Escher’s lizards, angels and devils lock together to cover an entire wall. Fix your eyes on one small patch and you realize the whole picture is just “one base motif + a few repeating moves.” There are only four such moves: translation, rotation, reflection, and glide reflection (mirror then slide).
The astonishing fact: no matter how ornate the pattern, the number of essentially different ways to repeat it in the plane is exactly 17 — the “wallpaper groups.” The Moorish artisans of the Alhambra had, by intuition, laid out nearly all 17 centuries before the theorem was proved. Beauty, boxed in by a precise finite number.
A wallpaper group is a discrete subgroup of the plane’s isometries (distance-preserving maps) that contains two independent translation directions. Plane isometries come in just four kinds: translation, rotation, reflection, glide reflection. How they may combine is ruled by the crystallographic restriction theorem: a periodic pattern’s rotational symmetry can only have order 1, 2, 3, 4, or 6 — never 5. Exhaust the classification and the answer is exactly 17.
A purely artistic question — “how many kinds of wallpaper are possible?” — has a clean, finite answer: 17. This is one of those rare moments where an aesthetic space is fully exhausted by mathematics. More beautiful still, Escher broke out of the Euclidean plane late in life: in hyperbolic geometry (his Circle Limit prints), one fish can shrink infinitely to tile a disk, because a hyperbolic plane has “infinite area” — here 5-fold and 7-fold symmetry are legal, and the tilings become infinitely many. The flat world’s constraint loosens in a curved one.
Crystallography: X-ray diffraction reads off a lattice’s symmetry group, which directly determines a material’s properties (in 3D there are 230 space groups). Deep learning: the core assumption of a convolutional network is translation equivariance — a cat shifted over is still a cat, and the CNN welds this symmetry into its architecture, which is why it is parameter-thrifty and generalizes well. Quasicrystals (2011 Nobel Prize) exhibit the forbidden 5-fold symmetry, overturning the century-old dogma that crystals must be periodic.
Railroad tracks are plainly parallel, yet in a painting they meet at a single point on the horizon. Renaissance painters discovered the “vanishing point” without realizing they had invented a geometry. The core insight: treat the eye as a center of projection casting the 3D world onto a canvas — parallel lines meet because the point where they “meet at infinity” is projected onto a real, visible point on the horizon.
Projective geometry is simply “the geometry of the plane after adding points at infinity.” Here “parallel” is no longer special; any two lines intersect, some just at infinity.
The projective plane = the ordinary plane + one “line at infinity” (each direction gives one point at infinity). A projective transformation changes lengths, angles, and parallelism, yet always preserves one quantity called the cross-ratio. For four collinear points:
The four letters are four points on one line; the fraction is a ratio of signed lengths. Perspective stretches everything, but leaves this ratio untouched — it is the “conserved quantity” of projection, like energy in physics.
In Euclidean geometry “point” and “line” are unequal; in projective geometry they are perfectly symmetric — swap “point ↔ line” and “lies on ↔ passes through” in any theorem and you get another theorem. This is the principle of duality. Two points determine a line ↔ two lines determine a point, buy one get one free. And the parallel postulate that vexed Euclid for two millennia simply evaporates: there is no such thing as “parallel.” A trick invented for painting grew into a geometry more symmetric than Euclid’s.
The bedrock of computer vision: camera imaging is essentially a projective transformation, and SLAM, 3D reconstruction, and AR all rest on it. The homography matrix stitches panoramas and pastes virtual images onto real planes. Cross-ratio invariance is used to recover true dimensions from an obliquely shot photo (surveying, forensics). GPUs use homogeneous coordinates (i.e. projective coordinates) to unify translation, rotation, and projection into a single matrix multiply.
“The golden ratio $\varphi\approx1.618$ is the most beautiful proportion; the Parthenon, the Mona Lisa, even credit cards hide it” — this most-repeated claim is largely pseudoscience. Split the truth in two: as an aesthetic law it is almost entirely a retrofit, forced on with a magnifying glass (for that “golden rectangle” you first get to choose where to measure from and to); but as the mathematics of nature, $\varphi$ is real and deep. Strip off the myth and what remains is more beautiful.
$\varphi=\dfrac{1+\sqrt5}{2}$ is the unique positive number satisfying $\varphi^2=\varphi+1$ — the cut where “whole : long part = long part : short part.” It is the limit of ratios of consecutive Fibonacci numbers ($1,1,2,3,5,8,13\ldots$), and it is the continued fraction made entirely of 1s:
All 1s means it converges the slowest and is the hardest of all numbers to approximate by any fraction — hence its nickname: “the most irrational number.”
The real beauty of $\varphi$ is not in art criticism but in the sunflower. Plants grow leaves and seeds at the “golden angle” $\approx137.5°$ tied to $\varphi$. Why that angle? Because “hardest to approximate by a rational” means every new seed always lands in the largest gap left by earlier ones, never lining up into a wasteful row — so they pack most densely. This is not an aesthetic choice but the mathematical optimum evolution was forced into for packing efficiency. The beauty of $\varphi$ is the structure nature reveals when driven to the optimum — not a preference of the human eye.
Botany uses it to model phyllotaxis, predicting the spiral counts of pinecones, pineapples, and sunflowers (always consecutive Fibonacci numbers). Algorithms have “golden-section search,” which exploits $\varphi$’s self-similarity to pin down the extremum of a 1D function with the fewest probes. It is useful in the negative too: the golden-ratio myth is a textbook case of numerology and confirmation bias — the moment you hear “some mystical constant is everywhere,” be on guard.
The curve of every letter on your phone, the rounded corner of every app icon, the “accelerate then decelerate” rhythm of an animation — behind almost all of it is one thing: the Bézier curve. Its magic: you never write a complex equation; you just drag a few “control points” and the curve follows smoothly, as if pulled by magnets — yet it never passes through the two middle control points, only gets “pulled” into an elegant arc by them. The control points are the reins; the curve is the tamed horse.
A cubic Bézier curve is defined by four points, tracing a curve as the parameter $t$ runs from 0 to 1:
The four coefficients are the expansion of the binomial $(1-t+t)^3$ (the Bernstein polynomials), and they sum to exactly 1. So every point on the curve is a weighted average of the four control points, held firmly inside their convex hull, never straying out. $P_0,P_3$ are the endpoints; $P_1,P_2$ only control the direction of departure and arrival.
One curve has two faces: an algebraic formula (Bernstein polynomials) and a pure geometric construction — de Casteljau’s algorithm: take a point at the same proportion on each control edge, connect, take points again, “cut corners” repeatedly, and the point it converges to lies on the curve (the gold and red points in the figure). This recursive corner-cutting, expanded out, equals the binomial coefficients exactly. More beautiful still, it is affine invariant: rotate or scale the control points and the curve follows automatically, no recomputation — because a weighted average inherently respects affine maps. Simple enough to do by hand, powerful enough to hold up the entire digital visual world.
Every vector font: TrueType uses quadratic, PostScript/OpenType outlines use cubic Béziers to trace each glyph. The pen tool in Illustrator and Figma draws exactly this; CSS animation’s cubic-bezier() easing uses it to define pacing. It also powers robot path planning and car body surface design — in fact Bézier curves were invented by Renault engineer Pierre Bézier to design car bodies.