Chapter 1: In situ electron paramagnetic resonance (EPR) – a unique tool for analysing structure and reaction behaviour of paramagnetic sites in model and real catalysts Free

EPR spectroscopy has been used to characterize catalysts since very early on, because of its ability to provide detailed information on paramagnetic species such as their geometric and electronic structure or their chemical environment. In terms of the systems being investigated, EPR covers a wide range of applications from heterogeneous, via homogeneous catalysts all the way to enzymatic systems. Paramagnetic species in catalytic systems range from sites considered crucial for catalytic turnover, such as transition metal ions, to paramagnetic reaction intermediates. Analyzing their behavior in situ, i.e. under conditions as close as possible to those of a catalytic reaction, can be most helpful for deriving structure–reactivity relationships and reaction mechanisms. Nevertheless, applications of in situ EPR spectroscopy for monitoring catalytic reactions are not as numerous compared to other common methods such as vibrational spectroscopy or X-ray techniques, due to the fact that it is restricted to systems containing unpaired electrons. However, for catalytic redox processes in which electrons are transferred between catalyst and reactants, in situ EPR (or operando EPR, as it is called when together with the EPR spectrum catalytic conversion/selectivity data are measured) is a unique tool, since it can visualize this electron transfer directly as long as paramagnetic species are involved. This has been illustrated in the past for a variety of hydrocarbon conversion reactions. Thus, mixed oxide and oxynitride bulk phases such as vanadium–phosphorus oxides,¹  vanadium–molybdenum oxides,²  mixed VTiSbSi oxides,³  VAlON and VZrON oxynitrides⁴  or heteropolyacids of defined structure⁵  have been used for selective oxidation of alkanes and aromatics. Such reactions have also been monitored by in situ EPR over supported vanadia and bismuth molybdate catalysts¹  while supported chromia¹  and nickel catalysts^6,7  were studied during non-oxidative aromatization of alkanes as well as during interaction with butenes. Another important class of heterogeneous catalytic systems analysed by in situ EPR comprises zeolites containing transition metal ions either incorporated in the framework or located in pore positions. Many of such catalysts have been used to remove nitrogen oxides (NO_x) from exhaust gases. Relevant examples comprise the use of Co–BEA⁸  and Fe–ZSM-5⁹  for selective catalytic reduction of NO_x, but also Cu–ZSM-5 has been widely used for the same purpose by Kucherov et al., whose work has been reviewed in ref. 1. The interaction of benzene with oxygen has been recently studied on the same type of catalysts to understand the gas-phase oxidation of benzene to phenol.¹⁰  For a more comprehensive selection of application examples for in situ EPR in heterogeneous catalysis the reader is referred to a number of reviews.^1,2,11–13  Surprisingly, the use of in situ EPR spectroscopy for elucidating structure–reactivity relationships in homogeneous redox catalysis is even more limited than in heterogeneous catalysis.

In EPR spectroscopy the sample is placed in an external magnetic field to lift the degeneracy of the electron spin states and microwave radiation is used to induce magnetic dipole transitions between these states. Historically, most of the information has been obtained using continuous wave EPR spectroscopy operating at a microwave frequency of ≈10 GHz in the so-called X-band. Spectrometers of this kind are still by far the most abundant ones, however, the last decades have seen a significant diversification of experimental capabilities, namely, the commercial availability of spectrometers operating at different microwave frequencies as well as pulse spectrometers, which enables the use of these techniques also outside of laboratories dedicated to instrumental developments in EPR spectroscopy.^14,15 

The aim of this chapter is to acquaint the reader with the basic principles and application opportunities of in situ EPR in redox catalysis. The introductory part starts with a presentation of instrumental aspects and experimental procedures (Section 2), followed by a discussion of the most important parameters, the g and A matrix components, that can be derived from EPR spectra (Section 3). Here we will focus on the examples of Au atoms and O₂^•− radicals deposited on a MgO(001) surface, which both represent the simplest case of paramagnetic species with a single unpaired electron and a spin of S=1/2. We will restrict ourselves to the analysis of cw EPR spectra being still the most common ones and typically the initial step in an EPR spectroscopic investigation. This discussion will be amended by only very few examples of pulse EPR techniques (Section 4), which provide valuable additional information on hyperfine interactions in paramagnetic species that can reflect peculiarities of their environment with much higher resolution than cw EPR.

In the application oriented part of this chapter (Section 5), we present two case studies, which illustrate the potential of in situ EPR for deriving structure–reactivity relationships in both heterogeneous and homogeneous catalytic redox processes, whereby special benefits arising from the coupling of in situ EPR with other techniques are explicitly pointed out. The first example is dedicated to a supported VO_x/TiO₂ catalyst. Such catalysts are of paramount importance for a variety of heterogeneous catalytic redox processes. Thus, V₂O₅/TiO₂ modified by WO₃ (also known as EUROCAT oxide) is an industrial catalyst for selective catalytic reduction of NO_x in power plant exhaust gases, which has been studied in a round robin test by many European catalysis laboratories.¹⁶  Besides, there is a multitude of papers, among them those containing operando EPR studies^17–19  in which the use of supported V₂O₅/TiO₂ catalysts is described for a variety of catalytic oxidation reactions. In the first case study, the influence of surface sulfates on the structure of VO_x species dispersed on the surface of titania²⁰  and their catalytic behaviour in the oxidative dehydrogenation of propane to propene is analyzed by operando EPR coupled with UV-vis diffuse reflectance and laser-Raman spectroscopy.¹⁸  The second case study illustrates the special benefits of in situ EPR, supported by vibrational spectroscopy, for elucidating the mechanism of photocatalytic water reduction in a homogeneous catalytic system. A similar approach has been used very recently to unravel different wavelength dependent electron transfer mechanisms in plasmonic Au/TiO₂ water splitting catalysts.²¹  The selection of these case studies was done with the aim to highlight the versatility of in situ EPR spectroscopy for a wide variety of reaction conditions, as long as paramagnetic species are involved.

Prior to a discussion of the appearance of EPR spectra it is appropriate to spend some time on experimental considerations. Herein, the discussion will be restricted to cw spectroscopy. With respect to pulse spectroscopy the interested reader is referred to the literature (e.g. ref. 22). EPR spectroscopy probes the properties of paramagnetic species by means of magnetic dipole transitions. Because of the small oscillator strength of such magnetic dipole transitions resonators are widely used in EPR spectroscopy to enhance the magnetic field strength at the sample. The properties of the resonator, which can be characterized by quantities such as the quality factor, filling factor or conversion efficiency, crucially determine the performance of the experiment. The use of resonators typically implies that monochromatic radiation has to be used, which forms a standing wave pattern inside the resonator. Additionally, the resonator structures lead to a separation of the magnetic from the electric field component helping to reduce undesired electric dipole excitations. The reader is referred to the literature for a more comprehensive discussion.²³ 

The EPR spectrum is obtained by sweeping the external magnetic field. Despite the fact that resonator structures are used, the amount of absorbed radiation is still very small. To improve the signal-to-noise ratio of the spectrum, cw-EPR spectrometers use a lock-in detection scheme. To this end, the external magnetic field is modulated with a known frequency and phase, which allows phase sensitive detection of the corresponding Fourier component of the detector signal. While this sounds like a pure technicality it has the important consequence that cw-EPR spectra are recorded as the first derivative of the absorption spectrum, more precisely: the first derivative of the imaginary part of the complex high frequency susceptibility by the field (dχ″/dB) as a function of the magnetic field. The number of spins may be determined from such a spectrum by double integration. This has to be done with care to avoid artefacts, e.g. due to imperfections of the baseline. Because of the direct relation of the EPR signal with the susceptibility of paramagnetic samples, the intensity of the EPR signal should obey Curie's law, which predicts an inverse proportionality of the magnetic susceptibility and, hence, the EPR signal intensity with temperature. A simple and often used test for Curie behavior is to plot the EPR signal intensity multiplied by the sample temperature (I×T) as a function of the temperature. For a paramagnetic sample with a constant number of spins, this product should be constant and deviations from this may be discussed as changes in the amount of paramagnetic species detected by the spectroscopy at different temperatures. For catalytic systems containing species, which may adopt paramagnetic as well as diamagnetic states, such as transition metal centres, temperature dependent measurements of the EPR signal intensity are a rather simple test to monitor the amount of paramagnetic species for different conditions. It is important to note that it is the signal area of an individual species and not simply the peak-to-peak amplitude of a line or the integral intensity of the entire spectrum, which needs to be considered. The latter two quantities may be used under appropriate conditions, too. The analysis of the temperature dependent intensity as outlined above also assumes that the experimental conditions will not induce additional changes in the measured spectrum. To this end, saturation effects may be one source of error, because relaxation times are a function of temperature. Unfavourable relaxation properties may also be responsible for the difficulty to observe paramagnetic species at certain temperatures. It is well known that not all paramagnetic species can be observed by EPR at all temperatures. Prominent examples are certain transition metal ions such as Co²⁺ or Ni²⁺, which typically require very low temperature to be observed. The relaxation behaviour of paramagnetic centres is one of the important issues to be considered in detail when one tries to apply EPR spectroscopy as an operando technique in catalysis. Despite the importance of relaxation phenomena for EPR, a more detailed discussion goes beyond the scope of this introductory part and the interested reader is referred to the literature on this topic.^24–28 

The key step in the interpretation of EPR spectra is to identify the various underlying physical effects by analysing appropriate EPR spectra. In order to get acquainted with the language typically used in EPR spectroscopy within this introductory chapter we will restrict the discussion to systems with a spin doublet ground state. Placing such a doublet state in a magnetic field leads to a splitting of the two spin states S=±1/2 according to Zeeman interaction, which gives rise to a linear dependence of the energy difference between the spin states from the magnetic field. This shall be exemplified using Au atoms, which are characterized by an unpaired electron in the 6s orbital leading to a ²S ground state. In the particular example presented here Au atoms are deposited on a single crystalline MgO(001) surface at 30 K to prevent diffusion of the atoms, which would lead to the formation of nanoparticles. The experiments were done under ultrahigh vacuum conditions (p<1×10⁻¹⁰ mbar) to ensure that the results are not perturbed by the interaction with molecules from the gas phase. For the magnetic field being oriented parallel to the surface, Au atoms on MgO(001) show an EPR spectrum as presented in Fig. 1a.²⁹ 

In contrast to the simple consideration based on the Zeeman interaction of the doublet ground state, it does not show a single line, but four non-equidistant lines of considerably different amplitude. When observing such a spectrum it is not obviously clear that these four lines correspond to the same paramagnetic species. However, a closer look into the properties of Au atoms helps to rationalize the appearance of the spectrum. Au consists exclusively of the isotope ¹⁹⁷Au having a nuclear spin of I=3/2. The unpaired electron couples to the nuclear spin of the Au atom. This is called hyperfine interaction. Therefore, it is expected that the Zeeman line splits into four peaks due to the interaction of the unpaired electron with the 4 different nuclear spin states of Au. Note that a statistically picked Au atom within the ensemble will have a certain nuclear spin state and give rise to one of the four lines observed in the spectrum. Because of the equipartition of the nuclear spin states even at 30 K, all four nuclear spin states are populated equally and thus the Boltzmann weight of each line is identical for an ensemble of spins as observed here. A double integration of the spectrum reveals that within experimental error the four lines have the same intensity despite the fact that the observed signal amplitudes are different. The latter is due to differences in line width observed for the different lines.

The hyperfine interaction plays a pivotal role for the analysis of EPR spectra of many catalytically important systems, because it often allows extracting valuable information on the electronic structure as well as on the environment of the paramagnetic centres. From a qualitative point of view the preceding discussion seems to be sufficient to understand the appearance of the EPR spectrum of Au atoms. However, additional experiments at different angles between the surface and the magnetic field summarized in the lower diagram of Fig. 1a reveal a significant dependence of the resonance positions on the orientation of the magnetic field. A more detailed description is required to understand this behaviour, but also to extract information on the structural and electronic properties of the species. The framework to achieve this goal is the description of the system by a so-called spin Hamiltonian (eqn (1)), which – for the current discussion – consists of two terms reflecting the electron Zeeman and hyperfine interactions.

µ_B denotes the Bohr magneton, B⃑, S⃑ and I⃑ are the vectors of magnetic field, electron spin and nuclear spin, respectively. The dependence of the resonance position on the angle between the surface and the magnetic field indicates anisotropic interactions, which imply that scalar values of g and A will not be sufficient to describe this system. Such anisotropic interactions are well known in physics and can be described by means of tensors. Eqn (1) gives the Hamiltonian for paramagnetic centres characterized by Zeeman and Hyperfine interaction. The corresponding g and A tensors have the form of 3×3 matrices as exemplarily shown in eqn (2) for the g matrix.

The g and A matrices are symmetric and can thus be diagonalized, which means that all off-diagonal elements vanish. The components g_xx, g_yy, g_zz and A_xx, A_yy, A_zz (or g₁₁, g₂₂, g₃₃ and A₁₁, A_22, A₃₃) of the diagonal matrices are also called principal components. It is important to note that these principal components correspond to a well-defined Cartesian coordinate system within the local framework of the paramagnetic centre. In general, symmetry is a powerful tool to predict the properties of the coupling matrices. Table 1 summarizes the relationship between g and A matrix components and their relative orientation with the point group symmetry of a paramagnetic site.³⁰ 

Information on the system can be deduced from the experiments by simulations, which allow extracting the principal components of the g- and A-matrices. There are different levels of sophistication depending on the size of the hyperfine interaction with respect to the Zeeman interaction. For systems with rather large hyperfine interaction such as Au atoms discussed here, it is not sufficient to consider the hyperfine interaction as a perturbation on the Zeeman interaction. This implies that the eigenvalue problem associated with the Hamiltonian of eqn (1) has to be solved. This is e.g. implemented in the freely available program package easy spin, a very powerful toolbox to simulate EPR spectra.³¹  Fits of the angular dependent EPR spectra using this program package were made and the results are shown as grey traces in the upper and lower plot of Fig. 1a. Which assumptions have been made to simulate the spectrum? First, it is important to realize that the spectrum contains four lines. Given the fact that the resonance positions are a function of the orientation of the centre with respect to the external magnetic field, all centres contributing to an individual EPR line behave similarly. Therefore, within one spectrum the principal components as well as the orientation of the coupling matrices are very similar for all centres. This is the typical scenario for a single crystal containing one well-defined paramagnetic species.

The principle components of the g and the A matrix extracted from the fits are shown in Table 2. In addition to the values one gets the orientation of the matrices with respect to the surface. It is clear that the latter information can only be extracted for macroscopically ordered systems such as planar single crystalline surfaces used here. It is found that one of the principal components is oriented perpendicular to the surface while the other two components lying in the surface plane are degenerated, which was shown by an independent experiment with an appropriately rotated crystal (not shown). The knowledge about the orientation of the coupling matrices gives the possibility to infer that the Au atoms contributing to the spectrum are located on the islands of the MgO(001) surface, because adsorption of Au atoms at structural defects such as steps, corners, or kinks would cause a tilt of the orientation of the magnetic interaction matrices away from the surface normal and in turn a significant change of the angular dependence of the spectra. It does not mean that Au atoms do not adsorb to step edges or corner sites, which they do as seen from the STM image (Fig. 1b) taken at 4 K on a single crystalline MgO(001) film, but these atoms do not contribute to the observed EPR signal, because of their comparably low abundance.

The behaviour of the signal amplitude is due to small differences in the hyperfine coupling constant within the ensemble. The interested reader is referred to the literature for details.²⁹ 

The situation on a single crystalline surface differs of course from that in a powder. Assuming the same adsorption behaviour, the EPR spectrum of such a powder material is considerably different, because the sample contains all orientations at ones, which were investigated separately in the single crystal case. Hence, the spectrum is an appropriately weighted superposition of the spectra for each orientation. Fig. 1c shows a simulation of the expected line shape neglecting the dependence of the hyperfine coupling constant on the adsorption sites. It is seen that each of the hyperfine lines from Fig. 1a is now a pattern determined by the g-anisotropy, where the low field maximum corresponds to g_⊥ and the high field minimum to g_‖ as indicated in Fig. 1c. In this particular case the situation is rather simple, because the hyperfine interaction is almost isotropic and large. Thus, the effect of the g-matrix anisotropy on the individual hyperfine lines is clearly visible. It is obvious that the spectrum interpretation of a powder sample can be much more complicated when both the g- and A-matrix are anisotropic and the resulting lines overlap. For a complex EPR spectrum some experience is required to deduce an appropriate spin Hamiltonian together with the matrix elements of the coupling matrices from a given line shape. In general it requires some knowledge about the system and possible paramagnetic centres in the first place.

One of the core results of the spectral analysis was the determination of the characteristic coupling parameters namely the anisotropic hyperfine and the Zeeman interaction. However, a very important question remains: What additional insight can be obtained from the values of the g and A matrices? The discussion of this crucial aspect is in general rather involved, but valuable insight can often be achieved based on a qualitative or semi-quantitative discussion of the expected electronic structure of the site under consideration, e.g. an analysis of the crystal field of a transition metal ion. Significant progress in this respect has been made in recent years by computational approaches mostly based on DFT methodologies, which are now capable to calculate g and A matrix components with sufficient precision for meaningful comparisons (see e.g. ref. 32–38). This has advanced the understanding of paramagnetic centres tremendously. Even though the following examples will be taken from the field of paramagnetic centres on solid surfaces, the impact is perhaps even more severe in the field of molecular or biological systems.

Information on the electronic properties of the system at hand is encoded in both the g- and the A-matrix. Conceptually, it is perhaps easier to start with the discussion using the hyperfine interaction. As mentioned above the hyperfine interaction is a symmetric (3×3) matrix, which can be diagonalized. This diagonal matrix has in general a trace, but it is possible to decompose the matrix into two parts: an isotropic, scalar part (a_iso) and an anisotropic dipolar part (T), which is a traceless (3×3) matrix. Mathematically this can be written as follows:

The reason for this decomposition is that these two parts of the hyperfine interaction can be associated with different physical effects. The isotropic hyperfine coupling constant is also known as the Fermi contract term and is due to the finite probability to find the electron at the nucleus. For Au atoms on MgO a large isotropic hyperfine coupling constant is expected, because the spin is predominately of s character (unpaired electron in the 6s orbital) and should thus have an appreciable density at the nucleus. While this is qualitatively expected, it is interesting to note that the isotropic hyperfine coupling constant of the adsorbed Au atoms is more than 50% smaller than the corresponding value measured for Au atoms in a rare gas matrix.³⁹  What is the reason for the reduced s electron spin density on the MgO surface? On the one hand the spin density itself could be reduced e.g. by partial charging of the Au atoms on the surface. On the other hand it is possible that the s character of the spin density is reduced. These two effects cannot easily be disentangled based on the experimental results alone. However, theoretical calculations can help to answer this question. Proper density functional calculations revealed that the charge transfer from or onto the Au atoms is small. One the other hand the shape of the spin density as shown in Fig. 2a is no longer spherical as expected for s orbitals, but the spin density and thus the wave function is polarized away from the surface as a result of the Pauli repulsion with the oxygen anions of the surface. In a simple orbital picture this implies that the corresponding wave function contains components with higher angular momentum l (p or d orbitals), which reduces the s contribution and, thus, the isotropic hyperfine interaction. From an energetic point of view this implies that the Au 6s orbital is destabilized due to the Pauli repulsion with the lattice oxygen ions of the MgO. This effect is not restricted to coinage metals on MgO, but has been seen before e.g. for alkali metals on MgO or coinage metals on alkali chloride surfaces.^40–42 

Hyperfine interaction extends also beyond the atom or atoms where the spin density is mainly localized. The hyperfine interaction found on more distant atoms is often called superhyperfine interaction (shf). To exemplify the kind of information, which can be accessed, Fig. 2b shows the largest EPR line of Au atoms observed for MgO grown with conventional oxygen (bottom trace), and the spectrum measured after growing the MgO(100) surface with ¹⁷O₂ (I=5/2; 90% enrichment) (top trace).²⁹  The single line is split into six lines and a small signal at the original position. The latter is due to Au atoms interacting only with ¹⁶O in the film. The observed intensity is about 10% of the original one. Given an isotopic enrichment of 90%, this indicates significant coupling of the Au atom to one oxygen ion only. This is in line with the sextet of lines, which can be readily understood by a coupling of the unpaired electron of the Au atom to one ¹⁷O in the film. Hence all adsorption sites having more than one equivalent oxygen neighbour such as the Mg cation site can be excluded. In particular, it is perfectly consistent with the preferred adsorption site according to theory being Au adsorbed on top of the oxygen anions of the film.^43,44 

Apart from the hyperfine interaction also the g-matrix contains valuable information. The extraction of this information is intimately linked to the electronic structure of the system at hand. For transition metal ions being a prominent class of paramagnetic centres in catalytic systems, crystal field theory is the basis of the qualitative and semi-quantitative discussions. The various situations found for transition metal ions have been investigated early on and there are excellent books and reviews summarizing their properties (e.g. ref. 30, 45 and 46). Here the information encoded in the g-matrix components will be exemplified using O₂^•− radicals on MgO(001). Adsorption of molecular oxygen at 30 K on a 4 ML thick MgO(001) film grown on a Mo(001) single crystal leads to the spontaneous formation of O₂^•− radicals characterized by the angular dependent EPR spectra shown in Fig. 3a.⁴⁷ 

The analysis of the presented spectra together with spectra taken at a different azimuthal orientation of the (001) surface with respect to the magnetic field (not shown) reveal an orientation of principal components of the O₂^•− g matrix along the surface normal and the [110] equivalent directions within the (001) surface (the third one is orthogonal on the other two). This proves that the radical is adsorbed on the terraces of the MgO(001) islands and aligned with [110] equivalent directions (Fig. 3b). This is in perfect agreement with theoretical predictions for the adsorption site.^48,49  More important for the present discussion are the values of the g-matrix elements, in particular the g_zz component. For this discussion O₂^•− radicals on the thin films are compared to O₂^•− centres on MgO powder. The radicals do not form spontaneously on the stoichiometric surface of the powders, but extra electrons are required as introduced e.g. by alkali metal atom doping.^50–53  A detailed analysis of the powder data in comparison with theoretical calculations revealed that the g_zz component of the matrix strongly depends on the adsorption site. In particular, radicals adsorbed on morphological defects such as edges or corners show a significant reduction of the g_zz component as compared to the regular terraces site (Table 3). The reason is that for such a 13 electron radical the g_zz component of the molecule can be given to first order by the following expression:⁵⁴ 

in which λ is the spin orbit coupling constant and Δ is the energy difference between the singly and the doubly occupied π* orbitals as depicted in Fig. 3c. The equation also reminds on the fact that the origin for the deviation of the g matrix components from the free electron values is due to spin orbit interaction. The reduction of the g_zz component with a reduction of the local coordination of the adsorption site on an ionic crystal is directly associated with an increase of the local electric field at the adsorption site. An increased electric field leads to an increased splitting between the π* orbitals of the O₂^•− radical and hence to an increase of Δ. In turn, the g_zz component will be found closer to the value of the free electron, which is the explanation for the experimental observation made on powders. What is the reason for the reduced g_zz values measured for O₂^•− radicals on terraces of the thin film compared to terrace sites on the MgO powders? The explanation is intimately related to the question of the stability of the O₂^•− radicals on the thin MgO film. Theory predicts that the electron transfer from the metal substrate (Mo(001)) through the MgO film onto the oxygen molecule is stabilized by a so-called polaronic distortion of the MgO lattice, an effect considered important not only for molecules but also for metal adsorbates with high electron affinity such as Au.^48,49,55  This means that the ions underneath the O₂^•− radical are pulled out of their regular lattice positions as indicated in Fig. 3d which, according to theory, is the important mechanism to stabilize the charge transfer state. The polaronic distortion gives rise to an increase of the electric field encountered by the oxygen molecule on the surface and thus a reduction of the g_zz component of the g matrix. EPR spectroscopy provides the first experimental evidence for the existence of the polaronic distortion for such a system, which is difficult to observe experimentally for such systems.

The paramagnetic Au atoms discussed above were considered as static entities and at a measurement temperature of 30 K this assumption is justified. However, catalytic systems are often investigated at elevated temperatures or molecular catalysts are studied in the liquid phase. For such systems the assumption that the paramagnetic species are static needs to be revisited. The first important question concerns time scales. To this end there is no definite answer. The time scale depends on the actual experiment. For cw-EPR spectroscopy at X-band (ν≈10 GHz) the rigid limit assumption is valid if the rotational correlation time is longer than about 100 ns. The effect of rotational motion may be illustrated using the EPR spectrum of vanadyl(iv) species characterized by a spin of 1/2 and nuclear spin of the vanadium of 7/2. Figure 4a shows the EPR spectrum of a frozen solution of vanadyl(iv)acetylacetonate (c=5×10⁻⁴ mol l⁻¹) in toluene at 77 K.⁵⁸ 

The spectrum is in line with expectations for a spin 1/2 species with anisotropic hyperfine as well as anisotropic Zeeman interaction. Similar line shapes are also found for VO²⁺ species on solid surfaces such as titania particles loaded with vanadium presented in Section 5.1. As discussed above, the line shape is due to a superposition of lines arising from centres, which have a different orientation with respect to the magnetic field for the different nuclear spin states. This picture implies that the orientation of the species is stationary on the characteristic time scale of the EPR spectroscopy. When the orientation changes on the time scale of the experiment, the resonance condition and hence the shape of the spectrum will change. To illustrate the effect it is easiest to consider the extreme of fast reorientation dynamics on the time scale of the EPR experiment. In such a case the anisotropy of the matrices can no longer be probed by the experiment and the resulting spectrum is determined by the isotropic g-value g_iso=(1/3(g₁₁+g₂₂+g₃₃)) as well as the isotropic hyperfine interaction constant (a_iso, see eqn (3)). For the vanadyl species one expects an eight line spectrum centred at g_iso with a splitting of a_iso between the lines within the so-called high field approximation. Figure 4b shows the spectrum of the same sample measured at 236 K.⁵⁸  Qualitatively, the spectrum is in line with the expectations. However, the line width and thus the amplitude of the lines are different for the different nuclear spin states. The reason for the different line width is largely due to relaxation effects, which are still present even if the anisotropy of the interaction is already averaged out. While the fast rotational limit as well as the static limit can be treated rather simply, the region of slow motion, in which most of the motional averaging of the matrix anisotropy occurs, is rather difficult to treat theoretically. One approach based on a stochastic Liouville equation (SLE), which can address rotational motions within diffusion models, has been developed by Freed et al.^59,60  In essence, the signal will undergo drastic changes in the line shape between the rigid and the fast rotational limit. This effects both, the line position as well as the line width for the different manifolds. Application of the SLE approach to simulate the effect of rotational diffusion on the vanadyl system discussed here, indicates that changes of the line shape occur for rotational correlation times smaller than 10 ns while a spectrum as shown in Fig. 2b corresponds to a rotational correlation time of about 0.1 ns.

While paramagnetic centres in bulk solids and low molecular weight molecules in solution can be easily divided into the rigid or the fast rotational limit, respectively, a back of the envelop calculation using the Stokes–Einstein equation reveals that objects of a few nanometer in diameter will have rotational correlation times in the slow motion regime of cw-EPR at X-band. Paramagnetic molecules adsorbed on surfaces may also exhibit rotational correlation times, which lead to appreciable changes in the EPR line shape.⁶¹  With respect to the spectral analysis one should note that in case of dynamic effects a determination of the characteristic magnetic interaction parameters requires particular care and a spectrum taken at a certain temperature may only give apparent magnetic parameters. This should be kept in mind when comparing magnetic parameters with literature values to assign the observed signals to species.

It was already mentioned that hyperfine interaction is an important source of information in EPR spectroscopy. However, this information cannot always be extracted from cw-EPR spectra straightforwardly. For real catalysts, which are much more complex materials than the ideal model systems discussed above, the assignment is usually more difficult. The problem is often caused by a lack in spectral resolution either due to overlapping lines or due to splittings smaller than the line width of the cw-EPR signal. Additional information on such systems can be obtained using pulse spectroscopic techniques. Coarsely, the pulse techniques addressing interaction of electrons to adjacent nuclei can be subdivided into experiments relying on nuclear modulation effects and double resonance experiments called electron nuclear double resonance (ENDOR). Such studies go back to experiments made in the 1960's which have shown that the decay of primary electron spin echoes are modulated by frequencies corresponding to nuclear frequencies as well as their differences and sums.^62,63  The term ESEEM (electron spin echo envelop modulation) was coined for these experiments. ENDOR was first demonstrated by Mims in 1965 and Davies added a second important variant of it in 1974.^64,65  Both techniques have their advantages and disadvantages, but it is not possible to discuss them in detail here.

In the following we will discuss two examples using ESSEM or advanced variants of it. There are several aspects both on the experimental as well as on the analysis side of ESEEM experiments, which need to be considered and the interested reader is referred to the appropriate literature (see e.g. ref. 66 and 67). As a starting point consider TiO₂ nanoparticles, which play an important role in a variety of technical applications. Photocatalytic processes are one of them and the fate of the electron hole pairs to be separated after the photoexcitation event is crucial for an understanding of these processes. In TiO₂ both the “electrons” and the “holes” are associated with paramagnetic states. To illustrate the capabilities of the method we will focus on a very special question namely the environment of the hole and the electron center created subsequent to the photoexcitation. From cw-EPR spectra it is readily inferred that the hole centers can be associated with oxygen based radicals, which give rise to characteristic EPR signals well separated from the EPR signals of the electron centers associated with Ti(iii)-centres. Apart from the g matrix anisotropy the spectrum does not reveal any additional splitting, which could arise from the coupling to other nuclei. A 2-pulse ESEEM experiment schematically shown in the inset of Fig. 5a was performed at 7 K by setting the magnetic field to an absorption line of the O^•− centers and the Ti(iii)-centres, respectively. The Fourier transform of the echo intensities as a function of the delay time τ between the pulses reveals a signal at 14.9 MHz for the hole centre but no signal for the corresponding electron centre.⁶⁸  The signal is very close to the nuclear Larmor frequency of hydrogen at the field strength used. This proves the oxygen centered radical to be located in the vicinity of protons (OH groups), typically on the surface of the TiO₂ nanoparticles. However, the hyperfine coupling is weak, suggesting that the paramagnetic centres have a certain distance from the coupled protons. On the other hand the electron related signal located on titanium sites does not show indications for proton coupling, hence these centers have a significantly larger distance to the surface OH groups.

In case of a single weakly coupled nucleus the analysis of the ESEEM traces is straightforward, however, in case of more than one coupled nucleus and different coupling constants the analysis can become rather intricate and one would like to enhance the spectral information content. A well-established strategy in this respect is the use of correlation spectroscopy, namely the correlation of the different hyperfine levels present in such an ESEEM experiment, which allows to simplify the analysis of the ESEEM experiment. The corresponding experiment first performed by Höfer et al. is called HYSCORE (hyperfine sublevel correlation spectroscopy).^69,70  To illustrate the ability of this approach, potassium atoms adsorbed to the surface of MgO powders will be considered. Figure 5b shows a HYSCORE experiment taken at 10 K for potassium atoms adsorbed on high quality ¹⁷O enriched MgO powder (enrichment appox. 10%).⁴⁰  The cw-EPR spectrum shows a clear coupling of the potassium atom to one oxygen atom as we have seen, too, in case of the Au atoms. A close look at the HYSCORE spectrum reveals at least three oxygen atoms couple to the potassium atom. Two of these interactions are found in the (−,+) quadrant of the spectrum shown on the left as pairs of cross peaks located at (−5.8,1.7) and (−1.7,5.8) MHz, and at (−4.2,1.1) and (−1.1,4.2) MHz. These two sites show a rather large hyperfine interaction – a condition for which the two cross-peaks should be separated by approximately twice the nuclear Zeeman frequency (ν_O). A third signal is found in the (+,+) quadrant located at approximately the nuclear Zeeman frequency. While this indicates that the coupling to the latter oxygen ion is small, one should note that the signal is elongated perpendicular to the diagonal of the quadrant (width about 1.2 MHz), which indicates that a sizable hyperfine coupling exists even for these oxygen sites. With the knowledge that the atom couples to three different oxygen ions in the lattice and the aid of theoretical calculations, it is possible to deduce the adsorption site of the potassium atoms on the MgO surface to be a so-called reverse corner site.⁴⁰  These examples show, that more advanced EPR techniques are able to provide additional information, which go beyond conventional cw-EPR spectroscopy. Even though the two examples discussed above focus on the environment of the paramagnetic center, the additional information is not restricted to these questions. With respect to catalytic questions it has to be borne in mind that pulse spectroscopic experiments are mostly done at low temperature, which renders experiments under in situ conditions difficult.

As mentioned in the introduction, supported VO_x/TiO₂ oxides catalyse, e.g., the oxidative dehydrogenation of propane to propene as well as the oxyhydrative scission of 1-butene and n-butane to acetic acid. In the latter reaction, a strong beneficial effect of surface sulfate was found, which remained in the titania support from its synthesis via hydrolysis of TiOSO₄. A reference catalyst prepared in the same way with the same loading of vanadium but on a sulfur-free TiO₂ support showed a significantly lower activity.¹⁷  Depending on the vanadium content, supported vanadia catalysts can contain a variety of different V species, ranging from single vanadyl surface sites in different coordination (distorted tetrahedral or distorted square-pyramidal/octahedral) via small oligonuclear V_xO_y clusters to V₂O₅ (nano)crystallites. However, not all of them are detectable by in situ EPR at elevated temperatures. These comprise diamagnetic V(v)-species (S=0) as well as V(iii) species (S=1), due to their large zero-field splitting and/or short relaxation times. Fast spin relaxation is also the reason why V(iv) (S=1/2) in tetrahedral environment can usually not be detected temperatures relevant for catalytic reactions. Therefore, only V(iv) in distorted square-pyramidal/octahedral coordination is accessible by EPR under reaction conditions. In supported VO_x/TiO₂ catalysts, such V(iv)-species can exist as single sites or as clusters of different nuclearity on the surface of the support, the smallest possible unit being a dimer (Fig. 6).

Single V(iv)-centres in distorted square-pyramidal/octahedral symmetry (Fig. 6, left) give rise to a complex signal (Fig. 7). As already mentioned in Section 3.2, this signal arises from the so-called hyperfine structure (hfs) coupling of the electron spin with the nuclear spin of V (I=7/2) which splits the signal for the electron spin transition into 2I+1=8 subsignals separated to a first approximation by the hfs coupling constant. The isolated vanadyl site has C_4v symmetry with the fourfold axis being aligned with the short VO double bond. In turn, both Zeeman and hyperfine interaction show axially symmetric tensors (s. Table 1) with the so-called parallel components (g_‖=g₃₃, A_‖=A₃₃) being aligned with the VO double bond. In a powder, there is a random orientation of the V-centres with respect to the B₀ field direction. For a given orientation a set of eight lines (for the eight nuclear spin states) positioned according to the appropriate g- and A-tensor components is expected and the observed powder spectrum is the appropriately weighted superposition of the spectra for all orientations. The principle components g_‖, g_⊥, A_‖ and A_⊥ of the g and A matrices (Fig. 7) can be determined readily from this spectrum. The two octets corresponding to the orientation of the B₀ field parallel and perpendicular to the VO bond (Fig. 6) are marked on top of Fig. 7. By spectra simulation (dotted lines in Fig. 7), precise values for the g and A tensor components can be derived which sensitively reflect changes in the local environment of the respective single VO²⁺ sites. This will be demonstrated below.

When the distance between neighbouring VO²⁺ sites decreases, e.g. in small clusters or partly reduced V₂O₅ crystals (Fig. 6, middle and right), dipole–dipole and spin–spin exchange interactions come into play which broaden the lines. In the latter case, a single line at the average g value g_av=(2 g_⊥+g_‖)/3 is observed in which hfs is no longer resolved. The reason is that for a given spin, this exchange causes a fast change of the orientation of neighbouring spins and, thus, a local fluctuating field. When the rate of this fluctuation is in the order of the resonance frequency, g and A splittings are no longer resolved. These effects give rise to a more or less broad isotropic singlet, which superimposes on the hfs multiplets of isolated VO²⁺ sites. This effect can be seen particularly pronounced in Fig. 7 for elevated temperatures.

As mentioned above, V(v), V(iv) in distorted tetrahedral geometry as well as V(iii) cannot be seen in in situ EPR experiments. This means that only a part of the V sites in the whole sample is detected by EPR, which calls for the application of other techniques particularly sensitive for EPR-silent species such as pentavalent vanadium. Therefore, the reactor used for operando EPR studies of supported VO_x/TiO₂ studies has been coupled with fibre optical sensors for simultaneous acquisition of Raman and UV-vis diffuse reflectance spectra. A detailed description of the experimental setup as well as the Raman and UV-vis spectra is given elsewhere.¹⁸  In this chapter, we focus on the EPR results only.

The catalyst used in this case study for oxidative dehydrogenation (ODH) of propane was prepared by thermal spreading of V₂O₅ on a commercial anatase support. It contained 6 wt% of vanadium in the form of single vanadyl sites, small oxide clusters and even some V₂O₅ nanocrystals (too small to be seen by XRD but visible by Raman spectroscopy).¹⁸  In terms of catalytic performance, this material does not belong to the best catalysts for propane ODH, yet it has been selected since the response of the different V species can be studied beneficially in parallel during reaction.

In the EPR spectrum at room temperature after thermal pre-treatment at 450 °C in air, just a small signal of some residual VO²⁺ species is seen (Fig. 7). However, when the sample is exposed at room temperature to the reactant gas mixture containing propane and O₂, the total intensity raises and two hfs signals S1 and S2 with different g and A tensor parameters (Table 4) are resolved suggesting, remarkably, that even under those mild conditions single V(v) sites are reduced to single VO²⁺ species. In the corresponding Raman and UV-vis spectra¹⁸  (not shown), this is reflected by the disappearance of the isolated V(v)O vibrational band and by an increase of light absorption in the range being characteristic for d–d transitions of reduced V species. Upon rising the temperature to 250 °C, these effects become more pronounced, leading to a clear separation of signals S1 and S2 and to a significant intensity gain of a broad isotropic background signal arising from reduced V_x⁴⁺O_y cluster species. At even higher temperatures between 250 and 450 °C the total intensity decreases and the S2 signal of isolated sites vanishes completely. This is seen even more clearly in the spectrum of the catalyst after stopping the reaction and cooling to room temperature again. In this spectrum, only the hfs multiplet S1 remained, suggesting that the VO²⁺ species reflected by signal S2 might have been reduced to V(iii) during the course of reaction, which is not detectable under the chosen measurement conditions, due to short relaxation times and/or too high zero field splitting.

In general, the intensity (double integral of the EPR signal which is recorded as first derivative) is a measure for the number of spins contributing to it. However, according to the Curie–Weiss law this signal intensity depends inversely on temperature. Assuming a Curie–Weiss behaviour the product of intensity and temperature (here normalized to I×T_ref with T_ref=20 °C) plotted against the temperature should give a horizontal line. With this assumption the observed deviation from the horizontal line is directly proportional to the change in the content of EPR active V(iv) sites (Fig. 8).

From Fig. 8 (filled symbols) it is readily evident that the concentration of V(iv) formed by reduction of initial V(v) increases in the range between 20 °C and 250 °C and then remains constant. Interestingly, a very similar behaviour was observed for the absorbance measured by simultaneous UV-vis spectroscopy at a wavelength of 800 nm in the range of d–d transitions of reduced V species (Fig. 8, open symbols). Comparison with the catalytic performance measured by on-line gas chromatography at the reactor outlet suggests that the reduction of V(v) to V(iv) on the catalyst surface (Fig. 8a) in the initial period of the reaction goes along with an increase in propene selectivity – a beneficial effect that may be related to the lower reduction potential of V(iv) in comparison to V(v) which might suppress total oxidation of propene to CO_x. By comparing the UV-vis data in Fig. 8a with those of a coupled TPR/UV-vis experiment in which the absorbance difference at 800 nm was related to the consumption of H₂ and, thus, to the O/V ratio, a lower limit of 4.86 has been derived for the mean vanadium valence state under reaction conditions.¹⁸  This indicates that the majority of the V species remains pentavalent during reaction and probably only those directly exposed to the feed on the surface are getting reduced.

A detailed inspection of the operando EPR spectra in Fig. 7 reveals that two different types of isolated VO²⁺ species S1 and S2 exist on the catalyst surface, reflected by two different sets of spin Hamiltonian parameters (Table 4), besides a broad isotropic background signal superimposed on the hfs signals in Fig. 7, which arises from magnetically interacting VO²⁺ species. From the spin Hamiltonian parameters A_‖, Δg_‖/Δg_⊥ (calculated with Δg_‖=g_‖−g_e and Δg_⊥=g_⊥−g_e and g_e=2.0023) and β₂*² (eqn (5)),⁷¹  more information on reaction-dependent changes of the surface V sites can be obtained.

In eqn (5) P is the term for the dipole–dipole interaction of the magnetic moment of the electron and the nucleus. For the free V(iv) ion, P=184.5 G was obtained using eqn (6), in which g_e and g_N are the free electron and nuclear g factor, μ_e and μ_N are the Bohr and nuclear magneton and r_av is the average distance between electron and nucleus.⁷² 

β₂*² is the so-called in-plane π-bonding coefficient. It is a measure for the delocalization of the single electron from V to the equatorial O ligands, that means, for the degree of covalence of these V–O bonds. The delocalization occurs via in-plane π-bonding of the d_xy orbital containing the unpaired electron with the π orbitals of the basal ligands. β₂*² is equal to one for a pure VO²⁺ ion. This is for example the case in the EPR spectrum of a diluted frozen solution of a vanadyl salt (e.g. of VOSO₂ containing [VO(H₂O)₅]²⁺ cations). β₂*² decreases with rising electron delocalization towards the ligands, i.e. with increasing covalence of the basal V-ligand bonds.⁷³ 

A_‖ characterizes the strength of the VO bond. The shorter this bond, the higher is the value of A_‖. This has been observed for three differently distorted VO²⁺ species deposited on the surface of alumina.⁷¹  From an analysis of the ν(VO) stretching frequency it was concluded that the VO bond strength decreases with rising electron donation ability of the in-plane ligands.⁷⁴  A similar effect was found for the hfs constant, showing that A_‖ decreases as well with rising donor strength of the in-plane ligands.⁷² 

The overall distortion of the VO²⁺ species is reflected by the ratio Δg_‖/Δg_⊥. The higher this ratio, the more distorted is the site, i.e., the shorter is the VO bond and the longer are the V–O single bonds in the equatorial plane (Fig. 6, left). Inspection of Table 4 shows that the A_‖ and Δg_‖/Δg_⊥ values of V sites S1 are markedly lower than those of sites S2, indicating that the VO bond is longer and the overall distortion of this kind of VO²⁺ sites is lower in comparison to sites S2. Moreover, the coefficient β₂*² is lower for sites S1, which means that the equatorial V–O bond might be more covalent than those of sites S2.

As mentioned above, V sites S1 are only formed on sulfate-containing TiO₂ while sites S2 are present on both sulfate-free and sulfate-containing TiO₂ supports. This suggests that V sites S1 might be connected via oxygen bridges to surface sulfate while sites S2 contain V–O–Ti bridges only. The more covalent character of a V–O–S bridge in comparison to a V–O–Ti bridge could be understood in terms of the lower electronegativity difference between O and S in comparison to O and Ti.

During ODH of propane, the EPR intensity of V site S2 rises in the initial period up to 250 °C and vanishes gradually at higher temperature (Fig. 7). This is most probably due to a stepwise reduction from V(v) via V(iv) to EPR-silent V(iii) which, on the other hand, suggests that connection via V–O–S bridges could stabilize V sites S1 against deep reduction to V(iii) and, thus, retain their catalytic activity. In certain oxidation reactions (e.g. in selective oxidation of butane to maleic anhydride) it has been shown that V(iii) is catalytically inactive.⁷⁵  The last reduction step to V(iii) does obviously not occur for site S1. This suggests that surface sulfate might have a stabilizing impact on V sites in catalytically active V(iv) and V(v) valence states and could explain why supported V₂O₅ was markedly more active in oxyhydrative scission of butane to acetic acid when sulfate-containing TiO₂ was used as support.

In summary, this application example demonstrates the potential of operando EPR for identifying different V sites and their relation to catalytic performance in a supported vanadia catalyst during selective oxidation of hydrocarbons. Such reactions are known to proceed via a Mars–van Krevelen mechanism in which nucleophilic oxide species from the catalyst lattice react with the organic substrate and the temporarily formed anion vacancies are refilled by incorporation of electrophilic oxygen from the gas phase. This process is not supposed to imply organic radical intermediates but a redox cycle of the active vanadium species. Although only a certain part (namely the paramagnetic VO²⁺ sites) and not all V sites can be monitored by EPR, these species can serve as probes for structural changes in the sample. However, this intrinsic limitation illustrates also the added value that can be derived from such in situ studies when several complementary techniques are coupled in the same experiment.

Photocatalytic water splitting is another instructive example for redox catalysis. It implies transfer of electrons from negatively charged oxygen to protons forming gaseous hydrogen and oxygen (Scheme 1). Clearly, from an application oriented point of view, the development of photocatalysts which promote simultaneous stoichiometric evolution of O₂ and H₂ from water upon irradiation is desirable. However, up to now, the number of catalysts promoting both partial reactions, water oxidation and reduction (Scheme 1), simultaneously is still very limited.^76,77  There are many more catalysts known for water reduction than for water oxidation. Moreover, for deriving mechanistic details, a separate study of both partial processes is more suitable. However, this requires the use of sacrificial reagents to donate electrons for water reduction (e.g. amines or alcohols) or to scavenge electrons from water oxidation (e.g. Ce salts or iodates).

In recent years a strong increase in the development of new catalysts for both homogeneous and heterogeneous photocatalytic water splitting can be observed. But surprisingly, the use of in situ EPR spectroscopy for elucidating mechanisms is virtually unknown, although it might be promising for the detection of the one-electron processes occurring in such systems.

In homogeneous water reduction catalysis, some systems with high turn-over numbers (TON) have been developed,⁷⁸  however, detailed mechanistic information verified by spectroscopic evidences is widely missing.^79,80  Recently, Beller and co-workers described an efficient catalytic system based on [Fe₃(CO)₁₂] as water reduction catalyst and [Ir(ppy)₂(bpy)]PF₆ as photosensitizer (IrPS, Scheme 2).^81,82  In this system, charge separation is generated by excitation of the IrPS and subsequent reduction of its excited state by a sacrificial reductant (triethylamine (TEA), cycle I). From the reduced state of the IrPS, an electron is transferred to the iron water reduction catalyst, which itself reduces aqueous protons to H₂ (cycle II). Recently, mechanistic studies including coupled in situ EPR/Raman spectroscopy combined with in situ FTIR spectroscopy and DFT calculations were performed.⁸³  Here we focus on the EPR results in more detail. The formation and the nature of the reduced iridium species formed as well as comprehensive studies on the iron catalyst cycle, including activation and deactivation of the iron species are discussed.

In situ EPR spectra were recorded in X-band using a rectangular cavity with a grid in the front side. For irradiation with light, the beam of a 300 W Xe lamp that simulates the spectrum of sunlight was focused through the grid on the sample within the cavity. For EPR/Raman measurements, the light beam of the Xe lamp was focused with an optical fibre in an angle of 60° while the laser beam of a fibre optical Raman spectrometer was focused on the sample in an angle of 90° (Fig. 9). EPR measurements of the initial IrPS and the reduced IrPS⁻ were performed at 27 °C and −183 °C. The EPR measurements of the reaction mixture of Fe-WRC and IrPS were performed at −73 °C. At this temperature isotropic signals of all iron radicals with minimal linewidth are observed. Rising temperature leads to an increase of the linewidth, which may be due to a shortening of the relaxation time. The exact reaction conditions can be found elsewhere.⁸³  In the following the two cycles of the water reduction cascade (Scheme 2) are considered separately.

As mentioned above, a reduced iridium complex should be formed by quenching of the excited state of the IrPS by TEA. Regarding the electron density, two possibilities must be taken into account for the electron location: (a) reduction of the bpy ligand leading to the formation of Ir(iii) and a bpy radical anion or (b) reduction of the metal centre to form Ir(ii) with a neutral bpy ligand. Both possibilities are discussed quite controversially in literature.^84,85 

Indeed, irradiation of a IrPS solution containing THF/TEA/H₂O in a ratio of 8/2/1 in the absence of [Fe₃(CO)₁₂] at 27 °C gives rise to an intense isotropic signal at g=1.9840 (Fig. 10a). At −183 °C the isotropic signal turns into an anisotropic line with strong axial distortion (Fig. 10b), due to frozen mobility of the complex. This spectrum can be reproduced by spectra simulation with the Sim14S program⁸⁶  using a total spin of S=1/2, an axial g tensor with g_⊥=2.0027 and g_‖=1.9498 and a line width of ΔB_⊥=13.4 G and ΔB_‖=35.1 G (Fig. 10c). A similar signal was neither formed in pure THF nor in THF/H₂O, suggesting that the presence of TEA as a reducing agent is needed and excitation by light is essential to initiate the electron transfer.

DFT calculations in combination with UV-Vis and XANES spectroscopy revealed that in this system the electron is mainly located in the bpy ligand with a distinct delocalization to the iridium center.⁸⁷  This partial delocalization leads to larger spin–orbit coupling and to a larger deviation of the g factor from the free electron value.

Upon electron transfer to the iridium photosensitizer, the sacrificial reagent TEA should be converted to a N^•+(CH₂CH₃)₃ radical cation with the unpaired electron located at the N atom. Such a radical species would give rise to an EPR spectrum with hyperfine splitting (hfs) being characteristic of the coupling of the electron with the nuclear spin of ¹⁴N (I=1) and six ¹H (I=1/2) from the three α-CH₂ groups.^88–91  Surprisingly, not this species but another radical without any hfs from nitrogen was observed (Fig. 11). It has been identified as a R₂NCH₂C^•H₂ radical (R=Et or H) by spectra simulation being consistent with coupling of the single electron on the β-C atom with two sets of two equivalent protons on the α- and β-C atoms. Considering the stability of the radicals of tertiary, secondary and primary amines (at the N, α- and β-C), this signal was assigned to the primary H₂NCH₂C^•H₂ radical or the H₂N⁺CH₂C^•H₂ radical cation which is formed as a result of oxidative degradation of TEA, during which other short-lived radical intermediates may be passed.^83,92 

The spectrum has been satisfactorily reproduced by assuming g=2.0017 and hfs by two sets of two equivalent protons (2H, A_H1=28.3 G, ΔB_H1=2.4 G and 2H, A_H2=22.3 G, ΔB_H2=2.4 G, Fig. 11b), while hfs from the N nucleus is missing. The assignment was supported by the use of deuterated d₁₅-TEA. Due to a strong kinetic isotope effect of deuterium, a decrease of the reaction rate was observed and the corresponding H₃N⁺(-CD₂-C^•D₂) radical can be detected at −73 °C at even higher Ir–PS concentration (Fig. 11c). The experimental spectrum has been simulated using g=2.0017, 4×I=1, A_Dα=A_Dβ=3.6 G, ΔB=0.9 G with A_Dα and A_Dβ being the hfs constants for the α- and β-D (Fig. 11d). It can be seen that the A_D hfs values are approximately by a factor of 7 smaller than the corresponding A_H values. Given that this should also be true for the difference between A_Dα and A_Dβ, the latter should be less than 1 G, which is obviously not resolved in the spectrum.

The Fe-WRC [Fe₃(CO)₁₂] contains low-spin Fe(0) which has a 4s² 3d⁶ configuration with no unpaired spins and is therefore diamagnetic. However, when this Fe-WRC and the IrPS are solved together in a solution containing THF/TEA/H₂O in a ratio of 8/2/1, three EPR signals appeared in a ratio of 1 : 66 : 32 (Fig. 12). These signals are assigned to the radicals [Fe₃(CO)₁₂]^•−, [Fe₃(CO)₁₁]^•−, and [Fe₂(CO)₈]^•−, which are formed upon one-electron from the reduced IrPS. The assignment of the EPR signals is based on comparison of their g values with literature g values of known iron carbonyl anions.^93–96  In these previous studies, such radical anions were prepared by reduction of Fe(CO)₅ solutions with alkali metals, followed by controlled reoxidation of the resulting diamagnetic iron carbonylate species with AgBF₄. Here, for a reliable identification of the different radical anion species, the same reactions were performed with labelled ⁵⁷Fe(CO)₅ (I_Fe=0.5) and Fe(¹³CO)₅ (I_C=0.5) and the super hyperfine splitting (shfs) patterns arising from the coupling of the single electron with the nuclear spins of ⁵⁷Fe and ¹³C were analysed. Spin concentration determined using TEMPO (2,2,6,6-Tetramethylpiperidine-1-oxyl) as a spin standard indicated that 85% of the total Fe content in solution is present as such radicals.

Upon UV-vis irradiation, the three radical signals disappeared quickly and a new triplet at g=2.0432 with a hyperfine coupling constant A_H=22.3 G arises. This triplet is attributed to the [H₂Fe₂(CO)₇]^•− radical anion being a deactivation product of the Fe-WRC as discussed below. The triplet structure arises from the shfs coupling of the single electron with the nuclear spins of two equivalent protons in the species [H₂Fe₂(CO)₇]^•−. The observed shfs constant A_H=22.3 G agrees very well with that observed for the radical [HFe₂(CO)₈]^•.⁹⁵  This radical, however, comprises only ∼3% of the total iron content. Nevertheless, it has been shown that the catalytic system is still active under these conditions.⁸¹  This indicates that the active species in cycle II might be EPR silent, which evidences the intrinsic limitation of this method for catalytic in situ studies. To overcome this drawback, the concept of coupling EPR with other spectroscopic techniques, as realized previously for heterogeneous catalytic gas phase reactions,¹⁸  has been adapted for photocatalytic water splitting. By this new two-in-one EPR/Raman spectroscopy, supported by separate in situ FTIR measurements and DFT calculations, the active species in cycle II has been identified as a diamagnetic [HFe₃(CO)₁₁]⁻ anion.⁸³ 

When a [HNEt₃][HFe₃(CO)₁₁] complex, in which this anion is already existing, was used instead of [Fe₃(CO)₁₂] as WRC, the reaction was running equally well and the same triplet like in Fig. 12 was observed after extended reaction time. It must be mentioned that during catalytic tests, catalyst deactivation was observed with time, which was accompanied by CO deliberation. This suggested that decomposition of the [Fe₃(CO)₁₂] WRC catalyst might be the major reason for deactivation. Therefore, a detailed study of the behaviour of the WRC catalyst in the presence of different reaction mixture components with and without light irradiation was performed.

The [Fe₃(CO)₁₂] WRC catalyst is sensitive to light already in the absence of water and IrPS. When a solution of [Fe₃(CO)₁₂] in THF/TEA 8/2 is monitored by EPR under light irradiation, the radicals [Fe₃(CO)₁₂]^•− (g=2.0016), [Fe₃(CO)₁₁]^•− (g=2.0497), [Fe₂(CO)₈]^•− (g=2.0385) and [Fe₄CO₁₃]^• (g=2.0134) are observed (Fig. 13).^93–96  The same species, though with slightly changed intensity ratios, are also formed in the presence of water. A possible formation pathway of these species is given in Scheme 3.

The coordination sphere of iron carbonyl radical anions is quite labile and capable of fast ligand and electron exchange. This enables interconversions of radical ion clusters. The reaction of [Fe₃(CO)₁₂] with Lewis bases such as THF and TEA generates the unstable electron rich radical [Fe₃CO₁₂]^•− under light irradiation, which is transformed into the more stable electron deficient species [Fe₃CO₁₁]^•− by decarbonylation. When the IrPS is present, the latter species is quickly transformed by electron transfer to the diamagnetic [HFe₃(CO)₁₁]⁻ (detected by FTIR and Raman spectroscopy⁸³ ), which itself undergoes decomposition, finally leading to deactivation of the [Fe₃(CO)₁₂] WRC catalyst within 24 h. A possible deactivation pathway of the Fe-WRC via dinuclear iron complexes is proposed in Scheme 4. In the first step [HFe₃(CO)₁₁]⁻ decomposes to [Fe₂(CO)₈]^•−. The latter species can also be formed directly from [Fe₃(CO)₁₂]^•− as shown in Scheme 3. It is further transformed into complex [Fe₂(CO)₇]^•− by decarbonylation. Note that gaseous CO has been experimentally detected in the system with time. By two one electron transfer steps involving IrPS and TEA combined with protonation reactions, the corresponding dihydride [H₂Fe₂(CO)₇]^•− is formed, which is detected by EPR. Loss of hydrogen and CO coordination can principally lead to regeneration of [Fe₂(CO)₈]^•−. These steps are in good agreement with the observed EPR signals. The deactivation cycle in Scheme 4 is particularly favoured in the presence of UV light. When a 420 nm cut-off filter is used to eliminate the UV light, [Fe₂(CO)₈]^•− and [H₂Fe₂(CO)₇]^•− are hardly detected at extended irradiation times, yet production of hydrogen was still observed under the same conditions in catalytic tests.⁸²  Thus, it can be excluded that both dinuclear iron complexes are essential for catalytic hydrogen production.

By selecting two case studies from heterogeneous and homogeneous catalysis, it has been demonstrated that in situ EPR spectroscopy is indeed a powerful technique to investigate structure-reactivity relationships of catalytic redox processes, which include the transfer of single electrons, such as selective oxidation of hydrocarbons or photocatalytic water splitting. Particularly detailed information could be obtained on structural and electronic changes of supported VO²⁺ sites in selective propane oxidation over V/TiO₂. Supported by spectra simulation VO²⁺ sites in different geometrical environment have been identified and their stability under propane oxidation conditions was probed. However, it has also been emphasized that it is frequently very beneficial to couple in situ EPR spectroscopy with other techniques that can compensate for the intrinsic limitation of EPR, namely its restriction to paramagnetic species only. Especially the combination with UV-vis and vibrational spectroscopy as well as DFT calculations has proven to be powerful for mechanistic studies. In the case of homogeneous photocatalytic water splitting, in situ EPR spectroscopy provided essential information on the nature of different intermediates formed during operation and deactivation of the iron carbonyl water reduction catalysts. In combination with complementary results from Raman and FTIR spectroscopy, EPR spectroscopy allowed to derive a complete reaction mechanism.⁸³