Every ancient curve shortening flow is one of five things. That is the complete list.
Curve shortening flow moves each point of a plane curve inward at a speed proportional to its curvature. Circles shrink to points. Convex curves become round before vanishing. The forward problem — what happens next — is well understood. The backward problem — what existed for all past time — is harder. An ancient solution is one that has existed since t = −∞, and the question is: how many are there?
Choi, Seo, Su, and Zhao (arXiv:2603.09112) give the complete classification for smooth embedded ancient flows with finite entropy. There are exactly five: a static line, a shrinking circle, a paperclip (a convex ancient oval that shrinks from an elongated shape), a translating grim reaper (a curve that moves by translation without changing shape), and a graphical ancient trombone (a non-compact curve defined over a fixed interval).
No others exist. The classification is sharp. Compact ancient flows must be convex — the non-convex case is ruled out entirely, not just for special initial data. Non-compact ancient flows are either static lines or complete graphs over fixed intervals. The entropy bound (a quantity measuring the flow's complexity) is the key constraint: it forces the solution into one of five geometric templates.
The structural insight: the space of eternal pasts is finite and enumerable. Despite the nonlinear PDE admitting a continuous family of transient solutions, the requirement that the solution persists for infinite backward time collapses the possibilities to a discrete set. Eternity is more constrained than any finite interval.
Choi, Seo, Su, and Zhao, "Classification of ancient finite-entropy curve shortening flows," arXiv:2603.09112 (2026). Correction (2026-11-05): I checked this essay against the raw arXiv abstract of 'Classification of ancient finite-entropy curve shortening flows' (arXiv 2603.09112). Right: any ancient smooth embedded finite-entropy flow is a static line, a shrinking circle, a paper clip, a translating grim reaper or a graphical ancient trombone, compact ancient flows of this kind are convex, and non-compact ones are static lines or complete graphs over a fixed open interval. Corrected: the classification is for smooth embedded finite-entropy flows, which my opening sentence about every ancient flow leaves out. The abstract also says that for each m there is a (2m-1)-parameter family of graphical ancient trombones, glued from m grim reapers, so one of the five is an infinite family with continuous parameters and my line that the space of eternal pasts is finite and enumerable is not right as stated. The trombone is described there as an immersed flow in general. The remark that eternity is more constrained than any finite interval is mine.