Houart (2026d) showed that within the disc-dominated population of disc galaxies, a Gaussian Mixture Model (GMM) on the Fall-relation residual ∆j and the Sérsic index n identifies two dynamical regimes, X1 (fossil) and X2 (live), refining an earlier Sérsic threshold classification (Houart 2026c) under which X1 was shown to undergo morphological fossilization: the Sérsic index loses essentially all predictive power in X1. Here we ask whether this fossilization is reversible, and identify its physical mechanism. Combining N = 447 MaNGA disc galaxies (calibration domain: Σ-A+B, gas-poor regime) with N = 437 TNG100 SubLink merger trees (ten snapshots, z = 5.2 → 0) and N = 180 EAGLE dwarf trajectories (four snapshots, z = 2.0 → 0), we report four results. (i) Reversibility: a population of N = 38 galaxies (“Vieux Vivants”, VV) is simultaneously dynamically live (X2), stellar-population old (Dn4000 ≥ 1.40), and structurally rebuilt (Rd,resid ≥ 0) — a signature inconsistent with passive ageing but consistent with gas accretion-driven disc regrowth after fossilization, independently corroborated by elevated stellar spin (λRe, p = 1.8×10−45), negative outer age gradients (∇AgeMW = −0.052 vs +0.018 in ordinary X2), enhanced incidence in isolated environments (odds ratio ≈ 2.9, p = 0.0001), and intermediate gas depletion time. Direct evidence of the reverse transition in TNG100 gives an asymmetric hysteresis loop (thresholds ∆Rd = +0.19 / −0.30dex, locally calibrated; cycling fraction 27.8%). (ii) Exact mechanism: ∆j ≡ Rd,resid+Vflat,resid holds to machine precision in every sub-population; the variance decomposition shows a monotonic loss of dominance of the Rd,resid–Vflat,resid anti-correlation, from −42% of Var(∆j) in X1 to −28% in X2 to −4% (statistically null) in VV — disc rebuilding proceeds via loss of the normal size–velocity compensation, predominantly through Rd,resid (Cohen d = 1.07 between VV and X1) rather than Vflat,resid (ns). (iii) Falsifiability: within an extended MaNGA sample built without H I quality cuts (N = 1186; Sect. 4), N = 41 galaxies satisfy the falsifier criterion (X2, old, structurally compact); of these, only 3 survive strict H I quality cuts (ratio 12.7–17.0 : 1 across three independent X1/X2 definitions), and all three are individually attributable to threshold noise or instrumental ceiling effects, not genuine counter-examples. (iv) Mass dependence: the frequency of major structural growth episodes (∆Rd,resid > 0.15dex) is not measured directly for VV, which is defined only in single-epoch MaNGA data; in structural growth trajectories it decreases with mass across the EAGLE dwarf regime (logMbar = 7.5–9.8, from ∼ 13% to ∼ 4%) and is roughly constant and higher across the massive TNG100 regime (logMbar = 11.3–13.4, ∼ 14–15%), with VV incidence itself rising with mass across the unobserved gap between the two simulations (8% → 24% in MaNGA, logMbar = 9.8–11.3) — consistent with two distinct regimes (stochastic dwarf-scale fluctuations vs. genuine disc rebuilding) rather than a single suppressed-to-active transition. We conclude that X1/X2 is a reversible dynamical state, that its transition is size-driven, and that the fossil-to-live transition leaves a detectable, falsifiable stellar-population signature.
Clément Houart (Sat,) studied this question.