Speed limits to fluctuation dynamics

Many physical systems exhibit randomness due to stochastic noise or intrinsic quantum uncertainty. Two key descriptors of such systems are the mean of an observable and its fluctuations. Under nonequilibrium conditions, the probability distribution of observables evolves, leading to changes in both the mean and the fluctuation. Recent research has provided speed limits that constrain how fast the mean of an observable can change in time, with roots in the Mandelstam–Tamm bound and modern formulations linked to information-theoretic inequalities such as Cramér–Rao. However, despite their importance for characterizing nonequilibrium processes, rigorous bounds on the speed of fluctuation dynamics (e.g., standard deviation) have remained unclear because the fluctuation is not simply the expectation value of a fixed observable. This gap is significant in scenarios like particle spreading, where the change in dispersion is as relevant as the mean displacement, and in quantum systems where fluctuation dynamics connect to entanglement and coherence growth. The central research question is whether there exist general, tight, and experimentally useful speed limits governing the rate of change of fluctuations of observables, and how these relate to the speed of the mean.

Classical and quantum speed limits have been developed extensively for state evolution and for the mean of observables. Foundational results include the Mandelstam–Tamm bound placing limits via energy uncertainty and generalizations to open dynamics and geometric formulations. For observables, bounds relate the speed of the mean to products of the standard deviation and Fisher information, connecting to the Cramér–Rao inequality and information geometry. These bounds can be tighter and more experimentally relevant than state-distance bounds. Prior work provides limited treatment of fluctuation dynamics; an earlier result addressed specific classical continuity-equation settings but lacked a general tradeoff between speeds of mean and fluctuation and did not cover unitary quantum or information-theoretic contexts. Consequently, prior bounds are often loose in transitions where dispersion changes are significant, and they do not yield a unified principle for fluctuation speed across diverse dynamics.

The work introduces a general framework establishing rigorous bounds on fluctuation dynamics. The core idea is to define a velocity observable V_A for an observable A such that ⟨V_A⟩ = d⟨A⟩/dt. Under broad conditions, the standard deviation σ_A obeys a speed limit dσ_A/dt ≤ c σ_{V_A} with c = 1 or c = 1/2 depending on the dynamics and time dependence of A. An equivalent tradeoff bound holds: (d⟨A⟩/dt)^2 + (dσ_A/dt)^2 ≤ v_A^2 σ_A^2 ≡ C_A. General derivation proceeds by representing expectations as an inner product (A|p), where p is a probability density or density matrix that evolves as dp/dt = L[p]. A dual map C satisfies (A|L[p]) = (C[A]|p). Defining V_A = Ȧ + C[A] yields the key equality for the variance δA^2: d(δA^2)/dt = 2 c (δA, δV_A), where (X, Y) = ⟨{X, Y}⟩ denotes a symmetrized correlation (anti-commutator for quantum, product for classical). Applying Cauchy–Schwarz gives the main inequality for σ_A. Sufficient conditions for c = 1 include dynamics admitting a local conservation/continuity law, or unitary evolution with C[A] = −i[H, A]/ħ. Instantiations and bounds on C_A:

• Continuous classical systems with continuity equation ∂_t p = −∇·(p V): V_A = V·∇A + Ȧ, c = 1. For time-independent A, Hölder inequalities yield C_A ≤ ||∇A||^2 ∫ dx p |V|^2, which can be related to kinetic energy or entropy production depending on the dynamics.
• Fokker–Planck dynamics ∂_t p = −∇·(p v) with mobility μ, temperature T: for time-independent A, C_A ≤ μ T Σ ||∇A||, where Σ = ∫ dx p |v|^2/μ is the entropy production rate.
• Classical and quantum hydrodynamics with continuity for density ρ and flow u: V_A = u·∇A + Ȧ, c = 1, and for time-independent A, C_A ≤ 2 E_kin/M with kinetic energy E_kin = ∫ dx ρ |u|^2.
• Discrete systems with a graph-based continuity structure: define edge measures Q_{ij} and velocities V_{ij} = J_{ij}/Q_{ij}; construct V_A from edge differences (A_i − A_j). One obtains c = 1 and bounds using discrete gradients ||∇_d A||.
• Unitary quantum dynamics ∂_t ρ = −i[H,ρ]: take C[A] = −i[H, A]/ħ and V_A = Ȧ − i[H, A]/ħ, yielding c = 1 and C_A = ⟨[Ĥ, Â]†[Ĥ, Â]⟩/ħ^2 for time-independent A (hats denote centered operators). This remains meaningful in the semiclassical limit.
• Information-theoretic scenario for discrete probabilities with surprisal I_i = −ln p_i: for time-independent A, although (7) may not hold, one can establish c = 1/2 versions: (d⟨A⟩/dt)^2 + (1/4) (d⟨A^2⟩/dt)^2 ≤ C_A with C_A bounded by classical Fisher information F_c via Hölder: C_A ≤ ||δA||^2 F_c.
• General quantum (possibly dissipative) dynamics: c = 1/2 inequalities are obtained with upper bounds using quantum Fisher information F_Q and energy variance ΔE under unitary evolution. Extensions include higher-order absolute central moments μ_n with bounds d⟨|δA|^n⟩/dt ≤ n ⟨|δA|^{n−1} |δV|⟩ (classical), and relations to Wasserstein geometry giving additional bounds on C_A via optimal transport costs.

• Main speed limit and tradeoff: For appropriate velocity observables V_A, the speed of fluctuation is bounded by its fluctuation, dσ_A/dt ≤ c σ_{V_A}, leading to the tradeoff (d⟨A⟩/dt)^2 + (dσ_A/dt)^2 ≤ C_A with c ∈ {1, 1/2} depending on dynamics and observable time dependence. This establishes that mean and fluctuation cannot change rapidly at the same time beyond C_A.
• Broad applicability (Table 1): The framework covers classical/quantum hydrodynamics (c = 1), Fokker–Planck dynamics (c = 1), discrete macroscopic transport in quantum systems (c = 1), evolutionary dynamics without mutation (c = 1/2), general nonlinear dynamics (c = 1/2), unitary quantum dynamics (c = 1 and c = 1/2 forms), and dissipative quantum dynamics (c = 1/2). For many cases, C_A is further bounded by physical or information-theoretic quantities: kinetic energy E_kin, entropy production rate Σ, modified transition strength, variance of growth rates σ_s^2, classical Fisher information F_c, quantum Fisher information F_Q, or energy variance ΔE.
• Fokker–Planck example (transition with non-negligible dispersion): For A(x) = |x| starting from a localized state, previous separate bounds on the mean and on σ are loose, while the combined bound from (2) tightly tracks μ T Σ (Fig. 2b). For time-independent A with ||∇A|| = 1, C_A ≤ μ T Σ, enabling tight constraints or inference of Σ from measured dynamics of ⟨A⟩ and ⟨A^2⟩.
• Hydrodynamics: For time-independent A, C_A ≤ 2 E_kin/M, implying simultaneous suppression of mean and fluctuation speeds when kinetic energy is small, regardless of potentials or interactions.
• Unitary quantum dynamics: For time-independent A, (d⟨A⟩/dt)^2 + (d⟨A^2⟩/dt)^2 ≤ ⟨[Ĥ, Â]†[Ĥ, Â]⟩/ħ^2 (Eq. 12). In spin-1/2 example H = gσ_z, A = σ_x, the inequalities become equalities for a broad class of initial states.
• Work extraction speed limits: For time-dependent unitary dynamics with initial state diagonal in H(0), instantaneous work and energy fluctuation speeds obey (dW/dt)^2 + (d⟨H⟩/dt)^2 ≤ ||dH/dt||^2 bounds; integrated forms yield W(t_fin)^2 + ⟨H(t_fin) − H(0)⟩^2 ≤ ∫_0^{t_fin} ||dH/dt||^2 dt, implying lower bounds on transition time given work and fluctuation change.
• Entanglement growth in non-interacting fermions: Entanglement entropy S_R is bounded via number fluctuation σ_{N_R} and an integrated fluctuation cost S = ∫ (dσ_{N_R}/dt)^2 dt, yielding S_R(t_fin) ≤ B with B depending on |R| and S. Crucially, V_{N_R} = [H, N_R] acts only on the boundary of R, enabling boundary-only evaluation. Numerical simulations show improved short-time bounds over prior state-independent entangling-rate bounds.
• Information-theoretic and population dynamics (c = 1/2): For time-independent A in discrete classical settings, (d⟨A⟩/dt)^2 + (1/4)(d⟨A^2⟩/dt)^2 ≤ C_A with C_A ≤ ||δA||^2 F_c. In evolutionary dynamics without mutation (ṗ_i = δs_i p_i), C_A ≤ ||δA|| σ_s^2; for A = s, one recovers Fisher’s fundamental theorem and a higher-order relation d⟨δs^2⟩/dt = ⟨δs^3⟩.
• General quantum dynamics (c = 1/2): One obtains (d⟨A⟩/dt)^2 + (1/4)(d⟨A^2⟩/dt)^2 ≤ const · ||δA|| F_Q^2; for unitary pure states, this further yields bounds in terms of ΔE^2/ħ^2.
• Higher-order moments and geometry: The method generalizes to higher-order absolute central moments in classical settings and connects C_A to optimal transport via Wasserstein geometry, giving additional bounds and highlighting cases of tightness (e.g., Gaussian processes).

The findings answer the open question of whether one can place rigorous, general limits on how fast fluctuations of observables can evolve in time. By identifying appropriate velocity observables and establishing the key variance identity, the work yields tight and broadly applicable speed limits for fluctuations, alongside a clear tradeoff with the speed of the mean. This tradeoff is not captured by earlier bounds focused solely on the mean and leads to stronger, experimentally relevant constraints in processes where dispersion changes significantly. The results are relevant across diverse nonequilibrium contexts: thermodynamic processes (via entropy production), hydrodynamic flows (via kinetic energy), quantum many-body dynamics (via commutators and boundary operators), and evolutionary dynamics (via growth-rate variance). They also provide practical tools: tighter diagnostics of entropy production from observable dynamics, bounds on work protocols and control costs, and boundary-based estimates of entanglement growth in free fermions. Compared with prior formulations, the present inequalities are both tighter (incorporating the mean–fluctuation tradeoff) and more general (extending beyond classical continuity equations to unitary quantum and information-theoretic regimes).

The work establishes a general law for far-from-equilibrium systems: the speed of fluctuation dynamics is bounded by the fluctuation of a suitable velocity observable. Equivalently, the sum of the squared speeds of the mean and standard deviation is upper-bounded by a quantity C_A, which itself can be bounded by physical (e.g., entropy production, kinetic energy) or information-theoretic (Fisher information) measures. Unlike conventional Cramér–Rao-based speed limits, these results directly constrain the rate of change of fluctuations. The framework extends to multiparameter settings, suggesting applications in metrology and nonequilibrium statistical mechanics. The broad applicability across classical and quantum, deterministic and stochastic, and many-body systems indicates a path toward quantitative control and understanding of fluctuation dynamics, with future work aimed at extending to general open quantum and discrete-time dynamics and leveraging these bounds for control at the level of fluctuations.

• For general (non-unitary) quantum dynamics, a universal choice of velocity observable satisfying the strongest (c = 1) conditions was not found; available results use c = 1/2 and depend on quantum Fisher information.
• For cases with c = 1/2, the main results are established for time-independent observables; extensions to time-dependent observables are not fully developed.
• The discrete-time analogue of the theory remains open.
• While bounds on C_A are provided for many settings, their tightness may depend on system specifics (e.g., state dependence, geometry), and evaluating optimal bounds can require additional information (e.g., entropy production, kinetic energy, or Fisher information).