Finite-temperature toroidal moment amenable to direct observation in an Fe10Dy10 molecular ring Authors: Alessandro Soncini, Kieran Hymas, Jonas Braun, Yannik F Schneider, Simone Calvello, Amer Baniodeh, Yanhua Lan, Wolfgang Wernsdorfer, Marco Affronte, Christopher E Anson, Annie K Powell Published: 2026 ABSTRACT AbstractSingle-molecule toroics host closed magnetic vortices carrying toroidal moments τ, whose electric-dipole symmetry enables magnetoelectric spin control. Yet opposite toroidal chiralities are degenerate in conventional magnetic fields, making direct detection of τ challenging. Current approaches probe toroidal dynamics only indirectly through weak residual magnetism, while finite-temperature toroidal polarisation and realistic preparation/readout conditions remain unestablished. Here we show that the Fe10Dy10 molecule hosts a 62-billion-dimensional low-energy manifold pervaded by toroidal character, rendered tractable by an ab initio-informed transfer-matrix framework that reproduces experimental data. The model reveals a large toroidal response robust to thermal fluctuations, quantified by a finite-temperature toroidal susceptibility ξ. We then propose a preparation-and-readout protocol in which a train of temporally asymmetric near-infrared pulses accumulates toroidal polarisation, converted through magnetoelectric response into a measurable electric-field-induced magnetic signal. These results establish Fe10Dy10 as a molecular system where τ can be prepared, accumulated and read out under realistic conditions.Subject terms: Magnetic properties and materials, Magnetic materials, Electronic structure of atoms and molecules, Magnetic properties and materials Introduction The interplay between spin-orbit coupling and low-symmetry electrostatic potentials at the nanoscale, such as those present in chiral matter, is emerging as a powerful route to engineer quantum functionalities. It underpins phenomena of direct interest to nanoscience and quantum technologies, including chirality-induced spin selectivity (CISS)1, non-reciprocal dichroism2, magneto-electric coupling3,4, and the creation of skyrmionic or antiskyrmionic excitations5–7. These effects exemplify how tailored symmetry breaking at the molecular or nanostructural level can give rise to robust, topologically protected quantum states. Among the most striking manifestations of such symmetry breaking is the toroidal moment, first introduced by Zeldovich to describe electromagnetic multipoles mediating parity-non-conserving weak interactions8,9, and later observed in atomic Cs10. This electromagnetic multipole can be pictured as a poloidal current encircling a torus or as a closed loop of magnetic dipoles arranged head-to-tail, resulting in no net magnetic poles-hence the alternative name anapole. Despite its magnetic origin (odd under time reversal), the toroidal moment transforms under spatial inversion like an electric dipole, so that a counter-clockwise magnetic vortex is converted into a clockwise one under parity reversal3,8,9,11. This dual magnetic and electric nature of the toroidal moment implies that a system with toroidal polarisation can support linear magneto-electric coupling3, providing a direct route to spin-electric effects, i.e., the electric control of atomic and molecular spins and of their interactions4,12–15. Moreover, in chiral nanostructures based on SMTs16, where parity symmetry is intrinsically broken, uniform magnetic fields can in principle induce toroidal polarisation through mixed magnetic–toroidal response effects, with magnetic and toroidal responses partially superimposed11. For achiral systems, producing a toroidal polarisation typically requires inhomogeneous magnetic fields with finite curl ∇ × B3,11, although recently an EPR-based spectroscopic alternative has been proposed by some of us for non centro-symmetric samples17. While classical analogues of toroidal moments have been engineered and detected in non-collinear low-dimensional magnets18, and in metamaterials19,20, to date, microscopic quantum states carrying a molecular toroidal moment have remained experimentally elusive. The reason is mainly two-fold. On the one hand, molecular quantum states carrying clockwise and anticlockwise toroidal moments of equal magnitude are necessarily doubly-degenerate in the absence of a finite curl ∇ × B, difficult to produce and control at the scale of a small molecule, which yields a zero net toroidal polarisation. On the other hand, for strongly spin-orbit-coupled, non-frustrated single-molecule toroics, a quantitative framework connecting the collective contribution of thermally populated toroidal excitations to finite-temperature toroidal linear response and experimental readout has not yet been established. Both challenges to the observation of a toroidal polarisation will be addressed in this study, where we establish a quantitative framework linking finite-temperature toroidal response, dynamical preparation, and experimental readout. To illustrate this point, we note that the only work in the literature claiming observation of a molecular toroidal state consists of a recent report of a record-long magnetic relaxation time detected via microSQUID magnetometry, attributed to the low-lying ferrotoroidic and antiferrotoroidic states of a Ga7Dy6 Single-Molecule Toroic (SMT)21. Of course, since the ferrotoroidic states of a centrosymmetric molecule cannot carry a magnetic moment by definition, while an antiferrotoroidic state can be weakly magnetic but cannot carry a net toroidal moment, the involvement of toroidal dynamics may only be inferred indirectly from studying the relaxation of the weakly magnetic non-toroidal states generated by the time-varying uniform magnetic field21. Related observations have also been reported in recent work on nuclear-spin-dependent toroidal ground states in 163Dy4L4 SMTs22, as well as in studies employing 161Dy-Mössbauer spectroscopy in a Co3IIIDy3III SMT23. While such inference is interesting, the precise relaxation mechanism of the toroidal states remains silent to typical magnetometry experiments, urging the need for new experimental proposals to detect direct signatures of these elusive molecular states. Recent studies on transition-metal spin triangles have explored electric-field control of chiral and toroidal-like spin states4,24. In such weak spin–orbit spin–frustrated systems, toroidal character is closely intertwined with spin chirality and electric-field-induced modifications of spin-Hamiltonian parameters, whereas the direct preparation and readout of a finite-temperature toroidal polarisation in strongly spin–orbit-coupled and non-frustrated SMTs remains an open challenge. Nevertheless, SMTs remain indeed the most promising candidates where toroidal moments in molecules could in principle be directly observed25–28, also using their ability to sustain linear magnetoelectric coupling29,30. They consist of metal ions with strong on-site magnetic anisotropy-typically 4f lanthanides-weakly coupled in ring-like topologies that stabilise vortex arrangements of 4f electron moments31–34. SMTs are thus crucial platforms both to probe fundamental molecular physics beyond a simple magnetic dipole picture26,31,35, and to develop hyper-dense molecular memories3, molecular spintronics devices35,36 and toroidal qubits37 Their ground states possess vanishing net magnetic dipole moments but support doubly degenerate, counter-rotating quantum vortices whose toroidal moment is where ri locates ion i and Mi is its magnetic moment. Equation (1) suggests two chemical design principles for maximising ground-state toroidal moments: employ ions with large magnetic moments, and enlarge the molecular ring radius. Dysprosium(III) ions, with unquenched orbital momentum and ~ 10μB moments, are ideal, and Dy-containing molecular wheels dominate SMT research25–27. While hexa-, octa- and decanuclear Dy wheels are known32,38, clear evidence of ground-state toroidicity has thus far been limited to Dy6 molecules and only with theoretical support28,39. Hybrid architectures introduce additional design freedom. Incorporating 3d transition metals to couple Dy3 vortex motifs can generate intramolecular ferrotoroidic states34,40–42, while ultra-large heterometallic rings push size and magnetic complexity43,44. Building on this strategy, some of us reported a family of icosanuclear [Fe10Ln10] wheels with alternating FeIII and LnIII ions in a nanoscale torus45, where the Gd-analogue showed proximity to a quantum critical point44. The fixed point in these previous studies was to keep fixed the 3d metal as FeIII, while the effect of varying the LnIII on magnetic and optical properties was tested. Here we investigate the Fe10Dy10 member of that family, whose large size and complex magnetic structure give rise to a dense manifold of low-lying toroidal microstates. This system provides a unique opportunity to quantify finite-temperature toroidal polarisation as an emergent thermodynamic property arising from multiple thermally accessible toroidal excitations. Combining multi-reference ab initio calculations with a tailor-made perturbatively corrected transfer-matrix approach describing a quantum-decorated non-collinear Ising chain, we quantitatively reproduce the magnetic and calorimetric experimental data and demonstrate an exceptionally large and thermally resilient toroidal response. We further propose and develop a quantitative framework for the ultrafast dynamical preparation and magnetoelectric readout of molecular toroidal states. In particular, we show that temporally asymmetric near-IR ultrafast electric-field driving can generate an experimentally detectable nonequilibrium toroidal polarisation through the accumulation of population imbalance between time-reversed toroidal states. Through the development of an ab initio-informed model of the magnetoelectric response, we show that the resulting toroidal polarisation generates a linear magnetoelectric effect predicted to yield an electric-field-induced magnetic signal within the sensitivity range of solid-state high-sensitivity magnetometers. Analysis of the ultrafast accumulation mechanism further reveals that the accumulated toroidal polarisation scales as ∣τ∣4, establishing a design principle for large molecular toroidal systems amenable to direct observation. Results We investigate the microscopic origin, finite-temperature robustness, and ultrafast preparation and magnetoelectric readout of toroidal states in Fe10Dy10. Even when restricting the model to the low-energy manifold, the resulting Hilbert space spans ~ 62 billion states, beyond the reach of direct ab initio approaches. We therefore adopt a hybrid atomistic strategy combining ab initio crystal-field calculations, transfer-matrix methods, nonequilibrium open-system dynamics, and perturbative evaluation of the magnetoelectric response. Ab initio local crystal field states The local states are determined via multiconfigurational scalar relativistic ab initio calculations on wheel fragments comprising a single open-shell ion (DyIII or FeIII) at a time, surrounded by an appropriate ligand environment using the ab initio multiconfigurational software CERES developed by some of us ref. 46. For all the DyIII ions, our results indicate energetically well-isolated ground KDs with strongly axial magnetic anisotropy (see “Methods” in CAHF/CASCI-SO calculations). Interestingly, the DyIII ions with the more axial g-factors and larger gaps to the first excited KD (Dy1, Dy2, Dy5) sit on higher curvature regions of the ellipse. The magnetic axes are not perfectly planar with respect to the median plane of the Fe10Dy10 wheel but instead observe an alternating canting pattern about the ring (Fig. 1a), consistent with the molecular symmetry. From Fig. 1b it is nonetheless clear that the planar projections of the axes describe a vortex like arrangement which is a necessary (though not a sufficient) condition for the existence of a ground-state toroidal moment. Previous calculations on a related Fe8Dy8 ring were reported, where, perhaps surprisingly, the orientations of the Dy Ising axes are not consistent with the molecular S8 symmetry. Further calculations for each of the symmetry unique FeIII ions using MOLCAS47 revealed energetically well-isolated orbitally non-degenerate SFe = 5/2 pure spin ground multiplets with negligible zero-field splitting. Schematic depiction of the Fe10Dy10 paramagnetic skeleton. a Side profile of the wheel. FeIII ions are depicted as red spheres and DyIII ions as blue spheres. The ab initio-calculated principal magnetic axis ui of each of the DyIII ground doublets (reported in Table 2) are shown as an orange arrow. b Bird’s-eye view of Fe10Dy10 showing the `long' and `short' axes of the ellipse. c Exchange connectivity of a repeating fragment of the ring. The Dy-Fe exchange coupling between DyiIII and FeiIII and between FeiIII and Dyi+1III are JA and JB, respectively. The weaker FeIII-FeIII antiferromagnetic exchange coupling is JC. DFT magnetic coupling constants Exchange coupling in Fe10Dy10 was modelled via broken symmetry density functional theory (DFT) calculations for each FeIII-DyIII-FeIII fragment (see Methods in Broken symmetry density functional theory calculations). We utilised Yamaguchi’s method48 to extract Heisenberg exchange coupling constants JA and JB between the FeIII and neighbouring DyIII spins, as well as a next-nearest neighbour constant JC to account for FeIII-FeIII exchange. A schematic of the exchange connectivity is given in Fig. 1c, while the calculated constants are reported in Table 1. Table 1Exchange coupling constants determined via broken symmetry density functional theory calculations for symmetry unique fragments of Fe10Dy10FragmentJA/cm−1JB/cm−1JC/cm−1Fe10-Dy1-Fe10.32671.041− 0.1682Fe1-Dy2-Fe20.28541.125− 0.1656Fe2-Dy3-Fe30.57710.998− 0.1657Fe3-Dy4-Fe40.36140.862− 0.2597Fe4-Dy5-Fe50.13251.028− 0.0263JA and JB are nearest-neighbour FeIII-DyIII and JC are next-nearest neighbour FeIII-FeIII exchange couplings as depicted in Fig. 1c. Exchange coupling constants determined via broken symmetry density functional theory calculations for symmetry unique fragments of Fe10Dy10 JA and JB are nearest-neighbour FeIII-DyIII and JC are next-nearest neighbour FeIII-FeIII exchange couplings as depicted in Fig. 1c. Our calculations revealed asymmetric ferromagnetic exchange couplings between FeIII and DyIII nearest neighbours, with Dy ions on lower curvature regions (Dy3, Dy4) associated with stronger JA and weaker JB, and a systematically weaker antiferromagnetic coupling between FeIII-FeIII next-nearest neighbours. Theoretical modelling of an analogous Fe10Gd10 system44 demonstrated a similar pattern. To further assess the robustness of these exchange parameters, we performed an independent statistical χ2 analysis based on a three-parameter fit of the magnetic susceptibility data (Supplementary Note 8). The best-fit values, JAfit=0.8−0.55+0.2 cm−1, JBfit=0.6−0.15+0.7 cm−1 and JCfit=−0.10−0.05+0.10 cm−1, are consistent with the broken-symmetry DFT estimates, which lie within the 68% confidence region of the fit. This independent analysis validates both the magnitude and sign of the dominant exchange interactions used in the ab initio-parameterised transfer-matrix model. Eigenstates and finite-temperature response properties The magnetic excitations of Fe10Dy10 were obtained via the following spin Hamiltonian, whose parameters are evaluated via ab initio calculations: where Hex and Hdip are the isotropic exchange and dipolar interaction terms, respectively (see Eqs. (18), (19) in “Methods”), HZee is the Zeeman Hamiltonian describing the coupling to a uniform external magnetic field (Eq. (20) in Methods), and HTor is an extension of the Zeeman Hamiltonian (Eq. (21) in “Methods”) accounting for the coupling of the wheel’s toroidal moment to ∇ × B (vide infra). From the ab initio results, it is clear that for T < 90K Eq. (2) can be projected on the product basis ∣σM, where the 210 Ising configurations are labelled by σ=σ1,…,σ10, with σi = ± 1 selecting between the two components of the ground state easy-axis KDs of the i th DyIII ion, consisting of almost pure MJ = ± 15/2 total angular momentum states projected on their respective non-collinear principal magnetic axes ui (see Table 2), while the 610 spin product states are labeled byM = M1, M2, …M10, with Mi = − 5/2, − 3/2, ⋯ + 5/2 are the degenerate spin states of the S = 5/2 centred on the ith FeIII ion. In this approximation, the ring Hamiltonian is block diagonal with respect to the 210 possible Ising configurations, and so each block σ may be considered individually (see “Methods” in The interaction Hamiltonian). Table 2Ab initio computed g-values and the principal magnetic axis (u) associated with the largest g-component (gZ) of each symmetry unique DyIII ion ground doubletFragmentgXgYgZ(u)x(u)y(u)zDy10.050.1119.55− 0.75030.37840.5422Dy20.070.1719.40− 0.14230.9042− 0.4026Dy30.200.5819.09− 0.8837− 0.0688− 0.4629Dy40.220.6818.850.12860.3877− 0.9128Dy50.000.0019.64− 0.5492− 0.7106− 0.4398Lower-case x, y, z denote the molecular Cartesian frame; upper-case X, Y, Z denote the magnetic axes frame. Ab initio computed g-values and the principal magnetic axis (u) associated with the largest g-component (gZ) of each symmetry unique DyIII ion ground doublet Lower-case x, y, z denote the molecular Cartesian frame; upper-case X, Y, Z denote the magnetic axes frame. Note that Eq. (2), while block-diagonal on the Ising σ configurations, would still give rise to 210 diagonalisation problems with dimension 610 ≃ 60.5 million, which is too computationally demanding. However, ab initio results show that FeIII-FeIII exchange is significantly weaker than DyIII-FeIII coupling. Thus, we adopted the perturbative strategy described in Methods section: Perturbative treatment of Fe-Fe exchange interaction, to account for the spin-frustrating contribution of FeIII-FeIII exchange, which keeps the problem non-diagonal only on the local spin basis of the FeIII quantum decoration, and solely dependent on the specific σ. Finally, the free energy F and the ensuing magnetic response properties are computed via a transfer matrix approach, as detailed in Methods section 3.5. Low energy spectrum Using Eq. (24), (25) we plotted the spectrum of the 210 DyIII Ising spin configurations σ (unflipped Fe spins). In Fig. 2a, we highlight classes of Ising excitations according to the number of angular nodes distorting the zero-noded vortex pattern (s-wave) of the Dy magnetic moments, analogously to angular spin-wave excitations49 (explicit representations in Fig. 2b). Note that most of the non-h-wave excitations are endowed with both toroidal and magnetic moments of variable sizes (see Supplementary Note 2). The modest ellipticity of the Fe10Dy10 ring (major/minor axis ratio ~ 1.22) may contribute to reducing the energy gap (~ 0.26 cm−1) between the toroidal ground state (s-wave) and the competing onion-like (p-wave) magnetic excitations, consistent with previous studies of vortex states in elliptical magnetic rings.50 In Fig. 2c, we also show the stacks of all Fe-spin single and double excitations. Low energy spectrum of Fe10Dy10. a Spectrum of the 210 DyIII Ising configurations for unflipped Fe spins, with configurations grouped by the number of angular nodes distorting the zero-node vortex (s-wave) pattern of the Dy magnetic moments. b Archetypal examples of s, p, d, f, g and h-wave Ising spin configurations analogous to the spin wave nomenclature. Colour coding corresponds to the band structure shown in (a). c Ground and excited state band structure of Fe10Dy10 from all possible one and two FeIII excitations around the ring for fixed DyIII Ising spin configurations. The Ising spin configurations are indexed by mapping the vector σ to a binary string and straightforwardly converting to a decimal representation offset by 1. d Calculated ground state FeIII on-site spin expectation values for the toroidal DyIII Ising spin configuration calculated with (red arrows) and without (grey arrows) FeIII-FeIII antiferromagnetic exchange. For the calculations involving FeIII-FeIII exchange, first-order perturbation theory corrections to the wavefunction were considered, taking corrections from 1-Fe exchange. Source data are provided as a Source Data file. Similarly to Eq. (25), we derived perturbative corrections to the on-site ground state spin expectation values of the FeIII ions, considering the next-nearest neighbour FeIII-FeIII exchange in first order perturbation theory, including only single excitations to correct the local Fe ground-state wavefunctions, a choice supported by Fig. 2c. In Fig. 2d, we plot the corrected average spin values of the FeIII ions (red) superimposed to their unperturbed values (grey). Unsurprisingly, the antiferromagnetic Fe-Fe exchange coupling directs the Fe spin moments towards a Néel configuration, which tends to diminish the ring toroidal moment45. Comparison with experiments Using a Quantum Design SQUID magnetometer MPMS-XL, we performed isothermal powder magnetisation and variable-temperature (1.8–300 K) magnetic susceptibility measurements on Fe10Dy10. Further magnetic data are provided in the Supplementary Information. From our transfer-matrix model, the magnetisation (Mα) and magnetic susceptibility (χαβ), together with their powder averages, are readily calculated from Eq. (29) via derivatives of the free energy with respect to the external magnetic field B: To account for thermal population of excited DyIII doublets when simulating χαβ, we employed χmT = (χxxT + χyyT + χzzT)/3 + (χT)ab initio − (χT)0, where the first term is the low-temperature powder average from Eq. (3), (χT)abinitio is the sum of single-ion ab initio χT contributions, and (χT)0 avoids double counting of the DyIII ground doublets and FeIII ground spin manifold40. In Fig. 3, we report excellent agreement between magnetic measurements and our ab initio-parameterised simulations. Notably, magnetic data seem to indicate a ground state with a large magnetic moment. However, our simulations show that a state with magnetic moment ~ 83μB lies ~ 0.26cm−1 above the magnetically compensated toroidal ground doublet, so that a field of ~ 6mT along the easy axis is sufficient to make the ground state magnetic. Towards higher fields, our simulations undershoot the experiment slightly, which could be a sign of magnetic torquing in the experiment not accounted for in our simulations. To demonstrate the spin-frustrating effect of FeIII-FeIII exchange coupling, in the inset of Fig. 3a, we present simulations of the T = 2 K powder magnetisation with (solid blue curve) and without (solid red curve) FeIII-FeIII coupling, unequivocally demonstrating that our perturbative inclusion of FeIII-FeIII exchange is crucial to reproduce the low field (∣B∣ ≤ 3 T) powder magnetisation. Magnetic measurements and simulated properties of Fe10Dy10. a Comparison between experimental measurements (orange circles) and simulated (solid blue line) isothermal powder magnetisation at T = 2 K. Inset: comparison between the experimental and simulated T = 2 K powder magnetisation when FeIII-FeIII next-nearest neighbour exchange JC is switched on/off (blue/red curves) in our model. b Comparison between experimental (orange circles) and simulated (solid blue line) molar magnetic susceptibility χmT as a function of temperature. Source data are provided as a Source Data file. In addition to magnetic measurements, we also collected specific heat data for Fe10Dy10 using a commercial PPMS 3He system from Quantum Design. Heat capacity was measured on pressed pellets of micro-crystals with approximate weight 1-2 mg by using the two-tau relaxation method. In Fig. 4, we present the data collected at a range of applied magnetic fields (coloured circles) as well as theoretical simulations using our transfer matrix model (solid lines). Specific heat of Fe10Dy10. Specific heat measurements (circles) on a Fe10Dy10 powder sample as a function of temperature for several values of applied magnetic field ∣B∣ = 0 T (blue), 1 T (yellow) and 3 T (green). Simulations (solid lines) using our transfer matrix mode averaged over the angular distribution. Source data are provided as a Source Data file. Once again our simulations, using the formula51 provide excellent agreement with the experiments. Importantly, we capture the shifting of the Schottky barrier to higher temperatures with the application of an external magnetic field. In addition to the specific heat from our transfer matrix model, we included the terms aT3 and bT3/2 in Eq. (4), accounting phenomenologically for lattice contributions to the specific heat. The first contribution is the well known Debye term, while the second accounts for anharmonic corrections which were necessary in order to reproduce the high temperature data51, and are here described by the only two fitting parameters a=1×10−3kB−1 K−3, and b=1×10−3kB−1 K−3/2. Finite-temperature molecular toroidal response While the magnetic measurements at T = 2 K indicate sizable low-temperature magnetic response of Fe10Dy10, they do not provide a characterisation of the zero-temperature ground state, nor do they exclude the existence of finite-temperature toroidal properties. In Fig. 5a, we simulate the static single-crystal magnetisation of Fe10Dy10 at T = 0.01 K, consistently showing a sharp rise in the magnetisation at a finite value of the field, suggesting a magnetically compensated ground state of Fe10Dy10. Evidence for a toroidal ground state. a T = 0.01 K single-crystal magnetisation simulations using our ab initio-parameterised transfer matrix model. The field is oriented along the long axis (blue curve), the short axis (red curve) and perpendicular to the Fe10Dy10 ring (yellow curve). b Low temperature zero-curl toroidal susceptibility simulations with (green) and without (blue) FeIII-FeIII exchange coupling included in the model. c Dependence on an applied uniform magnetic field of the toroidal susceptibility, calculated at various temperatures ranging from 1 K to 10 K. Source data are provided as a Source Data file. To probe the toroidal properties of the SMT at finite temperature, we define and evaluate, for this molecular setting, the toroidal susceptibility tensor ξαβ, describing the linear response of the system to a magnetic-field curl ∇ × B. Quantities related to finite-temperature toroidal response have previously appeared in Landau expansions of ordered states52, and in calculations of spin toroidization in periodic systems beyond the linear-response regime53. Building on these related formulations, here we treat ξαβ as an explicit finite-temperature molecular linear-response tensor54, directly analogous to the Van Vleck magnetic susceptibility, and specifically adapted to discrete, thermally populated toroidal spectra: Considering the free energy expressed in cm−1 and the magnetic-field curl in T a0−1 (a0 the Bohr radius), ξαβ is reported in units of cm−1a02 T−2. In direct analogy with the Van Vleck magnetic susceptibility, ξαβ can be expressed as an analytical derivative of the free energy54: involving matrix elements of the toroidal moment operator τα defined in Eq. (1). Here all toroidal matrix elements are evaluated with respect to the molecular inertia frame. In general, toroidal moments are strictly origin independent only for magnetically compensated states3,11,55. In this system, the maximally toroidal ground-state is thus rigorously origin independent, whereas finite-temperature corrections to ξαβ arising from thermally populated excited states can inherit a corresponding origin dependence. The molecular inertia frame nevertheless provides a useful reference frame both for analysing single-molecule toroidal structure-property correlations and for constructing a compact multipolar energy expansion of the linear response to locally constant magnetic-field curls in single-molecule settings, or to globally constant magnetic-field curls in molecular ensembles. Eq. (6) therefore describes the finite-temperature molecular toroidal response of an ensemble of non-interacting SMTs in the toroidal analogue of a conventional paramagnetic regime, rather than a spontaneously ordered ferrotoroidic phase. In zero magnetic-field curl the equilibrium ensemble remains globally unpolarised, while the induced linear toroidal polarisation generated by an external curl perturbation is We plotted the ξαβ component along the ring axis as function of temperature in Fig. 5b with and without FeIII-FeIII exchange. In the limit T → 0, ξααT=∣0τα0∣2/kB, hence the low-temperature value ξzzT ~ 2.5 × 106 cm−1 K a02 T−2 yields a ground-state toroidal moment ∣0τα0∣~1.3×103 a0μB. Using the peak magnetic-field curl estimated for the ultrafast near-IR driving protocol discussed in Section 1.7, (∇×B)z ~ 5 × 10−4 T nm−1~2.65×10−5Ta0−1, Eq. (7) predicts induced toroidal moments τα(T ) ~ 6.1 × 102, ~ 5.3 × 101, and ~ 1.5 × 101 a0μB at T = 0.1, 0.5, and 1 K, respectively, showing that even at 0.1 K the induced finite-temperature toroidal polarisation remains about 45% of the fully saturated ground-state toroidal moment, before progressively degrading at higher temperatures. We show in Fig. 5b that in both regimes of FeIII-FeIII exchange, a sizable ground state toroidal moment can be expected at zero temperature, although antiferromagnetic FeIII-FeIII exchange reduces its magnitude due to spin-frustrating effects in toroidal rings with N > 3 centres31. Also, we note that in the absence of FeIII-FeIII exchange coupling ξαβ has a slower decay with temperature, suggesting that FeIII-FeIII exchange lowers in energy magnetic states, becoming thermally populated at lower temperatures. However, as shown in Fig. 5c, we also find that for the same reason, applying a magnetic field will uniformly spread a large number of toroidal states across the spectrum, so that up to T ~ 10 K the magnetic field stabilises a constant toroidal polarisation that at finite temperatures is larger than what is achievable at zero field. Ultrafast-driven generation and magnetoelectric readout of toroidal polarisation While the equilibrium toroidal susceptibility ξ demonstrates that large finite-temperature toroidal moments are thermodynamically accessible, direct observation requires lifting the degeneracy between opposite toroidal chiralities. We show here that an ultrafast temporally asymmetric electric-field waveform can generate a cumulative toroidal population imbalance detectable via the magnetoelectric response of the polarised sample. In the absence of free currents, the magnetic-field curl associated with a temporally asymmetric electric-field waveform is governed by the Ampère–Maxwell relation56 Although ∂E(t)/∂t integrates to zero over each optical cycle, a temporally asymmetric waveform produces a strongly asymmetric distribution of ∂E/∂t. In the presence of finite relaxation times, this prevents complete cancellation of the toroidal response within each cycle, leading to cumulative toroidal polarisation. To clarify the physical origin of this mechanism, we first analyse an effective two-level open quantum system within a Lindblad framework (Methods in Non-equilibrium Toroidal Dynamics under Pulsed Electric Driving and Supplementary Note 7.1), where opposite toroidal chiralities ± τ are coupled to the driving field via Hdrive ∝ τ ⋅ (∇ × B(t)) and to a phonon bath characterised by a spin-phonon rate constant κ57. Treating dissipation perturbatively yields a closed-form expression for the toroidal population imbalance accumulated over one optical period Tl, which to leading order in κ becomes where ϵ controls the temporal asymmetry of the waveform, E0 is the electric field amplitude, and c is the speed of light. Although the expansion is formally carried out in powers of κ, the perturbative coefficients grow with irradiation time through repeated integration of the dissipative transition rates. Consequently, the linear accumulation regime described by Eq. (9) is transient and higher-order relaxation processes eventually become important. However, explicit numerical integration of the dissipative many-state dynamics under repeated pulse–wait sequences shows that, for realistic relaxation parameters, the linear regime remains dominant over experimentally relevant timescales. Equation (9) highlights three key features: (i) the accumulated toroidal polarisation initially grows linearly with time under repeated driving cycles, (ii) it depends critically on the asymmetry parameter ϵ, vanishing in the symmetric limit ϵ = 4, and (iii) within the linear regime it scales as ∣τ∣4, providing a strong nonlinear amplification mechanism favouring systems with large intrinsic toroidal moments such as Fe10Dy10 over smaller toroics such as Dy3. While this analytical model establishes the mechanism and scaling laws, quantitative predictions require the full low-energy spectrum and relaxation pathways of Fe10Dy10. We therefore perform numerical simulations based on the 144 low-energy ab initio-derived collective states, combined with a Lindblad master-equation approach (Methods in Non-equilibrium Toroidal Dynamics under Pulsed Electric Driving; Supplementary Note 7.2) accounting for multi-channel relaxation beyond the two-level approximation. A realistic implementation of the required temporally asymmetric waveform can be achieved using a superposition of harmonics of a near-infrared laser58. Specifically, we consider an ultrafast electric field constructed from a fundamental wavelength λ = 1600 nm and its first three harmonics (800, 533 and 400 nm), yielding the carrier-level waveform already providing an accurate approximation to the target sawtooth asymmetric waveform, as shown in Fig. 6a, b. Ultrafast-driven generation and magnetoelectric readout of toroidal polarisation. a Temporally asymmetric optical waveform used to drive toroidal dynamics, obtained from the superposition of the first four harmonics of a near-infrared laser field (λ = 1600 nm), compared with the ideal sawtooth profile. b Corresponding time-dependent magnetic-field curl (∇×B)z, generated through the Ampere--Maxwell relation using the k ≤ 4 waveform and E0 = 107 V m−1, showing strongly asymmetric positive and negative excursions within a single optical cycle. c Magnetoelectric-induced bulk magnetic moment Mxind as a function of the number of pulse--wait sequences for different waiting times twait. The induced moment is obtained by averaging the state-resolved magnetoelectric tensor over the full non-equilibrium time-dependent density matrix of the 144-state low-energy manifold generated by the ultrafast dissipative dynamics, and subsequently rescaled to the full irradiation volume containing 4.75 × 1013 molecules under a static readout electric field Ex = 109 V m−1. Dashed curves additionally report the contribution arising solely from the population imbalance Δρτ(t) of the maximally toroidal ground-state doublet ∣±τ, demonstrating that the magnetoelectric response is strongly dominated by the driven polarisation of the toroidal ground state. Horizontal dashed lines indicate the estimated microSQUID sensitivity threshold. For twait = 6.3 ns and twait = 1 ns, the threshold is reached after approximately 210 and 217 pulse--wait sequences, respectively, as indicated by the vertical dashed lines, corresponding to total irradiation times of only ~ 210 ns and ~ 1.37 μs, respectively. d Corresponding accumulated temperature increase ΔT as a function of the number of pulse--wait sequences for different waiting times twait. Despite faster thermal accumulation for shorter waiting times, the temperature increase remains modest throughout the experimentally relevant regime, reaching only ~ 5 mK and ~ 9 mK at the microSQUID sensitivity threshold for twait = 6.3 ns and 1 ns, respectively. Source data are provided as a Source Data file. We consider pulse envelopes of 107 fs, corresponding to Ncycles = 20 optical cycles. Using an electric field amplitude E0 = 107 V m−1 over a 20 μm spot size (pump fluence 1.42 μJ cm−2), the resulting magnetic-field curl varies asymmetrically within each cycle in the range Considering the measured low-temperature heat capacity reported in Fig. 4 (C/kB ≈ 3 for T < 1 K), a laser intensity I ≈ 1.33 × 1011 W m−2, and a laser spot area A = π × 10−10 m2, we estimate the laser-induced heating assuming a weakly absorbing near-transparent regime away from strong optical resonances. A Beer-Lambert estimate using a molar extinction coefficient ϵ ~ 1M−1cm−1 yields an absorbed fraction of only ~ 2.5% of the incident pulse energy (Supplementary Note 7.2). The resulting temperature increase is therefore only ΔTpulse ~ 5 × 10−2 mK per pulse. Heating scales linearly with absorbed fraction, making these estimates proportional to the assumed near-infrared extinction coefficient; even tenfold higher absorption would remain below 0.1 K over the detectability window. Although the thermal diffusion time is estimated to be tdiff ~ 630 ns, the present non-resonant weakly absorbing regime differs from conventional resonant pump–probe conditions, where repetition times are typically chosen much longer than tdiff to avoid cumulative heating. Here, however, longer waiting times would also favour relaxation of the accumulated nonequilibrium toroidal polarisation between pulses. The relevant operating regime therefore emerges from a competition between thermal accumulation and toroidal relaxation dynamics. Within the present thermal diffusion model59,60 (“Methods” in Non-equilibrium Toroidal Dynamics under Pulsed Electric Driving), pulse–wait sequences with twait = 6.3–1 ns (~ 160 MHz–1 GHz) already yield a nonequilibrium toroidal population imbalance associated with a detectable microSQUID magnetoelectric readout signal (Fig. 6c)61,62, while repeated-pulse heating remains bounded despite repetition times about two orders of magnitude shorter than tdiff. The twait = 6.3 ns regime provides a conservative operating point with modest thermal accumulation, whereas the 1 ns limit illustrates a more aggressive accumulation regime where the temperature increase nevertheless remains below the characteristic energy scale of the relevant magnetic excitations. Specifically, the accumulated temperature increase reaches only ~ 5 mK and ~ 9 mK at the microSQUID detectability threshold for twait = 6.3 ns and 1 ns, respectively (Fig. 6d), more than one order of magnitude below the thermal scale associated with the first low-lying magnetic (p-wave/onion-state) excitation at ~ 0.26 cm−1 (~ 370 mK). Within this realistic driving scheme, our simulations indicate a nonequilibrium toroidal accumulation per pulse per molecule of approximately 〈τz〉 ≳ 10−8a0μB (with a0 and μB the Bohr radius and Bohr magneton), arising from incomplete cancellation of the response within each asymmetric cycle. Explicit numerical propagation of the dissipative many-state dynamics further confirms sustained nonequilibrium accumulation over experimentally relevant pulse–wait sequences and relaxation regimes, with the shortest waiting times approaching an approximately linear growth regime over the timescales considered here (vide infra, Fig. 6c). To identify an experimental readout mechanism for the nonequilibrium toroidal polarisation, and hence the minimal ultrafast-driving duration required to accumulate a measurable signal, we numerically evaluate the magnetoelectric (ME) response3,29,30 associated with the evolving nonequilibrium population distribution. In the presence of toroidal polarisation, the system supports a linear coupling between an external electric field (Eβext) and an induced molecular magnetic moment (Mαind), where ααβME is the magnetoelectric tensor. Here we develop a tailor-made ab initio-informed computational protocol to evaluate the magnetoelectric response (see Methods in Magnetoelectric Coupling Tensor and Supplementary Note 6), inspired by previous work on toroidal systems29,30. In this framework, the magnetoelectric tensor associated with each collective Ising configuration of the Fe10Dy10 wheel included in the ultrafast-driven simulations is computed perturbatively as where ∣σ denotes one of the 144 low-energy collective Ising states of the ring, with 0≤Eσ ≲ 2.5 cm−1 relative to the maximally toroidal ground states ∣±τ, while the excited states ∣n are constructed from local ab initio crystal-field excitations of the DyIII ions on the corresponding Ising background. Since the local crystal-field excitation energies (≳ 100 cm−1) are much larger than the splittings within the 144-state low-energy Ising manifold (≲ 2.5 cm−1), the denominators (En − Eσ) are well approximated by the local crystal-field excitation energies (see Methods in Magnetoelectric Coupling Tensor and Supplementary Note 6). The resulting state-dependent tensors are then averaged over the nonequilibrium populations generated by the dissipative ultrafast dynamics. The resulting magnetoelectric tensor for the ground toroidal state ∣±τ of Fe10Dy10, expressed in the molecular inertia frame where z is approximately normal to the ring plane while x and y correspond to the long and short in-plane axes of the Dy ring, is (in atomic units, where 1a. u. of αME = 3.8894 × 10−12μB (V/m)−1). As detailed in the Methods section, the antisymmetric part of this tensor is dual to an induced magnetoelectric toroidal moment3, here estimated to be nearly aligned with the ground-doublet toroidal moment and of magnitude ∣τME∣ ≃ 5.13 a.u. Restricting first to the toroidal ground doublet ∣±τ, which has the largest magnetoelectric response, the most favourable readout channel is obtained by applying the static electric field along the long in-plane axis of the ring and detecting the induced magnetic moment along the same direction, governed by αxxME=28.2 a.u. To account more accurately for the nonequilibrium population distribution over the full 144-dimensional manifold explored by the ultrafast-driven dynamics, the induced bulk magnetic moment is evaluated as where ρNonEq(t) is the nonequilibrium density matrix obtained from the dissipative ultrafast dynamics, αxxME is the diagonal matrix whose entries are the state-dependent tensor components αxxME(σ) associated with the 144 low-energy collective Ising states σ, Nmol = 4.75 × 1013 is the number of molecules within the irradiation volume (Supplementary Note 7.2), and Exext=109 V m−1 is the static readout electric field. Because the induced magnetic signal scales linearly with the readout field, lower fields require proportionally more pulse–wait sequences to reach threshold. We regard the previously proposed Exext=109 V m−129,30 as a high-field limit relevant to micro- or nano-electrode, thin-sample, or pulsed-field configurations. Fig. 6c reports the calculated accumulation of the magnetoelectric readout signal, expressed as the bulk magnetic moment Mxind(t) induced by a static electric field Exext acting on the toroidal polarisation generated by the ultrafast pulse train. For the maximally toroidal ground-state doublet, the expression reduces to where Δρτ(t) is the time-dependent population imbalance between the + τ and − τ toroidal ground states accumulated over the pulse train, since the opposite toroidal branches satisfy αxxME(+τ)=−αxxME(−τ). The corresponding contribution to the accumulated magnetic moment is reported as dashed curves in Fig. 6c, demonstrating that the nonequilibrium magnetoelectric response remains strongly dominated by the driven population imbalance of the toroidal ground-state manifold Δρτ(t), thus providing direct evidence of nonequilibrium toroidal polarisation. From Fig. 6c it is evident that for both experimentally realistic repetition regimes considered here (twait = 6.3 ns and the more aggressive accumulation-driven 1 ns limit), the induced bulk magnetic moment exhibits sustained quasi-linear accumulation over at least 300 pulse–wait sequences, reaching values exceeding the estimated microSQUID detection threshold of ~ 104 μB61,62. The threshold is exceeded after approximately 210 − 217 pulse–wait sequences, corresponding to total irradiation times of only ~ 1.37 μs (twait = 6.3 ns) and ~ 210 ns (twait = 1 ns). Finally, the precise pulse counts and optimal waiting times should be regarded as relaxation-regime dependent, as they are controlled by plausible but phenomenological relaxation rates within a complex dissipative framework. Rather than attempting exhaustive optimisation of the dynamical model, our simulations show that experimentally detectable toroidal accumulation emerges within physically plausible dissipative regimes. Discussion This work shows that Fe10Dy10 operates in a regime where toroidal polarisation is thermodynamically accessible, dynamically accumulable, and experimentally detectable. The key ingredients are the large molecular toroidal moment, yielding strong coupling to magnetic-field curls, and the ultrafast-driven ratchet mechanism introduced here, whose accumulated toroidal polarisation scales as ∣τ∣4. Temporally asymmetric optical fields generate a net toroidal population imbalance through coherent driving and dissipative relaxation, yielding cumulative toroidal polarisation detectable through the magnetoelectric response. An alternative route would be direct magnetoelectric preparation29,30. In this case, the relevant energy scale is governed by the electric–magnetic-field coupling mediated by the magnetoelectric tensor. For an isotropic setup, Using the same fields considered previously for Dy330, E = 109 V m−1 and B = 1 T, the calculated Fe10Dy10 magnetoelectric tensor gives ΔE ≈ 3 × 10−2 cm−1, roughly one order of magnitude larger than in Dy330. This comparison is qualitative, since our estimate uses a more detailed ab initio-informed microscopic magnetoelectric tensor than the simplified Dy3 model of ref. 30. However, magnetic fields of order 1 T could destabilise the toroidal ground doublet of Fe10Dy10 by favouring low-lying magnetic states. In a regime preserving toroidal character (E = 109 V m−1 and B = 10 mT), the magnetoelectric splitting is of order 10−4 cm−1. Static magnetoelectric preparation, therefore, remains possible, but would require careful control of field geometry and competing magnetic states. By contrast, the ultrafast-driven protocol directly exploits the natural coupling τ ⋅ ∇ × B, avoiding competing field-induced magnetic states while enabling accumulation and detection of toroidal polarisation in a molecular system. More broadly, field-curl-driven control may provide a framework for manipulating toroidal degrees of freedom beyond molecular toroics, including skyrmionic magnetic textures where toroidal and magnetoelectric multipoles describe the spin structure7,63,64. Methods CAHF/CASCI-SO calculations We performed Configurationally Averaged Hartree Fock (CAHF)/Complete Active Space Configuration Interaction-Spin Orbit (CASCI-SO) calculations on the five symmetry unique [DyIIIFe2III(MeTeaH2−)(MeTea3−)2(CH3O−)2NO3−]2− (extended) fragments of the Fe10Dy10 molecular wheel using the in-house developed ab initio quantum chemistry code CERES46,65, a methodology which allows for the full representation of the spin-orbit coupling problem in the complete basis of the 2002 states arising from full set of 4f9 Russell-Saunders terms. To make the calculations tractable, we replaced the trivalent Fe ions with 3d10 trivalent Ga ions analogues. We employed the ANO-RCC-VTZP basis set for the Dy atom, ANO-RCC-VDZP for the eight O and N directly coordinating atoms and ANO-RCC-VDZ on all other atoms. A basis set convergence study using higher quality basis sets vindicated our initial finding of energetically well-isolated ground doublets for all DyIII ions in each fragment. Here below in Table 2 we report the results for the computed g-values and the Cartesian components of the principal magnetic axis (u) with the largest g-component (gZ) of each symmetry unique DyIII ion ground doublet. See Supplementary Note 1 and Supplementary Tables S1–S10 for further details about the crystal field spectra, magnetic properties and molecular fragments associated with each Dy(III) ion. As it can be seen from Table 2 and from Supplementary Tables S1–S5 in the Supplementary Information file, for all DyIII ions, the results of our calculations revealed energetically well-isolated ground KDs (with first excited KDs residing 60 cm−1 to 150 cm−1 higher in energy) which were comprised of almost pure mJ=±15/2 total angular momentum states arising from the 6H15/2 ground spin-orbit multiplet. These states displayed rather axial g-tensors with average values ḡX<ḡY=0.3 and ḡZ=19.3 (see Table 2 in the Methods section). The ab initio principal magnetic Z-axes ui for each DyIII are given in Table 2 and are shown as orange arrows centred on each DyIII ion in Fig. 1. Broken symmetry density functional theory calculations To determine the sign and magnitude of the fifteen symmetry-unique intramolecular exchange coupling constants between nearest- and next-nearest-neighbour paramagnetic ions, we employed broken-symmetry density functional theory (DFT) calculations66 on isostructural analogues of the FeiIII-DyiIII-Fei+1III fragments of the molecular wheel with appropriate metal substitutions (vide infra). Calculations were performed with ORCA 5.067 using the TPSSh functional68 and SARC-DKH-TZVP basis sets69 for all atoms. For the JA coupling constants, we converged high-spin and broken-symmetry calculations for each of the five symmetry-unique Gai−1III-GdiIII-FeiIII fragments, where Fei−1III was replaced by diamagnetic GaIII and DyiIII by orbitally non-degenerate GdiIII. The resulting couplings were scaled using SDy/SGd = 5/7 and extracted via Yamaguchi’s formula48. The same strategy was employed for JB and JC using the fragments Fei−1III-GdiIII-GaiIII and Fei−1III-LaiIII-FeiIII, respectively. The interaction Hamiltonian Within a given Ising configuration σ, the exchange Hamiltonian is where JA and JB are nearest-neighbour FeIII-DyIII exchange couplings and JC is the next-nearest-neighbour FeIII-FeIII coupling. The DyIII Ising spins are SiDy=52σiui, thus fixed by the Ising configuration σ (where ui the ab initio magnetic axis of the ith ion as reported in Table 2), whereas SiFe acts on the local SFe = 5/2 true quantum spin manifold of each FeIII ion. Intramolecular dipolar interactions are described by where Rij = ri − rj and MiFe=gμBSiFe with g = 2, while MiDy=12μBgzziσiui. Only dipolar interactions within each Dy-Fe-Dy fragment are retained. Magnetic dipole-dipole coupling between FeIII next-nearest neighbours is approximately μ0(μBgSFe)2/4πR3~0.05 cm−1 and thus can be safely neglected as can dipolar couplings to further removed FeIII ions in the ring. The Zeeman and toroidal interactions are and The toroidal coupling provides direct access to thermodynamic toroidal response properties via free-energy derivatives, as discussed in finite-temperature molecular toroidal response. Perturbative treatment of Fe-Fe exchange interaction Since our broken symmetry DFT calculations indicated systematically a smaller value of JC with respect to FeIII-DyIII exchange and magnetic dipole coupling (see Table 1), we further treat the antiferromagnetic FeIII-FeIII exchange as a first order correction to the zeroth order energies Eσλ(σ)(0) obtained from the following zeroth order block-diagonal Hamitonian (H0(σ)), in which the problem is partitioned into a sum of local FeIII Hamiltonians as: where each hi(σi, σi+1) details the exchange, magnetic dipole and external field interactions of the FeIII ion at site i. Since hi(σi,σi+1),hj(σj,σj+1)=0, each local FeIII Hamiltonian can be diagonalised on the local six dimensional mi basis of each FeIII site to give where the λi’s index the six energy eigenstates at each site. Naturally, the ring eigenstates for a given dysprosium configuration σ and FeIII excitation pattern λ(σ) are product states of the local FeIII eigenstates at each site λ(σ)=λ1σ1,σ2⊗⋯⊗λ10σ10,σ1 with total energy hence the perturbative correction due to antiferromagnetic Fe-Fe exchange coupling is evaluated as: where the expectation values are simply the diagonal matrix elements ⟨SiFe⟩λi=λiσi,σi+1SiFeλiσi,σi+1. The free energy and response properties for the whole system are finally computed, making use of the transfer matrix approach as detailed in the next subsection Free energy calculation. The effect of spin frustration obtained by this perturbative approach on the ring’s spectrum and states is discussed in the “Methods” subsection Low energy spectrum, and also in the main text. Free energy calculation To compute thermodynamic quantities (e.g., magnetisation, magnetic susceptibility) from our model requires knowledge of the partition function Z which, as has already been noted, is the sum of 210 × 610 terms. Rather than carrying out this sum explicitly, we show illustrate how the Fe10Dy10 partition function can be rewritten as the trace over a product of ten 24 × 24 dimensional transfer matrices70 allowing for the efficient calculation of the ring free energy and, subsequently, thermodynamic quantities of interest. The partition function for Fe10Dy10 is the sum of the exponentiated energies defined in Eq. (24) and Eq. (25) for each possible Ising configuration of the DyIII magnetic axes σ and FeIII excitation pattern λ(σ) where in the second line we have collected energies corresponding to each FeiIII-Fei+1III ‘bond’ in the term Defining the matrices Ai,i+1 with matrix elements where δij is the Kronecker delta, we identify the sum of products on the last line of Eq. (26) with the trace of a product of matrices A1,2…A10,1. Clearly, from the definition of the compound indices λiσiσi+1, the Ai,i+1 matrices are just (6 × 2 × 2)2 = 24 × 24 dimensional. Thus the Fe10Dy10 free energy (F) can be expressed as which is certainly computationally more palatable than a sum over 210 × 610 ≃ 62 billion states. To calculate thermodynamic quantities of interest, we take the appropriate derivatives of Eq. (29) as discussed in comparison with experiments. Non-equilibrium toroidal dynamics under pulsed electric driving To rationalise the dynamical accumulation of toroidal polarisation, we first consider an effective two-level model formed by two time- and parity-reversed toroidal states carrying opposite toroidal moments ± τ. The system is driven by the magnetic-field curl generated by a temporally asymmetric electric waveform, which couples to the two-level system through Dissipative population transfer between the two toroidal branches is described by a Lindblad equation, where L[O]=OρO†−12O†Oρ−12ρO†O, and the rates g(t) and h(t) obey detailed balance with respect to the instantaneous toroidal splitting Expanding the density matrix perturbatively in the dissipative rate, ρ(t) = ∑rκrρ(r)(t), shows that coherent driving alone only generates phases, while the combination of dissipation and temporal asymmetry produces a population imbalance. For the asymmetric sawtooth waveform defined in Supplementary Note 7, the leading contribution after one optical period Tl is where E0 is the electric-field amplitude and ϵ controls the temporal asymmetry. This expression shows that accumulation vanishes for the symmetric waveform, ϵ = 4, and scales as ∣τ∣4, motivating large molecular rings with large intrinsic toroidal moments. For Fe10Dy10, we then replace the two-level model by a reduced low-energy Ising basis. The full Hilbert space is made tractable by freezing the FeIII spin states in their zero-curl ground configurations and retaining only DyIII Ising configurations with energies ≤ 2.5 cm−1, appropriate for low-temperature and weak-driving dynamics. The populations ρm of the retained configurations obey the time-dependent master equation with phonon-assisted transitions and tunnelling rates Here δmkr−Dy=1 if configurations m and k differ by r Dy moment flips and zero otherwise, while Ek(t) is the instantaneous energy of configuration k in the applied curling field. For our simulations, we adopt the leading-order spin-phonon coupling constant Γ1 = 102 cm3 s−171,72, and the quantum tunnelling parameters γ1 = 1016 s−2 and ξ = 1010 s−1, corresponding to an on-resonance tunnelling rate γ1/ξ ~ 106 s−173, compatible with tunnelling rates reported for other dysprosium single-molecule magnets40,42. We further imposed a reasonable hierarchy on the higher-order multi-flip spin-phonon rate constants, Γ2 = 10−2Γ1 and Γ3 = 10−4Γ1. A similar hierarchy was adopted for the multi-site quantum tunnelling rate constants, γ2 = 10−1γ1 and γ3 = 10−2γ1, which conservatively represent the fastest identified two-site and three-site tunnelling processes in Fe10Dy10. The relative efficiency of such processes was estimated from products of the transverse ab initio g-tensor components reported in Supplementary Tables S1–S5, whose strong variability reflects the broad distribution of local transverse anisotropies and tunnelling efficiencies in this system. The driving field is generated from a truncated Fourier representation of a temporally asymmetric sawtooth waveform, implemented using the fundamental wavelength λ = 1600 nm and its first harmonics, as described in Supplementary Note 7. The toroidal polarisation at time t is then evaluated as The experimentally detectable signal is obtained by combining the nonequilibrium populations generated by the ultrafast dynamics with the state-dependent magnetoelectric tensors derived above. The nonequilibrium populations ρm(t) are assembled into the diagonal density matrix ρNonEq(t), while the corresponding state-dependent magnetoelectric tensor components αxxME(σ) for the selected 144 Ising states σ define the diagonal matrix αxxME. The induced bulk magnetic moment is then evaluated as where Exext is the static readout electric field applied along the long in-plane axis of the ring. For a laser volume containing the induced magnetic moment becomes experimentally detectable within the accumulation regime discussed in the main text. To evaluate the thermal load induced by the non-resonant laser pulse train, we adopt the standard linear superposition framework for multi-pulse heat accumulation60,74. For an initial Gaussian temperature profile of characteristic radius R, the solution of the three-dimensional heat diffusion equation gives a peak temperature excess at the centre of the laser spot (r = 0) that decays as59 where ΔT0 is the instantaneous temperature increase induced by a single pulse and tdiff is the characteristic thermal diffusion time. For a sequence of Npulses pulses separated by a waiting time twait, the accumulated temperature increase is obtained by linear superposition of the residual contributions from all previous pulses, In the experimentally relevant regime considered here (twait ≪ tdiff), the sum is well approximated by the continuous-limit expression The parameters ΔT0 and tdiff are obtained from the absorbed pulse energy, measured low-temperature heat capacity and estimated thermal diffusivity, as detailed in Supplementary Note 7. Full details of the perturbative two-level derivation, rate parametrisation, pulse shaping, numerical integration, magnetoelectric readout and heating estimates are given in Supplementary Note 7. Magnetoelectric coupling tensor Here, we derive the magnetoelectric (ME) response of a polynuclear rare-earth complex within perturbation theory as the leading mixed derivative of the energy with respect to electric and magnetic fields. In toroidal systems, this response is linear and corresponds to the lowest-order spin-electric coupling3,29. Owing to the weakness of the 4f–4f exchange, the response is dominated by single-ion contributions30. Within the single-ion mechanism, the electric field couples the local 4f states of each Ln(III) ion to the 5d manifold of the same ion through the odd-harmonic components of the effective crystal field, in close analogy with Judd-Ofelt theory for optical transitions in lanthanide systems75. Building on ref. 29, we retain the same second-order perturbative structure and constant 4f–5d energy-denominator approximation, but restrict the 4f states to the Hund’s-rule ground Russell–Saunders term and its spin-orbit multiplet. Within this manifold, crystal-field matrix elements are evaluated directly via the Wigner–Eckart theorem, eliminating the need for coefficients of fractional parentage while remaining exact for the quantities of interest. Moreover, we generalise previous symmetry-restricted models29,30 by explicitly constructing the internal electrostatic potential from effective point charges placed on the experimental Fe10Dy10 structure, in the spirit of the Magellan electrostatics approach76. We further retain the full hierarchy of odd-harmonic crystal-field contributions (k = 1, 3, 5), corresponding to the coupling of the electric field and its spatial derivatives to the dipole (k = 1), octupole (k = 3) and dotriacontapole (k = 5) moments of the 4f–5d overlap distribution, thereby extending previous symmetry- and dipole-restricted (k = 1) treatments29,30. Finally, the resulting local electric and magnetic Hamiltonians are projected onto the ab initio decomposition of a representative Dy(III) local wavefunction (Supplementary Note 6), assumed transferable across the ten Dy sites consistently with the approximate level of theory adopted. The resulting local response is then projected onto the collective toroidal states of the molecule in the presence of an external magnetic field. Effective electric Hamiltonian within the 4f manifold. We start from the electric perturbations acting on the 4fN shell of a single lanthanide ion29,75, where where ab† (ac) are the creation (annihilation) operators varying the occupation number of the associated one-electron spin-orbital basis {ϕb}, while the one-electron matrix elements vbc(1) read: and, likewise, the matrix elements vbc(2) read: where E−qext are the spherical components of the external electric field, Cq(k)(r) are Wybourne irreducible tensor operators built from an open-shell electron position r, and AkqCF=e∑n(Qn/Rnk+1)Cq(k)*(Rn) are the irreducible tensor components of the crystal field potential generated by point charges eQn placed at position Rn. Both operators V^i have vanishing matrix elements within the 4fN space, and contribute only via virtual excitations to states from opposite-parity configurations. Using quasi-degenerate perturbation theory and restricting the opposite-parity intermediate space to 4fN−15d1 (see Supplementary Note 6 for details), after explicit elimination of the intermediate space, we arrive at the following effective Hamiltonian: where P^0 is a projector onto the 4f N space (see Supplementary Note 6), Wfd is an average 4f–5d excitation energy29,30,75, and: Further manipulation of Eq. (49) yields an explicit one-electron effective Hamiltonian acting solely within the 4f shell, with E^m1m2=am1,α†am2,α+am1,β†am2,β the second-quantisation singlet excitation operators, and with coupling matrix elements: where is the effective coupling constant proportional to the product of the radial transition dipole 〈r〉fd and the radial transition multipoles ⟨rg⟩fd of the crystal field harmonics of rank g = 1, 3, 5, and with an analogous definition for Oγμg,1(m1,m2). The three radial coupling constants Eq. (52) were approximated here with the radial integrals of second-order Judd–Ofelt for Dy(III) tabulated by Wybourne and Smentek in Table 21-9 at page 294 of ref. 75, which already account for the excitation energy gap. The three values used here, in atomic units (a.u.), are thus Gfd(1)=−0.2949 a.u., Gfd(3)=−1.0005 a.u., and Gfd(5)=−4.8183 a.u. Using standard angular momentum recoupling arguments (see Supplementary Note 6), we further find that where where 〈2∥C(g)∥3〉 are standard reduced matrix elements of spherical tensor operators between f and d single-electron orbital angular momentum eigenstates77. Substituting Eq. (54) and Eq. (51) into Eq. (50), and introducing the coupled tensor operators we obtain (see Supplementary Note 6) the effective Hamiltonian of the Ith Dy(III) ion: where and Note that the superscript (12) indicates that these coefficients arise from the symmetrised second-order perturbative coupling V1V2 + V2V1 between the external electric field (V1) and the crystal-field potential (V2). Explicitly, BKQ(12)(I) and bQμK,(12)(I) are obtained from the contribution of Oμγ1,g(m1,m2) in Eq. (54) (i.e., the V1V2 term), while the complementary term Oγμg,1(m1,m2) (i.e., the V2V1 term) yields an identical contribution, resulting in the overall factor of two in Eq. (56). Further defining we can explicitly express the linear dependence of Eq. (56) on the three Cartesian components of the external electric field Eαext: with the on-site effective electric dipole operator for the Ith Dy(III) ion: Supplementary Note 6 further shows that the matrix elements of Eq. (63) on the ∣J,MJ basis of each Dy(III) centre, hence for the ab initio crystal field states of interest here, can be straightforwardly evaluated without recurring to coefficients of fractional parentage, in as far as the evaluation is restricted to a ground multiplet 6H15/2 originating from the single Hund’s rule 6H Russell-Saunders term. Local and global magnetoelectric tensors. The local magnetoelectric tensor for the Ith Dy(III) ion is obtained by combining the electric Hamiltonian in Eq. (62) with the magnetic Hamiltonian where M^(I)=−μBgJJ^(I), and applying second-order perturbation theory. This yields where h.c. denotes the Hermitian conjugate contribution with P^α(I) and M^β(I) exchanged. The local tensor is evaluated in a reference frame aligned with the ab initio magnetic axes reported in Supplementary Tables S1–S5 of Supplementary Note 1. The local ground Kramers doublet ∣KD0 and excited doublets ∣KDn are expressed as linear combinations of the local ∣J,MJ basis, with coefficients taken from the ab initio wavefunctions of a representative Dy(III) ion and transferred to all sites, consistently with the approximate level of theory adopted. The corresponding wavefunctions are reported in Supplementary Note 6. The local tensors ααβME(I) defined above describe the magnetoelectric response associated with the local ground Kramers doublet of the Ith Dy(III) ion. In order to construct the collective magnetoelectric response entering the non-equilibrium toroidal dynamics simulations, we approximate the low-energy collective states of Fe10Dy10 as Ising-like direct products of local ab initio ground Kramers-doublet states. For a given collective Ising configuration σ = (σ1, σ2, …, σ10), 7D1σI = ± , the corresponding low-energy collective state is approximated as where ∣±(KD0) denote the two partners of the local ground Kramers doublet on the Ith Dy(III) ion. The excited states entering the perturbative magnetoelectric expression Eq. (13) are correspondingly approximated as local crystal-field excitations on top of the same Ising background, where KDe denotes the eth excited Kramers doublet of the Ith Dy(III) ion. Within this approximation, the collective-state-dependent magnetoelectric tensor entering the ultrafast dynamical simulations is written as Since the local crystal-field excitation energies (≳ 100 cm−1) are much larger than the splittings within the low-energy collective Ising manifold (≲ 2.5 cm−1), the denominators are well approximated by the corresponding local crystal-field excitation energies obtained from ab initio calculations. Under these assumptions, the collective tensor reduces to a sum of rotated local contributions weighted by the Ising orientations, where σI = ± 1 labels the orientation of the local ground Kramers doublet on site I, and the rotation matrices RI transform from the local magnetic reference frames to the global molecular frame, and are constructed from the ab initio principal magnetic axes of the Ith Dy(III) ion expressed in the common reference frame of the Fe10Dy10 structure (Supplementary Tables S1–S5 and Supplementary Note 1). The magnetoelectric tensor is, in general, non-symmetric with respect to the exchange of its Cartesian indices α and β. It can therefore be decomposed into a scalar (trace), antisymmetric, and traceless symmetric part: The antisymmetric component defines an induced magnetoelectric toroidal moment, Further implementation details, including projector algebra, tensor recoupling, and explicit evaluation of matrix elements, are provided in Supplementary Note 6. Supplementary information Supplementary Information file Transparent Peer Review file Source data Source Data Acknowledgements The authors thank Jürgen Schnack for valuable discussions. Author contributions A.S. and K.H. developed and implemented the theoretical models and the proposed protocol to generate and observe toroidal polarisation in molecular systems. S.C. performed the DFT calculations. J.B., Y.F.S., A.B., C.E.A. and A.K.P. performed the synthesis and structural characterisation of the Fe10Dy10 ring. Y.L. performed the SQUID magnetometry measurements, and W.W. performed MicroSQUID experiments. M.A. performed the specific heat experiments. A.S. and K.H. wrote the manuscript with help from J.B., C.E.A., A.K.P. and M.A. All authors discussed the results, analysed the data, and contributed to the manuscript. Peer review Peer review information Nature Communications thanks Xiao-Lei Li, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Funding A.S. acknowledges funding via the grant P-DiSC BIRD2023-UNIPD from the Department of Chemical Sciences of the University of Padova, from the University of Padova and Monash University Joint Initiative in Research (2024 Seed Fund scheme), and from the CINECA award of HPC resources and support under the ISCRA initiative, Project Grant SMTQUANT. J.B., Y.F.S., C.E.A., W.W. and A.K.P. acknowledge funding by the German Research Council (DFG) through the CRC 1573 “4f for Future”. Additional funding was provided by the Helmholtz Association through POF MSE. Data availability The data supporting the findings of this study are available within the paper and its Supplementary Information. Source data for all plots appearing in the manuscript and Supplementary Information are provided with this paper as a Source Data file. Source data are provided in this paper. Code availability The custom in-house codes used for the transfer-matrix, nonequilibrium-dynamics and magnetoelectric-response calculations are under active development and are not currently maintained as a public software release. Additional implementation details, representative scripts and input files are available from the corresponding author, A.S., upon request for academic reproducibility and clarification. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Alessandro Soncini, Email: alessandro.soncini@unipd.it. Annie K. Powell, Email: annie.powell@kit.edu Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-75612-6. References Associated Data Supplementary Materials Supplementary Information file Transparent Peer Review file Source Data Data Availability Statement The data supporting the findings of this study are available within the paper and its Supplementary Information. Source data for all plots appearing in the manuscript and Supplementary Information are provided with this paper as a Source Data file. Source data are provided in this paper. The custom in-house codes used for the transfer-matrix, nonequilibrium-dynamics and magnetoelectric-response calculations are under active development and are not currently maintained as a public software release. Additional implementation details, representative scripts and input files are available from the corresponding author, A.S., upon request for academic reproducibility and clarification.