5D Library › Power-Law Weighting and Bottom-Up Controllability in Multi-Tier Conditional Excitation Systems · 中文

Chapter 6 Potential and Realization

Two variables, not one

The proposition of the previous chapter says that if the endogenous prior cannot change, no input from the bottom can move it. Yet the author’s intuition is not baseless: many people do feel that sustained learning has let their “original gifts” show, as if the top tier had been “unlocked”. The two are not in conflict once one variable is split into two:

  • potential PP: the endogenous prior itself, a constant in the model (an uncontrollable state);
  • realization RR: the degree to which the potential actually shows, for example R=P⋅g(z)R = P \cdot g(z), where zz is the state of the other tiers and gg lies between 0 and 1.

“Unlocking” in the model means raising gg, not raising PP. The distinction is not a play on words: it marks the boundary between what can and cannot be done, and it determines how claims can be tested.

Note

The author’s definitions. “Potential” and “realization” are the author’s definitions, introduced in this book to clarify the author’s framework; the form of gg (here an S-shaped function) is an illustrative choice. In reality PP cannot be measured directly: what we observe is always some realization. Claims of the form “this person has high potential that was never realized” are therefore untestable without an independent measure.

Realization has a ceiling

Calculation

The author’s model. Potential PP is fixed. Tiers 2–5 are coupled as in the UP structure of Chapter 5, with a constant input uu on C5C_5. Realization is R=P⋅g(z‾)R = P \cdot g(\bar z), where z‾\bar z is the mean deviation of tiers 2–5 and g(z)=1/(1+e−(2z−2))g(z) = 1/(1 + e^{-(2z – 2)}), so g(0)=0.12g(0) = 0.12 and g→1g \to 1 only as z‾→∞\bar z \to \infty. The program computes steady states and realization for several inputs, compares two invented people with potentials 0.5 and 0.9, and checks “output controllability” in the linear model: C1C_1 is uncontrollable, but is the output y=z‾y = \bar z built from tiers 2–5 controllable?

# Potential versus realization (author's model, Chapter 6).
# Potential P (layer C1) is fixed: an uncontrollable state (p10). Layers 2-5 follow the UP
# coupling of p10 with a constant input u on C5. Realized level of C1: R = P g(zbar),
# g(z) = 1 / (1 + exp(-(2 z - 2))), zbar = mean of layers 2-5, so g(0) = 0.12 and g -> 1
# only as zbar -> infinity: R can approach P, never exceed it. Linear version: the output
# y = zbar is output-controllable although c1 is not. All numbers illustrative.
import math

A = [[-1.0, 0.4, 0.0, 0.0], [0.5, -1.0, 0.4, 0.0], [0.0, 0.5, -1.0, 0.4],
     [0.0, 0.0, 0.5, -1.0]]                      # layers 2-5 of structure UP
B, c = [0.0, 0.0, 0.0, 1.0], [0.25] * 4

def mv(v): return [sum(a * b for a, b in zip(r, v)) for r in A]
def g(z): return 1 / (1 + math.exp(-(2 * z - 2)))

def steady(u):                                   # solve 0 = A z + B u by elimination
    X = [row[:] + [-b * u] for row, b in zip(A, B)]
    for j in range(4):
        for i in range(4):
            if i != j: X[i] = [a - X[i][j] / X[j][j] * b for a, b in zip(X[i], X[j])]
    return [X[i][4] / X[i][i] for i in range(4)]

if __name__ == "__main__":
    print("constant input u on C5 -> steady state of layers 2-5 and realization R = P g")
    print("    u    C2    C3    C4    C5   g(zbar)   R (P = 0.5)   R (P = 0.9)   gap")
    for u in (0.0, 1.0, 2.0, 4.0, 8.0, 16.0):
        z = steady(u); gz = g(sum(z) / 4)
        print(f"  {u:4.0f} " + " ".join(f"{x:5.2f}" for x in z)
              + f" {gz:8.4f} {0.5 * gz:12.3f} {0.9 * gz:13.3f} {0.4 * gz:6.3f}")
    print("potential P is the same in every row; R / P = g < 1 for every finite input")
    k, v = [], B[:]
    for _ in range(4):
        k.append(sum(a * b for a, b in zip(c, v))); v = mv(v)
    print("output y = mean of layers 2-5: c A^j B, j = 0..3 = "
          + " ".join(f"{x:.4f}" for x in k))
    print("  not all zero -> y is output-controllable although c1 is not")
    h, vs, v = 0.005, [], B[:]
    for _ in range(2001):
        vs.append(sum(a * b for a, b in zip(c, v)))
        v = [a + h * b for a, b in zip(v, mv(v))]
    wy = sum(x * x for x in vs) * h
    print(f"minimum energy to raise y by one unit in T = 10: {1 / wy:.1f}")
    print("(to raise C5 alone: 1.75; to raise C1 itself: impossible)")
# constant input u on C5 -> steady state of layers 2-5 and realization R = P g
#     u    C2    C3    C4    C5   g(zbar)   R (P = 0.5)   R (P = 0.9)   gap
#      0  0.00  0.00  0.00  0.00   0.1192        0.060         0.107  0.048
#      1  0.15  0.36  0.73  1.36   0.3318        0.166         0.299  0.133
#      2  0.29  0.73  1.45  2.73   0.6457        0.323         0.581  0.258
#      4  0.58  1.45  2.91  5.45   0.9608        0.480         0.865  0.384
#      8  1.16  2.91  5.82 10.91   0.9998        0.500         0.900  0.400
#     16  2.33  5.82 11.64 21.82   1.0000        0.500         0.900  0.400
# potential P is the same in every row; R / P = g < 1 for every finite input
# output y = mean of layers 2-5: c A^j B, j = 0..3 = 0.2500 -0.1500 0.1400 -0.1640
#   not all zero -> y is output-controllable although c1 is not
# minimum energy to raise y by one unit in T = 10: 14.4
# (to raise C5 alone: 1.75; to raise C1 itself: impossible)

There are four results. First, the potential PP is the same in every row: raising the input from 0 to 16 changes PP in none of them. Second, the realized fraction gg rises from 0.12 to nearly 1: 0.96 at u=4u = 4 and 0.9998 at u=8u = 8. For the person with potential 0.5, realization rises from 0.060 to nearly 0.5; for the person with potential 0.9, from 0.107 to nearly 0.9. Realization can approach potential but not exceed it. The author’s “C5=100%C_5 = 100\%” corresponds in the model to a large input with gg close to 1: it pushes realization towards the ceiling, but the ceiling itself does not move.

Third, in a multiplicative model the same input widens the absolute gap between the two people: 0.048 with no input, 0.40 with a large one. This follows directly from the multiplicative structure (R1−R2=(P1−P2)gR_1 – R_2 = (P_1 – P_2)g) and points the same way as the cumulative advantage of Chapter 4. If the aim is to narrow gaps, the same investment should go disproportionately to those whose realized fraction is low, rather than being shared out equally.

Fourth, the output yy (the mean of tiers 2–5) in the linear model is output-controllable: the cAjBcA^jB are not all zero, and the minimum energy to raise yy by one unit is 14.4. This is the precise meaning of “unlocking”: the state c1c_1 is uncontrollable; the realization built from the other tiers is controllable.

Same input, opposite conclusions

The previous chapter showed that controllability depends only on the coupling structure. This section goes further: under different coupling assumptions, even the direction of the effect of “more knowledge investment” differs.

Calculation

A constant unit input is applied to C5C_5 and the change in every tier and in ln⁡S=c1+⋯+c5\ln S = c_1 + \cdots + c_5 is read at t=40t = 40. Three couplings: UP (Chapter 5); PL (plastic prior, ε=0.05\varepsilon = 0.05); CROWD (as UP, but hours of study are taken from networking: a coupling of −0.9 from C5C_5 to C3C_3). Then, keeping the zero pattern fixed, every non-zero coupling is redrawn at random 1,000 times (random sign, magnitude between 0.1 and 1) to see whether the Kalman rank changes with the values.

# Same input, different coupling assumptions, opposite conclusions (Chapter 6).
# A constant unit input on C5 runs from t = 0 to 40; read the change of every layer and of
# ln S = c1 + ... + c5 at t = 40. UP and PL as in p10; CROWD = UP plus crowding-out
# (study hours taken from networking): C5 -> C3 = -0.9. Then the zero pattern is kept and
# the nonzero couplings are redrawn at random 1,000 times: the Kalman rank belongs to the
# pattern, not to the numbers (structural controllability, Lin 1974).
import random

def structure(name, k=0.5, h=0.4, eps=0.05, d1=0.1):
    A = [[0.0] * 5 for _ in range(5)]
    for i in range(1, 5): A[i][i] = -1.0
    for i in range(4): A[i + 1][i] = k
    for i in range(1, 4): A[i][i + 1] = h
    if name == "PL": A[0][0], A[0][1] = -d1, eps
    if name == "CROWD": A[2][4] = -0.9
    return A

def respond(A, T=40.0, h=0.01):
    f = lambda x: [sum(a * b for a, b in zip(r, x)) + (i == 4) for i, r in enumerate(A)]
    step = lambda x, k, s: [a + s * b for a, b in zip(x, k)]
    x = [0.0] * 5
    for _ in range(int(T / h)):
        k1 = f(x); k2 = f(step(x, k1, h / 2)); k3 = f(step(x, k2, h / 2))
        k4 = f(step(x, k3, h))
        x = [a + h / 6 * (b + 2 * c + 2 * d + e)
             for a, b, c, d, e in zip(x, k1, k2, k3, k4)]
    return x

def kalman_rank(A, tol=1e-9):
    M = [[0, 0, 0, 0, 1.0]]
    for _ in range(4): M.append([sum(a * b for a, b in zip(r, M[-1])) for r in A])
    r = 0; scale = max(abs(x) for c in M for x in c)
    for j in range(5):
        p = max(range(r, 5), key=lambda i: abs(M[i][j]), default=None)
        if p is None or abs(M[p][j]) < tol * scale: continue
        M[r], M[p] = M[p], M[r]
        for i in range(5):
            if i != r: M[i] = [a - M[i][j] / M[r][j] * b for a, b in zip(M[i], M[r])]
        r += 1
    return r

if __name__ == "__main__":
    print("change at t = 40 after a constant unit input on C5")
    print("  coupling    C1      C2      C3      C4      C5    ln S")
    for name in ("UP", "PL", "CROWD"):
        x = respond(structure(name))
        print(f"  {name:6s} " + " ".join(f"{v:+7.3f}" for v in x) + f" {sum(x):+7.3f}")
    rng = random.Random(41)
    for name in ("UP", "PL"):
        A0 = structure(name); ranks = {}
        for _ in range(1000):
            A = [[rng.choice((-1, 1)) * rng.uniform(0.1, 1.0) if a else 0.0 for a in row]
                 for row in A0]
            r = kalman_rank(A); ranks[r] = ranks.get(r, 0) + 1
        print(f"  {name} pattern, 1,000 random couplings: Kalman rank counts {ranks}")
# change at t = 40 after a constant unit input on C5
#   coupling    C1      C2      C3      C4      C5    ln S
#   UP      +0.000  +0.145  +0.364  +0.727  +1.364  +2.600
#   PL      +0.096  +0.209  +0.405  +0.753  +1.376  +2.839
#   CROWD   +0.000  -0.445  -1.113  -0.195  +0.902  -0.851
#   UP pattern, 1,000 random couplings: Kalman rank counts {4: 1000}
#   PL pattern, 1,000 random couplings: Kalman rank counts {5: 1000}

Under UP a unit input raises ln⁡S\ln S by 2.60, with C1C_1 unchanged; under PL by 2.84, of which C1C_1 contributes only 0.096; under CROWD, C5C_5 still rises by 0.90 but C3C_3 falls by 1.11 and drags down C2C_2 (−0.45) and C4C_4 (−0.20), so that ln⁡S\ln S actually falls by 0.85. The same input “unlocks” the upper tiers under one assumption and “locks” them under another. Which assumption is closer to reality is not a question mathematics can answer; only measurement can.

The random redraws show that all 1,000 sets of values with the UP zero pattern give rank 4, and all with the PL pattern give rank 5. This is structural controllability: whether a tier can be reached is fixed by “who influences whom”; how costly reaching it is, and whether the effect is positive or negative, is fixed by the values.

Hypothesis · not established

The realization-ceiling hypothesis (the author’s hypothesis, not established). For a given skill, sustained structured investment moves performance towards a personal asymptote that differs between people, and investment mainly changes how fast and how closely the asymptote is approached. Testable prediction: in a long training study, with exponential or S-shaped learning curves fitted to each person as registered in advance, the estimated personal asymptotes do not differ systematically between groups randomized to different training doses, while the fraction of the asymptote reached does. If the asymptote itself rises clearly with dose, the assumption of “fixed potential” has to be dropped in favour of a “plastic” model such as PL.

Note

Avoiding two kinds of fatalism. The mathematics of this chapter does not support “nature decides everything”: in any structure with bottom-up feedback, realization can be raised, and in the model the further the realized fraction is from 1, the more the same investment raises it. Nor does it support “effort can do anything”: under the stated assumptions potential is a ceiling, and under crowding-out an investment in the wrong direction backfires. Both slogans turn a mathematical result that depends on assumptions into a claim of fact about people.

Chapter summary

Once the endogenous prior is split into “potential” and “realization”, both of the author’s intuitions can hold: potential is uncontrollable in the model, realization is controllable. Sustained investment can push realization towards potential but not beyond it, and in a multiplicative structure the same investment widens the absolute gap between people with different potentials. Different coupling assumptions can reverse even the direction of the effect of the same input; whether a tier can be reached is fixed by the zero pattern of the couplings, the cost and direction by their values, and these call for measurement rather than reasoning. The next chapter looks at how classical Chinese writers handled the tension between fate and action.

Fan, Y. (2026). Power-Law Weighting and Bottom-Up Controllability in Multi-Tier Conditional Excitation Systems: A Mathematical Synthesis of Endogenous Priors, Timing Windows, Network Topology and Reputational Games [多层条件激发系统中的幂律权重与逆向控制可及性] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23810093 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).