Fig 1.
A schematic of the procedure followed by the two agents selected to interact in one round of the ARLOD model, as originally described in [22].
Agent i expresses an opinion Oi to their neighbour j, who responds by punishing or rewarding agent i. Agent i updates the Q-value for the opinion they expressed accordingly. The numbers to the top left of the boxes indicate the suggested order for reading the schematic.
Fig 2.
(A) Tail probabilities () (on a log-log scale) and (B) a box and whisker diagram for the time to consensus for different values of the radius rg used in the random geometric graph model to sample networks. The linear nature of these plots are indicative of a heavy tailed distribution. The high number of outliers on the upper end of the time to consensus is indicative of a heavy-tailed distribution. The parameter settings are detailed in §4.1.
Fig 3.
Number of agents holding opinion o = 1 in a simulation run exhibiting metastable behaviour plotted with time on a logarithmic scale.
The state of the network is plotted for telling timestamps of this simulation run in Fig 4. In this simulation run rg = 0.25, the other parameters are as in §4.1.
Fig 4.
Opinions in simulation run with metastable behaviour at timestamp (A) t = 1, (B) t = 100, (C) t = 103, (D) t = 104, (E) t = 106, and (F) t ≈ 5.82×106.
Note the group with opinion o = −1 (blue) forms around t = 104 and switches to o = 1 (red) after t = 106. The corresponding total number of agents holding opinion o = 1 is plotted in Fig 3. In this simulation run rg = 0.25, the other parameters are as in §4.1.
Fig 5.
Illustration of the voter model dynamics.
We show the transition probabilities conditioned on voter 1 being selected to copy the opinion of one of their neighbours at random.
Fig 6.
The dynamics in one time step of the batched version of the ARLOD model at a high level of abstraction.
Agent i expresses an opinion bt times to their neighbour agent j who responds each time. Thereafter, agent i updates their Q-values with all the feedback they received.
Fig 7.
The number of agents holding opinion o = 1 in a simulation run of the SRLOD model plotted with time on a logarithmic scale.
Notice the switch from consensus on opinion o = 1 to opinion o = −1 shortly after t = 106.
Fig 8.
Opinions in simulation run of the SRLOD model at rounds (A) t = 1, (B) t = 103, (C) t = 104, (D) t = 2×105, (E) t = 3×105, and (F) t = 2×106.
Notice the switch from consensus on opinion o = 1 (red) to opinion o = −1 (blue) between t = 3×105 and t = 2×106.