Quantum Speed Limit (QSL) Overview

• Quantum speed limit (QSL) is a fundamental bound on the minimum time required for a quantum system to evolve between distinct states, determined by energy uncertainty, mean energy, and geometric distances.
• Canonical formulations like the Mandelstam–Tamm and Margolus–Levitin bounds, along with geometric and action-based methods, quantify state evolution speed across unitary and nonunitary dynamics.
• Extending to open systems, operator flows, and classical analogues, QSLs serve as diagnostic tools in experiments, revealing insights into decoherence, control, and dynamical phase transitions.

Quantum speed limit (QSL) denotes a lower bound on the time required for a quantum system to evolve from a given initial state to a distinguishably different final state. In its original form, it is the minimum orthogonalization time for a pure state under unitary dynamics; in contemporary usage it encompasses mixed states, time-dependent generators, open-system evolution, operator flows, phase-space dynamics, and even classical Liouville evolution. Across these settings, QSLs quantify a kinematic constraint of the form “distance divided by speed,” where the distance is set by a state- or operator-space geometry and the speed is set by energetic, generator-norm, or representation-dependent resources (Hadipour et al., 2020, Campo, 2020, Okuyama et al., 2017).

1. Canonical bounds and standard formulations

For a closed quantum system evolving under a time-independent Hamiltonian $\hat H$, the canonical bounds are the Mandelstam–Tamm (MT) and Margolus–Levitin (ML) inequalities. In the notation used in the literature surveyed here,

$$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$

where $$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$ is the energy uncertainty and $E=\langle \hat H\rangle$ is the mean energy above the ground state. A standard unified closed-system expression is

$$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$

which captures the tightest of the MT and ML constraints (Hadipour et al., 2020).

Geometric formulations recast these inequalities in terms of a distance on state space. For pure states, the fidelity $F=|\langle\psi_0|\psi_\tau\rangle|^2$ defines the Bures or Fubini–Study angle

$\vartheta=\arccos(\sqrt{F}),$

and the MT bound becomes

$\tau \ge \frac{\hbar\,\vartheta}{\Delta E}.$

In this perspective, the actual evolution traces a path of length $\gamma(\tau)=\int_0^\tau dt\,\Delta H(t)/\hbar$, while the QSL states that the geodesic distance cannot exceed the accumulated path length (Campo, 2020).

For open systems, the dynamics is nonunitary and the generator is no longer simply a Hamiltonian. One widely used form is the Deffner–Lutz-type bound

$$\tau_{\mathrm{QSL}} =\frac{1}{\Lambda_\tau^{\mathrm{op}}}\, \sin^2\!\bigl(\mathcal{L}(\rho(0),\rho(\tau))\bigr),$$

with $$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$0 the Bures angle from the initial pure state to the evolved mixed state and

$$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$1

where $$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$2 is the time-local generator (Teittinen et al., 2021). Another standard open-system route uses relative purity, for example

$$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$3

together with ML-type or MT-type norm bounds on $$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$4; in the specific moving-qubit study discussed below, the ML-type relative-purity bound is the tighter one (Hadipour et al., 2020).

2. Geometric structure, brachistochrone ideas, and generalized bounds

A central contemporary theme is that QSLs are geometric statements about trajectories in a manifold of states. In pure-state unitary dynamics this geometry is essentially unique, but for mixed states and open systems there is an infinite family of contractive Riemannian metrics. This enlarges the set of admissible distances and speeds, and shifts the interpretation of a QSL from a single universal time-energy relation to a metric-dependent optimality statement (O'Connor et al., 2020).

One important refinement is the distinction between path length and how that path is traversed. “Action quantum speed limits” define, for a path $$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$5 and metric $$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$6,

$$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$7

where $$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$8 is the geodesic distance between the endpoints and

$$\tau \ge \tau_{\mathrm{MT}}=\frac{\pi\hbar}{2\,\Delta E},\qquad \tau \ge \tau_{\mathrm{ML}}=\frac{\pi\hbar}{2\,E},$$9

is an action functional. Unlike the usual geometric bound

$$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$0

the action QSL depends explicitly on the instantaneous speed profile along the path. For a given path, it is saturated when the path is geodesic and the speed is constant (O'Connor et al., 2020).

A related development formulates QSLs through a quantum brachistochrone problem in a Riemannian metric. In that setting the minimal time is constrained not only by the global speed

$$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$1

but also by the dynamics of individual coordinates $$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$2. The resulting bound has the form

$$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$3

so a single “critical parameter” can determine the operative speed limit (Dou et al., 2022).

The structure of QSLs becomes richer still when several compatible observables constrain the dynamics simultaneously. Chau showed that by replacing single-observable bounds with inequalities built from multiple energy moments,

$$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$4

the allowed set of evolution times can become disconnected. In that case QSLs exhibit forbidden speed intervals and, in a finite-dimensional example, a first-order phase transition in the minimum evolution time as the target fidelity is varied (Chau, 2013).

For qubits and finite-dimensional systems, a distinct geometric construction uses the Bloch angle

$$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$5

A 2025 formulation defines a QSL $$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$6 by the condition

$$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$7

with the instantaneous Bloch-angle speed

$$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$8

This bound saturates when the Bloch-vector trajectory is geodesic and the speed is constant (Wang et al., 31 May 2025).

3. Open systems, decoherence, non-Markovianity, and control

In open systems, QSLs are often discussed alongside decoherence, memory effects, and control of system–environment coupling. A recurrent observation is that non-Markovianity can shorten the QSL time, but that this is not universal. In phase-covariant qubit models, speed-up can occur under P-divisible and even CP-divisible dynamics, so reduction of $$\Delta E=\sqrt{\langle \hat H^2\rangle-\langle \hat H\rangle^2}$$9 is not necessarily tied to a transition from P-divisible to non-P-divisible dynamics. In those examples, oscillations of populations and fidelity are more directly correlated with QSL reduction than divisibility alone (Teittinen et al., 2021).

A concrete model of velocity-controlled decoherence is the moving qubit inside a leaky cavity. There, the qubit couples to a Lorentzian reservoir through a time-dependent spatial mode function $E=\langle \hat H\rangle$0. Using a relative-purity open-system bound, the study finds that in both weak-coupling (Markovian) and strong-coupling (non-Markovian) regimes, increasing the qubit velocity increases the QSL time. In the strong-coupling regime the QSL as a function of the initial time shows oscillations, but higher velocity damps those oscillations and drives the speed of evolution toward a nearly constant value; the system thereby becomes more stable against state change (Hadipour et al., 2020).

This stabilization viewpoint is developed explicitly in a Markovian setting by treating QSL as a measure of robustness. For a pure initial state $E=\langle \hat H\rangle$1, one introduces the relative-purity angle

$E=\langle \hat H\rangle$2

and derives the explicit lower bound

$E=\langle \hat H\rangle$3

with

$E=\langle \hat H\rangle$4

Larger $E=\langle \hat H\rangle$5 means slower decoherence-induced departure from $E=\langle \hat H\rangle$6, and maximizing $E=\langle \hat H\rangle$7 reduces to a convex quadratic optimization problem over $E=\langle \hat H\rangle$8 (Kobayashi et al., 2020).

The total-system viewpoint yields a different route to open-system QSLs. Using quantum-state diffusion, one writes the total system–environment state in terms of stochastic pure-state trajectories and defines the Bures angle and Fubini–Study metric on the total state. For the spin–boson model, this exposes a spectral mechanism absent in reduced Born–Markov treatments: the “infinite speedup capacity” of the noiseless case is destroyed under the Born–Markovian approximation, but is recovered in non-Markovian dynamics whenever a bound state forms in the energy spectrum of the total system (Wu et al., 2022).

Continuous quantum measurements complicate the picture further. In the ensemble description, standard open-system QSLs apply to the averaged density matrix. At the trajectory level, however, the conditioned state follows a nonlinear stochastic master equation, and the trajectory-dependent speed can exceed the ensemble QSL bound. In the measured-qubit example studied, a substantial fraction of trajectories violate the standard ensemble-based QSL, and the variance of trajectory speeds grows with the measurement strength. The same analysis shows that continuous monitoring induces Brownian dynamics in Hilbert space, with mean squared Bures-angle increment proportional to the observable variance and the measurement strength (García-Pintos et al., 2018).

4. Beyond quantum-state trajectories: classical limits, operator flows, and phase spaces

QSLs are not confined to pure or mixed quantum states viewed in Hilbert space. A major conceptual extension is the observation that the underlying mechanism is geometric evolution under a Hermitian generator. In the Koopman–von Neumann formulation of classical mechanics, the Liouville equation defines a Hermitian flow on a Hilbert space of phase-space densities. This yields a classical speed limit (CSL), including a Mandelstam–Tamm-type form

$E=\langle \hat H\rangle$9

and analogous bounds for Fokker–Planck and detailed-balance master equations. In this sense, QSL is not uniquely quantum but a broader Hilbert-space dynamical property (Okuyama et al., 2017).

Another extension concerns operators rather than states. For unitary operator flows $$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$0, vectorization in Liouville space leads to MT-type and ML-type bounds on the operator overlap $$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$1. The MT-type operator bound is

$$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$2

while the ML-type form is

$$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$3

with $$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$4 the Liouvillian and $$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$5. These operator-flow QSLs translate directly into bounds on autocorrelation functions, dynamical susceptibilities, and the quantum Fisher information entering metrology (Carabba et al., 2022).

A further generalization replaces Hilbert-space geometry by arbitrary phase-space representations built from Stratonovich–Weyl correspondence. In that framework, for a phase-space symbol $$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$6 and relative purity $$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$7, the instantaneous speed obeys

$$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$8

with

$$\tau \ge \tau_{\mathrm{QSL}}^{(\mathrm{closed})} =\max\!\left\{\frac{\pi\hbar}{2\,\Delta E},\frac{\pi\hbar}{2\,E}\right\},$$9

$F=|\langle\psi_0|\psi_\tau\rangle|^2$0

This yields a universal phase-space QSL for both continuous-variable and finite-dimensional systems. In several examples, suitable $F=|\langle\psi_0|\psi_\tau\rangle|^2$1-parametrized phase spaces produce tighter speed bounds than either Wigner phase space or Hilbert space under the same relative-purity metric (Meng et al., 2022).

5. QSL as diagnostic, experimental observable, and performance indicator

QSLs have become practical probes of dynamics rather than solely abstract bounds. In ultracold gases confined in time-dependent harmonic traps, self-similar dynamics collapses the many-body evolution onto a single scaling factor $F=|\langle\psi_0|\psi_\tau\rangle|^2$2. The fidelity is then expressible as

$F=|\langle\psi_0|\psi_\tau\rangle|^2$3

while the energy variance is controlled by a nonadiabatic factor $F=|\langle\psi_0|\psi_\tau\rangle|^2$4: $F=|\langle\psi_0|\psi_\tau\rangle|^2$5 Because $F=|\langle\psi_0|\psi_\tau\rangle|^2$6 is directly inferred from cloud-size measurements, both the Bures angle and the path length can be reconstructed without full tomography, enabling an experimental probe of QSLs in many-body systems (Campo, 2020).

In many-body localization problems, QSLs can function as phase-transition diagnostics. For one-dimensional Aubry–André and Wannier–Stark models, the MT, ML, and dual ML$F=|\langle\psi_0|\psi_\tau\rangle|^2$7 bounds were compared after sudden quenches across localization transitions. The MT bound is always tighter in the short-time limit for arbitrary states, and for extreme quenches it remains the tighter orthogonalization bound. More strikingly, crossing points of QSL curves obtained from initial states deep in opposite phases identify the localization–delocalization transition point exactly in the Aubry–André case and reproduce the expected finite-size scaling in the Wannier–Stark case (Suman et al., 2023).

A photonic implementation has recently demonstrated direct measurement of a geometric QSL built from the Bloch angle. For qubits the Bloch angle can be written as

$F=|\langle\psi_0|\psi_\tau\rangle|^2$8

and measured via a swap test through the overlap relation

$F=|\langle\psi_0|\psi_\tau\rangle|^2$9

where $\vartheta=\arccos(\sqrt{F}),$0 is the antisymmetric-subspace probability. This avoids comprehensive state tomography and, in the Landau–Zener example implemented, the new Bloch-angle QSL was tighter than an earlier Bloch-angle bound for accelerating dynamics (Wang et al., 31 May 2025).

Relativistic settings provide another operational use of QSL. For a Dirac electron in a magnetic field, the relevant orthogonalization time is

$\vartheta=\arccos(\sqrt{F}),$1

and an effective evolution speed is defined by radial displacement over $\vartheta=\arccos(\sqrt{F}),$2. In uniform magnetic fields the saturated QSL reaches $\vartheta=\arccos(\sqrt{F}),$3, whereas spatially varying fields $\vartheta=\arccos(\sqrt{F}),$4 restructure the relativistic Landau spectrum and can raise the saturated value to $\vartheta=\arccos(\sqrt{F}),$5 for suitable positive $\vartheta=\arccos(\sqrt{F}),$6, still below $\vartheta=\arccos(\sqrt{F}),$7. The same framework is connected to the Bremermann–Bekenstein bound on information processing rate (Aggarwal et al., 2024).

6. Interpretive issues, misconceptions, and scope

Several recurring misconceptions have been corrected by recent work. First, QSL is not uniquely quantum in origin: classical Liouville dynamics, imaginary-time Schrödinger equations, and detailed-balance master equations all admit mathematically analogous speed limits once evolution is formulated as Hilbert-space motion under a Hermitian generator (Okuyama et al., 2017).

Second, QSL reduction is not a universal diagnostic of non-Markovianity. Non-Markovian information backflow can indeed shorten $\vartheta=\arccos(\sqrt{F}),$8, but speed-up can also occur under P-divisible and CP-divisible dynamics, while some non-Markovian evolutions do not exhibit speed-up. The operative mechanism in several qubit models is more directly associated with oscillations of populations and fidelity than with divisibility classes alone (Teittinen et al., 2021).

Third, open-system geometric QSLs should not automatically be interpreted as physically achievable minimum times. In the action-QSL analysis, geometric bounds are shown to indicate optimality with respect to the geodesic path in the chosen metric, whereas the actually reachable minimum time may be constrained by the path generated by the dynamics and by the available control over the instantaneous speed (O'Connor et al., 2020).

Fourth, standard open-system QSLs apply naturally to ensemble-averaged dynamics, not necessarily to continuously monitored single trajectories. Under continuous measurement, conditioned trajectories can exceed the ensemble QSL bound, so “violation” in that setting reflects a mismatch between ensemble and trajectory descriptions rather than a breakdown of quantum mechanics (García-Pintos et al., 2018).

Finally, most concrete QSL formulas are conditional on specific modeling assumptions. Examples in the literature assume, depending on context, pure initial states, time-independent Markovian Lindblad generators, rotating-wave and dipole approximations, single-excitation sectors, Lorentzian spectral densities, or particular phase-space kernels. Open-system bounds are often sufficient but not necessary, and their quantitative tightness depends strongly on the chosen distance measure, the control manifold, and the physical regime (Hadipour et al., 2020, Kobayashi et al., 2020).

Taken together, these developments establish QSL as a broad framework for dynamical constraints rather than a single inequality. It now spans unitary and nonunitary evolution, state and operator dynamics, quantum and classical descriptions, and both analytic and experimentally accessible formulations. Within that enlarged framework, the central question is no longer merely how fast a state can evolve, but which geometry, representation, and control constraints determine the relevant notion of speed for the physical process under study.