dynsys op• Data kinds: points → measurement
• Call: import fullseye as fs; fs.ledger.dynsys_correlation_dimension(points, n_radii=24, r_lo=None, r_hi=None, max_points=4000, seed=0) (to call the implementation directly, import mathops; mathops.dynsys_correlation_dimension(points, n_radii=24, r_lo=None, r_hi=None, max_points=4000, seed=0); from the registry, opsmath.get("dynsys_correlation_dimension"))
Grassberger-Procaccia correlation dimension — the slope of `log C(r)`.
`C(r) is the fraction of point pairs closer than r`; for a self-similar
set it grows like `r**D`, and *D* is read off the straight part of the
log-log plot (fitted on the middle 60 % of the radii, where the curve is free
of the small-`r noise floor and the large-r` saturation).
★Why this earns its place: unlike box counting it needs no grid, and its
answers are known for simple sets — a circle gives 1, a filled square
2, a Cantor set `log2/log3 = 0.6309`. It measures a different quantity
from the existing `fractal_dimension` (box counting), so the two are an
independent pair rather than two names for one number.
Returns a `measurement`: the fitted dimension.
Raises `ValueError`: fewer than 32 points; not a 2-D array; non-finite
input; a degenerate cloud (every point identical); a radius range that leaves
no pairs.
Limits: sub-sampled to *max_points* (pairs grow quadratically). ★The
dominant error is not the sub-sampling but the radius window: the
default range is the 1st-25th percentile of pair distances, and on a *bounded*
set its upper end runs into the boundary, where `C(r)` saturates and flattens
the slope. Measured on a unit square (true D = 2): 1.879 with the default
window and 1.873 / 1.879 / 1.871 at 400 / 1,500 / 3,000 points —— more points
do not help; narrowing the window to `r_lo=0.01, r_hi=0.1` gives 1.947
and `0.002 / 0.05` gives 2.050. Pass *r_lo* / *r_hi* explicitly when the
answer matters, and report the window with the number.
Every mathops op validates its input before computing (nothing slips through silently):
• **complex input raises ValueError** — coercing to float64 silently discards the imaginary part (numpy only emits a ComplexWarning and returns a plausible-looking wrong real number). State .real/.imag/abs() explicitly, or use complexops, which handles complex data.
• **masked arrays with masked elements raise ValueError** — the implicit conversion that peels off the mask and uses the raw values underneath is refused. Say explicitly whether to fill or to drop.
• **NaN/Inf raises ValueError on every input** (refused with the count stated — it propagates through the whole result).
• Shapes are strict: 1-D and 2-D are never implicitly promoted or broadcast (a matrix in a vector slot, or a vector in a matrix slot, raises ValueError; reshape explicitly).
• Size cap: ops that take a matrix, and the stat_histogram bins, raise ValueError beyond mathops.MAX_ELEMENTS (2^26 ≈ 67 million elements).
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• poc_what_a_picture_cannot_check — py -3.11 examples/poc_what_a_picture_cannot_check.py
measurement as input)—
dynsys)ode_flow_states · ode_vector_field_grid · dynsys_poincare_section · dynsys_lyapunov_spectrum · dynsys_bifurcation_map
*Provenance: mathops.py — MATH operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.