WorksPhysics and Astronomy
Fer expansion of driven spins: closed-form recursion and no-go theorems for termination
Abstract
The Fer expansion factorizes the propagator of a driven quantum system into exponentials of time-integrated effective Hamiltonians, which govern the evolution left over by the previous factors. Every truncation is unitary, and adding a factor squares the size of the remainder, but each effective Hamiltonian is an infinite series of nested commutators of the previous one. For , in every representation, from qubits to spins of any dimension, we represent the effective Hamiltonians by time-dependent vectors in and resum this series exactly. The result is a closed map from one vector to the next, in which the infinite series reduces to the previous vector crossed once and twice with its time integral, weighted by elementary functions. Using this map, we address the question of when the expansion can be stopped after finitely many factors. This is possible whenever the leftover evolution is known in closed form, for instance when the effective Hamiltonian commutes with itself at all times or precesses about a fixed axis with constant angular velocity. We prove that, for real-analytic drives, neither happens beyond the first factor unless the first exponential is already exact. Stopping the expansion therefore yields no new exactly solvable drives.
Introduction
A spin driven along a time-dependent axis is the working model of magnetic resonance [1] and, for qubits, of most quantum-computing platforms [2]. Its Hamiltonian is a time-dependent combination of the generators of su(2), up to a multiple of the identity that only contributes a global phase. Hamiltonians of the same form, in other representations of the algebra, describe a spin of any size, an ensemble of two-level atoms driven collectively, and, through Schwinger’s oscillator model [3], two bosonic modes coupled by a time-dependent beam splitter. In these settings, designing pulse sequences, decoupling schemes, and gates requires the evolution operator generated by the Hamiltonian , or an effective Hamiltonian that approximates it [4–6]. Formally, is given by the time-ordered exponential , whose expansion in powers of is the Dyson series [7, 8]. This is a formal solution only, and truncating it breaks unitarity at every finite order [9].
Several methods avoid this problem by keeping the propagator unitary at every order. The best known is the Magnus expansion [9–12], which writes the propagator as a single exponential, , with skew-Hermitian and built from nested commutators of at different times. Its convergence and error estimates are well charted [9, 13, 14], where is an infinite series, in which each new term adds one more commutator of the true Hamiltonian.
A second route exploits the algebraic structure of instead of expanding in nested commutators. If , with real coefficients , is
built from Hermitian generators that close under commutation into a finite-dimensional Lie algebra , the Wei–Norman theorem [15–17] factorizes the propagator over a basis of , locally in time, into the ordered product , and turns the Schrödinger equation into a closed system of scalar ordinary differential equations (ODEs) for the . Except for special cases such as solvable algebras [16], the can diverge at finite times [18], beyond which the factorization has to be restarted. The method has been revisited in tutorial form for Gaussian and non-Gaussian quantumoptical models [19], underlies unitary integration [20], and its equations and their integrability have been studied in detail [21, 22]; the same factorizations are behind the Lie-group integrators of numerical analysis [23].
The Fer expansion [24–26] lies between these two approaches and has been explored much less, although it has been used to build numerical integrators [27] and compared with the Magnus and Floquet expansions in solid-state NMR [28]. Like Magnus, it is built from nested commutators alone and needs no prior knowledge of the algebra generated by ; like Wei–Norman, it writes the propagator as a product of exponentials. The factors are peeled off one at a time, each generated by the time integral of an effective Hamiltonian that governs the evolution left over by the previous ones. Every truncation is therefore unitary, and what it leaves out is not an uncontrolled error but the evolution generated by the next effective Hamiltonian . Each factor also squares the size of this remainder, so that, within the convergence domain of the expansion [25], recently enlarged by about 30% [26], the error decreases doubly exponentially with the number of factors [9]. The price is that each effective Hamiltonian is an infinite series of nested commutators of the previous one, which in general has to be truncated or evaluated numerically at every order.
First we show that, for , the Fer expansion can nevertheless be made explicit at every order: the infinite series of nested commutators that defines each effective Hamiltonian can be summed exactly, in every representation of the algebra. Each effective Hamiltonian is a combination of the three generators of with time-dependent coefficients that form a vector in , and commutators become cross products of these vectors. One step of the recursion thus becomes an explicit map from each vector to the next: the new vector is the previous one crossed once and twice with its time integral, and the two terms are weighted by elementary trigonometric functions. Every factor of the Fer product is then an explicit rotation of the spin.
We then use the closed map to answer a question that the structure of the Fer product raises: can the product stop by itself? If at some step the effective Hamiltonian commutes with itself at all times, the next one vanishes identically, there is nothing left to peel off, and finitely many factors give the exact propagator. More generally, the product can be stopped as soon as the leftover evolution is known in closed form, for instance when the effective Hamiltonian precesses about a fixed axis with constant angular velocity, as in Rabi’s rotating field [29]. If a generic drive could be steered onto one of these two classes after finitely many steps, the Fer expansion would produce new exactly solvable drives, starting from the drive rather than from the solution.
Such drives would be valuable, since closed-form propagators are rare. For a qubit, the classic examples are Rabi’s rotating field [29] and the Landau–Zener sweep [30]. Larger families have been obtained by reverse-engineering the two-level Schrödinger equation [31, 32], or through the dynamical invariants of Lewis and Riesenfeld [33], which trade the linear problem for an auxiliary nonlinear equation, such as the Ermakov–Pinney equation of the time-dependent harmonic oscillator [34], and underpin the inverse engineering of shortcuts to adiabaticity [35]. In every case the drive is built from the solution, and not the other way around: for a drive given in advance, the propagator is in general not known in closed form.
Among the drives with a closed-form propagator, we consider two classes on which the Fer product can be stopped: effective Hamiltonians that commute with themselves at all times, and those that precess about a fixed axis with constant angular velocity, as in Rabi’s rotating field. Whether the product can reach either class at some order is not obvious in advance, since the recursion mixes the components of the drive nonlinearly at every step. We prove two no-go theorems showing that it cannot, except trivially. More precisely, for a real-analytic drive, an effective Hamiltonian beyond the first falls into either class only if it vanishes identically and the original drive already commutes with itself at all times, in which case the first exponential is already exact: the recursion creates no stopping point that was not already present in the original drive.
The paper is organized as follows. Section II recalls the Fer recursion, and Sec. III isolates what in it does not depend on the algebra, including the condition under which the product stops. Section IV derives the closed map and shows how it makes every order of the expansion explicit, Sec. V proves the two no-go theorems, and Sec. VI discusses where these results apply. The derivations are collected in Appendices A–C.
The Fer expansion
Throughout, we set and a dot denotes the total time derivative . The evolution operator of a Hamiltonian obeys
Following Refs. [24, 25], and in the spirit of the average-Hamiltonian and Floquet–Magnus methods of magnetic resonance [36] [4] and of dynamical decoupling [5], one writes it as an ordered product of exponentials,
in which the remainder obeys a Schrödinger equation of the same form,
with an effective Hamiltonian generated recursively by
where , conjugation by an exponential has been written as the exponential of the adjoint action (Appendix A), and the exponents are
with . This is the defining step of the expansion: at order the integrated part of the dynamics is peeled off into , leaving a smaller remainder for the next step [24]. Each step squares the size of that remainder, which makes the expansion superconvergent, in the sense of the accelerated perturbation schemes of Refs. [37, 38]; its convergence domain is analyzed in Ref. [25], which for a Hermitian guarantees convergence while , with the operator norm; this radius was recently raised to about 2.6058 [26]. Expanding the two exponentials of in Eq. (II.4) and doing the integral collapses it into a single series of nested commutators,
with no auxiliary integrals left to perform.
The Fer map on a Lie algebra
Two features of the recursion hold for any algebra. We isolate them here, before specializing to .
The recursion stays in the algebra
Let the initial Hamiltonian be a real combination of Hermitian generators of a Lie algebra , ,
and , with . Every term of Eq. (II.6) is an iterated commutator of with , so the next-order Hamiltonian is again a combination of the same generators, and by induction
for all . The recursion never leaves ; it only reshuffles the real coefficient vector. It therefore closes on the coefficient vectors themselves, taken as whole functions of time,
with the Fer map fixed by the structure constants of through Eq. (II.6). Iterating it generates the chain
to any desired order. The Fer map is not invertible in general: if commutes with itself at all times, it also commutes with its own integral , every nested commutator in Eq. (II.6) vanishes, and , so that sends all the drives of this kind to the zero drive.
Where the chain stops
Whatever the algebra, the evolution generated by is solvable in closed form whenever the Hamiltonian commutes with itself at all times,
We call family the class of drives obeying Eq. (III.5), and we say that the coefficient vector belongs to when its Hamiltonian does.
Every Fer chain that reaches family stops there, as anticipated in Sec. III A. Indeed, if , the Hamiltonian commutes with its own integral, Eq. (II.5),
because the integrand vanishes identically by Eq. (III.5). The innermost commutator of every term of Eq. (II.6) is , so the whole series vanishes term by term and
Reaching at order means that the factorization closes there: and is exact, with no remainder left. Since , one has , and the chain stays at forever after.
The question of § 1 is now precise: for which does the map of Eq. (III.3) reach after finitely many steps? We exclude because it is the trivial case: there already at the first step, so for all and the expansion stops after a single factor, , which is the time-ordered exponential of § 1 with the time ordering dropped. Answering the question means going back along the chain, one step at a time: finding the drives that one step sends into .
The Fer map on su(2)
For the map can be written down in closed form. A component of the Hamiltonian along the identity commutes with everything: it drops out of every commutator in Eq. (II.6) and only contributes a global phase to , so we take traceless. We take to be a vector of Hermitian matrices of any dimension, obeying [3]
that is, twice the spin operators of any nontrivial representation of , for instance the Pauli matrices for a qubit, and write, as in Eq. (III.2),
with a real three-dimensional vector, the drive vector about which the mean spin precesses, as the Bloch vector of a qubit in the representation of Feynman, Vernon, and Hellwarth [39], and . The norms of the two vectors are abbreviated as and , with the Euclidean norm. By construction
Since the commutator is bilinear, Eq. (IV.1) turns the commutator of two combinations of the into the cross product of their coefficient vectors,
where , so that the operator series of Eq. (II.6) becomes a recursion entirely in ,
with numerical coefficients . This series resums into elementary functions: summing its odd and even powers separately gives the closed map
built from the two auxiliary vectors
and weighted by the two scalar functions
Equation (IV.6) is the closed-form realization of the abstract map of Eq. (III.3) for : it is what we iterate from now on. Appendix B derives it twice, once by resumming the series of Eq. (IV.5) and once, independently, by inserting the Rodrigues rotation formula [3] into the recursion of Eq. (II.4) and integrating. Each factor of the product of Eq. (II.2) is an explicit rotation: rotates the spin by the angle about the unit vector , and in any representation it is given by the corresponding rotation matrix [3]. For a qubit, with the vector of Pauli matrices, .
Two properties of the map will be used repeatedly. First, both auxiliary vectors lie in the plane orthogonal to , Figure 1: is orthogonal to both and , and is times the projection of onto . Hence the whole output lies in ,
Figure 1. One step of the Fer map on su(2), Eq. (IV.6), drawn in the space of the coefficient vectors. The integrated drive , in green, is normal to the plane , in red, through the origin . The auxiliary vectors , in orange, and , in purple, span ; the latter lies along the projection of the drive , in brown, onto the plane (dotted parallelogram). The output , in blue, is assembled from them (dashed parallelogram) with the weights and of Eqs. (IV.8) and (IV.9), both evaluated at the norm , and never leaves .
Second, makes both auxiliary vectors vanish at the origin, so every Fer output starts there,
a constraint that the initial Hamiltonian does not carry.
Rigidity of the chain
With the map available in closed form, the question of Sec. III B can be answered by inspection of a single backward step. We do it first for family , the drives that commute with themselves at all times, and then for the uniformly precessing drives of Rabi’s rotating field, a textbook solvable class outside . In both cases the answer is negative: the recursion reaches the class at order only when the original drive was already in , and then only through the zero drive.
Throughout this section the original drive is taken to be real-analytic in . The weights and of Eqs. (IV.8) and (IV.9) are even, and their singularities at the origin are removable, hence they are analytic functions of , so every produced by Eq. (IV.6) is real-analytic as well.
Drives with a fixed direction
Read through the commutator of Eq. (IV.4), the defining condition of family , Eq. (III.5), becomes
i.e., the trajectory of never leaves a fixed direction in : it moves along a single line through the origin. Since by Eq. (III.7), the following theorem amounts to a single step back along the chain: a drive that one step sends into was already in .
Theorem 1. Let the original drive be real-analytic. In the Fer expansion associated with the algebra, if the coefficient vector at some order belongs to family , then , and the coefficient vector of the original Hamiltonian belongs to family as well.
Proof of Theorem 1. We prove the one-step implication, valid for every , from to and , and conclude by descending induction along the chain of Eq. (III.4).
Assume : both and lie on a fixed line through the origin. By Eq. (IV.10), is orthogonal to this line wherever , hence at all times unless , since the zeros of a real-analytic function are isolated. Either way . Differentiating Eq. (IV.6) and projecting on (Appendix C), every term is a cross product with and drops out, except one of the two generated by . What survives is the necessary condition
By the triple product identity and the Lagrange identity,
so that Eq. (V.2) reads
At every instant this holds either because or because , and the first alternative can only occur at isolated times. Indeed, as noted at the beginning of this section, the composition is real-analytic, even though the norm itself need not be where vanishes. Near the origin,
and since , the composition equals at , so it is not identically zero, and its zeros are isolated. Equation (V.4) then forces on the complement of a discrete set, which is dense, and by continuity everywhere:
that is, .
Two consequences follow. First, both auxiliary vectors of Eq. (IV.7) vanish identically, and with them the output of Eq. (IV.6): . Second, since , Eq. (V.6) reads . If , then . Otherwise, on any interval where , its derivative is parallel to it and its direction does not change; being real-analytic, then stays on that line at all times, and so does . This is .
Remark. The proof never inverts , which in general cannot be done (Sec. III A). It does not reconstruct from : it takes any drive that one step sends into and constrains it using only the forward map, Eq. (IV.6).
Together with Eq. (III.7), Theorem 1 says that the drives that one step sends into , or into , are exactly those already in ; in particular, no real-analytic drive outside produces, at any later order, an effective Hamiltonian that commutes with itself at all times.
Precessing drives
Family is not the only solvable class in . After steps the propagator reads, by Eq. (II.2), , with generated by . The product terminates when vanishes, which involves only family , Eq. (III.7), and is settled by Theorem 1. The expansion can be stopped just as well whenever belongs to a class whose dynamics is solved exactly: , and with it the whole propagator, is then known in closed form. Here we focus on the best-known such class, Rabi’s rotating field [29], which we call family : it collects the drives whose vector precesses about a fixed unit axis with constant angular velocity ,
Despite the richer time dependence, these drives are solvable too: in the frame rotating about , the propagator generated by factorizes as
where is generated by the time-independent Hamiltonian . Yet the chain cannot reach either.
Theorem 2. Let the original drive be real-analytic. In the Fer expansion associated to the algebra, if the coefficient vector at some order belongs to family , then , and the coefficient vector of the original Hamiltonian belongs to family .
Proof of Theorem 2. Since , is a genuine Fer output and therefore vanishes at by Eq. (IV.11). A member of has constant magnitude by Eq. (V.7), so one that vanishes at the origin vanishes identically. Hence , and Theorem 1 gives .
Conclusion
For a spin of any size, and in every representation of , we have summed the series of nested commutators that defines each step of the Fer expansion into a closed map on time-dependent vectors in , with two elementary trigonometric weights. Each factor of the product follows from the previous one through a single time integral and elementary functions of the norm of the previous integral, . The known strengths of the expansion thus become directly usable. The product is unitary at every truncation, and each factor squares the size of the remainder, so that the error decreases doubly exponentially with the number of factors [9], within a convergence domain that has recently been enlarged by about 30% [26]. A few factors therefore give an accurate propagator for any spin drive, such as the pulse sequences of magnetic resonance [4] and dynamical decoupling [5], or the control pulses that implement quantum gates [6].
With the closed form obtained here and the recently enlarged convergence domain, the Fer product is thus well suited also to quantum algorithms, in which product formulas are among the most widely used methods to simulate time evolution [40, 41]: a truncated Fer product is a product formula of explicit rotations, which can be applied directly as gates.
We have also used the closed map to ask whether the Fer product can be stopped after finitely many factors on one of two classes of effective Hamiltonians whose evolution is known in closed form: those that commute with themselves at all times, on which the product terminates, and those that precess about a fixed axis with constant angular velocity, as in Rabi’s rotating field. For a real-analytic drive, no effective Hamiltonian beyond the first falls into either class unless the first exponential is already exact. In , exactly solvable drives therefore still have to be engineered backward, from the solution to the drive [31, 32, 35], and whether the chain can reach other solvable classes remains open. Running the Fer map until the effective Hamiltonian becomes exactly solvable is nonetheless a direct method to search for solvable dynamics, since it starts from the drive rather than from the solution. For the families considered in this work, it fails for , but the first part of our analysis, in Sec. III, holds for any finite-dimensional Lie algebra, so the method can be applied to other algebras, for instance those that describe squeezing and parametric amplification in quantum optics [19], where the answer may differ.
Acknowledgments
The authors thank Riccardo Castellano (Department of Applied Physics, University of Geneva) for helpful discussions on the presentation of the introduction and conclusions.
Use of AI
The authors used Claude Code (Anthropic) to assist with spelling corrections, graphical refinement of figure 1, and searches for relevant literature. The scientific content, including the theorems and their proofs, is the authors’ own. The authors reviewed the output and take full responsibility for the manuscript.
Data availability
No data were created or analyzed in this article.
Conjugation formula
Consider two operators and the one-parameter family with . Differentiating times gives
as follows by induction, since one further derivative produces one further commutator with between the same two exponentials. Expanding in a Taylor series around and evaluating it at then yields
valid for arbitrary . This is what allows conjugation by an exponential to be written as the exponential of the adjoint action, in the recursion of Eq. (II.4) as in its double-integral form, Eq. (B.7).
We now translate Eq. (A.2) into vector form, with and for real and Hermitian generators . By the commutator of Eq. (IV.1), a single application of gives
The result is again of the form , so the same computation iterates: by linearity of , each further application of contributes one factor of and one more power of , whence by induction. Substituting into Eq. (A.2) gives
Closed form of the Fer map on
Here we derive Eq. (IV.6) in two independent ways.
From the series
The key structural fact is that generates only two directions out of : every power with is a multiple of either or . Indeed, from , a second application and the BAC–CAB identity give
and a third, using ,
Each further application swaps the two structures, and every second one brings a factor , so the pattern closes at all orders,
with and the auxiliary vectors of Eq. (IV.7). Splitting the series of Eq. (IV.5) into even and odd parts and inserting Eqs. (B.3) and (B.4),
and the two sums are precisely the Taylor series of the elementary functions of Eqs. (IV.8) and (IV.9),
which is Eq. (IV.6). In particular and are the closed-form sums of the even and odd subseries of the recursion, and the apparent singularities at are removable.
From the Rodrigues formula
The same result follows from the recursion of Eq. (II.4) without expanding in commutators. Written as a double integral, equivalent to Eq. (II.4) since both reduce to the series of Eq. (II.6), the recursion reads
The conjugation of Eq. (A.4) resums, for , into the Rodrigues rotation formula [3]
where is the Euclidean norm in . Substituting Eq. (B.8) into Eq. (B.7), carrying out the integrations over the auxiliary variables and using the orthogonality gives
whose vector of coefficients is again Eq. (IV.6) with the weights of Eqs. (IV.8) and (IV.9). This is an independent check of the resummation above.
Derivative of the Fer output
The proof of Theorem 1 needs the time derivative of the closed map of Eq. (IV.6). From and , the auxiliary vectors of Eq. (IV.7) obey
so that differentiating Eq. (IV.6) gives
Here a prime denotes the derivative of a function with respect to its argument, , and we used , and likewise for . Since and are even and regular at the origin, so are and , and Eq. (C.2) stays finite also where vanishes. Every term on the right-hand side is a cross product with , and therefore orthogonal to it, with the single exception of . Projecting Eq. (C.2) on thus leaves
which vanishes whenever , since then , as shown in the proof of Theorem 1. This is Eq. (V.2).
References
References
- [36]E. S. Mananga and T. Charpentier, J. Chem. Phys. 135, 044109 (2011).
- [37]W. Scherer, Phys. Rev. Lett. 74, 1495 (1995).
- [38]W. Scherer, J. Math. Phys. 39, 2597 (1998).
- [39]R. P. Feynman, F. L. Vernon, and R. W. Hellwarth, J. Appl. Phys. 28, 49 (1957).
- [40]S. Lloyd, Science 273, 1073 (1996).
- [41]A. M. Childs, Y. Su, M. C. Tran, N. Wiebe, and S. Zhu, Phys. Rev. X 11, 011020 (2021).
