Coalescence dynamics of platinum group metal nanoparticles revealed by liquid-phase transmission electron microscopy

Summary Coalescence, one of the major pathways observed in the growth of nanoparticles, affects the structural diversity of the synthesized nanoparticles in terms of sizes, shapes, and grain boundaries. As coalescence events occur transiently during the growth of nanoparticles and are associated with the interaction between nanoparticles, mechanistic understanding is challenging. The ideal platform to study coalescence events may require real-time tracking of nanoparticle growth trajectories with quantitative analysis for coalescence events. Herein, we track nanoparticle growth trajectories using liquid-cell transmission electron microscopy (LTEM) to investigate the role of coalescence in nanoparticle formation and their morphologies. By evaluating multiple coalescence events for different platinum group metals, we reveal that the surface energy and ligand binding energy determines the rate of the reshaping process and the resulting final morphology of coalesced nanoparticles. The coalescence mechanism, based on direct LTEM observation explains the structures of noble metal nanoparticles that emerge in colloidal synthesis.


INTRODUCTION
Understanding the formation mechanism of nanoparticles is critical for designing controlled synthesis of nanoparticles (Chang et al., 2019;Polte et al., 2010aPolte et al., , 2010bSelvam and Chi, 2011). Conventionally, the formation of colloidal nanoparticles has been understood based on classical crystallization theory characterized by nucleation during the initial period of synthesis and subsequent growth driven by monomer attachment (Kwon and Hyeon, 2011;Talapin et al., 2001). However, in a typical size regime where nanoparticles undergo these processes during the synthesis, their surface-to-volume ratios are significantly high enough to make the surface of nanoparticles reactive (Ivanova and Zamborini, 2010;Jia and Schü th, 2011). Such reactivity provides possibilities for alternative pathways to be involved during monomeric growth (Lee et al., 2016;Wang et al., 2014). Coalescence, which denotes the attachment between two or more particles, is one of these non-classical pathways, while evidence of its mechanism has been observed in experiments using small-angle X-ray scattering (Ingham et al., 2011) and in situ X-ray absorption fine structure spectroscopy (Harada and Kamigaito, 2012). These experiments concurrently indicate that the coalescence of growing nanoparticles occurs along with the monomeric growth process. However, these studies are limited in that they neither identify the exact moments of coalescence events nor elucidate the effect of coalescence on the final structures of synthesized colloidal nanoparticles.
Direct observation of coalescence events at the level of individual nanoparticles was recently accomplished by in situ liquid-cell transmission electron microscopy (TEM) of colloidal nanoparticle growth (Anand et al., 2016;Chen et al., 2020;Yang et al., 2019;Yuk et al., 2012). In this approach, liquid cells encapsulate a specimen in the liquid phase and seal it against the high vacuum of TEM, allowing for real-time measurement with nanometer-scale resolution (Kim et al., 2018). Using this method, coalescence events were observed to frequently occur along with classical growth by monomer attachment, and both pathways arrive at a uniform final size determined by thermodynamic factors in the solution (Zheng et al., 2009). In addition, in situ liquid-cell TEM observations revealed that iron oxyhydride nanoparticles rotate to obtain an energetically preferred lattice plane before coalescence (Li et al., 2012;Liu et al., 2020;Zhu et al., 2018). The effects of ligands and solvent molecules, and the process of neck formation within a nanoparticle pair during coalescence were also examined via liquid-cell TEM (Jin et al., 2018;Lim et al., 2020). The surface-to-volume ratios of nanoparticles are generally high, rendering the achievement of sufficient driving force for coalescence possible. Such events are supposedly ubiquitous in colloidal solutions of nanoparticles (Zheng et al., 2009). The surface energy of nanoparticles, which is determined by the surface-to-volume ratio, van der Waals interactions, and dipole moments can be an important factor in coalescence events because it is associated with the governing factor of nanoparticle interactions. Furthermore, surface ligands used to stabilize the high surface energy of the growing nanoparticles are also relevant to the coalescence event because they can be involved in nanoparticle interactions . The morphology of the coalesced particles is likely to be determined by such factors as well.
Here, using in situ liquid cell TEM, we investigate the coalescence-driven growth of nanoparticles with different metal compositions. The tracking of single nanoparticle trajectories and quantitative analysis of acquired trajectories confirm that the coalescence events and reshaping processes of platinum (Pt) and palladium (Pd) nanoparticles show different kinetics owing to their different surface and ligand binding energies. We also elucidate that different coalescence kinetics results in distinct final morphologies of the synthesized nanoparticles.

RESULTS
Direct observation of coalescence process in platinum and palladium nanoparticle growth using in situ liquid phase transmission electron microscopy We prepare a liquid cell compatible with normal TEM holders (Kim et al., 2017) for the in situ liquid-cell TEM experiment. The liquid cell comprises two 100-mm-thick cell bodies, two 50-nm-thick Si 3 N 4 windows, and a 100-nm-thick spacer ( Figure 1A). A precursor solution prepared by mixing 10 mg of metal precursor and 0.1 mL of oleylamine into 0.9 mL of dichlorobenzene solvent is loaded into the liquid cell. In this work, we compare the growth of two types of metal nanoparticles (Pt and Pd) using Pt(acac)2 and Pd(acac) 2 as precursor solutions for the growth of Pt and Pd nanoparticles, respectively. In both systems, oleylamine works as a surface ligand. The nanoparticle growth is monitored using JEOL 2100 at an acceleration voltage of 200 kV. The in situ TEM images are obtained with a frame rate of two frames per second (Videos S1-S5).
Irradiation of the electron beam in LTEM induces nucleation and growth of Pt or Pd nanoparticles by the reduction of the Pt(acac) 2 or Pd(acac) 2 precursors, respectively, followed by coalescence-mediated growth. The liquid TEM images of the growth of Pt and Pd nanoparticles reveal their different coalescence behaviors ( Figure 1B). The coalesced Pt nanoparticles undergo a fast reshaping process while coalesced Pd nanoparticles retain their snowman-shaped morphology for a prolonged time. To investigate the different growth processes between Pt and Pd nanoparticles, we track nanoparticle size and shape in each frame of the in situ liquid-cell TEM images ( Figures 1C-1J). The size of the nanoparticle is defined as the equivalent diameter (D) calculated from the projected area in the TEM images using the following equation, D = 2 3 ffiffiffiffiffiffiffiffiffi A=p p , where A denotes the projected area of the nanoparticle. The shape of the nanoparticle is quantified by the circularity (C) evaluated by the relationship, C = 4pA=P 2 , where A and P denote the projected area and the perimeter of the nanoparticle, respectively. The average size of both types of nanoparticles continuously increases within 50 s (from 3.2 to 5.1 nm for Pt and from 3.2 to 6.5 nm for Pd) ( Figures 1E  and 1I). During size growth, the circularity of Pt nanoparticles persists at high values, indicating that the coalesced nanoparticles rapidly transform into spherical shapes ( Figure 1F). On the other hand, Pd nanoparticles show a different coalescence mechanism from Pt nanoparticles ( Figures 1G-1J). The magnified TEM snap-shot images of Pd nanoparticles reveal that the attached nanoparticles do not reconstruct into spherical shapes, but rather retain their dumbbell shapes. When multiple particles are attached, they form an irregular worm-like shape. Owing to the irregularity of the coalesced Pd nanoparticles, size distribution is broadened during the growth of nanoparticles ( Figures 1H and 1I). The retention of the worm-like shapes of Pd nanoparticles is confirmed by the low circularity of about 0.5.

Quantitative analysis of coalescence dynamics in platinum nanoparticle growth by individual trajectories
We examine the individual trajectories of the merging nanoparticles to discover the detailed mechanism of the size and circularity change in the nanoparticles induced by the coalescence events. Four representative trajectories of Pt nanoparticle coalescence demonstrate the multiple coalescences in a single-nanoparticle growth trajectory (Figures 2A and 2B). Smaller nanoparticles are frequently attached to other particles which contribute to abrupt growth in size, while monomeric growth of nanoparticles simultaneously occurs as continuous growth. Interestingly, the trajectories and sizes of nanoparticles associated with coalescence ll OPEN ACCESS  Figures 2D and S1). The size narrowing is related to asymmetric coalescence. Small particles are consumed quickly, and the size distribution of particles can thus be narrow ( Figure S2). Most coalescence events occur between 0 and 60 s, while this period overlaps the period during which the size distribution of the nanoparticles decreases ( Figure S3).
We also observe the change in the shape of the nanoparticle during the coalescence and structural relaxation process. The dumbbell or snowman shapes formed after the coalescence can relax into a spherical structure (José -Yacamá n et al., 2005;Wang et al., 2016), and the relaxation process is quantitatively investigated using the circularity change. The circularity of nanoparticles is maintained asymptotically close to 0.9, which value is close to a spherical shape. The value abruptly drops to approximately 0.5 at the initial stage of coalescence, for every coalescence step in multiple coalescence events. After the relaxation process, the circularity recovers to 0.8-0.9 within 10 s, indicating that the coalesced nanoparticles are rapidly reconstructed to spherical shapes (Grammatikopoulos et al., 2019;Hawa and Zachariah, 2006).

Quantitative analysis of coalescence dynamics in palladium nanoparticle growth by individual trajectories
The Pd nanoparticles exhibit different coalescence dynamics from Pt nanoparticles. The in situ TEM images of individual Pd nanoparticles show that the coalesced nanoparticles form a snowman shape by the formation of the narrow neck between the two merging nanoparticles and retain this shape for a long time ( Figures 3A and 3B). Although maintaining this irregular shape, additional nanoparticles are frequently attached to these snowman-shaped nanoparticles ( Figures 3A and 3D). The combination of multiple coalescence and shape retention leads to worm-like shapes in the Pd nanoparticles. The intriguing shape changes are investigated by temporal changes in the circularity of individual Pd nanoparticles. As soon as two Pd nanoparticles are coalesced (red line in Figure 3C), the circularity drops by half. The reduced circularity is not restored to its initial value, but rather remains low, indicating that the shape is not reconstructed to a spherical shape. It is worth noting that the circularity value is less than 0.5 most of the time.
Considering that the circularity of a square, an equilateral triangle, and a five-point star is 0.89 and 0.78, and 0.52, respectively (Olson, 2011), the low circularity means that the Pd nanoparticles have multiple concave parts as dumbbell, snowman, and worm-like shapes do. The relaxation process of the coalesced Pt and Pd nanoparticle is evaluated using the temporal circularity change after the coalescence event ( Figure 4A). We select the coalescence processes among circular nanoparticles of similar sizes before their merging to minimize the influence of size and morphology before coalescence (15 pairs for Pt, 8 pairs for Pd) (Figures 4B and 4C). Figures 4B and 4C show the entire tracked circularity change in multiple Pt and Pd nanoparticles by setting t 0 for the time of surface contact. For Pt nanoparticles, the circularity drops to about 0.65 at the initial stage of coalescence and recovers to about 0.8 during the relaxation process. In contrast, the circularity of Pd nanoparticles drops to about 0.5 and then slowly recovers to about 0.7. To quantitatively investigate the relaxation process after coalescing, we calculate the rate constant (k) of the relaxation process based on the simple model where the deviation of the circularity of the merged particle from the perfect spherical shape (circularity = 1) decays exponentially . The relaxation process is fitted based on the change in the circularity (C) over time. The difference between the C and C f . (circularity when relaxation process ends) decays exponentially, C f À CðDtÞ = Cðt 0 Þ 3 e À kDt , where t is time. By fitting averaged circularity changes for the two cases, the rate constant of coalesced Pd nanoparticles is 0.15 s À1 while that of Pt is 0.31 s À1 (Table S1). The rate constant for merged Pt nanoparticles is two times higher than that of Pd nanoparticles, indicating that the relaxation process of a coalesced Pd nanoparticle happens much slower than for a Pt nanoparticle. The relaxation rate is different possibly because the surface energy of Pt is higher than that of Pd, which is calculated by density functional theory ( Figures 5E and 5F). Owing to the high surface energy of Pt, the Pt nanoparticles prefer to reduce their surface-to-volume ratio, resulting in the relaxation process. On the other hand, Pd nanoparticles with low surface energy require a longer time to be relaxed to a spherical shape. Interestingly, the size ratio between the two approaching nanoparticles does not affect the rate constant of the relaxation process ( Figure S4). In addition, Pt nanoparticles can remain circular despite multiple coalescences. As nanoparticles that do not undergo the coalescence process maintain their circular shape during the observation time, morphological changes occur only during the coalescence process ( Figure S5).
Owing to the differences in the relaxation rates of Pd and Pt nanoparticles based on the surface energy differences, nanoparticles synthesized from the two types of platinum group metal precursors exhibit different morphologies. We conduct the control experiment with the low electron beam dose rate ( Figures S6, S7, and Video S5). Coalescence behaviors of nanoparticles with different dose rates show similar behaviors in terms of the slow relaxation process, worm-like morphology, and low circularity value after coalescence, indicating that coalescence events of nanoparticles observed in LPTEM are not strongly influenced by the dose rate. In the later stages of nanoparticle growth, Pt and Pd nanoparticles exhibit different behaviors in inter-particle interactions. For Pt nanoparticles, instead of coalescence that mainly occurs in the earlier stage, two approaching nanoparticles form a nanoparticle pair without merging events by maintaining a persistent gap between each other ( Figure 5A). On the contrary, Pd nanoparticles still undergo coalescence with approaching nanoparticles (Figure 5B), forming irregular worm-like shaped nanoparticles in the late stage of growth. To quantitatively evaluate this trend, we calculate the radial distribution function (RDF) using the tracked nanoparticle trajectories (Lee et al., 2017;Liu et al., 2020). RDF is defined by g(r) = 1 pNrr 0 P N j = 1 P N i > j ðr À r ij Þ, and represents the density of nanoparticles as a function of distance from a particle. At the initial stage of nanoparticle growth (at 0 to 2 s), the RDF does not show a significant difference between the Pt and Pd nanoparticles, and the most probable peaks are similar between the two systems ( Figure 5C). In the later stage of Pt nanoparticle growth (at 40 to 42 s), the most probable peak is concentrated around 6 nm, which corresponds to the sum of the radii of two nanoparticles and the length of the ligand ( Figure 5D). This result suggests that the approaching Pt nanoparticles do not coalesce anymore but assemble with ligand inter-digitation. The prevention of coalescence for these large-sized Pt nanoparticles that appeared in the later stage of nanoparticle growth can be explained in terms of surface energy and activation energy. After sufficient growth of Pt nanoparticles, the surface energy of the nanoparticle decreases owing to the reduced surface-to-volume ratio, thereby preventing coalescence events. The absence of coalescence events for the large nanoparticles with high curvature can be also explained by high coalescence activation energy, which is mainly attributed to difficult ligand displacement in largesized nanoparticles with large curvature. The high activation energy prevents approaching nanoparticles from coalescing with each other (Grammatikopoulos et al., 2019). However, in the case of Pd nanoparticles ( Figure 5E), the nanoparticles cannot maintain the gap between each other, so the RDF peak disappears in the small radius. As the surface binding energy of the ligand is low, the energy barrier of the coalescence process of the Pd nanoparticle is lower than that of the Pt nanoparticle ( Figure 5F), and nearby particles easily coalesce into the larger nanoparticle. Moreover, the small curvature relative to size, which is attributed to the worm-shaped nature reduces the activation energy of coalescence and allows the coalescence event to occur, even at large sizes. The difference between surface energy and coalescence activation energy of Pt and Pd affects the relaxation process after the coalescence. The unique coalescence dynamics are supported by the previous study for ligand-dependent coalescence . In the literature, the low binding energy of ligands lowers the energy barrier of the coalescence process, which makes it difficult to create a gap between nanoparticles, leading to the synthesis of irregular worm-like shaped nanoparticles.

DISCUSSION
In this study, we systematically investigate coalescence dynamics of platinum group (Pt and Pd) metal nanoparticles using liquid-cell TEM. The analysis of the nanoparticle growth trajectories shows that nanoparticle growth is affected by frequent coalescence events. The coalesced Pt nanoparticles rapidly relax to iScience Article spherical shapes while coalesced Pd nanoparticles retain their snowman shapes owing to the retarded relaxation process. In the late stage of Pd nanoparticle growth, multiple coalescence events and the retention of shape lead to the formation of worm-like shapes. On the other hand, already-grown Pt nanoparticles are assembled with ligand inter-digitation without coalescence process owing to their high activation energies attributed to their large curvature and high surface energy. Our studies suggest that the coalescence process regulates the shapes and structures of the synthesized nanoparticles.

Limitations of the study
This research proposed the role of a binding process in nanoparticle growth mechanism and provides the perspective regarding morphological differences observed in the synthesis of metal nanoparticles with different compositions. This study is based on the real-time tracking of nanoparticle growth by LTEM. However, the measurements with low-resolution LTEM only allow understanding of processes based on morphological changes of nanoparticles. If this study were to be conducted with high-resolution LTEM, kinetics associated with coalescence events could be understood at a level that incorporates information on crystal structures and orientations of interacting nanoparticles. Other factors that may affect the change in surface energy and inter-particle interaction can be considered to extend understanding the underlying mechanism behind nanoparticle coalescence. The coalescence behaviors can be modulated by changing ligand species of nanoparticles  and liquid solvent-ligand interaction that modifies the surface energy. In addition, dipole-dipole interaction (Liu et al., 2020), lattice direction (Li et al., 2012), particle-surface interaction (Lu et al., 2014), van der Waals forces, and liquid film thickness (Kang et al., 2021) affect the particle-particle interaction for the nanoparticle coalescence and relaxation process.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following: d KEY RESOURCES TABLE

DECLARATION OF INTERESTS
iScience Article