01
Two entropies, one formula
Boltzmann’s entropy counts the microscopic arrangements behind a state of matter; Shannon’s counts the possibilities behind a message. Both take the form −Σ p log p, which is why von Neumann reportedly urged Shannon to borrow the name.
02
Waves inside every photo
JPEG applies Fourier’s idea to 8 × 8 patches of pixels: each block becomes a sum of cosine waves, and the finest ripples, which the eye barely notices, are thrown away.
03
Random kicks make a bell curve
A pollen grain’s displacement is the sum of countless random molecular kicks, so by the central limit theorem it follows the normal distribution, spreading in proportion to the square root of time.
04
The same S-curve
Verhulst’s logistic equation describes bacteria filling a dish as resources run short. Run in discrete yearly steps instead of continuously, the same equation is the gateway to chaos.
05
Google’s eigenvector
PageRank is the leading eigenvector of the web’s link matrix: the one distribution of visitors that the random surfer’s clicking leaves unchanged.
06
From apples to planets
Newton showed that an inverse-square force produces exactly Kepler’s rule T² ∝ a³, and the constant in Kepler’s law turned out to be a measure of the Sun’s mass.
07
How the Sun pays for its light
Four protons fuse into a helium nucleus that is 0.7% lighter than they were. E = mc² turns that missing mass into the energy that has kept the Sun shining for 4.6 billion years.
08
The Sun runs on quantum luck
The Sun’s core is far too cool for protons to overcome their electric repulsion by brute force. Quantum tunnelling lets a tiny fraction slip through the barrier, and that is just enough.
09
Spacetime as a lens
In general relativity mass curves spacetime, so it bends the paths of light. Eddington measured the effect for the Sun in 1919; whole galaxies and clusters act as natural telescopes.
10
Hearing planets in starlight
The same shift that drops a passing siren’s pitch moves a star’s spectral lines back and forth as an unseen planet tugs on it.
11
The universe’s afterglow is a black body
The cosmic microwave background matches Planck’s black-body curve at 2.725 kelvin more precisely than any source ever made in a laboratory.
12
Earth glows in the infrared
Earth radiates like a cool black body, peaking in the infrared at about 10 micrometres, the wavelengths that water vapour and carbon dioxide absorb.
13
Clocks in the rocks
The steady, random decay of unstable nuclei gives geologists clocks that date rocks, meteorites and Earth itself back 4.5 billion years.
14
The same Arrhenius
Svante Arrhenius explained how temperature speeds up chemical reactions in 1889, and in 1896 made the first calculation of how much carbon dioxide warms the planet.
15
Enzymes are catalysts
Each of the Krebs cycle’s eight steps is speeded up by its own enzyme, a protein catalyst that lowers the energy barrier without being used up.
16
Curves that saturate
An enzyme’s rate and haemoglobin’s oxygen load both level off as binding sites fill. Haemoglobin’s four cooperating sites turn the simple curve into an S, ideal for delivering oxygen.
17
A structure that suggests copying
The complementary base pairs that hold the double helix together are exactly what let each strand serve as a template for a new partner.
18
Weak bonds, strong shapes
Hydrogen bonds hold together the helices and sheets of proteins, and water’s own hydrogen-bond network pushes oily parts of the chain inward.
19
Diffusion across a membrane
Osmosis is diffusion of water through a barrier that holds back dissolved molecules; Fick’s law gives the rate.
20
Why storms spin
The Coriolis effect deflects air flowing toward a low-pressure centre, turning the inflow into a spiral: anticlockwise in the northern hemisphere, clockwise in the southern.
21
Rotation steers the ocean
The same deflection turns wind-driven water to the right of the wind in the northern hemisphere, layer by layer, until the net flow runs at right angles.
22
Rhythmic pushes
A swing pushed in time with its motion builds up. Moons whose meetings recur at the same places tug each other into lockstep for the same reason.
23
Seismic refraction
Earthquake waves bend as their speed changes with depth, obeying Snell’s law like light in glass, and the bending revealed the core.
24
A bead that behaves like light
Johann Bernoulli solved the brachistochrone by imagining the bead as a ray of light crossing layers where its speed increases, applying Snell’s law at every layer.
25
Two roads to chaos
A one-line population model and three equations for convection both show deterministic chaos: rules with no randomness whose futures still cannot be predicted in practice.
26
How magnets order
The Ising model explains why iron magnetises below a critical temperature. Earth’s core is far hotter than that, so its field must instead be generated by flowing molten iron.
27
Alan Turing’s two great ideas
The same mathematician defined what it means to compute in 1936, and in 1952 showed how chemistry alone can paint stripes and spots.
28
Patterns from reactions
Reaction plus diffusion produces travelling spirals in the BZ dish and stationary spots and stripes in Turing’s model.
29
Complexity from simple rules
Neither a starling nor a cell in Conway’s grid sees the big picture. Rich collective patterns emerge from each one following a few rules about its neighbours.
30
Evolution as an algorithm
Genetic algorithms copy natural selection: variation, inheritance and survival of the fittest, applied to candidate solutions instead of organisms.
31
Artificial neurons
The perceptron borrowed the idea of a nerve cell that fires once its combined inputs cross a threshold.
32
Following the slope
Gradient descent uses derivatives to find which way is downhill. Training a neural network is calculus applied to billions of variables at once.
33
Error correction in cells and computers
DNA polymerase proofreads its work and repair enzymes fix mismatches, much as Hamming’s check bits locate and correct a flipped bit.
34
Binomials draw curves
The weights that blend a Bézier curve’s control points are the numbers of Pascal’s triangle, multiplied by powers of t and 1 − t.
35
Security from primes
RSA is safe because multiplying two large primes is easy while recovering them from the product is, as far as anyone knows, impractically hard.
36
From bridges to route planners
Euler’s drawing of land and bridges started graph theory; Dijkstra’s algorithm searches the same kind of graph for the shortest route.
37
Tilt and the ice ages
The tilt that gives us seasons slowly rocks between 22.1° and 24.5°; together with changes in Earth’s orbit, that rhythm paces the ice ages.
38
The planet breathes
The yearly sawtooth in the Keeling curve is the northern hemisphere’s forests taking in carbon dioxide each summer and releasing it each winter.
39
Particles on magnetic rails
The Lorentz force makes charged particles spiral along Earth’s field lines toward the poles, where they light the aurora.
40
Weighing ions with magnets
A mass spectrometer uses the Lorentz force: in the same magnetic field, heavier ions curve less than lighter ones.
41
Electric force at storm scale
Charges separated inside a thundercloud build an electric field until the air itself breaks down and conducts.
42
Squeezable air
Because air compresses under pressure, the atmosphere is densest at the bottom and thins out exponentially with height.
43
Flow in pipes and arteries
Both describe fluid in tubes. Bernoulli ignores friction; Poiseuille is all about it, which is why narrowing vessels matter so much.
44
Fractal plumbing
One explanation for Kleiber’s three-quarter power law is that blood vessels branch like a fractal, reaching every cell with the least plumbing.
45
Exponential outbreaks
Early in an epidemic each case causes R₀ new ones, so cases grow exponentially, with e setting the pace of continuous growth.
46
Chemical and ecological clocks
Feedback loops with delays make both lynx and hare populations and the BZ reaction’s chemicals cycle again and again.
47
Colour from both ends
A complex’s colour is set by which light its electrons absorb; we perceive what is left through three kinds of cone cell.
48
A football made of carbon
C₆₀ has 60 vertices, 90 edges and 32 faces: 60 − 90 + 32 = 2, just as Euler’s polyhedron formula requires.
49
Order without repetition
Ordinary crystals repeat a unit cell. Penrose tilings and quasicrystals are perfectly ordered, yet never repeat.
50
The nucleus orders the table
Rutherford’s nucleus led Moseley to show that an element’s place in the periodic table is set by its nuclear charge, the atomic number.
51
Two perpendicular vibrations
Combine vibrations along x and y with a phase difference and you get ellipses: Lissajous figures at equal frequencies, and the paths traced by polarized light.
52
Electrons as waves
The wave nature that makes single electrons build up fringes behind two slits also shapes them inside atoms: an orbital is a standing electron wave, and its shape is set by how many nodes fit around the nucleus.
53
States as waves and arrows
A qubit is a quantum superposition like a particle’s wave function; the Bloch sphere is a picture of the simplest, two-state case.
54
Codes in cells and computers
Computers encode information in bits; cells encode proteins in three-letter codons drawn from four bases.
55
Worlds made of cells
Wolfram’s rules and Conway’s Life both update a grid of cells from their neighbours; Rule 110 and Life can each carry out any computation.
56
Tracing rays
Descartes traced rays through a raindrop by hand in 1637; ray tracers do the same through whole scenes, millions of times a second.
57
Dynamos in the Sun and the Earth
Flowing conductive matter generates the magnetic fields of both. The Sun’s flips every 11 years; Earth’s at irregular intervals of hundreds of thousands of years.
58
Drifting continents, branching species
As Gondwana broke apart, separated populations evolved independently, which is why marsupials dominate Australia.
59
Randomness that computes
Buffon’s needles estimate π by chance; modern film renderers estimate lighting the same way, by averaging many random light paths.
60
Bayes’ rule, applied to every email
A spam filter is Bayes’ theorem run once per word: the prior odds of spam are updated by the likelihood of each word in turn, with the naive assumption that the words are independent.
61
Data as a sum of bell curves
A Gaussian mixture explains a lumpy distribution as several overlapping normal distributions, each with its own centre and spread, weighted by how common it is.
62
The directions that matter most
The principal components of a dataset are the eigenvectors of its covariance matrix, and their eigenvalues measure how much of the variation each direction carries.
63
Trees built from distances
Average-linkage clustering, known in biology as UPGMA, was one of the first ways to build evolutionary trees: repeatedly join the two most similar species or groups.
64
Questions that reduce uncertainty
A decision tree chooses each question by information gain: the drop in Shannon entropy of the labels after the split. The best question is the one that tells you the most.
65
An S-curve for probability
Verhulst’s logistic curve describes a population levelling off as it fills its habitat. Logistic regression uses the same curve to turn a score into a probability between 0 and 1.
66
Random walks run backwards
The forward process of a diffusion model is Brownian motion: each point takes small random steps until the data is pure noise. Generating an image means learning to reverse that walk.
67
A law found by fitting a line
Plot the logarithm of each planet’s period against the logarithm of its distance from the Sun and the points fall on a straight line of slope 3/2. Least squares, invented to fit such orbits, finds it.
68
Bellman’s equation, twice
Shortest-path algorithms and Q-learning both rest on Bellman’s principle: the best total from here is the best immediate step plus the best total from where it leads.
69
Filters over small patches
JPEG describes each 8 × 8 patch with a fixed set of cosine patterns; a convolutional network slides small learned filters over the image instead, choosing patterns that help it recognise what it sees.
70
The best building blocks
Fourier analysis rebuilds any signal from the same fixed sine waves. The singular value decomposition finds the building blocks best suited to one particular dataset, sorted by importance.
71
The cocktail party problem
Ears and brain pick one voice out of a noisy room. Independent component analysis does a mathematical version, unmixing recordings into the separate sources that produced them.
72
Passing values along links
PageRank repeatedly passes importance from each page to the pages it links to. A graph neural network repeatedly passes learned messages from each node to its neighbours.
73
Similarity as distance
In 2019, word vectors trained only on materials-science abstracts placed the chemical elements in groups that echo the periodic table, recovering chemistry from the way scientists write.
74
Learning as a game
Hawk–dove and a GAN are both two-player games that settle at an equilibrium where neither side can gain by changing its strategy alone.
75
Do more of what worked
A bacterium tumbles less often while its food supply is improving; a policy-gradient agent makes actions more likely after they improved its reward. Both climb by trial and reward rather than by a map.
76
Polynomials that bend to fit
A Taylor series builds a function from powers of x; polynomial regression does the same from noisy data. Both are reliable near the data and can swing wildly far from it.
77
Machines with memory
A finite-state machine carries one remembered state from step to step; a recurrent network carries a learned vector of numbers, which lets it remember far more than a fixed set of states.
78
Squeezing out redundancy
Huffman coding compresses a message exactly by giving common symbols short codes. An autoencoder compresses approximately, learning a short code from which the input can be rebuilt.
79
Fourier, made fast
The fast Fourier transform computes the frequencies of Fourier analysis for sampled signals in N log N steps, which is what makes digital audio, images and radio practical.
80
Counting mutations
Comparing DNA sequences with an edit-distance table counts the substitutions, insertions and deletions between them, a measure of how long ago two species shared an ancestor.
81
Collisions come sooner than you think
A Bloom filter’s false alarms are hash collisions, the same effect as the birthday paradox: when many items share a limited number of slots, overlaps become likely surprisingly early.
82
A record in layers
Rock strata and a blockchain both record order by stacking: each new layer rests on the one before, so altering an old layer means disturbing everything above it.
83
Detail at every scale
The Koch snowflake adds the same bump at smaller and smaller scales. Perlin noise adds smooth randomness in octaves of halving size, producing the same fractal roughness found in coastlines and mountains.
84
Nearest-neighbour maps
A one-nearest-neighbour classifier gives every point the label of the closest training example, so its decision regions are exactly the Voronoi cells of the training points, coloured by class.
85
Why six sides
When three cracks meet at every corner, Euler’s formula V − E + F = 2 forces the average cell to have six sides, which is why cooling lava, soap foam and honeycombs all settle near hexagons.
86
Cells from seeds
Each step of the cooling front lets basalt columns shift toward the centres of their cracks, the same relaxation that turns a random Voronoi diagram into a near-perfect honeycomb.
87
Spin is a qubit
A silver atom’s spin is a two-state quantum system, the same as a qubit: its direction is a point on the Bloch sphere, and measuring along an axis turned by θ gives up with probability cos²(θ/2).
88
Patterns from diffusion
Both turn smooth chemical gradients into regular stripes: in Liesegang rings a threshold and depletion space the bands, in Turing patterns a fast inhibitor and slow activator do.
89
Supercooled clouds
Rising air cools below freezing long before its droplets turn to ice. Clouds hold supercooled water down to about −40 °C, and the ice crystals that do nucleate grow at the droplets’ expense, starting most rain outside the tropics.
90
Cold light
A firefly’s flash is chemiluminescence: the enzyme luciferase oxidises luciferin into an excited molecule that emits a photon, wasting almost nothing as heat.
91
Chaos in the sky
Poincaré met sensitive dependence on starting conditions in the three-body problem seventy years before Lorenz met it in weather: in both, tiny errors grow exponentially and long-range prediction fails.
92
Borrowed momentum
A flyby is an elastic collision at a distance. Momentum and energy are conserved overall, so the speed the spacecraft gains is paid for by an imperceptible slowing of the planet.
93
Apparently faster than light
Blobs in quasar jets seem to cross the sky at several times the speed of light. Nothing breaks relativity: the blobs move toward us at almost c, nearly catching up with their own light.
94
Order from neighbours
Coupled fireflies and coupled magnetic spins both switch from disorder to order abruptly once the coupling beats the randomness: synchrony in one, magnetisation in the other.
95
Networks without a planner
The slime mould’s pruned network sits between a minimum spanning tree, the cheapest way to connect every food source, and a fully meshed network, trading cost against resilience.
96
Ant colony optimisation
Marco Dorigo turned pheromone trails into an algorithm for the travelling salesman problem: virtual ants build tours, and short tours get more pheromone, steering later ants toward good routes.
97
Bubbles from falling pressure
Rising magma behaves like an opened fizzy drink: as pressure falls, dissolved gas comes out and expands, as Boyle’s law describes, and whether it escapes or shatters the magma decides the eruption.
98
Boiling under pressure
The boiling curve on water’s phase diagram rises with pressure, so water deep in a geyser’s tube stays liquid far above 100 °C until a drop in pressure tips it across the curve.
99
Will it ever stop?
Deciding whether the Collatz process reaches 1 from every starting number is a halting question, and no general method exists to settle such questions for every program.
100
One tenth in binary
In binary, 0.1 is the infinite repeating fraction 0.000110011…, a geometric series that a computer must cut off after 52 bits, which is why 0.1 is stored slightly wrong.
101
Rumours as epidemics
Gossip protocols borrow the mathematics of epidemics: informed nodes are infected, each infects a random contact per round, and the update spreads along the same S-shaped curve.
102
Memories as magnets
A Hopfield network is an Ising model whose couplings are learned: neurons are spins, stored patterns are low-energy states, and recall is the system settling into a valley.
103
Searching a tree of choices
Backtracking walks a game or puzzle tree exhaustively, pruning dead ends; Monte Carlo tree search samples the same kind of tree at random, spending its effort on the most promising branches.
104
Selfish traffic
Data networks face the same tension as road networks: each sender acting selfishly can make everyone slower, which is why TCP backs off politely when it senses congestion.
105
The red Moon
During a total lunar eclipse the only sunlight reaching the Moon has passed through Earth’s atmosphere edge-on, where Rayleigh scattering has stripped out the blue, turning the Moon coppery red.
106
Modes as eigenvectors
Each Chladni figure is a vibration mode, an eigenvector of the plate’s equations of motion, and its pitch is set by the matching eigenvalue.
107
Across many orders of magnitude
Benford’s law holds best for quantities spread over many powers of ten, like earthquake energies, which follow the Gutenberg–Richter power law.
108
The spinning Earth underfoot
The rotation that turns a Foucault pendulum’s swing also deflects moving air, bending the flow of the Hadley cells into the trade winds and the westerlies.