Permeability and Selectivity of PPO/Graphene Composites as Mixed Matrix Membranes for CO2 Capture and Gas Separation

We fabricated novel composite (mixed matrix) membranes based on a permeable glassy polymer, Poly(2,6-dimethyl-1,4-phenylene oxide) (PPO), and variable loadings of few-layer graphene, to test their potential in gas separation and CO2 capture applications. The permeability, selectivity and diffusivity of different gases as a function of graphene loading, from 0.3 to 15 wt %, was measured at 35 and 65 °C. Samples with small loadings of graphene show a higher permeability and He/CO2 selectivity than pure PPO, due to a favorable effect of the nanofillers on the polymer morphology. Higher amounts of graphene lower the permeability of the polymer, due to the prevailing effect of increased tortuosity of the gas molecules in the membrane. Graphene also allows dramatically reducing the increase of permeability with temperature, acting as a “stabilizer” for the polymer matrix. Such effect reduces the temperature-induced loss of size-selectivity for He/N2 and CO2/N2, and enhances the temperature-induced increase of selectivity for He/CO2. The study confirms that, as observed in the case of other graphene-based mixed matrix glassy membranes, the optimal concentration of graphene in the polymer is below 1 wt %. Below such threshold, the morphology of the nanoscopic filler added in solution affects positively the glassy chains packing, enhancing permeability and selectivity, and improving the selectivity of the membrane at increasing temperatures. These results suggest that small additions of graphene to polymers can enhance their permselectivity and stabilize their properties.


Introduction
The removal of CO 2 from gaseous streams produced in energy production and energy-intensive industrial processes is one of the most straightforward ways to reduce the global warming effect due to atmospheric CO 2 increase [1,2].
Membrane separations are considered suitable, and environmentally friendly, technologies to capture CO 2 both in post combustion (removal of CO 2 from N 2 ) and pre combustion (removal of CO 2 from H 2 ) [3][4][5][6]. Mixtures of H 2 and CO 2 are generated in the production of hydrogen from steam reforming or biomass gasification, and their purification is necessary not only for environmental aims, i.e., for the reduction of the CO 2 emitted in the atmosphere, but also for technological purposes, i.e., the purification of H 2 for use as a fuel or chemical. Most of those processes are carried out at high temperature, and a high-temperature purification process would allow exploiting the thermal level of the gas stream [4].

Permeability Tests
The permeability was measured with a variable pressure apparatus described in previous works [38], by applying an upstream pressure of about 1.4 bar and vacuum on the downstream side. The pure gas permeability at steady state can be calculated from Equation (1), in which dp i dt s.s. is the slope of pressure versus time curve at steady state, V d is the calibrated downstream volume, R is the universal gas constant, T is the system temperature, A is the membrane area, l is the thickness of the sample and p up i − p down i is the gas pressure difference across the membrane film. P i = dp i dt s.s. The apparatus is placed in a thermostatic chamber which can reach temperatures in the order of 80 • C. Tests were conducted at 35 • C; and for most samples we also carried out tests at 65 • C to investigate temperature effect on permeability of the different gases investigated, i.e., He, N 2 and CO 2 .
The permeability value is also affected by the uncertainty on the membrane thickness, which depends on the sample considered. The error on ideal selectivity of a single sample however is negligible, because such value is unaffected by downstream volume, permeation area and membrane thickness values as it is clear from Equation (1).
Ideal selectivity for the different gas couple of interest is calculated using Equation (2), which is valid in the case of negligible downstream pressure, as in the tests considered here: where the selectivity is defined as α ij ≡ In general, real selectivity can differ from the ideal selectivity, which is estimated using the permeability values performed on the pure gases, but the latter still represents a good indication of the material performance with respect to other membranes as most of the data reported in literature refer to this kind of selectivity [7]. In many cases, we were able also to determine the characteristic time of permeation, i.e., the time required to reach a stable flux across the membrane, the so-called time-lag, t L , that is related to the gas diffusivity in the membrane by the following relation [39]: The time lag is actually estimated as the intercept, on the time axis, of the asymptotic straight line representing the downstream gas pressure versus time. It can also be noticed that, due to the phase equilibrium between the gas and the polymer phase, according to which the gas is absorbed onto the polymer surface proportionally to its solubility coefficient S i , a straight forward relation holds true between the gas permeability, and its diffusivity and solubility coefficient in the polymer: According to which, the selectivity can be decomposed into a diffusivity-based, and a solubilitybased contribution: Usually, for a same gas couple, the two contributions are, respectively, lower than unity and higher than unity. In general, the diffusivity-selectivity prevails, so that smaller gases are generally more permeable than larger ones in most polymeric membranes. In some cases, the solubility prevails, as happens for instance when CO 2 is concerned. For this reason, not all polymeric membranes have a H 2 /CO 2 selectivity higher than 1, but some of them, due to a very high CO 2 solubility, exhibit a CO 2 -selective behavior [40].
The temperature dependence of permeability, diffusivity and solubility is governed by Arrhenius-like laws, as follows: where E P and E D are the permeation and diffusion activation energies, respectively, which are usually positive for gases in glassy polymers, and ∆H S is the sorption enthalpy, which is usually negative. Accordingly, the permeability and diffusivity increase with temperature, while the solubility decreases with it. Due to Equations (4)-(8), the permeation activation energy has a diffusivity and solubility contribution as follows: For the selectivity dependence on temperature is concerned, one can combine Equations (2) and (6) to obtain: which indicates that the selectivity increases with temperature if the more permeable gas has a higher activation energy than the less permeable one, while it decreases with temperature if the opposite is true. In general, the second situation is more frequent, as less permeable gases have higher activation energies. The couples H 2 /CO 2 and He/CO 2 in PPO form an exception, because H 2 and He are more permeable at room temperature than CO 2 , but also have a higher activation energy of permeation. The activation energy of CO 2 is low, due to a high contribution of sorption, an exothermic process, on the permeation. Therefore, for H 2 -selective membranes, as the ones of the present paper, increasing the temperature enhances the H 2 permeability and H 2 /CO 2 selectivity.

SEM Analysis
First, neat graphene in powder was characterized. The Graphene XT7 was chosen for this analysis and two images are reported in Figure 2a,b that show some wrinkles. The effect of dispersing the platelets in water is that of swelling and opening the structure, as shown in Figure 2c,d. Then, the effect of sonication on the sample morphology was analyzed. Pictures of the sample dispersed in water and sonicated for only 10 min exhibit lateral sizes, from the picture analyzed and reported in Figure S1 in the Supplementary Materials, larger than what is declared by the supplier, between 26 and 63 micrometers (based on seven platelets). In Figure S2 in the Supplementary Materials, we report similar pictures taken on a sample of Graphene XT7 (in water suspension) sonicated for 15 h. As can be seen, the platelets analyzed have a lateral size ranging between 34 and 43 micrometers, also based on seven platelets. Therefore, it seems that the sonication produces just a slight decrease of the average lateral size after 15 h, but the size distribution of the platelets becomes narrower.
Pictures of the cross section of PPO + graphene membranes are reported in Figure 3. It is apparent that, by increasing the concentration of graphene, the cross section becomes filled with graphenic domains. Those structures are aligned perpendicular to the membrane cross section, thereby increasing the tortuosity of diffusing molecules. The images also show that some voids form at the interface of PPO and graphene. Images with higher magnification were also taken ( Figure 4), and show that the surface of graphene platelets remains similar to the one observed in the neat powder ( Figure 2). apparent that, by increasing the concentration of graphene, the cross section becomes filled with graphenic domains. Those structures are aligned perpendicular to the membrane cross section, thereby increasing the tortuosity of diffusing molecules. The images also show that some voids form at the interface of PPO and graphene. Images with higher magnification were also taken ( Figure 4), and show that the surface of graphene platelets remains similar to the one observed in the neat powder ( Figure 2). average lateral size after 15 h, but the size distribution of the platelets becomes narrower. Pictures of the cross section of PPO + graphene membranes are reported in Figure 3. It is apparent that, by increasing the concentration of graphene, the cross section becomes filled with graphenic domains. Those structures are aligned perpendicular to the membrane cross section, thereby increasing the tortuosity of diffusing molecules. The images also show that some voids form at the interface of PPO and graphene. Images with higher magnification were also taken ( Figure 4), and show that the surface of graphene platelets remains similar to the one observed in the neat powder ( Figure 2).

Permeability
In Table 2 and Figure 5, we report the permeability values measured at 35 and 65 °C in membranes with increasing loadings of graphene.

Permeability
In Table 2 and Figure 5, we report the permeability values measured at 35 and 65 • C in membranes with increasing loadings of graphene. At 65 °C, no increase of permeability could be detected after addition of graphene: we believe that, at this temperature, the transition between the regime where the free volume increase prevails and the one in which the tortuosity increase dominates occurs at very small graphene loadings, or the first regime does not occur at all. At higher temperatures, indeed, the mobility of both diffusing gases and polymeric chains is higher, and the diffusion and permeation process relies less on the presence of intrinsic free volume in the solid matrix.  It can be seen that, while the addition of small amounts (below 1 wt %) of both types of graphene enhance the He permeability, by about 10%, increasing the filler loading to higher amounts (5 and 15 wt %) reduces the permeability, for all gases considered. This is because the wide graphene platelets, in this range of concentration, significantly increase the tortuosity of gases diffusive path in the membrane. The permeability of CO 2 seems unaffected by the presence of graphene up to a loading of 1 wt %. The permeability of N 2 , on the other hand, follows a same qualitative behavior as that of helium, increasing by about 20% for graphene loadings lower than 1 wt %, and decreasing for higher values. The permeability enhancement observed could be due to various factors, such as the formation of additional free volume at the interphase between polymer and filler, and the permeation of the gas through graphene layers.
At 65 • C, no increase of permeability could be detected after addition of graphene: we believe that, at this temperature, the transition between the regime where the free volume increase prevails and the one in which the tortuosity increase dominates occurs at very small graphene loadings, or the first regime does not occur at all. At higher temperatures, indeed, the mobility of both diffusing gases and polymeric chains is higher, and the diffusion and permeation process relies less on the presence of intrinsic free volume in the solid matrix.

Selectivity
The ideal selectivity values calculated with Equation (2) at 35 • C are reported in Table 3 and Figure 6. Three types of separations are considered: He/CO 2 , He/N 2 and CO 2 /N 2 . The first separation (He vs. CO 2 ) is favored by high temperatures, i.e., both permeability and selectivity increase with increasing temperature, due to higher activation energy of permeation for He than for CO 2 . The other separations considered, namely He/N 2 and CO 2 /N 2 are, in terms of selectivity, not favored by temperature, as the selectivity of polymers for such couples decreases with temperature, due to the unfavorable difference between the permeation activation energies of the two gases (see Equation (10)).

Selectivity
The ideal selectivity values calculated with Equation (2) at 35 °C are reported in Table 3 and Figure 6. Three types of separations are considered: He/CO2, He/N2 and CO2/N2. The first separation (He vs. CO2) is favored by high temperatures, i.e., both permeability and selectivity increase with increasing temperature, due to higher activation energy of permeation for He than for CO2. The other separations considered, namely He/N2 and CO2/N2 are, in terms of selectivity, not favored by temperature, as the selectivity of polymers for such couples decreases with temperature, due to the unfavorable difference between the permeation activation energies of the two gases (see Equation (10)).   The addition of graphene in all proportions produces an increase of He/CO 2 selectivity of the polymer at 35 • C, of about 10%. For the He/N 2 separation, the selectivity values decrease monotonically with increasing graphene content at 35 • C, going from 26.0 to 21.1 at the largest loading. For the CO 2 /N 2 mixture, the selectivity also decreases monotonically with increasing graphene loading, from 20.3 to 15.0. Such trends are due to the gas-dependent variations of permeability induced by addition of graphene. The gas that experiences the highest increase of permeability after addition of small loadings of graphene is the largest one, namely nitrogen. For samples containing larger amounts of graphene, the smallest reduction of permeability recorded is again that of nitrogen. Thus, both He/N 2 and CO 2 /N 2 selectivity decrease. Such trend is attributable mostly to the kinetic term of permeability, the diffusivity, which is strongly affected by molecule size, as will be seen in the following. When the He/CO 2 selectivity is considered, on the other hand, the size of the molecule plays a less important role, as CO 2 permeability is also strongly affected by solubility, which depends on the molecule condensability rather than on its size. Therefore, for this couple of gases, the size-sieving capacity of the membrane slightly increases with the addition of graphene, rather than decreasing. When the temperature is increased to 65 • C, the situation slightly changes, in particular for the case of the He/N 2 mixture, for which graphene addition positively affects the selectivity. The incorporation of filler is also beneficial for the He/CO 2 separation, similar to the lower temperature case. The CO 2 /N 2 selectivity, on the other hand, decreases with graphene addition also at 65 • C. The different trends of selectivity at different temperatures are due to the effect that graphene has on the thermal dependence of the gas permeability in the polymer, which we will discuss below.

Diffusivity
The diffusivity is a kinetic quantity that contributes to the permeability, according to Equation (4). In particular, diffusivity is strongly affected by variations of the polymer free volume, and by the chain packing of the membrane. Therefore, it is expected to be the quantity that is most strongly affected by the introduction of filler with a peculiar morphology, such as the high aspect ratio graphene platelets with one nanometric dimension. Furthermore, in size-selective polymers such as the one that we are considering, diffusivity dictates the selective behavior, rather than solubility, as happens for instance in the case of higher free volume glassy polymers such as PTMSP and PIM-1. In Table 4, we report the variations observed for the diffusion coefficients as a function of graphene loading. It can be seen that, when high filler loadings are considered, a marked decrease of diffusivity is measured, which is because the graphenic nanoplatelets act as physical barriers and increase tortuosity. At lower loadings, the behavior shows a large scattering. The diffusivity data obtained at 35 and 65 • C are also reported in Figure 7a,b.  It is often seen, in mixed matrix membranes comprising nanometric fillers, that the variations of permeability induced by filler are gas size-dependent [15,16]. Usually, porous fillers with a size selective ability increase more effectively the permeability towards smaller gases, while the opposite can be observed in the case of nanoscopic impermeable fillers such as fumed silica [15,16]. In our previous work, we have found that the addition of graphene reduces more strongly the permeation of small gases, such as Helium, and to a lower extent the permeation of larger molecules, such as Nitrogen [35]. In this work, we observe a similar trend: we reported diffusivity variations after graphene addition versus the kinetic diameter of penetrating molecule in Figure 7c. We see values of D/D0 close to 1 for the larger molecule, namely N2, and much smaller than unity for the smallest molecule, Helium.
To effectively correlate the observed variations in permeability to variations of diffusivity, we plotted a parity curve, reporting the observed variations in permeability after addition of graphene, It is often seen, in mixed matrix membranes comprising nanometric fillers, that the variations of permeability induced by filler are gas size-dependent [15,16]. Usually, porous fillers with a size selective ability increase more effectively the permeability towards smaller gases, while the opposite can be observed in the case of nanoscopic impermeable fillers such as fumed silica [15,16]. In our previous work, we have found that the addition of graphene reduces more strongly the permeation of small gases, such as Helium, and to a lower extent the permeation of larger molecules, such as Nitrogen [35]. In this work, we observe a similar trend: we reported diffusivity variations after graphene addition versus the kinetic diameter of penetrating molecule in Figure 7c. We see values of D/D 0 close to 1 for the larger molecule, namely N 2 , and much smaller than unity for the smallest molecule, Helium.
To effectively correlate the observed variations in permeability to variations of diffusivity, we plotted a parity curve, reporting the observed variations in permeability after addition of graphene, P/P 0 , to the ones observed for diffusivity, D/D 0 , in Figure 8. It can be seen that, while for CO 2 and, less markedly, for N 2 , the permeability trend is strongly associated to the diffusivity one, confirming that diffusion plays a strong role in such membranes, for He, the behavior is different. For this gas, the permeability of composite membranes is reduced less than the diffusivity by addition of graphene. Although the error associated to the evaluation of the diffusivity from the time lag measurement is larger in the case of Helium, which can cause some significant random scattering, such trend is systematic, i.e., all the data fall above the bisector. If the deviations were due merely to scattering associated to experimental error, they should have lied both above and below the bisector. According to the solution-diffusion model expressed by Equation (4), one should have: so that, if P/P 0 = D/D 0 , as it approximately happens in the case of N 2 and CO 2 , one concludes that S/S 0 = 1, i.e., that the solubility of such gases is not affected by the presence of graphene. On the other hand, the same equation applied to the case of helium indicates that S/S 0 > 1, i.e., that graphene enhances the solubility of helium in the membrane. Although the solubility of helium in solids is generally small in absolute value, we are considering relative variations here, which can be significant. Furthermore, the helium molecule, due to its small size, could be adsorbed onto the graphene layers interface but also between the layers of graphene platelets. This solubility behavior counterbalances the strong reduction of diffusivity observed in the case of helium. P/P0, to the ones observed for diffusivity, D/D0, in Figure 8. It can be seen that, while for CO2 and, less markedly, for N2, the permeability trend is strongly associated to the diffusivity one, confirming that diffusion plays a strong role in such membranes, for He, the behavior is different. For this gas, the permeability of composite membranes is reduced less than the diffusivity by addition of graphene. Although the error associated to the evaluation of the diffusivity from the time lag measurement is larger in the case of Helium, which can cause some significant random scattering, such trend is systematic, i.e., all the data fall above the bisector. If the deviations were due merely to scattering associated to experimental error, they should have lied both above and below the bisector. According to the solution-diffusion model expressed by Equation (4), one should have: (11) so that, if P/P0 = D/D0, as it approximately happens in the case of N2 and CO2, one concludes that S/S0 = 1, i.e., that the solubility of such gases is not affected by the presence of graphene. On the other hand, the same equation applied to the case of helium indicates that S/S0 > 1, i.e., that graphene enhances the solubility of helium in the membrane. Although the solubility of helium in solids is generally small in absolute value, we are considering relative variations here, which can be significant. Furthermore, the helium molecule, due to its small size, could be adsorbed onto the graphene layers interface but also between the layers of graphene platelets. This solubility behavior counterbalances the strong reduction of diffusivity observed in the case of helium.

Analysis of Temperature Effect
It is very interesting to analyze the effect of temperature on those composite membranes. Normally, the transport properties of polymers, especially the permeability, are strongly dependent on temperature, and, to our knowledge, this is the first analysis of the temperature effect on transport properties of polymer/graphene composites for separation. Figure 9a reports the temperature effect on permeability, as a function of graphene loading, and explains the observed trend of selectivity with temperature. It is clear from the graph that the role of graphene is to reduce strongly the positive dependence of gas permeability on temperature, and even cause an inversion of trend, in the case of CO2. These data appear for the first time and indicate that graphene nanoplatelets hinder the polymer mobility and flexibility. Indeed, it is known that the strong increase of gas diffusivity and permeability observed in polymers at increasing temperature

Analysis of Temperature Effect
It is very interesting to analyze the effect of temperature on those composite membranes. Normally, the transport properties of polymers, especially the permeability, are strongly dependent on temperature, and, to our knowledge, this is the first analysis of the temperature effect on transport properties of polymer/graphene composites for separation. Figure 9a reports the temperature effect on permeability, as a function of graphene loading, and explains the observed trend of selectivity with temperature. It is clear from the graph that the role of graphene is to reduce strongly the positive dependence of gas permeability on temperature, and even cause an inversion of trend, in the case of CO 2 . These data appear for the first time and indicate that graphene nanoplatelets hinder the polymer mobility and flexibility. Indeed, it is known that the strong increase of gas diffusivity and permeability observed in polymers at increasing temperature is mostly due to the thermally-enhanced polymer chain flexibility and mobility, which makes the diffusive jumps of gas molecules more frequent. When increasing amounts of graphene are added, up to 5 wt %, to PPO, the thermal activation of diffusivity and permeability are strongly inhibited. In particular, adding 5 wt % of graphene to PPO reduces the relative increase of permeability (corresponding to a 30 • C increase) from 0.7 to 0.2 for N 2 , and from 0.47 to 0.36 for He. The relative dependence of CO 2 permeability in pure PPO is much weaker, for the reasons discussed above, and numerically is expressed by a value of +0.13, which becomes −0.17 when the sample contains 5 wt % graphene. At graphene loadings higher than 5 wt %, the effect discussed above is partially lost, possibly because the graphene nanoplatelets start to form aggregates and lose part of their ability to immobilize the polymer.
Polymers 2018, 10, x FOR PEER REVIEW 15 of 19 is mostly due to the thermally-enhanced polymer chain flexibility and mobility, which makes the diffusive jumps of gas molecules more frequent. When increasing amounts of graphene are added, up to 5 wt %, to PPO, the thermal activation of diffusivity and permeability are strongly inhibited. In particular, adding 5 wt % of graphene to PPO reduces the relative increase of permeability (corresponding to a 30 °C increase) from 0.7 to 0.2 for N2, and from 0.47 to 0.36 for He. The relative dependence of CO2 permeability in pure PPO is much weaker, for the reasons discussed above, and numerically is expressed by a value of +0.13, which becomes −0.17 when the sample contains 5 wt % graphene. At graphene loadings higher than 5 wt %, the effect discussed above is partially lost, possibly because the graphene nanoplatelets start to form aggregates and lose part of their ability to immobilize the polymer. In Figure 9b, we reported the relative selectivity variation in the temperature range inspected. The behavior is extremely clear and indicates that, for all the gases considered, the addition of even small amounts of graphene affects positively the selectivity, in all types of separation. For He/CO2 selectivity, favored by temperature, graphene further enhances this trend. For CO2/N2 separations, unfavored by temperature, the negative effect of a temperature increase seems mitigated by graphene addition. For He/N2 the selectivity starts to increase, rather than decrease, with temperature, after graphene addition. Such trends are a natural consequence of the permeability trend previously discussed, and indicate that graphene can have a marked, and beneficial, effect on the polymer mobility, which reduces the negative effects of temperature on selectivity. Indeed, at higher temperatures, the polymer becomes more flexible and loses part of its size selectivity. The addition of graphene in small amounts seems to mitigate such trend, due to the peculiar, thin and long, shape of graphene platelets which hinder polymer mobility and the loss of discriminating ability. Such effect is totally consistent with what observed when adding graphene to other glassy polymers prone to ageing: it was shown that graphene reduces ageing via the same mechanism, i.e., by acting as a physical constraint, or stabilizer, of the polymer matrix [35].

Comparison with Other Polymers
The permeability variations observed with graphene have comparable order of magnitude as the ones obtained by adding 1 wt % of graphene into other glassy polymers, namely PTMSP [35] and PIM-1 [33]. In the case of PTMSP/graphene, the nominal amount of graphene in the composite In Figure 9b, we reported the relative selectivity variation in the temperature range inspected. The behavior is extremely clear and indicates that, for all the gases considered, the addition of even small amounts of graphene affects positively the selectivity, in all types of separation. For He/CO 2 selectivity, favored by temperature, graphene further enhances this trend. For CO 2 /N 2 separations, unfavored by temperature, the negative effect of a temperature increase seems mitigated by graphene addition. For He/N 2 the selectivity starts to increase, rather than decrease, with temperature, after graphene addition. Such trends are a natural consequence of the permeability trend previously discussed, and indicate that graphene can have a marked, and beneficial, effect on the polymer mobility, which reduces the negative effects of temperature on selectivity. Indeed, at higher temperatures, the polymer becomes more flexible and loses part of its size selectivity. The addition of graphene in small amounts seems to mitigate such trend, due to the peculiar, thin and long, shape of graphene platelets which hinder polymer mobility and the loss of discriminating ability. Such effect is totally consistent with what observed when adding graphene to other glassy polymers prone to ageing: it was shown that graphene reduces ageing via the same mechanism, i.e., by acting as a physical constraint, or stabilizer, of the polymer matrix [35].

Comparison with Other Polymers
The permeability variations observed with graphene have comparable order of magnitude as the ones obtained by adding 1 wt % of graphene into other glassy polymers, namely PTMSP [35] and PIM-1 [33]. In the case of PTMSP/graphene, the nominal amount of graphene in the composite membrane was 1 wt %, and the procedure used to fabricate the membrane was exactly the same used for PPO in this work, as well as the initial features of the graphene added [35]. In the case of PIM-1, the graphene content was varied between 0.1% and 2.43%, and the graphene was exfoliated in situ from graphite in the polymer solution, after a long sonication of 84 h [33]. Due to this process, it is expected that the graphene particles in such case have a smaller aspect ratio than the ones used in this work. The data of permeability variation in the various polymers inspected, after addition of 1 wt % of graphene in the case of PPO and PTMSP, and of 0.71 wt %, in the case of PIM-1, are reported in Figure 10a. Data were obtained at different temperatures in the range 25-35 • C. It can be noticed that the order of magnitude of permeability variation is the same in all polymers inspected, and the trend is increasing with increasing kinetic dimeter, indicating that the permeability of larger gases is enhanced more (or reduced less) than that of smaller gases. This could be because, for large gases, even a small adjustment of the internal free volume associated to graphene addition can make a big difference in the permeability value. It must be noted that both PIM-1 and PTMSP have much higher permeability values than PPO, and that PTMSP has a higher free volume than PPO. Plus, both such polymers, as far as the He/CO 2 separation is concerned, are CO 2 -selective, rather than He-selective: this is due to their high free volume, and to a strong impact of the solubility on the separation. In Figure 10b,c, we report, for comparison, the trend of permeability variation for two gases, He and CO 2 , in the PPO of this work, and in PIM-1. We can notice that, in the case of helium, no significant differences between the two polymers are observed. (c) Figure 10. (a) Permeability variation induced by graphene addition, as a function of the gas kinetic diameter, in three glassy polymers: PPO (this work), PTMSP [35] and PIM-1 [33]. The weight fraction of graphene added was 1 wt % in the case of PPO and PTMSP; 0.71 wt % in the case of PIM-1. He permeability variation (b); and CO2 permeability variation (c) after addition of graphene in PPO (this work) and PIM-1 versus graphene loading.

Conclusions
We fabricated mixed matrix membranes based on PPO and increasing amounts of few layer graphene, from 0.3 to 15 wt %, and tested them for the permeability of He, N2 and CO2 at 35 and 65 °C.
In general, the best effect of graphene addition is observed when the loadings are small, below 1%, and this was attributed to the fact that, at high filler loading, the effect of increasing the tortuosity Figure 10. (a) Permeability variation induced by graphene addition, as a function of the gas kinetic diameter, in three glassy polymers: PPO (this work), PTMSP [35] and PIM-1 [33]. The weight fraction of graphene added was 1 wt % in the case of PPO and PTMSP; 0.71 wt % in the case of PIM-1. He permeability variation (b); and CO 2 permeability variation (c) after addition of graphene in PPO (this work) and PIM-1 versus graphene loading.
When it comes to CO 2 (and similar results are noticed for N 2 ), that is a slightly larger molecule, the situation changes strongly: while in PPO the CO 2 permeability is not much affected by graphene, in the loading range inspected, marked increases of permeability are observed at very low loadings (0.1 wt %) in PIM-1. One can say that such difference can be due to the, probably, smaller aspect ratio of the graphene used in the paper by Althumayri et al., [33]. Furthermore, a significant difference may also arise from the fact that PPO is a size selective polymer, with He/CO 2 selectivity higher than 1, while PIM-1 has a He/CO 2 selectivity much smaller than unity.
The comparison with other polymers case indicates that, as far as graphene addition to gas separation polymers is concerned, it seems that is advisable to work in the low loading range to retain the beneficial effects. The quantitative enhancement obtained are still limited, but there is large room for improvement by working on the morphological aspects, i.e., the aspect ratio of graphene, which could be optimized by varying the preparation method. Furthermore, the effect of a surface chemistry modification of graphene platelets on the final composite properties has not been studied yet: chemistry affects the solubility contribution to permeability, but also the adhesion and morphology of the composite material. Both aspects will be investigated in future works.

Conclusions
We fabricated mixed matrix membranes based on PPO and increasing amounts of few layer graphene, from 0.3 to 15 wt %, and tested them for the permeability of He, N 2 and CO 2 at 35 and 65 • C.
In general, the best effect of graphene addition is observed when the loadings are small, below 1%, and this was attributed to the fact that, at high filler loading, the effect of increasing the tortuosity prevails on the effect of enhancing the polymer chain distribution. Such aspect is in agreement with previous findings on other glassy polymer filled with graphene. The selectivity for He/CO 2 and He/N 2 is also positively affected by addition of graphene, and the effect is observed at both 35 and 65 • C. Furthermore, the graphene addition allows adjusting the thermal dependence of permeability and selectivity, generally improving it, by acting as a physical constraint to the relaxation of polymer chains which compromises size selectivity at high temperatures.
The permeability trend follows closely the diffusivity one, confirming the validity of the solution-diffusion mechanism, and that solubility of gases is less strongly affected than diffusivity by graphene addition, for all gases except helium, which might be adsorbed onto graphene insertions.
These preliminary results suggest that small additions of graphene to PPO can enhance the permselectivity, as it does in other glassy polymers. The effect could be quantitatively improved by optimizing the aspect ratio of particles, and testing chemical modification of graphene.