Abstract
At the heart of competition for resources lie two simple principles: survival of the fittest, which reflects how intrinsic fitnesses of competing agents influence their ability to gather resources; and cumulative advantage, which reflects how accumulated resources help gather even more resources. In this letter we show that competitions embodying just the essence of these two principles exhibit a variety of surprising behaviors with respect to their duration, as measured by how long it takes for an indisputable winner to emerge, and intensity, as measured by how many times agents have equal amount of resources. We demonstrate that when the fitnesses of the agents are different, long-lasting competitions continue to occur in spite of a dramatic decrease in their intensity and the certainty that the fitter agent will win. We also unveil an unexpected phenomenon that we call struggle of the fittest, where long-lasting competitions become more probable when one agent is slightly fitter than when agents are equally fit. These results have important implications for our understanding of competitions and various applications that leverage such models.
Introduction
In many complex systems—from the popularity of hashtags to the wealth of individuals—resource accumulation is driven by two factors: fitness (intrinsic skill or quality) and luck (randomness). When these systems exhibit cumulative advantage (CA), where having more resources makes it easier to get even more, the "fittest" usually wins in the long run. However, a major gap exists in our understanding of the process of that competition. Most research focuses on the final market share of winners. This study addresses two critical, neglected questions:
- Duration: How long does it take for the most skilled agent to take an undisputed lead?
- Intensity: How many times do the competitors tie for the lead before a winner emerges?
Modeling Competition Dynamics
The authors employ a rigorous mathematical approach, comparing two theoretical frameworks: the Random Walk (RW) model, where the probability of winning a resource depends solely on the fitness ratio, and Pólya’s Urn (CA), where the probability depends on both intrinsic fitness and current accumulated resources.
As shown in Figure 1, the competition can be modeled as a lattice path $(X_t, Y_t)$. The duration $T$ is the time step after which one agent maintains a lead forever, and intensity $N$ is the number of times the path hits the diagonal $X_t = Y_t$.
Key Results: Duration and Intensity
The study reveals that CA acts as a "luck multiplier." In a standard random walk, a more skilled agent quickly overcomes an early streak of bad luck. In a CA model, an early "stroke of luck" for the less skilled agent can be amplified, allowing them to lead for an incredibly long time.
The Power-Law of Duration
For equally fit agents in CA, the duration follows a power law with an exponent of $-1/2$. Crucially, for unequal fitness, it remains a power law, meaning the expected duration is often infinite. This is in sharp contrast to the RW model, where duration tails are exponential (very short).
Figure 2 illustrates this heavy-tailed behavior. The "shape" of the competition (the power-law slope) is heavily determined by the fitness ratio and the initial wealth of the fitter agent, while the initial wealth of the weaker agent has a much smaller impact.
The Struggle of the Fittest
A counter-intuitive finding is the "Struggle of the Fittest": the probability of a very long competition increases when one agent is slightly fitter than the other ($r > 1$), compared to when they are equally fit ($r = 1$). This is due to a discontinuous jump in the power-law exponent of the duration distribution.
Long but "Boring" Competitions
While CA makes competitions last much longer, it actually makes them less "intense." Once an agent takes a lead in a CA system, ties become rare. The competition is long not because of a back-and-forth struggle, but because one agent is "hanging on" to a lead they will eventually lose—a phenomenon the authors call the "Delusion of the Weakest."
Conclusion
The findings suggest that in CA systems, the principle of "survival of the fittest" should perhaps be renamed "survival of the fittest and persistent," as the more skilled agent must be able to endure long periods of trailing behind a "lucky" competitor. Future work could explore multi-agent dynamics and non-linear cumulative advantage models to better reflect the noise of real-world systems.
Summary updated