Applied Catalysis B: Environmental Sol gel graphene/TiO 2 nanoparticles for the photocatalytic-assisted sensing and abatement of NO 2

Human exposure to volatile organic compounds and NO 2 can lead to health problems, therefore strategies to mitigate against the risks are required. Abatement and sensing are approaches which could both neutralise and monitor these species thus providing a safer environment and warning occupants of harmful levels. This paper presents pure TiO 2 and TiO 2 /graphene hybrids synthesized through a sol-gel route. Electron optical, helium ion microscopy, X-ray di ﬀ raction and spectroscopic methods have been applied to elucidate the physical and che- mical behaviour. NO 2 sensing properties of TiO 2 /graphene hybrids formed by the addition of graphene to the reaction vessel prior to initiating the sol gel reaction followed by annealing ( GTiO 2 S), and an alternative manufacturing method involving the addition of graphene to TiO 2 nanoparticles which had already been an- nealed ( GTiO 2 M ) were compared and evaluated. A conductometric sensor based on TiO 2 /graphene prepared using material GTiO 2 S showed a higher response to NO 2 compared to sensors based on pure TiO 2 and TiO 2 / graphene prepared with material GTiO 2 M . Under UV irradiation generated by a low power LED, the sensor showed a remarkably enhanced response to 1750ppb NO 2 , about double the response in the dark, and a limit of detection of about 50ppb of NO 2 (Signal/Noise=3). Photocatalytic tests to assess the degradation of NO x showed that TiO 2 /graphene hybrids using material GTiO 2 S were the most active amongst the whole series of TiO 2 -based materials. Our data highlights the unique characteristics of material GTiO 2 S TiO 2 /graphene and the suitability for multi-purpose applications in the ﬁ eld of environmental monitoring and remediation. The capability of the material for both sensing and abatement of NO x could be exploited to o ﬀ er a safer environment through providing a warning of the presence of NO x whilst also reducing levels.


Introduction
Awareness of the health risks associated with the presence of atmospheric pollutants and volatile organic compounds (VOCs) in the indoor and outdoor environment is stimulating research aimed at improving air quality. In the UK each year approximately 40,000 deaths can be attributed to outdoor air pollution. Many major health challenges of modern societies such as cancer, asthma, stroke, heart disease, diabetes, obesity and dementia can be linked to air pollution. The the semiconductor is irradiated by ultraviolet (UV) light [6]. The portion of UV in the solar spectrum only accounts for a few percent in the outdoor environment and even lower indoors [7]. The amount of UV in solar radiation greatly reduces the ability of TiO 2 to catalyse the oxidations or reductions of harmful compounds into less dangerous ones such as water and carbon dioxide. PC performance can be modified by several different methods including modification of the band gap and doping with additional elements such as transition metals, noble metals and non-metallic elements [8][9][10]. Coupling the TiO 2 with additional structures or compounds such as other semiconductors, organic molecules or oxides can improve PC performance [11][12][13][14]. Graphene with its unique electron properties has been combined with TiO 2 in several studies to improve performance [15][16][17][18][19]. Some studies have proven that graphene has electron accepting and transport properties that can be successfully used to improve the recombination time in the semiconductor [20].
TiO 2 in the form of nanoparticles (NPs) has a widespread use in the improvement of indoor and outdoor air quality [2,[21][22][23][24]. Poor indoor air quality (IAQ) can be attributed to emission of inorganic pollutants and VOCs from a variety of materials in buildings [25][26][27][28][29], including furnishing, paints, carpets and air fresheners. The presence of various pollutants contribute to sick building syndrome (SBS) [30,31], which is the negative impact of a buildings environment on the life and wellbeing of its inhabitants.
Among these pollutants, NO 2 contributes largely to poor IAQ. This toxic gas is produced from ground-level emissions related to the burning of fossil fuels from vehicles, power plants, industrial sources, and off-road equipment. Several air cleaning technologies have been therefore developed to improve IAQ by reducing NO 2 concentration levels [32,33]. PC materials based on TiO 2 are the most used for the design of highly efficient systems for environmental reduction of NO x [34].
Long term exposure to NO 2 concentrations higher than those normally found in the ambient air are known to cause an increased risk of acute respiratory illnesses, especially in vulnerable groups. To address concerns associated with NO x levels the US Environmental Protection Agency (EPA) has set a 1-hour NO 2 standard at the level of 100 ppb [35]. Sensors that continuously monitor these low NO 2 concentrations in air with the added capability of alerting occupants when NO 2 levels have reached harmful levels could be a powerful strategy in the prevention of unacceptable exposure. Both solid electrolyte electrochemical and conductometric sensors are available for detection of NO 2 [36][37][38][39]. Wide band gap semiconducting metal oxide (MO) based conductometric sensors are commonly used for this purpose, due to their simple functioning, low cost, mass production and easy integration with integrated microelectronic circuits. Among these sensing materials, titania was investigated for developing NO 2 sensors [40]. UV-or Vis-light operating TiO 2 sensors were also proposed for improving the performance at room to low temperatures and in addition reducing the power consumption of the device [45]. However, the UVor Vis-light operating TiO 2 sensors for room temperature NO 2 monitoring so far proposed, do not reach the sensitivity which is needed for environmental applications [41,42].
Owing to the importance of this issue, this paper focuses on the design and preparation of a multifunctional material, with gas sensing and catalytic activity, for the monitoring and abatement of NO 2 with the aid of UV-or Vis-light. For this purpose TiO 2 and graphene/TiO 2 hybrids were synthesized through a simple sol-gel method. As the morphology and microstructure play a fundamental role in both the PC and sensing properties, a detailed characterization has been undertaken. Field emission and conventional scanning electron microscopy (FESEM/SEM), helium ion microscopy (HIM), X-ray diffraction (XRD), BET surface area analysis, differential reflectance spectroscopy, Raman spectroscopy and energy dispersive X-ray analysis (EDX) were used to characterise the physical and chemical properties of both pure TiO 2 , the hybrid graphene/TiO 2 nanoparticles and commercially available Degussa P25 ® TiO 2 as a reference [43]. As the composition of commercial Degussa P25 ® is reported to be inhomogeneous between different batches [44], the very same batch was used for all the PC tests in the present work. Then, the PC and sensing properties of synthesized TiO 2 and Graphene/TiO 2 hybrids were investigated in detail and their performance was correlated with their microstructural characteristics.

TiO 2 synthesis
Preparation of TiO 2 through the sol-gel method is based on the hydrolysis and polymerization of a metal-organic precursor to form a colloidal suspension [45,46]. In the synthesis, 10 ml of tri-tetraisopropoxide (TTIP) (Sigma Aldrich, ≥ 98%, MM = 284.22; d = 0.96 g/mL) was mixed with 40 mL of 2-propanol (Sigma Aldrich, ≥ 99.7%, d = 0.785 g/mL) in a round bottom flask. The flask, which also acted as the reaction vessel, was positioned above a hotplate/magnetic stirrer within a fume cupboard. A flat crystallizer dish containing the reaction vessel was filled with water thereby ensuring the flask was heated evenly at a temperature of 80°C A separator funnel was filled with deionized water and adjusted to a pH of 2 using nitric acid measured with an Omega PHB23 pH-meter mounted above the round bottom flask. In the synthesis of undoped TiO 2 , acidified deionised water was added to the alcoholic solution with the titanium precursor. The acid provided the optimum pH for the hydrolysis reaction.
The 2-propanol and TTIP precursor were mixed under continuous vigorous stirring for 1 h at a temperature of 80°C before the acidified water was introduced into the reaction vessel at a rate of 2 mL min −1 to initiate the hydrolysis reaction. After continuous stirring for 6 h a white coloured colloidal liquid containing 5 g L -1 TiO 2 was obtained through the reaction described in Eq. (1): After completion of the reaction all the excess solvent was removed immediately by evaporation through continuous stirring on a hotplate at 80°C. The TiO 2 was initially amorphous and required annealing to form the anatase crystalline phase that exhibits higher PC activity compared to amorphous or rutile TiO 2 [47,48]. Annealing was carried out using an Elite BRF 14/10-2416 CG furnace with the material held in an alumina crucible of internal volume 20 ml. The annealing was achieved by ramping the temperature at a rate of 200°C/h followed by a dwell at 450°C, for 2.5 h before cooling.

Graphene/TiO 2 sample synthesis
Commercial graphene nanoplatelets, CamGraph ® G1, kindly supplied by Cambridge Nanosystems, were used. These were manufactured using a surfactant-free microwave plasma synthesis process. According to the manufacturers analysis these consisted of 3 ± 2 layers and had a surface area of 320 ± 20 m 2 /g with a typical elemental analysis of 97.8% carbon, 0.68% hydrogen, 0.29% nitrogen and 0.99% oxygen. A Raman spectrum and XRD pattern of the graphene are presented in Fig.  S1a and b (Supplementary Information File). Graphene/TiO 2 samples were prepared following two different protocols.
In the first synthesis, graphene was added to the reaction vessel together with the TTIP, equating to 10 wt% of graphene. The mixture was then stirred at 80°C for 2 h before initiating the hydrolysis reaction with acidified water. A grey colloidal solution was formed, with a pH of 1.8measured using an Omega PHB23 pH-meter. Following the completion of the reaction, the sample was recovered, by evaporation of the solvent in the hotplate, and annealed following the procedure described above in the Section 2.1: "TiO 2 synthesis"at 450°C by ramping at a rate of 200°C/h and a 2.5 h dwell time at the selected temperature. This material is referred to as GTiO 2 S.
In the second procedure, TiO 2 was prepared following the synthesis previously described in Section 2.1 "TiO 2 synhesis". It was then placed in a beaker, together with a mixture of ethanol and water (50% wt), and 10% wt of CamGraph graphene. This solution, was stirred at 70°C for two hours, and then the solvent was removed by evaporation on the hotplate at 80°C, to collect the powder. As the TiO 2 was annealed prior to the graphene being added, the graphene was not subjected to any heat treatment. This material was named GTiO 2 M.

Sample characterization
Analysis using X-ray diffraction (XRD) was carried out to evaluate the weight percentage of crystalline phases in the specimens prepared in addition to microstructural features. Semi-quantitative phase analysis (QPA) of the XRD data was performed using the Rietveld method. Rietveld refinements were assessed using the GSAS-EXPGUI software packages [49,50]. XRD data were collected on a PANalytical X'Pert Pro (NL) θ/θ diffractometer, equipped with a fast real-time multiple strip (RTMS) detector (PIXcel 1D, PANalytical), using Cu K α radiation (45 kV and 40 mA) with a virtual step size of 0.02°2θ and virtual time per step of 200 s over a 20-80°2θ range. The instrumental broadening was measured using the NIST SRM 660b standard (LaB 6 ) with data collected under the same conditions as those used for the TiO 2 samples. XRD was also used to determine microstructural features from the specimens. For this purpose, XRD data were collected in the same instrument with identical set-up as that used for QPA analysis. A higher signal-to-noise ratio was achieved in the range of 20-145°2θ, using a virtual step size of 0.1°, and a virtual time per step of 500 s. The instrumental contribution was obtained by parameterising the profile of fourteen (hkl) reflections from the NIST SRM 660b standard (LaB 6 ), according to the Caglioti et al. relationship [51]. Microstructural features of the specimens were analysed through the whole powder pattern modelling (WPPM) method [52], as implemented in the PM2K software package [53]. By means of this novel methodology, the size distribution of individual phases in NPs can be accurately defined [54]. In this work, crystalline domains were assumed to be spherical, and their diameter distributed according to a log-normal curve.
Raman spectra where acquired using a Renishaw System 2000 spectrometer equipped with an inVia Raman microscope using a He-Ne laser as an excitation source operating at 785 nm and a maximum power of 20 mW. The samples were analysed by focusing the laser with objective magnification ×100 onto the sample surface corresponding to a laser spot diameter of about 10 μm. The acquisition time of 1 s was used for each spectrum over the wavenumber range 100-3500 cm −1 with a 4 cm -1 resolution.
The morphology of the aggregates of TiO 2 and graphene/TiO 2 was studied using a JEOL JSM-6480lv scanning electron microscope (SEM) with Oxford INCA Energy X-ray Analyser correlated using an acceleration voltage of 20 keV. A higher magnification study of the particles using a helium ion microscope (HIM) was obtained using a Zeiss Orion NanoFab instrument. The images were acquired using He + ions with an accelerating voltage of 25 kV and beam current of 0.3 pA with the angle varied between 0 and 54 deg. Images of higher magnification were obtained using a JEM-2100Plus Transmission Electron Microscope (TEM) using an acceleration voltage of 200 keV and a spot size of 1 nm.
Diffuse reflectance spectroscopy (DRS) was performed with a Shimadzu UV 3100 (JP) spectrometer and spectra of the samples were acquired in the UV-Vis range (250-850 nm), with 0.2 nm step-size, and Spectralon ® white reference material. The Kubelka-Munk theory, Eq.
(2), was applied to allow conversion of the diffuse reflectance into the pseudo-absorption coefficient α, afterwards, the Tauc plot was used to estimate the electronic transition in the specimens prepared: where h is Planck's constant, ν is the photon frequency, α is the pseudoabsorption coefficient, E g is the band gap, and A is a proportionality constant. The value of the power coefficient γ denotes the nature of the electronic transition, whether allowed or forbidden and whether direct or indirect: for direct allowed transitions: γ = 1/2, while for indirect allowed transitions, γ = 2 [55]. The specific surface area (SSA) of the prepared samples was determined using the Brunauer-Emmett-Teller (BET) method (Micromeritics Gemini 2380, US), using N 2 as the adsorbate gas.

NO x photocatalytic activity
Gas-solid phase PC study was carried out in a reactor operated continuously, as previously described in detail by Lucas et al. [56]. It comprised a cylindrical chamber (∼35 L in volume, 70 cm in height, 25 cm in diameter) constructed from a high grade stainless steel with a stainless steel top with a sealed glass window to allow the light to reach the sample that was placed inside. A schematic diagram is shown in Fig. 1, whilst a digital photograph of the working rig is in Fig. S2. The light source employed was a solar lamp (Osram Ultra-Vitalux, 300 W, 280 to 780 nm), placed 65 cm from the photocatalyst. The light intensity reaching the samples, measured with a radiometer (Delta OHM, HD2302.0, IT), was found to be approximately 3.6 W m −2 in the UV-A range and 25 W m −2 in the visible-light range. A Petri dish 6 cm in diameter was filled up with a layer of powder. The tests were performed at 27 ± 1°C (temperature inside the reactor) with a relative humidity (RH) of 31%. These parameters were controlled by means of a thermocouple that was placed inside the chamber and a humidity sensor placed in the inlet pipe and they remained stable throughout the tests. The concentration of the pollutant gas exiting through the outlet was measured using a chemiluminescence analyser (AC-31 M, Environment S.A).
The experiment was conducted by firstly placing the petri dish containing the photocatalyst inside the reactor and covering the glass window with a sheet of aluminium. The inlet gas mixture (prepared mixing gas cylinders containing synthetic air and NO x ) was allowed to flow into the chamber until it stabilized at a concentration of 200 ppb. The mixture of air with that concentration of NO x was guaranteed using two mass flow controllers with a flow rate of 1 L min -1 . The condition of continuous flow was necessary to achieve sample saturation, ensuring that, during the test, the PC process was the only one involved (i.e., no absorption from the sample, nor from the reactor walls) [57]. Once a concentration of 0.2 ppm was reached, the window glass was uncovered, the lamp turned on, and the PC reaction started. When the pollutant concentration stabilized to a minimum concentration the NO x was no longer being decomposed by the catalyst. The PC efficiency was then expressed in terms of the conversion rate (%) of the initial NO x concentration which was calculated using Eq. (3) [58]: where (NO x ) 0 and (NO x ) t are the initial and final concentrations (both expressed as ppb) after an irradiation time, t.

NO 2 sensing tests
NO 2 sensing tests were carried out on conductometric platforms fabricated using alumina substrates (3 × 6 mm), supplied with interdigitated Pt electrodes and a heating element on the back side. Printable pastes of sensitive materials were obtained mixing TiO 2 based samples with distilled water, then each paste was screen printed onto the interdigitated electrodes to obtain a thick film (∼ 10 μm) with a sensitive geometrical area of 3 × 3 mm. The high porosity of the alumina meant that no binder was necessary to enhance the adhesion of the sensing layer on the alumina substrate.
All measurements were performed at room temperature (RT) in a 10 ml volume test chamber equipped with a 400 nm UV-vis low power LED (180 mW) [59,60]. For the entire duration of the measurement, the sensor was exposed to a constant synthetic air total stream of 100 sccm (standard cubic centimeters per minute) with a RH of about 40%. The resistance of the sensors was monitored using a Keithley 6487 Picoammeter at an applied voltage of 1 V. The sensing tests under UV excitation were performed exposing the sensor to the target gas while the sensitive film was constantly irradiated with the UV light. The gas response was defined as the ratio S = R 0 /R, where R 0 represents the electrical resistance of the sensor in air and R is the electrical resistance of the sensor at a given NO 2 concentration.

Morphological characterization
SEM, HIM and TEM images were acquired to investigate the morphology of the synthesized powder. Fig. 2a shows a low magnification image of one of large aggregates observed on the pure TiO 2 sample. The rough surface of TiO 2 aggregates, consisting of nanoparticles around 20 nm ± 5 nm, is clearly observed at high magnification (Fig. 2b). TEM analysis confirmed the nanometric size of TiO 2 particles and their aggregation even at nanometer level ( Fig. 2c and d).
Electron microscopy of the GTiO 2 S sample is reported in Fig. 3. The presence of graphene in the GTiO 2 S modified the nucleation and growth of TiO 2 aggregates as shown in Fig. 3a. The dimension of the particles measured in post processing does not change significantly when compared with pure TiO 2 . Fig. 3c) shows how graphene sheets behave as a nucleation point for the titania. Fig. 3d) focuses on the interface between the graphene sheet and the TiO 2 NPs. Fig. 4a) and b) show the intricate network of TiO 2 and graphene in the GTiO 2 M sample that was not observed in the GTiO 2 S sample. Graphene and TiO 2 are deeply embedded together, and this modified the shape and the roughness of the surface of this sample. TEM images were in agreement with those obtained by HIM: Fig. 4c) highlights the intricate network of graphene and TiO 2 produced through the mixing of the two components in alcoholic solution. Fig. 4d) shows the interface between TiO 2 , where atomic lattice planes are visible, and graphene, where atomic planes are visible.

Microstructural characterization
XRD patterns of the samples are reported in Fig. 5, while an example of the Rietveld refinement is depicted in Fig. S3. The agreement factors of the Rietveld refinements are reported in Table S1 in addition to the semi-QPA analyses data. The main microstructural features of the specimens are listed in Table 1. In these analyses, only the crystalline TiO 2 polymorphs were considered. The sol-gel synthesis of the unmodified TiO 2 specimen led to the formation of anatase (64.1 wt%) and brookite (35.9 wt%) TiO 2 polymorphs. This is in agreement with previous research as anatase is the thermodynamically stable TiO 2 polymorph at the nanoscale [61]. Furthermore, the presence of brookite is to be expected, as it is a common product of such sol-gel syntheses [62,63]. The specimen GTiO 2 M, quite obviously, had virtually the same mineralogical composition as the unmodified TiO 2 , cf Table 1. On the other hand, the specimen GTiO 2 S contained slightly less anatase, 57.6 wt%, in favour of a higher amount of brookite, i.e. 42.4 wt%. P25 ® is composed of 76.3 wt% anatase, 10.6 wt% rutile, and 13.0 wt% amorphous phase as showed by Tobaldi et al. [43].
Microstructural information in the form of the size, and size distribution of anatase and brookite NPs in the specimens was determined using X-ray methods (cf Table 1). Quantitative figures are reported in Table S2 and S3 and Fig. 6. Anatase unit cell parameters of TiO 2 and GTiO 2 M are virtually the same, as shown in Table S2; the change in the synthesis method (specimen GTiO 2 S), led to a slight expansion of the anatase c-axis, although the unit cell volume is the same in all the three synthesised samples. As per the brookite unit cell parameters these are, once again, virtually the samewithin the experimental errorin samples TiO 2 and GTiO 2 M. Mixing graphene to the sol, led also in this case to a slight expansion of brookite unit cell volume. Similar observations can be made by considering the average diameter of the anatase crystalline domains in specimens TiO 2 and GTiO 2 M. They are virtually the same, within experimental error, being 6.4 and 7.0 nm, respectively, see Table S3 and Fig. 6. Similarly to observations for the unit cell parameters, GTiO 2 S had a bigger average diameter, being 8.9 nm.
There is no trend for the brookite in the specimens, the average diameter of the crystalline domains being 7.2, 8.8, and 6.2 nm for the TiO 2 , GTiO 2 M, and GTiO 2 S respectively. All the size distributions for brookite had a positive skewness, as shown in Fig. 6.
Raman spectra were obtained to determine the crystalline phase of TiO 2 and to allow a comparison with the commercial sample. Fig. S4a shows the Raman spectrum of pure synthesised TiO 2 showing the presence of anatase and brookite TiO 2 polymorphs, thus confirming results from X-ray diffraction. The presence of anatase in crystalline form is indicated by the presence of its characteristic bands at 144, 197, 399, 519 and 639 cm -1 ; the main brookite Raman modes were detected at: 247, 323, 366 cm -1 (cf Figure S4a). A Raman spectrum obtained from a sample of P25 is reported in Fig. S4b. Anatase, rutile and brookite Raman modes are listed in Table S4 [64]. Raman spectra of the hybrid samples containing graphene are reported in Fig. 7. Both spectra indicate the same bands found in the pure synthesised sample attributed to anatase and brookite. Furthermore, the major bands assigned to graphene (i.e. D and a weak D′ band, together with the G band) are present in the samples. The presence of these Raman features, e.g. D and D′ bands, is an indication of irregular edge disorders and oxidised dangling bonds [65,66]. Furthermore, as they are with essentially the same intensities, this shows that the defect state in graphene virtually didn't vary between the two synthesis methods employed. Moreover, those defects might have acted as nucleation sites for TiO 2 [67].
DRS spectra (not shown) consisted of one single absorption edge, located at around 380 nm, and assigned to the band-to-band transition in titania [68]. The apparent optical E g of the specimens were calculated by means of the Tauc plot, assuming both a direct allowed and indirect allowed transitionsvalues listed in Table 1. The E g values obtained, assuming a direct allowed transition (cf Fig. 8), agree well with the expected E g of anatase (i.e. ∼3.2 eV; ∼388 nm) reported in the literature [47], thus the direct allowed model would seem more applicable for these specimens. However, the E g values extracted with the Tauc formalism are shifted toward higher energies compared with the expected E g value of anatase. This might be a contribution from the non-negligible amounts of brookite in all the specimens. Indeed, the optical band gap of brookite has been reported to be ∼3.4 eV (∼365 nm) [69], probably explaining the higher blue-shift of the specimen GTiO2S, which is the specimen with the higher amount of brookite (cf Table 1).Furthermore, although anatase is commonly reported to be a semiconductor having an indirect allowed transition [70], it has been shown by Serpone and co-workers that, for indirect band gap semiconductors such as anatase, a direct transition might appear, when in a colloidal nanomaterial form [71]. Besides, the presence of graphene did not significantly change the apparent optical E g , as shown in Table 1.
With regard to the structural properties, an increase in the BET specific surface area of the samples was noted, being 79.6, 87.9, and 100.3 m 2 g −1 for the TiO 2 , GTiO 2 M, and GTiO 2 S respectively as reported in Table 1. The sorption isotherms are shown in Fig. S5a-c. They all are of type IV, typical of a mesoporous material. Furthermore, we detected the presence of no micropores (all the pore being in the range of ∼2-25 nm, inset of Fig. S5a-c). The main microstructural and optical characteristics of the as prepared samples, which are of primary  importance in determining their sensing and PC performances, have also been summarised in Table 1.

NO 2 sensing tests
TiO 2 is a widely used gas sensor material which can be activated under light. However, so far, only very few TiO 2 based photo-activated sensors have been studied. Furthermore, the range of concentrations investigated is higher than that required for environmental control. Here, the sensing characteristics of the fabricated conductometric sensors have been tested for NO 2 at room temperature in the concentration    range from 70 ppb to 1750 ppb. Notably, to reduce the size and the power consumption of the photo-activated sensors, a low-power light emitting diode (LED) lamp (180 mW) was used. LEDs are promising for gas sensing applications because of their reduced size, instant on/off operation, long lifetime, low environmentally impact and ability to produce accurate wavelengths at specific intensities. Preliminary tests, carried out to investigate the effect of the UV-vis excitation on the behaviour of the pure TiO 2 and GTiO 2 S sensors toward NO 2 (1750 ppb), are shown in Fig. 9. In the dark, almost no difference in the sensor response is noted. Looking at the sensor response under UV-vis excitation with a 400 nm LED, no appreciable variation occurs for the unmodified TiO 2 sensor, whereas the GTiO 2 S sensor shows a two-fold increase in response. It worth noting that, under UV-vis excitation a quick and reversible recovery of the baseline was observed for both sensors.
The positive response of the synthesized TiO 2 and graphene/TiO 2 hybrids under UV excitation for different concentrations of NO 2 and the related calibration curves, are reported in Fig. 10. To make the comparison clearer, both the data sets have been plotted on the same scale. Both TiO 2 and GTiO 2 S based sensors showed changes in resistance when exposed to NO 2 in the range 70-1750 ppb. On the contrary, the GTiO 2 M based sensor was almost insensitive to NO 2 in the same range of concentration. This may be due to the high conductivity of the sample, through the well-interconnected graphene sheets which promote electronic transport. A decrease in the resistance has been clearly observed after the introduction of NO 2 ( Fig. 10a and b). As NO 2 is an oxidizing gas, this suggests that the tested materials behave as p-type semiconductors. Such behaviour can be explained assuming that NO 2 adsorption on the titania active layers involves the transfer of electrons toward the titania sensing layer, decreasing the resistance of the sensor device. Fig. 10c compares the response under UV excitation to 1750 ppb of NO 2 of the synthesized TiO 2 and graphene/TiO 2 hybrid-based sensors. The results, summarised in Fig. 10d at different concentrations indicate that the GTiO 2 S sensor was the most responsive, in agreement with previous behaviour of TiO 2 -based sensors reported in the literature [41,72]. The calibration curve for the GTiO 2 S sensor reported in log-log scale (see Fig. 11) has a good linear relationship (R 2 = 0.984) according Fig. 7. Raman spectra of GTiO 2 M and GTiO 2 S. Inset shows the magnified wavenumber region 1200-1850 cm -1 to highlight the major graphene bands, D (∼1350 cm -1 ), D′ (around 1620 cm -1 ), and G (1580 cm -1 ). Fig. 8. Kubelka-Munk analysis of the DRS data versus photon energy of the specimens to show the optical E g , calculated using the Tauc procedure. In the inset, the dotted lines represent the x-axis intercept of the line tangent to the inflection point of the sample, i.e. the optical E g according to the direct allowed E g model -(αhν) 2 , γ = ½.
where S is the response (R 0 /R) and c the concentration of NO 2 . The limit of detection (LOD) estimated at a signal to noise ratio of 3, was found to be about 50 ppb of NO 2 .
The gas sensing performances of the GTiO 2 S sensor are compared in Table 2 with previously reported TiO 2 -based NO 2 sensors. It is noteworthy that almost all TiO 2 sensors work at high temperature and only in the ppm range of concentration. In comparison the proposed GTiO2S sensor tested in this work possesses a lower operating range of concentration up to ppb level, functioning at room temperature, with a comparable or superior response to the other TiO2-based formulations with a relatively short response/recovery time.
The data reported above can help to propose a plausible mechanism for the enhanced response in the presence of UV-vis light. It is well known that light irradiation of the sensing layer affects gas detection in metal oxide semiconductors, enhancing the carrier generation. In addition, it is also responsible for the photo-dissociation of the target gas [59]. The results obtained, indicated that the resistance of TiO 2 decreased upon exposure to NO 2 and decreased further with LED excitation, increasing the response. This occurs through the combination of processes reported below, starting with the photo-generation of electron-hole pairs (Eq. 5). The photo-generated holes react with adsorbed oxygen ions, forming O 2 (gas) which leaves the sensing surface (Eq. 6). Adsorption of NO 2 target gas is then enhanced on the clean and more reactive surface. Removal of oxygen contamination from the illuminated surface results in a greater formation of reactive sites for NO 2 adsorption. The photo-induced adsorption of NO 2 has been suggested to be the key pathway for higher response [79,80]. Adsorbed NO 2 extracts electrons from the conduction band (Eq. (7)), increasing the number of holes in p-type TiO 2 and consequently decreasing the resistance.
NO 2 (gas) + e − → NO 2 − (ads) In summary, due to the increased availability of NO 2 molecules adsorbed, the resistance change upon exposure to NO 2 under photoexcitation was higher than in the dark. The substantial enhancement in the response of the GTiO 2 S sensor to NO 2 gas under the LED excitation compared to the other samples was attributed to its higher surface area (smaller particle size).

Photo-catalytic NO x degradation tests
Results of NO x PC degradation using the solar lamp are shown in Fig. 12. All the specimens exhibited PC activity when triggered with a lamp simulating solar radiation. Graphene modified TiO 2 specimens were the most active amongst the whole series of samples, with GTiO 2 S being more active than GTiO 2 M. On the other hand, P25 ® is more photocatalytically active than our unmodified TiO 2 sample.
These results are well confirmed by the initial (first 20 min of reaction) pseudo-first order kinetic constants, cf Fig. 13, in which are reported the values averaged over three consecutive PC activity tests (using the same specimen as in the first PC run) with the respective standard deviation. These data are supported by the literature which states that graphene is a high performance support for photocatalysis [81]. Thus, because electronic transport is enhanced in the graphene/ TiO 2 hybrids, as shown in Section 3.3, a decrease in the resistance of GTiO 2 S, improves PC activity. The photo-generated electrons, from the TiO 2 component, are swiftly delivered to the surrounding graphene sheets in a composite process [82]. Such electrons are involved in reductive reactions, i.e. the formation of active O 2 − species with the  oxygen from the environment, as in Eq. (8) [83]. Furthermore, and most importantly for PC purposes, the removal of these electrons from TiO 2 to graphene, allows for a boost in the separation of the photogenerated pair [84]. Moreover, the incorporation of TiO 2 on a graphene substrate is able, at the same time, to increase the adsorption of pollutants [85], all these factors being beneficial for PC activity. However, data also show that there is a clear difference in the interaction amongst TiO 2 NPs and graphene, depending on the preparation method. Consistent with the sensing experiments, GTiO 2 S behaves itself as the best performing sample in the PC tests despite the thermal treatment that was necessary after the preparation of the specimen. Indeed, such a thermal treatment might be beneficial in synthesising TiO 2 and graphene in more intimate contact (cf Fig. 3d); furthermore, the introduction of graphene directly into the sol was beneficial in producing NPs with a higher SSA, as shown in Table 1. Also, as shown by Trapalis and co-workers [84], the photocatalytic de-NO x performance of graphene-TiO 2 composites reaches a maximum efficiency at low graphene loadings. This was attributed by those authors to the interaction between TiO 2 nanoparticles and graphene sheets, acting as electron traps.
Plotting the degradation of NO and NO 2 versus reaction time of GTiO 2 S, Fig. 14 This observation is consistent with previous literature where Dong and colleagues (via electron spin resonance) have shown that superoxide radicals ( % O 2 − ) are the main active reactive oxygen species responsible for the PC degradation of NO x [87], thus we can state that NO was mainly oxidised by oxygen radicals. Finally, it is well known that HNO 3 is the final product of the NO x degradation. However, as shown    by Ohko and colleagues, this can be easily eluted from the catalyst into water [88].

Conclusions
A sol gel synthesis has been used to synthesize graphene/TiO 2 nanoparticles. Addition of graphene to the reaction vessel prior to initiating the sol gel reaction, followed by annealing, led to an intimately mixed composite (GTiO 2 S). An alternative manufacturing method involving the addition of graphene to TiO 2 nanoparticles which had already been annealed resulted in a material with lower performance (GTiO 2 M). XRD and Raman data were complementary in describing the different structures observed. The as-prepared samples were evaluated as sensing materials for monitoring NO 2 at low concentrations in air. Under UV-vis photo-excitation, the GTiO 2 S sensor was the most responsive. The significant improvement in sensitivity at room temperature under the UV-vis excitation was attributed to higher surface area (smaller particle size) compared to the other samples. The physical mechanisms, i.e. the photo-generation of free charge carriers and the resulting reactions between semiconductor surface and oxygen and NO 2 are likely under LED excitation. Interestingly, the PC properties which result in the abatement of NO x follow the same trend seen in NO x sensing, with GTiO 2 S being the most active material when compared to both GTiO 2 M and TiO 2 pure, making it the most suitable for multipurpose environmental applications.

Conflict of interest
No conflicts of interest.

Acknowledgments
This project has received funding from the European Union's Seventh Framework Programme for research, technological development and demonstration under grant agreement No. 609234. The authors wish to acknowledge Cambridge Nanosystems for supplying the graphene. Andrea Giampiccolo wishes to thank the BRE for funding his research studentship. This work was developed within the scope of the project CICECO-Aveiro Institute of Materials, POCI-01-0145-FEDER-007679 (FCT Ref. UID/CTM/50011/2013), financed by national funds through the FCT/MEC and when appropriate co-financed by FEDER under the PT2020 Partnership Agreement. The Zeiss Orion NanoFab HIM was purchased as part of a package funded by EPSRC's "Great Eight" capital funding Grant No. EP/K022679/1 and Newcastle University. The HIM is housed at the National EPSRC XPS Users' Service, an EPSRC Mid-Range Facility. All data supporting this study are provided as supplementary information accompanying this paper.