Effect of maghemization on the magnetic properties of nonstoichiometric pseudo-single-domain magnetite particles

The effect of maghemization on the magnetic properties of magnetite (Fe 3 O 4 ) grains in the pseudo-single-domain (PSD) size range is investigated as a function of annealing temperature. X-ray diffraction and transmission electron microscopy conﬁrm the precursor grains as Fe 3 O 4 ranging from (cid:2) 150 to (cid:2) 250 nm in diameter, whilst M € ossbauer spectrometry suggests the grains are initially near-stoichiometric. The Fe 3 O 4 grains are heated to increasing reaction temperatures of 120–220 8 C to investigate their oxidation to maghemite ( c -Fe 2 O 3 ). High-angle annular dark ﬁeld imaging and localized electron-energy loss spectroscopy reveal slightly oxidized Fe 3 O 4 grains, heated to 140 8 C, exhibit higher oxygen content at the surface. Off-axis electron holography allows for construction of magnetic induction maps of individual Fe 3 O 4 and c Fe 2 O 3 grains, revealing their PSD (vortex) nature, which is supported by magnetic hysteresis measurements, including ﬁrst-order reversal curve analysis. The coercivity of the grains is shown to increase with reaction temperature up to 180 8 C, but subsequently decreases after heating above 200 8 C; this magnetic behavior is attributed to the growth of a c -Fe 2 O 3 shell with magnetic properties distinct from the Fe 3 O 4 core. It is suggested there is exchange coupling between these separate components that results in a vortex state with reduced vorticity. Once fully oxidized to c -Fe 2 O 3 , the domain states revert back to vortices with slightly reduced coercivity. It is argued that due to a core/shell coupling mechanism during maghemization, the directional magnetic information will still be correct; however, the intensity information will not be retained.


Introduction
During formation, magnetic minerals record the direction and intensity of the Earth's magnetic field. Paleomagnetists use this information to investigate, for example, past tectonic plate motion and geodynamo evolution. However, subsequent to formation, the constituent magnetic minerals are commonly exposed to a range of weathering conditions and environments. One of the most common weathering processes is maghemization, which is the oxidation of (titano)magnetite at ambient temperatures, i.e., the slow oxidation of (titano)magnetite to (titano)maghemite [van Velzen and Zijderveld, 1995]. Maghemization is known to be very common in mid-ocean ridge basalts >2 Ma [Bleil and Petersen, 1983;Matzka et al., 2003], and understanding this general process is important as maghemization that is known to alter the original remanent magnetization [Kr asa et al., 2005].
Of the constituent magnetic minerals, particles in the single domain (SD) grain size range (<100 nm) are regarded as ideal paleomagnetic recorders because of their strong remanence and high magnetic stability, with potential relaxation times greater than that of the age of the Universe [Dunlop and € Ozdemir, 2001]. However, magnetic signals from rocks are often dominated by small grains with nonuniform magnetization that exhibit magnetic recording fidelities similar to those of SD grains (termed pseudo-SD (PSD)).
In this paper, the progressive oxidation/maghemization of PSD magnetite (Fe 3 O 4 ) to maghemite (c-Fe 2 O 3 ) is investigated. Oxidation of PSD Fe 3 O 4 grains is a critical process of alteration of remanent magnetic signals and must be fully understood to allow for reliable interpretation of paleomagnetic data. The oxidation process starts at the Fe 3 O 4 surface, where Fe 21 is either partially removed from the crystal or it reacts with oxygen to form a new crystal layer [O'Reilly, 1984]. Further oxidation is driven by the oxidation gradient where Fe 21 diffuses from the interior of the grain to the surface, leaving vacancies in the interior. Hence, depending on the degree of oxidation, the Fe 3 O 4 grains can be altered to structures ranging from a thin oxidized c-Fe 2 O 3 surface layer, a hetero-structured core-shell Fe 3 O 4 /c-Fe 2 O 3 grain or a fully oxidized c-Fe 2 O 3 grain.
Recent work using the transmission electron microscopy (TEM) technique of off-axis electron holography combined with environmental TEM visually showed a change in both strength and direction of remanent magnetization within PSD Fe 3 O 4 particles, as a consequence of in situ oxidation within the TEM [Almeida et al., 2014]. This study is very important as it directly confirmed that oxidation can modify the original magnetic signal within individual Fe 3 O 4 particles in the size range dominant in rocks. However, due to the reduced O 2 atmosphere pressures (<10 mbar) permitted for these in situ experiments, it was necessary to use elevated temperatures (8008C) to induce oxidation and hence is not directly representative of nature. Owing to the localized aspect of this study, it was limited to analysis of only a few individual PSD Fe 3 O 4 particles and it is therefore perhaps incorrect to generalize this magnetic behavior to the larger numbers of grains within rocks. In this context, the ex situ oxidation of a bulk powder comprised entirely of PSD Fe 3 O 4 grains, at comparatively lower temperatures, is reported. The chemical and structural transformation associated with low-temperature oxidation of the PSD Fe 3 O 4 particles is related directly to their corresponding magnetic properties. This work is similar in objective to several previous studies [e.g., Liu et al., 2004;Ge et al., 2014]; however, unlike the previous studies, advanced TEM methods and M€ ossbauer spectroscopy, in addition to magnetic hysteresis and X-ray diffractometry (XRD), are applied.

Experimental
Synthetic Fe 3 O 4 particles (diameter ranging from $150 to $250 nm) were purchased from Nanostructured and Amorphous Materials, USA. For heating experiments, discreet samples of 30 mg native Fe 3 O 4 powder (Sample S1) were mixed with 20 ml ethanol in a beaker and dispersed using an ultrasonic bath. Dispersions were poured into a glass petri dish and then inserted into a temperature-controlled furnace at reaction temperatures of 120, 140, 160, 180, 200, and 2208C under ambient atmosphere (denoted as Samples S120-S220). To promote a uniform powder distribution, dispersions were heated until the ethanol had evaporated ($5 to 10 min), leaving a thin layer of powder, and subsequently heated for a further 2 h. After heating, the petri dish was removed and the dry reaction products were allowed to cool at room temperature.
The synthesized reaction products were deposited onto single crystal silicon substrates for crystallographic identification using XRD (PANalytical X'Pert PRO Diffractometer). For the purpose of TEM, the particles were dispersed in acetone using an ultrasonic bath before deposition onto lacey carbon/copper mesh support grids (Agar Scientific Ltd., UK). Conventional diffraction contrast imaging was performed using an FEI Titan TEM operated at 300 kV (Center for Electron Nanoscopy (CEN), Technical University of Denmark) and selected area electron diffraction (SAED) patterns allowed for phase identification. Electron-energy loss spectroscopy (EELS) analysis was performed at an energy resolution of $0.3 eV, achieved through excitation of the monochromator, to provide information on sample oxidation state. High-angle annular dark field (HAADF) high-resolution scanning TEM (STEM) analysis was performed using a Nion UltraSTEM 100 at 100 kV (SuperSTEM, STFC Daresbury Laboratories, Daresbury, UK). EELS analysis was performed to measure the oxygen/Fe ratio and was quantified using Digital Micrograph software. The background signal was subtracted and, using the signal quantification parameters determined by the software, the relative area densities in the EEL spectra allowed the relative atomic ratios to be calculated. For the purpose of off-axis electron holography [Dunin-Borkowski et al., 1998], an FEI Titan with a C S probe corrector on the condenser lens, equipped with a Lorentz lens and biprism, was used at 300 kV (CEN). Off-axis electron holograms were recorded in Lorentz mode with the objective lens switched off and the specimen in magnetic-field-free conditions at room temperature. Mean inner potential contributions to the recorded phase shift were removed by using in situ magnetization reversal [Kasama et al., 2011].
The magnetic measurements were conducted at the Natural Magnetism Group Laboratory at the Imperial College London and Technical University of Denmark. High-temperature thermomagnetic curves were measured using a Princeton Measurements vibrating sample magnetometer (VSM) fitted with a furnace Geochemistry, Geophysics, Geosystems

X-Ray Diffraction Analysis
The XRD patterns of Figure 1 illustrate the effect of increasing temperature on the crystallography of the reaction products. The peaks in the XRD pattern of Samples S1 are in good agreement with the presence of Fe 3 O 4 (Joint Committee on Powder Diffraction standards (JCPDS) 75-449). Samples S120 and S140 exhibit XRD patterns similar to S1, whilst development of small peaks characteristic to c-Fe 2 O 3 (JCPDS 8-7119), can be seen with increasing reaction temperature in Samples S160-S220. The findings are in agreement with those of Ge et al. [2014] who made similar heating experiments for grains that were $80 to 240 nm in diameter. In addition, the 2h values were used to determine the interplanar lattice spacing (d) for the {220}, {311}, {400}, {511}, and {440} sets of lattice planes. The lattice parameter (a) was calculated from each set of (hkl) using the equation for the cubic crystal unit cell [Cullity and Stock, 2001], and the average was determined. The change in lattice parameter a as a function of reaction temperature is summarized in Table 1. Figure 2 provides information on the size, morphology, and crystallographic phase of the initial Fe 3 O 4 precursor (Sample S1), along with the c-Fe 2 O 3 sample (S220), produced by heating to 2208C (Sample S220). The bright-field (BF) images of Sample S1 ( Figure 2a) displays generally isotropic Fe 3 O 4 grains ($150 to $250 nm in diameter), as confirmed by SAED (Figure 2a, inset). EELS analysis of the O 1s K edge, in the region 520-580 eV (Figure 2c), is indicative of Fe 3 O 4 and composed of a prepeak (labeled a); a broad edge at higher energy (labeled b); along with higher-energy-loss structures, lying at about 548 eV (labeled c); and 563-565 eV (labeled d). Prepeak (a) results from the hybridization of Fe 3d and O 2p and usually splits into t2g and eg states by the ligand field. Peak (b) is composed of density of states from oxygen p character hybridized with metal 4s and 4p states [Colliex et al., 1991]. The higher-energy-loss maxima in the fine structures (labeled c and d) are induced during the final state of the emitted electron, described as superposition, on the core-excited site, of an outgoing wave and of the multiple backscattered waves on the successive shells of neighboring oxygen ions [Colliex et al., 1991]. EELS analysis of the Fe 2p L 2,3 edge, in the region 704-726 eV (Figure 2d), confirmed the assignment of Fe 3 O 4 . The L 2 edge for this sample shows the typical shape of a mixed-valence compound, i.e., three visible features of differing intensities (Figure 2d, black arrows), whilst the almost shapeless L 3 edge is attributed to the combined spectral contributions of different iron sites (i.e., Fe 21 at octahedral B-sites and Fe 31 at both tetrahedral A and octahedral B-sites), consistent with the more delocalized structure of Fe 3 O 4 , as compared with other mixed iron oxides [Crocombette et al., 1995;Fujii et al., 1999].  Geochemistry, Geophysics, Geosystems The HAADF images of Figure 3 provide a closer examination of an individual Fe 3 O 4 grain that was heated at 1408C (Sample S140). Figure 3a shows the grain to be rectangular in this projection, with approximate dimensions of $130 nm long and $110 nm wide. Higher magnification of the boxed region in Figure 3a is displayed in Figure 3b and highlights the characteristic lattice fringes of the {111} set of planes (inset). EEL spectra were acquired from the center, near-edge, and surface regions labeled A, B, and C, respectively, and the associated atomic weight composition (at %) was quantitatively determined from each cumulative EEL spectrum (summarized in Table 2). The data are suggestive of a slightly higher O content present at the surface of the nominally magnetite grain than at its near-edge and center.

Transmission Electron Microscopy
Off-axis electron holography allowed for the imaging of the in-plane magnetization of Samples S1 and S220 and is presented in Figures 4 and 5.

Magnetic Properties
The M€ ossbauer spectra of Figure 6, acquired at room temperature, depict the transition from nonstoichiometric Fe 3 O 4 to c-Fe 2 O 3 as a function of increasing reaction temperature. M€ ossbauer parameters were determined by fitting two sextets, St 1 (red) and St 2 (blue), with area ratios of the lines constrained to 3:2:1:1:2:3 for both sextets, using a Lorentzian fitting program called mfit (Table 3). The oxidation state from magnetite to maghemite of the samples was estimated as the magnetite content (wt %) in a magnetitemaghemite mixture, by considering the weighted average isomer shift d aver of the two sextets (Table 3)  The uncertainty of this model is estimated to about 5 wt % magnetite, given that d aver is determined with an error of 0.01 mm/s. The two sextets in the spectrum of S1 exhibit hyperfine parameters which at first may appear similar to pure Fe 3 O 4 , but the weighted average isomer shift of 0.46 mm/s (cf. d aver 5 0.55 mm/ s for pure magnetite) and the relative spectral area of the two sextets of 0.93 (cf. St 1 /St 2 5 1.9 for pure magnetite) [Da Costa et al., 2014] suggest nonstoichiometric magnetite. Using the model of Da Costa et al.
[2014], a magnetite content of 59% for Sample S1 is obtained. The spectra of Samples S120-S200 demonstrate further reduction in magnetite content and correspondingly oxidation to maghemite with increasing reaction temperature, as shown in Figure 7a and summarized in Table 3. Samples S200 and S220 display a typical spectrum of c-Fe 2 O 3 , composed of two sextets with very similar hyperfine parameters [Da Costa et al., 1995]. In addition, the oxidation parameter z was determined by comparing the F B /F A ratios in Table 3 to the average of the experimental data in Ramdani et al. [1987] and Schmidbauer and Keller [2006], where the oxidation parameter z varies from 0 to 1, representing a transformation from Fe 3 O 4 (z 5 0) to c-Fe 2 O 3 (z 5 1) [Readman and O'Reilly, 1972;Ge et al., 2014]. The relationship between the oxidation parameter z and the lattice parameter a is shown is Figure 7b, where z is shown to increase with the decreasing size of the cubic unit cell.   reaction products and FORC diagrams of Samples S1 and S220. The coercivity (B c ) distribution in each sample is shown by horizontal profiles along the B u 5 0 axis of FORC data measured from Samples S1-S220 ( Figure 8a). In addition, the measured values for coercivity of remanence (B cr ) and B c are summarized in Table 4. A general trend of increasing B cr values and B c profiles with reaction temperature and oxidation can be seen up to 1808C, and exhibits a marked decrease thereafter, which is in agreement with the study of Ge et al. [2014]. The ratios of M rs /M s versus H cr /H c are plotted in Figure 8b (a ''Day plot'' [Day et al., 1977]) for the different reaction products. All the samples lie within the PSD region [Dunlop, 2002], however, Sample S220 plots distinctly toward the SD region, in agreement with the previous studies [Liu et al., 2004]. The FORC diagram of Sample S1 (Figure 8c) displayed PSD/SD type behavior with a distinct peak centered at a B c value of $22 mT [Muxworthy and Dunlop, 2002]. The peak in the FORC diagram of Sample S220 ( Figure  8d) depicts similar behavior, but is centered at a lower B c value of $18 mT.
Curie temperatures (T C ) were determined using the second-derivative method [Tauxe, 1998] and the relationship with oxidation parameter z is presented in Figure 9a. Values of T C were calculated as 583 6 58C for Samples S1 and S120, which is marginally above that of stoichiometric Fe 3 O 4 [Dunlop and € Ozdemir, 2001]. The T C of Samples S140 and S160 was slightly higher and a linear increase of T C with temperature can be seen in Samples S180-S220, with a maximum T C of 649 6 58C, which is in good agreement with that of c-Fe 2 O 3 [Dunlop and € Ozdemir, 2001]. This trend of increasing T C with reaction temperature is similar to that seen in the oxidation process of titanomagnetites [Readman and O'Reilly, 1972]. The thermomagnetic curve of Sample S220 is presented in Figure 9b and the large difference between the heating and cooling curves

Overview
This study has followed the maghemization of Fe 3 O 4 grains in the PSD size range as a function of reaction temperature and the associated effect on their magnetic properties. The XRD pattern of Sample S1 is in good agreement of pure Fe 3 O 4 , which is reinforced by complementary SAED and characteristic O 1s K/Fe 2p L 2,3 edges in the associated EEL spectra. Further, TEM imaging showed the particles to be $150 to $250 nm in diameter and off-axis electron holography revealed their PSD nature. However, M€ ossbauer analysis of Sample S1 exhibited sextets with hyperfine parameters more indicative of nonstoichiometric Fe 3 O 4 , with an approximate composition of 59% magnetite and 41% maghemite. Increasing the reaction temperature led to the development of additional peaks in the XRD patterns of Samples S120-S200, and Sample S220 displayed an XRD pattern characteristic of c-Fe 2 O 3 . This structural phase transformation is reinforced by the SAED pattern of Sample S220 and development of small peaks in the EEL spectra of Fe L 23 edge further confirm the oxidation to c-Fe 2 O 3 . TEM imaging shows the reaction products heated to 2208C retained their shape and without a marked change during the transition to c-Fe 2 O 3 . M€ ossbauer spectra of Samples S120-S200 demonstrate a reduction in magnetite content and conversion to maghemite with increasing reaction temperature, where Sample S220 displays a typical spectrum of c-Fe 2 O 3 [Da Costa et al., 1995;Schmidbauer and Keller, 2006].

Oxidation Mechanism
Oxidation of inverse spinel structured Fe 3 O 4 under low-temperature conditions is known to involve topotactic transformation where the original morphology is maintained throughout [Gallagher et al., 1968;Cornell and Schwertmann, 2003]. The reaction proceeds by outward migration of Fe 21 cations toward the surface of the crystal, thereby creating cation vacancies, and absorbed oxygen atoms at the surface oxidize the migrating cations to form a shell of c-Fe 2 O 3 . In terms of diffusion pathways, when a-Fe 2 O 3 is reduced to Fe 3 O 4 , the newly formed Fe 3 O 4 {111} planes are parallel to the original a-Fe 2 O 3 {006} planes, with the latter considered favorable planes for oxygen diffusion during thermal reduction [Almeida et al., 2012]. Hence, in the reverse process of oxidation, the Fe 21 cations are suggested to preferentially diffuse along the Fe 3 O 4 {111} planes, Figure 6. M€ ossbauer spectra of Samples S1-S220 (black) follow the maghemization of Fe 3 O 4 as a function of reaction temperature. M€ ossbauer parameters were determined by fitting two sextets, St 1 (red) and St 2 (blue).  , where the octahedral sites are enclosed by the brackets and w denotes a lattice vacancy. The composition calculated quantitatively from the EEL spectra shows the surface of the Fe 3 O 4 grain in Sample S140 (Spectrum C, Figure 3) to exhibit a slightly higher O content than toward its center. By implication, the observed lower surface Fe content provides evidence for the surface oxidation mechanism with the formation of core-shell Fe 3 O 4 /c-Fe 2 O 3 structures with increasing reaction temperature. However, the change in stoichiometry during oxidation of significantly larger Fe 3 O 4 grains is considered to proceed more uniformly.    [Mathew and Juang, 2007]. In contrast, the ferrimagnetism of c-Fe 2 O 3 is the result of an average excess per formula weight Fe 8/3 O 4 of 2 / 3 of an Fe 31 ion on the B sublattice [Dunlop and € Ozdemir, 2001]. The T C values for Samples S1 and S220 were indicative of Fe 3 O 4 and c-Fe 2 O 3 , respectively, whilst their FORC diagrams display PSD/SD type behavior with distinct peaks centered at $22 and $18 mT, respectively. Construction of magnetic induction maps from the off-axis electron holograms complemented the FORC data by visualizing the PSD domain state nature of both the Fe 3 O 4 and c-Fe 2 O 3 grains. Interestingly, during the oxidation process from Fe 3 O 4 to c-Fe 2 O 3 , the B c and B cr values generally increased with reaction temperature up to 1808C, with subsequent decrease thereafter. Ge et al. [2014] reported a similar trend during oxidation of Fe 3 O 4 particles in the PSD size range, which they interpreted using a micromagnetic numerical model: for PSD grains near the critical SD grain size boundary, as the Fe 21 cations diffuse toward the surface of the particle and the c-Fe 2 O 3 shell thickness increases through oxidation, the diameter of the Fe 3 O 4 core decreases as a consequence. As the Fe 3 O 4 core becomes smaller, the overall magnetic strength of the grain decreases due to the lower M s of the c-Fe 2 O 3 skin, whilst exchange coupling of the Fe 3 O 4 core to the c-Fe 2 O 3 skin maintains the high magnetocrystalline anisotropy, and at a point, the minimum energy domain structure is more SD-like. As oxidation proceeds the small Fe 3 O 4 core shrinks sufficiently so that it is unable to provide a significant influence on the B c , which collapses markedly as the grain reverts to a vortex structure. For larger PSD grains, the grains remain in a vortex state throughout the oxidation process, but vorticity of the magnetic domain state reduces as the c-Fe 2 O 3 skin grows. The magnetically harder and slightly more uniform domain structures result in a trend of increasing B c until the Fe 3 O 4 core almost disappears when both parameters decrease significantly, which is consistent with the experimental data and supports the proposed mechanism of maghemization proceeding through the formation of core-shell Fe 3 O 4 /c-Fe 2 O 3 structures. Due to the rather subtle changes observed in the B c values and FORC diagrams, the transition from vortex state to vortex with reduced vorticity is considered to be the dominant mechanism in this study. It appears unlikely given the small increases  Figure 9. (a) Curie temperatures of Samples S1-S220 as a function of oxidation parameter z. (b) Thermomagnetic curve of Sample S220. Heating was performed in flowing helium in a field of 300 mT.

Implications for Paleomagnetic Studies
From the above observations, the reliability of the magnetic remanence of maghemitized rocks is called into question; the process of oxidation will lead to the acquisition of a new chemical remanent magnetizaton (CRM). Due to the shell growth mechanism, the original Fe 3 O 4 phase in the core will effect the magnetic alignment of the c-Fe 2 O 3 shell [Ge et al., 2014]. It is likely that on average for a distribution of the particles the original magnetization direction will be retained. However, obtaining an paleointensity estimate may not be accurate for two reasons: (1) the core/shell CRM acquistion model may not always couple and the CRM of individiual particles may not be aligned with the orignal remanence and (2) whilst it might be theoretically possible to recover the paleointensity informaton from a c-Fe 2 O 3 sample, c-Fe 2 O 3 is generally unstable to heating, as evidenced by the thermomagnetic curve of Sample S220, and samples dominated by this phase are, therefore, likely to fail Thellier-type or other heating paleointensity protocols. It is also recognized that potential differences exist between these experimental results, which involve oxygen addition, compared to that undergone in mid-ocean ridges that may favor a Fe-removal mechanism for oxidation.

Conclusions
This study has investigated the maghemization of Fe 3 O 4 grains in the PSD size range and the effect of this oxidation process on their magnetic properties. Complementary methods of XRD and Mossbauer spectroscopy followed this phase transformation as a function of temperature and showed the onset of oxidation was most pronounced at 1608C, with a reaction temperature of 2208C producing pure c-Fe 2 O 3 . The phase transformation is confirmed by EELS analysis of the O 1s K/Fe 2p L 2,3 edges acquired from the precursor Fe 3 O 4 grains and fully oxidized c-Fe 2 O 3 grains. High-resolution HAADF imaging and localized EELS analysis revealed evidence to suggest the surface to be more oxidized, supporting the formation of a core-shell Fe 3 O 4 /c-Fe 2 O 3 structure. Off-axis electron holography allowed for construction of magnetic induction maps of individual Fe 3 O 4 and c-Fe 2 O 3 grains, revealing their PSD (vortex) nature in both cases, which is supported by corresponding first-order reversal curves characteristic of PSD states. Thermomagnetic curves showed an increase of T C with reaction temperature that is consistent with maghemization process. The coercivity of the magnetite grains was shown to increase with reaction temperature up to 1808C, but decrease thereafter. This magnetic behavior is attributed to exchange coupling between a c-Fe 2 O 3 shell with the Fe 3 O 4 core resulting in an energy minimum of SD state or a vortex state with reduced vorticity, with the latter considered the dominant mechanism in this study. Once fully oxidized to c-Fe 2 O 3 , the domain states revert back to vortices with slightly reduced coercivity. It is suggested that heavily maghemized samples may not retain meaningful paleomagnetic signals.