Bounds for survival probabilities in supercritical Galton-Watson processes and applications to population genetics

Imagine a tiny, lucky mutation appearing in a vast population of organisms. This mutation gives its carrier a slight advantage—maybe they can run a bit faster or digest food a little better. The big question for evolutionary biologists is: Will this lucky mutation take over the whole population, or will it fizzle out and disappear?

This paper, written by Reinhard Bürger, is essentially a mathematical toolkit for answering that question with extreme precision, especially when the mutation is just starting out.

Here is the breakdown of the paper's ideas using simple analogies:

1. The "Family Tree" Game (Galton-Watson Processes)

Think of the mutation as a single ancestor starting a family tree. In every generation, this ancestor has children.

The Lucky Scenario: On average, the mutant has more than one child (say, 1.1 children). This is a "supercritical" process. The family could grow forever.
The Unlucky Reality: Even if the average is high, there is a real chance that in the first generation, the mutant has zero children. If that happens, the family tree ends immediately.

Biologists need to know the probability that this family tree survives for $n$ generations. This is called the Survival Probability, $S(n)$ .

2. The Problem: Math is Too Hard

For simple cases (like a coin flip for offspring), we have exact formulas. But in real biology, the number of offspring is messy. It might follow a "Poisson" distribution (like random raindrops hitting a roof) or a "Binomial" distribution (like rolling dice).

When the math gets messy, calculating the exact survival probability for a specific number of generations is like trying to predict the exact path of a leaf swirling in a storm. It's computationally heavy and often impossible to write down as a simple formula.

3. The Solution: The "Shadow" Method

The author's main trick is to build a mathematical shadow of the messy reality.

Imagine you have a weirdly shaped, bumpy rock (the real, messy mutation distribution). You want to know how much light it blocks. Instead of measuring the bumpy rock, you find a smooth, simple shape (a Fractional Linear distribution) that fits perfectly around it.

The Shadow: This simple shape is easy to calculate.
The Fit: The author proves that for many common biological distributions, this simple shape acts as a ceiling (an upper bound) or a floor (a lower bound) for the real probability.
The Magic: Even though the shadow isn't the rock, it mimics the rock's behavior so closely that by the time you look far enough into the future, the shadow and the rock are practically identical.

4. Why Do We Need Bounds? (The "Time to Settle" Problem)

In the real world, we don't just care if the mutation survives forever; we care about the speed of evolution.

Imagine a trait (like beak size) changing over time. This change is driven by many mutations sweeping through the population. To calculate how fast the trait changes, scientists need to integrate (add up) the effects of these mutations over time.

The Hurdle: You can't integrate a messy, unknown function easily.
The Fix: The author provides a "stopwatch." He calculates exactly how many generations it takes for the survival probability to settle down and become stable.
The Result: Once the mutation survives past this "settling time," we can stop doing complex math and just use a simple constant number. This turns a nightmare calculation into a simple arithmetic problem.

5. The "Haldane" Connection

The paper pays homage to J.B.S. Haldane, a giant in genetics from the 1920s. Haldane famously guessed that if a mutation gives a 1% advantage ( $s=0.01$ ), its chance of survival is roughly $2% $($ 2s$).

This paper says: "Haldane was right, but here is the exact correction."
The author shows that for very small advantages, Haldane's rule is a great first guess, but if you want to be precise (like a surgeon), you need the author's new formulas. These formulas tell you exactly how much Haldane's rule overestimates or underestimates the truth depending on the specific "family rules" of the organism.

6. Real-World Application: The "Quantitative Trait"

The paper ends with a practical application: Directional Selection.
Imagine a population of birds where the wind is blowing harder every year, and the birds need bigger wings to survive. This isn't just one mutation; it's a continuous stream of small mutations adding up.

Without this paper: Predicting how fast the birds' wings will grow is a black box.
With this paper: Scientists can use these "shadow bounds" to calculate the exact speed of the wing growth. They can say, "With these specific family rules, the average wing size will increase by X millimeters per century."

Summary

Reinhard Bürger has built a mathematical safety net.

He takes complex, messy biological scenarios.
He wraps them in simple, solvable "shadows" (bounds).
He proves these shadows are tight enough to be accurate but loose enough to be easy to calculate.
This allows evolutionary biologists to predict the future of populations with much greater confidence, turning a chaotic storm of probabilities into a clear, navigable path.

In a nutshell: It's about finding a simple ruler to measure a bumpy mountain, so we can finally predict exactly how fast the mountain is growing.

Here is a detailed technical summary of the paper "Bounds for survival probabilities in supercritical Galton-Watson processes and applications to population genetics" by Reinhard Bürger.

1. Problem Statement

The paper addresses the challenge of approximating the survival probability $S(n)$ of a single beneficial mutation up to generation $n$ in a finite, supercritical Galton-Watson (GW) branching process.

Context: In population genetics, while the fixation probability ( $S_\infty$ ) of a single mutant is well-studied (e.g., Haldane's approximation $2s $), the time-dependent survival probability$ S(n)$ is crucial for modeling evolutionary processes occurring on longer time scales, such as the adaptation of quantitative traits under directional selection.
The Gap: Existing methods (like diffusion approximations) struggle to provide analytically explicit expressions for the time-course of allele frequencies. Previous branching process approaches often relied on specific assumptions about offspring distributions (e.g., Poisson) or lacked rigorous, simple bounds for $S(n)$ for general distributions.
Goal: To derive sharp, analytically explicit, and simple upper or lower bounds for $S(n)$ for various offspring distributions, ensuring the bounds converge to the true survival probability $S_\infty$ at the correct asymptotic rate.

2. Methodology

The author employs a method pioneered by Seneta (1967) and attributed to P.A.P. Moran, utilizing fractional linear (FL) generating functions to bound a given probability generating function (pgf), $\phi(x)$ .

Core Construction:
- For a given supercritical process with pgf $\phi$ , mean $m > 1$ , and extinction probability $P_\phi^\infty$ , the author constructs a fractional linear pgf, $\phi_{FL}(x)$ , defined by parameters $\pi$ and $\rho$ .
- Matching Conditions: The FL function is uniquely determined to satisfy two conditions at the extinction point $P_\phi^\infty$ $P_{ϕ}^{\infty}$ :
  1. $\phi_{FL}(P_\phi^\infty) = P_\phi^\infty$ (same extinction probability).
  2. $\phi'_{FL}(P_\phi^\infty) = \phi'(P_\phi^\infty) = \gamma_\phi$ (same rate of convergence).
- Bounding Logic: If $\phi_{FL}(x) \leq \phi(x)$ for all $x \in [0, P_\phi^\infty]$ , then the extinction probability of the FL process provides a lower bound for the true extinction probability (and thus an upper bound for the survival probability $S(n)$ ). Conversely, if $\phi_{FL}(x) \geq \phi(x)$ , it provides a lower bound for survival.
Analytical Tools:
- Series Expansions: Since $P_\phi^\infty$ and $\gamma_\phi$ are often not analytically solvable for complex distributions, the author derives series expansions in terms of the selective advantage $s$ (where $m=1+s$ ) to approximate these parameters.
- Generalized Haldane Approximation: The paper connects these bounds to generalized versions of Haldane's approximation ( $S_\infty \approx 2s$ ) and bounds derived by Quine (1976) and Daley & Narayan (1980).

3. Key Contributions and Results

A. General Theoretical Framework

Proposition 3.1: Establishes that if the constructed FL pgf bounds the true pgf on the interval $[0, P_\phi^\infty]$ , then the survival probability $S(n)$ is bounded by the explicit formula:
$S_\phi^\infty \leq S(n) \leq \frac{S_\phi^\infty}{1 - \gamma_\phi^n (1 - S_\phi^\infty)}$
(or the reverse inequality, depending on the direction of the bound).
Convergence Rate: The constructed bounds converge to $S_\infty$ at the exact geometric rate $\gamma_\phi$ , ensuring high accuracy for large $n$ .

B. Validation for Specific Distributions

The paper proves the validity of the bounding method for several standard families:

Poisson Distribution: Proven that the FL bound is an upper bound for $S(n)$ (lower bound for extinction) for all $x \in [0, 1]$ . The proof utilizes the Lambert $W$ function.
Binomial Distribution: Proven that the FL bound is an upper bound for $S(n)$ for all $x \in [0, 1]$ .
Negative Binomial Distribution: Proven that the FL bound is an upper bound for $S(n)$ for $x \in [0, P_\phi^\infty]$ . The author conjectures it holds for $x \in [0, 1]$ .
Distributions with at most 3 offspring: A complete characterization is provided. Depending on the parameters ( $p_0, p_2, p_3$ $p_{0}, p_{2}, p_{3}$ ), the FL approximation can be:
- A strict upper bound.
- A strict lower bound.
- An approximation that switches from lower to upper (or vice versa) at a specific generation $n$ .
- The paper maps these regimes in parameter space (Figures 4.2 and 4.3).

C. Series Expansions and Generalized Poisson

Series Expansions: The author derives explicit series expansions for $S_\infty$ and $\gamma_\phi$ in powers of $s$ for Poisson, Binomial, Negative Binomial, and Generalized Poisson distributions.
Generalized Poisson (Consul-Jain): This distribution is analytically intractable for exact $S_\infty$ $S_{\infty}$ . The paper uses the series expansion method to derive approximate bounds.
- It identifies critical values of the parameter $\lambda$ where the nature of the bound changes (from upper to lower or vice versa).
- For small $\lambda$ , the FL bound is an upper bound; for large $\lambda$ , it becomes a lower bound.

D. Comparison with Existing Bounds

The derived bounds are compared with Pollak's (1971) and Agresti's (1974) bounds.
Result: Pollak's and Agresti's bounds are slightly more accurate for small $n$ because they match the second derivative at $P_\infty$ . However, the author's FL bounds are superior in analytical simplicity and tractability for general distributions where higher-order derivatives are difficult to compute.
The paper confirms that Quine's and Daley-Narayan's bounds for $S_\infty$ have errors of order $O(s^3)$ , consistent with the derived series expansions.

E. Application to Quantitative Trait Evolution

The paper applies these bounds to the problem of directional selection on a quantitative trait (building on Götsch and Bürger, 2024).
Significance: The response of a trait mean depends on the integral of the variance contributed by mutations over time. This integral requires $S(n)$ .
Result: Using the derived bounds, the author shows that for sufficiently large $n$ (specifically $n > T(\epsilon)$ ), $S(n)$ can be approximated by the constant $S_\infty$ . This allows the integral to be solved explicitly, yielding a simple formula for the asymptotic genetic variance and the response to selection.
The paper demonstrates that mutations lost early contribute negligibly to the total variance, validating the simplification used in quantitative genetics models.

4. Significance

Analytical Tractability: Provides a unified, simple method to bound survival probabilities for a wide class of offspring distributions without needing complex numerical iterations.
Bridging Theory and Application: Connects rigorous branching process theory with practical population genetics applications (quantitative trait evolution), offering explicit formulas where previous methods were semi-explicit or required numerical simulation.
Generalization: Extends the applicability of fractional linear bounds beyond the Poisson distribution to Binomial, Negative Binomial, and Generalized Poisson, characterizing exactly when these bounds hold.
Error Quantification: Provides explicit error estimates and convergence times ( $T(\epsilon)$ ), allowing researchers to determine precisely when the survival probability stabilizes to its asymptotic value.

In summary, this paper provides a robust mathematical toolkit for population geneticists to model the time-dependent dynamics of beneficial mutations, facilitating more accurate predictions of evolutionary responses in finite populations.