Coarse Scrambling: Hierarchies & Mechanisms

• Coarse scrambling is defined as the delocalization of information using reduced, coarse-grained observables rather than full microscopic resolution.
• It is characterized through diverse metrics such as Rényi entropies, tripartite mutual information, and operator-space distributions that reveal a hierarchy of scrambling effects.
• Operational approaches like partial trace over unobserved modes and experimental benchmarks provide practical tools to measure and distinguish coarse scrambling in many-body systems.

Searching arXiv for recent and foundational papers on “coarse scrambling” and related scrambling notions. Coarse scrambling is a family of notions in which “scrambling” is defined not at the finest microscopic level, but after a projection onto coarse observables, coarse sectors, coarse spectral data, or operationally accessible subsystems. Across the literature, the phrase covers several distinct but related constructions. In one line of work, the central mechanism is coarse-graining over unobserved environmental modes, so that fully unitary dynamics appears scrambling for accessible observables once the hidden sector is traced out (Su et al., 2023). In another, coarse scrambling denotes a weaker, macroscopic spreading criterion—such as delocalization of a constant-size message without the stronger requirement that all linear-size subsystems become maximally mixed (Brown et al., 2012). Other formulations use coarse operator-space distributions (Omanakuttan et al., 2022), coarse collective observables in long-range semiclassical spin systems (Marino et al., 2018), coarse spectral flow parameters (Kalsi et al., 2023), or coarse thermodynamic state counting after black-hole absorption and equilibration (Azuma, 25 May 2026). The shared theme is that scrambling is diagnosed through reduced, grouped, or operationally accessible data rather than through a complete microscopic account.

1. Coarse scrambling as a hierarchy of weaker notions

The literature does not use a single universal definition. Instead, “coarse scrambling” labels several weaker or reduced notions of information delocalization. In random quantum circuits, the weaker notion is explicit: “only a constant number of initial low weight Pauli strings are brought to linear weight strings with high probability,” whereas stronger results require that for every initial state and all subsystems $S$ with $|S|\le fn$, the reduced state $\rho_S$ is close to maximally mixed (Brown et al., 2012). In this hierarchy, coarse scrambling means that a constant-size message becomes so delocalized that a random subset $T$ of size $n-c$ is nearly decoupled from the reference, but without the stronger quantum-error-correcting guarantees associated with full linear-size decoupling (Brown et al., 2012).

A different hierarchy appears in the study of designs and Rényi entropies. There, ordinary scrambling is weaker than full Haar randomness, and the intermediate levels are organized by Rényi entanglement entropies and design order. The paper shows that ensembles of the same design order have almost maximal Rényi entanglement entropies of the corresponding order, and interprets these as a hierarchy “between information scrambling and Haar randomness” (Liu et al., 2017). This suggests that coarse scrambling can mean low-order entropic or moment-based pseudorandomness without full Haar-like complexity.

A third hierarchy is operational rather than formal. In “Benchmarking Information Scrambling,” scrambling is quantified through an overlap $F(t)$, its scrambling-induced asymptote, and extracted rates $\lambda_s$ and $\lambda_d$, so that scrambling becomes a graded property rather than a binary one (Harris et al., 2021). This suggests a coarse characterization in which one measures degree and rate of scrambling from a small number of effective observables.

These uses are compatible rather than identical. A plausible synthesis is that coarse scrambling denotes delocalization visible at a reduced level of description—constant-size messages, low-order moments, grouped operator sectors, coarse observables, or accessible reduced states—without requiring full microscopic randomization.

2. Environment-induced coarse scrambling in quantum field theory

A particularly sharp realization appears in “Scrambling Power of Soft Photons,” which studies QED scattering in a bipartite channel picture with hard and soft sectors (Su et al., 2023). The input and output are split into hard particles $(A,C)$ and soft photons $(B,D)$, with the scattering channel treated as a unitary $|S|\le fn$0 identified with the $|S|\le fn$1-matrix by $|S|\le fn$2 (Su et al., 2023). The exact coarse-graining operation is a partial trace over undetected soft modes, for example

$|S|\le fn$3

together with

$|S|\le fn$4

This is the formal implementation of coarse-graining: the soft photons are treated as inaccessible environmental degrees of freedom, and the observed hard sector becomes mixed and decohered (Su et al., 2023).

Scrambling is quantified through the tripartite mutual information

$|S|\le fn$5

and, more tractably in the QFT setting, through its 2-Rényi version

$|S|\le fn$6

The paper emphasizes the bounds

$|S|\le fn$7

so negative $|S|\le fn$8 implies scrambling also in the von Neumann sense (Su et al., 2023).

In the free limit, the result is

$|S|\le fn$9

whereas with leading interaction terms

$\rho_S$0

For the external-field example, after box and time regularization,

$\rho_S$1

The negativity is finite but “much greater than the lower bound” $\rho_S$2, so the effect is weak or non-maximal (Su et al., 2023).

The physical mechanism is environment-induced decoherence. Infrared dressing attaches momentum-dependent soft coherent clouds to hard charges, and different hard histories radiate distinguishable soft clouds. When the soft sector is traced out, coherent-state overlaps suppress off-diagonal hard-sector terms. The paper states that “soft photons remove information from the incoming hard state and transfer to the environment, where it is observationally inaccessible,” while the full $\rho_S$3-matrix remains unitary (Su et al., 2023). This is a canonical example of coarse scrambling: information is not fundamentally lost, but it becomes inaccessible after coarse-graining over unobserved radiation.

3. Weak, macroscopic, and operator-space formulations

In random quantum circuits, coarse scrambling is a weaker decoupling notion tied to the Hayden–Preskill setting rather than to uniform local thermalization. For the parallel complete-graph model, Theorem 4.1 states that for any fixed $\rho_S$4 and constant message size $\rho_S$5, there exists sufficiently large constant $\rho_S$6 such that for a random subset $\rho_S$7 of size $\rho_S$8,

$\rho_S$9

with probability

$T$0

after depth

$T$1

On a $T$2-dimensional lattice, the analogous depth is

$T$3

with probability $T$4 (Brown et al., 2012). This is weaker than the paper’s strong scrambling results, which require that every subsystem of size at most $T$5 be nearly maximally mixed and arise only after depth $T$6 after parallelization (Brown et al., 2012).

A related but distinct coarse formulation appears in “Scrambling and quantum chaos indicators from long-time properties of operator distributions.” There the Heisenberg-evolved operator is expanded in a complete orthonormal basis,

$T$7

and the squared amplitudes are grouped into a coarse-grained distribution over complexity sectors,

$T$8

For spin-$T$9 systems, the sectors $n-c$0 are shells of fixed Pauli weight; for collective-spin systems they are shells of fixed tensor rank (Omanakuttan et al., 2022). The paper then studies the mean

$n-c$1

the variance

$n-c$2

and the inverse participation ratio

$n-c$3

Its central claim is that the long-time properties of this coarse-grained operator distribution provide proxies for the onset of quantum chaos, and that fully chaotic behavior is associated with long-time agreement with the Haar-random benchmark

$n-c$4

together with suppressed temporal fluctuations (Omanakuttan et al., 2022).

A further restricted notion is developed in “Quantum circuits with classically simulable operator scrambling.” There the scrambling is confined to the operator subspace $n-c$5, represented as an operator-space Hilbert space with $n-c$6, $n-c$7. Under the gate set $n-c$8, initially simple $n-c$9-Pauli strings evolve into superpositions of exponentially many strings with operator entanglement linear in system size, yet the dynamics remains efficiently simulable because the induced evolution is Clifford in operator space (Blake et al., 2020). This is coarse scrambling in a restricted operator sector rather than on the full operator algebra.

4. Collective, spectral, and experimentally benchmarked coarse scrambling

A collective version appears in “A cavity-QED simulator of slow and fast scrambling,” where the system is a set of large spins with all-to-all-like inhomogeneous exchange,

$F(t)$0

Scrambling is diagnosed by the collective-spin OTOC

$F(t)$1

which in the semiclassical limit becomes

$F(t)$2

The paper finds exponential growth

$F(t)$3

only for non-separable couplings

$F(t)$4

and initial states near the equator of the Bloch sphere, whereas other cases show algebraic growth such as

$F(t)$5

(Marino et al., 2018). Because the observables are collective on-site spin components in a semiclassical large-spin model, the work is best read as collective or coarse scrambling rather than local few-body scrambling (Marino et al., 2018).

A spectral version is formulated in “Spectral chaos bounds from scaling theory of maximally efficient quantum-dynamical scrambling.” There scrambling is the approach to CUE spectral statistics, parameterized by a single scaling variable $F(t)$6 along the Poisson-kernel flow

$F(t)$7

with

$F(t)$8

The endpoint $F(t)$9 is complete scrambling, finite $\lambda_s$0 corresponds to incomplete but efficient scrambling, and deviations from the scaling predictions at the same $\lambda_s$1 indicate inefficient scrambling (Kalsi et al., 2023). This gives a precise notion of coarse spectral scrambling: the system’s distance from ergodicity is encoded in one coarse parameter and in the scale-dependent spectral form factor hierarchy.

An operationally coarse benchmark is provided by “Benchmarking Information Scrambling.” The protocol uses a forward–perturb–backward loop with overlap

$\lambda_s$2

where

$\lambda_s$3

Under sufficient scrambling, the asymptotic state averages to

$\lambda_s$4

with

$\lambda_s$5

and the overlap saturates at

$\lambda_s$6

In the single-qubit projective example, the asymptotic overlap approaches $\lambda_s$7 under ideal scrambling and $\lambda_s$8 under strong decoherence, and a two-stage fit

$\lambda_s$9

extracts a scrambling rate $\lambda_d$0 distinct from the decoherence rate $\lambda_d$1 (Harris et al., 2021). This is coarse scrambling in the sense of a low-dimensional operational summary.

5. Black holes, holography, and coarse-grained area or onset laws

Several papers use coarse-grained scrambling in black-hole contexts, but in different senses. “Negative entropy in scrambling black holes” is explicitly thermodynamic. It assumes that the positive subsystem $\lambda_d$2 of the black hole “undergo[es] scrambling while imposing conservation of energy $\lambda_d$3,” so that it relaxes to the microcanonical state

$\lambda_d$4

with entropy

$\lambda_d$5

The induced entropy increase after absorption takes the canonical form

$\lambda_d$6

at leading order, with finite-reservoir corrections contained in

$\lambda_d$7

The paper’s point is that the relevant area response is controlled not by the infalling object’s initial von Neumann entropy but by post-absorption microcanonical state counting after scrambling (Azuma, 25 May 2026). This is coarse scrambling as thermodynamic equilibration into microcanonical degrees of freedom.

A different coarse-grained onset notion appears in “Onset of scrambling as a dynamical transition in tunable-range quantum circuits.” There the diagnostic is the tripartite mutual information $\lambda_d$8, computed from second Rényi entropies of large subregions. At fixed short time $\lambda_d$9, tuning the interaction-range exponent $(A,C)$0 reveals a transition between a local-like regime where $(A,C)$1 and a scrambling regime where $(A,C)$2. For the weighted random all-to-all Clifford circuit, the scaling ansatz

$(A,C)$3

collapses the data with

$(A,C)$4

while deterministic sparse circuits motivated by neutral-atom arrays give similar exponents (Kuriyattil et al., 2023). This is coarse scrambling because the onset is detected through coarse entropic observables on large regions, well before full late-time thermalization.

“Notes on Scrambling in Conformal Field Theory” reaches a more cautious conclusion. Optimized wavepacket OTOCs can exhibit a logarithmic timescale

$(A,C)$5

under Virasoro identity block dominance, but the same paper argues that non-identity Virasoro blocks can dominate the relevant OTOCs on the second sheet, undermining any broad claim of universal fast scrambling in $(A,C)$6 CFT (Liu et al., 2018). This suggests that coarse, smeared OTOC probes can look fast-scrambled without establishing fine-grained microscopic scrambling.

6. Limits, distinctions, and recurring caveats

A recurrent theme is that coarse scrambling must be distinguished from both ordinary decoherence and stronger microscopic chaos notions. In QED with soft photons, the negativity of tripartite mutual information comes from interaction-induced entanglement with unobserved soft radiation, not from intrinsic many-body chaotic dynamics (Su et al., 2023). In experimental benchmarking, OTOC decay alone can be faked by decoherence or reversal errors, whereas the recovery-plateau structure in $(A,C)$7 is claimed to single out genuine scrambling (Harris et al., 2021). In collective cavity-QED models, exponential OTOC growth is a semiclassical instability of collective variables and is not by itself a demonstration of fully quantum many-body chaos (Marino et al., 2018).

Another recurring caveat is sector restriction. Super-Clifford circuits scramble a subspace of non-local $(A,C)$8-only operators but do not establish classically simulable scrambling for arbitrary operators on the full Pauli algebra (Blake et al., 2020). Coarse operator distributions group together all operators with the same Pauli weight or tensor rank and therefore forget the microscopic structure within each sector (Omanakuttan et al., 2022). Entropic design hierarchies show that low-order randomness or Rényi entropies do not imply higher-order flatness of the entanglement spectrum; indeed, there exist state 2-designs whose higher-order Rényi entropies are bounded away from the maximum (Liu et al., 2017).

Black-hole and holographic contexts add further qualifications. The microcanonical black-hole construction is explicitly a coarse thermodynamic model and does not derive the microscopic origin of the density of states or a fine-grained scrambling mechanism (Azuma, 25 May 2026). Holographic CFT analyses based on selected OTOCs or smeared probes can be sensitive to intermediate channels or bulk kinematic constraints, so apparent scrambling in coarse correlators need not imply universal microscopic fast scrambling (Liu et al., 2018).

Taken together, these works suggest that coarse scrambling is best understood not as a single formal definition but as a class of reduced descriptions of information delocalization. The reduction may be by tracing out inaccessible modes, restricting to constant-size messages, grouping operator sectors by size or rank, projecting onto collective observables, compressing spectral data to a single scaling parameter, or replacing microscopic states by microcanonical shells. What survives these reductions is the shared signature that information becomes inaccessible to simple local or coarse probes and recoverable only through more global, joint, or fine-grained access.