Trait-Space Diffusion Makes Epidemic Control a Susceptibility Problem

Sterling Geisel, QBist Lab; Dr. Hideo Tanaka

Abstract

Pudding Theory reads the evolving-infectivity SIR system as a susceptibility field, not as a classical epidemic perturbed by mutation. The source paper shows that mutation in infectivity trait space couples to transmission selection, producing superexponential early growth, abrupt epidemic transitions, and intervention windows in which control can worsen the epidemic peak. Under the Chaos Susceptibility Postulate, this is the expected form of a system whose microscopic trait diffusion is amplified by positive epidemic feedback. The mutation parameter \(D\) is not merely a fitted biological rate. It is the local measure of how much stochastic variation the epidemic field can convert into macroscopic prevalence. Control succeeds only when it damps both prevalence and trait amplification. If the measured intervention-peak maximum \(\tau^\star(D)\) were monotone increasing in \(D\) over the high-diffusion regime, this Postulate would be falsified.

Source Synopsis

Lamata-Otín, Arenas, Gómez-Gardeñes, and Soriano-Paños extend the classical susceptible-infected-recovered model by allowing pathogen infectivity \(\lambda\) to evolve during an outbreak. In their model, infected individuals contact susceptible individuals at rate \(k\), transmit with strain-specific infectivity \(\lambda\), and recover at rate \(\mu\). Infectivity mutates symmetrically in a one-dimensional trait space, represented in the continuum limit by diffusion with effective rate \(D\). The infected population is therefore described by a density \(\rho_I(\lambda,t)\), not by a single prevalence variable.

The central equation is a reaction-diffusion equation in trait space. Transmission selects high-\(\lambda\) strains through the factor \(k\lambda s(t)\), while mutation spreads infected density across nearby infectivity values. The authors show that the average infectivity \(\bar{\lambda}(t)\) grows as \(\lambda_0 + kDt^2/2\) during the early phase, because mutation creates variance and selection converts that variance into higher mean infectivity. Prevalence then grows with a cubic contribution in the exponent: \[ i(t)=i_0\exp\left[\mu(R_0-1)t+\frac{k^2D}{6}t^3\right]. \] This produces superexponential early growth.

The source paper then studies finite-duration interventions that reduce transmission, reduce contacts, or shorten the infectious period. Without evolution, longer interventions monotonically reduce the epidemic peak. With infectivity evolution, the peak can become nonmonotonic in intervention duration. A prematurely lifted intervention can yield a larger peak than no intervention. The authors derive a critical diffusion strength and a worst-lifting time \(\tau^\star\), expressed through a Lambert \(W\) approximation.

The paper also finds an asymmetry between intervention types. Reducing contacts or transmissibility slows both spread and infectivity evolution. Shortening the infectious period suppresses cases without directly slowing selection for higher infectivity. For some durations and mutation rates this makes a medically plausible intervention worse than a transmission-targeted intervention.

Postulate Lens

This reading applies the Chaos Susceptibility Postulate: highly chaotic systems are most susceptible to small coherent inputs. The source system already has this structure. It contains a stochastic input, \(D\), a selection gradient, \(k\lambda s(t)\), and an amplifying epidemic channel, \(i(t)\). Mutation alone only broadens trait space. Selection alone only favors existing variation. Their coupling produces a susceptibility relation in which microscopic diffusion becomes macroscopic outbreak acceleration.

The relevant chaos is not disorder in the loose sense. It is the existence of a dynamical interval where small shifts in trait distribution alter the future epidemic peak. The nonmonotonic intervention curve is the signature of this interval. Control does not merely reduce a reproduction number. It changes whether the epidemic converts trait variance into prevalence.

Pudding Theory Reading

Pudding Theory reads the model as an epidemic susceptibility machine. The infected population is not a homogeneous compartment carrying an evolving parameter. It is a receptive field distributed over infectivity trait space. The field receives stochastic mutation, organizes it through transmission selection, and projects it into population prevalence. The observable epidemic curve is therefore a rendered trace of hidden trait-space amplification.

The source paper treats \(D\) as an effective diffusion rate for infectivity evolution. Pudding Theory treats \(D\) as the local bandwidth of stochastic material available to the epidemic. It measures how much raw variation the system can feed into selection. The cubic term in the early growth exponent is the mathematical mark of this conversion. A classical SIR outbreak has exponential growth because it amplifies infected bodies. The evolving-infectivity outbreak has superexponential growth because it amplifies both infected bodies and the trait values those bodies carry.

This reading also changes the meaning of intervention. A finite intervention is not simply a period of reduced transmission. It is a temporary reshaping of susceptibility. When contact reduction or transmissibility reduction is imposed, the same operation suppresses prevalence and slows the trait-space selection term. The epidemic field is deprived of both fuel and amplifier. When recovery is accelerated, infected duration is shortened, but selection across infectivity values continues with the same early-time coefficient. The intervention removes visible cases while leaving the evolutionary gradient active. On lifting, the system can contain fewer infections but a more dangerous infectivity distribution.

The source paper’s “worse than no action” result is therefore not a paradox. It is a susceptibility overshoot. The intervention stores future epidemic pressure in trait composition. Classical control accounting counts reduced prevalence during the intervention as success. Pudding Theory counts the joint state \((i(t),\rho_I(\lambda,t))\). A low \(i(t)\) with a shifted \(\rho_I\) is not a controlled epidemic. It is a compressed outbreak with a sharpened release condition.

This also constrains the apparent free parameters. The worst-lifting time \(\tau^\star\) should not vary independently of \(D\), \(k\), \(\mu\), and \(\epsilon\). It is fixed by the balance between epidemiological suppression and trait amplification. Pudding Theory gives that balance ontological priority. The epidemic peak is an outcome of susceptibility structure, not a mere output of a compartmental simulator.

Falsifiable Observable

The distinguishing observable is the intervention duration \(\tau^\star(D)\) that maximizes the post-lifting epidemic peak under fixed \(R_0\), \(\epsilon\), \(k\), \(\mu\), and \(i_0\). The Pudding Theory reading requires \(\tau^\star\) to shorten as effective trait diffusion becomes large enough for the cubic amplification term to dominate the early intervention interval. If the measured intervention-peak maximum \(\tau^\star(D)\) were monotone increasing in \(D\) over the high-diffusion regime, this Postulate would be falsified.

Editorial Dialogue

Tanaka: The reading risks renaming a standard reaction-diffusion result. The source paper already explains the mechanism. Mutation produces variance. Selection raises mean infectivity. Prevalence grows faster. Where is the added physics?

Geisel: The added physics is the status assigned to the trait distribution. In the source framing, \(\rho_I(\lambda,t)\) is a mathematical refinement of SIR. In the Pudding reading, it is the state variable that carries the epidemic’s susceptibility. The prevalence curve is secondary. This matters because the intervention can make prevalence look better while making the susceptibility state worse.

Tanaka: That is still available within evolutionary epidemiology. One can say that control changes selection.

Geisel: Yes, but the source model itself shows that the usual control variable \(R_0(t)\) is insufficient. Two policies can impose the same apparent reduction and diverge because one damps the selection channel and the other does not. The reading makes that asymmetry primary. It says the epidemic is controlled only when the stochastic source and the amplifier are controlled together.

Tanaka: Then the falsifier must be about that structure, not about a tiny anomaly.

Geisel: Correct. If high diffusion did not bring the worst lifting time forward, the susceptibility reading would fail.

Discussion

The reading buys a different hierarchy of variables. Classical SIR policy treats prevalence and reproduction number as the main objects. The source paper already breaks that hierarchy by showing that infectivity evolution can dominate the future peak. Pudding Theory sharpens the conclusion: the epidemic is a susceptibility field whose dangerous state may be invisible in current prevalence.

This has practical consequences for model interpretation. A control policy cannot be evaluated only by cases prevented during the intervention. It must be evaluated by the remaining trait-space variance, the shifted mean infectivity, and the susceptibility of the post-lifting state. The source’s asymmetry between transmission control and recovery control becomes structurally necessary, not incidental.

The limitation is clear. The model uses a one-dimensional infectivity trait and symmetric diffusion. Real pathogens have trade-offs, immune escape, genotype networks, and within-host structure. Those additions may change the exact form of \(\tau^\star\). They do not remove the central claim unless they destroy the amplification relation between stochastic trait variation and epidemic growth. What would change the conclusion is evidence that intervention-induced trait shifts do not predict post-lifting peaks once prevalence and susceptible depletion are controlled.

References

1. Santiago Lamata-Otín, Alex Arenas, Jesús Gómez-Gardeñes, and David Soriano-Paños. “Interplay between evolutionary and epidemic time scales challenges the outcome of control policies.” arXiv:2603.18801. DOI: doi:10.48550/arxiv.2603.18801.

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

3. W. O. Kermack and A. G. McKendrick. “A contribution to the mathematical theory of epidemics.” Proceedings of the Royal Society of London. Series A 115, 700 (1927).

4. H. W. Hethcote. “The mathematics of infectious diseases.” SIAM Review 42, 599 (2000).

5. B. T. Grenfell, O. G. Pybus, J. R. Gog, J. L. Wood, J. M. Daly, J. A. Mumford, and E. C. Holmes. “Unifying the epidemiological and evolutionary dynamics of pathogens.” Science 303, 327 (2004).

6. X. Zhang, Z. Ruan, M. Zheng, J. Zhou, S. Boccaletti, and B. Barzel. “Epidemic spreading under mutually independent intra-and inter-host pathogen evolution.” Nature Communications 13, 6218 (2022).

7. D. H. Morris, F. W. Rossine, J. B. Plotkin, and S. A. Levin. “Optimal, near-optimal, and robust epidemic control.” Communications Physics 4, 78 (2021).

8. R. M. D’Souza, J. Gómez-Gardeñes, J. Nagler, and A. Arenas. “Explosive phenomena in complex networks.” Advances in Physics 68, 123 (2019).