Stroboscopic Dynamics

Stroboscopic dynamics observes a periodically driven system at one fixed phase of every drive cycle:

$$t_n=t_0+nT, \qquad n\in\mathbb Z.$$

At those times, continuous evolution collapses to repeated application of the one-period unitary

$$U_F(t_0) = U(t_0+T,t_0).$$

This reduction is exact at the sampling times. It is also incomplete: it discards motion within each period. Understanding both statements is essential when interpreting Floquet effective Hamiltonians, numerical maps, kicked systems, pulse cycles, and digital quantum simulations.

Floquet Operators develops the one-period spectrum. Here the question is what a sequence of snapshots can determine about the continuous trajectory.

Stroboscopic Map

For a $T$-periodic Hamiltonian,

$$H(t+T)=H(t),$$

propagator covariance gives

$$U(t_0+nT,t_0) = U_F(t_0)^n.$$

The sampled pure states are

$$\lvert\psi_n\rangle = U_F^n \lvert\psi_0\rangle.$$

Density operators obey the discrete unitary channel

$$\rho_n = U_F^n \rho_0 \left( U_F^* \right)^n.$$

It is useful to name the one-step superoperator:

$$\mathcal U_F(\rho) = U_F\rho U_F^*,$$

so that

$$\rho_n = \mathcal U_F^n(\rho_0).$$

For a closed system, $\mathcal U_F$ preserves trace, positivity, entropy, and purity. It does not generically drive states toward one attractor.

The sampling phase $t_0$ is part of the definition. Sampling at a different drive phase gives a unitarily conjugate Floquet operator, but the state and measurement representatives at that phase also change.

Stroboscopic Observables

In the discrete Heisenberg description,

$$A_n = \left( U_F^* \right)^n A U_F^n.$$

Expectation values agree:

$$\operatorname{Tr}(\rho_nA) = \operatorname{Tr}(\rho_0A_n).$$

The same construction defines sampled correlation functions. For example,

$$C_{AB}(n) = \operatorname{Tr} \left[ \rho_0A_nB \right].$$

If an observable is fixed by the one-period map,

$$U_F^*AU_F=A,$$

then its stroboscopic expectation is conserved. This is weaker than continuous conservation. The observable may vary strongly during each period and return only at the sampling phase.

Likewise, a density operator satisfying

$$U_F\rho U_F^*=\rho$$

is stroboscopically stationary. It need not commute with $H(t)$ at every instant.

Spectral Form of Sampled Motion

In finite dimension, choose Floquet eigenvectors

$$U_F \lvert\phi_\alpha\rangle = e^{-i\theta_\alpha} \lvert\phi_\alpha\rangle.$$

Expand

$$\lvert\psi_0\rangle = \sum_\alpha c_\alpha \lvert\phi_\alpha\rangle.$$

Then

$$\lvert\psi_n\rangle = \sum_\alpha c_\alpha e^{-in\theta_\alpha} \lvert\phi_\alpha\rangle.$$

For an observable $A$,

$$\langle A\rangle_n = \sum_{\alpha,\beta} c_\alpha^*c_\beta e^{in(\theta_\alpha-\theta_\beta)} A_{\alpha\beta}.$$

Sampled oscillations are controlled by eigenphase differences modulo $2\pi$. Degenerate eigenphases give stationary coherences; nondegenerate differences give discrete-time interference and recurrences.

An eigenstate of $U_F$ returns to the same ray after each period. A superposition generally does not, because its components accumulate different eigenphases.

Micromotion Versus Stroboscopic Evolution

Let

$$0\leq\tau\lt T.$$

At a fixed intra-period offset, exact composition gives

$$\begin{aligned} U(t_0+nT+\tau,t_0) &= U(t_0+nT+\tau,t_0+nT) U_F^n\\ &= U(t_0+\tau,t_0) U_F^n. \end{aligned}$$

Define the intra-period propagator

$$M(\tau;t_0) = U(t_0+\tau,t_0).$$

Then

$$\lvert\psi(t_0+nT+\tau)\rangle = M(\tau;t_0) U_F^n \lvert\psi_0\rangle.$$

$M(\tau;t_0)$ is the partial propagator within the chosen interval, not the periodic Floquet factor $P(\tau)$: generally $M(T;t_0)=U_F$, while $P(T)=I$. The Floquet operator controls cycle-to-cycle evolution; $M(\tau;t_0)$ reconstructs the motion inside a cycle. Knowing $U_F$ alone does not determine $M(\tau;t_0)$.

This is the operational meaning of micromotion. Measurements locked to $\tau=0$ see only $U_F^n$. Measurements at another fixed phase see a conjugated or dressed observable. Measurements with timing jitter can mix micromotion into what was intended to be a stroboscopic signal.

Averaging over an Unrecorded Sampling Phase

Suppose preparation is locked to $t_0$, but the final measurement offset $\tau$ is drawn from a known probability density $w(\tau)$ on $[0,T)$ and is not recorded. Each experimental run ends with one measurement; no intermediate measurement or random change to the Hamiltonian is assumed. For $\rho_n=U_F^n\rho_0(U_F^*)^n$, the measured mean is exactly

$$\begin{aligned} \overline{\langle A\rangle}_n &=\int_0^T d\tau\,w(\tau) \operatorname{Tr}\!\left[M(\tau;t_0)\rho_nM(\tau;t_0)^*A\right]\\ &=\operatorname{Tr}(\rho_n\overline A_w), \qquad \overline A_w=\int_0^T d\tau\,w(\tau)M(\tau;t_0)^*AM(\tau;t_0). \end{aligned}$$

Equivalently, average the final density operator inside the first line. This is a mixture over sampling times. It need not preserve purity even though each trajectory does. It is not an additional evolution step to apply after every cycle: for repeated preparations the average is performed once at the final measurement. For general $w$, the full partial propagator $M$ is needed, including any evolution within the period generated by $H_F$. No high-frequency approximation enters this formula. Bukov, D’Alessio, and Polkovnikov (2015, §2.4) discuss the broader distinction between phase-locked and phase-averaged measurements.

For a concrete calculation, use the commuting sinusoidal spin from the chapter overview:

$$H(t)=\frac{\hbar a}{2}\cos(\Omega t)\sigma_z, \qquad T=\frac{2\pi}{\Omega}, \qquad q=\frac{a}{\Omega}, \qquad \rho_0=\lvert+x\rangle\langle+x\rvert.$$

Preparation is at $t_0=0$, and $a$ and $\Omega>0$ are angular frequencies. Here $U_F=I$ and the intra-period Bloch vector is $(\cos(q\sin\Omega\tau),\sin(q\sin\Omega\tau),0)$. Uniform sampling, $w=1/T$, therefore gives

$$\begin{aligned} \overline{\langle\sigma_x\rangle} &=\frac{1}{2\pi}\int_0^{2\pi}\cos(q\sin\theta)\,d\theta=J_0(q),\\ \overline\rho &=\frac{I+J_0(q)\sigma_x}{2}, \qquad \operatorname{Tr}\overline\rho^{\,2}=\frac{1+J_0(q)^2}{2}. \end{aligned}$$

The integral defines the Bessel function $J_0$. At $q=\pi$, cycle-boundary measurements give $\langle\sigma_x\rangle=1$, whereas uniform phase sampling gives about $-0.30424$. At the first zero $q\simeq2.40483$, the averaged state is maximally mixed, despite exact pure-state returns at every cycle boundary. The loss of contrast comes from discarding the final timing record; with the same sampling distribution it does not grow with the number of cycles. It is not by itself evidence of dissipative evolution or irreversible heating.

Same Map, Different Micromotion

Consider two protocols on a period $T$.

The first has

$$H_a(t)=0,$$

so

$$U_{F,a}=I.$$

The second has

$$H_b(t) = \begin{cases} A, & 0\leq t\lt T/2, \\ -A, & T/2\leq t\lt T. \end{cases}$$

Because the two segments commute,

$$U_{F,b} = e^{iAT/(2\hbar)} e^{-iAT/(2\hbar)} =I.$$

Both protocols are identical at integer periods. Inside the period they are different: the first state is constant, while the second evolves away from its initial value and then returns.

This example shows why a stroboscopic effective Hamiltonian is not a full reconstruction of the laboratory drive.

Frequency Aliasing

Suppose a continuous signal contains

$$e^{-i\nu t}.$$

At the stroboscopic times,

$$e^{-i\nu t_n} = e^{-i\nu t_0} e^{-in\nu T}.$$

Replacing $\nu$ by

$$\nu'=\nu+m\Omega, \qquad m\in\mathbb Z,$$

does not change the cycle-to-cycle factors because

$$e^{-in(\nu+m\Omega)T} = e^{-in\nu T} e^{-i2\pi mn} = e^{-in\nu T}.$$

The raw signals differ by the constant phase $e^{-im\Omega t_0}$; their sequences normalized to the first sample agree. Without independent knowledge of that initial phase, sampling once per period identifies frequencies modulo $\Omega$. This is temporal aliasing.

Quasienergy modularity is the spectral version of the same fact:

$$\varepsilon \sim \varepsilon+m\hbar\Omega.$$

A stroboscopic record cannot determine which representative generated an eigenphase without additional continuity, bandwidth, or microscopic information.

Sampling at multiple intra-period phases can recover some micromotion harmonics. It does not remove all inverse-problem ambiguities automatically.

Effective Hamiltonian

Choose a logarithm branch and write

$$U_F = e^{-iH_FT/\hbar}.$$

Then

$$U_F^n = e^{-iH_FnT/\hbar},$$

so $H_F$ exactly generates the selected stroboscopic sequence.

The logarithm is multi-valued. If

$$U_F = \sum_\alpha e^{-i\theta_\alpha} \lvert\phi_\alpha\rangle \langle\phi_\alpha\rvert,$$

then

$$H_F = \sum_\alpha \frac{\hbar(\theta_\alpha+2\pi m_\alpha)}{T} \lvert\phi_\alpha\rangle \langle\phi_\alpha\rvert$$

generates the same $U_F$ for any integers $m_\alpha$.

Additional criteria may select a useful branch:

• continuity as the drive is turned on;
• locality or quasilocality;
• a chosen quasienergy zone;
• matching to a high-frequency expansion;
• agreement with known weak-drive limits.

Even after a branch is chosen, $H_F$ determines only the stroboscopic part. A micromotion operator is still required for general times.

Sampling Phase

Let

$$U_F(t_0) = U(t_0+T,t_0).$$

At another phase $t_1$,

$$U_F(t_1) = U(t_1,t_0) U_F(t_0) U(t_1,t_0)^*.$$

Thus the eigenphases are unchanged, but the eigenvectors and effective-Hamiltonian representative are dressed by micromotion.

A laboratory stroboscopic protocol should specify:

1. the drive phase used as $t_0$;
2. the measurement offset $\tau$;
3. whether state preparation is repeated at that phase;
4. timing jitter and finite measurement duration;
5. whether a reported observable is in the laboratory or a rotating frame.

Without this information, two nominally identical stroboscopic datasets may correspond to different sampled observables.

Examples

Slow dynamics under a fast carrier

Near-resonant two-level control often separates a rapid carrier rotation from a slow Rabi envelope. Sampling once per carrier period suppresses the visible carrier phase and reveals the slow cycle-to-cycle motion. A rotating frame explains the same simplification continuously.

The two descriptions are complementary. A rotating frame retains a continuous slow trajectory; stroboscopic sampling retains only selected points of the laboratory trajectory.

Repeated pulse cycle

A pulse sequence with one-cycle unitary

$$U_{\rm cyc} = U_m\cdots U_2U_1$$

has stroboscopic evolution $U_{\rm cyc}^n$. Reordering pulses generally changes the cycle map when the generators do not commute.

Cycle-boundary measurements can diagnose accumulated errors, while within-cycle measurements reveal transient populations and toggling-frame motion.

Delta-kicked rotor

For a rotor receiving an impulsive potential once per period, the cycle map factorizes exactly into a kick and a free-rotation unitary. Kicked Rotor Preview uses this map to compare classical momentum diffusion with quantum interference, dynamical localization, and resonant ballistic growth.

Subharmonic return

If

$$U_F^q\lvert\psi_0\rangle = e^{i\chi}\lvert\psi_0\rangle$$

and $q>1$ is the smallest positive integer with this property, the ray has minimal stroboscopic return time $qT$. Such subharmonic behavior can occur in finite systems by spectral commensurability. Robust many-body subharmonic response requires additional physics and should not be inferred from a short finite-time oscillation alone.

Quantum Simulation Connection

Digital quantum simulation naturally produces a repeated cycle:

$$U_{\rm step} = U_m\cdots U_1.$$

After $n$ cycles,

$$\lvert\psi_n\rangle = U_{\rm step}^n \lvert\psi_0\rangle.$$

This is stroboscopic dynamics whether the cycle represents:

• one Trotter step approximating a target Hamiltonian;
• an engineered Floquet period;
• a compiled gate layer;
• a discrete-time quantum walk;
• a kicked-map model.

The interpretation matters. For Trotterization, the within-step gate sequence is an implementation detail and the cycle map approximates a target continuous propagator. For Floquet engineering, the micromotion may be physical and intentionally designed.

Sampling only at cycle boundaries can conceal coherent within-cycle errors. Conversely, boundary sampling can isolate the effective long-time map from large but reversible micromotion.

The error bounds for product formulas are developed in Trotter Product Formula. The direct one-period series is developed in Floquet–Magnus Expansion, while the broader inverse-frequency and prethermal analysis is explained in High-Frequency Expansions.

Scope and Limitations

The unitary stroboscopic framework does not directly describe:

• periodic open-system channels or Liouvillians;
• measurement-conditioned maps;
• noisy cycle-to-cycle timing or control fluctuations;
• quasiperiodic drives with no common period;
• long-time many-body heating without further analysis;
• continuous observables reconstructed from one sampled phase alone.

For a periodic open system, one can still define a one-period quantum channel. Its spectrum can contain decay rates and attractors, unlike a finite-dimensional unitary Floquet map. Open-system dynamics describes this case.

Common Mistakes

• Treating stroboscopic equality as equality throughout a period.
• Omitting the sampling phase $t_0$.
• Assuming $U_F$ uniquely determines micromotion.
• Assuming a logarithm of $U_F$ is unique.
• Interpreting aliased frequencies or quasienergies as absolute.
• Calling a stroboscopically fixed observable continuously conserved.
• Ignoring timing jitter or finite measurement duration.
• Confusing a physical Floquet cycle with a numerical Trotter step.
• Inferring robust subharmonic order from a short recurrence.
• Applying unitary-map intuition to a dissipative one-period channel.

References

• M. Bukov, L. D’Alessio, and A. Polkovnikov, “Universal high-frequency behavior of periodically driven systems: from dynamical stabilization to Floquet engineering,” Advances in Physics 64, 139–226, 2015, arXiv:1407.4803.
• A. Eckardt, “Colloquium: Atomic quantum gases in periodically driven optical lattices,” Reviews of Modern Physics 89, 011004, 2017.
• F. Haake, Quantum Signatures of Chaos, 3rd ed., Springer, 2010.
• M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, Cambridge University Press, 2010.
• H. Sambe, “Steady states and quasienergies of a quantum-mechanical system in an oscillating field,” Physical Review A 7, 2203–2213, 1973.
• J. H. Shirley, “Solution of the Schrödinger equation with a Hamiltonian periodic in time,” Physical Review 138, B979–B987, 1965.

Exercises

1. Derive the exact reconstruction formula at a fixed intra-period offset.

Solution

Use composition:

$$\begin{aligned} U(t_0+nT+\tau,t_0) &= U(t_0+nT+\tau,t_0+nT) U(t_0+nT,t_0). \end{aligned}$$

Periodicity gives

$$U(t_0+nT+\tau,t_0+nT) = U(t_0+\tau,t_0),$$

and stroboscopic evolution gives

$$U(t_0+nT,t_0) = U_F^n.$$

Therefore

$$U(t_0+nT+\tau,t_0) = M(\tau;t_0)U_F^n.$$

2. Derive the sampled expectation value in the Floquet eigenbasis.

Solution

With

$$\lvert\psi_0\rangle = \sum_\alpha c_\alpha \lvert\phi_\alpha\rangle$$

and

$$U_F\lvert\phi_\alpha\rangle = e^{-i\theta_\alpha} \lvert\phi_\alpha\rangle,$$

the sampled state is

$$\lvert\psi_n\rangle = \sum_\alpha c_\alpha e^{-in\theta_\alpha} \lvert\phi_\alpha\rangle.$$

Thus

$$\begin{aligned} \langle A\rangle_n &= \langle\psi_n\vert A\vert\psi_n\rangle\\ &= \sum_{\alpha,\beta} c_\alpha^*c_\beta e^{in(\theta_\alpha-\theta_\beta)} A_{\alpha\beta}. \end{aligned}$$

3. Prove that sampling once per period aliases frequencies modulo $\Omega$.

Solution

At $t_n=t_0+nT$,

$$e^{-i(\nu+m\Omega)t_n} = e^{-i(\nu+m\Omega)t_0} e^{-in(\nu+m\Omega)T}.$$

The cycle-to-cycle factor is

$$e^{-in(\nu+m\Omega)T} = e^{-in\nu T} e^{-i2\pi mn} = e^{-in\nu T}.$$

The overall prefactor at $t_0$ depends on the sampling phase, but the discrete evolution cannot distinguish $\nu$ from $\nu+m\Omega$.

4. Construct two protocols with the same Floquet operator and different micromotion.

Solution

Take $H_a(t)=0$, so $U_{F,a}=I$ and the state never moves.

For the second protocol, use

$$H_b(t) = \begin{cases} A, & 0\leq t\lt T/2, \\ -A, & T/2\leq t\lt T. \end{cases}$$

Then

$$U_{F,b} = e^{iAT/(2\hbar)} e^{-iAT/(2\hbar)} =I.$$

At $T/2$, however,

$$U_b(T/2,0) = e^{-iAT/(2\hbar)},$$

which is generally not the identity. The two protocols agree stroboscopically and differ inside the period.

5. Explain the difference between a Floquet cycle and a Trotter step.

Solution

A Floquet cycle is the physical period of a time-periodic Hamiltonian. Its intra-period micromotion can be observable and may be part of the intended dynamics.

A Trotter step is usually a numerical or digital approximation to evolution under a target Hamiltonian. The within-step gate ordering creates approximation error relative to that target. Repeating the step produces a stroboscopic approximation:

$$U_{\rm target}(n\Delta t) \approx U_{\rm step}^n.$$

The mathematics of repeated unitaries is shared, but the physical interpretation and error question differ.

6. In the uniformly sampled sinusoidal-spin example, compute the variance of the recorded $\sigma_x$ outcomes. Why is squaring $\overline{\sigma_x}=J_0(q)\sigma_x$ the wrong way to obtain the second moment? Then replace uniform sampling over the whole cycle by a uniform timing error $\epsilon\in[-\delta,\delta]$ around $nT$, with $n\ge1$. Find the leading loss of contrast when $|a|\delta\ll1$ and $\Omega\delta\ll1$.

Solution

Every measurement still records $+1$ or $-1$. Thus the second moment is $\overline{\langle\sigma_x^2\rangle}=1$, and

$$\operatorname{Var}(\sigma_x)=1-J_0(q)^2.$$

The averaged operator reproduces the mean of this measurement; it is not a replacement observable with new outcomes $\pm J_0(q)$. Averaging does not preserve products: $\overline{\sigma_x^2}=I$, whereas $(\overline{\sigma_x})^2=J_0(q)^2I$. In particular, at a zero of $J_0$ the measured variance is one, not zero.

For a narrow timing window, expand the actual trajectory before averaging:

$$\begin{aligned} \overline{\langle\sigma_x\rangle} &=\frac{1}{2\delta}\int_{-\delta}^{\delta} \cos\!\left[q\sin(\Omega\epsilon)\right]d\epsilon\\ &=1-\frac{a^2\delta^2}{6} +O\!\left[(a^4+a^2\Omega^2)\delta^4\right]. \end{aligned}$$

Here $\langle\epsilon^2\rangle=\delta^2/3$. Resolving the drive period alone is insufficient when $|a|\gg\Omega$: the local spin rotation also requires $|a|\delta\ll1$.