Iron-catalyzed graphitization for the synthesis of nanostructured graphitic carbons

Carbons are versatile and diverse materials that have numerous applications across energy and environmental sciences. Carbons with a graphitic structure are particularly appealing due to their high chemical stability, large surface areas and high thermal and electronic conductivity. Numerous methods exist to produce nanostructured graphitic carbons but some of these can be energy-intensive and/or have problems with scalability. One option that is being increasingly explored is the process of iron-catalyzed graphitization. This simply involves the pyrolysis of carbon-rich precursors in the presence of an iron catalyst and has been used to produce carbons with a wide range of structures and properties. This review will examine the current ﬁ eld of iron-catalyzed graphitization, with a focus on molecular organic or biomass precursors. Bio-derived precursors are particularly attractive as a potential option for sustainable production of graphitic carbons. We start with a brief introduction to some key carbon structures, the current applications in which they are employed and some of the key methods that have been developed to produce nanostructured graphitic carbons. We will then review the history of catalytic graphitization before evaluating the wide range of conditions and precursors that have been employed in catalytic graphitization. Finally, this review will investigate the current challenges facing iron-catalyzed graphitization, looking particularly at the limitations of the current understanding of the mechanistic aspects of graphitization, with a view to outlining where research in this ﬁ eld might progress.


Introduction
Carbon has been at the heart of human technological development since the ancient discovery that metals could be extracted from rocks by heating them with charcoal (smelting). In the modern era, carbons nd broad applications in energy and environmental applications as well as in pigments and as llers for elastomers. 1 The value of carbon materials can be Robert Hunter is currently studying for a PhD in the School of Chemistry at the University of Birmingham in the group of Dr Zoe Schnepp. His research is focused on the synthesis of porous graphitic carbons by iron-catalyzed graphitization of biomass-derived precursors. His work involves the use of both experimental and theoretical techniques to study mechanistic aspects of the graphitization process.

Joaquin
Ramirez-Rico is a professor of Physics at the University of Sevilla, in Spain. He is the scientic director of the X-ray facility and heads a research group working on methods to convert biomass and natural resources into advanced materials for structural and functional applications. He was a Postdoctoral Fellow in Northwestern University and specialized in synchrotron radiation analysis for in situ characterization of materials. His current interests are in carbon materials for electrochemical energy storage and wastewater treatment.
evidenced by the fact that graphite is now classed as a critical material, 2 due to its importance in refractory materials and battery technology. 3 Many governments now recognise the need for future carbon materials to be sustainable, both for environmental concerns and economic security.
One class of carbon materials that is receiving increasing attention is that of nanostructured graphitic carbons. These boast a wide variety of useful properties that include high chemical stability, large accessible surface areas and high thermal and electronic conductivity. This makes graphitic carbon materials valuable in technologies such as batteries, 4 fuel cells 5 and in separation/purication science. 6 Numerous techniques have been developed to produce complex graphitic nanostructures, e.g., chemical vapour deposition, 7 arc discharge, 8 or laser ablation 9 for the synthesis of carbon nanotubes. However, many of these methods are energy intensive and/or difficult to scale up. Therefore, there has been an international drive to develop cheaper, scalable, and more sustainable routes to nanostructured graphitic carbons. 10 A promising method for the synthesis of nanostructured graphitic carbons is catalytic graphitization. Catalytic graphitization is broadly dened as the "transformation of nongraphitic carbon into graphite by heat treatment in the presence of certain metals or minerals". 11 The process normally occurs at a much lower temperature than graphitization without a catalyst, which makes it attractive from both an environmental and economic perspective. Various transition metals have been used to promote catalytic graphitization, but iron is particularly appealing, due to its abundance and low toxicity. Iron-catalyzed graphitization can be considered to include processes such as the production of graphite within blast furnaces or the chemical vapour deposition synthesis of carbon nanotubes. We will discuss these briey, but the bulk of this article will describe the procedure of pyrolyzing mixtures of iron and organic precursors in an inert atmosphere to produce a range of nanostructured graphitic materials (Fig. 1). The simple conditions and wide range of potential organic precursors available for this process (including raw biomass) make it particularly appealing in terms of sustainability. We will compare the many examples of iron-catalyzed graphitization later in the review, but it is useful to start with a brief overview of carbon chemistry.

Dening carbon structures
Dening (and naming) carbon materials is made challenging by the complexity and wide range of structures that are possible. Heating any organic material (synthetic or biological) in a lowoxygen atmosphere will generate a carbon-rich solid, with the chemistry and structure depending on the heating conditions and the nature of the precursor. Most of these carbon materials do not have a perfectly regular structure like that of diamond or graphite but instead are comprised of a mixture of features with signicant variability and disorder. Considerable effort has gone into characterizing different types of carbon materials, for example through 13 C NMR spectroscopy, 12 Raman spectroscopy, 13 X-ray diffraction, 14,15 transmission electron microscopy, 16 total scattering, 17 electron energy loss spectroscopy (EELS) and X-ray photoelectron spectroscopy. 18 These techniques seek to probe features such as hybridization of the carbon atoms, the presence of sheet-like structures, interlayer spacing, orientation of stacked sheets, and sheet curvature caused by pentagonal or heptagonal rings of carbon. These materials are not just academic curiosities. Carbon blacks and activated carbons, for example, are widely used for air and water purication, as pigments in inks and as llers in elastomers and plastics. 19 More recently, new carbon materials are also being developed for high-tech energy applications such as anodes for lithium and sodium batteries, where understanding of the ne structure is critical to optimizing their properties. [20][21][22] Amorphous carbons are dened as containing a mixture of sp 2 and sp 3 hybridised carbon atoms and display no long-range crystalline order throughout their structure. 23 The surface of amorphous carbon typically consists of many reactive, dangling bonds, 24 and may also contain heteroatoms such as hydrogen, oxygen, nitrogen, or boron throughout the carbon network, depending on the method of synthesis. In contrast, graphitic carbons are composed of sp 2 hybridized carbon atoms arranged in hexagonal layers with long-range order that can be detected by diffraction methods. Strictly speaking, the term 'graphitic' should be reserved for carbon materials where the individual graphene layers are packed in an ABAB vertical arrangement. 25 However, many other carbon materials are commonly referred to as 'graphitic' if they have sheet-like structures arranged with an interplanar spacing similar to that of graphite. These can include multiwalled carbon nanotubes (Fig. 2a), 26 'nanoscrolls', 27 bamboo-like carbon nanotubes (Fig. 2b), 28,29 which have a segmented structure, and onion-like carbons (Fig. 2c), [30][31][32] which consist of roughly spherical concentric layers of graphitic carbon surrounding a central 'core' that can either be hollow or contain a nanoparticle catalyst. Graphitic carbons have various appealing properties, including high electronic conductivity and chemical and thermal stability. 33 Another term that is commonly used in describing some carbon materials is 'turbostratic', derived from "turbo" (rotated) and "strata" (layer). Turbostratic carbons contain stacked sheets of sp 2 -bonded carbon like graphite, but the sheets may exhibit random translation of the graphene layers along with rotational disorder, resulting in areas of larger interlayer spacing. 34 This can be observed in X-ray diffraction patterns as a broadening and/or a small shi to a lower 2q value of the characteristic (002) peak.
An important factor in describing the chemistry of carbonization and carbon materials is the fact that some carbons resist graphitization. This was discovered in the early twentieth century and led to a seminal paper by Rosalind Franklin, who coined the terms 'graphitizing' and 'non-graphitizing' to describe different types of carbon. 35 She observed that some organic materials, such as polyvinylchloride and pitch, could be converted to crystalline graphite at >2200 C whereas others, such as polyvinylidene chloride and sugar retain a porous isotropic structure even up to 3000 C. Graphitizing and non-graphitizing carbons are commonly called 'so' and 'hard' carbons, terms that relate to their observed physical properties. Understanding of the structure of hard and so carbons has evolved with the development of increasingly sophisticated analytical techniques. The structures proposed by Franklin involved small crystallites of graphite, each of which contains several small graphene layers. In graphitizing carbons, these crystallites are approximately parallel to each other, facilitating graphitization (Fig. 3a). In non-graphitizing carbons, she proposed that the individual crystallites were crosslinked in a random orientation (Fig. 3b). More recent studies, using aberration-corrected TEM, 13 C NMR and total scattering, provide evidence for a 'fullerene-like' structure for microporous carbons, involving curved sheets of graphene arising from nonhexagonal carbon rings ( Fig. 3c and d). [36][37][38][39] The structures of the different types of carbon are important as they may have  implications for the process of catalytic graphitization as well as helping us to describe and characterize some of the materials produced by catalytic graphitization.

Applications of carbon materials
As noted above, there are numerous applications for carbon materials and some of these represent large industries, for example the production of carbon black as a ller to improve the wear resistance of tyres. 40 Other signicant industrial examples include carbon bres for lightweight composites 41 and activated carbon for purication and environmental remediation. 42 It is outside of the scope of this article to cover all the applications of carbons and there are numerous reviews on the subject. [43][44][45] Instead, we will focus on several key applications that are the motivation of much of the work in the eld of catalytic graphitization. These applications exploit the many useful properties of carbons produced by catalytic graphitization, such as porosity, electronic conductivity, chemical stability, and tuneable surface chemistry. Another driving force in this area is the potential for catalytic graphitization as a sustainable route to carbons since iron can produce graphitic structures at relatively low temperature ($800 C).
One of the biggest areas of interest for carbons produced by catalytic graphitization is as electrocatalysts for the oxygen reduction reaction (ORR) in proton exchange membrane fuel cells. 46 The ORR occurs at the cathode, where oxygen is reduced by electrons and protons to produce water. This process commonly uses a noble-metal catalyst such as platinum, decorated on porous carbons. However, the cost and scarcity of noble-metals has led to a drive to nd noble-metal free alternatives. Numerous authors have turned to functionalised carbons, many of which are produced by pyrolysis of organic species. This simple process can lead to a wide range of carbons, and it is possible to introduce various chemical and structural features to optimise catalytic activity. This can be achieved through the addition of iron species and by incorporating nitrogen functionality via the organic precursor or additives. 47 There are various potential ORR-active sites in Fe/Ndoped carbons, such as Fe-N x , N/C, and Fe 3 C@C 48 and many of the materials rely on the benecial properties of graphitic nanostructures (such as high electronic conductivity and thermal/chemical stability) that are produced during synthesis.
Pure graphite remains the material of choice for lithium-ion battery anodes as it displays excellent stability and reversibility. 49 However, natural and synthetic graphite both have limitations in terms of energy density, cost, and sustainability so there are considerable efforts to develop alternatives. 50 Catalytic graphitization offers a promising alternative and has been used to convert biomass into graphitic carbon materials that can be used in lithium-ion batteries. 51 The appeal of this method is that enables a relatively low processing temperature and the opportunity to use a renewable feedstock. Catalytic graphitization also has the advantage of being able to convert non-graphitizing carbons into graphitic structures. There are only a handful of examples of iron-catalyzed graphitization being used to produce materials for lithium-ion battery anodes but given the wide interest in using biomass as a precursor for this application, 52 it is likely to become a much larger area of research in the future. 53 It should be noted that there are also many efforts to produce nanostructured graphitic carbons for other energy storage applications such as sodium-ion batteries and supercapacitors. 54 The structural and chemical requirements for these applications are different and there are now many examples of carbons for energy storage produced by pyrolysis or catalytic graphitization of a range of precursors. 55,56 A nal promising application for nanostructured graphitic carbons is for use as adsorbents for separation or purication. As mentioned above, activated carbons are already used widely in industry for removal of various impurities or contaminants from air or water. Recently, many authors have investigated graphitic nanostructures such as carbon nanotubes and graphene for purication and separation applications. 57,58 There are various ways that graphitic features are believed to facilitate adsorption of contaminants, including p-p interactions with aromatic functionality of organic pollutants 59 or binding of heavy metal ions to defect sites or to tube ends. 60 Indeed, carbon nanostructures with signicant sp 2 character such as nanotubes or graphene have been shown to have specic advantages over activated carbon. [61][62][63] However, the high cost of these specialist carbon materials limits their range of applications, particularly in environmental remediation. 64 Iron-catalyzed graphitization may offer an economical and sustainable solution, given the potential to transform bio-derived materials 65 or even waste biomass 66 into complex carbons with promising adsorption properties.

Production of carbon materials
This review focuses on the simple method of pyrolyzing organic molecules or materials with an iron precursor to generate graphitic carbon nanostructures. However, it is useful to give a brief overview of other approaches to synthesizing carbon materials as studying them can offer insight into the mechanism of iron-catalyzed graphitization. The rst known production of carbon materials dates to the paleolithic era, where early humans used charcoal for cave paintings. 67 Charcoal was made in simple earth-covered kilns, where wood is slowly burned with a limited supply of air. The development of charcoal is tied to the discovery of metal smelting. 68 Charcoal res burn much hotter than wood res, which are limited by the release of signicant amounts of water and volatiles. The heat and reductive environment of a charcoal re is what facilitates the extraction of metals from their ores. Charcoal has endured as a useful material to humankind due to its ease of production and high purity compared to coal and coke. 69 Many modern carbon materials are produced using technology that is fundamentally very similar to that of ancient charcoal kilns i.e. pyrolysis of organic matter in a low-air environment. However, modern furnaces enable much better control over heating conditions and purity of atmosphere. The main factor affecting the properties of the resulting carbon is the nature of the organic precursor. For example, carbon black is a ne powder with high surface area and electronic conductivity that is typically produced by the reaction of hydrocarbon fuels with a limited supply of air. 19 In contrast, biochars are produced by the pyrolysis of biomass in the absence of oxygen and tend to maintain the macrostructure of the biomass source. 70 Biochars have reasonably high porosity but carbons that are produced by pyrolysis of solid materials can gain higher porosity through 'activation'. This can be achieved through chemical (e.g. phosphoric acid, sodium hydroxide or zinc chloride) or physical (e.g. steam) treatment of the char. 71 Carbon black and activated carbon are two of the most industrially important carbon materials but there have been substantial efforts to develop routes to more specialist carbon materials such as fullerenes or nanotubes. One of the earliest examples is arc discharge. In this method, an electrical arc between two electrodes in a non-conductive gas is used to generate a plasma. The high-temperature plasma vaporizes solid graphite from the anode, which then deposits onto the cooled cathode as structures such as carbon nanotubes 72 and fullerenes. 73 The graphite anode can be doped with various metals such as nickel, iron or cobalt, which act as catalysts for the growth of nanotubes and can facilitate growth of specic structures such as single or double-walled carbon nanotubes. 74 Arc-discharge continues to be an area of interest for many researchers 75 and there is still much to be learned about graphitization in this system, but that is outside the scope of this review. Another technique for growing carbon nanotubes is laser ablation, which was developed to create more controllable conditions. Laser ablation was rst reported by Guo et al., 9 and uses a pulsed laser beam to vaporize a graphite target with embedded metal catalyst particles. Like arc discharge, the resulting carbon vapour is deposited onto a cooled substrate in the form of nanotubes and fullerenes. 76 Arc-discharge and laser ablation both require a lot of energy and are difficult to scale up. As a result, an alternative method called chemical vapour deposition (CVD) was developed to facilitate the industrial manufacture of carbon nanotubes. In addition to the lower energy requirements, CVD has the advantage of not requiring pure graphite as a reagent. Instead, a gaseous hydrocarbon such as acetylene ows over a twodimensional substrate. The substrate is coated in nanoparticles of a metal catalyst (typically iron, cobalt, or nickel) and when the hydrocarbon gas is heated, it decomposes and dissolves into the catalyst. Carbon nanostructures grow from the catalyst nanoparticles by various mechanisms such as oat growth or base growth (Fig. 4). Compared to arc discharge and laser ablation, CVD can produce a relatively high purity product in a high yield. By changing various experimental parameters such as reaction temperature or the chemical nature of the catalyst or substrate, it is possible to inuence the structure of the nal product and CVD methods are widely used in industrial manufacture of carbon nanostructures. It is worth noting that many of the methods discussed in this section employ transition metal catalysts such as nickel, cobalt or iron. Considerable effort has gone into elucidating the mechanism by which these catalysts produce carbon nanostructures. While the conditions of iron-catalyzed graphitization are very different from those of arc discharge, laser ablation and CVD, there is much we can learn from the extensive mechanistic studies of these processes, and we will return to this later in the review.

A brief history of catalytic graphitization
Humankind has been exploiting catalytic graphitization for thousands of years, long before the technology evolved for us to understand the chemistry. For example, pottery from an ancient settlement called Keeladi in India has a durable black coating that was found to be made up of multi-walled carbon nanotubes. 77 These are believed to have been formed during ring from carbonization of plant matter alongside naturally occurring catalytic iron species. In another example, ancient Damascus steel blades, renowned for their strength and sharpness, have shown a complex microstructure of carbon nanotubes and iron carbide nanowires. 78 Catalytic graphitization by iron may even occur naturally throughout the universe, with evidence that iron has driven the formation of graphitic structures found on meteorites. 79 In more recent years, there have been more deliberate efforts to understand and exploit the process of catalytic graphitization. Many of these have stemmed from an interest in improving the efficiency of graphite production by reducing the typical synthesis temperature of >3000 C. 80 The ability of 'impurities' to drive catalytic graphitization at much lower temperatures was discovered by Acheson 81 in 1896. Since then, many different elements and alloys have been shown to promote the transformation of amorphous carbon to graphitic carbon. 82 Early attempts involved the pyrolysis of mixtures of metal powders with amorphous carbons. 83 The differences in the type of carbons formed (graphitic or turbostratic/graphitic) were linked to two different graphitization mechanisms. The rst of these was proposed to be formation and subsequent decomposition of metal carbides and the second was suggested to involve dissolution of carbon into the catalyst, followed by reprecipitation. 84 The latter mechanism was believed to be driven by formation of a solution that is saturated with respect to disordered carbon but supersaturated with respect to graphite. 85,86 The precise composition of the carbon precursor was found to affect the degree of graphitization, but importantly, it was shown that even 'non-graphitizable' carbons could be graphitized at relatively low temperatures (<1400 C). 87 The specic phenomenon of iron-catalyzed graphitization has been studied in considerable detail since the early reports of catalytic graphitization. This is partly due to the importance of graphitization within the iron and steel industry. For example, in blast furnaces, iron ore is heated with a porous carbon such as coke, which acts as a reducing agent and energy source and is also crucial in maintaining the permeability of the reactor contents for upward owing gases. 88 For environmental reasons, there is a need to minimize coke consumption, which has driven investigations into coke degradation and transformation within blast furnaces. 89 One mechanism by which this occurs is the graphitization of coke carbon. A model study used to probe this system combined coke with a ne iron powder (particle size <5 mm). Heating the mixture resulted in the formation of graphitic carbon above 1200 C. 90 The process is believed to result from a dissolution-precipitation mechanism within molten iron catalyst particles.
Another observation that prompted study into iron-catalyzed graphitization was the unwanted deposition of carbon on metals exposed to carbon monoxide. For example, the conversion of carbon monoxide to solid carbon on iron surfaces was observed to cause deactivation of Fischer-Tropsch catalysts, 91 damage to brickwork in blast furnaces, 92 and carburization in heat exchangers of nuclear reactors. 93 While carbon deposition can be problematic in these circumstances, it was recognised that the same process may enable industrial production of useful carbons. In studying carbon deposition, many authors observed tube like laments, with small crystals of iron or iron carbide (Fe 3 C), indicating that the iron had catalyzed the lament formation. Work quickly moved to the deliberate synthesis of these tubular laments from decomposition of carbon monoxide 94 or benzene 95 over iron catalysts. This research led to development of CVD synthesis of carbon nanotubes, 96 as discussed above.
Alongside development of gaseous precursors for catalytic graphitization, there has been increasing interest in ironcatalyzed graphitization of non-gaseous organic precursors. This typically involves mixing solvated iron precursors such as ferrocene, iron nitrate, iron acetate or iron chloride with a solid or dissolved organic material. 83,97,98 The mixture is then pyrolyzed in an inert atmosphere to drive thermal decomposition of the organic matter to amorphous carbon. Alongside this, the iron precursor also decomposes. In many cases this results rstly in iron oxide nanoparticles, which are then transformed by carbothermal reduction to Fe or Fe 3 C. These particles then drive graphitization, which is believed to occur either by dissolution/reprecipitation of carbon or formation/ decomposition of carbides. Due to the relatively small size of the catalytic iron nanoparticles, the graphitization process occurs at the relatively low temperature of 800 C or 900 C. In many instances, tubular nanostructures are formed, suggesting that the iron-containing catalyst particles move through the amorphous carbon matrix during graphitization, perhaps driven by dissolution and reprecipitation of carbon. Multiple variations of this procedure now exist, using a wide range of organic precursors and these will form the basis of the next sections of this review. We will then return to the question of the mechanism of graphitization.

Organic precursors used in ironcatalyzed graphitization
This section will present the main categories of organic precursors that have been used in iron-catalyzed graphitization.
A few examples from the literature will be discussed to highlight the diversity of the method, focusing particularly on those in which effort has been made to explain or inuence the formation of the graphitic carbon product.

Small organic molecules
The simplest organic precursors used in iron-catalyzed graphitization are small molecules such as sucrose and glucose. These sugars are highly soluble in water so can be combined with aqueous iron salts such as iron(III) nitrate to produce a homogeneous solution or gel. This is then dried and pyrolyzed to yield graphitic carbon structures (Fig. 5a). An example of this can be seen in the work of Yang et al., who synthesized graphitic shelllike structures containing metal particles by pyrolyzing a mixture of sucrose and iron(III) nitrate above 700 C (Fig. 6a). 99 The metal nanoparticles can be removed by acid-washing to leave graphitic "capsules" (Fig. 6b), alongside a smaller number of graphitic nanotubes, highlighting the possibility of the formation of different nanostructures within the same product. Sevilla et al. found that a combination of iron nitrate and glucose could be used to produce lamentous multi-walled nanotubes, suggesting that the catalyst in this system is highly mobile (Fig. 6c and d). 100 It is difficult to know whether sucrose and glucose themselves inuence the process of graphitization as the metal : organic ratio and heating conditions are quite different in these two papers. Also, the sucrose was mixed with iron nitrate in water, whereas the glucose system employed ethanol as a solvent. It seems most likely, however, that the metal : organic ratio is the dominant reason for the difference in these systems. The 'shell' like structures Fig. 5 Schematic showing some of the methods that have been used to prepare nanostructured graphitic carbons from small organic molecule precursors such as glucose. These include (a) simple pyrolysis of a mixture of the organic precursor with an iron salt, (b) pyrolysis of the organic precursor followed by infiltration with an iron salt and a second pyrolysis step, (c) hydrothermal carbonization (HTC) followed by adsorption of iron species onto the resulting carbon spheres and (d) hydrothermal carbonization around iron oxide nanoparticles, followed by pyrolysis. were produced from a system containing 3 millimoles of iron per gram of sucrose and the nanotubes came from mixtures of 0.4 or 0.8 millimoles of iron per gram of glucose. The glucose system would therefore have a lot more amorphous carbon available, which could allow for substantial movement of the catalyst particles during catalytic graphitization before all the carbon is consumed. In a further example using aqueous glucose and iron nitrate, shell-like structures were observed rather than nanotubes. 101 Again, it is difficult to conclude anything about the mechanism as the authors also used NaCl as a sacricial template but the ratio of 1.7 millimoles of iron per gram of glucose is similar to the levels used by Yang et al. to produce shell-like structures. This lends credence to the argument that the metal : organic ratio is most important in dictating the type of structures formed in these systems. Another study that used sucrose, glucose and urea with iron acetylacetonate showed that all organic precursors resulted in carbons that show a mixture of shell-like and short tubular structures. 102 Some differences were observed in the iron/iron carbide composition and the graphite peak height in XRD but the suggestion overall is that iron : organic ratio has a more signicant impact on graphitization than the chemistry of the different small-molecule precursors.
Sugars have also been used to produce carbons where graphitic nanostructures are combined with macrostructural features such as pores or spherical shapes. The simplest approach to this is to initially produce an amorphous carbon and then introduce iron before pyrolyzing (Fig. 5b). This can be achieved just by pyrolyzing a sugar such as glucose to produce a porous amorphous carbon, 103 but more elaborate structures can be formed by templating sugars with amphiphilic copolymers before carbonizing them. 104 In the copolymer-templated example, the resulting amorphous carbon was then decorated with iron oxide nanoparticles by hydrothermal treatment and then pyrolyzed a second time to drive graphitization. Hydrothermal chemistry can also be used to produce the amorphous carbon precursor e.g. via hydrothermal carbonization of sucrose solutions to generate carbon spheres. The surface of these can be impregnated with iron nitrate, followed by pyrolysis to generate carbon spheres with a highly graphitic surface (Fig. 5c). 105 Hydrothermal carbonization can also be used to deposit carbon from glucose around iron oxide nanoparticles before pyrolysis to drive graphitization (Fig. 5d). This has the advantage of maintaining a relatively consistent catalyst particle size. 106 It is not particularly instructive to continue discussing all examples of iron-catalyzed graphitization that employ small molecules. A selection of examples from the literature are listed in Table 1 and it can easily be seen that a wide range of precursors and conditions have been used, even within this small area focusing on small molecules. It is clear from these examples that there is scope for tailoring graphitic nanostructures even in very simple systems but there is a need for systematic investigation.

Synthetic polymers
A large range of synthetic polymers have been used as precursors for graphitic carbons and these are summarized in Table 2.
As with small organic molecule precursors, the simplest way to produce graphitic carbons from synthetic polymers is to produce a homogeneous mixture of the polymer with an iron compound and pyrolyze in an inert atmosphere. For example, Huang et al. produced bamboo-like hollow carbon bres from an aqueous mixture of polyethylene glycol and iron sulfate. 116 Interestingly, these authors found that the carbon nanobre yield decreased with a high iron loading, suggesting that catalyst particle size may be important in regulating graphitization. Rather than mixing an iron precursor directly with the polymer, it is also possible to mix the iron precursor with a monomer and then initiate polymerization, before pyrolyzing the resulting material. This method has been used to produce graphitic carbon nanostructures from polyfurfuryl alcohol. It should be noted that the structures produced in these systems are turbostratic (i.e. no regular stacking between layers) rather than truly graphitic, as indicated by an absence of hkl diffraction in selected area electron diffraction (SAED) studies. 97 This is probably true of most of the examples in this section but is typically not stated by authors.
A common type of synthetic polymer used for catalytic graphitization is phenolic resins. These are thermosetting polymers that are generally synthesized from the reaction of phenols with formaldehyde and are commonly used as industrial adhesives. The type of structures and porosity of graphitic carbons produced from phenolic resins depends on the type of iron precursor (e.g. iron nitrate or ferrocene) 117 and also the method of mixing, with mechanical mixing of solid precursors favouring micropores and homogeneous mixing of solvated precursors favouring mesopores. 118 This is due to the homogeneously-mixed sample displaying more graphitic features, compared to the micropores of amorphous carbon. Organic gels formed from the reaction between resorcinol and formaldehyde have also been successfully graphitized by ironbased catalysts. Again, the polymerization is performed in the presence of the iron precursor, allowing a homogeneous mixture of the two components. An interesting example by Hasegawa et al. uses resorcinol, iron chloride and formaldehyde in an ethanol/water solvent. 119 The iron chloride acts as an acid catalyst to initiate polymerization and as the reaction proceeds, the polymer network becomes more hydrophobic, leading to spinodal decomposition and the formation of a macroporous structure. The pore size can be controlled simply by varying the ethanol:water ratio and the macroporosity is maintained in the graphitic product aer pyrolysis (Fig. 7a).
Another route to introduce porosity into graphitic carbons produced from synthetic polymers is to use hard or so templates. These can be sacricial or can themselves form part of the carbon product or the catalyst. Maksimova et al., 120 embedded needle-shaped iron hydroxide nanoparticles in polyvinylalcohol and polyethylene before pyrolysis under nitrogen. TEM images of samples heated to 600 C showed needle-like iron oxide crystals (from iron hydroxide decomposition) coated in a layer of amorphous carbon (Fig. 7b). Pyrolysis to 750 C results in carbothermal reduction of the iron oxide to iron and a markedly different structure (Fig. 7c) where the carbon has retained the needle-like structure but the catalyst particles have moved out of the carbon shells to form much larger, rounded particles. This suggests that the catalyst phase is highly mobile and the authors propose that this indicates a melting transition of the catalyst particles. Sevilla et al. produced mesoporous carbons by impregnating silica xerogels with a solution of phenolic resin in methanol, followed by pyrolysis and etching of the silica template with HF. 121 The resulting mesoporous carbons were then lled with ethanolic iron nitrate and pyrolyzed to 400-900 C before removing the metal catalyst particles by acid washing. The combination of graphitic features and porosity from the silica template led to remarkably high surface areas and bimodal porosity. Interestingly, these authors found that a lower pyrolysis temperature (#600 C) favoured graphitization, perhaps because high temperatures drove sintering of the catalyst particles. In a similar method, Li et al. synthesized a mesoporous silica template ( Fig. 7d) with embedded iron oxide particles that was inltrated with phenolic resin, pyrolyzed at 900 C and washed with HF to generate a mesoporous graphitic carbon (Fig. 7e). 122 The challenge with using silica as a template is the need for harsh treatment (HF or strong NaOH) to remove the silica. An alternative approach is to use so templates which can be combusted during the pyrolysis process. For example, Wang et al. used Pluronic P123 to introduce mesoporosity into a phenolic resin. 123 Pyrolysis without iron produced a carbon with regular porosity and a high surface area of 800 m 2 g À1 . However, much of this ordered porosity appears to be lost if an iron precursor is also included in the phenolic resin, with the formation of graphitic nanostructures and a lower BET surface area. Another system showed more successful retention of ordered mesoporosity. Li et al. employed a triblock copolymer Pluronic F127 to produce resorcinol-formaldehyde resins with ordered mesoporosity. 124 They discovered that a high iron content led to loss of the ordered porosity but that it could be maintained at a low iron:organic ratio (Fig. 7f), with metallic iron nanoparticles dispersed throughout the structure.

Biopolymers
Polymers derived from biomass (oen called biopolymers) offer an attractive alternative to synthetic polymers as they are renewable and could therefore offer a sustainable route to materials. Many biopolymers may also be waste products of industrial processes so producing graphitic carbons may be a way to add value to unwanted materials. In this section, we will consider both soluble and insoluble biopolymers as it is instructive to do so, and some may 'swell' rather than truly dissolve. In the following section we will discuss raw biomass. A summary of biopolymer-derived graphitic carbons can be found in Table 3.
The most abundant biopolymer found in nature is cellulose, which is the main component in the cell walls of green plants. Cellulose is a polysaccharide consisting of linear chains of b(1/4) linked glucose units. Cellulose molecules align to form strong microbrils, with individual cellulose polymers bound together by hydrogen bonds and it is networks and arrays of these microbrils that provides much of the mechanical strength of plants. Cellulose can be extracted from plant material (or from cellulose-producing bacteria) but is very difficult to solubilise, requiring specialist conditions such as ionic liquids 135 or concentrated NaOH at controlled temperatures 136 to be dissolved. All the examples of iron-catalyzed graphitization of cellulose therefore use pure solid forms of cellulose such as bres or microcrystalline powders.
A range of iron precursors have been used for catalytic graphitization of cellulose. Hoekstra et al. pyrolyzed microcrystalline cellulose spheres impregnated with three trivalent iron salts: iron(III) nitrate, ammonium iron(III) citrate and iron(III) chloride. 65 The study showed several interesting results. X-ray diffraction of all three systems showed similar compositions in the resulting carbons, with peaks for Fe and Fe 3 C and a broad peak corresponding to the interlayer plane spacing of graphite. However, only the iron nitrate and ammonium iron citrate showed the shell-like and tube-like graphitic nanostructures characteristic of most examples of catalytic graphitization. The sample prepared from FeCl 3 showed much less evidence of graphitic nanostructures in TEM images and appeared to have a high amorphous carbon content. This observation was supported by porosimetry data, which showed mesoporosity in both the iron nitrate and ammonium iron citrate systems but only microporosity in the FeCl 3 system (Fig. 8a, d, g). Investigation of the structure of the materials and the mechanism of formation offered insight into the reason for this disparity. Temperature-dependent X-ray diffraction data (Fig. 8b, e and h) showed that all samples initially formed iron oxide (Fe 3 O 4 ) but for iron nitrate and ammonium iron citrate the peaks are very broad indicating nanoparticles. In contrast, the Fe 3 O 4 peaks for the sample prepared from FeCl 3 were sharp, indicating large crystallites. In the nitrate and ammonium iron citrate systems, the magnetite nanoparticles are reduced to wüstite (FeO) nanoparticles (again indicated by very broad diffraction peaks) before reduction to Fe/Fe 3 C. In contrast, the sharp magnetite peaks in the FeCl 3 system are transformed to sharp wüstite peaks before reduction to Fe/Fe 3 C. SEM images support these observations, with carbons prepared from iron nitrate and ammonium iron citrate containing small nanoparticles of iron, indicated by the bright spots on the backscattered electron images (Fig. 8c, f, i). The carbon derived from FeCl 3 , however, shows very large iron particles, which appear to be located mainly on the carbon surface. The authors demonstrate that the reason for the very large particles is the volatility of the FeCl 3 precursor driving evaporation of FeCl 3 from the loaded microcrystalline cellulose spheres and deposition of large particles on the surface. The work provides further evidence that graphitization is closely linked to catalyst particle size. Interestingly, the iron chloride also inuences the carbonization process and particularly the formation of micropores because of its Lewis acid character, which 'activates' the carbon by promoting dehydration rather than depolymerization reactions. 137,138 Another study compared the graphitization of cellulose, starch and glucose, using iron nitrate as the iron precursor. 108 Cellulose and starch are both composed of polymers that are based on glucose monomers but while glucose is fully water-soluble, cellulose is insoluble and starch only swells to form gels. Pyrolysis of each of these precursors with aqueous iron nitrate resulted in carbons with very different structures, where cellulose and glucose carbons were mesoporous and the starch carbon was primarily microporous and with a much broader graphite XRD peak. Further study of the system showed that a mesoporous graphitic carbon could be formed from starch, but only by extending the pyrolysis time (Fig. 9). Small-angle Xray scattering data indicated that the iron carbide particles in the starch system grow very slowly, as can be inferred by the broad peaks in X-ray diffraction data. The authors postulated that the complex gel-structure of starch granules allows intercalation of the Fe 3+ precursor within the constituent amylose and amylopectin molecules (Fig. 10). The thermal stability of starch means that the organic network persists for a long time around the iron precursor as it gradually agglomerates to form iron oxide nanoparticles. Smaller iron oxide nanoparticles would in turn lead to small iron carbide nanoparticles. Glucose, being a small molecule, decomposes much earlier, which would facilitate mass transport of iron species through the developing carbon. In the insoluble cellulose system, the iron is dispersed over the surface, again facilitating mass transport and particle growth. The implication from this data is that there is a critical size of Fe 3 C particle that must be reached before graphitization can occur. This is something that has also been suggested by other authors. 116,118 The other major component of many plant-based biomass sources is lignin. Lignin is an abundant polyaromatic molecule and a waste product of the Kra wood pulping process, making it a popular candidate for graphitization. The challenge of working with lignin is that it is insoluble in water. To address this, several authors report graphitization of lignin by rst dissolving lignin in tetrahydrofuran (THF) and then adding aqueous iron nitrate to maximise homogeneity between the lignin and iron. 139,140 Aer drying the mixtures and subjecting them to pyrolysis under argon, shell-like encapsulated metal nanoparticles were found in the carbon products. 140 Similar shell-like nanostructures were reported by Zhang et al., but rather than solubilizing the lignin in THF, the authors added aqueous iron nitrate to lignin powder. 141 This resulted in a large polydispersity in the catalyst particles, reecting the fact that the iron would just coat the lignin particles rather than be dispersed throughout the lignin molecules. Interestingly, the authors found some evidence that the number of graphene layers around a catalyst particle may be correlated to the diameter of that particle.
There are many other biopolymers produced by both plants and animals and several of these have been employed in the production of carbons by iron-catalyzed graphitization. Most of these examples involve polymers that are soluble in hot water, producing gels that can trap aqueous iron precursors. The gels can then be dried in an oven to produce a dense material or freeze dried to give a sponge-like structure which is retained during pyrolysis. 142 An advantage of many biopolymers is that they contain functional groups which can facilitate dispersion of an ionic iron precursor. For example, agar is derived from seaweed and consists of two polysaccharides (agarose and agaropectin) which contain hydroxyl groups that can coordinate to aqueous Fe 3+ to produce a precursor for catalytic  graphitization. 143 Chitosan is another polysaccharide that been used to generate porous graphitic carbons. The advantage of chitosan is that each saccharide monomer contains an amine group, which can facilitate doping of nitrogen into carbons during pyrolysis. 142 A nal example is gelatin, which is a polypeptide derived from collagen. Gelatin forms sticky liquids and resins when combined with metal nitrates and these expand to form foams on drying. 144 Like the freeze-dried chitosan sponge, the foam-like structure of the gelatin-metal nitrate mixtures is also maintained during pyrolysis and if gelatin is combined with iron nitrate, the resulting carbon foam contains iron carbide nanoparticles coated in 'onion-like' layers of graphitic carbon. 145

Raw biomass
There is considerable interest in the conversion of raw biomass into functional carbon materials for various applications and there are some excellent reviews of this specic eld. 20,158 For catalytic graphitization using iron-precursors, one of the most interesting features of products derived from biomass is that they maintain their macrostructure during pyrolysis (Fig. 11a). This is not just aesthetically attractive. Biological materials encompass a wide range of complex and highly evolved structures and capturing these in a functional material can give unique properties. 159 For example, the interconnected pore structure of wood is maintained during iron-catalyzed graphitization, leading to advantageous properties for the resulting carbon in supercapacitors. 54 While normal (iron-free) pyrolysis will also maintain the macrostructure of biomass, the addition of iron catalysts introduces mesoporosity through the formation of graphitic nanostructures. Doping biomass with iron salts leads to the formation of Fe or Fe 3 C nanoparticles during pyrolysis. These nanoparticle catalysts drive graphitization, as in examples earlier in this review. However, in raw biomass, the catalyst nanoparticles oen appear to be highly mobile, driving conversion of the biomass-derived amorphous carbon into a dense network of graphitic carbon nanotubes ( Fig. 11b-d). 160 The fact that the Fe or Fe 3 C nanoparticles produce the graphitic nanostructures means that the size of the catalyst particles directly impacts the porosity of the resulting carbon. This can be observed in nitrogen porosimetry data of graphitic carbons produced from Fe(NO 3 ) 3 /sawdust at various ratios (Fig. 11e), where a higher Fe-content leads to larger pores. This can also be observed in TEM images, (Fig. 11f/g) which show much smaller particles at a lower Fe/sawdust ratio. Interestingly, the degree of graphitization does not seem to be affected by a signicant drop in iron content, possibly indicating that the smaller Fe/Fe 3 C catalyst particles travel further through the amorphous carbon matrix. The broad graphitic peak in X-ray diffraction patterns of these samples suggests that the graphitic carbon nanotubes have a high level of disorder, resembling a turbostratic rather than a truly graphitic carbon. In a similar system, Gomez-Martin et al. showed that pyrolysis of FeCl 3 -doped wood samples to higher temperatures (1000-2000 C) leads to a much sharper graphitic peak, indicating more regular graphitic structure. 51 There is a drop in the BET surface area at higher temperatures (200 m 2 g À1 at 850 C to 31 m 2 g À1 at 2000 C) but this is still higher than a control sample (no Fe) pyrolyzed to 2000 C, showing that some of the porosity can be maintained. This is reected in TEM images, which still identify graphitic nanostructures in FeCl 3 -doped wood pyrolyzed to 2000 C. A factor that must be considered in iron-catalyzed graphitization of biomass is the ability of the iron salt to penetrate the biomass structure. Many authors mill biomass to a ne powder before inltration with an iron salt to maximize coating of the biological material in iron. Hunter et al. demonstrated this systematically, showing that milling a wide range of biomass precursors before iron-catalyzed graphitization led to porous graphitic carbons with consistent adsorptive properties. Another consideration is the nature of the iron precursor. Liu et al. describe an intriguing process using FeCl 3 as a catalyst for carbon nanotube production from sawdust. 161 Their process involves a fast pyrolysis step, where powdered sawdust is loaded with an iron salt, dried, ushed with N 2 and then inserted into a preheated quartz tubular reactor (Fig. 12a). 162 The resulting carbon powder was covered in a dense mat of graphitic nanotubes (Fig. 12b). The authors demonstrate that the rapid heating leads to degradation of the cellulosic precursor into volatile low molecular weight hydrocarbons. These dissolve into surface Fe or Fe 3 C nanoparticles to drive the formation of graphitic nanotubes in a process analogous to chemical vapour deposition. Interestingly, Fe(NO 3 ) 3 and Fe 2 (SO 4 ) 3 catalysts did not produce the surface mat of graphitic nanotubes (Fig. 12c) and neither did CuCl 2 or NiCl 2 , indicating that the combined catalytic effect of iron and chloride is required for this phenomenon to occur.
Zhang et al. reported similar structures from the fast pyrolysis of FeCl 3 -treated rice husks. 163 It is possible that the unique behaviour of FeCl 3 derives from the Lewis acid character of this salt, which is known to change the decomposition pathway of cellulosic materials. [164][165][166] There are many other examples of raw biomass being used to produce graphitic carbons and it is difficult to draw further conclusions due to the large range of iron precursors and pyrolysis conditions used. For the benet of the reader, these are summarized in Table 4.

Organometallics
The examples discussed in the preceding sections have all used a separate organic and metal precursor. An alternative method is to use a single reactant containing both components, i.e., an  organometallic complex containing iron (Table 5). This route has been less widely explored but there are some examples of direct pyrolysis of iron complexes such as iron gluconate. 191 One challenge with organometallic precursors is their lack of thermal stability or even volatility. Some authors have resolved this by polymerizing the precursor, e.g., the catalytic crosslinking of ferrocene. 192 Another approach is to exploit the volatility of the precursor. For example, Leonhardt et al. used one furnace to drive sublimation of ferrocene before passing the vapour into a second furnace at a much higher temperature, where carbon nanostructures were deposited onto a substrate. 193 A similar approach has been used to produce graphitic nanocages by bubbling a mixture of nitrogen and carbon-rich gas such as acetylene through liquid iron pentacarbonyl and passing the resulting vapour through a tube furnace. 194,195 High pressure reactors have also been used for catalytic graphitization of ferrocene. 196 A nal note on the use of organometallic prearccursors is that they can be readily templated. Lee et al. mixed iron phthalocyanine with mesoporous silica (SBA-15) in a pestle and mortar. Pyrolysis resulted in sublimation and decomposition of the iron phthalocyanine and deposition of graphitic carbon and iron nanoparticles through the silica structure. 197 The silica can subsequently be removed with HF or NaOH and the resulting carbon retains the high porosity and surface area (877 m 2 g À1 ) of the template.

Mechanism of catalytic graphitization
One of the current drawbacks of catalytic graphitization is the limited understanding of the mechanism. As seen in the sections above, there are many examples of graphitic nanostructures synthesized by pyrolysis of organic and iron precursors. The choice of precursors and synthesis conditions vary greatly between different studies and can all signicantly affect the chemistry and structure of the graphitic carbon product. As a result, it is difficult to directly compare different systems or predict which may result in graphitic carbon materials with enhanced properties. Future development of the method of catalytic graphitization requires a more detailed understanding of the mechanism. While there are few studies on the mechanism of catalytic graphitization in pyrolysis, there has been considerable effort to understand the process of CVD synthesis of carbon nanotubes. 200,201 The processes are fundamentally different. CVD synthesis of carbon nanotubes involves decomposition of a hydrocarbon gas and dissolution of carbon into a catalyst particle. In contrast, catalytic graphitization involves decomposition of an organic precursor into amorphous carbon and graphitization of that solid carbon by a catalyst particle. However, it may be possible to gain insight into the mechanism of catalytic graphitization by studying CVD chemistry.

Insights from CVD synthesis of carbon nanotubes: nucleation and growth
As noted in section 2.3, CVD synthesis of carbon nanotubes can occur by oat growth or base growth. In both situations, the rst step is the initial nucleation of solid carbon from the catalyst particle. A popular theory for nucleation of carbon nanotubes on catalyst particles is known as the "yarmulke" mechanism, rst proposed by Dai et al. 202 The yarmulke mechanism involves the initial formation of a graphene 'cap' over the catalyst nanoparticle, which is stabilized by binding of the edges to the catalyst nanoparticle. Computer simulations have been widely used to probe the precipitation of carbon from saturated Fe x C y catalyst nanoparticles and the formation of the graphene cap as the rst step to SWCNTs. 203 As new carbon is introduced to the catalyst particle, the cap lis of the catalyst and elongates to form a SWCNT (Fig. 13a). Successful detachment of the cap and thus nanotube growth is dependent on multiple factors such as temperature and work of adhesion of the graphitic cap on the nanoparticle, which is governed by the interfacial energy and thus linked to chemical and physical properties of the catalyst particle. 204 Cap li-off and nanotube growth has been observed experimentally using in situ environmental TEM (Fig. 13b). 205,206 Simulations have also shown the possibility of cap formation followed by growth to form a continuous shell around the catalyst particle. 207 This is believed to occur when there is not enough energy to overcome the interfacial energy between the cap and the catalyst and correlates to experimental observations of graphitic shells coating catalyst particles in CVD. Factors that appear to contribute to shell formation include particle size (very small particles are not encapsulated), catalyst phase (Fe or Fe 3 C) and temperature (encapsulated particles tend to be found in the lower temperature regions of CVD ovens). Interestingly, there is evidence that the number of 'shells' of graphite around a catalyst particle is linked to the diameter of that particle. Given that the addition of carbons to the edge of an existing graphene sheet is energetically more favourable than nucleation of a new sheet, it is believed that the rst shell is completed before a new shell nucleates underneath. This would lead to exhaustion of the carbon supply in the catalyst, which could explain experimental evidence that encapsulated particles in CVD synthesis tend to be formed of Fe rather than Fe 3 C. 208 Catalytic graphitization leads to similar 'encapsulated' and tube-like structures, suggesting that there may be parallels with the processes of CVD. However, while some examples of catalytic graphitization produce mainly shells and others mainly tubes, there are also plenty of examples where a mixture of products is formed.
Another feature of catalyst diameter is that it inuences the diameter of the resulting carbon nanotube. This is believed to be due to cap li-off occurring aer the cap reaches a similar diameter to the catalyst cluster. 209 Cheung et al. demonstrated experimentally that monodisperse iron nanoparticles could be used to produce carbon nanotubes with controlled diameter via CVD. 210 These authors also note that an efficient carbon supply is required to drive the growth of large-diameter nanotubes from large catalyst particles. On the other end of the scale, catalyst clusters that are too small (<20 atoms) have been shown in simulations to produce poor quality carbon nanotubes, due to inconsistencies in binding between the cluster and the dangling bonds of the nanotube. 211 This reects some experimental observations that nanoparticles must reach a critical  3 800 Different iron loading and pyrolysis conditions investigated 'Graphite-shell-chains' 186 Wood, coffee, tofu residue, cotton Electron microscopy shows tube-like structures Graphitic structures 187 Oil palm frond Fe(NO 3 ) 3 1000-1400 Silica also added size before they become catalytically active in CVD. 212 A similar level of control of nanotube diameter has also been seen in catalytic graphitization. If catalyst particles are smaller, they produce nanotubes and shells of smaller diameter, which is reected in the porosity of the carbon material. 65,160 A point to note is that many authors have observed Fe or Fe 3 C nanoparticles to be 'liquid-like' during the graphitization process and very quickly agglomerate to form larger particles. Controlling this process may be a route to a wider range of graphitic products.

Insights from CVD synthesis of carbon nanotubes: types of growth
Nanotube growth in CVD can proceed in two different ways depending on specic reaction conditions. One possibility is tip-growth, otherwise known as oat-growth, in which the catalyst nanoparticle detaches from the substrate and leaves a carbon nanotube trail between itself and the substrate (Fig. 4). 213 Alternatively, nanotube formation can proceed via base-growth, in which the catalyst nanoparticle remains in contact with the substrate and the nanotube grows outwards from the nanoparticle. 214 The dominating mechanism is dependent on the interactions between the catalyst nanoparticle and the substrate, demonstrated by the work of Wang et al. in which iron was used as a catalyst and the nature of the substrate was varied. 215 They suggested that if the interfacial energy between the catalyst particle and the substrate is greater than the surface energy of the substrate itself, tip-growth would dominate, which they observed with a silica substrate. Conversely, with a tantalum substrate, the iron/tantalum interfacial energy is lower than the surface energy of pure tantalum so base growth was the dominant pathway. While non-carbon interfaces are generally not relevant to ironcatalyzed graphitization, it will still be important to consider the different interfacial energies that may contribute to catalyst movement.
A second important observation from CVD nanotube growth is that of continuous and bamboo-like carbon nanotubes. As the name suggests, bamboo-like carbon nanotubes resemble the structure of bamboo, with regular compartment-like graphitic structures (Fig. 14a). 216 This contrasts with the straight channel formed in normal carbon nanotubes (Fig. 14b/c). Bamboo-like carbon nanotubes are believed to form by periodic nucleation steps, where a series of graphitic layers form across the trailing edge of the catalyst particle during oat growth (Fig. 14d). The catalyst particle then detaches from the graphitic layers and moves forward before stopping to generate another nucleation point. 217 Simulations suggest that bamboo-like carbon  nanotubes form in conditions of higher carbon concentration, 218 which correlates to an in situ TEM study which identied Fe 3 C as the catalyst for bamboo-like carbon nanotubes and Fe as active for continuous carbon nanotubes. 216 Continuous and 'bamboo-like' nanotubes have also been identied in many reports of catalytic graphitization, sometimes within the same sample. 160 Given that many reports of catalytic graphitization identify both Fe 3 C and Fe in X-ray diffraction data, it is possible that the different materials drive two different types of graphitization. Some interesting insight comes from work by Ichihashi et al., who grew amorphous carbon 'nanopillars' by electron-beam-induced chemical vapor deposition on an iron-doped carbon substrate. 219 On heating the sample to 650 C inside a transmission electron microscope, they observed the formation of iron nanoparticles in the substrate followed by 'liquid-like' movement of particles outwards along the carbon pillars, leaving a graphitic 'trail' behind ( Fig. 15). In one case, they observed the liquid-like particle moving at constant speed along the carbon nanopillar to form a multi-walled nanotube. In another case, the catalyst  particle moved more slowly, and the tail of the particle periodically paused for several hundred milliseconds, forming graphene caps within the nanotube. Crucially, the nanoparticle catalysts never moved back along the pillar once graphitization was complete, indicating that graphitic carbon does not redissolve in the catalyst once it has formed. The authors do not propose a reason for the different behaviour, but they do suggest that the driving force for movement is due to the difference in solubility of amorphous carbon in iron compared to graphitic carbon in iron. A difference in solubility of amorphous carbon in Fe compared to Fe 3 C could potentially then be a reason for slower or faster movement of a catalyst particle through amorphous carbon.

The chemical nature of the catalyst
In CVD synthesis and in catalytic graphitization, the chemical nature of the catalyst is a matter of debate. The high activity of iron-based catalyst particles is commonly credited to the high solubility of carbon in iron at temperatures of around 700-800 C. One of the proposed mechanisms by which graphitization occurs is dissolution-precipitation. Here, the iron particle dissolves carbon atoms until it reaches a point of supersaturation, at which point graphitic carbon is precipitated from the particle. This hypothesis has been used to explain why transition metals with a low carbon solubility such as copper show poor catalytic activity for graphitization, while metals with high carbon solubility such as iron and nickel show high activity. 139 An alternative theory suggests that the formation of metastable carbides is crucial to graphitization. These decompose into a more thermodynamically stable metal species and precipitate graphitic carbon in the process. There is evidence for Fe 220 and Fe 3 C 221 being the active catalyst for CVD synthesis of carbon nanotubes and other authors have shown that both phases can be active catalysts, but for different graphitic products. 216 Fe and Fe 3 C are both phases that are commonly iden-tied in in situ X-ray diffraction studies of iron-catalyzed graphitization, oen within the same sample. 146,149,222,223 This raises the possibility that both dissolution-precipitation and carbide decomposition are potential mechanisms in catalytic graphitization. However, other pathways may also be possible. For example, a study by Yan et al. proposed that as iron carbide is stable at the temperatures involved in catalytic graphitization, it is unlikely that the formation of graphitic carbon is due to metal carbide decomposition. Instead, they suggested that the iron carbide and pure iron species may both act as dissolutionprecipitation catalysts, contributing to a high catalytic activity in iron species. 139 An important point to note is that the crystal phases present during catalytic graphitization may change as the sample is cooled. For example, in situ synchrotron XRD data from Gomez-Martin et al. showed that the onset of graphitization of a wood precursor corresponded to the appearance of an Fe 3 C phase. 222 During cooling, the Fe 3 C phase transforms to g-Fe then a-Fe, triggering a second carbon precipitation step. This change in the crystalline composition on cooling highlights the importance of studying graphitization in situ. Systems that are only studied aer removal from the furnace will offer limited information on the mechanism by which they were formed.
Another question that is linked to the chemical nature of the catalyst is how carbon is transported within the catalyst particles. Again, much of the evidence so far on this phenomenon comes from studies of CVD processes. For example, Yoshida et al. collected in situ environmental TEM footage of MWCNTs growing from Fe 3 C nanoparticles. 221 All the cylinder layers of the nanotube were observed to grow at the same rate, regardless of diameter, which strongly suggests migration of carbon atoms through the bulk of the catalyst particle. Other examples propose diffusion of carbon atoms around the surface of the catalyst nanoparticles. Again, there is a possibility that different mechanisms operate in different systems. In the complex world of iron-catalyzed graphitization, there may even be multiple mechanisms operating within the same sample.
The mechanism of iron-catalyzed graphitization can be further complicated when considering doping, or the presence of impurities within the system. Doping of Fe-C materials with heteroatoms has been an intense eld of study as it has led to improved performance in various applications. For example, Ndoped carbons that contain iron species are known to be highly active in the oxygen reduction reaction 224,225 and N-doped carbons are also of interest in lithium-ion batteries. 226 There is also considerable interest in phosphorus 227 and sulfur 228doped carbons. Given that iron can form alloys with nitrogen, phosphorus and sulfur, it is reasonable to assume that the presence of these heteroatoms will inuence the chemistry of Fe-C graphitization catalysts, whether or not the heteroatoms become integrated into the resulting graphitic carbon. While there are few studies on the effect of heteroatom doping on graphitization catalysts, it is clear that the presence of these species affects graphitization. For example, sulfur doping was shown to affect the number of layers in carbon nanotubes grown by CVD. 229 Sulfur has also been used to prevent agglomeration of iron catalyst particles during CVD, which in turn controls the diameter of the carbon nanotubes. 230 It is possible that iron sulde (FeS) or iron nitride (Fe 3 N) may act as graphitization catalysts in their own right. For example, there have been reports of graphitic structures forming around iron sulde nanoparticles. 231 A nal point that is worth noting is that different nitrogen-containing precursors can lead to different types of nitrogen feature in the resulting graphitic carbon, which suggests that the precursor can affect the mechanism of graphitization. 232

The physical nature of the catalyst
Determining the physical state of the catalyst particle during graphitization is also challenging. A commonly cited theory describing the physical state of the catalyst in CVD processes is the vapour-liquid-solid mechanism rst proposed by Wagner and Ellis to explain the growth of silicon whiskers. 233 This mechanism suggests that the catalyst particle must be in a liquid state to promote growth. However, the melting temperature of bulk iron is 1538 C, 234 much higher than typical reaction temperatures in CVD and catalytic graphitization. Despite this, many in situ environmental TEM studies have observed 'liquid like' behaviour. 235 Harutyunyan et al. proposed that the formation of a liquid phase is essential for growth of carbon nanotubes in CVD and the formation of a solid particle hinders growth. 236 Other authors, however, show evidence that the catalyst particle remains in the solid state (demonstrated by the presence of lattice fringes in in situ TEM images) and that the 'liquid-like' behaviour is due to the constant uctuation of the catalyst nanoparticle phase. 221 While many of the environmental TEM studies have probed CVD processes, similar liquidlike behaviour has been observed in the iron-catalyzed graphitization of cellulose (Fig. 16). The catalyst particle was observed to move through the solid carbon matrix with the front edge of the particle creeping slowly forwards and dissolving the amorphous matrix. The trailing edge of the particle moves in a very different way and stops periodically to allow build-up of graphitic layers before suddenly detaching and moving forward very quickly. This implies that the interfacial tension between the catalyst and the graphene surface is very different to the interfacial tension between the catalyst and the amorphous carbon. 147 The video footage showed that the particle stopped moving once the amorphous carbon was exhausted. Interestingly, the experiment also showed that some particles remained stationary throughout the experiment, which may indicate that some particles became encapsulated in graphite or that they were composed of an inactive phase.
Computer simulations have been used to offer insight into the physical properties of the graphitization catalyst. Methods such as molecular dynamics have shown that the melting temperature of metal nanoparticles can be considerably lower than the corresponding bulk material, raising the possibility that graphitization catalysts may be in a liquid or liquid-like state. 237 One model, carried out by Ding et al., suggests that both states are possible. 238 Their study modelled the growth of single-walled carbon nanotubes on both solid and liquid iron particles and found that the two routes had similar growth mechanisms. The main difference was that the main diffusion pathway of the carbon atoms in liquid nanoparticles was through the bulk, while surface diffusion dominated in solid nanoparticles. In real systems, it may be that both routes take place at the same time, because catalyst particles are unlikely to be completely uniform in size. It is possible that the smaller particles are in a liquid state and the larger particles in a solid state at the same temperature. The study by Ding et al. not only suggests that this is plausible, but that the resulting nanostructure may be the same in both scenarios. A further factor to consider is that some metals, including iron, are known to undergo surface melting. This may provide an explanation for the liquid-like behaviour of the catalyst particles in some systems. Ding et al. have used molecular dynamics to demonstrate that surface melting occurs at temperatures below the melting temperature in both free 239 and supported iron nanoclusters, 240 with the depth of the surface melt increasing with temperature until the cluster undergoes complete melting. Therefore, as well as both solid and liquid states, it is possible that the nanoparticles occupy 'in-between' states in the process of graphitization. Modelling can never take into consideration all the factors of an experimental study. One important consideration when considering the size effect is the introduction of a surface, which is applicable to both CVD and catalytic graphitization processes. While stronger interactions caused by a lower contact angle can decrease the melting temperature, 241 favourable epitaxy between the nanoparticle and the substrate can raise the melting temperature by hundreds of degrees. 242 This implies that the suggested variation in melting temperature may not be a purely size-based argument. There are many more challenges and discoveries remaining in this eld.

5.5
In situ methods for studying the mechanism of graphitization The need to understand the mechanism of catalytic graphitization has required the development of novel in situ methods. By following the graphitization reaction as a function of time and temperature, it is possible to determine the extent of graphitization and the presence of phase transformations or reactions between carbon and the catalyst. Two in situ methods, X-ray diffraction and transmission electron microscopy, have received the most attention. In the case of X-ray diffraction, one of the earliest accounts is from Fitzer and Weisenburger, who devised a special heating stage to follow graphitization of coke (without any catalyst) using a laboratory diffractometer. 243 More recent papers from Hoekstra et al. studied the in situ pyrolysis and graphitization of microcrystalline cellulose spheres with copper, nickel, cobalt and iron. 65,146,149 These studies used a commercially available heating stage on a laboratory X-ray diffractometer to identify the crystalline phases present during pyrolysis and conrmed the presence of iron oxide intermediates and both Fe 3 C and Fe during the graphitization step. The availability of synchrotron sources has enabled experiments with better time resolution. These experiments are technically challenging, due to the need to hold the samples in a quartz capillary in the synchrotron beam, while heating and maintaining an inert atmosphere. Some studies used nitrogen gas ow through 223 or around 244 the capillary while another partially carbonized the sample before sealing it within a quartz capillary. 222 In all these studies, a hot air blower was used to heat the sample to the required temperature (Fig. 17). Despite the very different set-up to a laboratory furnace, the capillaries appear to replicate the conditions of a laboratory furnace effectively. 245 There are many other synchrotron techniques that have been used to study carbon materials or graphitization by other metals. For example, synchrotron based XPS was employed to study the growth of carbon nanotubes using ZrO 2 nanoparticles as catalysts in a low-pressure CVD system. While Zr metal can catalyze graphitization at very high temperatures, 246 Steiner et al. found that the catalytic effect for nanotube growth of ZrO 2 took place without carbothermal reduction, with no visible changes in the Zr 3d XPS spectrum during synthesis. 247 Other authors have used in situ synchrotron smallangle X-ray scattering (SAXS) to probe the (catalyst-free) pore structure of polyacrylonitride-based carbon bres during graphitization. 248 Undoubtedly, similar methods will be applied to the question of iron-catalyzed graphitization in the future.
In situ or environmental TEM has also been used to study graphitization of carbon by iron and other metals. The challenge in these experiments again is replicating the laboratory furnace environment as closely as possible. Since electron microscopes operate under a high vacuum, it is necessary to be cautious when interpreting results from in situ TEM experiments as the materials may behave differently in the vacuum compared to the N 2 or Ar atmosphere of the laboratory furnace. It is also important to consider how the high energy electron beam may be interacting with the sample. The earliest example of in situ TEM investigations of graphitization come from Krivoruchko et al. 235 The authors dispersed iron hydroxide particles over an amorphous carbon lm on a TEM grid. Heating the sample inside the microscope led rst to reduction of the iron hydroxide and then movement of the resulting 'liquidlike' Fe particles across the carbon lm, producing a trail of graphitic carbon. Glatzel et al., used a similar approach but dispersed a sample of cellulose/Fe(NO 3 ) 3 on a TEM grid before heating in situ and recording video footage. 147 They observed very similar 'liquid-like' movement as the Fe/Fe 3 C nanoparticles graphitized the sample.

Other graphitization catalysts
Although iron is the most studied catalyst for graphitization, many other elements are known to drive graphitization, with transition metals and especially iron, cobalt and nickel being the most efficient. From these three, the highest catalytic activity is normally found for iron, followed by cobalt and then nickel, as evidenced by the interplanar distance of the (002) planes and Raman intensity ratio between D and G bands of the resulting graphitic carbon. 139,167,191,[249][250][251] The reason for this order is not clear, although some authors argue that is related to the number of electron vacancies in their d-shell orbitals. 82,139 Group VIII metals have a d-shell occupied by 6-10 electrons, with electron congurations of [Ar] 3d 6 4s 2 for iron, [Ar] 3d 7 4s 2 for cobalt and [Ar] 3d 8 4s 2 for nickel, and the energy levels of their electronic conguration would only slightly change by accepting electrons from carbon, allowing for the formation of covalent bonds and the dissolution of carbon by the metal. According to this criterion, the catalytic activity of transition metals should have the order iron > cobalt > nickel, in agreement with experimental evidence.
An additional factor for explaining the high catalytic graphitization efficiency of iron may be related to carbon solubility at high temperatures and the ability to form metal carbides. While iron can form a stable carbide with carbon, nickel and cobalt carbides are metastable, due to their weaker carbon bonds, with the order of the enthalpy being: Fe-C < Co-C < Ni-C. 252 The maximum solubility of carbon in nickel is about 0.6 wt% at 1327 C, while the maximum solubility of carbon in cobalt is 0.9 wt% at 1320 C. For iron, the maximum solubility of carbon is 2.06 wt% at 1153 C for the austenite (fcc) phase. This drops sharply to 0.02 wt% at 723 C, where the transition to ferrite (bcc) iron occurs, due to the smaller interstitial positions in the ferrite lattice. 253 According to the maximum solubility, the order is again iron > cobalt > nickel for the catalytic activity of these three metals. The abrupt decrease in solubility upon the austenite to ferrite transition may also help explain the higher efficiency of iron as a graphitization catalyst from a classic solution-precipitation view.
Other elements have shown to be active catalysts for graphitization, as reviewed byŌya and Marsh. 82Ō ya andŌtani studied the graphitization of formaldehyde derived carbon at 2600-3000 C by 10 wt% of Mg, Si, Ca, Cu, and Ge, which formed only graphitic carbon, and Al, Ti, V, Cr, Mn, Fe, Co, Ni, Mo and W, which formed both graphitic and turbostratic carbon. 83 In addition, B was found to markedly accelerate homogeneous graphitization, while Zn, Sn, Sb, Pb and Bi had no catalytic effect. Yokokawa et al. studied the graphitization of furfuryl alcohol-derived carbon at 1400-2300 C by copper compounds and copper metal. 87 Weisweiler et al. used monolithic glass-like carbon crucibles lled with different metals and found Ni, Co, Fe, Pt, Mo, Cr and B to be highly effective in catalyzing graphitization, while Ag, Mg, Zn, Cd, Ge, Sn, Pb, Sb, Bi, Se, Te and Pd showed no reaction. 85 More recent works have studied the catalytic graphitization of rare earth elements, 254 yttrium, 255 manganese 121 and magnesium. 256

Conclusions and perspective
Iron-catalyzed graphitization has the potential to be a scalable and economical route to carbons with complex graphitic structures. Natural graphite is listed by many governments as a critical material and there is a pressing need to identify routes to synthetic graphite materials. The appeal of iron-catalyzed graphitization is the simplicity. Organic matter is combined with an iron precursor and pyrolyzed in an inert atmosphere. For example, a complex network of graphitic nanotubes can be produced by heating a mixture of sawdust with iron nitrate. The method has proven to be extremely exible, and authors have utilised a wide range of organic precursors and additional templates to produce carbons with diverse properties. One challenge that now needs to be addressed is how to translate this fascinating system into real-world applications. Ironcatalyzed graphitization appears to be scalable and exible but there are very few examples of systematic or large-scale studies of this process. Without these, it is difficult to understand exactly how different precursors and heating conditions can be tuned to optimize the graphitic carbons towards certain applications. The few systematic studies that do exist have shown that it is possible to achieve dramatic variations in carbon structure through simple changes in precursors. 65,108 Another challenge in iron-catalyzed graphitization is understanding the mechanism. It is still not clear why some organic precursors produce graphitic shells while others generate graphitic nanotubes. It is possible that the different precursors or conditions lead to different catalytic routes. This may relate to the chemical nature of the catalyst (Fe or Fe 3 C) or to its physical state. When the catalyst does become mobile and produce graphitic nanotubes, it is not known why this movement is apparently random. If the movement of the catalyst particles could be understood and then controlled, it may be possible to further optimize the properties of the resulting carbons by introducing directional order between the nanotubes. On a simpler level, it is also important to gain a further understanding of how catalyst size inuences the graphitization process and thus the porosity of the resulting materials. It is clear from all the examples in this paper that there is still much to be discovered in the eld of iron-catalyzed graphitization. Increasingly sophisticated simulations, and synchrotron and electron microscopy experiments will undoubtedly offer insight into the nature of the catalytic process. However, there is also a lot of scope in simple, systematic experimental studies of different precursors and conditions.

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