Weakly Interacting Time Evolution Operators

• Weakly interacting time evolution operators are defined as families that encode dynamics near free limits with a unique closest free-fermion representation and minimizing geodesics.
• They are examined through geometric, perturbative, and stochastic frameworks, employing methods like deformation retraction, mean-field approximations, and operator disentangling.
• These operators have practical implications in analyzing the stability of topological phases, real-time nonequilibrium dynamics, and effective evolution in quantum many-body systems.

Weakly interacting time evolution operators are operator families used to encode dynamics when interactions are either geometrically close to a free-fermion evolution, perturbatively subordinate to an exactly solvable part, or reduced to an effective stochastic or mean-field description. In the most specific sense, introduced by Müssnich–Vieira, a weakly interacting time evolution operator is an interacting fermionic time evolution that lies in the complement of the cut-locus of the free-fermion submanifold inside the full interacting operator space, so that it has a unique closest free representative and a unique minimizing normal geodesic back to that free point (Müssnich et al., 17 Mar 2026). In adjacent literatures, closely related operator constructions appear in Heisenberg-picture excitation-operator methods for nonequilibrium fermions (Wang, 2010), non-coherent Markovian reductions of closed weakly interacting systems (Meilakhs, 2024), second-order Bogoliubov corrections for weakly interacting bosons (Chen, 2010), and pre-relaxation propagators for weak integrability breaking (Bertini et al., 2015).

1. Definition and semantic range

The phrase “weakly interacting time evolution operator” is not used uniformly across the literature. In the geometric free-fermion/topological setting, it denotes a subset

$\mathscr M_{wi,N}\subset \mathscr M_{i,N}$

of interacting time evolutions singled out by Riemannian geometry relative to the free locus $\mathscr M_N\subset \mathscr M_{i,N}$. In this usage, weakness is not expressed by a small coupling constant alone; it is encoded by the existence of a unique closest free operator and a unique minimizing normal geodesic with no conjugate point on $(0,1]$ (Müssnich et al., 17 Mar 2026).

In other settings, “weakly interacting” refers instead to perturbative proximity to a solvable model. For spin chains one writes

$H=H_0+gV,\qquad g\ll 1,$

and studies the regime $t\to\infty$, $g\to 0$, with $T\equiv gt=O(1)$, where the full propagator can be replaced by an effective mean-field evolution (Bertini et al., 2015). For weakly interacting bosons, the Hamiltonian carries an $N$-dependent scaling, here a $1/N^2$ prefactor for three-body interactions, so that the total three-body energy per particle remains $O(1)$ as $\mathscr M_N\subset \mathscr M_{i,N}$0 (Chen, 2010). For discrete-time weakly interacting particle systems, the time-evolution operator is a nonlinear map on probability measures,

$\mathscr M_N\subset \mathscr M_{i,N}$1

rather than a unitary on Hilbert space (Budhiraja et al., 2014).

A common misconception is that these usages are interchangeable. They are not. The geometric notion in the tenfold-way context is a statement about deformation retraction and stable homotopy type; the stochastic and mean-field notions concern reduced dynamics of probabilities or fluctuations; and perturbative operator methods address explicit real-time propagators or Heisenberg evolutions. The common thread is that interaction effects are controlled without replacing the problem by a completely arbitrary interacting dynamics.

2. Geometric weakly interacting loci

In the finite-dimensional Nambu-space model with symmetric bilinear form $\mathscr M_N\subset \mathscr M_{i,N}$2, Müssnich–Vieira distinguish the free and interacting Hamiltonian spaces by

$\mathscr M_N\subset \mathscr M_{i,N}$3

and define the corresponding time-evolution spaces

$\mathscr M_N\subset \mathscr M_{i,N}$4

Any interacting point $\mathscr M_N\subset \mathscr M_{i,N}$5 may be written nonuniquely as

$\mathscr M_N\subset \mathscr M_{i,N}$6

The weakly interacting locus $\mathscr M_N\subset \mathscr M_{i,N}$7 consists of those points for which there exists a unique closest free operator $\mathscr M_N\subset \mathscr M_{i,N}$8 and a unique minimizing normal geodesic

$\mathscr M_N\subset \mathscr M_{i,N}$9

with no conjugate point in $(0,1]$0. Equivalently, $(0,1]$1 is the complement of the cut-locus of $(0,1]$2 (Müssnich et al., 17 Mar 2026).

This geometric condition yields an immediate strong deformation retraction

$(0,1]$3

At $(0,1]$4 this is the identity, and at $(0,1]$5 it returns every weakly interacting operator to its unique free representative. The significance is structural rather than perturbative: the interacting operator is organized by a canonical projection to the free submanifold, defined precisely on the complement of the cut-locus.

When the same Cartan–Altland–Zirnbauer symmetry constraints are imposed on interacting Hamiltonians, the construction restricts to symmetry-fixed weakly interacting loci

$(0,1]$6

and then to colimit spaces $(0,1]$7. This makes the weakly interacting condition compatible with the symmetry data underlying the tenfold way (Müssnich et al., 17 Mar 2026).

3. K-theoretic spectra and stability of the tenfold way

For irreducible free fermion systems with Cartan–Altland–Zirnbauer internal symmetries $(0,1]$8, the colimit of free time-evolution spaces produces the Bott-periodic sequence of classical compact symmetric spaces. In the notation of Müssnich–Vieira,

$(0,1]$9

together with the remaining four real spaces in the order of $H=H_0+gV,\qquad g\ll 1,$0-periodicity. These assemble into sequential spectra $H=H_0+gV,\qquad g\ll 1,$1 and $H=H_0+gV,\qquad g\ll 1,$2, with structural suspension maps $H=H_0+gV,\qquad g\ll 1,$3 realizing the twofold complex and eightfold real periodicities (Müssnich et al., 17 Mar 2026).

The weakly interacting analogue is defined by replacing each free space by its weakly interacting counterpart: $H=H_0+gV,\qquad g\ll 1,$4 The suspension maps are taken to be the same $H=H_0+gV,\qquad g\ll 1,$5, but pre-composed with projection to the unique free operator. Concretely, if

$H=H_0+gV,\qquad g\ll 1,$6

then

$H=H_0+gV,\qquad g\ll 1,$7

The cut-locus retraction commutes with the suspension maps and respects the Cartan-fixed and Cartan-inverted loci, so it assembles into a spectral strong deformation retraction

$H=H_0+gV,\qquad g\ll 1,$8

The resulting conclusion is that $H=H_0+gV,\qquad g\ll 1,$9 and $t\to\infty$0 represent the same generalized cohomology theories as the free $t\to\infty$1-theory spectra. In the paper’s formulation, no new stable homotopy classes of ground-state time evolutions appear under weak interactions, and the periodic table of topological insulators and superconductors is stable under weak interactions (Müssnich et al., 17 Mar 2026).

A frequent overstatement is that this proves robustness under arbitrary interactions. The actual statement is narrower: it is a stable homotopy-theoretic result for the weakly interacting locus defined as the complement of the cut-locus.

4. Explicit operator constructions for real-time dynamics

A separate line of work studies weakly interacting dynamics by constructing the time-evolution operator, or its Heisenberg-picture action, directly. In interacting fermion systems driven out of equilibrium, one may solve the Heisenberg equations of motion by introducing excitation operators $t\to\infty$2 satisfying

$t\to\infty$3

If an observable is expanded as $t\to\infty$4, then

$t\to\infty$5

With an ansatz

$t\to\infty$6

the commutator equation becomes a matrix eigenproblem $t\to\infty$7. At fixed truncation order, the number of independent coefficient blocks grows linearly with the number of single-particle states, so one diagonalizes $t\to\infty$8-dimensional matrices rather than an exponentially large object. In the toy Hubbard-inspired model and in the nonequilibrium current problem for the single-impurity Anderson model, the resulting phases $t\to\infty$9 contain interaction-modified excitation energies directly in the exponent, which is the paper’s sense in which the construction goes beyond traditional perturbation theory in the Keldysh-Green’s function formalism (Wang, 2010).

For two interacting oscillators with Hamiltonian in $g\to 0$0, Feynman’s disentangling rules provide another explicit operator construction. After extracting a displacement operator and rewriting the $g\to 0$1 sector in terms of Schwinger-boson generators $g\to 0$2, the remaining ordered exponential reduces to

$g\to 0$3

where $g\to 0$4 solves the complex Riccati equation

$g\to 0$5

In the “SU(2)-type” case with constant phase $g\to 0$6, closed forms for $g\to 0$7 yield a fully disentangled expression for $g\to 0$8. When the mode-mixing $g\to 0$9 is small, one has $T\equiv gt=O(1)$0, $T\equiv gt=O(1)$1, and the first-order weak-coupling approximation is valid so long as $T\equiv gt=O(1)$2, i.e. for times $T\equiv gt=O(1)$3 (Alvarez-Moraga, 2023).

These constructions share a characteristic feature: interaction effects are not appended only as post hoc corrections to observables. They modify the operator-valued evolution itself, through renormalized excitation energies, self-consistent coefficients, or disentangled noncommuting factors.

5. Effective semigroups, mean fields, and reduced evolutions

Weakly interacting dynamics are often recast as effective evolution operators on reduced state spaces. One example is non-coherent evolution in closed macroscopic quantum systems. If amplitudes $T\equiv gt=O(1)$4 are updated by a short unitary kick $T\equiv gt=O(1)$5 together with an unknown diagonal phase matrix $T\equiv gt=O(1)$6, then the probabilities $T\equiv gt=O(1)$7 transform as

$T\equiv gt=O(1)$8

Uniform averaging over the phases removes the off-diagonal terms and produces the bistochastic map

$T\equiv gt=O(1)$9

In the infinitesimal limit this becomes a continuous-time Markov chain

$N$0

For weak interactions, the transition rates are given by Fermi’s golden rule,

$N$1

and detailed balance $N$2 implies convergence to the uniform distribution $N$3. Under factorization assumptions on occupation probabilities, the same framework yields the irreversible Boltzmann collision integral. The paper presents non-coherence as the mechanism transforming time-reversible unitary evolution into time-irreversible stochastic evolution, and as a possible solution to the arrow-of-time problem (Meilakhs, 2024).

In weakly interacting bosonic many-body dynamics, the exact Fock-space evolution can be approximated by a mean-field condensate plus a second-order pair-excitation correction. With

$N$4

the fluctuation vector is approximated by

$N$5

The condensate solves the quintic Hartree equation, while the correction kernel $N$6 satisfies a nonlinear hyperbolic system, equivalently a Riccati-type equation

$N$7

Under the defocusing assumption $N$8 and appropriate initial mass, energy, and variance bounds, the coupled $N$9 system has a unique global solution and the error relative to the exact dynamics is uniformly bounded by $1/N^2$0 for all $1/N^2$1 (Chen, 2010).

For weak integrability breaking in spin chains, the interaction-picture propagator

$1/N^2$2

is replaced, in the pre-relaxation regime $1/N^2$3, by

$1/N^2$4

where $1/N^2$5 is the time-average of $1/N^2$6. In the thermodynamic limit, $1/N^2$7 becomes a polynomial in local charges of $1/N^2$8, and cluster decomposition produces a state-dependent mean-field Hamiltonian $1/N^2$9. Local observables then evolve under

$O(1)$0

Depending on whether the effective charges commute asymptotically, one obtains a second quasi-stationary plateau, trivial stationarity, or persistent oscillations at $O(1)$1 frequencies (Bertini et al., 2015).

An analogous reduction appears in the discrete-time weakly interacting particle system of Budhiraja–Majumder. There the evolution operator is the nonlinear Markov map

$O(1)$2

with

$O(1)$3

Under Lipschitz, integrability, and contraction assumptions, $O(1)$4 is a strict contraction in $O(1)$5, has a unique fixed point $O(1)$6, and $O(1)$7 converges exponentially fast to $O(1)$8. The empirical measure $O(1)$9 converges to $\mathscr M_N\subset \mathscr M_{i,N}$00 uniformly in $\mathscr M_N\subset \mathscr M_{i,N}$01, permitting interchange of the limits $\mathscr M_N\subset \mathscr M_{i,N}$02 and $\mathscr M_N\subset \mathscr M_{i,N}$03 and yielding propagation of chaos at $\mathscr M_N\subset \mathscr M_{i,N}$04 (Budhiraja et al., 2014).

6. Long-time asymptotics, transport, and limits of weak notions

Long-time control of weakly interacting dynamics often requires modifying the evolution operator or weakening the mode of convergence. In Heisenberg-picture simulations of energy transport in an interacting one-dimensional Majorana chain, Kuo et al. interleave exact small-$\mathscr M_N\subset \mathscr M_{i,N}$05 unitary steps with a dissipative superoperator that damps high-weight operator components. For DAOE one uses Pauli weight, while FDAOE uses fermionic weight defined on Majorana monomials $\mathscr M_N\subset \mathscr M_{i,N}$06. One full step is

$\mathscr M_N\subset \mathscr M_{i,N}$07

with $\mathscr M_N\subset \mathscr M_{i,N}$08. Accuracy is controlled by $\mathscr M_N\subset \mathscr M_{i,N}$09, the dissipation rate $\mathscr M_N\subset \mathscr M_{i,N}$10, and the SVD cutoff $\mathscr M_N\subset \mathscr M_{i,N}$11. In the weak-interaction regime, FDAOE preserves free-fermion conserved quantities that DAOE would damp, and both DMT and FDAOE produce a diffusion coefficient consistent with

$\mathscr M_N\subset \mathscr M_{i,N}$12

rather than the naive $\mathscr M_N\subset \mathscr M_{i,N}$13 scaling suggested by a first-pass Fermi’s golden rule estimate. The paper interprets this as support for weak-integrability-breaking theory (Kuo et al., 2023).

A distinct but complementary long-time issue is the topology of convergence. For interacting Fermi systems with Galilei-invariant interaction, the Heisenberg dynamics $\mathscr M_N\subset \mathscr M_{i,N}$14 are weakly asymptotically abelian in time-invariant states $\mathscr M_N\subset \mathscr M_{i,N}$15 if

$\mathscr M_N\subset \mathscr M_{i,N}$16

for all $\mathscr M_N\subset \mathscr M_{i,N}$17. Under clustering and trivial-center assumptions, this weak limit holds for local observables, yielding temporal clustering of multipoint correlations. However, norm asymptotic abelianness fails: for the odd unitary $\mathscr M_N\subset \mathscr M_{i,N}$18,

$\mathscr M_N\subset \mathscr M_{i,N}$19

for all $\mathscr M_N\subset \mathscr M_{i,N}$20. This establishes that weak convergence of commutators in expectation does not imply norm decay of commutators themselves (Narnhofer, 2020).

This distinction is essential for interpreting “weak” results throughout the subject. Weak interactions do not by themselves guarantee strong operator convergence, exact reversibility after coarse-graining, or persistence of every free-fermion invariant. What can be established depends on the framework: deformation retraction in the tenfold-way setting, controlled effective propagators in pre-relaxation and Bogoliubov theory, stochastic semigroups after phase averaging, or weak asymptotic abelianness in operator algebras. Together, these results show that weakly interacting time evolution operators are best understood not as a single formalism, but as a family of rigorously structured approximations and classifications for dynamics near a free, integrable, or otherwise tractable limit.