University of Birmingham Scalable synthesis of dispersible iron carbide (Fe3C) nanoparticles by ‘nanocasting’

Metal carbides have shown great promise in a wide range of applications due to their unique catalytic, electrocatalytic and magnetic properties. However, the scalable production of dispersible metal carbide nanoparticles remains a challenge. Here, we report a simple and scalable route to dispersible iron carbide (Fe 3 C) nanoparticles. This uses MgO nanoparticles as a removable ‘ cast ’ to synthesize Fe 3 C nanoparticles from Prussian blue (KFe III [Fe II (CN) 6 ]). Electron tomography demonstrates how nanoparticles of the MgO cast encase the Fe 3 C nanoparticles to prevent sintering and agglomeration during the high-temperature synthesis. The MgO cast is readily removed with ethylenediaminetetraacetic acid (EDTA) to generate Fe 3 C nanoparticles that can be used to produce a colloidal ferro ﬂ uid or dispersed on a support material.

Scalable synthesis of dispersible iron carbide (Fe 3 C) nanoparticles by 'nanocasting' † Introduction Transition metal carbides have historically been valued for their hardness and chemical resistance. However, more recently they are being investigated for their unique catalytic, 1 electrocatalytic 2 and magnetic properties, 3 particularly at the nanoscale. For example, iron carbide (Fe 2.2 C and Fe 5 C 2 ) nanoparticles are applied in magnetic hyperthermia for cancer treatment 4 and also Fischer-Tropsch synthesis of liquid hydrocarbons. 5 Fe 3 C (cementite) nanostructures are also being developed for fuel cell reactions such as oxygen reduction 6 and hydrogen evolution, 7 where they have the potential to replace noble metals. A major factor inhibiting the development of carbides for all of these technologies is the lack of simple, scalable synthetic routes to metal carbide nanoparticles. Many of the potential applications of carbides also require dispersible nanoparticles, 8 for example in the production of colloidal dispersions or for combining catalysts with a cocatalyst or support.
The scalable synthesis of dispersible transition metal carbides is challenging due to the ceramic nature of carbides.
The industrial manufacture of carbides generally relies on direct heating of metal or metal oxide powders with carbon at high temperature, oen >1000 C, which leads to powders of large (1-100 mm) agglomerated particles with low surface area (Fig. 1a). Sol-gel chemistry has provided some useful advances in the synthesis of some iron carbides, including Fe 3 C. Here, an iron salt is combined in solution with an organic precursor, then heated in an inert atmosphere (Fig. 1b). The organic precursor decomposes to form carbon, which drives formation of Fe 3 C nanoparticles. 9,10 However, while sol-gel chemistry offers good control over Fe 3 C crystallite size, the resulting nanoparticles are embedded within a matrix of carbon. Dispersing these nanoparticles in solution relies on physical or chemical methods to break individual nanoparticles away from the carbon matrix 11 and can result in discarding signicant portions of the material. 12 In this study, we address the challenge of producing dispersible iron carbide (Fe 3 C) nanoparticles by synthesizing the nanoparticles within a nanostructured 'cast' (Fig. 1c). Iron carbide nanoparticles are prepared by thermal decomposition of nanoparticles of Prussian blue (KFe III [Fe II (CN) 6 ]) within a magnesium oxide nanopowder. Electron tomography demonstrates this proof of concept, showing individual Fe 3 C nanoparticles 'wrapped' in the MgO matrix. The MgO cast is easily removed using aqueous ethylenediaminetetraacetic acid (EDTA) and the resulting Fe 3 C nanopowder is used to make a ferrouid, demonstrating the dispersible nature of the Fe 3 C particles. One of the major advantages of this method is that it is simple and scalable.

Synthesis and characterization of dispersible Fe 3 C nanoparticles
Prussian blue nanoparticles protected with polyvinylpyrrolidone were synthesized in a method modied from Uemura et al. 13 It is already known that bulk Prussian blue will decompose to Fe 3 C on heating in an inert atmosphere. 14 A MgO/Fe 3 C nanocomposite was prepared by vigorously mixing the as-synthesized Prussian blue nanoparticles with commercial MgO nanopowder (1 : 1 ratio by mass) as a slurry in methanol, drying and then calcining the pale blue powder to 800 C under N 2 . X-ray diffraction (XRD) patterns of the resulting grey powder showed characteristic peaks for MgO and Fe 3 C (Fig. 2). High-angle annular dark eld (HAADF) transmission electron microscopy (TEM) images (Fig. 3a) show the Fe 3 C particles as bright spots within a matrix of smaller MgO particles, which have protected the Fe 3 C nanoparticles from agglomerating or sintering. The separate phases are clearly identied by energy dispersive X-ray (EDX) maps ( Fig. 3b-e), with red, blue and turquoise highlighting the position of Fe, Mg and O in the sample, respectively. Green in the EDX map corresponds to carbon and shows both the lacey carbon lm of the TEM grid and the presence of carbon shells around many of the Fe 3 C nanoparticles. This is a common feature of Fe 3 C synthesis, 15 as iron is able to catalyze the conversion of amorphous carbon to graphitic carbon, the carbon in this case coming from the polyvinylpyrrolidone. The presence of the carbon shell is benecial as it offers protection for the Fe 3 C nanoparticles against oxidation. High-resolution TEM images conrm the presence of thin graphitic carbon shells around the Fe 3 C particles, with the fringe spacing of 0.34 nm corresponding to the (002) d-spacing of graphite (Fig. 3f). As a control, a sample of Prussian blue was heated in a furnace without MgO. The TEM image (Fig. S1 †) of the resulting sample shows a dense network of Fe 3 C particles that appear to be sintered together.
Further evidence for the dispersion of Fe 3 C nanoparticles within the MgO matrix comes from electron tomography. Fig. 4a shows screenshots from a HAADF tomographic video (ESI Video 1 †) produced by processing a series of images of a representative section of sample at 2 tilt intervals through 180 . The bright particles of Fe 3 C can be seen dispersed within the matrix of smaller MgO nanoparticles. Fig. 4b shows the screenshots from a corresponding EDX tomographic video (ESI Video 2 †), showing Fe-rich particles (red) dispersed within a Mgrich (blue) matrix. A magnied portion of the sample shows how the MgO matrix is able to partially encapsulate the Fe 3 C nanoparticles ( Fig. 4c), demonstrating the 'cast' concept.
To isolate dispersible Fe 3 C nanoparticles, aqueous ethylenediaminetetraacetic acid (EDTA) was used to dissolve the MgO cast. Washing with EDTA, rinsing with water then ethanol and drying in air produces a black solid with characteristic XRD peaks for Fe 3 C ( Fig. 5a) with all characteristic peaks for crystalline MgO removed. Elemental analysis (by Flame Atomic Absorption Spectroscopy) also indicated that 98% of the Mg had been removed by EDTA washing (ESI Section 1.2 †). TEM images of the EDTA-washed sample show particles of Fe 3 C alongside material with a rumpled sheet or bre-like appearance (Fig. 5b). This is attributed to amorphous carbon as elemental analysis of the sample shows 17.5% carbon (by mass) compared to 6.7% in pure Fe 3 C, due to excess carbon from the polyvinylpyrrolidone. The amorphous carbon can be removed by washing the sample with dilute H 2 O 2 (as reported by Schliehe et al., details in ESI †) 11 leaving only a graphitic carbon shell around each Fe 3 C nanoparticle (Fig. 5c) and without oxidizing the Fe 3 C (Fig. S2 †).

Magnetic properties
The magnetic properties of the Fe 3 C powder were studied using a superconducting quantum interference device (SQUID). It should be noted that analysis of the magnetic properties of this system is complicated by the polydispersity of the particles and the presence of graphitic shells. 16 Fig. 5d shows the temperature dependence of the zero-eld-cooled (ZFC) and eld-cooled (FC) magnetization of a packed powder of Fe 3 C from 10-400 K under an applied eld of 100 Oe. The data indicate superparamagnetic behaviour in at least part of the sample. The ZFC curve exhibits a broad maximum with the peak at 130 K. This is the average blocking temperature (T B ); below this point particle moments are not free to thermally uctuate. Below T B , the ZFC decreases sharply, whereas the FC curve increases gradually, a characteristic of non-interacting or weakly-interacting nanoparticles. 17 The ZFC and FC magnetization curves do not converge below 400 K. The point of convergence is the irreversibility temperature (T irr ), and corresponds to the blocking temperature of the largest particles in the sample. As samples are cooled, the  particle moments are blocked progressively, since larger particles have a higher energy barrier to thermal uctuation. The difference between T B and T irr therefore gives a measure of the distribution of energies, which in turn gives an indication of polydispersity. The large gap between T B and T irr in this sample indicates a large particle size distribution. This is consistent with the observed polydispersity (resulting from a large size distribution in the Prussian blue precursor). Fig. 5e and S3 † show plots of the eld dependence of the isothermal magnetization at 400 K and 300 K respectively. At 400 K, the coercivity is 140 Oe and remanence is 2.7 emu g À1 and at 300 K the coercivity is 300 Oe and the remanence is 7 emu g À1 , consistent with other reports of Fe 3 C nanoparticles. 18 The saturation magnetization (66 emu g À1 at 300 K) is slightly lower than other reports, possibly due to the presence of defects in the Fe 3 C structure, or the presence of carbon in the product.

Particle size distribution
Characterization of the particle size distribution was achieved using small-angle X-ray scattering (SAXS). Fig. 6a shows the SAXS and WAXS (wide-angle X-ray scattering) data t for a Prussian blue/MgO mixture before pyrolysis, compared to the same sample aer pyrolysis (Fe 3 C/MgO) and aer washing with EDTA (Fe 3 C). The peaks in the high q (WAXS) region correspond to the crystalline phases observed in X-ray diffraction analysis (Fig. S4 †). SAXS derives from an interaction between the incident X-ray radiation and regions of contiguous electron density within a sample, which here are nanoparticles (Prussian blue, MgO and Fe 3 C) and residual carbon. The SAXS data was analysed using a Monte Carlo method to extract form-free size distributions. 19 The resulting radius histograms, scaled by relative volume fraction, for samples of Prussian blue/MgO, Fe 3 C/MgO and Fe 3 C (aer EDTA washing) are shown on a log 10 scale in Fig. 6bd respectively. For the sample of Prussian blue nanoparticles in a MgO matrix, there are two main regions of interest. The rst, R1 ($2-20 nm) corresponds to the MgO nanopowder and is consistent with observations of the MgO in TEM images. R2 ($30-160 nm) represents the radius distribution of the larger Prussian blue nanoparticles, again consistent with observations from TEM. The tail in the data below 3 nm corresponds to surface roughness and is most like a result of scattering from the polyvinylpyrrolidone that is present on the surface of the Prussian blue nanoparticles from the synthesis method. The same data is plotted on a linear scale in Fig. S5a † and shows that $40% (by volume) of the scattering objects are less than 15 nm in radius, representing the MgO nanoparticles and polyvinylpyrrolidone residue. The cumulative distribution function rises steadily aer this point, indicating a large size distribution in the Prussian blue.
The size distribution data for the Fe 3 C/MgO mixture also shows a bimodal distribution (Fig. 6c), but with the peaks closer together than the parent Prussian blue/MgO. Region R3 (radii of $5-30 nm) represents the MgO particles, with the slightly larger particle radii probably caused by sintering of the MgO during This journal is © The Royal Society of Chemistry 2019 the 800 C synthesis. The distribution of radii for the Fe 3 C particles ($30-120 nm) indicates a contraction of the particles during the Prussian blue to Fe 3 C transformation (region R4). Given the mass loss the Prussian blue is likely to experience on transforming to Fe 3 C, a reduction in the particle size is expected. The overall reduction in particle size during the Prussian blue to Fe 3 C transformation is clearer in the linear plot of the same data ( Fig. S5b †), where 80% of the cumulative distribution function represents particles <60 nm in radius. Fig. 6d is the size distribution data for Fe 3 C (aer MgO removal with EDTA). Region R6 (radii of $15-160 nm) corresponds to the Fe 3 C nanoparticles. Region R5 (1-8 nm) is probably a result of scattering from pores and surface roughness of the residual carbon. Scattering features like this have been observed before in templated carbons, where the carbon-air interface produces signicant scattering. 15 Given that this scattering feature is not present in Fig. 6c, it seems likely that the scattering results from pores created by the removal of MgO. This would suggest that carbon encases some of the MgO particles, possibly resulting from some mixing of MgO with the polyvinylpyrrolidone during precursor synthesis. Aer treatment with H 2 O 2 , the peak of this scattering feature shis to a smaller radius ($1.3 nm from $3 nm, Fig. S6b †). This is consistent with the removal of amorphous carbon, leaving behind graphitic carbon where scattering occurs due to surface roughness in the sample.

Producing a ferrouid
A measure of whether the Fe 3 C nanoparticles are dispersible is the ability to produce a ferrouid. Ferrouids are colloidal dispersions of single-domain magnetic particles in a carrier uid. They show properties of magnetic materials and uids. 20 In order to produce enough Fe 3 C for a ferrouid, a commercial Prussian blue powder was ground in a pestle and mortar with MgO nanopowder. The mixture was pyrolyzed under N 2 and washed with aqueous EDTA to generate dispersible Fe 3 C nanoparticles (full details in ESI †). TEM images (Fig. S7 †) show that the scaled-up synthesis does not affect the formation of nanoparticles. SAXS data shows a very similar scattering pattern for the scaled-up sample compared to the sample prepared with as-synthesized Prussian blue (Fig. S8 †). The particle size distribution for the scaled-up synthesis no longer shows a bimodal distribution, indicating that the Fe 3 C and MgO nanoparticles are closer in size (Fig. S8c †), suggesting the commercial sample is more homogeneous. A ferrouid was prepared by sonicating the Fe 3 C nanoparticle powder in DI water and oleic acid for 1 hour. The black powder collected in the oleic acid layer, which was separated using a SmCo 26 magnet and washed with ethanol/kerosene. Images of the resulting ferrouid are shown in Fig. 7, with characteristic 'peaks' from the dispersed Fe 3 C particles following the lines of the applied magnetic eld. XRD of the magnetic uid conrmed that Fe 3 C had been retained (Fig. S9 †).

Dispersion of particles on a support
A nal demonstration of the dispersibility of the Fe 3 C nanoparticles was to combine them with a support material. Since metal carbides are promising catalysts, it is important to be able to disperse them on standard support materials. Fe 3 C nanoparticles prepared from as-synthesized Prussian blue nanoparticles were dispersed by sonication in water with cetyltrimethylammonium bromide (CTAB) surfactant to aid dispersion. The resulting solution was then sonicated with silica (kindly provided by Johnson-Matthey) then dried at room temperature. SEM images, taken with a backscattered electron detector, are shown in Fig. 7b and c. The Fe 3 C nanoparticles can be seen as bright particles dispersed on the darker grey surface of the silica support.

Experimental
Prussian blue was prepared by dissolving FeCl 2 $4H 2 O (0.6 g) and polyvinylpyrrolidone (1 : 1 molar ratio) in deionised (DI) water (240 ml). Separately, K 3 [Fe(CN) 6 ] (0.99 g) was dissolved in DI water (60 ml) and the resulting solution added dropwise to the FeCl 2 solution. Acetone (300 ml) was added to induce agglomeration of the Prussian blue particles. The particles were washed 3Â with acetone and the dark blue solid transferred into methanol (300 ml). MgO nanopowder was added and the mixture stirred thoroughly before drying. To prepare MgO/Fe 3 C nanoparticle composites, the Prussian blue/MgO mixture was heated at 5 C min À1 to 800 C under N 2 in a tube furnace. To isolate dispersible Fe 3 C nanoparticles, the MgO was removed from MgO/Fe 3 C samples by stirring overnight in 0.2 M aqueous EDTA followed by washing 3Â with DI water and ethanol then drying. Full experimental details can be found in the ESI. †

Conclusions
In summary, we have demonstrated the synthesis of dispersible iron carbide nanoparticles by 'casting' Prussian blue nanoparticles within a matrix of MgO nanoparticles. The MgO matrix surrounds the Prussian blue particles as they decompose to Fe 3 C, preventing sintering and agglomeration of the Fe 3 C. The MgO matrix can easily be removed using EDTA to leave dispersible Fe 3 C nanoparticles. We have demonstrated the scalability of this process (up to 15 g batches). The resulting Fe 3 C powder can be dispersed in oleic acid to produce a ferro-uid, or dispersed on a SiO 2 support. The Fe 3 C nanoparticles are quite polydisperse, but it should be possible to improve this by using alternative routes to Prussian blue as well as experimenting with the type of MgO cast and the MgO : Prussian blue ratio. While this proof of concept focusses on the synthesis of Fe 3 C, it should be possible to apply this method more generally, e.g. by using Prussian blue analogues or metal organic framework nanoparticles as metal/carbon sources within the same MgO 'cast'. The importance of this method is that it is simple and scalable and relies on well-established ceramics processing technology, which should enable large-scale manufacture of dispersible metal carbide nanoparticles.

Conflicts of interest
There are no conicts to declare.