnected opinion model in which social effort π is the bifurcation parameter and equilibria represent community decisions.

    Source: §I · p. 1

  • Results

    Greater frustration raises the social effort threshold for the first bifurcation, while the qualitative behavior remains similar to the structurally balanced case.

    Source: §I · p. 2, §V · p. 6, §VII · p. 9

  • Takeaways & Limitations

    For π between the first and second thresholds, two alternative decisions are possible, but greater frustration requires greater effort to reach them.

    Source: §V · p. 6

Abstract

from arXiv · show

In this work we consider a collective decision-making process in a network of agents described by a nonlinear interconnected dynamical model with sigmoidal nonlinearities and signed interaction graph. The decisions are encoded in the equilibria of the system. The aim is to investigate this multiagent system when the signed graph representing the community is not structurally balanced and in particular as we vary its frustration, i.e., its distance to structural balance. The model exhibits bifurcations, and a ``social effort'' parameter, added to the model to represent the strength of the interactions between the agents, plays the role of bifurcation parameter in our analysis. We show that, as the social effort increases, the decision-making dynamics exhibits a pitchfork bifurcation behavior where, from a deadlock situation of ``no decision'' (i.e., the origin is the only globally stable equilibrium point), two possible (alternative) decision states for the community are achieved (corresponding to two nonzero locally stable equilibria). The value of social effort for which the bifurcation is crossed (and a decision is reached) increases with the frustration of the signed network.

I. INTRODUCTION

The paper studies decision-making when agents have both cooperative and antagonistic interactions, using a signed-network model whose equilibria represent community decisions. Its central result is that greater network frustration raises the social effort needed to cross the first bifurcation.

B. Signed graphs

  • A signed network labels friendly interactions positive and unfriendly interactions negative; structural balance means agents can be divided into internally friendly, mutually antagonistic groups.For connected undirected networks, structural balance is equivalent to the smallest normalized signed-Laplacian eigenvalue being zero.Source: §I · p. 1

I. INTRODUCTION

  • The model uses sigmoidal opinion-transmission nonlinearities and a social effort parameter π to represent interaction strength; equilibria encode community decision states.The paper extends analysis to signed networks that are not structurally balanced, where the system is no longer monotone.Source: §I · pp. 1–2
  • As frustration increases, the first bifurcation requires greater social effort, while the qualitative behavior remains similar to the structurally balanced case.Frustration is the distance of a signed network from structural balance; the threshold grows with the smallest eigenvalue of the normalized signed Laplacian.Source: §I · p. 2
  • The paper develops global state-space analysis for nonmonotone interconnected systems and bounds the 1-norm of equilibria in terms of network frustration.The stated class includes systems that are not diagonally dominant, such as Hopfield-like neural networks,.Source: §I · p. 2
  • The discrete-time extension can exhibit period-2 cycles, and its first bifurcation is a pitchfork or period-doubling depending on the relative threshold values.The pitchfork threshold matches the continuous-time case; with zero or small frustration, the period-doubling threshold exceeds the usual bifurcation threshold.Source: §I · p. 2

C. Monotone systems

The paper formulates opinion dynamics on signed networks and uses social effort as a bifurcation parameter. For structurally balanced networks, the system is monotone and follows the established sequence of pitchfork bifurcations.

C. Monotone systems

  • The system is monotone exactly when its interaction graph is structurally balanced, characterized by a fixed signature matrix making the Jacobian's off-diagonal entries nonnegative.The monotonicity criterion is stated through preservation of a signature-dependent partial ordering by the solution operator.Source: C. Monotone systems · p. 3

A. Problem formulation

  • Agent opinions evolve through signed neighbor interactions weighted by social effort π, and the equilibria represent community decisions.The model assumes an undirected, connected network without self-loops and sigmoidal opinion-transmission functions.Source: A. Problem formulation · pp. 3–4

B. Structurally balanced case

  • For structurally balanced networks, π < 1 gives a unique stable origin; at π = 1 a pitchfork yields two stable nonzero equilibria, and a later threshold can produce more equilibria.At the second pitchfork, the new equilibria may be stable or unstable; that bifurcation applies when λ2(L) < 1 and is simple.Source: B. Structurally balanced case · p. 4

C. Structurally unbalanced case

For structurally unbalanced networks, the system retains the qualitative pitchfork pattern, but its first threshold depends on the network. The paper also establishes equilibrium bounds tied to frustration.

C. Structurally unbalanced case

  • The first threshold grows with frustration, so more frustrated networks require greater social effort to reach a nontrivial equilibrium.The paper relates the threshold to the smallest eigenvalue of the normalized signed Laplacian.Source: C. Structurally unbalanced case · p. 4
  • When π crosses π1 = 1/(1 − λ1(L)), the origin loses global stability and a pitchfork creates two locally stable nonzero equilibria, stable until π2.For π < π1, the origin is the unique globally asymptotically stable equilibrium; between π1 and π2, there are exactly three equilibria.Source: C. Structurally unbalanced case · p. 4
  • At π2, a second pitchfork creates additional equilibria; for structurally unbalanced networks, λ2(L) is below 1, making π2 well-defined when λ2(L) is simple.The additional equilibria may be attractors.Source: C. Structurally unbalanced case · p. 4
  • For π > π1, the equilibrium 1-norm is bounded by π(n − 2ϵ(G)); if L is symmetric, trajectories converge to the set defined by that bound.The symmetric-L condition is ∆ = δI.Source: C. Structurally unbalanced case · pp. 4–5

IV. DISCRETE-TIME

The discrete-time model is analyzed through the social-effort thresholds at which the origin loses stability. Below both thresholds the origin is the unique globally asymptotically stable equilibrium; at the first threshold, either a pitchfork or a period-doubling bifurcation can occur, depending on threshold ordering.

IV. DISCRETE-TIME

  • π1 < π1,d holds for structurally balanced networks and for structurally unbalanced networks when λ1(L) < 2 − λn(L).Thus, under these conditions the first bifurcation is the pitchfork case.Source: §IV · p. 5
  • A nontrivial equilibrium requires π > π1, whereas a period-2 limit cycle requires π > π1,d; the latter result assumes ε < 2 max_i δ_i.The step-size condition is an upper bound for the Euler approximation.Source: §IV · pp. 5–6
  • Below min{π1, π1,d}, the origin is the unique equilibrium and is globally asymptotically stable, assuming 1 − ε max_i δ_i ≥ 0.The theorem addresses the first bifurcation at the origin.Source: §IV · p. 6
  • At the first threshold, a simple eigenvalue crossing +1 yields a pitchfork when π1 < π1,d, while a simple crossing −1 yields period-doubling when π1 > π1,d.The case π1 = π1,d is not analyzed; the authors suggest a Neimark–Sacker bifurcation there.Source: §IV · p. 6
  • Theorem 4 covers only the first bifurcation at the origin, not secondary bifurcations associated with other eigenvalues reaching +1 or −1.The excluded cases include λn−1(Jπ) = +1 and λ2(Jπ) = −1.Source: §IV · p. 6

V. DISCUSSION AND INTERPRETATION OF THE RESULTS

In the continuous-time model, frustration raises the effort threshold for decisions without changing the system’s general behavior. In discrete time, high frustration can additionally create a range of effort where opinions fluctuate rather than settle.

V. DISCUSSION AND INTERPRETATION OF THE RESULTS

  • Higher frustration raises π1 and shrinks (π1, π2), the range in which only two alternative equilibria are admitted.For balanced networks π1 = 1; for unbalanced networks π1 > 1 and rises with λ1(L).Source: §V · p. 6, Figure 1 · p. 7
  • For π < π1 there is no decision, while π ∈ (π1, π2) yields two alternatives; unbalanced networks require more effort as frustration increases.The diagrams show these bifurcation behaviors for balanced and unbalanced networks.Source: §V · p. 6, Figure 1 · p. 7
  • For π > π2, overcommitment can produce several alternative decisions; the discrete-time model also admits stable periodic solutions.Figure 2 depicts the balanced case and two unbalanced cases with different orderings of π1 and π1,d.Source: §V · p. 6, Figure 2 · p. 7
  • When π1 > π1,d, the discrete-time process fluctuates throughout (π1,d, π1); further commitment is needed before the community can reach a decision.This ordering is associated with structurally unbalanced networks with high frustration.Source: §V · pp. 6–7, Figure 2 · p. 7

VI. NUMERICAL EXAMPLES

The numerical examples illustrate the bifurcation bound across symmetric and nonsymmetric signed networks and show how frustration shapes equilibrium thresholds and magnitudes. They also demonstrate multistability, symmetry-related equilibria, and a discrete-time periodic solution between thresholds.

Example 1

  • The bifurcation bound holds for both symmetric and nonsymmetric normalized signed Laplacians, and is tight at low frustration when λ1(L) < 2 − λn(L).Example 1 uses two sequences of networks with 500 agents; the tightness condition is reported numerically.Source: §VI · p. 7

Example 2

  • As frustration rises across three unbalanced networks, the first Laplacian eigenvalue increases while the second stays nearly constant, narrowing the interval with only two equilibria.The examples use frustration values 0.677, 4.285, and 5.536.Source: §VI · p. 8, Figure 4 · p. 8

Example 3

  • The maximum equilibrium 1-norm is bounded by π(n − 2ϵ(Gβ)), and this bound decreases as frustration increases.Example 3 varies edge-sign probability β while preserving the underlying unsigned adjacency matrix.Source: §VI · p. 8

Example 4

  • For π = 4 > π2 = 1.995, simulations from 50 random initial conditions illustrate multistability in the network used in Example 2(c).The plot tracks the first state component over time.Source: §VI · p. 8, Figure 6 · p. 8

Example 5

  • With identical nonlinearities, equilibria include x*1=α[1,−1,0]ᵀ and x*2=[β,β,γ]ᵀ; Figure 7 shows the origin and ±P x*2 branches for tanh as π increases.Source: §VI · pp. 8–9, Figure 7 · p. 9

Example 6

  • When π1,d < π1, the discrete-time system admits a periodic solution for π between these thresholds, while the continuous-time system still converges to the origin.For both systems, π must exceed π1 to reach a nontrivial equilibrium; when π1 < π1,d, both converge to the origin below π1.Source: §VI · p. 9

VII. CONCLUSIONS

The analysis extends the nonlinear decision-making model to structurally unbalanced signed networks and finds that frustration raises the bifurcation threshold. In the discrete-time comparison, the ordering of two thresholds determines whether periodic behavior appears between them.

VII. CONCLUSIONS

  • The model retains qualitatively similar bifurcation behavior for structurally unbalanced networks, but the threshold grows with network frustration.The paper provides necessary and sufficient conditions for equilibrium existence and stability.Source: §VII · p. 9
  • Greater frustration corresponds to more social effort being required for agents to achieve a common decision.The authors interpret this relationship as plausible given frustration's representation of social disorder.Source: §VII · p. 9
  • Figure 8 presents trajectories of discrete-time system (8) with ε = 0.3 (top panels) and continuous-time system (5) (bottom panels), for Network G1 with π1 < π1,d and G2 with π1 > π1,d.Source: Figure 8 · p. 10

APPENDIX A TECHNICAL PRELIMINARIES

The appendix supplies linear-algebra tools and proofs supporting the equilibrium analysis. It establishes uniqueness and global stability of the origin below the first threshold, and a pitchfork bifurcation with new equilibrium branches at that threshold.

APPENDIX A TECHNICAL PRELIMINARIES

  • A symmetrizable matrix becomes symmetric after multiplication by a positive diagonal symmetrizer, and the appendix's eigenvalue lemma relates eigenvalues of scaled matrix products.The lemma bounds the scaling factor θk using the diagonal entries of the positive definite matrix.Source: Appendix A · p. 11

APPENDIX B PROOF OF THEOREM 2

  • The proof of Theorem 2 is divided into arguments for the origin, nontrivial-equilibrium existence and stability, and uniqueness.The proof of the theorem's remaining part is omitted as identical to the existence argument.Source: Appendix B · p. 11

A. Proof of Theorem 2(i)

  • For π ≤ π1, a Lyapunov argument shows the origin is globally asymptotically stable and the system has no other equilibrium.The proof uses monotonicity of the nonlinear functions and a radially unbounded Lyapunov function.Source: A. Proof of Theorem 2(i) · p. 11

B. Proof of Theorem 2(ii.1): existence

  • At (0, π1), the system undergoes a pitchfork bifurcation, creating two new nontrivial equilibrium branches.The proof verifies the derivatives required for the bifurcation conditions.Source: B. Proof of Theorem 2(ii.1): existence · p. 12

C. Proof of Theorem 2(ii.2): stability

The stability proof shows that nontrivial equilibria are locally asymptotically stable between the first two thresholds. Once π exceeds the first threshold, the origin becomes unstable, while existing nontrivial equilibria are stable.

C. Proof of Theorem 2(ii.2): stability

  • For π ∈ (π1, π2), every nontrivial equilibrium, if present, is locally asymptotically stable.The proof bounds the relevant eigenvalues to show the linearization at such an equilibrium is Hurwitz stable.Source: C. Proof of Theorem 2(ii.2): stability · pp. 12–13
  • When π > π1, the origin is unstable because the linearization has at least one positive eigenvalue.The stability of any nontrivial equilibrium in this range follows from the stated lemma.Source: C. Proof of Theorem 2(ii.2): stability · p. 13

D. Proof of Theorem 2(ii.3): uniqueness

The proof rules out additional bifurcations between π1 and π2, so only the origin and the two equilibria from the first pitchfork remain. It also uses eigenvalue properties to establish structural balance by contradiction and rewrites frustration through a signed quadratic-form maximization.

D. Proof of Theorem 2(ii.3): uniqueness

APPENDIX C PROOF OF LEMMA 2

  • For a connected graph, structural balance is equivalent to signing Hsym so that SHsymS is nonnegative; assuming λ2(L) > 1 contradicts structural unbalance.The contradiction uses that Hsym is special, nonsingular, and elliptic.Source: Appendix C · p. 14
  • When λ2(L) = 1, the proof likewise constructs a signature matrix making Hsym nonnegative, contradicting structural unbalance.It uses a permutation and block decomposition of the singular Hsym.Source: Appendix C · p. 14

APPENDIX D PROOF OF THEOREM 3

  • The frustration expression is rewritten using a sign vector that maximizes the network’s signed quadratic form.The maximizing vector u∗ satisfies u∗ = sign(HSu∗1).Source: Appendix D · p. 14

A. Proof of Theorem 3(i)

Theorem 3(i) bounds the size of every equilibrium using the network’s frustration, including when H is nonsymmetric or equilibria have zero components. The following proof section uses a Lyapunov argument for the symmetric case.

A. Proof of Theorem 3(i)

  • Every equilibrium satisfies ∥x∗∥1 ≤ π(n − 2ϵ(G)), without requiring H to be symmetric or every component of x∗ to be nonzero.The proof treats the symmetric case, the nonsymmetric case with nonzero components, and then equilibria with zero components.Source: A. Proof of Theorem 3(i) · pp. 14–15
  • For symmetric H, the proof bounds the equilibrium using a maximization over vectors in the hypercube |u| ≤ 1.The maxima are obtained at its corners, where |u| = 1.Source: A. Proof of Theorem 3(i) · p. 15

B. Proof of Theorem 3(ii)

  • For symmetric normalized signed Laplacian L, the proof uses a Lyapunov function to establish attraction toward Ωϵ(G).The set contains all equilibria, and the argument considers points outside it.Source: B. Proof of Theorem 3(ii) · pp. 15–16

APPENDIX E PROOF OF PROPOSITION 1

These proofs connect frustration to bounds on equilibrium size and bifurcation thresholds, and establish stability and convergence results under stated graph and parameter conditions.

APPENDIX E PROOF OF PROPOSITION 1

  • The proof of Proposition 1 reduces the frustration-dependent threshold bound to an eigenvalue inequality and invokes Rayleigh’s theorem for symmetric H.It expresses frustration through 2ϵ(G) = n − 1T SϵHSϵ1.Source: Appendix E · p. 16

APPENDIX F PROOF OF PROPOSITION 2

  • The first part of Proposition 2 applies to connected graphs whether structurally balanced or unbalanced; its threshold argument treats two eigenvalue conditions.The cases are λ1(L) = 0 and λ1(L) < 2 − λn(L).Source: Appendix F · p. 16

APPENDIX G PROOF OF LEMMA 3 AND LEMMA 4

  • The Lemma 3 and Lemma 4 arguments exclude nonzero equilibria when π ≤ π1 and period-2 cycles when π ≤ π1,d.The latter argument uses monotonicity and a Lipschitz constant of 1 for each nonlinearity.Source: Appendix G · p. 16

APPENDIX H PROOF OF THEOREM 4

  • Theorem 4’s proof is divided into separate arguments for parts (i) and (ii).Part (i) is proved in Section H-A and part (ii) in Section H-B.Source: Appendix H · p. 16

A. Proof of Theorem 4(i)

  • For part (i), the Lyapunov argument yields asymptotic convergence to the origin when −2I + εLπ is negative definite.Convexity establishes positivity of the auxiliary function away from zero.Source: A. Proof of Theorem 4(i) · pp. 16–17

B. Proof of Theorem 4(ii)

The proof uses a center-manifold reduction of the discrete-time system. Depending on the ordering of π1 and π1,d, it establishes a pitchfork or a period-doubling bifurcation.

B. Proof of Theorem 4(ii)

  • When π1 < π1,d, the discrete-time system undergoes a pitchfork bifurcation at π = π1.The critical eigenvalue is simple and equals +1.Source: B. Proof of Theorem 4(ii) · pp. 17–18
  • In both cases, the proof decomposes the state along the critical eigenvector and its orthogonal complement before restricting the dynamics to a center manifold.The coordinates are x = uq + y, with u = qT x and y orthogonal to q.Source: B. Proof of Theorem 4(ii) · pp. 17–18
  • When π1 > π1,d, a period-2 cycle bifurcates from the origin at π = π1,d.The reduction uses a one-dimensional eigenspace associated with the critical eigenvalue −1.Source: B. Proof of Theorem 4(ii) · p. 18
Loading 2105.11396v1…