---
title: Floquet–Magnus Expansion in Periodic Quantum Systems
url: https://www.emergentmind.com/topics/floquet-magnus-expansion-d717c74f-c792-4731-8bdc-3dbea1add305
type: topic
---

# Floquet–Magnus Expansion in Periodic Quantum Systems

The Floquet–Magnus expansion is a perturbative construction of a time-independent Floquet Hamiltonian for a periodically driven system with \(H(t)=H(t+T)\), defined through the one-period propagator
\[
U(T,0)=\mathcal T\exp\!\Bigl[-\,i\int_0^T H(t)\,dt\Bigr]=e^{-iH_FT}.
\]
In periodic quantum dynamics it provides an effective stroboscopic generator, and in formulations that keep micromotion explicit it yields a factorization \(U(t)=P(t)e^{-iH_F t}\) with \(P(t+T)=P(t)\). The modern literature treats the Floquet–Magnus expansion not only as a formal high-frequency series, but also as an asymptotic tool with rigorous finite-order error control, and as an object whose divergence can coexist with localized or metastable dynamics rather than implying immediate heating or ergodicity [2401.12955][1508.05797][1903.00195].

## 1. Definition and canonical formulas

For a \(T\)-periodic Hamiltonian, the Floquet Hamiltonian \(H_F\) is defined by
\[
U(T)=\mathcal T e^{-i\int_0^T H(t)\,dt}=e^{-iH_FT},
\]
with the short-period requirement \(H_F\to H_0\) as \(T\to0\) in the convention used for periodically driven quantum systems [1903.00195]. A common form of the Floquet–Magnus expansion writes
\[
H_F=\sum_{n=0}^\infty \Omega_n T^n.
\]
The first coefficients are
\[
\Omega_0=\frac1T\int_0^Tdt\,H(t),
\]
\[
\Omega_1=\frac{1}{2iT^2}\int_0^Tdt_1\int_0^{t_1}dt_2\,[H(t_1),H(t_2)],
\]
\[
\Omega_2=\frac{1}{6T^3}\int_0^Tdt_1\int_0^{t_1}dt_2\int_0^{t_2}dt_3
\Bigl([H(t_1),[H(t_2),H(t_3)]]+[H(t_3),[H(t_2),H(t_1)]]\Bigr)
\]
[1903.00195].

A closely related convention expands the Magnus generator \(\Omega(T)\) in
\[
U(T)=e^{\Omega(T)}, \qquad \Omega(T)=\sum_{n=1}^\infty \Omega_n,
\]
and then identifies \(H_F=(i/T)\Omega(T)\). In that notation the first term is \(\Omega_1=-i\int_0^T H(t)\,dt\), and higher orders are ordered time integrals of nested commutators [2507.02736][2603.13821]. The two conventions differ only by indexing and normalization; both encode the same nested-commutator structure.

The leading contribution is the time average of the drive over one period. Higher orders quantify noncommutativity between Hamiltonians at different times. In this sense the Floquet–Magnus expansion is more than a static averaging prescription: it resolves the extent to which time ordering fails to collapse to a simple period average.

## 2. Floquet decomposition, micromotion, and coordinate transformations

Floquet theory permits a decomposition of the exact propagator as
\[
U(t)=P(t)e^{-iKt},
\]
where \(P(t+T)=P(t)\) is unitary and \(K\) is a time-independent Hermitian operator [2401.12955]. One constructive route introduces a time-dependent unitary transformation
\[
U(t)=e^{\Omega(t)}X(t), \qquad X(t)=e^{-iKt},
\]
and chooses \(\Omega(t)\) to be periodic. In this formulation the expansion determines both the effective Hamiltonian \(K\) and the micromotion generator \(\Omega(t)\) order by order [2401.12955].

This representation matters because a finite truncation preserves unitarity by construction:
\[
U_N(t)=\exp(\Omega^{(N)}(t))\,\exp(-iK^{(N)}t),
\]
so the approximation remains a product of unitary exponentials [2401.12955]. In practice, this is one of the main distinctions between Floquet–Magnus constructions and perturbative schemes that do not preserve the Lie-group structure term by term.

The same logic extends beyond linear finite-dimensional quantum systems. For general nonlinear periodic ODEs \( \dot x=f(t,x)\), one may seek a near-identity periodic change of variables \(x=\Psi_1(X,t)\) that transforms the original system into an autonomous equation \( \dot X=G(X)\). In the continuous-changes-of-variables approach, the recursive step is
\[
R_j=U_j-G_j,\qquad G_j=\langle U_j\rangle,
\]
with \(G_j\) fixed by period averaging and Lie brackets replacing operator commutators [1907.05362]. In the linear limit this reduces to the standard Floquet–Magnus structure.

In solid-state NMR, an allied formulation writes
\[
U(t)=P(t)e^{-itF}P^\dagger(0), \qquad P(t)=e^{-iA(t)},
\]
with periodic frame-functions \(A_n(t)\). That formulation emphasizes an operational point: unlike average Hamiltonian theory restricted to integer multiples of \(T\), the Floquet–Magnus representation can track dynamics between stroboscopic observation points [2403.16359].

## 3. Convergence, asymptoticity, and finite-order control

The classical sufficient condition for convergence of the Magnus series on one period is
\[
\int_0^T\|H(t)\|\,dt<\pi,
\]
or, in equivalent notation, \(\int_0^t\|A(s)\|_2ds<\pi\) for the generator \(A(t)=-iH(t)\) [2507.02736][2401.12955]. This criterion is sufficient rather than necessary, and it is often too restrictive for many-body or strong-drive settings.

For generic few-body periodically driven many-body Hamiltonians with local energy scale \(J\), Kuwahara, Mori, and Saito proved
\[
\|\Omega_n\|\le \frac{2V_0\lambda^n}{(n+1)^2}n!, \qquad \lambda=2kJ,
\]
so that \(\|\Omega_n\|T^n\sim n!(CT)^n\). The factorial growth implies that the Floquet–Magnus series is generically asymptotic: partial sums decrease until \(n\sim 1/(\lambda T)\) and then diverge [1508.05797]. In the high-frequency regime \(T\le 1/(4\lambda)\), the optimal truncation
\[
n_0=\Bigl\lfloor \frac{1}{16\lambda T}\Bigr\rfloor
\]
satisfies
\[
\bigl\|e^{-iH_FT}-e^{-iH_F^{(n_0)}T}\bigr\|\le 6V_0T\,2^{-n_0},
\]
and the corresponding stroboscopic approximation remains accurate up to times \(t\approx e^{c'\omega}\) [1508.05797]. This is the rigorous basis of Floquet prethermalization: for \(t\ll\tau\) with \(\tau=e^{c\omega}\), dynamics are governed by a quasi-conserved truncated Floquet Hamiltonian, whereas at much later times a generic nonintegrable system heats toward the infinite-temperature ensemble [1508.05797].

A complementary nonperturbative route derives effective Hamiltonians by iterated integration by parts. In that framework, if \(H_{F,L}\) denotes the Floquet–Magnus truncation through order \(L\), then at stroboscopic times
\[
\|U(mT)-e^{-iH_{F,L}mT}\|=\mathcal O(\Omega^{-(L+1)}),
\]
without assuming convergence of the full infinite series [2504.20533]. Applied to the semiclassical Rabi model, this analysis shows that the rotating-wave approximation can outperform the Bloch–Siegert Hamiltonian in most regimes, while the third-order approximation ultimately outperforms both over very long timescales [2504.20533].

## 4. Divergence, resonances, and unbounded spectra

For infinite-dimensional systems with unbounded Hamiltonians, the operator norms entering the standard convergence criteria may be ill-defined. In a driven anharmonic oscillator,
\[
H(t)=\frac{\hat p^2}{2}+\frac{\omega_0^2}{2}\hat x^2+\frac{\beta}{4}\hat x^4+\lambda(t)g\hat x,
\]
a basis-truncation procedure gives a practical definition of the convergence radius. One chooses an orthonormal basis, truncates to dimension \(D\), computes \(H_F^{(D)}\) and its Magnus coefficients \(\Omega_n^{(D)}\), forms the infinite-\(D\) limits of matrix elements, and defines
\[
T_c=\lim_{n\to\infty}\frac{|(\Omega_n^\infty)_{ij}|}{|(\Omega_{n+1}^\infty)_{ij}|}
\]
[1903.00195].

For the purely harmonic case \(\beta=0\), the ratio \(r_n\) approaches \(2\pi/\omega_0\), so \(T_c=2\pi/\omega_0\). For any \(\beta>0\), however, the asymptotic behavior is
\[
r_n\sim \frac{c(\beta)}{n}, \qquad c(\beta)\approx -\kappa\ln\beta,\quad \kappa\approx 4.4,
\]
hence \(r_n\to0\) and \(T_c=0\): the Floquet–Magnus expansion diverges for all driving frequencies once any nonlinearity is present, even if the anharmonicity is arbitrarily small [1903.00195]. The same work shows that this divergence coexists with exponentially localized Floquet eigenstates in energy space, bounded long-time average energy, and Poissonian rather than Wigner–Dyson level statistics. The divergence therefore does not require quantum ergodicity or heating to infinite temperature [1903.00195].

The interpretation proposed for this coexistence invokes quantum resonances: extremely narrow resonant peaks in \(\langle E\rangle\) versus \(T\) that proliferate arbitrarily close to \(T=0\), analogously to the dense resonances of the kicked rotor at rational \(T/4\pi\). This suggests that the analytic continuation of \(H_F(T)\) can fail at every nonzero \(T\) even when those resonances are unobservable at finite resolution [1903.00195].

Related conclusions appear in the periodically driven Friedrichs models. In the discrete model, where the spectrum is bounded, the Floquet–Magnus expansion converges for \(\omega\gtrsim 4g\), and the zeroth-order effective Hamiltonian correctly predicts an exact Floquet bound state. In the continuous model, where the spectrum is unbounded, no true bound state exists at finite \(\omega\); instead the Floquet–Magnus prediction survives as a metastable resonant state whose lifetime scales as \(\tau(\omega)\sim \sqrt{g\omega}/\lambda^2\) and diverges as \(\omega\to\infty\) [1412.6738]. In the low-frequency regime, the same model exhibits Floquet resonances with exponentially small imaginary parts of quasi-energy,
\[
\Im\,\epsilon_{\rm res}\sim -Ce^{-\alpha/\omega},
\]
which are interpreted as tunneling in energy space [1412.6738].

A recent operator-theoretic extension places these observations on a broader footing. For time-periodic unbounded Hamiltonians, the expansion can be reformulated on the invariant core \(\mathcal C^\infty(H_0)\) using relative bounds with respect to a reference self-adjoint operator \(H_0\), together with spectral clustering and conjugation symmetry. Under these hypotheses, the finite-order effective Hamiltonian
\[
H_{\mathrm{eff}}^{(T,L)}=\sum_{\ell=0}^L T^\ell h_\ell
\]
has coefficients that coincide order by order with the bounded Floquet–Magnus series, and the stroboscopic error obeys
\[
\|e^{-iqTH_{\mathrm{eff}}^{(T,L)}}-U^{(T)}(qT)\|=O(T^{L+2}),
\]
again without requiring convergence of the full series [2605.23734].

## 5. Classical, dissipative, and nonlinear generalizations

The Floquet–Magnus expansion is not confined to closed quantum systems. For classical driven dynamics, including stochastic equations, one can pass from the equations of motion to a periodic Fokker–Planck operator \(L_t\) and define an effective static generator through
\[
U(T,0)=\mathcal T\exp\!\Bigl[\int_0^T L_t\,dt\Bigr]=e^{TL_F}, \qquad
L_F=\sum_{n=0}^\infty \Omega^{-n}L_n.
\]
In Fourier components \(L_m=(1/T)\int_0^Tdt\,L_t e^{im\Omega t}\), the first three terms are
\[
L_0=L_{m=0},
\]
\[
L_1=i\sum_{m\ne0}\frac{[L_{-m},L_m]}{2m\Omega},
\]
\[
L_2=-\sum_{m\ne0}\frac{[L_{-m},[L_0,L_m]]}{2(m\Omega)^2}
-\sum_{m\ne0}\sum_{m'\ne0,m}\frac{[L_{-m'},[L_{m'-m},L_m]]}{3mm'\Omega^2}
\]
[1810.01103].

This master-equation formulation applies to the Langevin equation, the Gross-Pitaevskii equation, and the time-dependent Ginzburg-Landau equation [1810.01103]. In many-body classical systems it is again asymptotic rather than convergent in the strict infinite-order sense:
\[
\|\Omega_F^{(m)}\|\lesssim N\,m!\,\frac{[(\pi k\omega_0)/\Omega]^{m+1}}{(m+1)^2},
\]
with effective truncation order \(m_0\approx \Omega/(k\omega_0)\), and the truncated generator approximately commutes with one-period evolution up to exponentially small error [1810.01103].

Kapitza’s pendulum provides a concrete bridge between classical and quantum uses of the expansion. For a vertically driven pendulum with \(y_P(t)=A\cos(\omega t)\), the third-order Floquet–Magnus calculation yields an effective Hamiltonian
\[
H_{\mathrm{eff}}(\theta,p)=\frac{p^2}{2m\ell^2}-mg\ell\cos\theta+\frac{m(A\omega)^2}{4\ell}\sin^2\theta+O(1/\omega^4),
\]
or equivalently the effective potential
\[
U_{\mathrm{eff}}(\theta)=-mg\ell\cos\theta+\frac{m(A\omega)^2}{4\ell}\sin^2\theta.
\]
The inverted position becomes stable when
\[
A\omega>\sqrt{2g\ell}
\]
[2507.02736]. In the dissipative classical treatment with friction, the effective equation of motion becomes
\[
\ddot\theta=-\gamma\dot\theta-\omega_0^2\sin\theta-\Bigl(\frac{a\Omega}{2l}\Bigr)^2\sin 2\theta,
\]
reproducing the stabilization criterion \(\Omega>\Omega_c=\sqrt2(l\omega_0)/a\) and matching exact stroboscopic dynamics up to long times [1810.01103].

For genuinely nonlinear deterministic systems, the continuous-transformation formalism extends Floquet–Magnus methods to the Van der Pol oscillator and the nonlinear Schrödinger equation on the torus. In that setting the universal commutator integrals are replaced by Lie-bracket expressions for vector fields, while the periodic near-identity map eliminates explicit time dependence order by order [1907.05362].

## 6. Engineering and domain-specific applications

One contemporary use of the Floquet–Magnus expansion is inverse Floquet engineering. For a periodically driven oscillator in the rotating frame, Guo and Xu construct a real-space drive
\[
V(x,t)=V^{(0)}(x,t)+V^{(1)}(x,t)+V^{(2)}(x,t)+\cdots
\]
so that \(V^{(0)}\) produces the desired target Hamiltonian at zeroth order, \(V^{(1)}\) cancels the induced first-order correction, \(V^{(2)}\) cancels the net second-order correction, and so on [2410.10467]. Their six-step iteration computes the error Hamiltonian at order \(\Omega^{-(m+1)}\), converts it through a non-commutative Fourier transform, and defines a correction drive whose rotating-wave contribution precisely cancels that error [2410.10467].

A key technical ingredient is the NcFT “bracket” operation, which translates commutator hierarchies into low-dimensional c-number integrals without explicit operator reordering. This allows the systematic engineering of target Hamiltonians with discrete rotational and chiral symmetries in phase space [2410.10467]. In the example of
\[
\mathcal J(t)=\tfrac12(x^2+p^2)+\beta\cos[x+n\Omega t],
\]
the zeroth-order rotating-wave Hamiltonian has \(n\)-fold rotation and chiral symmetry, while \(\Omega_1\) and higher terms break those symmetries. Adding the analytically derived \(V^{(1)}\) restores the pairwise quasi-energy degeneracies; numerically, the Floquet eigenstate fidelities increase from \(\approx 90\%\) to \(99\%\), and the remaining micromotion and \(t_0\)-dependence are suppressed to \(O(1/\Omega^2)\) [2410.10467].

Two-level systems remain a major testing ground because the \(\mathfrak{su}(2)\) algebra collapses nested commutators to vector cross products. For
\[
H(t)=\tfrac12\Delta\sigma_z+g\cos(\omega t)\sigma_x,
\]
the semiclassical Rabi model has \(H_F^{(1)}=\tfrac12\Delta\sigma_z\), vanishing second-order correction in the specific convention used there, and third-order correction
\[
H_F^{(3)}=\frac{g^2\Delta}{8\omega^2}\sigma_z
\]
[2603.13821]. A parameter scan over \(\omega/\Delta\in[2,20]\) and \(g/\Delta\in[0.1,2.0]\) shows absolute quasienergy errors decreasing from \(1.2\times10^{-2}\) to \(3.0\times10^{-4}\) to \(5.0\times10^{-6}\) at \((\omega/\Delta,g/\Delta)=(10,0.5)\) as one goes from first to third to fifth order, with analogous improvements at lower frequency and stronger drive [2603.13821]. Suitable interaction-picture and adiabatic-basis transformations enlarge the regime of satisfactory agreement with exact results, and symmetry under half-period shifts can force all even orders to vanish [2603.13821].

In solid-state NMR, the Floquet–Magnus expansion has been revisited for the Triple Oscillating Field Technique. There it is emphasized that the method operates directly in the finite-dimensional system Hilbert space rather than in an infinite-dimensional Sambe space, and that its frame-function formulation supplies propagators and effective Hamiltonians that describe the evolution between stroboscopic points [2403.16359]. This places Floquet–Magnus methods at the intersection of effective-Hamiltonian theory, spectroscopy, and time-dependent quantum control.

Taken together, these applications show that the Floquet–Magnus expansion serves two technically distinct roles. First, it is a high-frequency or asymptotic reduction tool that exposes prethermal generators, suppressed heating, and metastable resonant structures. Second, it is a constructive design formalism for building target effective Hamiltonians and symmetry-protected Floquet phases. The current literature indicates that these roles remain meaningful even when the full infinite series diverges, provided that truncation, micromotion, and error control are handled with the appropriate operator-theoretic or model-specific framework [2504.20533][2605.23734][1903.00195].

Source: https://www.emergentmind.com/topics/floquet-magnus-expansion-d717c74f-c792-4731-8bdc-3dbea1add305