Minda
Chen
a,
Yong
Han
b,
Tian Wei
Goh
a,
Rong
Sun
ac,
Raghu V.
Maligal-Ganesh
a,
Yuchen
Pei
a,
Chia-Kuang
Tsung
d,
James W.
Evans
*be and
Wenyu
Huang
*ae
aDepartment of Chemistry, Iowa State University, Ames, Iowa 50011, USA. E-mail: whuang@iastate.edu
bDepartment of Physics and Astronomy, Iowa State University, Ames, Iowa 50011, USA. E-mail: evans@ameslab.gov
cDepartment of Chemistry, Beijing Normal University, Beijing, China
dDepartment of Chemistry, Boston College, Chestnut Hill, MA 02467, USA
eDivision of Chemical and Biological Sciences, Ames Laboratory, U.S. Department of Energy, Ames, Iowa 50011, USA
First published on 18th February 2019
The outstanding catalytic activity and chemical selectivity of intermetallic compounds make them excellent candidates for heterogeneous catalysis. However, the kinetics of their formation at the nanoscale is poorly understood or characterized, and precise control of their size, shape and composition during synthesis remains challenging. Here, using well-defined Pt nanoparticles (5 nm and 14 nm) encapsulated in mesoporous silica, we study the transformation kinetics from monometallic Pt to intermetallic PtSn at different temperatures by a series of time-evolution X-ray diffraction studies. Observations indicate an initial transformation stage mediated by Pt surface-controlled intermixing kinetics, followed by a second stage with distinct transformation kinetics corresponding to a Ginstling–Brounstein (G–B) type bulk diffusion mode. Moreover, the activation barrier for both surface intermixing and diffusion stages is obtained through the development of appropriate kinetic models for the analysis of experimental data. Our density-functional-theory (DFT) calculations provide further insights into the atomistic-level processes and associated energetics underlying surface-controlled intermixing.
In heterogeneous catalysis, as well as many other fields, it is generally desirable to synthesize IMC nanoparticles (NPs) with a small size, high mono-dispersity, and a single-phase nature.20,21 However, this remains challenging despite numerous efforts made to engineer the synthesis routes of IMCs.22,23 For example, poor control over the NPs often results in either mixed phases,24,25 heterogeneous surfaces,26 or severe aggregation.12 In addition, the underlying kinetics and thermodynamics of IMC formation at the nanoscale have been rarely discussed, although an in-depth understanding could lead to profound improvements in designing the synthesis routes for these IMC NPs.27 Many methods for synthesizing IMCs involve transformation from monometallic to bimetallic material, including successive impregnation, chemical vapor deposition, metallurgical alloying, and the seeded growth method.10,28 The seeded growth method is especially useful as it often retains the structure, shape and dispersity of the parent monometallic material.14,29,30
Colloidal synthesis of PtSn IMCs has been achieved and is widely used in heterogeneous catalysis. Komatsu et al. used PtSn supported on a zeolite as an effective catalyst for the dehydroisomerization of butane into isobutene.31 Eichhorn et al. developed PtSn as an efficient CO-tolerant electrocatalyst for H2 oxidation.32 Our group previously reported the use of PtSn for the selective hydrogenation of furfural to obtain furfural alcohol.14 Recently, PtSn NPs were also reported to improve pairwise selectivity in the hydrogenation of the CC bond for parahydrogen-induced hyperpolarization NMR.33 However, there are no systematic studies yet on the temperature and size dependence of the detailed formation kinetics of PtSn at the nanoscale. Suitably tailored kinetic modeling would enable reliable extraction of key activation barriers, which could ideally be compared with ab initio density-functional-theory (DFT) analysis. We note one recent study34 considered PtSn NP formation for a single size of Pt seed and under temperature ramping, wherein a generic bulk Avrami model was used to fit kinetics. In this work, we report our study on phase transformation from Pt to PtSn IMCs, using 5 and 14 nm Pt NPs that are encapsulated in mesoporous silica (Pt@mSiO2). The mSiO2 shell not only prevents aggregation of the NPs, but also allows uninhibited access to the surrounding chemical environment. Specifically, it allows the addition of the Sn precursor to form encapsulated PtSn IMCs (PtSn@mSiO2). Previous studies extensively investigated the role of mesoporous silica during the formation of intermetallic NPs, in particular revealing that etching of its inner surface by locally concentrated inorganic ions, in this case H+ and Cl−, creates space for uninhibited NP growth.14,35 The Pt to PtSn transformation was performed under various fixed temperatures and monitored by powder X-ray diffraction (PXRD). Analysis of the deconvoluted PXRD patterns suggests that the transformation involves two distinct stages: a surface-controlled intermixing step followed by a Ginstling–Brounstein (G–B) type solid-state diffusion-controlled step. We crafted appropriate kinetic models for these two stages in order to assess the relevant activation barriers. Further DFT analysis relevant to the surface-controlled intermixing stage provides additional insight into the underlying atomistic diffusion processes and the relevant energetics and pathways.
To monitor the transformation from Pt to the ordered PtSn phase, a time-evolution PXRD study was performed by quenching the reaction mixture at various times during the transformation. Fig. 2 shows the corresponding PXRD pattern of 5 and 14 nm NPs, where the various time periods of the conversion at 250 °C are indicated. PXRD patterns obtained at other temperatures are shown in Fig. S5.† Pt(111), Pt(200), PtSn(102), and PtSn(110) diffraction peaks were observed in the patterns. The Pt phase gradually diminished and PtSn phase grew, as suggested by the change in integrated peak intensity. In addition, the rate of transformation was apparently faster at the beginning of the process and gradually slowed down within the two hours. Despite the existence of other phases in the Pt–Sn phase diagram (Pt3Sn, Pt2Sn3, PtSn2, PtSn4), we did not observe their formation by PXRD. To identify the relative amount of each phase, a standard mixture of Pt and pure-phase PtSn in 1:
1 mole ratio was quantified by ICP-MS and its PXRD pattern was deconvoluted into four subpeaks (Fig. 3) within the region of 34 to 52°, and the integrated intensity (A) ratio of the two most intense peaks, Pt(111) and PtSn(102), was calculated to be 1.38. It is then used as the ratio of the response factor (RF) between Pt and the PtSn phase. Standard mixtures of Pt
:
PtSn = 1
:
3 and 3
:
1 are also measured, as shown in Fig. S6,† and the calculated RF ratios match well with the 1
:
1 case. Subsequently, the amount of Pt
:
PtSn in each sample was calculated as:
![]() | (1) |
![]() | (2) |
Fig. 4 shows α versus reaction time t at different temperatures. As observed in Fig. 4, the transformation occurs rapidly at the beginning of the process and gradually slows down for both sizes of Pt NPs at all temperatures. Comparing 14 nm Pt@mSiO2 with 5 nm Pt@mSiO2, the smaller particle initially converts slightly slower at a fixed temperature but ultimately reaches a higher conversion after 120 min. At least the latter is expected given the smaller particle size.
Since our Pt NPs are encapsulated inside mSiO2, we designed two control experiments to explore the effect, if any, of the mSiO2 on the growth of intermetallic PtSn. We first employed 5 nm Pt nanoparticles deposited onto 100 nm silica spheres (noted as Pt/SiO2) following a reported method.36 We observed similar kinetics for the conversion from Pt/SiO2 to PtSn/SiO2 as the encapsulated Pt@mSiO2 (Fig. S7†). A TEM image of the nanoparticles after conversion is shown in Fig. S8.† Next, we also tested the unsupported Pt NPs capped by myristyltrimethylammonium bromide (TTAB), but severe aggregation of the Pt NPs was observed before reaching the desired temperature (Fig. S9†). This emphasizes that the mSiO2 shell is critical to protect the NPs from aggregation. These two control experiments demonstrate that the major function of the mesoporous silica shell is to prevent the aggregation of Pt NPs, and the shell does not affect the transformation of the encapsulated Pt core to intermetallic PtSn.
It is however perhaps instructive to perform a benchmark analysis of our data with a simple fixed reaction-order model.38 Such a model for the kinetics assumes the form
![]() | (3) |
Given the above assessment of at least two distinct stages for the conversion of 14 nm Pt NPs, we have refined our modeling to incorporate this feature. We describe the first stage as a “surface stage” that involves the reduction of the Sn precursor and formation of PtSn starting at the surface of the Pt NPs, anticipating that a complete intermetallic shell around the periphery of the NP has not yet formed. Furthermore, we anticipate that the kinetics will reflect the feature that a portion of the surface of the original Pt NPs remains unconverted which can facilitate reduction and intermixing of Sn. To show that the Pt surface facilitates Sn reduction, we have performed a control experiment where SnCl2 is dissolved in TEG and heated to 280 °C without the existence of Pt, and we observed no reduction of Sn2+ to form metallic Sn. On the other hand, we describe the second stage after which a complete intermetallic shell has formed as a “Ginstling–Brounshtein (G–B) diffusion stage”, and will analyze associated behavior with a suitably adapted G–B type diffusion model.40 A schematic picture of the structural evolution of NPs through these two stages is shown in Fig. 5, where further discussion is provided in the following subsections.
![]() | ||
Fig. 5 Schematic structural evolution of NPs during complete conversion to IMC. Data shown in Fig. 4 does not correspond to complete conversion, but such extended data are shown in the ESI (Fig. S16B†). |
![]() | (4) |
![]() | (5) |
First, we analyze the data for the smaller 5 nm Pt NPs, where it is anticipated that the observed behavior will correspond mainly or entirely to the first surface stage. Least-squares fitting by adjusting all ms, αsurf, and Ks as free parameters produces ms = 1.78, 1.79, and 1.75 with αsurf = 0.79, 0.57, and 0.30 for T = 250, 240, and 230 °C, respectively. The fit for associated activation energy gives Ea,s ≈ 245 kJ mol−1. A key observation indicating the veracity of the model is the similar values of ms obtained for different T, noting that behavior in the surface stage is expected to be described by a single reaction order. From this perspective, it is natural to refit the data by imposing a single value of ms = 1.75, and adjusting just αsurf and Ks as free parameters. This analysis produces αsurf ≈ 0.788, 0.566, and 0.297 for T = 250, 240, and 230 °C, respectively, but preserves Ea,s = 245 kJ mol−1. These fits are shown in Fig. 4A, and the Arrhenius analysis is shown in Fig. 6A.
![]() | ||
Fig. 6 Arrhenius plot for the surface stage of: (A) 5 nm Pt NPs with ms = 1.75; (B) 14 nm Pt NPs with ms = 1.63. |
Second, we analyzed behavior in the first surface stage for 14 nm Pt NPs. Here, we first make a reasonable selection of the values of αsurf based on the form of the conversion curves, and then assess the associated reaction order and activation energy. Choosing αsurf = 0.81, 0.58, and 0.51, for T = 260, 250, and 240 °C, respectively, yields ms = 1.64, 1.62, and 1.64. From these choices of αsurf and ms, one obtains Ea,s ≈ 225 kJ mol−1. The similarity of the values of ms for different T, again, supports the veracity of the modeling. Furthermore, just as for the 5 nm Pt NPs, it is natural to reanalyze the data imposing a single value of ms = 1.63 which yields only a negligible modification in the values of αsurf = 0.795, 0.574, and 0.506, for T = 260, 250, and 240 °C, respectively, but preserves Ea,s ≈ 225 kJ mol−1. These fits are shown in Fig. 4B (blue curve), and the corresponding Arrhenius analysis is shown in Fig. 6B. Note that the above fitting is based on only 5, 4, and 4 data points for 260, 250, and 240 °C, respectively.
An additional expectation for a reliable model is that ms for 5 and 14 nm Pt NPs should not be greatly different. Thus, our determination of ms = 1.75 for the 5 nm Pt NPs versus ms = 1.63 for 14 nm Pt NPs seems reasonable. Similarly, one should expect somewhat similar values of Ea,s for the two different particle sizes, a feature which is produced in our modeling: Ea,s ≈ 245 kJ mol−1 for 5 nm Pt NPs versus 225 kJ mol−1 for 14 nm Pt NPs.
However, we should note that, perhaps unexpectedly, the initial rate of conversion (corresponding to Ks) is higher for larger Pt NPs than for smaller NPs (when the comparison is made at the same temperature). We interpret this trend as related to the feature that the larger NPs are much more likely to include grain boundaries. This is confirmed by additional PXRD analysis using the Scherrer equation, which consistently gives a grain size of around 11 nm for the larger Pt NPs, while TEM indicates a particle size of 14 nm. The existence of grain boundaries in the 14 nm Pt@mSiO2 is also confirmed by high resolution TEM (HR-TEM) (Fig. S11†). We anticipate that the diversity of “defect sites” along the grain boundary near the surface of the Pt NPs will facilitate the initiation of conversion to the IMC.41,42 Indeed, our DFT analysis discussed below indicates significant inhibition to the initiation of conversion on perfect Pt(100) facets.
Further characterization of the elemental distribution, especially the location of Sn during the transformation is provided by energy-dispersive X-ray spectroscopy (EDX) analysis. First, an EDX line scan analysis corresponding to the first stage (Fig. 7A) provides no clear indication of a reduced Sn shell surrounding the Pt core. It suggests that intermixing of Sn is more facile than Sn reduction, and therefore reduction would be rate limiting. In this case, Ea,s should reflect an activation barrier for reduction. However, it is possible that a thin shell of Sn has formed around the Pt NPs which is not distinguishable in the EDX analysis. In this scenario, Ea,s could be associated with intermixing. Second, a line scan corresponding to the second stage, as seen in Fig. 7B, indicates that, at 260 °C, a complete Sn shell has formed on the 14 nm NPs with an α value around 0.8. This observation implies that αsurf should not be above this value, which is consistent with the above analysis that αsurf = 0.795 at 260 °C. Additional data are available in the ESI (Fig. S12 and S13†) to show that the EDX line scan results are representative of the sample. The structure of the 14 nm NPs after a 30 min conversion at 260 °C was obtained from HR-TEM (Fig. 7C) and it shows a core of the Pt domain with (200) lattice fringes and the shell of the PtSn domain with (101) lattice fringes, which correspond to the beginning of the G–B type diffusion stage for the 14 nm NPs. This observation is consistent with our schematic picture of the NP evolution shown in Fig. 5.
Naturally, the traditional G–B model does not apply to the surface stage where there is no complete intermetallic shell. Furthermore, just at the end of the surface stage where a complete shell is first formed, this shell will not be uniform. At this point, the Pt NP surface intermetallic shell has just formed at some locations on the NP surface, so the shell has negligible thickness. In contrast, at a different location where the intermetallic shell has formed much earlier, the shell will have a significant thickness. However, the subsequent diffusion-mediated growth of this nonuniform shell should quickly lead to a more uniform shell thickness, as indicated in Fig. 5. Such growth of the interface between the intermetallic shell and the unconverted Pt core is characterized by so-called anti-DLA shape stabilization,43–45 as thinner portions grow faster and thicker portions grow slower due to the diffusion-mediated nature of growth. Thus, after some interval time at the end of the surface stage, it is reasonable to apply a G–B model. Therefore, the standard G–B formula should be refined so that the shell thickness has an appropriate non-zero thickness (in contrast to the default assumption of zero thickness) when the model is first applied. Such refinement (see the ESI Fig. S14 and S15†) leads to the result
![]() | (6) |
![]() | (7) |
![]() | (8) |
Finally, we believe that there are two reasons that, perhaps counterintuitively, the G–B diffusion stage is not observed during 120 min when we track the transformation kinetics for 5 nm Pt NPs, while it is observed when tracking 14 nm Pt NPs over the same time interval. First and most significantly, the initial rate of conversion is lower for the smaller 5 nm NPs, as discussed above, indicating that more time is required to reach the second stage. In addition, values of αsurf are higher for 5 than 14 nm NPs at the same temperature, further delaying the onset of the second stage for the former.
To illustrate that our model convincingly describes the transformation, we performed control experiments wherein 5 and 14 nm Pt@mSiO2 are converted to PtSn at 250 °C and 260 °C, respectively, with the reaction time extended but all other parameters following previous experiments. For the 5 nm Pt@mSiO2, the experimental conversion (Fig. S16A†) initially matches well with previous experimental results, but ultimately exceeded the surface intermixing stage limit αsurf. This confirms that for 5 nm NP, a G–B diffusion stage does occur later in the transformation. For the 14 nm Pt@mSiO2, we were able to predict before the experiment the conversion α at a given time, and the time necessary (around 930 min) for the transformation to go to completion, based on out fitting results. Experimental results (Fig. S16B†) matched well with both previous experimental result and our prediction.
Assessment of the energetics relevant for intermixing is provided by DFT analysis using the Vienna ab initio simulation package (VASP).46,47 We used the projector-augmented-wave method48 for the electron-core interactions. The analyses were performed with both the Perdew–Burke–Ernzerhof (PBE)49 and PBEsol50 generalized gradient approximation (GGA) functionals for exchange and correlation. The energy cutoff of 400 eV for the plane-wave basis was tested to be sufficient for energy convergence. For accuracy in energy minimization, the magnitude of the force acting on each atom undergoing relaxation was reduced to less than 0.1 eV nm−1. The selection of a supercell size and the corresponding Γ-centered k meshes were always tested carefully for energy convergence. We have performed benchmark calculations for bulk fcc Pt as well as both α- and β-Sn to ensure that experimentally determined lattice constants and cohesive energies are recovered. See Tables S17–S22 in the ESI† for more details. In addition, we confirm that DFT analysis for bulk Pt3Sn and PtSn alloys recovers experimental lattice constants and assures the thermodynamic stability of these alloys at T = 0 K.
First, we assess relevant hopping barriers using a 4 × 4 × 4 supercell with the k mesh of 5 × 5 × 5. For pure Pt containing a single vacancy, we found a barrier of Edv = 1.23 (1.41) eV from PBE (PBEsol) GGA for Pt hopping into the vacancy. These values should be compared with a previous theoretical estimate of Edv = 1.43 eV from the local-density approximation (LDA).41 For Pt containing an embedded Sn at a fcc site and a vacancy at an adjacent fcc site, we find a substantially lower barrier of Edv(Sn) = 0.68 (0.77) eV from PBE (PBEsol) GGA for Sn hopping into the vacancy site. Consequently, for hopping dynamics associated with Sn impurity diffusion, the rate-limiting process is the vacancy diffusion to a site neighboring the embedded Sn, rather than the Sn hopping to that neighboring vacancy location. For this reason, the first set of results for vacancy diffusion in pure Pt is relevant for the analysis of the effective barrier of Sn impurity diffusion.
Second, we consider relevant vacancy formation energies. To determine the vacancy formation energy Eform in a pure fcc metal, the standard procedure is as follows. One evaluates the total energy Evac for a periodic supercell of N fcc sites where N − 1 sites are populated by metal atoms, and one is a vacancy. If Ebulk denotes the total energy of the same cell populated with N metal atoms, then one has that Eform = Evac − (N − 1) Ebulk/N.41 Using this approach for fcc Pt, we obtain Eform = 0.65(0.84) eV from PBE (PBEsol) GGA. These results can be compared with previous theoretical estimates of Eform = 0.95 eV from LDA, and Eform = 0.68 eV from PBE. It is however well recognized that DFT energetics suffer from a “surface intrinsic error” for which correction procedures have been developed.52 These yields corrected estimates of Eform(corr) = 1.15 eV from LDA, and Eform(corr) = 1.18 eV for PBE.41 There are experimental estimates for Eform in Pt with a range of values from 1.25 to 1.6 eV, so the corrected estimates are on the lower end of this range.41
To supplement the above conventional analysis of the formation energy, we also determine the energy cost, Eform(Sn in Pt), for the creation of a vacancy adjacent to a single Sn impurity embedded in fcc Pt.53 To this end, we calculate the total energy Evac(Sn in Pt) of a supercell of N fcc sites with N − 2 sites occupied by Pt, one site occupied by Sn, and one vacancy adjacent to the Sn, as well as the total energy Ebulk(Sn in Pt) of a supercell of N fcc sites with N − 1 sites occupied by Pt and one site occupied by Sn. Then together with Ebulk for pure Pt mentioned above, one has that Eform(Sn in Pt) = Evac(Sn in Pt) + Ebulk/N − Ebulk(Sn in Pt). From such an analysis, we find that Eform(Sn in Pt) = 0.44 (0.62) eV for PBE (PBEsol) GGA. These formation energies are significantly below the uncorrected values for vacancy formation in pure Pt, so we reasonably assume that corrected values are also lower. This analysis indicates that the relevant formation energy controlling Sn impurity diffusion in Pt is the higher value for pure Pt of Eform(corr) ≈ 1.2 eV.
In summary, we conclude that the effective barrier, Ed(Sn in Pt), for the diffusion of Sn impurities in Pt corresponds to the effective barrier or diffusion of vacancies in pure Pt. However, this conclusion is only possible after the above comprehensive analysis. This effective barrier is given by Ed(Sn in Pt) = Eform(corr) + Edv ≈ 2.6 eV (or 250 kJ mol−1) using the higher PBEsol or LDA value for Edv, and is consistent with experimental estimates. A value of Ed(Sn in Pt) ≈ 250 kJ mol−1 is compatible with our kinetic analysis of experimental data determining the effective barrier for the surface stage of intermixing.
From slab calculations using a 4 × 4 supercell with the k mesh of 7 × 7 × 1 (slab thickness is four single-atom Pt(100) layers plus a vacuum separation of 2.2 nm in the direction perpendicular to the slab surface), we determine that Sn prefers the four-fold-hollow (4fh) site on Pt(100), the energy at the bridge site being 0.76 (0.85) eV higher from PBE (PBEsol) GGA. Thus, the diffusion barrier for surface hopping between 4fh sites is given by Ed(Sn) ≈ 0.8 eV. In addition, upon exchange of Sn at a 4fh surface site with one of the four supporting Pt, displacing that Pt to a second nearest-neighbor 4fh site of the initial Sn 4fh site, the energy of the system actually increases by ΔE = + 0.44 (+0.59) eV from PBE (PBEsol) GGA. Thus, this process is endothermic, i.e., not thermodynamically favorable. Previous analyses indicate that it is generally not favorable for more noble (or less cohesive) metals to substitute into less noble (or more cohesive) substrates due to an energy penalty in moving the latter to the surface as adatoms.48 This picture is consistent with our analysis. We find that incorporating additional Sn adatoms which become neighbors of the Pt displaced onto the (100) surface (but are not neighbors of the Sn which substitutes into the surface layer) still does not make the exchange process exothermic.
Thus, we conclude that the onset of intermixing at a Pt(100) surface cannot be described in terms of simple single-atom processes, but must reflect more complex concerted behavior (and is presumably facilitated by the presence of defects associated with grain boundary as suggested previously).
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c8nr10067e |
This journal is © The Royal Society of Chemistry 2019 |