Activating plasmonic catalysis through light-mediated steady-state spin modulation

Abstract

Light-mediated electronic spin modulation possesses intriguing potentials for photochemistry, enabling on-demand customization of catalysts towards distinct reactions and catalytic requirements. However, the significant photobleaching of the transient spin transitions as well as their temporal mismatch with slower chemical reaction dynamics substantially hinders its applicability. Herein, we demonstrate light-driven steady-state and on-demand catalyst spin modulation that effectively activates plasmonic catalysis. The rapidly oscillating plasmonic electromagnetic near-field spin-polarizes a low-spin CoFe₂O₄ catalyst and overcomes the photobleaching to produce stable high-spin states with astounding spin lifetimes >60 μs. The high-spin plasmonic catalyst effectively balances the tradeoff between spin polarization and carrier dynamics. For benchmark light-driven nitrate reduction catalysis, it achieves substantial photo-enhancement in ammonia production rate and selectivity as well as photocatalytic performance driven by sunlight, benefiting from polarization activation of nitrate reactant and preferential reaction pathway modulation. The highly generalized light-mediated strategy opens intriguing new avenues for on-demand and steady-state electronic spin engineering with profound implications for distinct disciplines.

Introduction

Light-mediated electronic spin modulation represents an emerging strategy to engineer the optical and electric responses of nanomaterials in the presence of external electromagnetic (EM) stimuli^1,2,3. It has, thus, demonstrated intriguing potentials in distinct disciplines, including quantum communications⁴, optoelectronic devices⁵, paramagnetic sensors⁶, and catalysis^7,8. In particular, nanocatalysts could benefit from effective modulation of their spin states, factoring in facilitated charge transfer dynamics^9,10, modulated electronic d band center^11,12, and/or activated surface reactive sites^13,14. Compared to conventional spin modulation approaches based on crystal structure engineering strategies or external (electro)magnetic fields^8,11,15, light modulation is distincted by its in-situ, on-demand and non-invasive nature without jeopardizing additional intrinsic properties of the catalysts, such as their charge carrier mobility^16,17, which is exceptionally valuable for light-driven catalytic applications. However, the temporal mismatch between transient spin transitions and slower chemical reaction dynamics significantly hinders its applicability, as the rapid, sub-nanosecond relaxation of the photoexcited high-spin state(s) back to steady-state low-spin counterpart^16,18 substantially limits the availability of the former for key chemical conversion steps, including adsorption and diffusion (micro-to-milliseconds). This issue is unlikely to be alleviated even with constant illumination, as the limited absorption cross-section of the catalysts and significant photobleaching effects of spin transitions prevent their continuous photoexcitation.

Localized surface plasmon resonance (LSPR) could provide a plausible solution to the above-mentioned challenge. It is signified by the collective charge oscillation upon resonant photoexcitation^19,20, creating rapidly alternating EM near-fields at pico-to-femtohertz frequencies^21,22, presenting intriguing possibilities for spin polarization. In this work, we achieve light-driven and surface plasmon-mediated steady-state electronic spin modulation that stabilizes the otherwise transient high-spin states in situ to effectively activate plasmonic catalytic performance. The photoexcited and rapidly oscillating plasmonic EM near-field spin-polarizes low-spin CoFe₂O₄ and effectively overcomes the photobleaching in its spin transition, producing stable high-spin form with astounding lifetimes as long as ~60 μs. Nitrate reduction reaction is used as a benchmark for the light-driven catalytic performance. Benefiting from the polarization activation of nitrate reactant, preferential modulation of ammonia production pathways, and facilitated ammonia desorption, the high-spin plasmonic catalyst balances the tradeoff between spin polarization and carrier dynamics, and achieves considerable photo-enhancement in ammonia production rate and selectivity as well as photocatalytic nitrate reduction performance driven solely by light. The highly generalized light-mediated strategy opens intriguing new avenues for on-demand and steady-state electronic spin engineering with profound implications for light-driven catalysis and beyond.

Results

Design principle of the surface plasmon-mediated spin modulation strategy

The key to the light-induced and surface plasmon-mediated catalyst spin modulation lies in the exploitation of the resonant plasmonic EM near-field to induce a stable magnetic dipole for the duration of plasmonic excitation, facilitating strong dipole interactions towards activated light-driven catalysis (Fig. 1a). The increased optical cross-sections and the rapidly oscillating field directionality is able to conserve the spin-multiplicity after initial polarization, enforcing spin-selectivity in the charge relaxation process, and thus overcoming the significant down-transition and photobleaching to achieve steady-state spin modulation concurrent to the plasmonic near-field. As depicted in our finite-difference time domain (FDTD) simulations, the plasmonic near field contains intense electric as well as magnetic components perpendicular to each other (Fig. 1a, insets). The role of electric field component in induced dipole interactions has been previously demonstrated for resonant plasmonic energy transfers in the form of “forced” plasmons^22,23. On this basis, it is hypothesized that the synergy between the oscillating near-field magnetic and electric components could respectively induce stable spin polarization and carrier dynamics modulation.

Fig. 1: Design principle of surface plasmon-mediated spin modulation.

Schematic illustrations of the spin modulation strategy (a) and the Au-CoFe₂O₄@SiO₂ spin-active plasmonic catalyst (b). Red spheres: Oxygen; Brown: Fe; Blue: Co; Yellow arrow: surface Fe unit cell; Blue arrow: bulk Fe unit cell; green: Co unit cell. (c) TEM and HRTEM micrographs of Au-CoFe₂O₄@SiO₂. Raman (d) and single-particle dark-field scattering (e) spectra of Au-CoFe₂O₄@SiO₂ and controls. Au-CoFe₂O₄@SiO₂ Fe 2p (f) and Co 2p (g) ISI-XPS spectra with light or in dark.

As the resonant plasmonic near-field is highly intense at nanoplasmonic interfaces but decays exponentially with distance, it is essential to localize the spin-active component within the direct vicinity of the plasmonic nanoantenna. To this end, a prototypical plasmonic antenna-reactor (A-R) catalyst architecture is constructed using gold nanoantenna and cobalt ferrite (CoFe₂O₄) catalyst (Fig. 1a, see also Fig. S1). As elaborated in the next section, the CoFe₂O₄ nanocrystals contains predominantly (~72%) normal spinel structure, which possesses Fe(III) species (3d⁵) within octahedral interstices (Fig. 1b). In accordance with the crystal field theory^11,15, the octahedral crystal field could then split the degenerate d orbitals into lower-energy t_2g and higher-energy e_g bands with different filling profiles at high-spin (HS) (e_g² t_2g³) or low-spin (LS) (e_g⁰ t_2g⁵), rendering the Fe(III)_oct highly spin-“active”. For surface ·Fe species, the absence of an axial lattice oxygen leads to a square pyramidal crystal field, producing additional elevations in the orbital energies of d_xy and d_x²−y² orbitals. This is corroborated by our density-functional theory (DFT) calculations of the electron configurations and orbital energies (Supplementary Data 1 and Fig. S2). On the other hand, the tetrahedral Co(II) (3d⁷) would exhibit identical electron configuration of e_g⁴t_2g³ and is, thus, spin-“inactive”.

The CoFe₂O₄ nanocrystals are encapsulated within a silica nanoshell to insulate possible plasmonic hot carrier or heat transfer processes for close examination of the plasmon-mediated spin modulation. The nanoshell is highly permeable by the near field as well as molecular reactants in later catalytic measurements. CoFe₂O₄@SiO₂ was tethered to Au nanoantenna via a thiolated silane linker to form the spin-active plasmonic nanocatalyst termed Au-CoFe₂O₄@SiO₂ (Fig. 1a, see also Fig. S1). The distance between the antenna and reactor components were controlled to be 3–5 nm for optimal near field-enhancement²⁴. Transmission electron microscopy (TEM) images (Fig. 1c) and element mapping (Fig. S3) indicate that the CoFe₂O₄ nanocrystals (10.1 ± 0.9 nm in diameter) are evenly dispersed around the Au nanoantenna (60.1 ± 2.9 nm in diameter). High-Resolution TEM (HRTEM) image exhibits lattice spacings of 0.252 and 0.219 nm in CoFe₂O₄ as well as 0.236 nm in Au NPs (Fig. 1c), respectively corresponding to the CoFe₂O₄ (311), (400) and Au (111) crystal facets, consistent with the dominant peaks from X-ray diffraction (XRD) spectra (Fig. S4). Raman spectra of Au-CoFe₂O₄@SiO₂ display clear signatures for CoFe₂O₄ (A_1g at 627 and 684 cm⁻¹, T_2g at 442 cm⁻¹, and E_g at around 300 cm⁻¹)²⁵ as well as for Au-linker²⁶ (Fig. 1d), further corroborating its successful formation.

The resonant constructive enhancement between Au plasmon resonance and CoFe₂O₄ interband transitions²⁴ leads to a prominent increase in the visible light absorbance of Au-CoFe₂O₄@SiO₂ (Fig. S5). Notably, the enhancement in absorbance exceeds a physical mixture control containing identical concentration of Au and CoFe₂O₄ colloids, which signifies the importance of localizing the spin-active component within the plasmonic near-field for efficient resonant enhancement. As a consequence of the strong resonance, significant energy transfer from the Au nanoantenna towards CoFe₂O₄ could be mediated by the resonant plasmonic near-field^21,27. Single-particle dark-field hyperspectral imaging and scattering spectra depict an obvious red-shift as well as an increase in FWHM for Au-CoFe₂O₄@SiO₂ in comparison to Au nanoparticle (Figs. 1e and S6), which indicates the dampening of the energies associated with the excited plasmons and energy transfer towards the CoFe₂O₄ nanocrystals owing to an increase in dielectric function in the surrounding environment of Au antenna in the presence of CoFe₂O₄²⁸.

To further characterize the oxidation state and photo-response of Au-CoFe₂O₄@SiO₂, in-situ irradiated X-ray photoelectron spectroscopy (ISI-XPS) was next performed. Fe and Co 2p XPS spectra respectively exhibit predominant feature peaks for Fe³⁺ and Co²⁺ (Fig. 1f, g), consistent with the prevalence of normal spinel structure and a chemical stoichiometry of around 2:1 (Table S1). When illuminated with Xe lamp, the Au 4f_5/2 and 4f_7/2 peaks for both Au-CoFe₂O₄@SiO₂ and a Au NPs-only control both depict a uniform shift towards lower binding energies (BE) (Fig. S7), consistent with the resonant excitation of the localized surface plasmons^29,30. Notably, the Fe³⁺ 2p_3/2 and 2p_1/2 features^31,32 of Au-CoFe₂O₄@SiO₂ depict apparent shifts towards lower BE (Δ_shift = 0.3 eV) when illuminated (Fig. 1f), which is not observed for the Co species (Fig. 1g) or with the CoFe₂O₄-only control (Fig. S8). These observations are highly consistent with the spin polarization of the spin-active Fe species, but not the spin-inactive Co in Au-CoFe₂O₄@SiO₂, from LS to HS, leading to lone-pair HS electrons having lower apparent binding energies³³.

Evidence for the surface plasmon-mediated spin modulation

At room temperature and when not exposed to external stimuli such as EM field or illumination, the Fe species preferentially occur in its LS state in the CoFe₂O₄ nanocrystal^9,34. To corroborate this phenomenon, the magnetic susceptibility (χ) of CoFe₂O₄ was first measured as a function of temperature (Fig. S9), where very low χ on the order of 10⁻² is obtained, consistent with its superparamagnetic nature³⁵. Compositing the nanocrystal with Au antenna further decreased the overall χ to 3.3 × 10⁻³ (Fig. S9). As expected, both CoFe₂O₄ and Au-CoFe₂O₄@SiO₂ exhibit no evident hysteresis behavior (Fig. S10). The spin state of the nanocrystal was next analyzed based on the Curie-Weiss law^36,37, taking into consideration the magnetic moments of both Co and Fe cations. A Curie–Weiss constant (C) of 0.821 was fitted from the magnetic susceptibility profile (Fig. S9b), which corresponds to a HS/LS ratio of around 0.08 (~7.4% HS + 92.6% LS), confirming the predominance of LS state without external stimuli.

To corroborate the light-induced and surface plasmon-mediated spin modulation, operando X-ray emission spectroscopy (XES) was first performed for Au-CoFe₂O₄@SiO₂ and controls (Fig. 2a). XES represents a benchmark to probe the spin states of molecular or nanoscale systems, where the Kβ emission lines of spin-active elements are highly sensitive to the effective number of unpaired 3 d electrons^18,38,39, and the Kβ’ satellite could also be correlated with local 3d spin moments⁴⁰. In our measurements, a main Fe³⁺ Kβ_1,3 emission peak as well as a pronounced Kβ’ satellite line could be observed, respectively at ~7058 and ~7044 eV. Importantly, when Au-CoFe₂O₄@SiO₂ is illuminated with Xe lamp (1-sun), both Kβ_1,3 and Kβ’ lines demonstrate an apparent redshift towards lower emission energies (Δ_shift = 0.4 eV and 1.1 eV, respectively), consistent with the lower binding energies of the HS e_g electron. Furthermore, an observable increase in the relative intensity of the Kβ’ satellite peak is observed with illumination (relative area ratio of 0.20) compared to in dark (0.17) (Fig. S11 and Table S2). These phenomena are highly consistent with light-induced spin polarization, as HS electronic states favor the 3p-3d spin-orbit coupling and could lead to the redshift of Kβ emission as well as favor the Kβ’ satellite feature³⁹. In contrast, the CoFe₂O₄-only control showed no evident changes with illumination (Fig. 2a, inset), which underscores the necessity of surface plasmon in spin modulation.

Fig. 2: Evidence for surface plasmon-mediated spin modulation.

a XES spectra of Au-CoFe₂O₄@SiO₂ and CoFe₂O₄-only control (inset) with/without illumination. b Raman spectra of Au-CoFe₂O₄@SiO₂ under non-resonant (top) and resonant (bottom) incident wavelengths. c–e Au-CoFe₂O₄@SiO₂ Fe K-edge XANES spectra (Inset: Magnification of the absorption rising edge and pre-edge peak) (c), Fourier transforms of k²-weighted EXAFS in R-space (d) and wavelet transformed-EXAFS (e) with/without illumination. f Calculated magnetic moment, Fe–O bond length and O–Fe–O bond angles at LS and HS. Red spheres: Oxygen; brown: Fe; yellow: lone-pair electron.

Spin polarization has also been shown to induce distinct alterations to Raman vibrational features^41,42. In accordance, surface-enhanced Raman spectroscopy was carried out for Au-CoFe₂O₄@SiO₂ at on-resonance and off-resonance excitation wavelengths (Fig. S12). Consistent with our measurements in Fig. 1d, non-resonant 633-nm excitation led to two obvious A_1g Raman features at 627 and 684 cm⁻¹ (Fig. 2b). Importantly, with resonant 785-nm excitation, a prominent new peak appears at higher frequencies (709 cm⁻¹), indicating a higher-energy vibrational mode corresponding to HS-CoFe₂O₄. This observation is consistent with previous reports that spin polarization into the HS state typically leads to blueshift of the vibrational modes or new signatures at larger frequencies⁴¹. As expected, this wavelength-dependent new peak is not seen with any excitation conditions for CoFe₂O₄-only (Fig. S13). Notably, the Raman spectra were also measured for Au-CoFe₂O₄@SiO₂ with in situ heating up to 40 °C, where no evident changes to the two A_1g peaks were detected (Fig. S14), providing direct proof that it is through the resonant plasmonic near-field, rather than plasmonic heating, that the spin modulation is achieved.

To obtain in-depth understandings on the coordination and electronic structure, X-ray absorption spectroscopy (XAS) was next performed^43,44. The average coordination number of Fe was first analyzed from the X-ray absorption near-edge structure (XANES) spectra as well as Fourier transformation (FT) of k²-weighted Extended X-ray Absorption Fine Structure (EXAFS) spectra (Fig. S15 and Table S3). A predominant proportion of 86% of Fe cations with a coordination number of 6 (positioning in octahedral interstices) is calculated, translating to a ratio of 0.72:0.28 between regular and inverse spinel structures, with clear prevalence of the former. This confirms the crystal and electronic structure outlined in the previous section that octahedral Fe(III) represents a most dominant spin-active component.

To illustrate the light-induced alterations to the electronic structure, in situ illumination was next applied in XAS measurements. When illuminated, the Au-CoFe₂O₄@SiO₂ Fe K-edge XANES spectrum exhibits a conspicuous shift of the absorption rising edge towards lower absorption energies, as well as increases in the white-line intensity and width (Figs. 2c and S16). These observations are highly consistent with the lower binding energies as well as the higher coordination number associated with the unpaired 3d electrons at HS⁴³. Transport of plasmonic charge carriers from Au towards CoFe₂O₄ could, in theory, also lead to similar shifts towards lower energies; yet it is effectively excluded by the silica nanoshell and the XPS peak shift upon illumination (Fig. S8). Moreover, the pre-edge absorption peak around 7115 eV is also correlated with the spin configuration, particularly for surface Fe(III) in square pyramidal crystal fields. This peak is derived from the Fe 1s-3d spin transition, whose intensity is inversely related to the degree of centrosymmetry of the local coordination environment^45,46. The lone-pair electrons in HS Fe could induce moderate distortion of the equatorial O ligands owing to steric repulsion, and thus decrease its degree of centrosymmetry. As expected, the Fe pre-edge intensity in Au-CoFe₂O₄@SiO₂ increased with illumination (Fig. 2c, inset), consistent with the LS-to-HS spin transition.

Fourier transformation of the Fe K-edge EXAFS spectra (Fig. 2d) as well as wavelet-transform analysis (Fig. 2e) indicate that when illuminated, the fitted Fe–O scattering in Au-CoFe₂O₄@SiO₂ exhibits significant lengthening, which can typically be correlated with elongated bond length associated with the HS population owing to its abundant lone-pair electrons^47,48. In addition, evident increase in R-space signal intensity is also observed with illumination, which can be accounted for by the increase in local degree of disorder caused by the spin polarization into HS (Table S4).

The observed changes in coordination structure as well as magnetic properties of the plasmonic spin-active catalyst are further confirmed by our DFT calculations (Fig. 2f). For fully LS surface Fe in square pyramidal crystal field, a net average number of lone-pair electron of 1.50 (effective magnetic moment μ_eff = 2.28) is calculated, consistent with our Curie-Weiss analyses (Fig. S9b). As anticipated, increasing magnetic moment and bond lengths are found to be associated with the high-spin catalyst, consistent with our EXAFS analyses above. Calculations of the O–Fe–O bond angles shed further light on the spin modulation effect for surface Fe(III). For LS CoFe₂O₄, the equatorial O ligands primarily assume a planar configuration, with the sum of adjacent O–Fe–O bond angles (α + β) close to 180°. Importantly, with illumination, the HS CoFe₂O₄ shows an increase in the sum of adjacent O–Fe–O bond angle, exceeding 180^° (Fig. 2f). This agrees well with the strong steric repulsion of HS lone-pair electron that pushes the equatorial oxygen ligands out-of-plane.

Photochemical implications of plasmon-mediated spin modulation

The rapidly oscillating plasmonic near-field could not only achieve spin polarization to facilitate HS production, but also exert spin-selectivity to the relaxation processes, overcoming photobleaching and maintaining spin multiplicity. This could, in turn, significantly prolong the lifetime of HS excited states, thus bridging the temporal mismatch between the transient light-induced spin transitions and chemical reactivity. To obtain insights on the relaxation kinetics, transient absorption (TA) spectroscopy was next performed. After excitation with resonant pump laser (515 nm), significant ground state bleaching (GSB) signal could be observed for Au-CoFe₂O₄@SiO₂ at 550 nm due to the rapid relaxation of the excited plasmons⁴⁹ (Fig. 3a), which is also observed in the absorption pattern of the Au NP-only control (Fig. S17). Importantly, for Au-CoFe₂O₄@SiO₂, a strong broadband photo-induced absorption (PIA) feature is observed in the 650–950 nm range (Fig. 3a, b), which is much weaker in the CoFe₂O₄-only control sample owing to the latter’s rapid photobleaching of the light-induced spin transition (Fig. S18). The PIA persists far beyond the time window of our TAS measurements, as evidenced by a significant non-zero background signal before time zero. By comparing the amplitudes before and after time zero (see SI), we estimate its lifetime to be as long as ~60 μs (Fig. 3c). Such a long-lived species is consistent with a Fe HS excited state that relaxes slowly to the ground state. Notably, Au-CoFe₂O₄@SiO₂ exhibits a slow buildup time of ~62 ps. This feature corresponds to the plasmonic near-field-mediated energy transfer from Au nanoantenna to CoFe₂O₄, leading to an order-of-magnitude enhancement of the long-lived HS population compared to the CoFe₂O₄-only control, in which the slow buildup feature is not present. Time-resolved circular dichroism further reveals that the spin polarization of Au-CoFe₂O₄@SiO₂ decays with a time constant of ~6.8 ps (Fig. 3d), much slower than the sub-picosecond spin-flip times typically observed⁵⁰. This suggests that the spin polarization is likely preserved via resonant plasmonic energy transfer concurrent with illumination following the initial spin transition.

Fig. 3: Photochemical implications of plasmon-mediated spin modulation.

Transient absorption spectra (a), pseudo-color map (b), fitting of the absorption feature at 700 nm (c) and time-resolved circular dichroism spectrum (d) for Au-CoFe₂O₄@SiO₂. e DFT calculations of the PDOS for HS and LS Au-CoFe₂O₄@SiO₂. f Pulsed chemisorption for Au-CoFe₂O₄@SiO₂ with/without illumination using CO as a molecular marker. Inset: saturated quantity of adsorption.

Light-induced spin polarization into high-spin Fe states can significantly modulate the local density of EM states as well as the band structure of the plasmonic catalyst^11,12. In this term, DFT calculations of the projected density of states (PDOS) was performed for the illuminated and non-illuminated conditions (Figs. 3e and S19). Upon resonant irradiation, the light-induced spin polarization leads to noticeable dichotomy in state population distribution in HS CoFe₂O₄ owing to the spin transition and the difference in energy between different spin states, leading to an overall lower d-band center and wider distribution deviation. HS contains substantially increased and more localized state population at higher energy levels, which could present significant changes to its adsorption capability towards distinct reaction intermediates⁵¹. In the HS configuration, characterized by weaker crystal field splitting⁴⁷, unpaired electrons occupy both t_2g and e_g orbitals. This results in a high magnetic moment and significant contributions from e_g orbitals near the Fermi level (E_Fermi) (d_z² and d_x²−y²). Conversely, the LS configuration, stabilized by stronger crystal field splitting, pairs electrons in t_2g orbitals, reducing the magnetic moment.

Stable HS catalysts have been previously shown to be associated with increased surface available active site for chemical reaction as well as modulated adsorption capabilities^52,53. To demonstrate theses effects, pulsed chemisorption (PC) experiments were performed for Au-CoFe₂O₄@SiO₂ with or without illumination (Fig. 3f), where a CO molecular marker was exposed to the catalyst surface in a pulsed manner to determine chemical adsorption capability. The rate of adsorption (slope) for non-diffusion-controlled region can be considered to reflect the intrinsic adsorption capability of individual active sites, while the time (or number of pulses) at which saturation adsorption is reached is considered to reflect the number of active sites available for adsorption. As expected, upon resonant irradiation, the HS Au-CoFe₂O₄@SiO₂ exhibited both larger slope and much later saturation than the LS counterpart in dark, pointing at both improved adsorption capability and increased number of available sites for adsorption (Fig. 3f, inset).

Spin-modulated plasmonic catalytic performance

To investigate the effect of surface plasmon-mediated spin polarization on light-driven plasmonic catalytic performance, we next utilized the photoelectrochemical nitrate reduction reaction (NO₃RR) into ammonia as a benchmark. NO₃RR involves the relatively non-polar nitrate reactant molecules in the catalytic process, which could benefit from polarization activation by spin-modulated nanocatalysts^54,55. NO₃RR has also been shown to be sensitive to catalysts’ electronic spin state^56,57, which renders it a meaningful platform to test the effect of the plasmon-mediated spin modulation. To characterize the light-driven catalytic performance, linear sweep voltammetry (LSV) was first performed in 1 M KOH and 0.1 M NaNO₃ in dark or under 1-sun (100 mW/cm²) Xe-lamp illumination (Figs. 4a and S20). When illuminated, the LSV curves for Au-CoFe₂O₄@SiO₂ exhibited a tremendous gain in photocurrent density (J_photo) (106.8 mA/cm²), which corroborates the distinct light-driven catalytic reactivity. A similar J_photo (91.4 mA/cm²) was obtained with 590 nm LED irradiation, which resonates with the red-edge of Au surface plasmons (Fig. 1e). Considering the extremely low current density in dark, the plasmon-mediated spin modulation effectively activates plasmonic catalysis for light-mediated NO₃RR. To probe the effect of surface plasmons in promoting the reaction kinetics, the kinetic order (n) for nitrate⁵⁸ was determined by varying its concentrations in LSV measurements (Fig. S21). Notably, higher kinetic orders are calculated with illumination (n = 2.83) than in dark (n = 1.59) for Au-CoFe₂O₄@SiO₂.

Fig. 4: Spin-modulated catalytic performance.

a Photoelectrochemical nitrate reduction LSV curves for Au-CoFe₂O₄@SiO₂ and controls. b, c Incident wavelength and power dependence of photocurrent. d Light-modulated photocurrent measurement with alternating lights on/off cycles (100 s-on/100 s-off) (Inset: Magnification of the photocurrent response). e Chronoamperometric I-t curves at −1.5 V with light or in dark. f Photoelectrochemical ammonia production rate (Column chart) and selectivity (Line graph).

The role of photo-illumination on light-driven NO₃RR catalysis is next probed by monitoring the photocurrent as a function of incident wavelength and light power. In wavelength-dependence tests, maximum J_photo is attained with 590 nm illumination (Fig. 4b), which matches the plasmon resonance maximum of the nanocatalyst. For measurement of the catalytic photocurrent on illumination light power, it has been previously demonstrated that a linear dependence indicates a photochemistry-dominating process; whereas non-linear, exponential profile is typically considered a signature of the photothermal effect²². As expected, both 590 nm and polychromatic Xe lamp illumination depicted a linear dependence profile (Fig. 4c), consistent with the photochemical origin of the enhanced photocatalytic performance driven by the HS lone-pair electrons. This is further evidenced by our light-modulated photocurrent measurements⁵⁹, where illumination led to evident surges in the absolute value of the NO₃RR photocurrent (Fig. 4d). The increase in photocurrent upon irradiation contains primarily a fast-response region with rapid increase in photocurrent owing to the dominant photochemical effects. The moderate slow-response region with more attenuated increase due to photothermal effect is much less pronounced⁶⁰.

To quantify the contribution of photochemical and photothermal effects to light-driven catalytic performance, the temperature of Au-CoFe₂O₄@SiO₂ was monitored using infrared (IR) camera in situ. A 16 °C increase in catalyst surface temperature was observed under identical illumination and aqueous electrolyte conditions, which led to very minor changes to the electrolyte temperature (<2 °C) (Fig. S22). As demonstrated above, such extent of temperature increase does not have significant impact on the spin state of the plasmonic catalyst (Fig. S14). Its promotion role on catalytic performance is also moderate, as indicated by the much lower thermal current (11.2 mA/cm²) obtained with in situ external heating compared to the photocurrent (J_photo= 106.8 mA/cm²) (Figs. S23 and S24). These results further illustrate that it is the surface plasmon-mediated spin modulation that effectively activates light-driven catalytic performance.

The catalytic stability was next evaluated through chronoamperometric I-t measurements with illumination (Fig. 4e, see also Fig. S25), where the photocurrent density remains steady over a measurement period of 10 hours. This is further evidenced by the TEM images and Raman spectra for “spent” catalyst after the light-driven catalytic reaction, which depict no noticeable changes (Fig. S26). As a result of the stable photo-response, robust and highly selective ammonia production was achieved under photoelectrochemical catalytic conditions (Fig. 4f, see also Fig. S27). A highest NH₃ yield of 1.135 mmol mg_cat.⁻¹ h⁻¹ is attained by Au-CoFe₂O₄@SiO₂ with illumination. The selectivity of ammonia depicts a parabolic dependence pattern on applied bias, peaking at −1.6 V (87.3%) (Fig. 4f, see also Fig. S28), consistent with the increased water splitting at higher biases. These findings further corroborate the substantial light-mediated catalytic modulation of the spin-active plasmonic catalyst.

Catalytic mechanism and applicability

To illustrate the spin-modulated catalytic pathway, in situ Raman spectroscopy was utilized to probe the catalytic intermediates for Au-CoFe₂O₄@SiO₂ during operando photoelectrochemical NO₃RR measurements (Figs. 5a and S29). For all spectra obtained, two strong peaks from the D (~1313 cm⁻¹) and G (~1587 cm⁻¹) bands of the substrate as well as a clear band between 1000 and 1050 cm⁻¹ from the NO₃⁻ reactant are present⁶¹. No additional peaks could be observed at time 0 (Fig. 5a, bottom). According to the previous reports, the reduction and hydrogenation of the nitrogen species would lead to sequential production of *NH, *NH₂, and ammonia species⁶². Consistent with these reports, as the catalytic reaction progressed with illumination and −1.5 V bias, the signal of *NH at 1507 cm⁻¹ gradually emerged (30 min) but soon disappeared, accompanied by the appearance of distinct bands at 1130, 1240, 1380, and 1570 cm⁻¹ at 40 min, which are respectively assigned to the *NH₂ intermediate and ammonia products⁶³. These features became more pronounced as the reaction further proceeded. For the measurement with illumination at −1.6 V, evident acceleration of the reaction could be seen, where signature peaks for the ammonia products appeared in the spectrum obtained at 30 min (Fig. S29). In contrast, no noticeable intermediate/product signals could be observed under identical illumination conditions for CoFe₂O₄-only (Fig. S30), due to the lack of plasmonic near-field-mediated spin modulation.

Fig. 5: Catalytic mechanism and applicability.

a In-situ Raman spectra for photoelectrochemical NO₃RR by Au-CoFe₂O₄@SiO₂. b DFT calculation of the free energy diagrams for NO₃RR with the HS (red) and LS (blue) plasmonic catalyst. c In-situ Raman investigation of the nitrate activation behavior. d ICOHP calculations of ammonia energy levels adsorbed on HS or LS plasmonic catalyst. e Radar map of the light-driven plasmonic catalytic performance for Au-CoFe₂O₄@SiO₂ with/without illumination (red/black) and CoFe₂O₄-only in dark (blue). f Purely photocatalytic nitrate reduction performance for Au-CoFe₂O₄@SiO₂. Retest: 3 times. g NO₃RR catalytic performance as a function of external field intensity. h Generality of the plasmon-mediated spin modulation strategy.

To corroborate the experimental findings on reaction pathway, DFT calculations of the Gibbs free energy diagram were performed with the HS and LS plasmonic catalyst (Fig. 5b). The effect of excited Au plasmons on the Fe species was modeled by altering the magnetic moment of Fe active sites. Gibbs free energy diagram of the NO₃RR reaction indicates that the adsorbed *NO₃ is reduced to *NO, and then hydrogenated sequentially to *NH₃ via *NH, *NH₂ intermediates, consistent with our in-situ Raman results. For HS Au-CoFe₂O₄@SiO₂, in general, lower free energy barriers were calculated compared to the LS counterpart, indicating the facilitated adsorption and formation of the key reaction intermediates. For instance, the transition from *NO₃ to *HNO₃ requires only 0.28 eV in the HS state, compared to 0.44 eV in the LS state. Similarly, intermediate steps such as *NHO to *NH₃ exhibit significantly reduced Gibbs free energy changes in the HS state. The lower energy barriers are attributed to the unpaired e_g electrons, which enhance bonding interactions with adsorbed species and facilitate critical bond-breaking and forming processes that underpin catalytic efficiency. Notably, the HS state achieves a final Gibbs free energy of −0.40 eV during NH₃ formation, in contrast to 0.82 eV observed for the LS state, underscoring its thermodynamic favorability. Therefore, desorption of the *NH₃ into ammonium product is apparently facilitated by the HS nanocatalyst.

The role of light-induced HS Au-CoFe₂O₄@SiO₂ plasmonic catalyst in activating the nitrate reactant is highly consistent with our PDOS calculation and pulsed chemisorption results (Fig. 3e, f). The plasmon-mediated spin modulation creates substantially increased population of state at higher energy sections above the vacuum level, which demonstrates significant overlap with the LUMO of nitrate that facilitate their bonding interactions. This is corroborated by our calculation of the nitrate adsorption energy of the CoFe₂O₄, where the (400) crystal facet, calculated to be the predominant active facet, demonstrates significant enhancement in nitrate adsorption capability at light-induced HS state (Fig. S31a). In addition, the (311) crystal facet, containing the spin-active Fe(III)_oct species, is also found to be catalytically active (Fig. S31b). Furthermore, the plasmonic near-field as well as HS electrons effectively polarize the relatively non-polar N–O bondage, as indicated by our further in-situ Raman investigations, where the stretching frequencies of free-stranding and adsorbed nitrate are compared (Fig. 5c). Significant shift of N–O vibration towards lower frequencies are observed for adsorbed nitrate on Au-CoFe₂O₄@SiO₂ when exposed to resonant illumination, indicative of bond elongation through polarization activation. This change is not seen with non-resonant illumination, or with control groups missing in one or more of the active components (Fig. S32).

HS Au-CoFe₂O₄@SiO₂ could also modulate the reaction pathway for ammonia production by inhibiting side reactions including nitrogen and hydrogen production. HS Au-CoFe₂O₄@SiO₂ strongly disfavors formation of *N, which is crucial for forming nitrogen side product (Fig. 5b). As a result, no obvious production of N₂ gas or participation of atmospheric nitrogen are detected in our ¹⁵N isotope-labeling experiments (Fig. S33). On the other hand, light illumination considerably increases the proton adsorption energy barrier for HS species, inhibiting hydrogen production (Fig. S34). Once produced, desorption of the *NH₃ products are facilitated by the HS Au-CoFe₂O₄@SiO₂. The total Integrated Crystal Orbital Hamiltonian Population (ICOHP) of the Fe–NH₃ coordination in the HS configuration is −1.50 eV, which is substantially less negative than that in the LS counterpart (−2.14 eV), indicating reduced bonding strength of the former (Fig. 5d). The COHP spectrum of the HS configuration exhibits pronounced anti-bonding states near the Fermi level, which diminish orbital overlap and destabilize the Fe–NH₃ interaction, thereby facilitating *NH₃ desorption. In contrast, the LS state displays a more stabilized bonding landscape, with fully occupied bonding orbitals and negligible anti-bonding character, accounting for the stronger NH₃ retention and higher desorption barrier. This is consistent with our observation from the final step of ammonia desorption in the free energy diagram, and is also experimentally supported by our in-situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) measurements, where a light-induced repeatable change in the N–H bending vibration intensity at 1625 cm⁻¹ is observed (Fig. S35).

The improved polarization activation of nitrate reactant, modulated reaction pathway and facilitated ammonia desorption collectively activates the favorable light-driven plasmonic catalytic performance for the plasmonic spin-active catalyst (Fig. 5e). To shed further light on the unique role of light illumination, purely photocatalytic nitrate reduction was next performed for Au-CoFe₂O₄@SiO₂ under 1-sun illumination only, and the ammonia product was monitored over time (Fig. 5f). Remarkable photocatalytic properties were observed over 1.5-hour constant measurement, and an ammonia production rate of 0.041 mmol mg_cat.⁻¹ h⁻¹ was recorded, which presents intriguing potentials for sunlight-driven reactivity.

The unique feature of the surface plasmon-mediated strategy to simultaneously achieve in situ steady-state spin modulation while promoting charge carrier dynamics is extremely valuable and unlikely to be achieved with external fields. To demonstrate this aspect, we compared the spin state and carrier dynamics of Au-CoFe₂O₄@SiO₂ in the presence of external magnetic fields. Whilst external magnetic fields ≥100 Oe can also induce HS Raman features (Figs. S36 and S37), they could typically also bring about increased electron transport resistances (R_ct) as well as considerably reduced charge carrier population according to our photoelectrochemical impedance and Mott-Schottky measurements (Fig. S38). These results are in sharp contrast with the effect of illumination, where reduced R_ct and promoted carrier transport behaviors are consistent observed. As a result of the tradeoff between the spin polarization and carrier dynamics, the external field-mediated catalytic performance manifests a parabolic dependence pattern on field intensity (Fig. 5g). Even the highest magneto-current obtained (86 mA/cm²) is far lower than the photocurrent obtained with illumination (150 mA/cm²). For the illuminated condition, the localized plasmonic near-field guarantee effective spin polarization for long-lasting magnetic moments, while their pico-to-femtohertz oscillation frequency exerts spin-selectivity in the recombination and facilitates carrier extraction^64,65.

Another important advantage associated with the surface plasmon-mediated spin modulation strategy is the highly general applicability towards different catalyst configuration as well as plasmonic and spin-active components. To demonstrate the generality, plasmonic Ag and Cu nanoantenna, as well as CoO nanocrystals (having spin-active 3d⁷ octahedral Co²⁺) were utilized in place of the Au antenna and/or the CoFe₂O₄ catalyst. With their respective resonant excitation wavelengths, the Ag and Cu antenna have also been shown to induce effective spin modulation, as evidenced by the noticeable HS Raman features (Fig. 5h), as well as significant activation in light-driven catalytic performance (Fig. S39). The generalized applicability is likely to enable the on-site and on-demand customization of spin modulation towards distinct types of catalytic reactions.

Discussion

In summary, we develop in this work an effective and generalized strategy for in situ, steady-state and on-demand electronic spin modulation through resonant light excitation of localized surface plasmons. The rapidly oscillating plasmonic EM near-field spin-polarizes low-spin spinel CoFe₂O₄, and effectively overcomes its photobleaching to produce stable high-spin states with an astounding lifetime of ~60 μs, thus overcoming the temporal mismatch between transient spin transitions and slower chemical reaction kinetics, activating its light-driven catalytic performance. Benefiting from the polarization activation of nitrate reactant, modulated reaction pathways and facilitated ammonia desorption, the plasmonic high-spin catalyst balances the tradeoff between spin polarization and carrier dynamics, and achieves considerable photo-enhancement in ammonia production rate and selectivity as well as photocatalytic nitrate reduction performance. Our strategy provides valuable and generalized insights for on-demand and steady-state electronic spin engineering, and opens new avenues for light-induced spin modulation towards distinct fields of applications.

Methods

Materials

All chemicals were used as raw material without further purification. Gold acid chloride trihydrate (HAuCl₄·3H₂O, ≥99.9%), oleylamine (OAm, 80–90%), iron acetylacetonate (Fe(acac)₃, >98%), cobalt acetylacetonate (Co(acac)₂, >97%) and (3-Mercaptopropyl)trimethoxysilane were purchased from Aladdin. Trisodium citrate dihydrate (C₆H₅Na₃O₇·2H₂O) was purchased from Enox. Sodium borohydride (NaBH₄, 99%) was purchased from Thermo Scientific. L-Ascorbic acid (AA, 99.99%) and polyvinylpyrrolidone (PVP, average mol wt 10,000) were purchased from Macklin. The silica precursor tetraethyl orthosilicate (TEOS) was obtained from TCI (Shanghai). Sigma-Aldrich supplied potassium iodide (KI, 99%), ammonium hydroxide (28 wt%) and the surfactant Igepal CO-520. All solvents (ethanol, acetone, and cyclohexane) were supplied by Sinopharm Chemical Reagent Co., Ltd.

Preparation of Au NPs

Au NPs were synthesized through a modified seed-mediated growth approach. Seed solution was first prepared by mixing water solutions of HAuCl₄ (5 mM, 1 mL) and C₆H₅Na₃O₇ (1 mL, 5 mM), and diluting to a total volume of 18 mL DI water. Then, a reducing agent containing 10 mM NaBH₄ was added under vigorous stirring. The color of the seed solution turned dark brown after adding the reducing agent. For NPs growth, a growth solution containing aqueous solutions of PVP (1 wt%, 1.25 mL) and HAuCl₄ (0.25 M, 300 µL) was prepared. Then, the mixed solution containing 1.25 mL AA (0.1 M) and 1.25 mL KI (0.2 M) was added into the growth solution. Finally, 30 µL seed solution was added into the growth mixture with violent stirring for 20 min, then washed three times with deionized water, and stored at 4 °C for further use.

Preparation of CoFe₂O₄ nanocrystals

The synthesis of CoFe₂O₄ nanocrystals followed a modified solvothermal protocol. Specifically, a precursor solution was prepared by dissolving 3.6 mM Fe(acac)₃ and 1.8 mM Co(acac)₂ in 32 mL of oleylamine under vigorous stirring for 10 min at room temperature. This solution was then sealed in a 100 mL autoclave and heated to 200 °C for a duration of 12 h. Following the reaction, the autoclave was cooled to room temperature. The solid product was isolated via centrifugation, purified through sequential washing with acetone and ethanol, and finally dispersed in acetone for future use.

Preparation of CoFe₂O₄@SiO₂ Nanocrystals

A typical synthesis began with dissolving 1 g of the surfactant Igepal CO-520 in 11 mL of cyclohexane via 30 min of ultrasonication. Subsequently, 1 mL of a pre-prepared cyclohexane dispersion containing CoFe₂O₄ nanocrystals (10 mg) was introduced to this mixture under stirring. After the combined solution was stirred for 5 h, tetraethyl orthosilicate (TEOS, 0.05 mL) and aqueous ammonia (28 wt%, 0.1 mL) were added sequentially. The reaction was allowed to proceed for another 5 h under continuous stirring. The final core-shell nanoparticles were isolated by centrifugation, purified through multiple ethanol washes, and ultimately stored in ethanol.

Preparation of Au-CoFe₂O₄@SiO₂ Nanocrystals

Au NPs and CoFe₂O₄@SiO₂ nanocrystals were linked by the thiolated silane linker. For the thiolated silane linking with Au NPs, 30 mg of Au NPs were dissolved in 30 mL of ethanol under ultrasonication for 10 min and then mixed with 0.75 mL of (3-Mercaptopropyl)trimethoxysilane. After stirring for 12 h at room temperature, the mixed solution was washed three times with ethanol and stored in ethanol. Next, 20 mg of the above solution (Au NPs with the thiolated silane linker) were dissolved in 10 mL ethanol and 15 mL DI water. Then, 0.6 mg of CoFe₂O₄@SiO₂ was added in three times with a 10-min interval under continuous ultrasonication. Finally, the products were collected by centrifugation, cleaned with ethanol for several times and stored in ethanol.

Structural characterizations

Transmission electron microscopy (TEM) images and Energy-Dispersive X-Ray Spectroscopy (EDS) mappings were obtained with an FEI Talos F200X S/TEM at 200 kV with a field-emission gun. Samples were prepared by casting several drops of 0.1 mg/mL solutions onto copper-grid mounted “holey” carbon films, dried at room temperature. The Au, Fe and Co contents were measured by inductively coupled plasma-optical emission spectroscopy (ICP-OES) (Avio200, PerkinElmer).

Measurement of UV-vis absorbance spectra

UV-vis absorbance spectra were obtained using a Lambda 750 UV/VIS/NIR spectrometer from PerkinElmer at a scan rate of 120 nm min⁻¹ over the range 200–800 nm. Two microliters solution of Au-CoFe₂O₄@SiO₂ and other controls were separately recorded in a standard quartz cuvette with path length of 1 cm, volume of 3.5 cm. The blank control group ethanol solvent was performed in the same condition with the solvent above supplying the testing background.

Measurement of Raman spectroscopy

Raman spectroscopic measurements were performed using a Renishaw InVia confocal Raman Microscope equipped with a Leica 100× air objective (Numerical Aperture = 1.25). Samples, including Au-CoFe₂O₄@SiO₂ and controls prepared as drop-casted thin films on silicon wafers (<100>, Ferrotec Shanghai), were analyzed either on the substrate or within an in-situ electrolysis cell. The measurements employed 633 nm (1800 l/mm grating) and 785 nm (1200 l/mm grating) excitation lasers under the following parameters: 14 mW power, 60 s exposure time, and 2 accumulations.

Measurement of single-particle scattering spectroscopy

The scattered light from individual particles was first collected using a Zeiss 50× air-spaced objective (numerical aperture = 0.8). This signal was then directed through a hyperspectral imaging spectrograph (Princeton Instruments, Acton SpectraPro 2156) equipped with a slit aperture. The dispersed spectrum was ultimately detected by a thermoelectrically cooled, back-illuminated CCD camera (Pixis 400BR, Princeton Instruments). The entire detection assembly was mounted on a computer-controlled translation stage (Newport, LTA-HL).

In-situ irradiated X-ray photoelectron spectroscopy (ISI-XPS) characterization

ISI-XPS measurements were conducted using a Thermo Escalab 250Xi photoelectron spectrometer. A non-monochromatized Al Kα X-ray source (1486.6 eV) was employed for excitation. A full-spectrum light source (HF-GHX-XE-300) at an intensity of 100 mW/cm² was used as the illumination source. The light was introduced from the top of the analysis chamber through a quartz window. To probe spin polarization behavior, then the XPS spectra (specifically Au 4f, Fe 2p, and Co 2p) were acquired and compared under both dark and illuminated conditions.

X-ray absorption spectroscopy (XAS)/X-ray emission spectroscopy (XES) characterization

Fe K-edge XAFS and XES spectra were performed using a lab-based instrument (easyXAFS300+ from easyXAFS, LLC) with an industrial Pd tube source (VAREX) and a silicon drift diode detector (Amptek SDD-X123). Data merging, background subtraction, normalization, and fitting of the XANES and EXAFS spectra were modeled using Artemis and Athena software, respectively. The k²-weighted Fourier transforms for all the EXAFS data were conducted in a Hanning-shaped window. The amplitude-reduction factor S02 was defined at 0.7. The R-ranges for the fitting of all the EXAFS data were set as 1–4.5 Å. The coordination numbers (CN), distances (R), and Debye–Waller factors (σ) were determined independently. The quality of the fitting was evaluated by the R-factor. We performed EXAFS curve fitting for the spinel oxides by referencing previous benchmark.

Measurement of transient absorption (TA) spectroscopy

The TA measurements were performed on a custom-built setup. A 1030 nm femtosecond output from a laser (BFL-1030-20BS, BWT) was split into two beams in an 80:20 intensity ratio. The more intense (80%) beam served as the pump. It was frequency-doubled to 515 nm, modulated by a 500 Hz optical chopper, and then focused onto the sample using a convex lens. The weaker (20%) beam functioned as the probe. It was focused into a 5 mm YAG crystal to generate a supercontinuum white-light spectrum. The transmitted probe light was collected by a fast fiber-optic spectrometer, whose acquisition was synchronized with the chopper frequency for lock-in detection.

Photoelectrochemical characterizations

Using a three-electrode setup on a CHI 660E electrochemical workstation, we performed LSV, chronoamperometry (i-t), EIS and Mott-Schottky measurements. The working electrode was a 5 mm Glassy Carbon (GC) disk, modified by drop-casting 5 μL of the catalyst colloid (Au-CoFe₂O₄@SiO₂ or controls). An Ag/AgCl electrode and a Pt wire served as the reference and counter electrodes, respectively. The potential of −1.2 V, −1.5 V and −1.6 V with a similar collection rate of 1 Hz were adopted for the chronoamperometric i-t measurement. The wavelength dependence test used monochromatic-light LEDs with wavelengths of 420, 470, 520, 590, 620, 740 nm (MC-LED-M-λ, Beijing MerryChange Technology Co., Ltd.) with the light intensity of 100 mW/cm². The intensity-dependent photoelectrochemical test also used the full-spectrum Xe-lamp and 590 nm LED illumination by adjusting the light intensity of illumination from 50 to 200 mW/cm², respectively. EIS Nyquist plots were collected at an open circuit −1.6 V, with a frequency range from 100,000 to 10 Hz.

Current density (mA/cm²) of Au-CoFe₂O₄@SiO₂ and other control samples was calculated from the measured current and the total geometric surface area (SA) of different samples on the working electrode. For the particulate photocatalyst, the calculated geometric surface area gives a conservative estimation of the current densities and photocurrent densities. Five microliters Au-CoFe₂O₄@SiO₂ colloid with an Au concentration of 175.05 ppm calculated from ICP-OES measurements (Table S1), was dropcasted on GC electrodes for PEC measurements, which gives a mass of 8.925 × 10⁻⁷ g Au. Based on the mass density of 19.32 g/cm³, the total particle volume and the volume of an individual Au NP can be calculated. The number of Au-CoFe₂O₄@SiO₂ NPs is calculated as 1.5 × 10⁸, SA amounts to 0.033 cm².

Photocurrent densities (J_photo, mA/cm²) were calculated by subtracting the current densities in dark from those measured with illumination. Conversion between Ag/AgCl reference electrode and Reversible Hydrogen Electrodes is based on: E (vs RHE) = E(vs Ag/AgCl) + 0.0591 pH + E^ө(vs Ag/AgCl), where the E^ө(vs Ag/AgCl) = 0.1976, the pH = 14.

Catalytic mechanism characterizations

The NH₃-desorption in situ DRIFTS experiments were performed using an FTIR spectrometer (Thermo Nicolet iS50). The CO pulse chemisorption experiments were performed in the flow reactor with gas chromatography (GC).

Density functional theory (DFT) calculations

All spin-polarized DFT computations were performed with the Vienna Ab initio Simulation Package (VASP) code. The computational setup utilized the projector-augmented wave (PAW) pseudopotentials and the Perdew–Burke–Ernzerhof (PBE) generalized gradient approximation (GGA) exchange-correlation functional. A uniform 3 × 3 × 3 k-mesh grid was used to optimize the crystal structure of bulk CoFe₂O₄, with lattice parameter a = b = c = 8.45 Å. For the slab models, the high-spin (HS) CoFe₂O₄ structure was constructed with lattice parameters of a = b = 8.53 Å, c = 21.80 Å, and α = β = γ = 90°. Pending inevitable variations in the actual crystal lattice, this model provides a closest replication of the nanocatalyst to the best of our capabilities. The low-spin (LS) CoFe₂O₄ slab dimensions were reduced to 89% of the original values. Both HS and LS CoFe₂O₄ slab models consisted of five atomic layers with (001) exposed surface separated by ~16 Å of vacuum. For the HS structures, only the bottom three atomic layers were fixed during optimizations, while all other atoms were allowed to relax. For the LS structures, both sides of the slabs were fixed to maintain the single e_g orbital occupation. The first Brillourin zone was sampled with a gamma-center 3 × 3 × 1 k-point grid for the slab models. The kinetic energy cutoff for the wavefunctions was set at 450 eV for the bulk and slab structures without and with molecular adsorption. The structural optimizations were considered convergent when the force on each atom was less than 0.01 eV/Å. The density of states (DOS) calculations were performed using the DFT + U approach, with U values of 3.5 and 4.0 eV applied to Fe and Co, respectively, in both HS and LS CoFe₂O₄ models. Following the previous study, which indicates that adsorption energies are independent of the U value, the U correction was only applied for electronic structure calculations.

The Gibbs free energy values were calculated using the following equations:

$$G={E}_{{{{\rm{DFT}}}}}+{E}_{{{{\rm{DFT}}}}}+\int\limits^{298.15}_{0}{C}_{v}dT-TS$$

(1)

$${E}_{ZPE}=\frac{1}{2}\sum h\upsilon$$

(2)

$$\int\limits^{298.15}_{0}{C}_{v}dT={k}_{{{{\rm{B}}}}}\sum {\left(\frac{h\upsilon }{{k}_{{{{\rm{B}}}}}{{{\rm{T}}}}}\right)}^{2}\frac{exp(\frac{h\upsilon }{{k}_{{{{\rm{B}}}}}{{{\rm{T}}}}})}{{[exp(\frac{h\upsilon }{{k}_{{{{\rm{B}}}}}{{{\rm{T}}}}})-1]}^{2}}$$

(3)

The terms in equations are defined as follows: E_DFT stands for the DFT-computed energy, E_ZPE signifies the zero-point vibrational energy, ${\int }_{0}^{298.15}{C}_{v}{dT}$ corresponds to the integrated heat capacity and S is the entropy. The constants ℎ (Planck’s constant) and k_B (Boltzmann constant) are used alongside the vibrational frequency. For entropy, only vibrational contributions were considered due to the insignificance of translational and rotational terms.