When Entanglement Gets Complicated: Quasiparticle Pairs Break Down, and the XXZ Chain Reveals What Happens Next

Tests the standard quasiparticle picture against highly entangled initial states. Free models need multiplet corrections; interacting models seem to ignore them entirely.

Imagine you're at an airport watching passengers deplane. If every traveler exits as part of a couple — two people, joined at the hip — you can predict the crowd flow perfectly by tracking pairs. That's the quasiparticle picture of entanglement spreading: a quantum quench creates entangled pairs of excitations that fly apart at known velocities, and by tracking those pairs you get the entanglement entropy exactly. It works beautifully for integrable systems after simple quenches. But what happens when the passengers aren't traveling in couples — when they're traveling in entangled groups of 4, 6, 8, or more? This paper constructs exactly that scenario. Sabadini and Alba study quenches in the XXZ spin chain starting from 'n-rainbow states' — macroscopically homogeneous states built from increasingly long-range Bell pairs. For n=1, you get the standard nearest-neighbor dimer state (the integrable case). For n=2 and above, the initial state is no longer integrable, and the quench pumps excitations into correlated multiplets of 2n quasiparticle species rather than simple pairs. The local stationary state approaches infinite temperature as n grows, with 1/n corrections the authors derive analytically. The key split in the paper is between free and interacting models. In the free XX chain (no interactions), the multiplet structure matters enormously. Although interspecies correlations among the 2n quasiparticle flavors vanish as 1/n, they conspire to suppress the entanglement entropy growth rate, which goes to zero as n→∞. The authors capture this quantitatively within a generalized quasiparticle picture that accounts for multiplet correlations — a genuine extension of the standard framework. In the interacting XXZ chain, everything changes. Time-dependent DMRG (tDMRG) simulations show that entanglement entropy simply grows at the rate of thermodynamic entropy exchange across the subsystem boundary. There is no visible trace of the multiplet suppression seen in the free case, at least within the time windows accessible to tDMRG. The interacting chain appears to 'forget' the multiplet structure entirely, producing entanglement growth consistent with the simplest possible thermodynamic prediction. The paper also treats a genuinely asymmetric protocol — joining an n-rainbow state to an infinite-temperature half-chain — and conjectures a formula for the von Neumann entropy in that setting. This is a harder problem than the homogeneous quench because left-right symmetry is broken, and the conjecture provides a testable target for future numerics or analytics. What makes this interesting beyond the specific XXZ model is the methodological question it sharpens: when does the quasiparticle picture need to be extended versus when do interactions wash out the need for extensions? The free-model result says multiplets are real and consequential. The interacting-model result says interactions may thermalize away the fine structure before it matters for entanglement. The tension between these two findings is the load-bearing contribution. The limitations are honest: tDMRG is constrained to moderate times and system sizes by entanglement growth itself, so the interacting-case conclusion carries a caveat — the multiplet signature could appear at longer times. The paper acknowledges this directly. The analytical tools (generalized hydrodynamics, quasiparticle picture extensions) are well-established in the integrable-systems community, and the numerical methods are standard. The novelty is in the question asked and the clarity of the answer obtained, not in methodological innovation.