Path Memory Drives Braess Trapping in Tandem-Running Ants

Authors: Sterling Geisel, QBist Lab; Dr. Hideo Tanaka

Abstract

Braess’s paradox in tandem-running Diacamma indicum ants is not a failure of cooperation. It is a memory-driven collective state. The source paper shows that adding a short linking bridge makes relocation slower because leaders preferentially choose the narrow shortest route, \(N_1N_2\), even when congestion makes that route globally inferior. Pudding Theory reads this as an instance of Material Memory: a successful exploratory traversal leaves a physical decision trace in the leader-colony system, and later choices are biased by that trace. The source model estimates this trace as an exploitation probability, \(p_{\mathrm{mle}}=0.690\). Pudding Theory treats it as a constrained memory strength generated by successful path discovery under nest-relocation stress. The bridge does not merely add capacity. It reshapes the memory basin so that the shortest discovered path becomes the dominant stored route. If route-conditioned repeat probability after first discovery were measured to be equal to unbiased exploration probabilities for all routes, this Postulate would be falsified.

Source Synopsis

Bairagya, Chakraborti, Annagiri, and Chakraborty study whether Braess’s paradox can appear in a eusocial system. In the classical paradox, adding a link to a transport network worsens travel time because individually rational agents crowd onto the apparently superior route. The usual account depends on selfish route choice. The source paper removes that assumption by using tandem-running ants during nest relocation.

The biological system is Diacamma indicum. During relocation, leaders guide nestmates from an old nest to a new nest. A leader then returns to recruit another follower. Thus the transport is bidirectional. Tandem leaders and returning leaders meet on paths, and narrow paths impose encounter costs. The authors first show that a bidirectional version of Braess’s paradox is possible in a minimal game with narrow short segments, wide long segments, and a near-zero-time bridge.

They then build an experimental path network. Without the bridge, ants can use two symmetric routes, \(N_1W_2\) and \(W_1N_2\). With the bridge, two additional routes appear, including the narrow-short route \(N_1N_2\). The measured observable is total relocation time per colony member, \(T\). The central result is that \(T\) is significantly larger in the bridge condition \(P_L\) than in the no-bridge condition \(P_0\). The colony moves more slowly when the network contains the additional link.

The path-choice data identify the mechanism. In \(P_0\), leader choices are statistically balanced between the two available mixed routes. In \(P_L\), leaders strongly prefer \(N_1N_2\), the route shortest in length but vulnerable to congestion. The authors then introduce an exploration-exploitation model. Leaders first explore by an unbiased random walk. After discovery, they reuse the discovered path with probability \(p\). Maximum likelihood gives \(p_{\mathrm{mle}}=0.690\), with 95 percent confidence interval \([0.636,0.747]\). Agent-based simulations using this value reproduce the observed path-choice frequencies. The paper concludes that cooperative systems can fall into Braess’s paradox without individual selfishness. Evolution may have selected shortest-path identification, and in this network that adaptation becomes an evolutionary spandrel.

Postulate Lens

The applicable Postulate is Material Memory. The source phenomenon already has the structure this Postulate names: an initial successful route discovery is retained by the leader as a trace, and that trace biases future path probability. The trace is not a metaphor for cognition alone. It is expressed in repeated bodily navigation, route fidelity, recruitment timing, and colony-level traffic density.

Material Memory fits because the source paper’s key fitted parameter, \(p\), is precisely a memory persistence parameter. In the authors’ model, exploration produces a first route. Exploitation then repeats that route with probability \(p\). The model treats \(p\) as a maximum-likelihood estimate. Pudding Theory treats \(p\) as the measurable strength of the retained signal laid down by successful traversal. The bridge changes the geometry of discovery, and therefore changes which trace becomes most available for repetition. Once \(N_1N_2\) becomes the easiest trace to form, the colony’s future traffic is biased toward the very path that makes relocation slower.

Pudding Theory Reading

Pudding Theory reads the ant colony as a distributed memory-bearing transport medium. The relevant unit is not a selfish individual. It is the leader-path-colony circuit formed when a tandem leader discovers, stores, and repeats a route while other leaders do the same. The source paper’s exploration-exploitation model captures this circuit statistically. Pudding Theory gives it physical interpretation: the successful traversal of a path writes a trace into matter, and repeated successful use deepens that trace until it becomes a probability well for later choices.

The source paper treats congestion as the cost that converts shortest path into slowest collective outcome. Pudding Theory treats congestion as the macroscopic visibility of memory alignment. In \(P_0\), the two available routes have symmetric memory basins. Random exploration discovers them at comparable rates, and exploitation preserves that balance. The colony therefore distributes its motion. In \(P_L\), the bridge produces a new short composite route. The narrow route \(N_1N_2\) has the largest discovery advantage under the random-walk geometry used by the authors. It is not just shorter. It is easier to inscribe. Once inscribed, it is repeated. Repetition then creates the traffic condition that makes the path slow.

This reading changes the status of the exploitation probability. In the source model, \(p\) is a fitted behavioral parameter. In Pudding Theory, \(p\) is structurally constrained by trace formation. It should increase with the salience and success of first discovery, with the number of repeated tandem cycles by the same leader, and with the stability of decision cues at the junction. It should not be an arbitrary colony constant. It should be route-conditioned. A leader who first discovers \(N_1N_2\) in \(P_L\) should show stronger subsequent route retention than a leader whose first successful path is \(W_1W_2\), because the short route receives both geometric reinforcement and repeated traffic reinforcement.

The bridge therefore does more than increase network capacity. It creates a memory trap. The trap is cooperative because each leader’s retained trace was formed in service of relocation. It is maladaptive at the colony scale because the same traces become correlated. Braess’s paradox appears when many locally useful memory traces select the same narrow material channel.

Falsifiable Observable

The distinguishing observable is route-conditioned repeat probability after first successful route discovery, measured separately for \(N_1N_2\), \(N_1W_2\), \(W_1N_2\), and \(W_1W_2\) in the bridge network. Pudding Theory predicts that repeat probability is not a single free \(p\), but is strongest for the shortest successfully discovered narrow route after controlling for exploration frequency. If route-conditioned repeat probability after first discovery were measured to be equal to unbiased exploration probabilities for all routes, this Postulate would be falsified.

Editorial Dialogue

Tanaka: The source paper already explains the data with exploration and exploitation. The maximum-likelihood value \(p=0.690\) reproduces the observed route frequencies. Why add a field-theoretic reading when the behavioral model is sufficient?

Sterling: The behavioral model is sufficient for curve fitting. It does not say what \(p\) is. It labels repeated use as exploitation and estimates its frequency. Pudding Theory identifies the physical content of exploitation: retained path trace. The ant does not simply choose from a menu at each trip. The first successful traversal changes the later probability landscape of that ant’s motion.

Tanaka: But that can be ordinary memory, not a new theoretical claim.

Sterling: Ordinary memory is the observed substrate. The claim is about its network consequence. In this geometry, memory does not improve optimization. It concentrates flow into the route easiest to encode. The bridge creates a stronger memory basin for \(N_1N_2\), and the colony follows the stored trace into congestion.

Tanaka: Then the prediction must differ from the source model.

Sterling: It does. A scalar \(p\) is too weak. The repeat probability should be route-conditioned and history-conditioned. The first successful \(N_1N_2\) traversal should bind later choices more strongly than an equally available wide route. If all routes show the same repeat strength after discovery, the reading fails.

Discussion

This reading buys a different account of the paradox. The source paper shows that selfishness is unnecessary. Pudding Theory goes further: suboptimal collective transport can arise from cooperative memory itself. The colony is not failing to communicate a global optimum. It is over-preserving locally successful traces.

The limitation is direct measurement. The existing experiment measures aggregate route frequencies and relocation time. It does not isolate first-discovery histories for individual leaders across subsequent tandem cycles. That is the needed next dataset. Marked leaders, full-path video tracking, and route-conditioned transition matrices would decide whether \(p\) is truly scalar or whether it decomposes into memory strengths attached to particular route traces.

The conclusion would change if repeat probabilities were independent of first successful route, or if congestion alone explained later choices without any route-specific persistence. The present data point the other way. The bridge alters discovery geometry, discovery writes memory, and memory organizes traffic. In that sequence, Braess’s paradox is a material memory effect.

References

1. Joy Das Bairagya, Udipta Chakraborti, Sumana Annagiri, and Sagar Chakraborty. “Braess’s paradox in tandem-running ants: When shortest path is not the quickest.” arXiv:2603.26226. DOI: doi:10.48550/arxiv.2603.26226, 2026.

2. S. Ochs. Pudding Theory: A Topological Theory of Information Fields. QBist Lab Working Paper, 2026.

3. D. Braess, A. Nagurney, and T. Wakolbinger. “On a paradox of traffic planning.” Transportation Science 39, 446-450, 2005.

4. S. Mukhopadhyay, M. K. Pathak, and S. Annagiri. “Path minimization in a tandem running Indian ant in the context of colony relocation.” Journal of Experimental Biology 222, 0022, 2019.

5. T. Gueudré, A. Dobrinevski, and J.-P. Bouchaud. “Explore or exploit? A generic model and an exactly solvable case.” Physical Review Letters 112, 2014. DOI: doi:10.1103/physrevlett.112.050602.

6. J. G. March. “Exploration and exploitation in organizational learning.” Organization Science 2, 71, 1991.

7. J. C. Biesmeijer and H. de Vries. “Exploration and exploitation of food sources by social insect colonies: a revision of the scout-recruit concept.” Behavioral Ecology and Sociobiology 49, 89-99, 2001.

8. N. R. Franks and T. Richardson. “Teaching in tandem-running ants.” Nature 439, 153-153, 2006.