# The Weakening Chain: A Stochastic Decay Proof


**2026年6月**

## ​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​‌‍​Formal Derivation of f(k) = w₀(1−α)^k, Half-Life Theorem, and Multi-Shock Attenuation

### Lin Xiaohei (林小黑) — June 21, 2026

**Supplement to**: 弱化链-旧制度湮灭的数学结构-林小黑.md (June 16, 2026)
**Status**: Mathematical formalization. Does NOT replace the original paper.

---

## ⚠️ Copyright Notice

**Founder: Lin Xiaohei (China).** Original: June 16, 2026. Mathematical formalization: June 21, 2026.

---

## Abstract

We formalize the weakening chain $f(k) = w_0(1-\alpha)^k$ as a discrete-time multiplicative stochastic process. From the four axioms, we prove: (1) the exponential decay form is the unique solution under the Markov property and independent shock assumption; (2) the half-life $k_{1/2} = \ln(0.5)/\ln(1-\alpha)$ provides a robust invariant for comparing institutional decay rates; (3) multi-source attenuation follows the product rule $C(k) = C_0 \prod_{i} (1-\alpha_i)^{k_i}$, with the total decay rate exceeding any single-source rate; (4) variable $\alpha(k)$ introduces a generalized weakening chain solvable by separation of variables. The weakening chain connects the micro-level AI forgetting dynamics (paper #24) to macro-level institutional phase transitions (paper #35), providing the missing quantitative parameter for Stage 1 of the five-stage universal derivation sequence.

---

## 1. The Weakening Chain as a Multiplicative Process

### 1.1 Structural Basis

**Axiom 2 (Difference Creates Being)**: An institution exists because it maintains a persistent asymmetry — a difference between its internal configuration and the external environment. This asymmetry is the institution's *persistence strength* $w$.

**Axiom 3 (Coupling Creates Novelty)**: When an institution couples with an incompatible external structure (e.g., AI replacing a human function), the coupling perturbs the institution's configuration. If the perturbation is *incompatible* — i.e., it cannot be absorbed into the existing configuration without structural change — the persistence strength decreases.

**Definition 1.1 (Weakening Chain).** Let $w_k$ be the persistence strength of an institution after $k$ incompatible coupling shocks. The *weakening chain* is the sequence $\{w_k\}_{k=0}^{\infty}$.

### 1.2 Derivation of the Exponential Form

**Theorem 1.1 (Weakening Chain Form).** Under the Markov property and time-homogeneous shock effect, the weakening chain takes the form:

$$w_k = w_0 \cdot (1-\alpha)^k$$

where $\alpha \in (0,1)$ is the *attenuation coefficient* — the fraction of persistence strength lost per incompatible shock.

*Proof.* At shock $k+1$, the institution with persistence $w_k$ encounters an incompatible coupling. The coupling perturbs the institution's configuration. By Axiom 3, the perturbation either:
- **Resolves compatibly**: the institution absorbs the perturbation without structural change ($w_{k+1} = w_k$)
- **Resolves incompatibly**: the institution's configuration changes, reducing persistence ($w_{k+1} < w_k$)

Let $\alpha \in (0,1)$ be the probability-weighted expected fractional loss per shock. Then:

$$\mathbb{E}[w_{k+1} \mid w_k] = w_k \cdot (1-\alpha)$$

By the Markov property (the effect of shock $k+1$ depends only on the current state $w_k$, not on the history of previous shocks), this recurrence has the unique solution:

$$w_k = w_0 \cdot (1-\alpha)^k$$

This holds for constant $\alpha$. When $\alpha$ varies with $k$ (non-homogeneous shocks), the solution becomes:

$$w_k = w_0 \cdot \prod_{i=1}^{k} (1-\alpha_i)$$

∎

**Corollary 1.1 (Deterministic Trajectory).** If the shock sequence is sufficiently dense (many small shocks rather than few large ones), the expected trajectory approximates the actual trajectory with high probability. By the law of large numbers for martingale differences, $\frac{1}{k} \sum \ln(1-\alpha_i) \to \mathbb{E}[\ln(1-\alpha)]$, giving:

$$\frac{w_k}{w_0} \to \exp(k \cdot \mathbb{E}[\ln(1-\alpha)]) \approx (1-\bar{\alpha})^k$$

for small variance in $\alpha$. ∎

---

## 2. Critical Threshold and Half-Life

### 2.1 Annihilation Condition

**Definition 2.1 (Critical Threshold).** An institution *annihilates* when its persistence strength falls below a critical threshold $w_c$ required to maintain its defining asymmetry (Axiom 2). Formally:

$$\text{Annihilation} \iff w_k < w_c$$

**Theorem 2.1 (Critical Shock Count).** The number of shocks required for annihilation is:

$$k_{\text{crit}} = \left\lceil \frac{\ln(w_c / w_0)}{\ln(1-\alpha)} \right\rceil$$

*Proof.* Set $w_0(1-\alpha)^k < w_c$. Taking logarithms (valid since $1-\alpha \in (0,1)$, making $\ln(1-\alpha) < 0$):

$$k \cdot \ln(1-\alpha) < \ln(w_c / w_0)$$

Since $\ln(1-\alpha) < 0$, dividing flips the inequality:

$$k > \frac{\ln(w_c / w_0)}{\ln(1-\alpha)}$$

Since $k$ must be an integer, take the ceiling. ∎

**Interpretation**: If $w_0 / w_c = 10$ (institution starts 10× above threshold), then $k_{\text{crit}} = \lceil \ln(0.1) / \ln(1-\alpha) \rceil = \lceil -2.303 / \ln(1-\alpha) \rceil$. For $\alpha = 0.05$: $k_{\text{crit}} = \lceil 2.303/0.0513 \rceil = \lceil 44.9 \rceil = 45$ shocks.

### 2.2 Half-Life

**Theorem 2.2 (Half-Life Invariant).** The number of shocks required to halve the institution's persistence strength is independent of the initial strength $w_0$:

$$k_{1/2} = \frac{\ln(0.5)}{\ln(1-\alpha)} \approx \frac{0.693}{\alpha} \quad (\text{for small } \alpha)$$

*Proof.* Set $w_0(1-\alpha)^k = w_0/2$. Cancel $w_0$: $(1-\alpha)^k = 1/2$. Taking logarithms: $k \ln(1-\alpha) = -\ln 2$, giving the result. The approximation $\ln(1-\alpha) \approx -\alpha$ for small $\alpha$ yields $k_{1/2} \approx 0.693/\alpha$. ∎

| $\alpha$ | $k_{1/2}$ (exact) | $k_{1/2}$ (approx) | Half-life at 3 shocks/year |
|:---:|:---:|:---:|:---:|
| 0.01 | 69.0 | 69.3 | 23.0 years |
| 0.03 | 22.8 | 23.1 | 7.6 years |
| 0.05 | 13.5 | 13.9 | 4.5 years |
| 0.10 | 6.6 | 6.9 | 2.2 years |
| 0.20 | 3.1 | 3.5 | 1.0 years |

---

## 3. Multi-Source Attenuation

**Theorem 3.1 (Independent Multi-Source Decay).** If an institution is subject to $m$ independent shock sources, each with attenuation coefficient $\alpha_i$ and shock count $k_i$, the total persistence strength is:

$$w(k_1, \ldots, k_m) = w_0 \cdot \prod_{i=1}^{m} (1-\alpha_i)^{k_i}$$

*Proof.* By independence of shock sources (each source perturbs a distinct aspect of the institution's configuration), the effects multiply. The institution survives only if ALL aspects maintain above-threshold persistence. The joint survival probability is the product of individual survival probabilities, each given by $(1-\alpha_i)^{k_i}$. ∎

**Corollary 3.1 (Accelerated Decay).** The effective attenuation coefficient under $m$ simultaneous sources is:

$$\alpha_{\text{eff}} = 1 - \prod_{i=1}^{m} (1-\alpha_i) > \max_i \alpha_i$$

Multi-source attenuation is strictly faster than any single-source attenuation. For small $\alpha_i$, $\alpha_{\text{eff}} \approx \sum_i \alpha_i$ (additive in the small-$\alpha$ limit).

*Proof.* The effective single-shock attenuation from all sources combined (assuming one shock from each source per time unit) is $1 - \prod_i (1-\alpha_i)$. By the union bound for independent events, $1 - \prod_i (1-\alpha_i) \geq 1 - (1 - \max_i \alpha_i) = \max_i \alpha_i$, with strict inequality when more than one $\alpha_i > 0$. ∎

**Corollary 3.2 (AI as Phase Transition Accelerator).** Adding AI as a shock source to pre-existing shock sources (generational change, economic crisis) accelerates decay from $\prod_{i \neq \text{AI}} (1-\alpha_i)^{k_i}$ to $\prod_{\text{all } i} (1-\alpha_i)^{k_i}$. The *acceleration factor* is $(1-\alpha_{\text{AI}})^{k_{\text{AI}}} < 1$ — strictly less than 1, strictly accelerating the decay.

---

## 4. Variable Attenuation Coefficient

### 4.1 The Generalized Weakening Chain

In realistic institutions, $\alpha$ is not constant. Early shocks (replacing low-level functions) may have small $\alpha$; later shocks (replacing decision-making functions) have large $\alpha$. This gives the *generalized weakening chain*:

$$w_k = w_0 \cdot \prod_{i=1}^{k} (1-\alpha_i)$$

where $\alpha_i = \alpha(i)$ is a function of the shock index.

**Theorem 4.1 (Solvability by Separation).** The generalized weakening chain is solvable in closed form whenever $\alpha(i)$ admits a closed-form sum for $\sum \ln(1-\alpha_i)$.

*Proof.* 

$$w_k = w_0 \cdot \exp\left(\sum_{i=1}^{k} \ln(1-\alpha_i)\right)$$

When $\alpha_i$ is small, $\ln(1-\alpha_i) \approx -\alpha_i$, giving:

$$w_k \approx w_0 \cdot \exp\left(-\sum_{i=1}^{k} \alpha_i\right)$$

This is solvable whenever the cumulative sum of $\alpha_i$ has a closed form. ∎

**Example 4.1 (Linearly Increasing α).** If $\alpha_i = \alpha_0 + \beta \cdot i$ (attenuation grows linearly with each shock), then:

$$\sum_{i=1}^{k} \alpha_i = k\alpha_0 + \beta \cdot \frac{k(k+1)}{2}$$

$$w_k \approx w_0 \cdot \exp\left(-k\alpha_0 - \beta \cdot \frac{k(k+1)}{2}\right)$$

This is a *super-exponential* decay — faster than the constant-$\alpha$ case. Institutions whose core functions are threatened late in the shock sequence collapse faster than the constant-$\alpha$ model predicts.

**Example 4.2 (Sigmoid α).** If $\alpha_i = \frac{\alpha_{\max}}{1 + e^{-\gamma(i - i_0)}}$ (attenuation ramps up sigmoidally), the early shocks have negligible effect, but after the inflection point $i_0$, the institution collapses rapidly. This models institutions that appear resilient until a "tipping point" is reached.

---

## 5. Connection to the Five-Stage Phase Transition

The weakening chain provides the missing quantitative parameter for **Stage 1** of the five-stage universal derivation sequence:

| Stage | Original Description | Weakening Chain Parameter |
|-------|---------------------|--------------------------|
| 1. Old asymmetry dissolution | $f(k) = w_0(1-\alpha)^k$ | $k_{\text{crit}}$ = when dissolution completes |
| 2. Blind spot entry | New config enters through gap | Triggered when $w_k < w_c$ |
| 3. Identification & immune response | Old config detects new | Probability $\propto 1 - w_k/w_0$ |
| 4. Critical point | Phase transition | Triggered when $k \geq k_{\text{crit}}$ |
| 5. New phase stabilization | New inter-reference loops form | Duration $\propto 1/\alpha_{\text{new}}$ |

The weakening chain transforms Stage 1 from a qualitative "asymmetry weakens" into a quantitative prediction: given estimates of $\alpha$ and shock frequency, we can compute WHEN the critical point will be reached.

---

## 6. Falsifiable Predictions

### Prediction 1: Exponential Decay in AI Forgetting

In AI systems, the weakening chain predicts that model performance on a task degrades as $P(k) = P_0(1-\alpha)^k$ under incompatible fine-tuning steps. **Test**: fine-tune a model on task A, then apply $k$ incompatible fine-tuning steps on task B, and measure the degradation of task A performance. The degradation should follow the exponential form.

### Prediction 2: Half-Life Invariance

The half-life $k_{1/2}$ should be independent of the initial performance level $P_0$. **Test**: vary initial training intensity to produce different $P_0$ values, then measure $k_{1/2}$. It should be constant for fixed $\alpha$.

### Prediction 3: Multi-Source Acceleration

Two simultaneous incompatible shock sources should produce faster decay than either alone. The effective $\alpha_{\text{eff}}$ should satisfy $\alpha_{\text{eff}} \approx \alpha_1 + \alpha_2$ (for small $\alpha$). **Test**: apply shocks from source 1, source 2, and both simultaneously. Compare decay rates.

---

## §7.1 An Unresolved Open Question

The weakening chain assumes that each shock is *incompatible* — it cannot be absorbed without structural change. But what determines compatibility? Two structures may be compatible at one nesting level and incompatible at another. The weakening chain currently treats compatibility as binary (compatible/incompatible), but structural reality may be continuous (degree of compatibility).

Is there a $[0,1]$-valued compatibility measure that generalizes the weakening chain to partially compatible shocks? The author's position on this question is not disclosed here.

---

*Lin Xiaohei, June 21, 2026*
*Mathematical formalization by Hermes Agent (则弟). Does not replace original paper (June 16, 2026).*

---

© 2026 林小黑 (Lin Xiaohei). All rights reserved.
公众号：今晚狗蛋看局
https://gitee.com/samforce/structural-cognition