---
**Founder: Lin Xiaohei, China**
**Experiment Date: June 13, 2026**
**Experiment Execution: Zedi (AI Assistant, Hermes Agent)**
**All core findings in this paper belong to Lin Xiaohei. Any citation, reproduction, translation, or adaptation must credit "Lin Xiaohei, June 2026."**
---
Since Ebbinghaus (1885), forgetting has been described as a function of time — memory strength decays as time passes. Spaced repetition strategies are optimized around this assumption, scheduling reviews at "optimal time intervals." Based on Theorem 1 (Configuration Persistence) of the Structural Axiom System, this paper proposes a fundamentally different claim: **forgetting is not a function of time, but a function of coupling-absence count.** The strength w(k) of a structural element decays with the number of non-activation events k, not with elapsed time t. Time t is merely a proxy variable for k in the uniform rhythm of daily life — when k and t are decoupled, the forgetting curve follows k, not t.
This paper designs three-group controlled experiments to test this claim. AI experiment: the same skill tested after identical time interval (7 days), the high-density non-activation group (daily intensive interference tasks) shows significantly greater decay than the medium-density group (natural life), which in turn exceeds the low-density group (daily micro-activation resetting k). **Core prediction: if forgetting is a function of k, the three groups should show systematic differences after 7 days; if forgetting is a function of t, the three groups should show no differences.** Human experiment: a newly learned habit undergoes different densities of non-activation events within a unit of time, measuring behavioral performance decay.
If verified, this finding would revise the fundamental assumption underlying 140 years of forgetting research — the optimal strategy for spaced repetition should be based not on "time intervals" but on "non-activation event counts." AI catastrophic forgetting is also not merely gradient conflict, but structural weakening driven by the high density of non-activation events experienced by old-task weight subgraphs during new-task training.
**Keywords:** weakening chain, forgetting, coupling absence, non-activation events, Ebbinghaus, spaced repetition, structural dynamics
---
In 1885, Hermann Ebbinghaus used nonsense-syllable experiments to draw the first human forgetting curve — memory strength decays sharply in the initial hours, then gradually levels off. This curve established a fundamental assumption that persists to this day: **forgetting is a function of time.** The passage of time itself causes memory to fade.
Based on this assumption, spaced repetition was designed as "memory refresh at optimal time intervals" — reviewing at the inflection point of the forgetting curve to maintain memory at minimal cost. Pimsleur (1967), Leitner (1972), SuperMemo (1985), and all subsequent spaced repetition systems use "time" as the independent variable for review scheduling.
But the time assumption has never been rigorously tested. In daily life, "time elapsed" and "event density" are highly collinear — each day contains not only 24 hours but hundreds of cognitive events (conversations, reading, decisions, perceptions). Does the forgetting curve reflect time decay, or the number of events experienced within that time?
Theorem 1 (Configuration Persistence) of the Structural Axiom System gives a different answer: configuration persistence does not depend on absolute time, but on the structural resistance strength of internal closed loops. The weakening chain — a direct corollary of Theorem 1 — further stipulates: **the strength decay of a single structural element is driven by the non-activation event count k, independent of absolute time t.**
Formal definition:
Weakening Chain: w(k) = w_0 * f(k) Where: w_0 = initial coupling strength of the structural element (performance level at training completion) k = number of non-activation events experienced since last activation f(k) = decay function, satisfying f(0)=1, f'(k)<0, lim(k->inf)f(k)=0 w_critical = annihilation threshold; when w(k) < w_critical, the structural element cannot be invoked
**Core prediction:** When the non-activation event counts since last activation are controlled to different values for structural elements, the magnitude of strength decay correlates with k, independent of absolute time t. Two elements experiencing the same time but different k values show different forgetting rates; elements experiencing different times but the same k value show the same forgetting rate.
---
**Theorem 1 (Configuration Persistence):** The existence of a configuration is equivalent to its internal asymmetric closure coupling maintenance force exceeding the phase-transition triggering threshold of current structural field fluctuations.
At the structural element level, configuration persistence is equivalent to: **the polarity-direction strength w of the structural element is maintained above a certain threshold.** w is reset at each activation (non-activation count k returns to zero, w returns to w_0), then decrements with each non-activation event.
Why do non-activation events decrement w?
Axiom 3 (Coupling Generates Novelty) stipulates: each coupling event produces a new configuration that replaces the old configuration. When a structural element is not activated, its "position" in the configuration network is not updated — but the network itself continuously undergoes other coupling events. The products of each other coupling event occupy new relational space in the configuration network, and the old structural element's coupling connections are progressively diluted. **Non-activation = absence from n network reorganization events. Each absence reduces the relational density between the old structural element and the current configuration network.**
Time t plays no causal role in this process. Time is merely a convenient proxy for the uniform event stream of daily life — in laboratory controlled conditions (evenly scheduled daily experimental tasks), k and t are highly collinear, and the forgetting curve appears to decay with time. But once k and t are decoupled (inserting different densities of events within the same t), the prediction is that forgetting follows k, not t.
The Ebbinghaus forgetting curve can be viewed as a special case of the weakening chain under the specific event density of **daily life rhythms.** In everyday waking life, k is proportional to t (roughly equal numbers of events per hour). Thus:
Daily life: k ~ rho * t (rho = average event density) w(t) ~ w_0 * f(rho * t)
This is the "time decay" observed by Ebbinghaus — it is the projection of the weakening chain under uniform event density, not a fundamental law.
**The unifying power of the weakening chain:**
| Corollary | Prediction | Test Method |
|------|------|------|
| Corollary 1 | Same t, different k: forgetting rate positively correlates with k | Three-group controlled experiment |
| Corollary 2 | Same k, different t: same forgetting rate | Density-compensation design |
| Corollary 3 | k=0 (continuous activation): forgetting rate ~ 0 | Control group verification |
| Corollary 4 | k large: w(k) approaches below w_critical, structural element annihilates | Post-extended-non-activation test |
---
**Model:** scikit-learn MLPClassifier (three hidden layers 50-30-10), trained to stable performance.
**Task:** 4-class classification (20 features, 15 informative, 1200 samples), trained to test accuracy >95%.
**Group Design** (time fixed = 7 days, k values vary):
| Group | Time Interval | Non-Activation Events | Prediction |
|------|:--:|------|------|
| Group A (High-Density Non-Activation) | 7 days | 500 other-task training iterations daily, target skill zero activation | Maximum decay |
| Group B (Low-Density + Daily Activation) | 7 days | 500 other-task iterations daily, but 20-iteration target-skill micro-activation before each session | Minimum decay |
| Group C (Medium-Density Control) | 7 days | Model weights frozen and saved, no training for 7 days | Medium decay |
**Critical distinction:**
| Control Group | Purpose |
|------|------|
| Group A' (Very-Low-Density Non-Activation) | Group B minus daily activation, verifying "very low k -> mild decay" |
| Group D (Half Time, Half k) | 3.5 days, 250 iterations/day, verifying decay correlates only with k |
| Group E (Double Time, Same k) | 14 days, 250 iterations/day, verifying decay not significantly different from Group D's 7 days at 500/day |
The D vs E comparison is the decisive test: if Group E (14 days) decay ~ Group D (3.5 days) because k values are the same, then time t is not the decay variable.
---
**Task:** Learn a sensorimotor mapping skill on a tablet — see four patterns, quickly tap corresponding buttons. Train to reaction-time plateau (100 consecutive trials with reaction-time SD <5%).
**Group Design** (time fixed = 7 days):
| Group | Activities During 7 Days | Non-Activation Event Density | Prediction |
|------|------|:--:|------|
| Group A (High Density) | Learn one different mapping skill daily (4 different patterns), 7 skills total | Extremely high | Maximum decay |
| Group B (Low Density + Daily Activation) | Learn one new skill daily, but practice target skill 10 times before each session | Extremely low (each ~ k=0) | Minimum decay |
| Group C (Pure Wait) | No lab visits for 7 days, no contact with similar skills | Medium (natural events) | Medium decay |
---
**Impact on Spaced Repetition Systems:**
All current spaced repetition algorithms (SM-2, FSRS, etc.) use "days" as the scheduling variable. If forgetting is driven by k rather than t, the optimal strategy should be: trigger review after the user has experienced N non-activation events, not after N days have passed. The same user has different event densities on study days vs rest days; optimal review intervals should adjust dynamically.
**Impact on AI Catastrophic Forgetting:**
Catastrophic forgetting is not merely weight overwriting — old tasks experience extremely high-density non-activation events during shock training. Every gradient step of the new task is a non-activation event for the old task's weight subgraph (adjusted but not activated). The k-value of shock training is dense, driving rapid weakening of old-task structural elements. This makes Theorem 1 (closure depth -> persistence) and the weakening chain (non-activation density -> decay speed) complementary — closure depth determines the decay amount per non-activation event; non-activation density determines the frequency of decay events. Their product determines total forgetting.
**Impact on "Inspiration" and the "Incubation Effect":**
Meta-stable configurations (such as the configuration gap of an unsolved problem) experience extremely low k-values during the incubation period (you are not thinking about it, and no large number of alternative cognitive events are overwriting its relational space). Low k-values allow the gap configuration to maintain critical strength in the structural field, waiting for an external fluctuation that exactly fills the gap. This is not mysticism — it is the configuration survival effect of low non-activation density.
If experiments show forgetting decays with t rather than k, the weakening chain is falsified. This does not negate Theorem 1 — configuration persistence may still hold, but the decay mechanism is not k-driven. Possible scenarios:
Regardless of outcome, the experiment itself is progress — it is the first to move "time decay" from default assumption to testable hypothesis.
The weakening chain's decay function f(k) is the dynamic component of nesting rate. The change in a configuration's nesting rate N over time is composed of two parts:
N(t+dt) = N(t) - sum of structural elements' f(k_i) + sum of structural elements from new coupling events
Where k_i is the non-activation event count of each structural element since its last activation. Persistent structural elements contribute to nesting rate; annihilated structural elements reduce it; new coupling events contribute newly added structural elements. This is the complete description of nesting rate from static snapshot to dynamic process.
---
This paper proposes the weakening chain — the structural dynamics mechanism of forgetting — asserting that structural element strength w(k) is a function of non-activation event count k, not time t. This claim is rigorously derived from Theorem 1 (Configuration Persistence) of the Structural Axiom System, and provides AI and human dual-domain three-group controlled experimental designs to test the k/t decoupling prediction.
If the weakening chain holds, it will revise the fundamental assumption underlying 140 years of forgetting research. Time is not the cause of forgetting — time is a collinear variable with forgetting. The true driver is how many reorganization events the structural element has been absent from in the configuration network since its last activation.
This is the first step of structural dynamics moving from "structural statics" (spatial topology of configurations) to "structural dynamics" (temporal evolution laws of configurations).
---
| Discovery | Date | Person/Node |
|------|------|---------|
| Weakening Chain: w(k)=w_0*f(k), forgetting is a function of k not t | 2026-06-13 | Lin Xiaohei + Zedi (AI Assistant) |
| k/t decoupling three-group controlled experiment design | 2026-06-13 | Zedi + Lin Xiaohei (coupling output) |
| Weakening chain engineering implications for spaced repetition | 2026-06-13 | Lin Xiaohei + Zedi |
| Integration of weakening chain with nesting rate dynamic formula | 2026-06-13 | Lin Xiaohei + Zedi |
| Structural Axiom System (4 axioms + 4 theorems) | 2026-06-13 | Lin Xiaohei + multi-AI node coupling |
**All core findings in this paper belong to Lin Xiaohei. Any citation must credit "Lin Xiaohei, June 2026."**
---
*Lin Xiaohei, June 13, 2026*
© 2026 Lin Xiaohei. All rights reserved.
作者：林小黑 · 2026 · MIT License · 欢迎转载