Emergence
Properties of the whole not derivable from the parts without irreducible computational effort.
'Emergence' is a word often used to say nothing at all. In the physics of complex systems it has at least three precise formulations, and they are worth keeping apart, because they answer different questions.
The first is Anderson's, 1972: each level of the scientific hierarchy has laws of its own, irreducible to the level below, and the mechanism is spontaneous symmetry breaking. Microscopic laws are more symmetric than the states they produce: the Hamiltonian of a crystal is translation invariant, the crystal is not. There is a technical point that always gets dropped, though: a sharp phase transition exists only in the thermodynamic limit. For a finite system everything stays analytic, and emergence is an excellent approximation rather than an exact fact.
The second is computational, and comes from Crutchfield: a property is emergent if the minimal machine reproducing the macroscopic dynamics sits in a higher computational class than the microscopic one. That gives an operational criterion with a number to compare. Not 'the whole is mysteriously more than the parts', but: deriving the properties of the whole from the parts takes a computation with no shortcuts.
The third is Batterman's, and it is the most general. Emergence lives wherever the limit between two theories is singular, that is, wherever behaviour at the limit differs qualitatively from behaviour near it. It holds for 1/N → 0 in phase transitions, for ħ → 0 in the semiclassical regime, for λ → 0 between wave and ray optics. A rainbow is emergent in this sense, with no part-whole hierarchy holding it up: a limit that fails to commute is enough.
Sources
- Anderson, More Is Different (1972)
- Batterman, The Devil in the Details (2002)
- Crutchfield, Inferring Statistical Complexity