John Hopfield published two unrelated-looking papers eight years apart: 1974's kinetic proofreading (already this book's Ch 13 citation) and 1982's associative-memory networks. Both rest on the same underlying logic — an irreversible process that monotonically approaches a stable state — and in the 1982 case that logic is not just analogical, it is a proved mathematical fact: the network's energy function is a genuine Lyapunov function, and the network's storage capacity (~0.138N patterns before catastrophic failure) is itself a rigorously derived, widely reproduced result. This chapter, unlike several others in this book, needed almost no correction to the index's original claims — they were already accurate.
Hopfield (1982) defined a network of N binary neurons with symmetric connections Wᵢᵧ = Wᵧᵢ, updated asynchronously, and showed that the quantity E = −½∑WᵢᵧSᵢSᵧ never increases under any single-neuron update that flips a neuron to agree with its net input. This is precisely the definition of a Lyapunov function for the network's dynamics: E is bounded below, and monotonically non-increasing, which forces the network into a stable fixed point (a local energy minimum) from any starting state. Memories are stored by choosing the weights (classically via a Hebbian outer-product rule) so that the desired patterns sit at energy minima; recall is exactly the process of the energy function doing its job.
J.J. Hopfield's 1974 kinetic-proofreading paper (already cited in this book's Ch 13, thymic selection) and his 1982 neural-network paper are both about the same mathematical shape: a system driven toward increasing discrimination or decreasing energy through an irreversible process that cannot run backward. This is a genuinely real link — the same person, working on two very different biological/computational substrates eight years apart, applying what is recognizably the same style of argument each time. This chapter treats that as what it is: an interesting fact about one scientist's recurring mathematical instinct, not evidence that biochemistry and neural memory are "the same system." The K operator in this corpus's convention names the shared shape (irreversible, monotone approach to a stable state) that both papers instantiate, honestly, without claiming more.
Hopfield, J.J. (1982). Neural networks and physical systems with emergent collective computational abilities. PNAS 79, 2554–2558.
Amit, D.J., Gutfreund, H., Sompolinsky, H. (1985). Storing infinite numbers of patterns in a spin-glass model of neural networks. Phys. Rev. Lett. 55, 1530–1533. — Source for the 0.138N capacity result.
McEliece, R.J. et al. (1987). The capacity of the Hopfield associative memory. IEEE Trans. Inf. Theory 33, 461–482. — Independent confirmation.
Hopfield, J.J. (1974). Kinetic proofreading: a new mechanism for reducing errors in biosynthetic processes requiring high specificity. PNAS 71, 4135–4139.
See also: Ch 13 — Thymic Selection (cites the same 1974 paper) · Book VI Index