Concept

Saddle point

A saddle point is a critical point that is a maximum along one direction and a minimum along another, like a mountain pass. On a surface it is where two pieces of a region above a height join or a loop closes, and it counts minus one towards the Euler characteristic.

Named by 4 essays across 3 fields — each of them below, with the objects they name alongside it.

One minimum, or several. Two curves side by side with their local minima marked: a convex one with a single minimum, and a fourth-power well with 2.

Where the guarantee stops

Convexity converts every downhill method into a correct one, and its absence removes the guarantee entirely rather than degrading it. What is left is a collection of partial answers, and knowing which of them apply to a given problem is most of what non-convex optimisation is.

analysis · Convexity
A random surface and the region above height 1. A smooth Gaussian random field on a 64 by 64 torus and its excursion set above 1: 14 pieces, 0 holes, Euler characteristic 14; expected 13.8.

Pieces minus holes on a random surface

Cut a random landscape at some height and the region above it breaks into pieces with holes in them. How many pieces, and how many holes, has no formula. Their difference does: the average Euler characteristic of the region above height u is a constant times u multiplied by the bell curve at u, exactly, for every smooth Gaussian surface — because the Euler characteristic is a sum of peaks, pits and saddles, and the average number of those can be computed point by point.

topology · Euler characteristic
The pieces of a random vibration. Random plane wave, 240×240 grid, wavelength 24: 23 positive and 18 negative domains.

The pieces no formula counts

A random vibration of a large membrane splits the surface into regions where it bulges up and regions where it dips down. Pieces minus holes has an exact average; the number of pieces has none. Nazarov and Sodin proved it is proportional to the area, Bogomolny and Schmit predicted the constant from percolation, 0.0624, and counting the pieces of 51 random waves gives 0.0599 — below the prediction by several standard errors. The sizes of the pieces follow percolation's law; their number does not.

topology · Euler characteristic
Three random games and the cells where best replies meet. Three random 5 by 5 games with best replies marked; pure equilibria (cells best for both players): 0, 1 and 2.

A third of random games have no pure equilibrium

Fill a game's payoff table with random numbers and ask whether some cell is each player's best reply to the other. Exactly one cell is expected to be, at every size, and the chance that none is climbs from one in eight to 1/e — the same constant that counts the shuffles in which nobody gets their own hat back, and for the same reason.

applied · Equilibrium

Named alongside it

The objects these essays reach for when they reach for this one.

Critical pointEuler characteristicRandom fieldComplexityConvexityCounterexampleDerangementGradientInclusion exclusionLocal minimumMorse theoryNash equilibrium

All concepts