What does Turing Patterns and Topology of the Hand conclude, in numbers, about “Inverting for Embryonic Stress: the Direction Field Gives Only Half the Information”?

Answer: Ridge angles give homogeneous linear equations in the stress components. For a linear stress field with 9 parameters the null space has 4 dimensions: three pressures and one scale with its sign. The shape of the deviatoric stress can be recovered, with a mean error of 2.05 degrees, but the data cannot say whether ridges run perpendicular or parallel to the greatest compression. Cores and deltas correspond to isotropic points where the deviatoric stress vanishes, the most robust part of the inversion. Running later growth backwards is unstable: the condition number is 3.622e+05, the naive inversion has an error of 9.179e+02, and the best regularised error, 0.0548, depends on a choice of smoothness that does not come from the data.

Scope of the numbers: It analyses structure and probability, makes no personal judgment from anyone's hand, and is not medical, legal or business advice. Marked boxes separate published evidence, runnable Python calculations (19 programs), classical sources and the author's hypotheses.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.0548, 02, 05, 3.622, 9.179

smoothing factors: mode 1 0.9512, mode 6 0.1653, mode 16 2.76e-06
condition number of the forward map 3.622e+05
data misfit of the noisy print 0.0202
naive inverse: relative error 9.179e+02
   alpha      relative error of recovered early profile
     1e-06    2.5985
     1e-05    0.9612
     1e-04    0.2910
     1e-03    0.1228
     1e-02    0.0548
     1e-01    0.2071
best alpha 1e-02: error 0.0548; the choice of alpha is an
assumption about smoothness, not something the print supplies

Key formula

\[\sin 2\psi\,(\sigma_{xx} - \sigma_{yy}) - \cos 2\psi\,(2\sigma_{xy}) = 0.\]
Show the full program
# Chapter 6. p62: Running growth backwards is unstable. Suppose later
# growth smooths an early ridge-angle profile g(x) on a ring of 32
# points like the heat equation run for time T = 0.05: Fourier mode k
# is multiplied by s_k = exp(-k^2 T). The adult print gives the
# smoothed profile plus measurement noise of size 0.01 (fixed seed). We
# undo the smoothing naively, dividing by s_k, and with Tikhonov
# regularisation, multiplying by s_k / (s_k^2 + alpha). Profiles and
# noise are illustrative; the transform is an explicit DFT.
import cmath
import math
import random

rng = random.Random(20261112)
N, T, NOISE = 32, 0.05, 0.01


def dft(x):
    return [math.fsum((x[j] * cmath.exp(-2j * math.pi * k * j / N)).real
                      for j in range(N))
            + 1j * math.fsum((x[j] * cmath.exp(-2j * math.pi * k * j / N))
                             .imag for j in range(N)) for k in range(N)]


def idft(c):
    return [math.fsum((c[k] * cmath.exp(2j * math.pi * k * j / N)).real
                      for k in range(N)) / N for j in range(N)]


def wave(k):
    return min(k, N - k)


def rel_err(a, b):
    num = math.fsum((x - y) ** 2 for x, y in zip(a, b))
    return math.sqrt(num / math.fsum(y * y for y in b))


if __name__ == "__main__":
    xs = [2 * math.pi * j / N for j in range(N)]
    g = [0.6 * math.sin(x) + 0.3 * math.cos(3 * x) + 0.15 * math.sin(6 * x)
         for x in xs]
    s = [math.exp(-wave(k) ** 2 * T) for k in range(N)]
    f_clean = idft([c * sk for c, sk in zip(dft(g), s)])
    f = [v + rng.gauss(0, NOISE) for v in f_clean]
    F = dft(f)
    print("smoothing factors: mode 1 {:.4f}, mode 6 {:.4f}, mode 16 {:.2e}"
          .format(s[1], s[6], s[16]))
    print("condition number of the forward map {:.3e}".format(
        max(s) / min(s)))
    print("data misfit of the noisy print {:.4f}".format(rel_err(f, f_clean)))
    naive = idft([c / sk for c, sk in zip(F, s)])
    print("naive inverse: relative error {:.3e}".format(rel_err(naive, g)))
    print("   alpha      relative error of recovered early profile")
    best = None
    for alpha in (1e-6, 1e-5, 1e-4, 1e-3, 1e-2, 1e-1):
        rec = idft([c * sk / (sk * sk + alpha) for c, sk in zip(F, s)])
        e = rel_err(rec, g)
        best = (e, alpha) if best is None or e < best[0] else best
        print("   {:7.0e}    {:.4f}".format(alpha, e))
    print("best alpha {:.0e}: error {:.4f}; the choice of alpha is an"
          .format(best[1], best[0]))
    print("assumption about smoothness, not something the print supplies")

Single source and citation

Fan, Y. (2026). Turing Reaction-Diffusion Patterns in Palm Dermatoglyphics, Flexion-Crease Topology and Embryonic Stress Inversion (Version 1.0.0), Chapter 6: Inverting for Embryonic Stress: the Direction Field Gives Only Half the Information. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-158 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026158,
  author    = {Fan, Yafu},
  title     = {Turing Reaction-Diffusion Patterns in Palm Dermatoglyphics, Flexion-Crease Topology and Embryonic Stress Inversion},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book