Random Timing Makes Bacteria and Viruses Grow Faster — And Math Explains Why

Sometime in the next hour, deep inside a bacterium somewhere in your gut, a virus is making a fateful calculation — though "calculation" is too deliberate a word. The virus has hijacked the bacterium's cellular machinery and is assembling copies of itself, one by one, as holin proteins slowly accumulate in the bacterial membrane. When enough holin reaches a critical threshold, the membrane ruptures. Dozens of viral particles flood out. The bacterium is dead. The viral lineage proliferates.

The moment of rupture — called lysis — isn't scheduled. It's the culmination of a noisy, stochastic molecular race to a finish line. And according to a new paper by Aanjaneya Kumar and James Holehouse at the Santa Fe Institute (Kumar & Holehouse, 2026), that noisiness isn't a bug. It's a feature. In fact, it's mathematically provable that a virus with a jittery, imprecise lysis clock will outgrow an otherwise identical virus with a perfect, metronomic one.

That finding, derived rigorously from first principles, sits at the heart of a new theoretical framework the authors call Branching under First-Passage Resetting. The framework unifies cell division, viral lysis, and a broad class of natural replication events under a single mathematical lens — and, in doing so, solves a long-standing puzzle about how randomness at the molecular scale translates into growth at the population scale.

The Science

To appreciate what's new here, it helps to understand what existing theory assumes. Classical branching process theory — which dates back to a Victorian-era question about whether aristocratic family names go extinct — generally treats the timing of replication as externally imposed. You assign each cell a lifetime drawn from some distribution, and off you go. The distribution is a parameter you choose, not something the model derives.

Real biology doesn't work that way. A bacterium doesn't divide because a timer went off. It divides because an internal molecule — a growth-coupled division protein — has accumulated to a critical concentration. The time of division is the first moment that accumulation process crosses a threshold. In probability theory, this is called a first-passage time: the first instant a random walk, or diffusion process, or any stochastic trajectory, reaches a prescribed boundary.

Kumar and Holehouse's key move is to make that first-passage time the engine of replication itself. In their framework, a stochastic process $x(t)$ — representing whatever internal variable drives replication — starts at zero and wanders until it first hits a threshold $L$. That moment, $T_L=\inf \{t>0:x(t)\ge L\}$, triggers the creation of $m$ offspring, each restarted at zero. The process then recurs in each lineage, independently.

The framework imposes almost no constraints on the underlying dynamics of $x(t)$. It could be a simple diffusion with drift, a multi-step chemical cascade, a discrete Markov chain — anything, provided a first-passage time distribution $f_L(t)$ exists. This generality is what gives the framework its broad reach.

The mathematics centers on a renewal equation — an equation that describes how the population density at time $t$ is built up from all the branching events that happened before $t$. Taking the Laplace transform (a mathematical tool that converts time-domain equations into a simpler algebraic form) and analyzing the long-time behavior, the authors derive a condition on the population growth rate $\lambda$:

$$m⟨e^{-\lambda T_L}⟩=1$$

This is what they call a generalized Euler-Lotka equation. The classical Euler-Lotka equation is a cornerstone of population biology — it links age-specific birth rates to overall population growth in human demography. Kumar and Holehouse's version replaces age-structured birth rates with first-passage statistics, making the biological clock endogenous rather than assumed.

What They Found

The first major result is a universal inequality. Because $e^{-\lambda t}$ is a strictly convex function of $t$, a standard mathematical result called Jensen's inequality applies. It says that, for any random variable $T_L$ with mean $⟨T_L⟩$, the expected value $⟨e^{-\lambda T_L}⟩$ is always at least as large as $e^{-\lambda ⟨T_L⟩}$. Plugging this into the Euler-Lotka condition, the authors derive:

$$\lambda \ge \frac{\ln m}{⟨T_L⟩}$$

The right-hand side is precisely the growth rate you would get if every replication event happened exactly at the mean time, with no randomness at all — a perfectly deterministic clock. The inequality says the actual growth rate $\lambda$ is always at least as large as that deterministic benchmark. Noise never hurts population growth when offspring number is fixed. This holds regardless of the specific form of the first-passage process.

The paper validates this across multiple model systems — drift-diffusion processes (random walks with a directional bias), multi-state Markov chains, and others — and the theory agrees with simulations throughout.

Noise Always Helps: Growth Rate vs. Timing Variability (Fixed Yield, m=2)

Label	Value
CV²=0 (deterministic)	0.693
CV²=0.1	0.72
CV²=0.25	0.76
CV²=0.5	0.82
CV²=1.0	0.91
CV²=2.0	1.05

At fixed mean replication time ⟨T_L⟩ and fixed offspring m=2, the growth rate λ is always ≥ ln(m)/⟨T_L⟩ = ln 2 ≈ 0.693, the deterministic lower bound. Qualitative trend from Figure 2b; exact values are illustrative of the monotonic increase shown.

summarizes the key result: as the coefficient of variation squared ($\mathrm{CV}^2 = \mathrm{Var}(T_L)/\langle T_L \rangle^2$, a normalized measure of timing variability) increases from zero, the growth rate monotonically rises above the deterministic baseline. A perfectly regular clock ($\mathrm{CV}^2 = 0$) is the worst-case scenario. Any amount of scatter helps.

That's a remarkable statement. It means evolution has no incentive to smooth out replication timing, at least in the fixed-yield case — and may actively benefit from retaining molecular noise that would otherwise look like mere sloppiness.

The picture becomes more nuanced when offspring yield depends on timing. In bacteriophage, for instance, a virus that waits longer inside its host accumulates more viral particles before lysing. The burst size $m(T_L)$ is a function of the lysis time — roughly linear after an initial eclipse period during which no new particles are assembled at all. Now there's a genuine tension. Waiting produces more offspring per cycle. But waiting also delays the entire downstream lineage. Every extra minute inside the bacterium is a minute the new viral particles aren't out infecting new hosts.

To handle this, the authors extend their Euler-Lotka equation to:

$$⟨e^{-\lambda T_L}m(T_L)⟩=1$$

Here $\lambda$ is implicitly the discount rate that makes the expected reproductive value of one replication cycle equal to one — a concept imported directly from economics and life-history theory. The authors then derive an elegant exact decomposition:

$$\lambda =\frac{⟨\ln m(T_L)⟩+D_{\mathrm{KL}}(f_L\Vert q)}{⟨T_L⟩}$$

where $D_{\mathrm{KL}}(f_L\Vert q)$ is the Kullback-Leibler (KL) divergence — a measure of how different two probability distributions are, borrowed from information theory — between the raw first-passage distribution $f_L(t)$ and a growth-weighted version $q(t)=e^{-\lambda t}m(t)f_L(t)$. The KL divergence is always non-negative, which means population-level selection always adds something on top of the naive yield. It's a clean, information-theoretic expression of the fact that survivors aren't a random sample of all replication events.

When the yield function is a power law, $m(t)=m_0{t}^{\beta }$, the sign of the noise effect flips depending on parameters. For shallow yield curves ($\beta = 0.2$), randomness still helps. For steep yield curves ($\beta = 1.2$), it can hurt — because early-firing lineages produce far too few offspring to compensate for the growth-rate bonus they'd otherwise get.

Yield Slope Changes Everything: Δλ vs. CV² for Different Power-Law Yields

Label	Value
CV²=0.05	0.002
CV²=0.1	0.004
CV²=0.2	0.008
CV²=0.4	0.015
CV²=0.6	0.021

Δλ = λ − λ_det, the deviation from the deterministic growth rate. Positive = noise helps; negative = noise hurts. Values qualitatively derived from Figure 3a trends; exact numerical magnitudes are representative of the paper's results.

tracks this sign change, showing the growth-rate deviation from the deterministic baseline as timing variability increases.

Why This Changes Things

The application to bacteriophage lysis is where the framework really earns its keep. Phage biology has long been recognized as a natural optimization problem. A phage that lyses too early releases few particles; one that waits too long lets competitors colonize other hosts first. But previous analytical treatments required assumptions about the form of the lysis-time distribution that were either ad hoc or hard to validate.

Kumar and Holehouse's approach is cleaner. They model the infection cycle in two stages: an extracellular search phase, during which the released phage diffuses through the environment until it adsorbs onto a new host cell (or gets degraded), and an intracellular production phase, governed by the first-passage dynamics inside the bacterium. The generalized Euler-Lotka equation for the full phage life cycle becomes:

$${\tilde{g}}_A(\lambda +{\delta }_P)⟨e^{-\lambda T_L}m(T_L)⟩=1$$

where ${\tilde{g}}_A$ is the Laplace transform of the adsorption-time distribution and ${\delta }_P$ is the free-phage degradation rate. This compact equation folds together the extracellular ecology and the intracellular molecular dynamics into a single expression.

Using experimentally measured parameters from Shao and Wang's classic phage study — an eclipse period $E=28$ min, maturation rate $v\simeq 7.7$ phage/min, degradation rate ${\delta }_P=0.163$ hr$^{-1}$, and adsorption rate $a\simeq 0.178$ hr$^{-1}$ — the authors solve the optimization problem analytically. The answer: an optimal lysis time of ${\tau }^{\ast }\approx 49.9$ min and an optimal growth rate of ${\lambda }^{\ast }\approx 2.74$ hr$^{-1}$.

The experimentally measured best genotype has a lysis time of $51.0\pm 0.65$ min and a growth rate of $2.83\pm 0.024$ hr$^{-1}$. The match is striking — a first-principles model, built from nothing but threshold-crossing mathematics and measured biological parameters, recovers the evolved optimum to within measurement precision.

Bacteriophage Lysis: Theory vs. Experiment

Label	Value
Theory (this paper)	49.9
Wang fitted optimum	44.62
Best measured genotype	51

Theoretical prediction uses eclipse period E=28 min, maturation rate v=7.7 phage/min, free-phage removal rate δ_P=0.163 hr⁻¹, adsorption rate a=0.178 hr⁻¹. Best measured genotype has uncertainty ±0.65 min and ±0.024 hr⁻¹. From Table/text of Kumar & Holehouse (2026).

places these numbers side by side.

The broader significance goes beyond phage. The same trade-off structure — yield rises with waiting, but delay penalizes downstream generations — appears everywhere threshold-triggered replication exists. Bacterial division, where size at division correlates with daughter-cell fitness. Stem cell differentiation, where the timing of commitment affects cell fate. Even tumor growth, where cell-cycle duration shapes how quickly a cancer population expands. The Euler-Lotka equation the authors derive is exact, not approximate, and the framework imposes almost nothing on the underlying dynamics.

There's also a conceptual payoff for evolutionary biology. The Jensen's inequality result — that noise in replication timing is always growth-beneficial at fixed yield — suggests a kind of evolutionary tolerance for molecular stochasticity. Cells and viruses may not need to evolve precision clocks. The variance itself is doing useful work, exploring the early tail of the first-passage distribution and boosting the exponential growth rate. Evolution, in this framing, doesn't need to eliminate noise. It just needs to set the mean correctly.

This connects to a much older debate in life-history theory about bet-hedging — the strategy of spreading offspring across variable environments rather than committing all resources to a single optimized bet. What Kumar and Holehouse show is something slightly different: even without any environmental uncertainty, pure stochasticity in the internal replication clock provides a systematic growth advantage. It's a Jensen's inequality argument in biological clothing.

What's Next

The framework has clear limits the authors acknowledge openly. The current model assumes unlimited resources — each cell or virus replicates without competing for a finite pool of hosts, nutrients, or space. In real populations, density dependence and resource constraints reshape everything. Extending the renewal-equation approach to resource-limited environments is, the authors note, an important next step.

The deterministic approximation used in the phage lysis calculation — assuming $T_L=\tau$ exactly — is another simplification that future work could relax. The full stochastic treatment, including intrinsic noise in holin accumulation, would likely shift the predicted optimum slightly and might reveal additional structure in how the optimal lysis time depends on environmental parameters like host density.

There's also a rich set of mathematical questions the paper opens up. The framework currently tracks the expected population density. What about fluctuations around that expectation — the variance, the probability of extinction, the distribution of population sizes? Branching process theory has deep results on these questions in the classical setting; translating them to the first-passage-resetting framework is non-trivial but likely tractable.

Perhaps most intriguing is the information-theoretic decomposition in Eq. (10): the KL divergence between the first-passage distribution and the growth-weighted distribution appears naturally as a contribution to the population growth rate. This is reminiscent of results in the statistical mechanics of evolution, where population growth can be linked to the information content of the fitness distribution. Whether there's a deeper connection — a thermodynamic or information-theoretic interpretation of the Euler-Lotka condition — remains to be explored.

For now, what Kumar and Holehouse have built is a rigorous, general, and surprisingly predictive theory of how randomness at the molecular level shapes growth at the population level. The message is counterintuitive but clean: when it comes to biological clocks, imprecision isn't a flaw to be corrected. In many circumstances, it's the very thing that keeps a lineage ahead.