Punica granatum L. is a tree belonging to the Punicaceae family grown in Iran, Afghanistan, Turkey, the USA and Far East countries. The world pomegranate production is estimated as 1.5 million tonnes annually (1). Pomegranate juice is popular and is claimed to include antibacterial, anticancer, antioxidant, antiallergic and anti-inflammatory compounds (2, 3). Recent reports also showed its potential use in the treatments of cardiovascular diseases and diabetes (4).

The pomegranate peel amounts to more than half of the weight of the pomegranate. It is a fruit processing-waste but it could be used as a source of antioxidants, phenols, flavonoids and organic acids. Extracts of pomegranate peels including rind, husk, pericarp and membranes are rich in polyphenols (ellagitannin and punicalagins), gallic acid, flavones, flavanones and anthocyanidins (5–7). These phenolic compounds and organic acids can be extracted by using different solvents. The efficiency of extraction is mostly dependant on the type of solvent, time and temperature. Therefore, it is important to determine the best solvent and extraction method to obtain the best bioactive compound rich extracts. Many extraction methods are reported for extraction of phenolic compounds from pomegranate by using ethanol, methanol, acetone, ethyl acetate and water (8, 9).

The presence of punicalagins, punicalin, ellagic acid and gallic acid in PPEs could determine their antimicrobial activities on microorganisms (10, 11). PPEs prepared by using ethanol, methanol or their mixtures with water were shown to be effective on Staphylococcus aureus, Enterobacter aerogenes, Klebsiella pneumoniae and Salmonella typhi strains (10). Abdollahzadeh et al. (12) reported that the methanolic extract of pomegranate peel exhibited antimicrobial activities against oral pathogens including S. aureus and Staphylococcus epidermidis strains. Moreover, PPEs containing polyphenols, tannins and other secondary metabolites showed effective antibacterial activity against shiga toxin producing E. coli (STEC) (13). Extracts from other parts of pomegranate (such as rinds, membranes and seeds) also had antimicrobial effects on S. aureus and Bacillus megaterium (14). To the best of our knowledge there is no report for evaluation of antimicrobial activities of different PPEs for Turkish Hicaz variety. Therefore, in the present study the most efficient extraction methods were investigated to obtain potential natural antimicrobial compounds from Turkish Hicaz pomegranate peel that can be used as a source of safe preservatives in the food industry. For this, different solvents (ethanol, methanol, either alone or in combination with acid, acetone and water) were used for extraction of phenolics and organic acids from pomegranate fruit peels obtained as a waste from fruit juice processing industry. Extracts were then screened for their antimicrobial activities against some important microorganisms.

2.1 Preparation of Pomegranate Waste

2.2 Chemicals

2.3 Preparation of Pomegranate Peel Extracts

2.4 Qualitative Analysis of Phenolics and Sugars

2.5 Quantitative Analysis of Phenolics, Sugars and Organic Acids

2.6 Antimicrobial Analysis

2.7 pH Measurements

2.8 Statistical Analysis

3.1 Phenolics

3.2 Sugars

3.3 Organic Acids

3.4 pH Values

3.5 Antimicrobial Activities of Pomegranate Peel Extracts

In the present work, PPEs were prepared from Turkish Hicaz pomegranate variety by using ethanol, methanol or their combinations with acid, acetone and water. All extracts were found to have antimicrobial effects on different bacteria and a fungus. Amongst the evaluated extracts, MA and EA exhibited the largest inhibition zones for all tested microorganisms. It was shown that high amounts of organic acids (for MA and EA) and phenolics (for EA) could be responsible for the antibacterial activities. Further studies are required to isolate other potential bioactive compounds of the PPEs and to identify their molecular mechanisms of action. In addition, potential PPEs from different varieties of pomegranate should also be investigated to obtain the most valuable bioactive compounds that could be used as safe food preservatives in the food industry.

The global effort to produce solar fuels by means of molecular photocatalysis continues to intensify following the establishment of basic design principles in the 1970s and 1980s (1–5). Many of the challenges encountered at the time remain relevant today, namely molecular systems for photocatalytic water oxidation or reduction often rely on a sacrificial electron source to drive the thermodynamically demanding multielectron reaction, most of which are not environmentally benign or renewable (6). However, a direct photocatalytic system would avoid the fabrication and systems costs required with photovoltaics coupled to electrolysis. Tremendous effort has been expended in an attempt to understand photocatalytic processes in solution either as a bimolecular process for homogeneous catalysis or on a semiconductor surface in a heterogeneous system (7–11). Over the years a vast catalogue of new molecular catalysts and sensitisers has been generated, a prototypic example being the well-known ruthenium water oxidation catalyst reported by Meyer et al., the “blue dimer” (12). Recent efforts have produced catalysts based on earth abundant metals such as iron and cobalt which typically underperform compared to prior systems, leaving the economic and environmental considerations for moving away from precious metal systems (platinum, palladium, rhodium or ruthenium) under some debate (13–15).

Sensitisers based on the thoroughly understood ruthenium(II) trisbipyridyl complex remain popular as a starting point for new avenues of research (16–19). More recent efforts have focused on covalently linking the catalyst and sensitiser to optimise charge separation (20). The fundamental principle of photocatalyst design where donor and acceptor are joined by a suitable covalent bridge is elegant and inspired by the photosynthetic reaction centre of green plants, but challenging to optimise in a working photochemical reactor, even where the molecular systems are relatively simple (21, 22). To date, practical implementations of molecular photocatalytic systems are scarce, despite the variety of systems proposed. Surface and solvent parameters introduce a large set of variables for optimising a photocatalytic reactor. At this point there are now an abundance of molecules and materials but developing a mechanistic understanding of such systems to aid design is still an ongoing process (23).

Over the past decade, research into dye-sensitised solar cells (DSSCs) has converged with heterogeneous photocatalysis (1, 24–27). The potential advantages include high atom efficiency for the catalyst, low cost assembly through solution-based processing, and devices that operate under ambient conditions. Dye-sensitised photocathodes have been constructed in photoelectrochemical devices to reduce protons to hydrogen, usually under a small applied bias for a half-cell in a photocatalytic system (28–31). These can be coupled to photoanodes which provide the electrons for proton reduction as a byproduct of water oxidation in a tandem device (31). The tandem system, where both anode and cathode are decorated with photocatalysts, removes the need for sacrificial reagents and enables a sustainable system (32). These dye-sensitised photoelectrochemical cells overcome many of the disadvantages of molecular systems in solution, while taking advantage of the catalogue of sensitisers and catalysts already available. Light absorption, charge transport and catalysis are separated between three tuneable components (33–35). Attention then turns to the interfaces between these components, for example, how photocatalysts attach and behave on the semiconductor surface, which requires considerable optimisation.

In this contribution we focus on optimising the proton reduction half reaction, generating hydrogen by means of new integrated photocatalysts adsorbed onto the surface of a transparent p-type semiconductor, nickel(II) oxide. A proof-of-concept for this approach was reported by us recently for two supramolecular photocatalysts based on a bipyridyl ruthenium photosensitiser coupled to either a platinum or palladium catalytic centre via a terpyridine or triazole bridging ligand (36). In this follow up paper, we explore a series of new photocatalysts with structural modifications made to the bridge and the catalytic centre, and evaluate the impact these have on the photoelectrocatalytic reduction of protons (Figure 1). Consideration is given as to how the photocatalysts bind to the electrode surface and interact with the surrounding environment. Device testing in a photoelectrochemical cell is used to determine photocurrent and hydrogen produced as a result, and stability is evaluated over prolonged periods of illumination. The photophysical and electrochemical properties of the photocatalysts are examined to develop a mechanistic understanding of the system.

Fig. 1.

2.1 Optical Properties

The preparation and characterisation of mesoporous nickel oxide cathodes has been described previously (37). Adsorption of the photocatalysts onto the mesoporous nickel oxide films on NSG TEC^TM 15 conductive glass from acetonitrile solutions was recorded by ultraviolet-visible (UV-vis) spectroscopy. Normally in titania-based systems the ester group is hydrolysed prior to adsorption or the titania is treated with base to promote binding (38), but for nickel oxide we have found that these ester-functionalised systems bind as well as the carboxylic acid derivatives. To calculate the dye loading, we assumed that the molar absorption of the dyes did not change significantly on the nickel oxide surface compared to the dye in solution. In dry acetonitrile, all five dyes produce steady-state absorption spectra with characteristic metal to ligand charge transfer (MLCT) bands from ruthenium to the diethyl ester bipyridyl ligands (39). When immobilised on nickel oxide films, the overall trend is towards a broader, red-shifted spectrum. While we do not have a model regarding what causes this spectral shift, it indicates that the dye interacts with the nickel oxide surface through the ester anchoring groups. This spectral broadening is typical for dyes adsorbed on metal oxide surfaces, and is usually observed when carboxylic acid anchoring groups are used and it is possibly caused by deprotonation of the acid on binding (40, 41). In this case, where ester anchoring groups have been used it could be due to hydrolysis or, possibly, the result of overlap from several different orientations of the catalyst on the nickel oxide surface, including some aggregates (42). The steady state absorption spectra are shown in Figure 2 and the data are summarised in Table I.

Fig. 2.

UV-vis absorption spectra of PC1–PC5: (a) in acetonitrile solution; (b) adsorbed on nickel oxide. Solution spectra contained micromolar concentration of dye

Table I

Steady-State UV-Visible Absorption Data for the Photocatalysts

Catalyst	(solution), nm	(film), nm	ɛ, M–1 cm–1 at (MeCN)	Dye loading, nmol cm–2
PC1	480	510	28,800	5.9
PC2	498	520	27,600	7.0
PC3	480	520	42,500	2.9
PC4	527	530	42,000	6.2
PC5	467	470	30,500	6.2

2.2 Transient Absorption Spectroscopy

2.3 Photoelectrocatalysis

2.4 X-Ray Photoelectron Spectroscopy

As with our previous studies with integrated photocatalysts based on a similar architecture, ester groups on the bipyridyl ligands serve as an anchoring site to nickel oxide (36). Our previous findings clearly suggested that there was sufficient interaction between the ester groups and the nickel oxide surface to provide a system resistant to dye desorption during photocatalysis (under suitable conditions). XPS results largely corroborate these claims. The XPS results show that the buffer salts can assemble on the electrode surface, and it is possible that there is competition between organic salts and the photocatalyst, leading to some desorption. We continue to favour ester substituted photocatalysts due to the simplified synthesis and purification of the dye. However, the diethoxy ester bipyridyl ligand serves two functions, both as the anchoring group to the nickel oxide surface and as an electron withdrawing ligand which stabilises the MLCT excited states. Locating the electron density close to the nickel oxide surface could increase the rate of charge-recombination, so the best photocatalysts should promote charge-transfer to the catalyst centre via the bridging ligand (20).

All the photocatalysts in this study produced photocurrent and hydrogen under each varied condition. Some photocatalysts perform better at pH 3, whereas others performed better at pH 7. The performance with pH 5 acetate buffer was the most consistent between different catalysts, but was generally lower than the other two buffers. Changing to MES buffer improved the performance at pH 5. Different authors report contrasting results when changing the pH of the electrolyte for analogous. The type of buffer affects the performance in two ways. Lower pH should increase the rate of hydrogen formation from the catalyst (50), however, the effect of pH on the valence band edge of the nickel oxide is a shift to more positive potentials as the H⁺ concentration increases (51). Further transient absorption spectroscopy studies are necessary to probe the effect of pH on the dynamics. It is expected that charge transfer from the nickel oxide to the photocatalyst would be slower at higher pH due to the lower valence band edge and, therefore, smaller driving force (52). Charge-recombination should occur in the Marcus inverted region and would be expected to be slower at higher pH, and increase the charge-collection efficiency of the device. However, in most kinetic studies, charge-recombination at the photosensitiser/nickel oxide interface appears to follow Marcus normal behaviour, accelerating with increasing driving force, probably due to recombination with more energetic intra-bandgap defect states (53, 54).

While this study shows that there are a variety of ways to improve the performance of dye-sensitised photoelectrochemical devices, including tuning the structure of the catalyst or the environment, the performance of dye-sensitised photocathodes based on nickel oxide still lags behind equivalent photoanode devices based on titania. Further improvements are necessary, for example, improving the porosity of the electrodes to facilitate mass transport. Additionally, bubble formation on the electrode surface leads to a drop in the photocurrent and good diffusion is necessary to prevent a build-up of OH^– in the pores. At the same time, high surface area must be maintained to adsorb enough photocatalyst to absorb all the incident light. A limitation of the ruthenium bipyridyl chromophores is the low absorption coefficient (ɛ = 27,600–42,500 M^–1 cm^–1 for PC1–PC5). Organic photosensitisers have two to three times larger absorption coefficients, enabling thinner or more porous films to be used (33, 55, 56). Future considerations for dye-sensitised photoelectrochemical assemblies should also consider the prospect of engineering the thermodynamic driving force for charge recombination towards the Marcus inverted region (similar to titania-based systems). Ongoing work within our laboratory looks to find an alternative semiconductor to nickel oxide with a more typical band structure to address this (32, 57).

The present study describes a series of novel sensitiser-catalyst dyads that were tested in a photoelectrocatalytic half-cell set-up to determine the optimum conditions for water splitting. At a range of pH levels the dyads remained intact with a stable photocurrent, demonstrating their robustness. However, when using a non-coordinating buffer we saw an increase in photocurrent. All our systems produced hydrogen under a small applied bias. Under optimal conditions, Faradaic efficiencies up to 90% with turnover numbers of up to 33 in an hour were achieved, which are comparative to our previously reported systems. Transient absorption measurements confirmed that charge separation on the surface of a p-type semiconductor, nickel oxide, is very efficient when compared to the same system in dilute solution. While rapid charge injection facilitates reduction of the catalyst, thus aiding photocatalysis, recombination of the oxidised sensitiser with nickel oxide remains equally efficient. The insight from transient absorption data in particular highlights a major challenge for integrated molecular photocatalysts on a semiconductor interface and further work will seek to address this.

Of the technologies discussed, NIR operates at the longest wavelength in the range of 700–1500 nm with the peak spectral output usually between 800–1200 nm. The lamps are typically tungsten–halogen filaments (16) and the emitted spectrum is dependent upon the power setting of the lamp. Heating the material is dependent on how the material absorbs the emitted radiation and the thickness of the material (16). The effect of reducing the lamp’s power output on the emission spectra is shown in Figure 3(a), where reduced lamp power moves the lamp output towards the higher wavelengths (18). This means that the energy imparted to the material is not a linear relationship with lamp power. The power output from a single lamp ranges from 6 kW for laboratory sized emitters to over 250 kW per emitter in steel coating lines (16). Maximum temperatures that can be achieved are dependent upon the lamps’ energy density (in the range of 1200 kW m^–2) and the materials to be heated. Typical working temperatures are up to 800°C (19), however machines are available that can heat materials >1000°C, for example, the adphosNIR^® 6 x 6 kW ceramic based NIR machine.

Fig. 3.

IPL, also known as photonic or flashlight, involves the use of an inert gas lamp (such as xenon) which converts short duration and high-power electric pulses into radiation. Millisecond (or shorter) pulses of light from a xenon lamp are produced by applying high voltage and current between two tungsten electrodes through an inert xenon gas which generates an arc plasma. A super charged capacitor is used to deliver high electrical current in a short time (milliseconds). Xenon is most widely used because it is the most efficient gas to convert the applied electrical energy to white light. The spectrum of the lamp ranges as broad as UV to NIR (150–2500 nm) but the majority is in the visible region, where conversion is 50% in the range 200–1000 nm.

The power provided by pulses of light is greater than if the same equivalent total energy was provided by continuous light. The shorter the duration of the pulse, the higher the pulse power (Figure 3(b)), allowing a much higher penetrating capability through the material compared to continuous light (20). Crucially the short duration of light pulses also reduces the time available for thermal conduction inside the material, as well as other processes such as oxidation. This enables very rapid heating of a thin layer to much higher temperatures than the equivalent total energy not pulsed, without significantly increasing the temperature of the bulk (20).

The benefits of intense pulsed light combined with the high absorption of metal in the visible region has lent photonic to the application of curing printed circuits on visibly transparent, polymeric substrates which cannot withstand high temperatures. As described by Kinney et al. in their 1969 patent (21), brief intense pulses of light directed onto a printed pattern composed of conducting metal or semiconducting metal precursor can be converted to an electrical circuit pattern adhered to the substrate. “The heat profile of the composite element are such that the circuit is formed and energy dissipated before the substrate can be charred or decomposed, allowing heat sensitive substrates to be used” (21).

UV radiation operating at wavelengths of 240–580 nm (Figure 3(c)) is utilised in the laboratory, with 365 nm being the most common (peak) wavelength employed. This is mostly implemented using mercury UV lamps. When electricity is passed through the mercury vapour in the lamp the excitation causes UV light to be emitted, with the wavelength being dependent on the pressure.

UV curing can be used to tailor the cross-linking and porosity of a material, often enhancing the conductivity of the chosen material without deterioration (3, 22). This was shown by Zhang et al. where the conductivity of a polyethylene(oxide)-based solid polymer electrode was improved after UV curing compared with conventional heating (23).

The irradiation of a photopolymer via UV curing can cause photoinitiated chain growth and photoinduced crosslinking (Figure 4) via free radical photopolymerisation, initiated by free radicals formed from photoinitiators (PIs). PIs generate active species either through unimolecular dissociation or bimolecular photoinduced electron transfer reactions upon exposure to UV or visible light (3, 25, 24).

Fig. 4.

By adjusting the light intensity or concentration of photosensitive compounds, the rate of initiation and polymerisation can be controlled. However, photoinitiated free radical polymerisation can be inhibited by oxygen which reacts with the carbon free radicals. In some cases UV curing of a material is performed in an inert environment (26, 27) to avoid this effect. The polymerisation process, in some materials, can also lead to reduced strength and increased brittleness unless prevented through the addition of additives or alterations to curing time (24, 25).

Radiative heating to initiate a chemical reaction is commonly seen during UV curing, involving the irradiation of polymers to form crosslinked network polymers. Commercial inks are available that have been developed for this purpose. Krebs et al. examined metal printed back electrodes for organic polymer solar cells that can be applied in a roll‐to‐roll process using UV-curable silver-based inks. Silver flakes and a UV-curable binder (EBECRYL^®, allnex, Germany) were mixed, screen printed and UV cured for the solar cell. Results showed current extraction was efficient over the full area of the solar cell that was coated in silver, and increased current in areas coated in silver compare to those without (48).

UV-curing in situ is often used to solve the solid‐solid interface contact problem caused by the conventional assembly process in the layer‐by‐layer fabrication of solid state batteries. In the case of in situ photopolymerisation the fluidity of the monomer allows the electrolyte precursor to penetrate the microporous structure of the electrode, allowing a strong adhesion to form with the electrode after polymerisation and greatly reducing the interfacial resistance (4). Kim et al. investigated the fabrication of bipolar all-solid-state lithium-sulfur batteries (ASSLSBs) via a UV curing-assisted stepwise printing process (2). Fabrication involved the ethyl methyl sulfone gel electrolyte, which incorporated a sulfur cathode paste being printed onto the desired object and subsequently solidified by exposure to UV irradiation. The tetraethylene glycol dimethyl ether gel composite electrode was then added to the assembly and cured by the same process; the lithium metal anode was then placed on top to create an ASSLSB unit cell. This printing and UV curing process was repeated to create a bipolar ASSLSBs structure with two thermodynamically immiscible gel electrolytes in the printed battery (2). This ASSLSB showed high cycling performance with a discharge capacity of 680 mAh g^–1 after 200 cycles.

Yang et al. used similar technology to prepare solid polymer electrolytes (SPEs) for all-solid-state LIBs (ASSLIBs) via UV curing using a one‐step process. The electrolyte was UV-cured in situ on the cathode using a UV lamp, facilitating free radical photo-polymerisation to create the three-dimensional (3D) polymer network structure (4). The electrolyte had three components: the (ethylene glycol)₉ methyl ether acrylate (mPEGA) matrix, ethoxylated trimethylolpropane triacrylate (ETPTA) cross linker and succinonitrile (SN) additive. Crosslinking of ETPTA and mPEGA via UV curing formed a 3D network. The polyacrylate backbone provided mechanical strength, while the amorphous polyethylene oligomer side chains facilitate the transport of lithium ions. The SN additive (uniformly distributed in the crosslinked network) accelerated ion transport by promoting the dissociation of the lithium salts (4). Characterisation indicated the cathode layer and the SPE layer were tightly coupled without visible delamination, and the polymer was successfully crosslinked. The in situ polymerisation of the electrolyte on the electrode reduced the interfacial resistance and polarisation, as well as demonstrating an improved electrochemical performance than equivalent conventional ASSLIBs (4).

Similarly, Imperiyka et al. investigated the performance of solid-state ion conducting polymer electrolytes, but instead as an application for DCCSs. The polymer electrolytes were formulated from poly(glycidyl methacrylate) in the presence of a photoinitiator 2,2-dimethoxy-2-phenylacetophenone. UV curing allowed the linear polymers to form rapidly and trap the redox couples, which is favourable in PV conversion devices. A homopolymer structure was formed which facilitated charge transport and ionic migration. In this research the conductivity of the linear polymer electrolytes was good compared to traditional processes (for example, thermal treatments and catalysed reactions) (22).

UV curing can also be utilised in proton exchange membranes (PEM) for fuel cell applications. Rao et al. applied the technique for Nafion^® PEMS in micro-direct methanol fuel cells, irradiated the PEM at wavelengths between 240–450 nm. Nafion^® PEMs consist of hydrophobic fluorocarbon backbone chains with perfluoroether side chains containing some strong hydrophilic sulfonic acid groups (3). The optimal dose of UV irradiation was determined by studying the membrane samples water uptake, swelling ratios, porosity and conductivity. Rao demonstrated that increasing the dose of UV radiation increased the proton conductivity, reaching an optimal dose of 198 mJ cm^–2 (3). The UV irradiation created crosslinking with the PEMs to form more hydrophilic sulfonic acid functional groups, in turn tuning the porosity of the Nafion^® membranes and increasing the proton conductivity (which doubled at that optimal UV dose). Polarisation plots also showed significant improvements in voltage as well as power density (3).

Other forms of radiation induced chemical reaction include perovskite crystallisation (Equation (i)) which has been achieved using both photonic sintering (49) and NIR (50). The absorber layer in perovskite solar cells, which is composed typically of methyl ammonium lead iodide or similar and absorbs strongly in the visible region, lends itself well to photonic processing. The heating rate must be controlled to ensure dense crystallisation and to avoid the evaporation of methyl ammonium iodide leaving lead iodide (Equation (ii)) rather than lead halide perovskite.

(i)

(ii)

The first photonic processing of perovskite films was demonstrated by Troughton et al. (49). They used a PulseForge^® 1300 to flash anneal spin coated CH₃NH₃PbI_3–xCl_x perovskite films on alumina/FTO glass. The mesoporous alumina acted as a scaffold layer for the perovskite where some perovskite film is embedded within it and there is a capping layer of perovskite on top. Perovskite films heated by 1.15 ms of photonic exposure had a poorer performance in solar cells than the standard films heated for 90 min on a hot plate (up to 11.3% compared to 15.2%) which was likely due to the capping layer becoming ejected and removed in places due to the near instantaneous phase transformation (49). Lavery et al. reported perovskite solar cells using photonic annealing of CH₃NH₃PbI₃ perovskite films made from the two-step deposition approach on titania/FTO glass (51). This involves depositing a PbI₂ film first (which they heated on a hot plate at 100°C for 5 min) and then dipping the film into a solution of CH₃NH₃I in isopropanol to convert the PbI₂ to perovskite. This was then exposed to flash white light to grow the perovskite crystals. With increasing white light pulse energy the cubical particles grew and became a dense compact film (Figure 7) (51). In this work the crystals were already nucleated and were being grown, whereas in the previously mentioned work (49) the film needed both nucleation and crystallisation and there would have been residual solvent in the film.

Fig. 7.

4. Practical Considerations

All radiative technologies suffer from a difficulty in measuring the exact temperature which the material being processed experiences. Thermocouples can be soldered to the surface of the substrate being processed (19) but inaccuracies are caused should the thermocouple react differently to the radiative heating compared with the substrate, and should be taken as comparative measurements within identical geometries rather than absolute values. For substrates where it is not possible to solder a thermocouple, for example glass, an IR camera can be used. When using an IR camera it is essential calibration is undertaken at exactly the same set up (such as camera angles) as for the actual measurement as described by Hooper (53). Roy et al. (54) when microwave sintering used both optical pyrometers and sheathed thermocouples close to the sample to measure temperatures and whilst these could give comparative readings within +/–5°C they could not confirm the actual temperature so precisely.

If the material contains a component which decomposes near the processing temperature (such as a polymeric binder) a temperature can be approximated, as in the case of Baker et al. (30) where thermal gravitational analysis was used to quantify remaining binder after sintering of a carbon cathode. Since binder removal was also time and energy input dependent it was not a quantitative assessment. Brennan (16) utilised heat equations to calculate temperature of spectrally selective surface coatings on any substrate material with good agreement to the experimental data of steel coating products.

In addition to difficulties determining the precise working temperature, there can be negatives to the fast heating rate with respect to thermal stresses. In cases where the material is not uniformly coated internal cracking may be initiated due to thermal stresses within the substrate, Figure 8. To mitigate this (if slower heating does not work) systems can be set up with multiple inline lamps to provide a preheat, heat and then post-heat such that cooling can be controlled and thermal cracking avoided. A simpler solution can be to use a substrate carrier that also absorbs the heating radiation to help reduce the thermal stresses within the substrate.

Fig. 8.

The rapid heating of inks with solvent remaining can also cause defects such as solvent boil (16, 28). The steel coating industry has highlighted this issue with some of their coatings noting this could be solved by using coatings designed with the right absorption properties, specifically for NIR processing (16). Huang et al. (55) demonstrate the importance not only of the absorbance of the substrate but also of the solvent when designing for NIR, with butanol showing particularly strong absorbance within the NIR wavelengths.

5. Summary

Radiative heating technologies have been embraced by industry to reduce processing times and energy input. However, some of the key advantages of these techniques have not been fully exploited. The precision of heating such that the coating can be sintered whilst the substrate remains cool is a key advantage in the field of printed electronics. But it has other advantages in that it can reduce diffusion between two reactive layers and enable component architectures not possible with conventional heating technologies. It is hoped that this study encourages material scientists to consider radiative response during their design of coatings in order to take advantage of these techniques, not just for efficient heating, but for new energy device designs.

Glossary

At present, it is evident that a significant reduction or, better yet, a complete eradication in the use of fossil fuels is required towards achieving a more sustainable future, and 195 countries have signed the Paris agreement as of January 2021 towards limiting the global temperature increase to less than 2°C this century (1). This commitment will be achieved by transitioning to renewable energy generation and phasing out petrol and diesel vehicles in favour of electric vehicles (EVs). In this scenario, battery technology will have a major role to tackle the energy crisis and environmental problems. To date, LIBs are the most widely used battery technology due to their superior gravimetric and volumetric energy densities and cycle life (2, 3). However, the increasing demand for energy storage devices has raised concerns regarding their sustainability and cost-effectiveness, leading to the resurgence of NIB research in recent decades (4–11).

Sodium is significantly more abundant and uniformly distributed in the Earth’s crust compared to the scarcer and unevenly distributed lithium (Table I) (11–16). Consequently, the cost of sodium salts is much lower compared to lithium salts (for example, US$154 tonne^–1 for sodium carbonate vs. US$8000 tonne^–1 for lithium carbonate in 2020) (12). Furthermore, the working principle of both systems is similar (see Figure 1), and therefore, NIBs are considered a drop-in technology that can benefit from existing LIB manufacturing facilities. Another advantage of NIBs is the possibility to use low-cost and sustainable elements in their cathode materials (for example, iron, manganese, magnesium and titanium), moving away from cobalt, which is widely used in LIB cathodes (for example, nickel manganese cobalt (NMC) materials) (17, 18). However, the overall lower cost of NIBs can be mainly attributed to the use of aluminium (US$0.3 m^–2) as the current collector for the negative electrode (anode) compared to the more expensive copper (US$1.2 m^–2) used in LIBs (17). Unlike lithium, sodium does not alloy with aluminium at low potentials, allowing its use at both electrodes (19). Furthermore, since aluminium is lighter than copper, its use boosts the total energy density of the NIB.

Fig. 1.

Some companies such as Faradion (UK), Hi-Na (China) and Tiamat (France), have already demonstrated the feasibility of NIB technology (20). For instance, in 2018, Faradion’s 12 Ah NIB full cells exhibited a maximum energy density of 140 Wh kg^–1, similar to the typical range of energy densities currently offered by commercially available lithium iron phosphate (LFP) // graphite systems (140–175 Wh kg^–1). Additionally, it was predicted that by the end of 2020, 160 Wh kg^–1 would be achieved in 32 Ah full cells, with cycle life matching that offered by some LFP // graphite cell suppliers (~3000 cycles) (20). Furthermore, current NIBs offer a 25–30% cost reduction in terms of US$ kWh^–1 compared to current LFP // graphite systems (20). Today’s commercial NIBs are already beginning to replace lead-acid batteries in low energy density applications, such as starter-lighting-ignition batteries in internal combustion engine vehicles and backup energy storage applications. In the near future, NIBs will likely compete with LFP // graphite LIBs for mid-range energy density applications, that require a high degree of safety or rate performance, such as electric scooters, buses, power tools and stationary energy storage systems supplying the grid (14, 20). However, current commercial NIBs cannot compete in terms of energy density with the state of the art NMC/lithium nickel cobalt aluminium oxide (NCA) // graphite lithium-ion systems in high energy density applications, such as long-range EVs. This is due, in part, to the absence of high-performance anodes that can compete with the current graphite-based anodes in LIBs. Despite this, the development of NIBs has been much faster compared to that of commercial LIBs since their conception (20). Therefore, NIBs may become competitive with high energy density LIB systems very soon.

Although the raw materials that make up a NIB are more cost-effective than the equivalent ones for LIBs, it is their cost as a function of energy density that truly matters (17, 21, 22). Hence, successful commercialisation of NIBs will depend on how much improvement is attained in terms of energy density, long-term stability and power density, while keeping production costs low. This will be partly achieved through the development of new materials and the optimisation of existing ones. Currently researched anode materials can be categorised into three groups based on their sodium storage mechanism: (a) intercalation/insertion (23–25); (b) alloying (26–29); and (c) conversion (26, 30). Figure 2 shows examples of materials in each group and their theoretical energy density when used in a full cell. The most used materials in today’s NIBs are those in the first category, which can achieve specific capacities within the range of 100–400 mAh g^–1. Examples include carbonaceous materials such as expanded graphite (31) and non-graphitisable carbon/HC, as well as titanium-based materials such as titanium oxide (TiO₂) and titanates (for example, Na₂O·nTiO₂) (32, 33, 34).

Fig. 2.

When used in NIB systems, graphite, the most commonly used anode in commercial LIBs, is unable to intercalate a significant amount of Na⁺ ions with commercial carbonate-based electrolytes, achieving a specific capacity of ~35 mAh g^–1 (35, 36). However, disordered carbons are able to reversibly store significant amounts of Na⁺ ions, especially HCs. However, there is no consensus on how Na⁺ ions are stored during the first discharge (i.e. sodiation process). This appears to be a result of the diverse structural properties exhibited by HCs when synthesised using different methods and precursors. In this review, we will summarise the current understanding of the sodium storage mechanism occurring in HCs, and draw correlations between the intrinsic structural properties found in these materials and their effect on the storage mechanism, as a means to advance in their optimisation. Finally, we provide a summary of the main challenges and possible directions towards the future optimisation of HCs.

In NIBs, graphite intercalates a negligible amount of Na⁺ ions when using commercial carbonate ester-based electrolytes, with specific capacities as low as 35 mAh g^–1 (NaC₆₄) (35, 37). Theoretical calculations show that the intercalation of Na⁺ ions into graphite to form graphite intercalation compounds (GICs) NaC₆/NaC₈, is thermodynamically unfavourable, unlike lithium insertion to form LiC₆ (38–40). Instead, it is more favourable for the Na⁺ ions to be deposited on the surface of the graphite as metallic sodium (37). This is often attributed to the larger ionic radius of Na⁺ vs. Li⁺ (Table I), which prevents sodium intercalation between the graphitic layers. Yet, sodium is the exception among all alkali metals, as MC₆ and MC₈ GICs are readily formed with heavier Group I elements (for example, potassium, rubidium and caesium) (41). Thus, it is clear that the ionic radius is at least not the only accountable factor for the unsatisfactory electrochemical performance of graphite in carbonate-based electrolytes. However, the possibility of the use of graphite as an anode in NIB systems has been reignited in recent years, as in 2014 Jache et al. reported the reversible co-intercalation of Na⁺ ions into graphite through the use of diethylene glycol dimethyl ether (diglyme) electrolyte solvent (42). A reversible capacity of ~100 mAh g^–1 was achieved with a high initial coulombic efficiency (ICE) of 90%, excellent cycling stability over 1000 cycles and high rate capability, maintaining ~75 mAh g^–1 at 1 C. This finding sparked further work exploring the Na⁺ co-intercalation mechanism at graphite (43–46). However, most of the capacity resides as part of a high voltage plateau in the discharge/charge profile between ~0.8–0.6 V (vs. Na⁺/Na), leading to low energy density values.

On the other hand, HCs are the material of choice among all anode materials currently tested in NIBs. They are already used in the first generation of commercial NIBs (20), despite exhibiting lower specific capacity compared to graphite in commercial LIBs (200–300 vs. ~372 mAh g^–1 respectively) (47–49). All the synthetic routes developed to produce HCs share in common a core pyrolysis step (i.e. high-temperature carbonisation using temperatures from 800°C to 2000°C) of an oxygen-rich organic precursor under an inert atmosphere (for example, argon or nitrogen) (50, 51). Although precursors can be synthetic, research mainly focuses on sustainable plant-derived precursors such as glucose, sucrose, cellulose and lignin (48, 52, 53). HC precursors cannot be converted into graphite, even at extremely high temperatures (>3000°C), and therefore, are often termed as non-graphitisable carbons (50, 52–54).

Structurally, HCs differ significantly from graphite, i.e. HCs exhibit a disordered structure, with short-range crystalline domains (Figure 3(a)) (55–59), while graphite exhibits long-range ordered stacking of graphene layers with a well-defined interlayer spacing (3.3 Å) (25). However, as the temperature of synthesis increases, HCs become structurally more ordered, with increasingly large graphitic domains (Figure 3(a)). Franklin first proposed a structural model for HCs (59), followed by Dahn et al. with the so-called ‘house of cards’ or ‘falling cards’ model (56–58). According to this model, HCs consist of randomly oriented graphitic domains (two or three stacked layers) with expanded interlayer spacing (3.6–4.0 Å), and different curvatures that are interconnected by highly disordered regions. This mismatch of ordered and disordered domains causes the presence of closed pores (dotted red lines in Figure 3(b)), between the randomly orientated graphitic crystallites (56–58). Also, HCs tend to contain a higher content of heteroatoms (mainly oxygenated groups) than graphite, providing more defects and sites for Na⁺ ion adsorption (52).

Fig. 3.

Although the typical specific capacity values achieved by HCs are significantly lower compared to alloying- and conversion-based anodes (300 mAh g^–1 vs. 2600 mAh g^–1 for phosphorus, for example) (Figure 2), they exhibit superior cycling stability due to minimal volume changes during cycling (~1–2.5%) (60). As such, HC containing cells can exhibit lifetimes up to 1000 cycles compared to 50 and 20 cycles for antimony and lead alloying materials, respectively, which show large volume changes in the order of ~300% during cycling (24, 26, 27). Additionally, the working potential of HC (0.1 V vs. Na⁺/Na) is one of the lowest among intercalation anode materials (Figure 2), which is a great benefit for attaining high energy density (61, 62). Lastly, they are cost-effective, enabling the production of batteries at a lower cost.

Nevertheless, HC materials face various challenges that need to be addressed before achieving their full commercialisation. To date, the sodium storage mechanism remains elusive, as many research groups have reported contradictory studies in the last years. Attaining this understanding is paramount to developing feasible strategies to accomplish superior electrochemical performance. The complex structure of HCs implies that there are multiple sodium storage pathways, and different synthesis methods or precursors result in different structural features. As a result, significant variations in the electrochemical behaviour are observed, making it difficult to establish a universal storage mechanism model. Therefore, it is reasonable to consider that multiple storage mechanisms exist, depending on the (micro)structure and properties of the as-synthesised HC.

Furthermore, cycling typically shows a low ICE induced by a large surface area exposed to the electrolyte (usually below 80%, (47, 48, 63) which contrasts to the 85–95% ICE values reported for graphite (63, 64)). This leads to an extensive reduction of the electrolyte during the first discharge, with subsequent formation of a solid electrolyte interface (SEI) layer, and partially irreversible trapping of Na⁺ ions in the structure (65–68). The composition of the SEI formed on graphite in LIBs has been extensively studied and optimised in such a way that parameters including cycling stability, rate capability and safety have greatly improved (69–72). However, similar studies in NIBs are still in their infancy, partly due to the higher instability of the SEI components in NIBs (73–77).

Extensive research efforts have been made in the last two decades to provide insight into the sodium storage mechanism in HCs. Many studies have reported that the structural and microstructural features of HCs, including pore size, pore volume and terminal groups, are influenced by the synthetic conditions (for example, carbonisation temperature, dwelling time and atmosphere) and precursors choice. This, in turn, has a direct repercussion on several parameters that define the electrochemical behaviour of HCs, including specific capacity, cycling stability, coulombic efficiency, rate capability, wettability, and ultimately, the storage pathway of Na⁺ ions. For reference, a typical galvanostatic discharge/charge voltage profile of HC at low current rates (for example, 0.1 C and 0.2 C) is shown in (Figure 4(a)). This consists of two distinct features: a sloping region at a potential above ~ 0.1 V vs. Na⁺/Na, and a plateau region below ~ 0.1 V (78). Given the complex structural nature of HCs, researchers concur with the existence of a multi-step storage mechanism upon cycling which consists of the following reactions (52, 53): (a) intercalation of Na⁺ ions between the pseudo-graphitic layers; (b) adsorption of Na⁺ ions at reactive surfaces and defect sites (for example, vacancies and dangling bonds at the edges of the pseudo-graphitic domains (78)); and (c) filling of closed nanopores with Na⁺ ions (see Figure 4(b)). Overall, five different mechanistic models have been proposed in the literature (Figure 4(c)–4(g)) (78–84).

Fig. 4.

Dahn et al. were the first to report a sodium storage model for HCs, which was based on an intercalation-pore filling mechanism (Figure 4(c)) (37, 79, 85). The sloping region was assigned to the insertion of Na⁺ ions into the interlayer spacing of pseudo-graphitic domains, and the plateau region to the insertion/adsorption of sodium into closed nanopores. Observations were supported by in situ wide and small-angle X-ray scattering data, wide-angle X-ray scattering (WAXS) and small angle X-ray spectroscopy (SAXS), respectively (37). WAXS data showed a progressive decrease in intensity of the (002) reflection (attributed to the interlayer spacing of the graphitic layers) in the sloping potential region during discharge. This is consistent with the introduction of Na⁺ ions between the layers. However, no accompanying shift in 2θ values was seen for the (002) reflection. This process was found to be partially reversible upon desodiation. Furthermore, SAXS data showed a reversible decrease in intensity of scattering in the plateau region, suggesting the presence of scattering species (Na⁺ ions) entering the closed nanopores. From this, they were able to calculate the difference in electron density between the surrounding carbon matrix and the closed pores. A partially reversible decrease in electron density contrast upon sodiation across the plateau region was observed, which was attributed to the filling of closed nanopores with sodium (37).

This model was further supported by Komaba et al. who used ex situ X-ray diffraction (XRD) to show a gradual expansion of the (002) interlayer spacing in the sloping region during sodiation, demonstrating the occurrence of an intercalation process in this region (86). They also monitored changes in electron density with ex situ SAXS upon sodiation, corroborating previous results from Dahn et al. for closed nanopore filling in the plateau region (37, 85). More recently, in situ electrochemical dilatometry (ECD) studies have shown that there is a non-linear expansion of HC during sodiation, which increases in the sloping region and then levels off in the plateau region (60). Furthermore, it has been shown that the interlayer spacing in HCs decreases with increasing carbonisation temperature (82, 87–90), which is followed by a decrease in sloping capacity. Overall, these findings provide evidence for an intercalation process occurring in the sloping region and pore filling in the plateau region in accordance with the original model proposed by Dahn et al (37, 85).

However, as HCs have become extensively investigated in recent years, alternative storage models have been proposed as a result of conflicting experimental data (Figure 4(d)–4(g)). As mentioned earlier, differences arise from the different choices of precursors or synthesis conditions which strongly influence the properties of HCs and their respective electrochemistry (82, 87, 89–92). For instance, some authors have observed a shift in the (002) reflection solely in the plateau region from ex situ XRD data (67, 78, 91, 93) and in situ XRD data (81), in discordance with data reported by Stevens and Dahn (37, 85) and Komaba et al. (86), while others have not observed any changes in the lattice spacing over the whole sodiation process from in situ XRD and transmission electron microscopy (TEM) data (82, 83).

It has been widely observed that the surface area and defect site concentration of HCs decrease with increasing carbonisation temperature (Figure 5(a)) (78, 89, 92–94). Many authors have also observed a simultaneous decrease in the specific capacity of the sloping region (Figure 5(b)), with many reports showing a positive correlation between a decrease in sloping capacity and a decrease in surface area and defect site concentration. This link between the surface area, defect site concentration and sloping capacity, with increasing carbonisation temperature, suggests an adsorption process of Na⁺ ions occurring at the open pores and defect sites present at the surface of HCs, as opposed to an intercalation process during the sloping region (Figure 4(d)–4(f)) (78, 89, 92, 94). Bommier et al. suggested that different defect sites may have their own Na⁺ ion binding energy, and therefore, sodiation potential, resulting in a sloping shape in the galvanostatic profile (78). However, it is noteworthy to point out that HCs with higher surface and defects may trigger more side reactions with the electrolyte, increasing the specific capacity along the sloping region during the initial cycles due to SEI formation, in addition to the pseudocapacitive contribution of the adsorption process (54, 90).

Fig. 5.

Conclusion

Low temperature fuel cells and electrolysers such as proton exchange membrane fuel cells (PEMFC), proton exchange membrane water electrolysers (PEMWE) and alkaline water electrolysers (AWE) are leading clean power conversion and hydrogen generation devices for the transport, stationary and industrial sectors. Anion exchange membrane fuel cells (AEMFC) and anion exchange membrane water electrolysers (AEMWE) are still at the research and development (R&D) stage, although significant advances have been achieved over the past 15 years, and component degradation has been identified as the limiting factor affecting their market deployment (1).

PEMFC, PEMWE and AEMWE contain membrane electrode assemblies (MEAs) where the electrochemical reactions take place. The MEA consists of a proton exchange membrane (PEM), catalyst layers (CL), gas diffusion layers (GDL), microporous layers (MPL) and transport porous layers (TPL). Typically, these components are produced individually and then pressed together at high temperatures and pressures. The electrodes are made of catalyst inks deposited either onto the GDL or the PEM leading to gas diffusion electrodes (GDE) and catalyst coated membranes (CCM) respectively. The catalyst ink is usually deposited by either the decal, blade process, screen-printing, painting, spraying (air and ultrasonic), electrospraying or electrophoretic methods (2–5).

There are numerous well-described methods detailing catalyst ink preparation, made for example of platinum supported on a high surface area (SA) and conductive carbonaceous material (for example, VULCAN^® XC-72, Cabot Corporation, USA, Ketjenblack^®) to yield platinum/carbon mixed with a solubilised polymer electrolyte (ionomer, such as Nafion^®), a solvent (either 1-propanol or 2-propanol) and ultra-pure water (Figure 1). To obtain an efficient homogeneous catalyst dispersion, the catalyst inks are thoroughly mixed (2–5) either by magnetic stirring, by high-shear mixing (using a rotor-stator mixer at high rotation speeds), by ball-milling (also known as bead milling, a batch process in which a slurry is mixed with zirconia balls), by hydrodynamic cavitation (6) or by acoustic cavitation (ultrasonic agitation) (7–29).

Fig. 1.

Scientists and engineers in a typical research laboratory fabricating MEAs for fuel cell and water electrolyser applications, most often employ ultrasound for catalyst ink dispersion by either using a laboratory-grade ultrasonic cleaning bath, or an ultrasonic probe (sonifier) without paying attention of the ultrasonic frequency and acoustic power used, and more importantly to the effects of ultrasound on the catalyst ink components, particularly the catalyst and the ionomer. Generally, acoustic cavitation is neglected by the fuel cell and electrolyser communities. This is because this area of research has not been extensively explored, although a few studies exist (7–18), and the use of ultrasound in chemistry or sonochemistry is still seen as a niche R&D area (20–34). Moreover, in many cases, the literature does not report on: (a) the ultrasound source type and make, the ultrasonic frequency, acoustic power and ultrasonication time; and (b) the catalyst ink temperature, which is often not controlled or regulated during the experiment. In this short review paper, we will attempt to: (a) highlight the important aspects, advantages, and disadvantages of using ultrasound for the homogenisation of catalyst ink slurries; and (b) present some recommendations.

However, before considering the use and effects of ultrasound on catalyst ink slurries, a section on ultrasound and sonochemistry merits some attention.

The use and application of ultrasound in chemical, physical and biological sciences can be divided into two distinct groups: (a) low frequency ultrasound or power ultrasound (20 kHz–2 MHz); and (b) high frequency ultrasound or diagnostic ultrasound (2–10 MHz) (30–34). Power ultrasound is regarded as the propagation and effect of an ultrasonic wave when transmitted through a liquid, leading to: (a) the creation of cavities (or voids) known as acoustic cavitation bubbles (microbubbles); as well as (b) acoustic streaming (29, 30).

Acoustic cavitation phenomena are usually associated with the formation, growth and collapse of cavitation bubbles induced by the propagation of an ultrasonic wave in a liquid, and consequently, to the generation of very high pressures, shear stresses and temperatures, locally. At standard temperature and pressure, the collapsing of acoustic bubbles leads to the formation of microjets of liquid being directed towards the surface of the solid material (Figure 2) at speeds of up to 100 m s^–1. It is well-accepted in the field that cavitation bubble collapse leads to near adiabatic heating of the vapour that is inside the bubble, creating a so-called ‘hotspot’ in the liquid, where high temperatures (ca. 5000 K) and high pressures (ca. 2000 atms) are generated with a collision density of 1.5 kg cm⁻² and pressure gradients of 2 TPa cm⁻¹, with lifetimes shorter than 0.1 μs and cooling rates above 10^9–10 K s⁻¹ during the collapse of cavitation bubbles.

Fig. 2.

Acoustic streaming can take place within three different regions: (a) in the bulk solution; (b) on the reactor walls; and (c) at the boundary layer. The power of acoustic streaming is directly proportional to the intensity of ultrasound, the SA of the ultrasonic emitting device and the attenuation coefficient of the medium. It is inversely proportional to the bulk solution viscosity and the speed of ultrasound (30).

Sonochemistry is a relatively new concept that received attention in the late 1970s and has been defined as the application of ultrasound in chemistry. In the late 1980s and early 1990s, the area was revived by Tim J. Mason (32) and Kenneth S. Suslick (34). A significant effect caused by acoustic cavitation is ‘sonoluminescence’ (33). Sonochemical reactions in an ultrasonic cleaning bath or with an ultrasonic probe (horn or sonifier) are most of the time subjected to transient elevated temperatures and pressures generated during cavitation bubble collapse. These cavitation bubbles usually contain gas molecules such as nitrogen and oxygen and vapour from the solvent. Here, water vapour is ‘pyrolysed’ into these ‘microreactors’ and dissociates to lead to the formation of extremely reactive species and other species such as hydroxyl radicals (OH•), hydrogen radicals (H•), hydroperoxyl radicals (•OOH) as well as hydrogen peroxide, ozone and oxygen atoms through endothermic chemical reactions (30, 33, 34) – a process known as sonolysis (Equations (i)–(iv)):

(i)

(ii)

(iii)

(iv)

These species diffuse out from the interior of the bubble into the surroundings and react with solutes present in the aqueous solution (30). The hydroxyl radicals are the most dominant species in sonochemical reactions. Since the standard reduction potential for the redox couple OH•,H⁺/H₂O (+2.730 V vs. SHE) is much higher than that of H₂O₂/H₂O (+1.776 V vs. SHE) (35), OH• plays therefore a more critical role in sonochemical reactions than hydrogen peroxide. Near the bubble wall, the concentration of hydroxyl radical is usually in the micromolar and millimolar range and the lifetime of these are about 20 ns (30).

Overall, it is well accepted in the fields of power ultrasound and sonochemistry that ultrasonic waves propagating in liquids lead to acoustic cavitation, acoustic streaming, sonolysis (in situ radical generation), areas of extreme mixing close to the ultrasonic source (transducer), degassing, surface cleaning (and surface erosion) and significant rises in bulk temperature (especially at low ultrasonic frequencies). Ultrasonics are therefore often used for deagglomeration and for reducing particle size, dispersion, homogenisation and emulsification, especially in the case of catalyst ink preparation. Figure 2 shows a summary of the physical and chemical effects of ultrasound.

In 2010, Takahashi and Kocha (7) described the importance of catalyst ink optimisation when evaluating PEMFC electrocatalyst activities towards the oxygen reduction reaction in liquid electrolytes. They highlighted the importance of producing good dispersion with the catalyst ink prior to deposition on carbonaceous or polymeric substrates to produce electrodes. In their study, good catalyst ink dispersions were qualitatively identified by ultrasonicating (ultrasonic bath; ultrasonic frequency and acoustic power not mentioned in the paper) the catalyst ink and allowing it to rest without mixing for a period. Electrochemical surface area (ECSA) was used as the quantitative parameter to decide whether the catalyst ink dispersion was optimal. In their study, they used several commercially available catalysts, namely, TEC10E50E (~46 wt% Pt/C), TEC10E50-HT (~46 wt% Pt/C-heat treated (HT)), as well as an alloy catalyst TEC36E52 (~46 wt% PtCo/C) from Tanaka Kikinzoku Kogyo (TKK, Japan).

The catalyst inks were prepared by mixing the catalyst powder with Nafion^®, isopropyl alcohol (IPA) and ultra-pure water using ultrasound. They observed that the important factor in the catalyst ink optimisation was the ratio of the solvent (IPA) to water (found to be ~35%) in the catalyst ink formulation as well as the energy and duration of the subsequent ultrasonic treatment. They demonstrated that for a given catalyst ink composition, ultrasonication durations of less than 5 min showed irreproducible results, and durations greater than 10–15 min were found to be sufficient for all the catalyst inks studied. Above 15 min and for up to 3 h, prolonged ultrasonication was found to produce no observable degradation in terms of loss in ECSA or particle growth in transmission electron microscopy, indicating a reasonably strong adhesion of platinum to the carbon support under these extreme conditions.

In 2014, Pollet and Goh (10) showed that the ECSA of a series of catalyst inks, made of commercial catalysts supplied by TKK (TEC10E50E, 45.9 wt% Pt/C) and E-Tek, USA, (HP ~50 wt% Pt/C) prepared in Nafion^®, IPA and water, were affected by ultrasound (up to 40 kHz), acoustic powers (up to 12 W) and ultrasonic exposure durations (up to 2 h). Two types of ultrasonic systems were used, namely: a 40 kHz ultrasonic bath (375H, Langford Electronics Ltd, UK) and a 20 kHz ultrasonic probe (Vibra-Cell VCX 750 with a tip diameter of 6 mm, Sonics & Materials Inc, USA). They also showed that the mechanically shear-mixed (up to 19,000 rpm) catalyst inks were not affected by intense agitation, but longer ultrasonication times were detrimental to the catalyst ink composition and morphology, due to erosion (jets of liquid of high velocity, up to 400 km h⁻¹ near the surface) and sonolysis phenomena induced by acoustic cavitation. It was observed that platinum nanoparticles were detached from the carbon support which led to a decrease in ECSA. It was speculated that a possible partial or complete platinum dissolution followed by an increase in the rate of platinum particle growth via Ostwald ripening and platinum agglomeration of platinum nanoparticles along with the action of Van der Waals forces occurred.

In later work, Shinozaki et al. (12) reported on the effect of cooling and type of ultrasonication, ultrasonication duration and power on the ECSA. They found that under ambient and ultrasonic conditions, heat was generated, and the temperature of the catalyst ink rose markedly leading to lower ECSA values by ~30%. They solved this issue by inserting the vials containing the catalyst inks in an ice-water ultrasonic bath during ultrasonication. In this set up, they observed smaller platinum nanoparticles and reproducible ECSA values. Figure 3 shows the changes in ECSA, SA and mass activity (MA) for two types of ultrasonic systems (bath – FS30H, Fisher Scientific, UK, output: 42 kHz, 100 W – and probe – S-4000, Qsonica LLC, USA, output: 20 kHz, 600 W maximum), at various output powers in the absence and presence of ice-water. It was found that increasing the output power led to a significant decrease in the catalyst SA, MA and ECSA even when the catalyst ink was placed in an ice-water bath. To obtain efficient catalyst ink homogenisation and to avoid catalyst ink degradation, they concluded that an ice-water bath is required for catalyst ink preparation with an ultrasonic bath and ultrasonic probe (lowest power settings) operating at ultrasonication times of ~20 min and 30 s respectively.

Fig. 3.

In 2019, initial studies were followed up and deepened by Argonne National Laboratory, USA, and National Renewable Energy Laboratory, USA, by using ultra-small angle X-ray scattering (USAXS)–small angle X-ray scattering (SAXS)–wide-angle X-ray scattering (WAXS) (14). Wang et al. (14) confirmed previous studies (7, 10, 12) which showed that combining very short horn (probe) tip ultrasonication (SFX 250 digital sonifier (Branson Ultrasonics Corporation, USA), 250 W; micro tip diameter = 3 mm; ultrasonic frequency and acoustic power not mentioned in the paper) followed by bath ultrasonication (FS30, 100 W; ultrasonic frequency and acoustic power not mentioned in the paper) was found to be more effective in breaking up platinum agglomerates, yielding maximum catalyst activity and MEA performance. However, it was observed that prolonged ultrasonication was too destructive and led to platinum nanoparticle detachment from the carbon black support in turn yielding poor ECSA values and MEA performance.

Very recently Bapat et al. (15) introduced new tools and approaches to investigate simultaneously the dispersion state and stability characteristics of the mechanically stirred (no ultrasound – or ‘silent’ condition) and ultrasonicated catalyst inks. The catalyst inks were ultrasonicated using either an ultrasonic bath (Elmasonic S 10 (Elma Schmidbauer GmbH, Germany), 50/60 Hz, 30 W; ultrasonic frequency not mentioned in the paper) or an ultrasonic probe (SONOPULS HD 2200 (BANDELIN electronic GmbH, Germany), amplitude of up to 70%, sonotrode tip = 5 cm long, diameter = 13 mm; ultrasonic frequency not mentioned in the paper). The catalyst inks were also mixed using a magnetic stirrer (IKA Works GmbH, Germany, 50/60 Hz, 50 W) at 500 rpm over 24 h. They found that probe ultrasonication led to a highly stable catalyst ink (only up to an optimum ultrasonic amplitude).

A paper entitled ‘A Comparison of Rotating Disc Electrode, Floating Electrode Technique and Membrane Electrode Assembly Measurements for Catalyst Testing’ by Martens et al. (23) describes detailed testing protocols for characterising benchmarked and newly developed catalysts, for different laboratories to easily compare data. In this work, ultrasound in the form of an ultrasonic probe or a sonifier (either a BANDELIN Sonopuls HD 3200, a Branson Sonifier 150 or a Branson Digital Sonifier 450, with a 3 mm outer diameter horn tip) operating at the lowest ultrasonic power was used to disperse the catalyst homogeneously.

In these works, either the ultrasonic frequency, the acoustic power values or both were not mentioned; but more importantly the sonifier horn tip was immersed directly into the catalyst inks (7, 14, 23, 25, 26).

It is well-known in power ultrasound and sonochemistry that inserting the ultrasonic horn tip directly into the solution leads to sample contamination. It is because the probe, or more correctly the ultrasonic horn is often made of a strong and corrosion resistant titanium alloy, Ti-6Al-4V, which is unfortunately subjected to erosion due to acoustic cavitation at its surface.

For example, Mawson et al. (36) showed that the production of metallic micro-particulates (titanium, vanadium, aluminium (and iron)) from ultrasonic tip erosion occurred and was more pronounced at lower (for example, 18 kHz with power input of 180 W, UIP 1000, Hielscher GmbH, Germany, and 20 kHz with power input of 103 W, Branson Digital Sonifier Model 250, USA) than at high ultrasonic frequencies (2 MHz).

The Pollet group has recently shown (37) that the sonochemical production of platinum from platinum(IV) by direct immersion of the ultrasonic horn (20 kHz ultrasonic probe, 50% amplitude, 43 W acoustic power, 700 W maximum power output, 19 mm probe diameter, Q700, Qsonica) into the reaction vessel led to: (a) contamination of the reaction solution with micrometre size titanium, vanadium and aluminium on which platinum nanoparticles were deposited (Figure 4); and (b) a faster reduction of platinum(IV) in the presence of these contaminants and at longer ultrasonication times due to the constant supply of titanium-, vanadium- and aluminium-particles being eroded from the ultrasonic probe. It was observed that the gradual introduction of these impurities also resulted in faster reduction rates of platinum(IV) through heterogeneous nucleation. It was concluded that direct sonication at lower frequencies (such as 20 kHz) should therefore be avoided if the target catalytic material needs to be of high purity.

Fig. 4.

When using ultrasound, apart from neglecting the physical effects induced by acoustic cavitation, the chemical effects, such as sonolysis (in situ radical formation), are often not mentioned by the fuel cell and electrolyser communities. During sonolysis, it is known that water molecules are dissociated into hydrogen radicals (H•) and hydroxyl radicals (OH•) according to Equation (i). Ashokkumar et al. (38) found that in water, the concentration of OH• increased with increasing ultrasound frequency from 20 kHz to 358 kHz and, then for higher ultrasonic frequencies, for example above 450 kHz, the OH• yield drastically decreased. They attributed this observation to the fact that at higher ultrasonic frequencies (in the megahertz region), the acoustic cycles are extremely short, in turn limiting the growth and eventual collapse of the generated cavitation bubbles, and therefore affecting water sonolysis and OH• generation.

The presence of these sonolytic radicals could also trigger the sonochemical production of metallic nanoparticles (8). In (37), it was shown that the ultrasonication at two ultrasonic frequencies (20 kHz ultrasonic probe, 50% amplitude, 43 W acoustic power, 700 W maximum power output, 19 mm probe diameter, Q700, Qsonia – 408 kHz ultrasonic bath, 100% amplitude, 54 W acoustic power, Honda Ultrasonics, Japan) of 2 mM of platinum(IV) chloride in 0.8 M 96% ethanol and Milli-Q water led to the formation of platinum(0) without the use of a chemical reducing agent (for example, sodium borohydride). Figure 5 shows the concentration of platinum(IV) in the solution as a function of ultrasonication time under 20 kHz and 408 kHz. The size of the nanoparticles synthesised sonochemically at ultrasonic frequencies of 408 kHz and 20 kHz were found to be 1.4 nm and 2.7 nm respectively, while the nanoparticles synthesised through sodium borohydride reduction were found to be 4.1 nm. Therefore, the sonochemical synthesis produced nanoparticles of smaller crystallite sizes than the chemical reduction method, which was possibly due to the deagglomeration induced by ultrasonication and higher cavitational events at that ultrasonic frequency. From the findings, it was clear that ultrasonication leads to complete reduction of Pt⁴⁺ in as short as 40 min of reduction time, via Equations (v) and (vi):

(v)

(vi)

Fig. 5.

Early preliminary results (not shown here) on 30 min ultrasonication (40 kHz ultrasonic bath, acoustic power = 8 W) of a catalyst ink (prepared following (39)) containing a commercial Pt/C in Nafion^®, IPA and ultra-pure water using a laboratory ultrasonic bath, indicated that dissolved platinum concentration (monitored by single particle inductively coupled plasma, sp-ICP-MS) decreased significantly up to 10 min ultrasonication and then increased until 30 min (below the initial platinum concentration at t = 0). The possible explanation for this observation could be due to the reduction Pt⁴⁺ to Pt⁰ then followed by a partial platinum dissolution caused both sonolysis and acoustic cavitation. This speculative argument may explain these early observations and a full and systematic study is currently underway.

The use of ultrasound in the generation and modification of carbon-based materials, for example mesoporous and macroporous carbon micro/macro-particles, carbon nanotubes (CNT), and graphene and graphene oxide dispersions has been studied (40). In these investigations, ultrasound has been shown to be very useful in scattering and solubilising CNT as well as directly functionalising their surfaces by breaking the C=C double bonds in the presence of strong acids (40). It has also been shown that ultrasonicated carbon-supported mono- and bi-metallic catalysts yielded excellent electrochemical activity due to surface functionalisation of the carbonaceous support and better dispersion induced by ultrasound (8). In a detailed study, Sesis et al. (41) showed that acoustic cavitation leads to chemical modification of the CNT surface and helps CNT exfoliation and length reduction, and efficient dispersion of CNT in aqueous solutions is mainly due to mechanical forces. To this date and from the authors’ knowledge, there have been no detailed studies on the effects of ultrasound on highly conductive carbon blacks of turbostratic structures with high SAs as support materials for fuel cell and electrolyser catalysts, such as VULCAN^® XC-72R (250 m² g^–1), Shawinigan Black (Chevron Phillips Chemical, USA, 80 m² g^–1), BLACK PEARLS^® 2000 (Cabot Corporation, 1500 m² g^–1), Ketjenblack^® (Ketjenblack^® EC-600JD and Ketjenblack^® EC-600J Ketjen International, 1270 m² g^–1 and 800 m² g^–1 respectively) and DENKA BLACK (Denka, Japan, 65 m² g^–1) (8).

Power ultrasound is also used for polymerisation and depolymerisation in polymer chemistry as it causes atomisation, molecular degradation as well as chemical rate and yield improvement. However, it has also been demonstrated in numerous studies that power ultrasound can greatly enhance the decomposition and degradation of some polymers due to the formation of radicals induced by sonolysis (42–46). Power ultrasound is now regarded as a powerful method for the depolymerisation of macromolecules, usually observed in the reduction of the polymers’ molecular weights, mainly caused by acoustic cavitation (45, 46). In all these studies, long-time ultrasonic irradiation of the polymer led to a permanent reduction in solution viscosity, which was in most cases, irreversible (46).

The Holdcroft (16) and Pollet research groups (9, 16) undertook systematic investigations that were performed to shed some light on the effects of power ultrasound on a series of Nafion^® ionomer dispersion concentrations under various ultrasonication durations at 42 kHz (acoustic power = 2.1 ± 0.2 W, 1510 Ultrasonic Cleaner, Branson Ultrasonics Corporation) and using a simple laboratory ultrasonic cleaning bath. They found that, under controlled bulk solution temperature (298 K), as the ultrasonication irradiation duration increased, the Nafion^® dispersions’ viscosity decreased; however, this effect was more evident for all samples treated under unregulated bulk solution temperature and lower dispersion concentrations. For these samples, mass loss was observed possibly due to a more defined cleavage of the acidic side chains.

From these early studies, in 2020, Safronova et al. (17) showed that ultrasonic treatment (35 kHz ultrasonic bath, ultrasonic power = 4 W, BANDELIN electronic GmbH) of Nafion^® polymer solutions led to: (a) an increase in proton conductivity of membranes obtained via casting procedure up to 40% after ultrasonication for 30–45 min; and (b) a deterioration of the mechanical properties as compared with the initial membrane. Very recently, Safronova et al. (18) also demonstrated that ultrasonication (20 kHz ultrasonic probe, ultrasonic powers = 2.7–9.4 W, Vibra-Cell VSX 130, Sonics & Materials Inc) of Nafion^® polymer solutions affected the microstructure and transport properties of the resultant membranes due to the improvement of the connectivity of pores and mobility of charge carriers. The ultrasonic treatment resulted in both an increase of conductivity and a decrease in the selectivity of cation transfer in recast Nafion^® membranes.

It is well known that low frequency ultrasound yields rapid temperature rises within the solution with ΔTs of up to ~+50ºC in short exposure times starting from room temperature. In other words, the bulk solution ‘heats’ up. Although this can benefit endothermic reactions as well as reaction kinetics, some issues may arise regarding data reproducibility. It is therefore important that the bulk temperature is carefully regulated and controlled during ultrasonication to separate the effects of acoustic cavitation from temperature effects. Moreover, a fast temperature rise often leads to volatilisation of the analyte (water, organic solvent) in other words to molecular degradation. In addition, as the temperature is increased, the physical characteristics of the solution are affected, and no acoustic cavitation is observed — this phenomenon is known as the ‘decoupling effect’. In the case of catalyst ink homogenisation and in the literature, ultrasonic time is stated and varies from 5 min to 24 h ultrasonication from one study to another.

Temperature during ultrasonication has therefore a major effect on the homogenisation efficiency through increased vapour pressure and changes in solvent and bubble dynamics. There are several strategies to control the ultrasonicated the catalyst ink slurry. The first is to use an ice bath (to ensure rapid heat dissipation), in which the glass vessel containing the catalyst ink slurry is inserted. Although a simple method, the main issue is that the ultrasonic energy is absorbed by the ice, and thus, the ice needs to be replaced if longer ultrasonication times are applied. The second option is to use specially designed reactor cells, for example, coolant-jacketed cells (linked to a thermostatic bath or a cryostat), and the third is to use the ‘pulse’ mode of ultrasonication (only available in newer ultrasonic systems).

8.2 Sonochemical Efficiency

8.3 Electrical Power vs. Acoustic Power

9.1 Ultrasonic Cleaning Bath

9.2 Ultrasonic Probe

9.3 Glassware for Ultrasonication

Apart from using acoustic cavitation for effective homogenisation (7–29), there are other methods for thoroughly mixing (2–5) the catalyst inks by either magnetic stirring, ball-milling, hydrodynamic cavitation or high-shear mixing (6, 47–49). For all techniques, studies on the effect of high shear mixing on the performance of catalyst ink slurries are scarce. High shear mixing appears to be a technique that is used in some laboratories for homogenising catalyst ink slurries. It involves using a high-shear mixer (such as Silverson, Heidolph, IKA) containing a metallic rotating blade operating at rotation speed up to 20,000 rpm (10, 13, 14). Fairly recently, Jacobs (13) showed that high shear mixed catalyst inks are more effective at higher stirring rates as high mixing energies lead to effective production of catalyst/Nafion^® interfaces, in turn improving the so-called ‘three phase boundary’ observed during in situ fuel cell testing.

Two types of cavitation exist: acoustic cavitation and hydrodynamic cavitation. The latter is caused by flowing liquid static pressure drops and, as the liquid flow passes through constricted geometries (for example, an orifice plate, a venturi or a throttling valve), the flow rate increases and in turn a decrease in static pressure is observed. When the pressure drops below the local saturated vapour pressure, cavitation nuclei present in the liquid start to grow as their internal pressures become greater than their surface tensions. Eventually they become unstable as soon as the flow pressure recovers and then implode by creating jet of liquids of high velocity (48, 49). In hydrodynamic cavitation, the resultant bubble collapse intensity is low compared to acoustic cavitation. But in terms of developing a large-scale system, it is comparatively easy (tank, pump, control valve and pipes/tubing) and maintenance is also easy. For acoustic cavitation, the acoustic field is usually not uniform when the size of the reactor increases (in terms of diameter as well as length) and hence uniform cavitation field is difficult to obtain. Overall, effective cavitation regime is reduced, although the collapse intensity of the bubbles is high.

In general, for achieving the desired physical effects (mixing, dispersion, extraction, cleaning, homogenisation, deposition, coating) acoustic and hydrodynamic methods are powerful and in terms of energy efficiency, hydrodynamic cavitation is better (see Table I (49)). From experience, the hydrodynamic cavitation method is superior for physical effects whereas acoustic cavitation is more suitable for chemical effects (especially when scaled up).

In 2019, for the first time, Kuroki et al. (6) used the hydrodynamic cavitation method to efficiently disperse PEMFC catalyst ink slurries. In their experimental set up, the catalyst slurry was subjected for 30 min to hydrodynamic cavitation bubbles generated by rotating the impeller in a centrifugal pump at a frequency of 80 Hz (ω = 4800 rpm) operating at ~1.3 kW power output. They also used the batch-type ball-milling method, produced the electrodes (CCM prepared by the decal method)/MEAs and carried out the PEMFC testing. They found that MEAs prepared using catalyst ink dispersed by hydrodynamic cavitation and ball-milling methods exhibited higher fuel cell performance than those prepared by the ultrasonication method (Figure 8). They suggested that the insufficient dispersion of platinum/carbon catalyst ink slurries was due to the poor breakup of the catalyst agglomerates (mainly from carbon) and cracks on the CCM surface during ultrasonication leading to an unfavourable pore structure for oxygen transport in the CL (Figure 9).

Fig. 8.

Fig. 9.

The answer to the question of this short review paper title is: “Yes, the approach is still acceptable”, however special care and great attention should be taken before performing the ultrasonic experiments. We recommend the following:

Low frequency ultrasound provided by either an ultrasonic cleaning bath or an ultrasonic probe involve acoustic cavitation, in which cavitation bubbles are created. When these bubbles grow in size, they become unstable and then violently collapse creating localised transient high temperatures and pressures, together with the formation of a high velocity jet of liquids directed to surfaces (mainly responsible for surface erosion and activation). Catalyst ink dispersion is an important factor if better fuel cell and electrolyser catalyst utilisation and thus performance are to be achieved. There are various methods that are being used to efficiently homogenise catalyst ink slurries, such as magnetic stirring, high-shear mixing, ball-milling, hydrodynamic cavitation and acoustic cavitation (ultrasonic agitation). The latter method has been extensively used in fuel cell and electrolyser laboratories, as most of them possess cheap and available ultrasonic cleaning baths. However, ultrasound should be used appropriately and with great care to avoid catalyst dissolution and ionomer degradation induced by acoustic cavitation as well as metallic contamination originating from the ultrasonic probe. Moreover, for laboratory and data comparison purposes, the ultrasonic equipment (and make), ultrasonic frequencies, acoustic powers, ultrasonication durations and sonochemical efficiencies need to be clearly reported. Figure 10 shows a graphical summary of the effects of ultrasound on catalyst ink slurries.

Fig. 10.

The word graphene refers to a single atomic plane of graphite and the word graphite has its origin in the Greek word ‘Γραφɛιη’ (pronounced as graphein), which means to draw or to write (13). Graphite is one of the first known materials to be studied using X-ray diffraction (XRD) by A. W. Hull in 1917 (14). He proposed that the normal structure of graphite has hexagonal D_6h symmetry (14). Later this structure was confirmed by J. D. Bernal, O. Hassel and Z. H. Mark in 1924 (15, 16). According to their model of graphite, it consists of a stack of parallel hexagonal net planes of carbon atoms with C–C bond length of 1.42 Å and these net planes are separated by a distance of 3.35 Å and the unit cell contains four atoms in a planar stacking sequence ABABA… However, later reports based on X-ray spectroscopy accounted for these hexagonal planes being stacked in ABCABC… sequence (17–22). These reports revealed that in ordinary temperatures and pressures natural graphite occurs in two crystal structures: (a) Bernal with an ABABA… sequence of arrangements and (b) rhombohedral with an ABCABC… sequence of arrangements (23).

During the early period of research on graphite, many studies on its physical properties were also reported. For example, in 1915 G. E. Washburn measured the electrical resistance parallel to the basal plane of a perfect sample approximating to single crystals or highly oriented polycrystalline samples (24). Similar reports on electrical resistance measurements were reported by Ryschkewitsch in 1923 (25). Measurements related to the temperature-dependence of the electrical resistance parallel to the layers were made by D. E. Roberts in 1913 and W. Meissner, H. Franz and H. Westerhof in 1932. In these reports the resistivity was found in the range of 4–8 × 10^–5 ohm cm. Later, G. E. Washburn, N. Ganguli and K. S. Krishnan reported that the resistivity perpendicular to the basal plane was 10²–10³ times greater than that of the parallel planes (26). Simultaneous theoretical studies to explore the electronic structure of graphite were also carried out. The first known energy band structure of graphite crystal was studied as early as 1935 by Hund and Mrowka (27, 28). However, the first known theoretical calculations with an attempt to relate them to properties such as electrical conductivity of graphite were made by Wallace and Coulson in 1947 (29, 30). Wallace employed ‘tight-binding’ approximation to develop the electronic energy bands and Brillouin zones of graphite (31).

In addition to the investigations on the structural, physical and electronic properties, the chemical properties of graphite such as the reaction of graphite with gases, particularly oxygen and carbon dioxide in presence of catalysts were also reported, but the use of graphite as catalyst support was seen much later (32). For example, Brownlie et al. used graphite as a catalyst support to investigate how properties of metals are influenced by its interaction with the surface (31). Under standard temperatures and pressures, graphite is the most stable allotrope of carbon but due to its low surface area it was used very little as catalyst support (33).

In recent years however, many studies on the use of expanded graphite as catalyst supports have been reported (33, 34). We note that a modified graphite having an enlarged interlayer lattice distance of 4.3 Å, and having a long-range-order layered structure of graphite is referred to as expanded graphite (35). For example, Chen et al. recently showed that platinum-cobalt nanoparticles supported on expanded graphite have higher electrocatalytic activity and anti-poisoning ability and long-term stability for methanol oxidation (34). The expanded graphite also finds its application in fuel cell electrode material due to its good electrical and thermal conductivity, chemical stability and excellent mechanical properties (36). Other interesting experimental studies have been also reported on the use of graphite nanoplatelets (37).

Recently, Novoselov et al. reported the isolation of graphene by employing the technique of micromechanical cleavage, which involves mechanical exfoliation of small mesas of highly oriented pyrolytic graphite (38, 39). In recent years, the use of graphene as catalyst support has been also a topic of extensive research especially because of its interesting properties, which include excellent electronic and thermal conductivity, mechanical and thermal stability and high surface area (as compared to graphite) (40, 41). However, in addition to the well-structured materials such as fullerene, carbon nanotubes, graphene and graphite there are many other forms of carbon material such as carbon black, graphite oxide, activated carbon and carbon fibres (CFs) in which graphene is the building block. These carbon materials have wide range of applications, for example carbon blacks find their use as catalyst supports in fuel cells, in car tyres and as fillers in polymers used in conductive packaging for electronic components. Similarly, graphene oxide (GO), activated carbon and CF also find their application in catalyst support (42–51). However, theoretical studies on structural and chemical properties of these amorphous carbons have not yet been done in greater detail. In the following sections a brief discussion on the structural and chemical properties of these amorphous graphene-related carbon materials is presented.

Since the early 1900s, several models of the internal structure of carbon blacks have been proposed and in this regard, the review paper by Donnet provides an excellent overview of some of these models (52). One of the first models of carbon blacks was proposed by Riley, which was based on the X-ray studies that basically corresponded to the average structure of particles reflecting the random distribution of crystallites and it represented the organised part of the carbon black particles (Figure 1(a)). The improved models were proposed by Bouland (Figure 1(b)) and Shultz (Figure 1(c)) in which the skin of the particle was believed to be more organised than its core. A more systematic model was proposed by Hekman (Figure 1(d)), which was based on the X-ray and electron microscopy. In this model the skin of the particle was also proposed to be better organised than its randomly arranged crystallites in the core. Later, on the basis of results obtained from lattice fringe imaging technique, a more detailed structure of carbon black was proposed by Heidenreich, Hess and Ban (53). In this model, (Figure 2) they also proposed that the ordering of the crystallite size and their ordering towards the centre of the particle are less than on the surface. According to all these models the commercial carbon blacks are made up of crystallites which are on average four graphite layer planes. The layers are roughly parallel and equidistant but do not exhibit the ABA stacking of hexagonal graphite. These graphite layer planes are referred to as BSUs. The distance between the layers in the BSUs is larger than 3.35 Å. The size of these BSUs in the plane can differ between a few Ångströms and 20–30 Å. If the samples become more amorphous, the number of stacked layers decreases until only one layer is left and simultaneously, the number of condensed rings in the plane also decreases (54). Recently Jäger et al. reported that the carbon black samples condensed in 10 mbar and 100 mbar argon gas atmosphere were similar to the Riley model as shown in Figure 1(a) (54). They further reported that the samples produced by them at 100 mbar argon were close to the paracrystalline model (see Figure 3) where the curved and strongly distributed graphene layers are stacked together to form a particle.

Fig. 1.

Fig. 2.

Fig. 3.

In addition to the studies on the structural features of carbon black since the 1950s it is known that the surface of these particles have functional groups, which include hydroxyl, carboxyl, aldehydes, ketones and quinones (52, 55, 56). Several researchers including Donnet et al., Garten et al., Studebaker et al., Puri et al. and Boehm et al. in the 1950s and 1960s independently arrived at this conclusion (52, 57–61). Since then various spectroscopic studies have also led to similar conclusions. For example, the recent Fourier transform infrared (FTIR) spectral studies by Juan et al. have shown that in the carbon black structures there are lots of aromatic C–C bonds and a large amount of oxygen containing organic functional groups (OFGs) (62). Other recent studies by Pantea et al. have shown that in addition to oxygen containing groups, sulfur containing groups may be also present (42, 63).

Preparation of graphite oxide was first reported in 1850, which was obtained by oxidising graphite with KClO₃/HNO₃ (64, 65). Since then several other preparation methods of graphite oxides have been established and some of the most widely known methods are by Brodie, Staudenmaier, Hummers and Offeman. All these methods use strong oxidising agents such as KClO₃, KMnO₄ with HNO₃ (66–68). While the preparation methods for graphite oxide are well established, its structural features are still not fully understood. In this regard, several models have been proposed, which include:

In another study, Lerf et al. employed ¹³C and ¹H nuclear magnetic resonance (NMR) techniques to propose a newer structural model of graphite oxide, which consists of two kinds of regions i.e., (a) aromatic regions with unoxidised benzene rings and (b) regions containing aliphatic six-membered rings (70). In this model, the aromatic entities, double bonds and epoxide groups give rise to a nearly flat carbon grid and only the carbons attached to –OH groups are in a slightly distorted tetrahedral configuration, resulting in some wrinkling of the layers. It was further reported that the functional groups lie above and below the carbon grid forming a layer of oxygen atoms of variable concentration. Like graphite itself, graphite oxide has a layered structure, but its carbon layers contain large quantities of functional groups, mainly –OH and –CHO (71). We note that modifications of graphite oxide with soft donor atoms of sulfur have been also reported by Talanov et al. (72). Further to this, recent studies on the graphite oxide structure, using various microscopy and spectroscopic techniques, proposed a new model of GO, which indicates the presence of –O–, –OH and –C=O groups on the surface. In these models, the two neighbouring –OH groups are located on opposite sites in order to reduce the electrostatic repulsions in between them and for the same reason the –OH groups are located away from each other. The epoxy and ketone groups are also away from the –OH groups (see Figure 4) (73). Additionally, the presence of five- and six-membered-ring lactols has also been proposed (74).

Fig. 4.

The surface structure of activated carbon is highly complex and depends on the raw material used and method adopted for their production. It is believed that activated carbon generally consists of graphite crystallites with highly disordered, irregular, rough and heterogeneous surfaces (75). There are many experimental studies which have reported the surface chemistry of activated carbon and methods to evaluate it (76, 77). The infrared study on the surface structure of activated carbons has indicated the presence of oxygen-containing OFGs such as C=O in lactones and carboxylic anhydrides, quinine and aceto-enol groups and C–O group in ethers, lactones, phenols and carboxylic anhydrides (78). The structure of activated carbons is still not fully understood.

CFs are generally in the range of 5–50 μm, which are commercially produced by converting a carbonaceous precursor into fibre. The polyacrylonitrile-based (PAN-based) and mesophase pitch-based CFs are two of the dominant CFs (79). Several studies have shown that the PAN-based CFs have extensively folded and interlinked turbostratic layers (i.e., the sheets of carbon are haphazardly folded and crumbled together) with interlayer spacing larger than those of graphite (79, 80). Mesophase pitch-based CFs on the other hand may have radial (graphene planes radiating out from the centre of the fibre), random (with random orientation of graphitic planes) or onion or quasi-onion like transverse microstructures (79, 81). The mesophase-pitch based CFs usually have a larger diameter (10–15 μm) as compared to the PAN-based CFs (5–7 μm) (80). In addition to understanding the detailed structural features of CF there have also been numerous experimental studies on the characterisation of the surficial reactive functional groups in CFs (82–84). In this regard, the studies by Szabó et al. showed that CFs consist of mainly –OH groups (~97%) and small amounts of –NH₂ (~2.7%) and –COOH (~0.3%) groups (84).

Finally, we briefly comment on the structural and chemical features of soot, as carbon black is sometimes confused with soot even though they are very different materials (85). While carbon blacks are made under closely controlled conditions, soot is produced as a byproduct of incomplete combustion of a hydrocarbon or pyrolysis (86, 87). The gas-phase soot contains polycyclic aromatic hydrocarbons (PAHs). It has been found that soot is made of spherical-like particles with diameters in the 20–50 nm range. These particles are composed of graphite-type microcrystallite layers which are concentrically arranged in an onion-like structure. The graphitic layers are found to contain surface OH and C=O groups (88).

In 1964, Pierre Hohenberg and Walter Kohn proposed and proved that for a nondegenerate ground state all the ground state electronic properties are determined by the ground state electron density (ρ₀), which could be mathematically represented by Equation (i) (89, 90):

(i)

Here, V_ext is the Coulomb potential energy between electrons and the nuclei, and Ψ₀ is a unique ground state wave function (91). The electron density is dependent only on three spatial variables (r) whereas the wave function depends on 3N variables for a system of N electrons.

Thus, the ground state energy can be written as a functional of the ground state electron density (Equation (ii)),

(ii)

Here, , and are the average kinetic energy, potential energy due to nucleus-electron and electron-electron interactions, respectively. The term in Equation (ii) can be further written as Equation (iii):

(iii)

where v (r_i ) is the nuclear-electron potential energy for the i^th electron.

Combining Equations (ii) and (iii) yields Equation (iv):

(iv)

The term in Equation (iv) is unknown.

In the Hohenberg-Kohn variational theorem, Hohenberg and Kohn further proved that the true ground state electron density minimises the energy functional E_v [ρ_tr], where ρ_tr is the trial density function that satisfies ∫ρ_tr (r) dr = N and ρ_tr ≥ 0 for all r and the inequality E₀ ≤ E_v [ρ_tr] holds where E_v is the energy functional in the Equation (iv). Hohenberg and Kohn proved their theorems only for nondegenerate ground states and later Levy proved the theorems for degenerate ground states (92).

While the Hohenberg-Kohn theorem tells us that if we know the ground-state electron density ρ(r), we can determine the ground state wave function, it however does not tell how to calculate E₀ from ρ₀. It also does not tell how to obtain ρ₀ without finding the wavefunction (89, 93). In 1965 Kohn and Sham revisited this problem, which is briefly outlined below.

Kohn and Sham considered a fictitious system of N non-interacting electrons, which experiences an external potential energy v_s(r_i) (index s is reserved here for the fictitious systems) (93). It was further assumed that the electron density of this reference system ρ_s(r) is equal to the exact ground state electron density ρ₀(r) of the system of N interacting electron system under consideration (molecule). The Hamiltonian of this system is given by Equation (v):

(v)

where, is the one-electron Kohn-Sham Hamiltonian for the fictitious system (Kohn-Sham Hamiltonian). Additionally, since this system consists of noninteracting electrons, the ground state wave function Ψ_s,0 is the anti-symmetrised product (using the Slater determinant) of all the lowest energy one-electron spin-orbitals, which are the product of one-electron spatial orbital and a one-electron spin function. Further to this, they defined the difference between the kinetic energy of the interacting N-electron system and that of the fictitious system as Equation (vi):

(vi)

where and are the average electronic kinetic energies of the interacting N electron system and the fictitious noninteracting N electron system respectively. Similarly, the quantum mechanical many-body electron-electron interaction energy minus the classical one can be represented as Equation (vii):

(vii)

Here the term ½ is added so that the repulsion energy between the charges is not counted twice.

For simplicity, we can omit subscript 0 from ρ and by rearranging and substituting Equation (vi) and (vii) in Equation (iv) we have Equation (viii):

(viii)

which can be also rewritten as Equation (ix):

(ix)

Where Equation (x),

(x)

which is referred to as the exchange-correlation (xc) energy functional (93) and it is responsible for all the many-body electron-electron interaction. The term E_xc can be also written as Equation (xi):

(xi)

where E_x and E_c are exchange and correlation energy respectively.

We note that even though the xc energy is a small part of the total energy of a typical system it plays the vital role in binding atoms together and therefore, Perdew coined it as “nature’s glue” (94). This quantum mechanical phenomenon arises as electrons move in such a way to avoid one another, which in turn, lowers the expectation value of the electron-electron Coulomb interaction. The exchange energy in the xc functional (Equation (xi)) is a consequence of the system obeying the Pauli principle and is free from the spurious self-interaction of an electron, when the exact Fock exchange is used. The correlation energy (E_c = E_xc − E_x) should account for the remaining effects of spatial and spin correlation in the many electron system.

Some of the popular exchange and correlation functionals are summarised in Table I, which is not a complete list of all the exchange and correlation functionals currently available.

The earlier theoretical studies on the adsorption of TM adatoms on carbon materials were mainly reported on a single layer of sp² hybridised carbon support with some reports on a few layers of graphene and these studies we‐re mainly focused on structure, bonding, magnetic properties of the adsorbed TMs on graphene and their migration to the high symmetry sites (117). Duffy et al. for example, used the linear combination of atomic orbitals approach as implemented in DMol package to study the effect of the surface on the 3d TM adatoms and dimers on a cluster model of graphite (118–121). In their calculation they used the local spin density approximation using the Vosko-Wilk-Nusair (VWN) exchange and correlation functional (95). They reported that the preferable adsorption sites of scandium, titanium, chromium and manganese are above the carbon atoms while iron, cobalt and nickel prefer over the rings. Due to the hybridisation between these metals with the π-orbitals of graphite there is a small electron transfer. It was also seen that the total magnetic moments are higher for scandium, titanium and vanadium by 1μ_B than the free atom and lower by 2μ_B for iron, cobalt and nickel, while chromium and manganese have reversed their free atom values. To understand this behaviour, they considered the local density of states of the iron/graphite (with C_6v symmetry: iron stable over the hole-position) and vanadium/graphite (with C_3v symmetry: vanadium stable over-atom position) systems. In the iron/graphite system, the molecular orbitals (MOs) mainly comprise 3d and 4s orbitals and a small admixture of 4p orbitals. These MOs are labelled by the C_6v group i.e., a₁, a₂, b₁, b₂, e₁ and e₂. However, only the a₁, e₁ and e₂ orbitals have metallic as well as carbon π-orbital components. A careful analysis of these orbitals showed that the adsorbed atom has the tendency to shift the 4s orbital energy up with respect to the 3d orbitals meaning an electron configuration of 3d ⁿ⁺²4s⁰ is preferred over 3d ⁿ4s² where 1 ≤ n ≤ 8 of the elements under consideration (note that chromium has an electron configuration of 3d ⁵4s¹ due to half-filled electron configuration being more stable). From such analysis they were able to explain the reduction of the atomic spin moment by 2μ_B in iron. Similar analyses were done on vanadium/graphite system, which showed that there is also a tendency for increased population of 3d related MOs but due to its enhanced hybridisation with the graphite surface this effect is reduced, which leads to an increase in the spin moment by 1μ_B.

In another study, Valencia et al. performed periodic spin polarised DFT calculations using a projector-augmented wave (PAW) scheme, and the PW91 exchange and correlation functional as implemented in the VASP code to explore the trends in adsorption of the 3d TM atoms on graphene (122). They found that scandium, titanium, iron and cobalt could remain isolated on the graphene surface but other metals in this series may diffuse with a possibility to form aggregates. They employed density of states and Bader charge analysis to explain the observed trends. Another interesting aspect of this study was the use of the organometallic M(η⁶-C₆H₆) MOs to bridge the language used to describe molecular states and that used to describe solid states. Figure 5 shows the qualitative interaction orbital diagram for the M(η⁶–C₆H₆) C_6v system using Hückel calculations by means of the Computer Aided Composition of Atomic Orbitals (CACAO) program (123–126). This simplified MO diagram shows six 2Π-orbitals of benzene and the valence orbitals of the 3d TM atom and as described above, these orbitals are labelled by the C_6v group. Of the six 2Π-orbitals, three with a₁ and e₁ symmetries are occupied, which interacts with the 4s and 3d orbitals of similar symmetries giving three bonding orbitals mainly localised on the benzene participating in the M–C bonds directing downward. The three antibonding orbitals, which remain unoccupied are mainly localised on the metal 4s and 3d orbitals and directed upward. The metallic group orbitals are shown in the centre of the interaction diagram and they are formed of 1e₂ (from mixing of 3d _{x² − y²} and 3d_xy orbitals, with antibonding π^*_CC orbitals of benzene participating in metal-carbon δ bonding), 2a₁ (from mixing of 4s, 3d_z², 4p_z orbitals and the in-phase π_CC MO, which leads to a nonbonding σ MO), 2e₁ (formed by mixing of 3d_yz, 4p_x and 4p_y orbitals interacting with the bonding π_CC ) and 3a₁ (derived mainly from the metal 4s orbital). They employed this qualitative M(η⁶-C₆H₆) MO description along with the calculated partial density of states for the periodic M@graphene to show that the energy levels follow a global order i.e., 1e₂ ~ 2a₁ < 2e₁ < 3a₁ and they used this scheme to clarify the observed trends on chemisorption of 3d TM with an exception of half-filled chromium and manganese and fully filled copper 3d atoms, which physisorb on the graphitic surface. Further to this, they used qualitative MO diagrams and charge transfer considerations to explain the trends observed in computed magnetism. They also concluded that scandium, titanium, iron and cobalt are stable as isolated atoms on the graphitic surface and other metals diffuse to eventually form aggregates.

In another interesting study Nakada et al. reported the adsorption and migration energies of adatom with atomic numbers 1–83 on graphene at the local density approximation (LDA) level of theory using the PAW method as implemented in VASP (127). In this study they reported that the TM elements mainly adsorb on the hollow sites. They noted that the adsorption types can be of two types i.e., with fixed adsorption and those with no fixed sites. For the systems with no fixed adsorption sites they estimated the migration energy barrier as 0.5 eV where the minimum limit of the migration energy was obtained by taking the difference between the adsorption energies for each site. They concluded that for metal-graphene junctions in addition to the adsorption energies the migration energy plays an important role. Recent studies by Manadé et al. also reported a systematic study on the adsorption of 3d, 4d and 5d TM adsorbed on graphene. In this study they also concluded that TMs prefer the hollow sites when chemisorbed. However, if they are physisorbed, for example in the case of d⁵ and d¹⁰ configuration, then they prefer the bridge or the top sites. Their study showed that inclusion of dispersive forces simply increases the adsorption energies of the TM by ~0.35 eV. They also reported the electron transfer behaviour of the TMs and found that it decays along the series because of the increase in electronegativity (128).

Fig. 5.

In addition to the adsorption of TMs on a single layer graphene, studies on the adsorption of metal adatoms have been reported on bilayer and trilayer graphene (129–132). For example, Hardcastle et al. employed plane-wave DFT to investigate the adsorption properties of gold, chromium and aluminium atoms on the armchair and zigzag sites of monolayer graphene and on the adsorption sites of single, bi- and tri-layer graphene. They investigated the adatom mobility on the pristine substrates (132, 133). They used the CASTEP code with Perdew-Burke-Ernzerhoff (PBE) exchange and correlation functional in conjunction with Tkatchenko-Scheffler Van der Waals correction scheme (134). In this study they concluded that the graphene sublayers make a significant contribution to the total binding energy of the adatom and the adatoms are extremely mobile on graphene at room temperature. In recent years, the trends of theoretical studies on the interaction of TMs on graphene have moved towards exploring the interaction of more realistic metal nanoparticles on graphene (135, 136). For example, Engel et al. recently systematically studied the electronic properties of gold clusters on graphene using DFT-D+U (135). Further to this, the interactions of molecules for example, NO₂, H₂S and glucose on TM atoms and clusters supported on pristine graphene have been also reported (137–139).

Graphene can display single or multiple vacancies but it is worth mentioning that the filling of such a vacancy may occur due to the reservoir of loose carbon atoms readily available near these vacancies (140, 141). The single vacancy (SV) refers to a missing carbon atom from the lattice and the double vacancy can be created by combining two SVs or by removing two neighbouring carbon atoms. Double vacancies may result in different types of configuration i.e., two pentagons and one octagon (5-8-5), three pentagons and three heptagons (555-777) and four pentagons, one hexagon and four heptagons (555-6-7777) (140). Therefore, in addition to the interaction of TMs on pristine graphene, studies on graphene with single and double vacancies have been also reported, which is not only interesting from the catalysis point of view but also from the context of spintronics and Kondo physics (142). Krasheninnikov et al. for example, presented a DFT study on the structure, bonding and magnetic properties of the first TM series, platinum and gold on the single and double vacancies of single graphene sheets (142). They found that on the SVs, all the metal atoms form covalent bonds with the under-coordinated carbon atoms and as the atomic radii of TMs are larger than carbon, the metal atom bulges out of the graphene plane. The TM-carbon bond lengths were seen to decrease from scandium to iron due to the decrease of the TM radii and then it increases as bonding becomes weaker. The calculated binding energies of the TMs on the SVs were typically about –7 eV with an exception of fully filled d-shell elements such as copper and zinc. They further noted that SV complexes are magnetic for vanadium, chromium, manganese, cobalt and copper and nonmagnetic for iron and nickel. On the other hand, for the TMs on double vacancies, all the TMs from vanadium to cobalt are magnetic, which is related to the fact that on double vacancies due to the larger ‘hole’ the interactions of the TMs are weak and therefore, the complexes are in higher spin states.

In another study Krasheninnikov et al., using strain fields, further demonstrated that metal atoms have high affinity to the non-perfect and strained regions of the graphene with defects (143). In this work they visualised the strain fields as the difference between the bond lengths between the graphene sheet with and without defects, which suggested that all vacancies result in strain fields in the vicinity of the defects in the range of 2–3 nm or even more. In 2013, Robertson et al. used aberration corrected transmission electron microscopy (ACTEM), high‐resolution transmission electron microscopy (HR‐TEM), electron microscopy and DFT to show that single iron atoms can be trapped on the single and double vacancies of graphene, forming covalent bonds and causing significant displacement, up to 0.5 Å, of the surrounding carbon atoms. They further reported that iron on such vacancies are more stable than when they are incorporated into the graphene edge.

In recent years, theoretical studies have moved towards understanding the interaction of small TM clusters on the single and double vacancies of graphene. In this regard, Sen et al. investigated the stability and electronic properties of Pd_n (n = 1–5) clusters on different types of double vacancies mentioned above using DFT and molecular dynamic simulations (144). Their study revealed that the adsorption of Pd₄ clusters on the defect bridge site of double vacancy of (555-777) is favourable, which they concluded is because of the hybridisation between carbon 2p and palladium 4d and 5s orbitals and higher charge transfer to the graphene sheet. Further to this, the latest studies in this area are focused on understanding the interaction of molecules on TM clusters adsorbed onto the graphene lattice vacancies. For example, Hamamoto et al. used DFT to demonstrate that the local electronic properties of CO adsorbed on Pt₄ clusters supported on graphene with one to four vacancies are remarkably altered due to its interaction with the dangling bonds of carbon atoms in the defect sites (145).

The introduction of heteroatoms such as nitrogen and boron into the carbon framework of graphene can be used to tailor the electronic and local geometrical properties of graphene for applications in catalysis (146–152). The substitution of nitrogen (n-type doping) in the graphene lattice usually leads to four common bonding configurations, which include: (a) quaternary nitrogen (obtained by substituting the carbon atoms of the hexagonal ring by the nitrogen atoms); (b) pyrrolic nitrogen (obtained by nitrogen atoms due to the bonding on the five-membered rings as in pyrrole); (c) pyrazolic nitrogen (obtained by the insertion of an aromatic N₂ moiety in a five-membered ring as in pyrazole); and (d) pyridinic nitrogen (obtained by substitution of the nitrogen atoms at the defect sites or on the edge of graphene as in pyridine) (146, 153, 154). Similarly, it has been shown that doping with boron (p-type doping) or more complex functional groups such as boronic esters and boronic acids can be also done on graphene (155). Such incorporation of dopant atoms leads to significant alteration of the electronic properties of graphene, for example Zhang et al. theoretically showed that substitution of nitrogen on the graphene framework introduces asymmetric spin density and atomic charge density making it favourable to oxygen reduction reactions (156). Recently, reports on doping of graphene with beryllium, sulfur, silicon and phosphorus have been also made (157, 158). It is worth mentioning that doping on graphene can be also achieved by electron exchange between adsorbed species on graphene surfaces (148).

It has been now shown that the durability and activity of TM catalysts such as platinum increases significantly on nitrogen-doped carbon supports (159). In this regard, numerous theoretical studies have also been performed. For example, Groves et al. used DFT as implemented in the Gaussian 03 package on 42 carbon atom cluster models to calculate the binding energy between nitrogen-doped graphene structure and platinum adatoms in order to explore the MO and natural bonding orbital data (160). They showed that nitrogen-doping increases the binding energy of platinum to the substrate and the binding energy increases if the nitrogen-atoms are closer to the carbon atom bonded directly to the platinum adatom. In another study Zhou et al. employed the DMol³ code to study the effect of nitrogen and boron doping on the stability of defective graphene supported Ni_n (n = 1–6) clusters (2). They found that the binding energies of these clusters on doped defective graphene were higher than on pristine graphene meaning the stability of nickel clusters was further improved. It was also seen that Ni_n clusters were more stable on the boron-doped defective graphene than the nitrogen-doped ones. The probable reason for this was attributed to the fact that boron heteroatoms promote stronger hybridisation between carbon atoms and nickel atoms. Another interesting study by Kropp et al. also found that nitrogen doping resulted in increased binding energies of TMs (161). These studies clearly suggested that nitrogen and boron-doped graphene can be good catalyst support. We note that in recent years there have also been some studies in which the adsorption of small gas molecules on TM doped graphene have been reported (162).

Theoretical studies on the geometrical and electronic properties of graphene (mainly graphene nanoflake (GNF) cluster models) with OFGs to model the BSUs of carbon black, GO, activated carbon and soot have been reported (7, 163–167). For example, Efremenko et al. reported the adsorption properties of phenol on activated carbon, Hamad et al. reported the adsorption of water on soot particles and we have also investigated the detailed electronic properties of GNFs with OFGs to mimic the BSUs of amorphous carbon materials discussed earlier in Section 1 with various shapes and sizes (75, 88, 163–165). In this regard, we investigated the influence of orientation and stacking patterns of the model BSUs on the electronic properties. In these models, the oxygen containing OFGs such as –OH, –CHO and –COOH were decorated on the periphery of the GNF models. In one of these studies we used –OH and –SH groups to understand the effect of chemical constitution of these materials (Figure 6). The models constituted of pristine GNF models (referred to as G in Figure 6(a)), with only –OH groups (i.e., G(OH)), with only –SH groups (i.e., G(SH)) and models with mixed –OH and –SH groups (i.e., G(OS-C_3h) and G(OS-C_s); C_3h and C_s refers to the symmetry of the models). Additional studies on the effect of different stacking conformations with two, three and four layers of G(OH) and G(SH) models on the electronic properties were undertaken (see Figure 7). From these studies it was seen that in the fully relaxed geometries, the CSH angles (due to the –S…HS– bonds) were closer to 90° when compared with the COH angles (due to the –O…HO–bonds), which was related to different hybridisation abilities of the oxygen and sulfur atoms and due to the charge-charge interaction in O…H and charge-quadrupole interaction in S…H. Further to this, it was concluded that by substitution of different OFGs and by using different stacking patterns (such as AAA or ABA) the electronic properties of these systems could be significantly altered. Based on these studies, we also investigated the adsorption properties of platinum-adatom and Pt₄ cluster on GNFs with varying concentration of hydroxyl groups (f-GNFs) (7). In this study, based on the DFT results it was proposed that the loading of platinum catalyst on carbon supports could be engineered by controlling the concentration of oxygen containing OFGs.

Fig. 6.

Fig. 7.

In addition to the studies on decorating graphene with oxygen containing OFGs there are several studies in which the oxygen-doped graphene have been used to investigate the adsorption properties of TM clusters, which include Pt₄, and Pt₃M (M = scandium, titanium, vanadium, chromium, manganese, iron, cobalt and nickel) (168, 169). In one of these studies, Cui et al. reported that Pt₃Ni supported on the oxygen-doped graphene exhibit enhanced catalytic performance (169). In some recent work, more complex studies of TM adsorption on the OH decorated nitrogen-doped graphene have been reported. For example, Sahoo et al. used transmission electron microscope (TEM), HR-TEM, hard X-ray photoelectron spectroscopy (HAXPES), combined with DFT calculations to report the charge transfer of palladium catalyst to pyrrolic nitrogen-doped reduced GO (NRGO) (170). In this study, they showed that strong bond between palladium and pyrrolic-fraction of the nitrogen-moieties can be formed, which can be further strengthened by the presence of oxygen containing functional groups. This study further reported that the charge-transfer from the palladium 3d-orbitals to the 2p-orbitals of carbon, nitrogen and oxygen of NRGO results in high oxygen reduction reaction. In another recent study Rout et al. synthesised highly active and well-defined gold-copper nanoparticles supported on reduced GO (171). These nanoparticles were then characterised using various techniques, which include ultraviolet-visible (UV-vis), XRD, HR-TEM, scanning TEM (STEM) Raman and X-ray photoelectron spectroscopy (XPS). They found that Au₃Cu₁ bimetallic clusters exhibited superior catalytic activity. In this study, they used DFT on cluster models of reduced GO to clarify the reasons for the high catalytic activity of these nanoparticles. They concluded that efficient adsorption of the highly dispersed Au₃Cu₁ on reduced GO led to the enhanced catalytic activity.

As we can see from the above studies, the models on the amorphous carbons such as carbon black, GO, activated carbon and soot are based on a single layer of graphene. This is especially because of the complexity in modelling these systems. However, there are some interesting studies made towards modelling of complex amorphous carbons. For example, recently Caro et al. reported a systematic study in which they combined machine learning and DFT to theoretically understand the surface chemistry of amorphous carbons (172). In another study Ranganathan et al. employed molecular dynamics simulation to produce models of amorphous carbons (173). Clearly, more research in this area is required to reproduce as realistic models of amorphous carbons as those of the experiments and to understand the interaction between these catalyst supports and catalysts to design novel and efficient catalysts.