A Review of Alkali Tungsten Bronze Nanoparticles for Applications in Plasmonics

The optimal material for plasmonic applications is an electrical conductor with low damping losses, high chemical and thermal stability, simple low-cost synthetic methods, and a resonance frequency that can be tuned to suit a desired application. To date, plasmonic applications have predominantly relied on Au or Ag, but these materials are limited respectively by high damping losses and rapid corrosion. In the search for alternative plasmonic materials, the alkali tungsten bronzes have been identified as possible candidates, as they display many of the features of the optimal plasmonic material. In this review, we first describe the crystallography, electronic structure, synthesis methods and plasmonic behaviour of the tungsten bronzes. A range of plasmonic applications for tungsten bronze nanoparticles, including solar-control filtering, plasmonic photocatalysis and plasmonic photothermal therapy, are then discussed.


Introduction to Plasmonic Materials
In the presence of an external time-varying electric field, the conduction electrons (or "plasma") of an electrical conductor experience a polarisation and oscillate in response. Resonant oscillations can be established under certain circumstances, which depend on the geometry of the material, the nature of the external field and the dielectric response of the material. These resonant oscillations are referred to as plasmon resonances or plasmons [1].
A critical determinant in the nature of the plasmon responses of a material and its suitability for plasmonic applications is the relative permittivity or dielectric function, ε. The dielectric function describes the response of a material to an external electric field. It varies with frequency, ω, and is complex valued [2]: The zeroes of ε(ω) identify the frequencies at which a longitudinal collective excitation, the volume or bulk plasmon (BP), may occur. This is designated the bulk plasma frequency, ω BP . The longitudinal nature of this excitation means that the bulk plasmon cannot be excited by the field associated with transverse electromagnetic waves, but they are observed under excitation by fast electrons. Nevertheless, a purely free-electron conductor will be reflective for ω < ω BP and transmitting for ω > ω BP. The value of ω BP can be derived from the Drude model as: where N is the number density of electrons, m* is their effective mass, e is the fundamental electronic charge and ε 0 is the vacuum permittivity [1]. The introduction of damping into the free electron response due to collisions or interband transitions appears as finite values of the imaginary part of the dielectric function, ε 2 , and is thus connected to absorption in the material [2,3]. The ω BP will then also be modified from the free electron value of Eq. 2 [3].
In non-infinite systems, such as interfaces or nanoparticles, the introduction of boundary conditions generates different Ne 2 0 m * solutions of Maxwell's equations. For metal-dielectric interfaces (with dielectric functions ε and ε d , respectively), propagating transverse waves called surface plasmon polaritons (SPPs) can be excited. For a free electron metal and at large wavevectors, the following condition for the surface plasmon frequency, ω SPP , is satisfied.
In metallic nanoparticle systems, confinement of the free electrons gives rise to non-propagating solutions called localised surface plasmon resonances (LSPRs) where the resonance frequencies, ω LSPR , depend on the size and shape of the nanostructure and the dielectric functions for both the particle and the surrounding medium. Analytical solutions to Maxwell's equations are available for some simple geometries such as spheres and spheroids [4], but numerical calculations are usually required for complex geometries [5]. For the simplest case of a sphere that is sufficiently small such that the electrostatic approximation applies, a dipole LSPR will be observed at the frequency, ω LSPR , where: is satisfied. In typical metals, ∂ε 1 /∂ω > 0 below ω BP , meaning ω LSPR < ω SPP < ω BP and therefore nanostructures with higher aspect ratios are expected to have lower ω LSPR . For plasmonic applications, the ideal material will also have ε 2 ≈ 0 at ω SPP or ω LSPR to minimise losses.
Most applications exploit either the enormous magnitude of the optical extinction (i.e. strong scattering and/or strong absorption); the tunability of the optical extinction; or the local electric field enhancement [1,9]. The ideal plasmonic material should be a good electrical conductor with low damping losses at the wavelengths where plasmons are observed, it should be corrosion resistant and readily and inexpensively fabricated into nanoparticle form. Depending on the application, the desired plasmonic resonance may be in the ultraviolet (UV), visible or infrared (IR) spectral regimes and so appropriate choices regarding materials and the geometry of the system must be made.
The most widely used plasmonic materials today are the noble metals Au and Ag, and the techniques for control over particle size and shape are well-established for these materials [10,11]. Figure 1 shows the experimental dielectric functions for Au and Ag [12,13]. The high carrier concentrations of the noble metals (N ≈ 10 23 cm −3 [14]) leads to a free-electron response at low frequencies and a ω BP in the visible or ultraviolet. Consequently, SPPs and LSPRs in the noble metals are typically at visible frequencies or lower [15]. Figure 1 also shows that Au has much larger values of ε 2 than Ag for most visible and UV frequencies which is due to the onset of interband transitions at around 2.45 eV [13,14]. Despite the relatively high damping losses at LSPR frequencies, Au is widely used in applications because of its high chemical stability and ready functionalisation [16]. In contrast, the low values of ε 2 for Ag mean that it is used in applications requiring very strong plasmon resonances [15], but corrosion in air limits device lifetimes [10,14,17]. Combined with their relatively high cost, the shortcomings of the noble metals have fuelled the search for alternative plasmonic materials [14,15].
Many other metals have been considered for plasmonic applications. The dielectric function of Cu is shown in Fig. 1. Like Au, it also displays high values for ε 2 indicating plasmon damping will be significant [18]. This is consistent across the transition metals because the shallow d-band gives rise to significant optical absorption from interband transitions. Al more closely resembles a free electron metal but the plasmon resonances are usually in the UV, which is unsuitable for many applications [15]. Al and Cu are also affected by corrosion in ambient conditions. Alkali metals such as Na and K exhibit optical properties superior to those  [12] Au [13], Cu [18], Na [25], TiN [26] and Al-doped ZnO (AZO) [27]. Lower horizontal axis ticks show energy in eV; upper horizontal axis ticks show wavelength in nm of Ag [19], but are unusable for most applications due to their high chemical reactivity [15]. Several intermetallic and alloy systems have been considered for plasmonics with the goal of combining the most desirable properties of multiple metals [20][21][22]. Of the systems explored, the most promising candidates are intermetallics in the noble:noble system (e.g. AgAu) [22], and intermetallics in the akali:noble system (e.g. NaAu, KAu and KAg) [23,24]. Overall, most alloys or intermetallics do not offer better performance than their constituent metals, often due to the introduction of new interband transitions near plasmon frequencies [20].
The most widely discussed alternative materials are the transition metal nitrides, particularly TiN and ZrN [27,28]. Comprehensive reviews of these systems can be found in [27][28][29]36]. The carrier concentration of these materials can be as high as N ≈ 10 22 cm −3 due to N vacancies (i.e. TiN 1-δ ) [28], resulting in plasmon frequencies similar to those of Au [36][37][38]. The dielectric function of TiN [39] is shown in Fig. 1. A significant advantage of TiN in deviceon-chip plasmonic systems is its compatibility with CMOS processes [40,41], in addition to standard nanoparticle [39] or thin-film [28,42] fabrication techniques. However, optical losses are greater than those of Au for all nitrides. Regardless, the high-temperature stability and simple nanofabrication options for metal nitrides have led to applications in nanophotonics [41], plasmonic photovoltaics [43,44] and solar control filtering [45].
A wide variety of semiconducting metal oxides and chalcogenides have been investigated for plasmonics. These include highly doped metal oxides such as Al-doped ZnO (AZO) or Sn-doped In 2 O 3 (ITO) [27], as well as substoichiometric oxides or chalcogenides such as WO 3−δ [46] or Cu 2−δ S [47]. Several reviews of plasmonic metal chalcogenide systems are available [27,32,33,38,48,49]. Many metal chalcogenide systems offer wide tunability in carrier concentration (across N ≈ 10 19 -10 23 cm −3 ) and optical properties, meaning plasmon resonance frequencies can be varied by composition and by changing nanoparticle geometry [38]. Although materials such as AZO have significantly lower optical losses than even Ag (see Fig. 1 for ε of thin-film AZO [27]), their plasmon frequencies are at near or mid-infrared frequencies [27], restricting the range of applications to which they can be applied. The alkali tungsten bronzes are another metal oxide system that also offer tuneable resonance frequencies and low losses when compared to Au, but have plasmon frequencies at nearinfrared and visible light frequencies.

Alkali Tungsten Bronzes
The tungsten bronzes are sub-stoichiometric compounds described by M x WO 3 , where M is an intercalated species (usually a metal) and 0 ≤ x ≤ 1 [50][51][52]. They were first reported by Friedrich Wöhler in 1825 [53] and they have been the subject of extensive study in the subsequent decades due to their noteworthy crystallographic, electronic and optical properties [50][51][52]54]. The tungsten bronzes have been investigated for use as superconductors [55][56][57], in electrochromic displays [58], as electrodes for pH measurements [59], and as catalysts in a variety of applications [60][61][62]. In 2007, Adachi et al. proposed that M x WO 3 nanoparticles could be used in plasmonic solar-control filters [63]. Since then, there has been a growing body of research demonstrating that they form a promising class of new plasmonic materials [64,65]. Although M x WO 3 can be prepared with a large number of metals and species in the M site [50], tungsten bronzes where M is an alkali metal or NH 4 have been the most thoroughly investigated for plasmonics and are the focus of this review.

Crystal Structure
The tungsten bronzes are a perovskite-like system and, as such, display a range of crystal structure variants based on a network of corner-sharing WO 6 octahedra, with the M atoms occupying the interstices in the network. The crystal structures depend on M, x and the synthesis route and conditions, with the cubic (c-M x WO 3 ), hexagonal (h-) and two tetragonal (TI-and TII-) structures as the most relevant for plasmonics. These structures are shown in Fig. 2. Although orthorhombic and monoclinic structures have also been observed for x ≤ 0.1 [66], plasmonic properties have not been reported at these compositions. Figure 3 indicates the typical composition ranges for which the different crystal structures have been observed in some important tungsten bronzes.
In the cubic tungsten bronzes, the WO 6 octahedra are arranged to form 4-sided tunnels, into which the M species intercalates, forming a solid-solution of concentration M:W = x [79]. They have been observed across a continuous composition range for Li x WO 3 (0.12 ≤ x ≤ 0.5) and Na x WO 3 (0.3 ≤ x ≤ 1.0) [50], and as a line phase at x = 1 for K x WO 3 [74]. The representative structures for c-Na x WO 3 are shown in Fig. 2a and b. At room temperature, the WO 6 octahedra are rotated and distorted from the ideal arrangement of Fig. 2a. This unit cell can be described as a 2 × 2 × 2 supercell of WO 6 octahedra in space group Im3 [79,80] and is shown in Fig. 2b. As the supercell reflections are difficult to detect using X-ray diffraction [81], the smaller 1 × 1 × 1 unit cell in the Pm3m aristotype ( Fig. 2a) is sometimes sufficient for Rietveld analysis of powder patterns [82]. For Na x WO 3 , the lattice parameter of the Pm3m structure increases linearly with Na content, which permits the determination of the Na content of an unknown c-Na x WO 3 sample using diffraction techniques. In contrast, the lattice parameter of c-Li x WO 3 decreases with x between 0 ≤ x ≤ 0.5 [83]. Some cubic β-pyrochlore tungsten bronzes have also been synthesised [77,84,85], but reports of their optical properties are limited.
Two tetragonal structures have been observed in the tungsten bronzes. The tetragonal-I structure (TI-) is a tetragonal elongation of the perovskite-like structure observed in the cubic bronzes with alternating W sites displaced along the c-axis [86]. In the tetragonal-II structure (TII-), shown in Fig. 2c, the WO 6 octahedra are arranged to form 3, 4 and 5-sided tunnels. Due to its prevalence in other oxide and fluoride ceramics, the TII-structure shown here is also more broadly known as the tetragonal tungsten bronze (TTB) structure. In TII-M x WO 3 , the M ions preferentially occupy the 5-sided tunnels below x < 0.4, then begin to fill the 4-sided tunnels for 0.4 ≤ x ≤ 0.6 [72]. As with c-Na x WO 3 , the lattice parameters of TII-M x WO 3 increase with x for Na and K [74,87].
Hexagonal tungsten bronze structures are typically found when the ionic radius of M is large [50], such as in RbxWO3, CsxWO3 [70] and (NH4)xWO3 [78]. The structure of h-RbxWO3 is shown in Fig. 2d. In these structures, the WO6 octahedra are arranged to form large 6 and small 3-sided tunnels. M intercalates into the 6-sided tunnels only, leading to an upper composition limit of x ≤ 1/3 [50]. Crystal detects and vacancies appear to be more significant for the hexagonal tungsten bronzes [88][89][90] than in the other structures [91], though comparative studies are limited. LixWO3 can also be prepared with R3c (LiNbO3-type) structure across 0.5 ≤ x ≤ 1.0, but reports of material properties are limited [92].

Synthesis Techniques and Nanoparticle Morphology
A variety of different nanoparticle shapes and sizes can be prepared, depending on the synthesis method and composition, as illustrated in Fig. 4. One widely used technique  [71], Na x WO 3 [50,67,72], K x WO 3 [50,73,74], Rb x WO 3 [75], Cs x WO 3 [76,77] and (NH 4 ) x WO 3 [78]. The perovskite-like c-K 1.0 WO 3 [74] and the β-pyrochlore c-Cs 2 W 2 O 6 [77] are observed as line phases for M x WO 3 synthesis is the high-temperature method first reported by Straumanis [93] and further elucidated by Brown and Banks [67]. For M x WO 3 with M = Li, Na or K: This technique was first reported for Na x WO 3 , but has also been demonstrated for other alkali metals [83,94]. WO 2 [95,96] or flowing H 2 [63,97] can also be used as a reducing agent in place of W. Dry powdered reagents are ground and mixed, then sintered at T ≈ 850 °C under inert atmosphere [50,82]. This technique is typically carried out in a furnace, although synthesis in a microwave oven has been demonstrated [98]. At lower temperatures (T ≈ 600 °C), synthesis takes hours to weeks [67,99] resulting in large particle sizes. At higher temperatures (T ≈ 900 °C), Na x WO 3 nanoparticles can be synthesised with furnace hold times as short as t = 90 s [100]. When Na x WO 3 is prepared by the Straumanis method, the reaction completes in two stages: the reagents first combine to produce a low-x TI-Na x WO 3 , which is then sodiated by the Na-enriched mixture [99]. For Na x WO 3 , Dickens and Neild calculated the standard enthalpy of reaction of Eq. 5 to be ΔH = − 17.9 kJ⋅mol −1 for x = 0.53 and ΔH = − 23.3 kJ⋅mol −1 for x = 0.77 [95]. The upper limits on M concentration for this technique are approximately x ≤ 0.5 for perovskite-like c-Li x WO 3 , x ≤ 0.85 for c-Na x WO 3 and x ≤ 0.6 for TII-M x WO 3 [50,67,72,92]. The reason for these upper limits on x is not yet known for the cubic Li and Na structures. For the TII-structures, x = 0.6 corresponds to the complete filling of all the 4 and 5 sided interstices.
Nanoparticle morphology varies with Na content in the Na x WO 3 system. Na-rich (x ≥ 0.7) samples prepared via the Straumanis method consist of cube shaped nanoparticles (as shown in Fig. 4b) with faces parallel to {100} plane family [101]. Na-poor (x ≈ 0.4) samples show pseudospheres and rods [98,100] which correspond to the c-and TII-phases respectively. The long-axis of the rods is parallel to the < 001 > direction [87], but the relationship between crystal orientation and particle morphology of the pseudospheres has not been clarified. Na x WO 3 nanocubes have an approximately 5 nm Na-depleted layer around the edges of the particle [101], but whether this is a result of the synthesis technique or due to atmospheric degradation has not been determined.
The Straumanis method is not the only high-temperature M x WO 3 synthesis technique. Alkali M x WO 3 have been prepared by reducing WO 3 with the appropriate alkali azide (MN 3 ) under ≈ 6 MPa at T ≈ 800 °C [110]. Fully occupied Li 1.0 WO 3 , Na 1.0 WO 3 and K 1.0 WO 3 were prepared by reacting M 2 WO 4 , WO 3 and WO 2 at high temperatures and pressures [74,80,92]. W can be used to reduce metal halides and WO 3 into the corresponding tungsten oxyhalide and M x WO 3 for a wide variety of M [111,112]. Despite being a reduced metal oxide, Na x WO 3 can be synthesised in air at T = 600 °C by reduction of Na 2 WO 4 by B 4 C [113]. Several groups have synthesised tungsten bronzes from the reduction of ternary alkali-tungsten (VI) oxides by flowing H 2 at elevated temperatures [63,97,[114][115][116][117][118][119]. Post-synthetic ball-milling of Cs x WO 3 has been shown to deplete Cs from the edges of the particle [120], which affects the optical properties.  [107], g β-pyrochlore CsW 2 O 6 hollow tubes [108] and h h-Cs x WO 3 hexagonal prisms [109]. Reproduced from the references given above with permission from the Royal Society of Chemistry, the Institute of Physics, Elsevier, and the American Chemical Society
Nanoparticle morphology can be controlled by varying the reaction conditions. K x WO 3 grown on W foil in KOH forms nanorods when heated to 450 °C, and nanosheets (shown in Fig. 4d) when heated to 600 °C [104,124]. Ligand selection can also control particle morphology, as shown by Huang et al. with h-Na x Cs y WO 3 synthesis. SO 4 2− ligands promoted growth along the (002) plane to form nanorods, whereas Cl − ligands inhibited growth along the (111) plane to form truncated tetrahedra, as seen in Fig. 4a [102]. Mattox et al. synthesised Cs x WO 3 from WCl 4 and CsCl using oleic acid (OAc), with oleylamine (OAm) in toluene as a surfactant and capping ligand [109,130]. The CsCl content had little effect on the composition of the products, but varying the ratio of OAm to OAc changed the particle shape between hexagonal prisms (Fig. 4h), truncated cubes and pseudospheres. OAm was also used as a capping ligand and solvent in the synthesis of Na x Cs y WO 3 [85] and (NH 4 ) x WO 3 synthesis (Fig. 4f) [107], and OAc as a surfactant in the synthesis of Gd-doped Na x WO 3 nanorods [131].
In addition to the furnace assisted and wet chemical methods, a wide range of alternative approaches to synthesis have been demonstrated. Azimirad et al. fabricated Na x WO 3 nanowires by sputtering W films onto soda-lime glass then annealing, which drives Na 2 O to the surface to oxidise the W [132]. A similar method has been used to make Na x WO 3 thin films [133]. Cs x WO 3 thin films have been synthesised by the electron-beam evaporation of a Cs x WO 3 powdered target [134,135]. Cisternas Fig. 4g, by electrospinning a precursor solution onto Al foil with a 20 kV potential difference [108].
In summary, there is great diversity in the available synthesis routes for M x WO 3 . The wet chemical methods reported to-date appear to produce M x WO 3 with smaller particle sizes but with lower x than furnace-assisted techniques. This is particularly evident for the Na x WO 3 system where the absolute upper limit on x is the greatest. Although synthesis techniques for Cs x WO 3 nanoparticles are fairly mature, the level of control and homogeneity of particle size and shape that is achieved for Ag and Au nanoparticle synthesis has not yet been achieved for the tungsten bronzes. Development of robust nanostructuring techniques is perhaps the most important frontier to be explored before the potential of the alkali tungsten bronzes can be fully realised.

Electronic Structure and Optical Properties
Broadly speaking, the alkali tungsten bronzes exhibit metallic electronic properties in M-rich samples (x ≥ 0.25) and semiconducting properties in M-poor samples (x < 0.25) [137]. The electronic structure of Na x WO 3 is the most thoroughly investigated and is a model for other M x WO 3 . In metallic Na x WO 3 , the inserted Na is ionised and the 3 s electron is donated to the WO 3 conduction band. This means the free-electron density should be directly proportional to the Na content (N ∝ x) [138,139], although there have been suggestions that this is only true for x < 0.7 [140]. A similar dependence on x is observed in Li x WO 3 , suggesting the electrical properties of the alkali tungsten bronzes do not vary with the intercalated species M [137]. Electrical conductivity at room temperature increases with x, from ≈ 15,000 Ω −1 ⋅cm −1 at x ≈ 0.45 up to ≈ 65,000 Ω −1 ⋅cm −1 at x ≈ 0.90 [137,139]. As expected from a metal, resistivity (ρ) increases with temperature (T), although there is uncertainty on how the shape of ρ(T) varies across x [139,141,142].
The band structure of metallic Na x WO 3 around the Fermi energy, E F , has been investigated by several groups. The density of states (DOS) for c-Na 1.0 WO 3 , calculated using density functional theory (DFT) [143], is shown in Fig. 5a and is consistent with reports by other authors [144,145]. Akin to other perovskite-like materials, the filled valence band of WO 3 is primarily composed of O 2p orbitals and the empty conduction band primarily of W 5d orbitals [51,144,[146][147][148]. As the Na 3 s orbitals have an energy far higher than E F , Na is ionised when it is inserted and the 3 s electron instead enters the WO 3 conduction band. An increase in x raises E F further, as depicted in Fig. 5b [51,82,[144][145][146][147][148][149][150][151]. There has been debate [150][151][152] over whether metallic Na x WO 3 can be described by a rigid-band model. Some studies found the rigid-band model to be approximately correct [147,149], particularly in describing optical properties [82,144], whilst others have suggested that hybridisation between Na 3 s and O 2p states narrows the conduction band with increasing x [51,[150][151][152].
There are conflicting reports regarding the electrical properties of the semiconducting phases of M x WO 3 . In general, room temperature conductivity increases with x [137,153] but the relationship is not as well characterised as for the metallic M x WO 3 phases. Below room temperature, resistivity data on TII-K 0.56 WO 3 shows ρ(T) to decrease with T, typical of semiconductors [154], with similar results reported for TI-Li 0.097 WO 3 [137], TI-B 0.08 WO 3 [155], TIand c-Ca x WO 3 [153]. However, ρ(T) has been shown to increase with T for TII-K x WO 3 when x ≤ 0.45, a property typical of metals [56,74]. Anisotropic conductivity has been observed in TII-Na x WO 3 , with lower resistivity parallel to the unit cell c axis than in the a/b axes, and with metalliclike ρ(T) in both directions [139]. Similar anisotropy would be expected in other TII-and h-M x WO 3 . Although metallic Na x WO 3 is well characterised, more detailed and systematic studies of the electrical properties on the other alkali tungsten bronzes would help resolve some of the observed discrepancies.
The most obvious and striking property of the alkali tungsten bronzes are their vivid and lustrous colours, which vary with the M concentration, x [50]. Visible light microscope images of three Na x WO 3 samples are shown in Fig. 6. Na x WO 3 is dark blue at x ≈ 0.3, and as the Na increases, the colour transitions through purple, red, orange and finally yellow at x ≈ 1 [82]. As hexagonal tungsten bronzes are limited to x ≤ 0.33, these all dark blue in colour.
The optical properties of the alkali tungsten bronzes at visible wavelengths are dominated by the free-electron response arising from the ionised electrons of the intercalated atoms, particularly for M-rich samples. The dielectric functions for a selection of some Na x WO 3 and Cs x WO 3 are shown in Fig. 7. The zeroes of ε 1 occur at visible or near-infrared (NIR)  [82] frequencies and increase with x, although they are shifted from the values that would be predicted on the basis of the Drude model (Eq. 2) due to the effect of higher energy interband transitions on the dielectric functions. It follows that the bulk plasma frequency, ω BP , and the corresponding reflectivity minima also appear around the visible frequency range [156]. As the conduction band is filled with increasing x, the plasma edge rises: in Na x WO 3 , the bulk plasmon frequency increases from ω BP = 1.52 eV at x = 0.25 to ω BP = 2.38 eV at x = 0.82 [100,157,158]. Similar trends are observed in the reflectivity spectra of other alkali tungsten bronzes [159]. The shifting of this sharp reflectivity minimum through the visible frequency range with x is what leads to the vivid variation of colours of the alkali tungsten bronzes [82,160]. Qualitatively, there is little difference in the reflectivity between single crystals of Na x WO 3 , K x WO 3 , Rb x WO 3 and Cs x WO 3 [159,161], or between thin-films of H x WO 3 , Li x WO 3 and Na x WO 3 [162]. This is consistent with the finding that the electrical properties of the alkali tungsten bronzes are similar for different M [137], and suggests the same is true for the optical properties. Figure 7 shows that for Na x WO 3 and Cs x WO 3 , the zeroes of ε 1 occur at frequencies where ε 2 is small, suggesting that these materials should support high-quality plasmon resonances similar to those of Ag and Au. Although the nanoscale optical responses of many tungsten bronzes have been interpreted in terms of the free-electron responses [107,163,164], the direct identification of plasmon responses using electron energy-loss spectroscopy (EELS) has been limited to Na x WO 3 [157,158], K x WO 3 [87] and Cs x WO 3 [165]. Figure 8 shows bulk EEL spectra for c-Na x WO 3 where a strong BP around ≈ 2 eV, which blueshifts with increasing x, is observed. The strong peak at ≈ 7 eV arises from O 2p to W 5d interband transitions (IB) [158,166], and these also appear in WO 3 EEL spectra [166]. For Cs x WO 3 , the overall shapes of the spectra are similar to those for Na x WO 3 in Fig. 8 but some uncertainty still remains regarding the origin of a double peaked structure for the low-lying plasmon peak [130,165]. For all of the anisotropic structures, TI-, TII-and h-, different ω BP are expected along the a/b and c unit cell directions [87,167]. Absorptions from IR-energy polarons, arising from oxygen vacancies (analogous to WO 3-x ) [168,169] or M intercalation [170], have also been reported.

Plasmon Responses
In a scanning transmission electron microscope (STEM), spatially resolved EELS can be performed. Maps of the peaks corresponding to energy losses from the excitation of LSPRs can give insight into nature of these resonances, particularly when coupled with numerical simulation. Fig. 7 Dielectric functions for some c-Na x WO 3 [156] and h-Cs x WO 3 [76], both determined from spectroscopic ellipsometry. The dielectric functions of Ag [12] and Au [13] from Fig. 1 are shown for comparison Fig. 8 Bulk electron energy-loss spectra for Na x WO 3 [157]. Peaks from the bulk plasmon (BP) and valence-conduction interband transitions (IB) are labelled. The small peak at ≈ 0.8 eV in the Na x WO 3 spectra is the combination of a weak surface plasmon excitation, and the difficulty in subtracting the zero-loss peak from experimental data Figure 9 illustrates this process with an application to a single Na 0.82 WO 3 nanocube studied using spatially resolved EELS and numerical simulations with the boundary element method (BEM) method [171]. Distinct modes corresponding to "corner", "edge", "face" and bulk modes are observed [100], and these are consistent with those observed in Ag nanocubes [172]. The maps shown in Fig. 9h and m correspond to the bulk plasmon. In Na x WO 3 , LSPR frequencies will redshift with decreasing x, as would be expected give the dielectric functions in Fig. 7 [100]. Although detailed maps like those in Fig. 9 have not been reported for other M x WO 3 , Sato et al. observed surface plasmon excitations at the edges of irregularly shaped Cs 0.33 WO 3 nanoparticles using EELS [165].
More commonly, the plasmon responses of M x WO 3 nanoparticles are investigated using spectrophotometry. The nanoparticle extinction spectrum, σ, of M x WO 3 is characterised by strong absorption in the UV associated with interband transitions, and broad peaks in the visible and NIR variably associated with LSPRs or polarons, depending on the specific M x WO 3 in question [82,100,109,168]. Figure  10, reproduced from [130], shows (a) experimental and (c) simulated σ for h-Cs x WO 3 nanorods with varying aspect ratio, along with (c, inset) calculated charge distribution and (d) E-field maps at the LSPR wavelengths. The tunability in LSPR wavelength with nanoparticle aspect ratio is reminiscent of the responses of noble metal nanorods [130]. Experiment [109] and simulation [82] have shown that the peak in the NIR redshifts as the refractive index of the surrounding dielectric (n d = √ε d ) increases, a property of LSPRs and not polarons or defect states. However, it has been reported that tungsten or oxygen defects increase absorption in the NIR alongside LSPRs in Cs x WO 3 [88,168], similar to WO 3-δ [173]. It is likely that the absorbance spectrum of a M x WO 3 sample will depend on the preparation method, as this will determine the level of oxygen vacancies. to right) as two corner modes, edge, face and bulk modes. Although the particle is slightly rectangular, the LSPRs were not observed to split into transverse and longitudinal components in the experimental spectra, likely due to the energy resolution of the spectra (ΔE ≈ 0.5 eV). Similar results from a different nanocube were also presented in [100] There are similarities in the nanoparticle extinction spectra between M x WO 3 with similar x and different M [63,174], although different particle geometries will intrinsically support different LSPRs. The link between morphology and σ for h-Cs x WO 3 nanoparticles was explored in detail by the Milliron group, possible due to their mature techniques for control over nanoparticle geometry [109,130]. They observed that σ for nanorods was influenced not only by aspect ratio but also by the anisotropy in ε associated with the hexagonal crystal structure [130], as shown in Fig. 10. For samples with a wider distribution of particle sizes, the ensemble σ can be considered the sum of the optical responses from different sized particles [100,169] leading to a broadening of the LSPR feature in the optical response.
Overall, the nanoparticle optical responses of the tungsten bronzes are qualitatively similar to those of the noble metals, with plasmon resonances adding absorption peaks at frequencies below the onset of interband transitions. Whilst the ω LSPR of the noble metals are tunable only through particle geometry and environment, the tungsten bronzes have the advantage that plasmon frequencies can also be varied with composition. However, with a few notable exceptions [102,109], the synthesis techniques for M x WO 3 have not yet been developed to separately allow fine control over particle Fig. 10 LSPRs from h-Cs x WO 3 nanoparticles of varying aspect ratio, AR [130]. a Experimental optical absorption spectra of h-Cs x WO 3 nanoparticle samples of platelets (green), isoprisms (orange) and rods (blue), dispersed in tetrachloroethylene, showing the aspectratio dependence of LSPR wavelengths. b Plot of LSPR peak splitting, Δω lsp against particle aspect ratio, comparing experimental data (open circles) to theoretical modelling, assuming isotropic (blue stars) or anisotropic (pink stars) dielectric functions from [159]. c Simulated absorption spectra for (i) platelets, (ii) isoprisms and (iii) rod-shaped nanoparticles using the anisotropic dielectric function for h-Cs x WO 3 . Solid lines show the overall absorption profiles, summing one longitudinal mode (dashed lines) and two identical transverse modes (dotted lines). The insets show the simulated dipolar surface charge distributions at each peak frequency of the two modes. d Colour map of the simulated near-field enhancement for a rod excited at the (i) longitudinal and (ii) transverse LSPR wavelengths. Figure from reference [130], available from https:// pubs. acs. org/ doi/ 10. 1021/ acs. nanol ett. 6b013 90. Further permissions related to the material excerpted should be directed to the American Chemical Society morphology and material composition. As mentioned earlier, this is the main obstacle facing the broad adoption of M x WO 3 for plasmonics.

Applications
During the mid to late twentieth century, most of the research into the tungsten bronzes focused on understanding their unique structural and electronic properties. Now with advances in nanofabrication and the emergence of plasmonics, these materials have been approached from more application-oriented perspectives. Although the alkali tungsten bronzes have only recently been considered plasmonic materials [76,82,109], there has been significant progress in developing technologies and systems which use M x WO 3 .

Optical Enhancement
Compared to other plasmonic materials such as the transition metal nitrides, or even Au, the alkali tungsten bronzes (particularly Na x WO 3 ) typically have lower ε 2 at LSPR wavelengths. This suggests that existing plasmonic devices or systems could achieve better performance if M x WO 3 is used as the conducting material. Figure 11 shows some common figures-of-merit (FoMs) for characterising plasmonic materials. Shown are (a) the Q factor of Wang and Shen and (e) the α/V ratio of Arnold and Blaber [177]: In these equations, ℏ = h/2π is the reduced Planck constant and Re(ε d ) is the real part of the dielectric function of the insulating medium. These FoMs nominally quantify material performance for different nanoplasmonic systems. An evaluation on their usefulness is beyond the scope of this work; here, they will be used to broadly rank the conventional plasmonic materials Au [13] and Ag [12] against c-Na x WO 3 [82] and the popular alternative material, TiN [26]. Based on these FoMs, Ag outperforms all other materials under consideration, so in applications where cost and chemical stability are of no concern, Ag remains the material of choice. The Q, Q LSPR and α/V for Na x WO 3 are greater than those of Au at the energy of the dipole mode of a nanosphere, i.e. ℏω(ε = − 2) [1]. This suggests that nanospheres of Na x WO 3 should exhibit greater resonance qualities than those made from Au. However, at the energies which are practically Fig. 11 Some figures-of-merit for conventional and alternative plasmonic materials. a Q from Wang and Shen [175], b Q LSPR from West et al. [15], c Fa and d Jo from Lalisse et al. [176], and α/V from Arnold and Blaber [177], for Ag (using the dielectric function of [12]), Au [13], TiN [26] and c-Na x WO 3 [82]. The solid circles indicate the energy where ε 1 = − 2, which is the energy of the dipole LSPR of a quasistatic sphere. Silver dominates the Fa and Jo spectra, and so axes have been clipped to highlight features from the spectra of the other materials accessible for Na x WO 3 nanoparticles (i.e. below ≤ 2 eV for x = 1), Au should support higher-quality resonances than Na x WO 3 nanoparticles as those FoMs are greater at lower energies. This means Au requires complex nanostructuring in order to achieve resonance qualities equivalent to those that Na x WO 3 shows for simple geometries. The peak Fa and Jo are higher for Na x WO 3 than Au or TiN at ℏω(ε 1 = − 2), and although the definitions of Fa and Jo can be modified to describe ellipsoids [176], it is not clear if they can be modified to model larger particles or other geometries. Still, these results suggest that applications which require nearfield electric field enhancement or localised heating could use Na x WO 3 in place of either Au or TiN. Indeed, the strong local field enhancement of M x WO 3 is exploited in plasmonic photocatalysis (see Sect. 6.4), and strong heat generation is used in plasmonic photothermal therapy (see Sect. 6.3). These results would also suggest that M x WO 3 could be used to improve performance in plasmonic solar cells [6,178] or in surface-enhanced Raman spectroscopy [90,179,180], but few studies have investigated applications in this area.
The tungsten bronzes can also be used for plasmonic chemical sensing, where changes in the refractive index surrounding the nanostructure (n d = √ε d ) can be detected from the shift in LSPR wavelength. For a given particle geometry, larger wavelength shifts (i.e. a greater Δλ LSPR / Δn d ) are predicted for Na x WO 3 than for the noble metals [82]. Where the LSPR absorption peak is broadened due to a broad particle size distribution, the absorption minimum can also be used [181]. Similarly large shifts were predicted for Cs x WO 3 nanoparticles from simulation, although the experimental shift was not found to be as large [109]. Inhomogeneity in structure size and composition (x) in real samples broadens the observed LSPR response, meaning that progress in applications such as chemical sensing or nanophotonics has been hampered by the lack of high-precision nanofabrication techniques [100,169]. In chemical sensing systems, this increases the minimum resolvable Δn d , and in nanophotonics systems leads to reduced performance overall. Until fabrication control is improved, the tungsten bronzes are unlikely to replace Au or Ag in applications where narrow linewidth resonances are needed. Instead, most of the modern applications of plasmonic M x WO 3 take advantage of the broad LSPR responses, as discussed below.

Solar-control Filtering
Most of the publications regarding M x WO 3 nanoparticles in the past decade have focused on their NIR absorbance and scattering. This has led to the development of M x WO 3 solar control filters. These filters work by blocking UV and infrared light but allowing visible light to pass. Such a filter coated on the exterior windows of vehicles or buildings would allow sunlight to be used for interior lighting but would limit the heating associated with thermal infrared. This reduces the amount of energy needed to cool a building using air conditioning. Figure 12 shows transmittance functions of regular glass and an example Na x WO 3 :PVP filter (based on methods in [182]) alongside the AM 1.5 solar spectrum.
The ideal solar filtering material would have high extinction in the UV and NIR and high transmission in the visible. Such a material is difficult to design from a band-engineering perspective, so research in this area has focused on metals with low plasma frequencies: interband transitions provide absorbance in the UV, and LSPRs scatter and absorb light in the infrared. Candidate solar-control materials include transparent conducting oxides such as ITO [186] or AZO [187], or rareearth hexaborides such as LaB 6 [188]. Due to their ideal optical properties, M x WO 3 (particularly Cs x WO 3 ) has also been widely investigated as materials for solar control filters.
A typical M x WO 3 solar control filter consists of M x WO 3 nanoparticles dispersed in a transparent polymer which is layered onto glass. Some of the tungsten bronzes which have been investigated for solar control filtering include Na x WO 3 [63,100,133,189,190], K x WO 3 [191][192][193][194][195], Rb x WO 3 [193], Cs x WO 3 [76,114,117,126,127,134,184,190,[196][197][198][199][200][201][202], Li x K y WO 3 [197], Na x Cs y WO 3 [85,102,116] and (NH 4 ) x WO 3 [123,193]. In order to produce nanoparticles, it is common to ball-mill the products of some other M x WO 3 synthesis technique [63,76,184,201]. The samples are often milled in a solvent or a dispersant, which may be the same material used to produce the polymer film. Some commonly used polymers and dispersants include colloidon [192][193][194]197], ethocel [102], polyvinyl alcohol (PVA) [191,200,203], polyvinyl butyral (PVB) [184,199] and polyvinylpyrrolidone (PVP) [123,204], though reports with Fig. 12 The transmittance of a typical Na 0.7 WO 3 :PVP solar-control filter compared to ordinary glass and the AM1.5 solar spectrum. As with many M x WO 3 filters [183][184][185], the LSPR response extends to red wavelengths, so this filter is light blue in colour new polymers are frequently published. Solar-control filtering can also be applied to textiles for use in clothing [205]. Evidence of the rapid growth of this field comes from Zhou et al. [184] and Liang et al. [206], who have both reported bulk-fabrication of Cs x WO 3 -based solar-control filter via a roll-to-roll process, and for the acceptance of several patents related to M x WO 3 nanoparticles for solar-control filtering [207][208][209][210].
One underexplored aspect of M x WO 3 -based solar-control filters is their long-term chemical stability. Lee et al. observed that Na x WO 3 and Cs x WO 3 gradually lost their NIR shielding properties in the presence of water and oxygen and found that silicon-based polymers were most effective at maintaining strong NIR extinction over time [190]. Several groups have reported fabrication of films containing M x WO 3 mixed with metal oxides such as SiO 2 [118] TiO 2 [194,197] or ZnO [193,199]. Some authors targeted multifunctionality in their films (for example, to combine photocatalytic and solar-control properties, as shown in Fig. 13) [193,194,197], but for others, the aim was to improve the environmental stability of the films by coating M x WO 3 with oxides which are impervious to H 2 O and O 2 [118,199]. Notably, Zhou et al. found that their films turned dark when exposed to UV light, as free H• radicals (generated from the surrounding polymer R via RH → R + H•under UV irradiation) intercalate into the Cs x WO 3 [184]. This "antibleaching" could be minimised by careful choice of polymer. It is essential that the corrosion mechanisms are understood and device stability is controlled for the long-term commercial viability of M x WO 3 solar-control filters.
There are several ways that are used to quantify the performance of solar-control filters, making it difficult to compare the performance of filters from different authors. Some compare the transmittance at arbitrarily selected wavelengths in the visible and NIR [85,116,196,202], but transmittance is sensitive to the optical path length (and thus the film thickness) and says little about the films apparent colour or its ability to reduce thermal heating. There have been some attempts to define a figure-of-merit (FOM) to quantify performance. One proposed FOM is the ratio: where T(λ) is the transmittance of the film. Ψ r can represent either the efficiency function for visible light (Ψ v , integrated over 400-700 nm) or a standard solar spectrum (Ψ s over approximately 240-2600 nm) [184]. Another proposed FOM is the solar energy transmittance selectivity (SETS). Several variations have been proposed [118,201], but one of the most general is [199]: This extends on T r from Eq. 11 to produce a single-valued FOM. SETS = 1 is a perfect NIR filter, SETS = 0.5 is a filter which blocks no light and SETS = 0 is a filter which blocks (thus minimising air conditioning use), and photocatalytic activity to degrade atmospheric pollutants. For winter, the orientation of the window can be reversed, reducing NIR heat loss via the window and minimising condensation due to the films hydrophilicity. Reproduced from [194] under CC BY 4.0 [105] only visible light [118]. Overall, the most compelling evidence for the merits of M x WO 3 solar control filters come from testing the filters in a simulated solar environment. In a typical experiment, light from an incandescent lamp is directed at the windowed-face of a sealed box. The window is coated in different solar control filters, and the temperature inside the box is monitored with time. Some have shown that windows coated with M x WO 3 solar control filters are slower to heat up than windows without any filters [119,184,191,192,195], but comparing filters between different authors is also difficult as the exact performance will depend on factors such as lamp power, window size and box geometry and materials. Comparative studies between different solar control materials (such as LaB 6 or ITO) are limited and would be useful to predict the most economic solar control material for industrial-scale filter production.
In summary, the optical properties of M x WO 3 make it ideal for fabrication of solar-control filters, and significant progress has been made in this field in terms of optimising material synthesis, device fabrication and environmental stability. As the technology matures and commercialisation approaches, more efforts into standardising performance characterisation will be necessary.

Plasmonic Photocatalysis
The performance of some photocatalyst semiconductors can be enhanced by combining them with plasmonic materials [7,211,212]. In such a system, a plasmonic metal nanoparticle (often Au, Ag or Pt) is placed in close contact to a photocatalytic semiconductor (such as TiO 2 , ZnO or WO 3 ) [7]. Several effects work to modify the photocatalytic responses: in brief, a Shottky junction at the metal-semiconductor interface facilitates the separation of electrons and holes, and LSPRs in the metal nanoparticle inject hot electrons into the conduction band of the semiconductor [7,211]. These processes, amongst others, work together to enhance the photocatalytic response of the semiconducting oxide.
Some M x WO 3 exhibit intrinsic photocatalytic properties. Wang et al. reported that hydrated h-Na x WO 3+x/2 •nH 2 O (x ≈ 0.25) nanowires photocatalytically degraded methylene blue (MB) at a higher rate than commercial WO 3 . They proposed that W 5+ sites introduced by Na + doping act as reducing centres [213]. A similar mechanism was reported in the photocatalytic degradation of MB by Cs x WO 3 [128]. Doping WO 3 with refractory metals such as Nb [214] or Ta [215] adjusts the band edges to enable water-splitting. Chiang et al. described the photocatalytic degradation of MB from β-pyrochlore CsW 1.6 O 6 using a mechanism more typical of semiconducting metal oxides [216].
More commonly, tungsten bronzes are used as part of a plasmonic-photocatalytic composite. Some recent reports of M x WO 3 -based systems include the degradation of methyl orange by a fluorinated-TiO 2 :K x WO 3 composite [194] (partially illustrated in Fig. 13), rhodamine 6G by TiO 2 :Na x WO 3 [182], NO n by Nb-doped TiO 2 :Cs x WO 3 [197], 4-nitrophenol by Ag 2 O:Na 0.2 WO 3 [217], reduction of N 2 to NH 3 by graphitic-C 3 N 4 :Cs x WO 3 [218] and the degradation of various dyes by WO y :Cs x WO 3 [117]. It has been observed that the optical properties which are ideal for plasmonic photocatalysis are similar to those for solar-control filtering, allowing the development of dual-functionality composites, such as F-TiO 2 :K x WO 3 [194] or Ag 2 O:K 0.3 WO 3 [219].
Not all authors attribute NIR absorption to LSPRs in their M x WO 3 composites. Cui et al. reported water-splitting from a WO 2 :Na x WO 3 composite [220]. They note that the electronic band structures of WO 2 and Na x WO 3 are suitable for water reduction and oxidation respectively, and thus only work in tandem. Similar results were reported by Zhao et al. who synthesised WO 2 :Na x WO 3 composites decorated with carbon dots (CDs) [221]. Neither of these groups attributed the strong NIR absorption of their samples to LSPRs, and both interpreted their results in terms of electronic transitions. Detailed study by Li et al. showed that this is reasonable for a AgBr:Cs x WO 3 composite [222], but more work is needed for other M x WO 3 -based composites to untangle the contributions of electronic and plasmonic effects.
If the NIR optical properties of the M x WO 3 :photocatalyst composites are interpreted in terms of LSPRs, then comparison can be made between the tungsten bronzes and conventional plasmonic materials. Some M x WO 3 nanoparticle fabrication techniques are highly scalable to bulk quantities and use inexpensive reagents [100,184], suggesting that M x WO 3 is cost-effective replacements for the noble metals in largescale plasmonic photocatalytic systems. Inhomogeneities in particle size and composition in a M x WO 3 sample lead to broadening of the ensemble LSPR responses [100,169]. Although this is detrimental to applications requiring narrowbandwidth resonances (see Sect. 6.1), this is not an issue for technologies which utilise solar radiation, where capturing more of the solar spectrum is beneficial. The promising early results of the application of the tungsten bronzes in photocatalytic systems, and the potential societal and environmental impacts of low-cost efficient photocatalytic processes, warrant much further investigation in this area.

Plasmonic Photothermal Therapy
Plasmonic photothermal therapy (PPTT) is an emerging technique for the treatment of tumours. Some biomolecules overexpose in cancerous tissue compared to healthy tissue.
The specific biomolecule can be conjugated with a plasmonic nanoparticle and injected into the patient. The biomolecules, and thus the nanoparticles, will overexpose in the cancer tissue. Irradiation of the tumour at ω LSPR will excite plasmons, which generate heat as they decay, raising the local temperature and killing the cancerous tissue [223].
The required optical properties for plasmonic nanoparticles for PPTT are more specific than for other applications. The material must be non-toxic and should have LSPR wavelengths tuned to fall in one of the two biological NIR windows: 650-950 nm or 1000-1350 nm [224]. Typically, Au is used for PPTT due to its relatively low LSPR frequencies (compared to Ag), ease of functionalisation and biological inertness. However, several authors have demonstrated PPTT based on tungsten bronze nanoparticles. Unlike Au, where the resonance frequencies can only be modified by particle geometry, M x WO 3 also allows tuning by composition.
PPTT using a tungsten bronze was first reported in 2013 [163] and has since been demonstrated using Na x WO 3 [225][226][227], K x WO 3 [228], Rb x WO 3 [163], Cs x WO 3 [229], (NH 4 ) x WO 3 [107,[230][231][232] and Gd-doped Na x WO 3 [131]. Typical laser wavelengths for PPTT include 808 nm [163,231], 915 nm [225,229], 980 nm [131,163,226,233] and 1064 nm [107,228]. An example is shown in Fig. 14, where TeO 2 /(NH 4 ) x WO 3 nanoribbon composites were irradiated with a 1064 nm laser to induce localised heating in cancerous cells [230]. Cells injected with the nanoribbon composites experienced (a) dramatic localised heating, resulting in (b) a significant reduction in cell viability. The broad NIR response of M x WO 3 nanoparticles means that PPTT can be performed at a relatively wide range of wavelengths. Although some cytotoxicity assays have shown high cell viability after incubation with pure M x WO 3 without laser irradiation [231,233], others have shown intrinsic and phototoxicity to zebrafish at concentrations around 1 mg/ mL [234]. Regardless, nanoparticles are often coated with polyethylene glycol (PEG) to improve biocompatibility [131,225,226,229]. More study is needed in this area.
There has been progress in designing dual-purpose M x WO 3 -biomolecule conjugates. M x WO 3 has large X-ray attenuation coefficients (owing to the strong scattering of W) and can be used as an X-ray contrast agent in X-ray computed tomography (CT) scans in addition to PPTT [163,225]. The large absorption cross-section in the NIR also leads to a strong photoacoustic (PA) response, meaning M x WO 3 nanoparticles can also be as a contrast agent in PA imaging [163,226]. Ni et al. reported that oxygen-deficient Gd 3+ -doped Na x WO 3 can be used as a contrast agent for magnetic resonance imaging, in addition to its use in PPTT [131]. Lastly, Tian et al. conjugated PVP-coated Rb x WO 3 with doxorubicin (DOX), a chemotherapy drug [163]. The DOX desorbed from the Rb x WO 3 nanoparticle upon NIR irradiation, leading to a simultaneous PPTT and chemotherapy treatment.

Summary and Outlook
In this review, we have described how the tungsten bronzes (M x WO 3 ) can be used as alternative plasmonic materials. Their novel electronic properties can be understood in terms of their varied crystal structures: the inserted M species reside in the tunnels formed by corner-sharing WO 6 octahedra is ionised and donates free electrons to the WO 3 conduction band such that the free-electron density is proportional to x. The plasma frequency, tunable by varying x, lies below the onset energy of interband transitions, leading to freeelectron-like optical and plasmonic properties at visible and NIR frequencies. LSPR frequencies and plasmon damping losses are similar to those of Au, suggesting that the tungsten bronzes could be used as a lower cost alternative material for some applications.
The lack of robust nanofabrication techniques is perhaps the greatest barrier to the wider use of M x WO 3 in plasmonics. Decades of research have led to mature methods for the nanostructuring and functionalisation of noble metal nanoparticles. This review has shown that M x WO 3 supports plasmon resonances with quality comparable to Au, but sample inhomogeneity (both chemical and in particle size) broadens the responses. Although this is acceptable, or even desirable, for some applications, it is an enormous detriment to others.
Another area which warrants further study is the chemical stability of the tungsten bronzes. Though they have long been described as being chemically stable [50,95] (or having "very great chemical inertia" [52]), the studies reviewed here have shown that corrosion may be observable in nanoparticulate samples [120,190,199]. Research is needed to determine: the conditions under which M x WO 3 can corrode; the corrosion kinetics; and the corrosion products. These questions will affect the chemical environments where M x WO 3 can be used, the lifetime of plasmonic devices which use M x WO 3 , and the toxicity of the "spent" material. If corrosion is proven to be a significant concern, then further studies into passivation or coatings are needed.
Overall, the tungsten bronzes have some material properties that uniquely distinguish them from typical plasmonic materials. The cubic tungsten bronzes have low-loss and highly tuneable plasmonic properties, and the hexagonal or tetragonal tungsten bronzes appear to exhibit anisotropic electronic and optical properties, with different plasma frequencies in each crystallographic direction and more complex LSPR spectra. With variable plasma frequencies, the LSPR response of tungsten bronze nanoparticles can be varied by composition, in addition to by particle morphology. Several groups have reported bulk-scale nanoparticle fabrication routes from inexpensive reagents, making the tungsten bronzes well suited to industrialscale applications of plasmonics, such as solar-control filtering and plasmonic photocatalysis. These industrial-scale synthesis routes are one of the most promising areas of new research for these materials. If these synthesis techniques are further refined for nanoparticle shape control, we anticipate few other barriers to the broad use of the tungsten bronzes in current and nextgeneration plasmonic devices and applications.