Interfacial Polymerization of TEPA and HMDI: The Role of Water

Abstract

The interfacial polymerization (IP) between tetraethylenepentamine (TEPA) and 4,4′methylenebis (cyclohexyl isocyanate) (HMDI) at the inorganic–organic solvent interface happens so fast that it is difficult to control the reaction and the morphology of the polymer. Through quantum mechanical (QM) calculations and molecular dynamics (MD) simulations, we clarify the catalytic effect of water on the rapid IP process in the organic phase. QM results reveal that a single water molecule acts as a proton transfer bridge by establishing hydrogen bonds with both TEPA and HMDI, and the hydrogen bonds lower the free energy barriers from 15.1–16.6 kcal·mol^–1 (in the absence of a single water molecule) to 2.3–3.1 kcal·mol^–1 (in the presence of a single water molecule) in implicit hexadecane. The interaction energy analysis, performed via the fine-tuned MACE FF, indicates that HMDI preferentially resides in the hexadecane phase, while TEPA is more likely to diffuse across the interface. The result provides an energy-based perspective for the partition in the IP process, and experimental observations confirm that polymerization occurs toward the organic phase. These findings provide direct evidence that a single water molecule facilitates IP, offering critical insights for the rational design and precise control of polymerization at the interface to enable improved regulation of reaction kinetics and polymer morphology.

1. Introduction

Interfacial polymerization (IP) is a reaction that takes place at the interface between two immiscible solvents, where monomers from each phase rapidly react to form a polymer layer. − This IP technique enables the fabrication of ultrathin polymer films with controlled morphology, which are essential in a wide range of applications, including membrane separations, , batteries, and supercapacitors. ,

A fundamental question about IP concerns the exact location at which the reaction occurs, presumably at the interface where the two reactive solutions meet. Traditionally, monomer diffusion at the organic–inorganic interface is considered crucial for polymer membrane formation. Experimental research suggests that polymerization takes place predominantly in the organic phase, driven by the higher solubility of amines in alkanes compared to that of acyl chlorides in water. , Some researchers have proposed a diffusion-reaction model to describe how the monomers diffuse and react in the interphase. − They found that polymerization occurs when acyl chloride diffuses from the alkane phase into the ionic liquid phase to react with amine monomers. This model describes the opposite diffusion direction of monomers compared with conventional IP at the alkane–water interface. The polyamide nanofilms gradually grow along the same direction of trimesoyl chloride (TMC) diffusion in the ionic liquids rather than always in the alkane phase, which is verified by tracking the diffusion kinetics of monomers across the interfaces with UV spectroscopy. This can also be explained by the partition coefficients. The use of aromatic solvents in the organic phase or the addition of polar additives to alkanes can increase the partition coefficients of the reactants in the organic phase. This increase in partition coefficients can promote amine diffusion at the interface, leading to a shift in the site of membrane formation. − The location at which the polymer membrane forms is important, since it is linked closely to the resulting morphology, such as thickness and surface roughness, which, in turn, affects its function and performance. For example, polyamide films formed through the reaction of piperazine (PIP) and TMC commonly develop diverse surface features, including nodular, striped, and ring-like structures. In contrast, when m-phenylenediamine (MPD) reacts with TMC, the resulting films predominantly exhibit a characteristic ridge-valley morphology, , which may affect the separation performance of films. Unfortunately, the fundamental factors that determine the reaction velocity and site of polymer formation so crucial for controlling the morphology of membrane polymerization remain unanswered:

• Why does IP proceed so rapidly?

• What governs the reaction kinetics at the molecular level?

• How does solvent influence the energy barriers in the reaction?

TEPA and HMDI exemplify an ultrafast reaction, where the polyurea polymer films form within seconds, making kinetic characterization particularly challenging. In extreme cases, rapid interfacial reactions can even disrupt the emulsion structure, making it difficult to control the encapsulation process. Most existing studies, such as Zhao et al. (2024), rely primarily on experimental observations to investigate IP. For simulation study of IP, various modeling approaches have been developed to explore the IP process. Early studies such as those by Meakin and Freger and Srebnik , introduced cluster–cluster aggregation (CCA) models to simulate film growth, incorporating basic monomer diffusion and reaction kinetics. Further refinements, such as the work of Oizerovich-Honig et al., highlighted the multistage nature of film formation and its dependence on monomer concentration, sticking probability, and spatial distribution. More recently, coarse-grained (CG) simulations, Langevin dynamics, and dissipative particle dynamics (DPD) have been applied to approximate the multiscale dynamics of the IP process, including cluster aggregation, cross-linking, and supramolecular organization at fluid interfaces. Han et al. developed a quantitative rational design framework for IP-based microcapsules through machine learning models; the IP process can be precisely controlled by the model variables, but it only provides statistical relationships lacking in-depth mechanism understanding. These models, while insightful, often rely on highly simplified monomer geometries and time/length scales far removed from real experimental systems. − Current computational studies of IP primarily focus on mesoscale phenomena and structural predictions. These models lack the atomistic-level resolution needed to uncover how reaction mechanisms, interfacial interactions, and diffusion behavior work together to drive film formation.

Our study aims to clarify systematically the reaction mechanisms underlying the polymerization of TEPA and HMDI at the water/hexadecane interface by integrating QM, MD, and experiments. We used QM to calculate the atomistic reaction pathways of IP between TEPA and HMDI, assessing the impact of solvation effects, the facilitation by a water molecule, and amine types (primary and secondary) in TEPA, as well as the conformers of HMDI (trans–trans HMDI and trans–cis HMDI). The diffusion of TEPA and HMDI in water and hexadecane was studied via classical molecular dynamics, while their interaction properties were further explored using fine-tuned machine learning force fields in the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS), providing insights into monomer diffusion and intermolecular interactions. Finally, we verified the interfacial reaction characteristics through experimentation, validating the computational findings and providing macroscopic evidence of the reaction localization. This work provides a fundamental understanding of amine-isocyanate interfacial chemistry, which can be applied to the rational design of interface-based polymerization, encapsulation processes, and advanced material synthesis. −

2. Results and Discussion

Figure shows the structures of TEPA, HMDI, and the intermediate. Here, the (a) primary amine and secondary amine of TEPA, (b) trans–trans-HMDI, and (c) cis–trans-HMDI were used to investigate proton transfer in IP. (d) cis–cis-HMDI was not studied, since trans–trans-HMDI and cis–trans-HMDI can serve as representatives for all possible isocyanate group arrangements in HMDI. Figure e is the intermediate of the initial step formed between TEPA and trans–trans-HMDI. In this study, we investigated cases without water and with the assistance of a single water molecule as a proton transfer bridge between the amine and isocyanate groups.

1.

Molecular structures of the IP process studied in this work. (a) Molecular structure of TEPA. (b–d) Three conformers of HMDI: (b) trans–trans-HMDI, (c) cis–trans-HMDI, and (d) cis–cis-HMDI. (e) Representative intermediate formed in the initial step of the reaction between TEPA and HMDI, which ultimately leads to a cross-linked polymer network.

2.1. Proton Transfer Mechanism

The proton transfer pathways between TEPA and HMDI are calculated to explore possible reaction mechanisms, including both direct proton transfer (with the absence of a water molecule) and proton transfer promoted by a single water molecule. The middle secondary amine was chosen to represent the secondary amine in TEPA. The models investigated in this study are illustrated in Figure and Figure S1 of the Supporting Information (SI). Table summarizes the details of the reaction associated with these models. Owing to the high energy barriers (>20 kcal·mol^–1) reported for reactions between isocyanate groups and water, these reactions are kinetically hindered under ambient conditions and are therefore excluded from consideration in our system.

2.

Schematic of the reactant and product for the direct proton transfer reaction between TEPA and HMDI investigated. (a) Primary amine of TEPA with trans–trans-HMDI (Model 1). (b) Secondary amine of TEPA with trans–trans-HMDI (Model 2). (c) Primary amine of TEPA with cis–trans-HMDI (Model 3). (d) Secondary amine with cis–trans-HMDI (Model 4). The corresponding reaction pathways assisted by a single water molecule for Model 5 to Model 8 are detailed in Figure S1.

1. Molecules of TEPA and HMDI Studied in This Work.

Model no.	TEPA	HMDI	Water molecule number
1	primary amine	trans–trans-HMDI	0
2	secondary amine	trans–trans-HMDI	0
3	primary amine	cis–trans-HMDI	0
4	secondary amine	cis–trans-HMDI	0
5	primary amine	trans–trans-HMDI	1
6	secondary amine	trans–trans -HMDI	1
7	primary amine	cis–trans-HMDI	1
8	secondary amine	cis–trans-HMDI	1

The reaction schemes of direct proton transfer and a single water molecule assisted proton transfer are shown in Figure a,b, in which R, TS, and P represent the reactant, transition state, and product, respectively. The average free energy barriers (ΔG) of direct proton transfer between TEPA and HMDI are 14.7, 16.0, and 25.3 kcal·mol^–1 in the gas phase, implicit hexadecane, and implicit water, respectively (Figure c–e). Notably, the average ΔG in implicit water is 9.3 and 10.6 kcal·mol^–1 higher than implicit hexadecane and gas phase, separately. Based on the Eyring equation at room temperature, the average half-lives of reactants are 6.6, 59.7, and 4.5 days in the gas phase, implicit hexadecane, and water, respectively. The 4.5-day half-life in water is inconsistent with the rapid IP process observed experimentally, indicating that the reaction preferentially proceeds in hexadecane rather than water.

3.

Calculated reaction pathways between TEPA and HMDI. (a, b) Electron-pushing mechanisms for the direct and single water molecule assisted proton transfer, respectively. (c–e) Free energy profiles of direct proton transfer in the gas phase, implicit hexadecane, and implicit water, respectively. R, TS, and P represent reactant complexes, transition states, and products, respectively. (f–h) Free energy profiles of single water molecule assisted proton transfer. The energies for separated reactants are available in Tables S3–S6. Additionally, the energies for reactions (c)–(h) and their corresponding Cartesian coordinates are provided in Tables S14–S17.

Figures f–h show the ΔG profiles of proton transfer facilitated by a single water molecule. The average ΔG values are 2.7, 2.8, and 9.8 kcal·mol^–1 in the gas phase, implicit hexadecane, and implicit water, respectively. The corresponding half-lives are 10.6 ps, 12.6 ps, and 1.7 μs, which are about 9, 10, and 11 orders of magnitude lower than those of in the direct proton transfer cases (6.6 ms, 59.7 ms, and 4.5 days), respectively, consistent with the fast reaction rates observed experimentally.

When the ΔG values of trans–trans-HMDI or cis–trans-HMDI isomers are compared with the same amine site in TEPA, no definitive preference for either is observed. For instance, the ΔG values are 3.1 and 2.3 kcal·mol^–1 for the primary amine, as shown in Figure g. In contrast, both configurations exhibit a barrier of 2.9 kcal·mol^–1 for the secondary amine. Similarly, for the primary and secondary amines of TEPA, no discernible trend is found. With trans–trans-HMDI, the barriers are 3.1 and 2.9 kcal·mol^–1 for the primary and secondary amines, respectively, whereas with cis–trans-HMDI, they are 2.3 and 2.9 kcal·mol^–1.

This mechanism can be further extended to dimeric and oligomeric species as the functional groups remain identical. We investigated the reaction mechanism of subsequent steps by introducing an additional TEPA or HMDI molecule into the initial product. Detailed results are provided in Tables S1 andS2 and Figure S2.

To investigate the effect of additional explicit water molecules on the proton transfer between TEPA (secondary amine) and HMDI (trans–trans-HMDI), the reactions assisted by two or three water molecules were further examined. As shown in Figure a, the direct proton transfer pathway exhibits the highest ΔG. In contrast, the single-water-assisted pathway shows the lowest ΔG, with values decreasing from 14.6-20.4 kcal·mol^–1 (direct proton transfer) to 2.9–3.5 kcal·mol^–1 (single-water-assisted proton transfer). The ΔG values are 10.5 and 9.7 kcal·mol^–1 in implicit hexadecane and implicit water, respectively, when three water molecules are involved. The energies are listed in Table S7.

4.

Influence of the number of water catalysts and counterpoise (CP) correction on ΔG. (a) ΔG for the reaction between TEPA (secondary amine) and HMDI (trans–trans-HMDI) without CP corrections. (b) ΔG with CP. (c) Energy differences between results without and with CP correction. (d–g) Transition-state structures for the four investigated cases in the gas phase.

To account for basis set superposition error (BSSE) in these multibody proton transfer systems, counterpoise (CP) corrections were applied. The CP-corrected ΔG values for proton transfer assisted by varying numbers of water molecules are shown in Figure b. As illustrated in Figure c, the ΔG values obtained without CP correction are consistently slightly lower than the CP-corrected values. However, the differences remain within 0.8 kcal·mol^–1 for all cases, indicating that BSSE has a negligible effect on the calculated energetics. Consequently, the CP-corrected free energy barriers confirm the reliability of the observed trends. The transition-state structures for the four systems are shown in Figure d–g, with their corresponding coordinates detailed in Table S8.

The effect of explicit water molecules as a microsolvation environment were studied further; 0, 1, 6, and 26 additional water molecules were introduced into Model 6, as shown in Figure S3. The corresponding ΔG decreased from 3.5 to 2.4, 1.0, and 0.3 kcal·mol^–1, respectively (Figure S4), values significantly lower than that of direct proton transfer (14.6 kcal·mol^–1). The monotonic reduction in ΔG with increasing water content highlights the catalytic role of water and the importance of aqueous micro solvation in facilitating proton transfer. The energies, structure parameters of transition states, and Cartesian coordinates are detailed in Tables S9–S11.

We note that the present QM calculations do not provide quantitatively converged ΔG due to the limited number of explicit water molecules. While increasing the extent of the solvation environment consistently stabilizes the proton transfer transition state, the associated ΔG reduction is discussed here only qualitatively. Quantitative assessment of the barrier magnitude would require fully converged micro solvation approaches beyond the scope of the present study.

2.2. Hydrogen Bond Structures

The proton transfer process is commonly described by a classical Grotthuss mechanism, involving sequential ″proton hops″ mediated by water molecules. We analyzed the hydrogen bond (HB) structures for both direct proton transfer and single water molecule assisted proton transfer between the amine group of TEPA and the isocyanate group of HMDI. Figure a–f depict DFT-optimized structures of the reactant, transition states, and products for the direct and single-water-assisted proton transfer pathways. We compared the HBs formed by TEPA (secondary amine) and HMDI (trans–trans-HMDI) in hexadecane. In the direct proton transfer pathway, the NH···N angle is 83.7° and the H···N distance is 2.200 Å. These HB structure parameters indicate a weak interaction, as the angle deviates significantly from the ideal HB range of 150–180° and the distance exceeds the typical HB length of approximately 2 Å. In contrast, the presence of a single water molecule significantly alters the HB geometry. As shown in Figure d, the NH···O and OH···N angles are 156.2 and 155.2°, respectively, indicating the formation of strong and stable HBs. Figure b,e presents the HB structures of the corresponding transition states. For the direct proton transfer, the NH···N angle is 110.2°, whereas the NH···O and OH···N angles for the single water molecule-assisted case are 157.5 and 157.7°, respectively. The HB structures in the single water molecule-assisted pathway closely align with the ideal HB geometry, which facilitates a more effective energy pathway for proton transfer.

5.

Hydrogen bond structures and solvation analysis for the TEPA-HMDI reaction. (a–c) Optimized geometries of the reactant, transition state, and product for direct proton transfer. (d–f) Optimized geometries of the reactant, transition state, and product for single water molecule assisted proton transfer. (g) Radial distribution functions for atom pairs in the water/hexadecane solvated systems.

To further characterize the solvation environment of TEPA at the molecular level, radial distribution functions (RDFs, g(r)) were computed from MD simulations. Figure g shows the RDFs for the atom pairs of N_TEPA–H_Water, N_TEPA–H_Hexadecane, and N_TEPA–O_Water, calculated using the OPLS-AA force field in LAMMPS. The results demonstrate that the amine groups of TEPA consistently form HBs with water molecules. Specifically, the first peaks in the g(r) of N_TEPA–H_Water and N_TEPA–O_Water are located at r = 1.85 and 2.85 Å, respectively, which are in good agreement with previously reported values. , In contrast, the g(r) for N_TEPA–H_Hexadecane exhibits a monotonic increase toward the bulk value (g(r)→1) without any discernible peaks, confirming no hydrogen bonding between TEPA and hexadecane.

2.3. Interaction Energies

To elucidate why the reaction preferentially occurs in the hexadecane phase, we investigated the interaction energies of TEPA and HMDI with water and hexadecane to reveal the partition mechanism from an energy-based perspective. These energies were derived from MD simulations using a fine-tuned MACE force field from the recently developed MACE-MPA-0 (medium-size) foundation model, which provides near-quantum accuracy at a low computational cost. The mean absolute error (MAE) and root mean square error (RMSE) for both energy and forces are listed in Figure . For the original MACE-MPA-0 model, the MAE and RMSE values were 0.0540 and 0.0844 eV·Å^–1 for force, 4.7987 and 5.7407 eV for energy (Figure a and Figure b), respectively. After fine-tuning, these errors significantly decreased to 0.0077 and 0.0124 eV·Å^–1 for force, 0.0361 and 0.0541 eV for energy (Figure c and Figure d). This substantial reduction in error demonstrates that the fine-tuned MACE FF achieves excellent agreement with the DFT reference data. Details of the training protocol are described in Section .

6.

Accuracy of the fine-tuned MACE FF and calculated interaction energies. Comparison of predicted (a) forces and (b) energies between the Original MACE-MPA-0 model and DFT references. Comparison of predicted (c) forces and (d) energies between the fine-tuned MACE-MPA-0 model and DFT references; the corresponding MAE and RMSE values are provided. (e) Interaction energies between TEPA or HMDI and H₂O or hexadecane molecules.

Interaction energies were calculated for systems containing a single TEPA or HMDI molecule interacting with either 32 water molecules or 6 hexadecane molecules utilizing the fine-tuned MACE FF. Each system was equilibrated in the NPT ensemble at 298.15 K for 0.5 ns with a time step of 0.5 fs followed by a 0.5 ns data collection run. The potential energy of each system was averaged every 100 steps. As illustrated in Figure e, TEPA exhibits a significantly stronger interaction with water (−31.7 kcal·mol^–1) than with hexadecane (−15.4 kcal·mol^–1), indicating its greater stability in the aqueous phase. In contrast, HMDI shows a stronger preference for the hexadecane phase, with an interaction energy of −18.2 kcal·mol^–1, compared to an unfavorable value of 17.9 kcal·mol^–1 with water. The larger energy gap between HMDI-H₂O and HMDI-C16 indicates a hydrophobic driving force that excludes HMDI from the aqueous phase. The energy difference between TEPA-H₂O and TEPA-C16 is 16.3 kcal·mol^–1, which is 19.8 kcal·mol^–1 smaller than the corresponding gap for HMDI (HMDI-C16 and HMDI-H₂O). This indicates that TEPA faces a much lower energy barrier to partition into or interact with the hexadecane phase compared to HMDI’s tendency to enter the aqueous phase, which provides a molecular-level explanation for the preferential occurrence of IP on the hexadecane side of the interface.

2.4. Diffusivity and Interfacial Tension

To further substantiate that polymerization preferentially occurs on the hexadecane side of the interface, a combination of MD simulations and interfacial tension measurements via the pendant drop method was employed to probe the solvation behavior and interfacial properties of the system.

Figure a–h presents the MD simulation results, while Figure i–l correspond to experimental observations. Since higher concentrations can reduce the diffusion coefficient (D), dilute solutions (maintained at approximately 0.5 and 0.9 M) were employed to ensure more accurate and representative measurements of molecular mobility in this work. Based on these conditions, the solute morphologies and diffusion behaviors of TEPA and HMDI were systematically analyzed. Each system was first equilibrated for 1 ns in the NPT ensemble at 1 atm and 298.15 K to relax the volume, followed by a 5 ns NVT production run for data collection. The details of these benchmark systems are provided in Table S12 and Figure S5. The calculation of D via the Einstein relation based on mean square displacements (MSD) analysis , and the details of the simulation setup used to evaluate diffusion are provided in S8 and Table S13. As demonstrated by the potential energy profiles in Figure S6, all four systems converged and remained stable after the initial 1 ns. To ensure the reliability of the force field, the diffusion coefficients of pure water and hexadecane were benchmarked using GAFF2 in Figure S7, with the results showing excellent agreement with expected values. These calculations were primarily performed using the GAFF2 force field (Figure ) and further cross-validated with OPLS-AA to ensure consistency in Figure S8. For TEPA, D _Water = 7.726 × 10^–6 cm²·s^–1 exceeds D _Hexadecane = 5.117 × 10^–7 cm²·s^–1, which is consistent with the simulation snapshots showing a well-dispersed aqueous phase versus small clusters in hexadecane (Figure a,b). For HMDI, D _Water = 2.236 × 10^–6 cm²·s^–1 is larger than D _Hexadecane = 5.013 × 10^–7 cm²·s^–1; furthermore, HMDI packs tightly in water but remains more loosely organized in hexadecane (Figure c,d).

7.

Simulation and experiment-based diffusion and solubility behaviors of TEPA and HMDI in water and hexadecane. (a–d) Solute morphologies of TEPA in water, TEPA in hexadecane, HMDI in water, and HMDI in hexadecane studied by MD simulations, respectively. (e–h) Mean square displacement curves and corresponding diffusion coefficients of TEPA and HMDI in water and hexadecane obtained from MD simulations. (i–l) Experimental observations of solubility behavior. (i) TEPA is completely miscible in water, forming a homogeneous solution. (j) TEPA exhibits phase separation in hexadecane, with hexadecane in the upper layer and TEPA in the lower layer. (k) HMDI forms a hemispherical aggregate in water, surrounded by hydration shells. (l) HMDI is homogeneously dissolved in liquid hexadecane. (m–n) Pendant drop images used for interfacial tension measurements. (m) Water drop in hexadecane, (n) TEPA drop in hexadecane, and (o) HMDI drop in water.

To experimentally validate the simulation results, the solubility behavior of TEPA and HMDI in aqueous and organic phases was examined, with the results presented in Figure i–l. TEPA exhibited complete miscibility in the aqueous phase, consistent with its hydrophilic behavior due to multiple amine functionalities. However, when introduced into liquid hexadecane, TEPA displayed pronounced phase separation, manifesting as a distinct boundary, which indicates negligible solubility in the nonpolar medium. In contrast, HMDI demonstrated different behavior. In the aqueous phase, HMDI molecules self-assembled into hemispherical aggregates surrounded by water, a phenomenon attributed to the minimization of the interfacial energy. This configuration is characteristic of hydrophobic compounds in aqueous media, where surface tension effects drive the formation of discrete spherical domains. Conversely, HMDI showed complete miscibility in liquid hexadecane, as evidenced by the absence of phase boundaries or aggregation phenomena, confirming its lipophilic character. This differential solubility behavior of the two reactants establishes the fundamental basis for their IP dynamics. The experimentally measured interfacial tension images and values obtained from pendant drop measurements are summarized in Figure m–o and Table . The measured interfacial tensions, γ_Hexadecane = 4.58 ± 0.38 mN·m^–1 and γ_Water = 17.47 ± 1.93 mN·m^–1 confirm the immiscibility. Notably, the smaller γ_Hexadecane implies enhanced wetting and accumulation of TEPA on the hexadecane side, which enables the reaction to occur at or near the oil-side interface.

2. Interfacial Tensions Measured by Pendant Drop Method at 298.15 K.

System	Interfacial tension γ (mN·m–1)
HMDI/water	17.47 ± 1.93
TEPA/hexadecane	4.58 ± 0.38
water/hexadecane	36.00 ± 0.01

2.5. Experimental Results

Microscopic analysis of the reaction interface revealed a distinct reaction zone demarcated by white boundary lines, where the formation of polyurea oligomers and polymers occurs in the experimental group encountered after 30 min, as shown in Figure a. Fluorescence imaging further demonstrated that the polymerization reaction predominantly proceeds within the oil phase, as evidenced by Figure b. The spatial distribution of reactive species and products across the interface was subsequently characterized through detailed spectroscopic analysis (Figures S10 and S11).

8.

Overview of the formation of interfacial polyurea. (a, b) Microscopic and fluorescence images of polyurea formation and clear growth after 30 min at the interface of the experimental group. (c, d) Microscopic and fluorescence images of polyurea formation and unclear growth after 30 min at the interface of the control group. (e–h) Early formation process of polyurea upon direct mixing of 99 wt % TEPA and 50 wt % HMDI with increasing time. The scale bars in panels (b)–(h) are identical to that in (a). The arrows in (a)–(d) and (h) indicate the polymer growth direction.

To further investigate the mechanistic role of water in IP, comparative analyses were conducted on the experimental group and control group using the microfluidic platform (Figure a–d). Microscopic examination revealed striking differences in polymerization behavior between the water-assisted and the anhydrous systems, as illustrated in Figure a,b. The water-assisted system exhibited substantially enhanced polymer gel formation, characterized by a distinct broader reaction zone, whereas there is no visible polymer formed in the control group under anhydrous conditions after reaching 30 min. These observations highlight two distinct but complementary functions of water in the IP process: first, establishing and maintaining a well-defined biphasic interface, and second, acting as a previously unidentified catalytic bridge that significantly enhances reactivity between TEPA and HMDI. The dramatic enhancement in reaction kinetics under water-assisted conditions suggests that water molecules play a more sophisticated role than merely providing phase separation, likely forming hydrogen-bonding networks that facilitate reactant orientation, stabilize reaction intermediates, and lower activation barriers through cooperative hydrogen bonding, thereby enabling a more efficient reaction.

The early-stage rapid polymer gel formation upon the direct contact of 99 wt % TEPA and 50 wt % HMDI in the experimental group was further investigated. As shown in Figure e–h, the interfacial region between the two reactants was imaged from 0 to 0.3 s. Notably, substantial polymer gel formation was observed within 100 ms of the initial reactant contact, indicating remarkably rapid polymerization kinetics. This ultrafast reaction can be attributed to the pre-existing hydrogen-bonding network between the amine groups of TEPA and surrounding water molecules, which creates an activated complex that facilitates rapid nucleophilic addition with HMDI. The observed ultrafast gelation kinetics underscores the significance of hydrogen-bond-mediated activation in enhancing the reaction efficiency under ambient conditions. The physical and structural properties of the samples, including membrane thickness, cross-link density, and film morphology, were systematically characterized as shown in Figures S12–S14. To further rationalize this observation, we computed the binding energies of TEPA-H₂O and HMDI-H₂O, as shown in Figure S15. The TEPA-H₂O binding energies fall in the range 8.33–10.76 kcal·mol^–1, indicative of strong hydrogen bonding. Moreover, as the hydration number decreases from 4 to 1, the TEPA-H₂O binding energies consistently remain higher than those of HMDI-H₂O, supporting the conclusion that the rapid reaction of neat TEPA with HMDI is driven by the hydrogen-bonding interactions of surface water molecules with TEPA.

3. Conclusions

In this study, we investigated the mechanism of interfacial polymerization between TEPA and HMDI through QM calculations, MD simulations, and experimental validation. QM calculations reveal that while direct proton transfer between TEPA and HMDI exhibits average free energy barriers of 14.7, 16.0, and 25.3 kcal·mol^–1 in the gas phase, implicit hexadecane, and implicit water, respectively, the presence of a single water molecule dramatically reduces these barriers to 2.7, 2.8, and 9.8 kcal·mol^–1 under the same conditions. This demonstrates that the single water molecule between the amine group in TEPA and the isocyanate group in HMDI facilitates a highly efficient proton transfer mechanism, enabling rapid polymerization at room temperature. Furthermore, the MD simulations employing the fine-tuned MACE FF reveal that TEPA more readily diffuses into hexadecane because the interaction energy difference between hexadecane and water is small, whereas HMDI scarcely enters the aqueous phase owing to a much larger (and unfavorable) interaction energy difference in water versus hexadecane. Experimental observations confirm that polymerization occurs in the hexadecane phase of the interface and reacts faster for TEPA with the assistance of water. These findings provide fundamental insights into the interfacial behavior and reaction mechanisms of TEPA and HMDI, elucidating the key factors governing polymerization and thus providing a theoretical foundation for controlling reaction kinetics and tuning product morphology.

4. Methodology

4.1. Classical Molecular Dynamic Simulations to Study Diffusivity

MD simulations were performed in LAMMPS using General AMBER Force Field 2 (GAFF2) with Austin Model 1 with Bond Charge Correction (AM1-BCC) charges, results were cross-validated with Optimized Potentials for Liquid Simulations - All Atom (OPLS-AA). ,, The velocity Verlet algorithm was employed by using a time step of 1 fs. Long-range electrostatic and van der Waals interactions were calculated by using the particle–particle particle–mesh (PPPM) algorithm with a cutoff distance of 1 nm. All MD simulations employing classical GAFF2 and OPLS-AA force fields (FFs) were performed at 298.15 K, with the temperature controlled using a Nosé–Hoover thermostat , and a coupling time constant of 100 fs. For each case, eight independent simulations with different initial velocities were performed to obtain the average values. The Visual Molecular Dynamics (VMD) and Open Visualization Tool (OVITO) were used to visualize the calculation results. ,

4.2. Fine-Tuning the MACE-MPA-0 Model

To achieve near-DFT accuracy for the specific chemical environments of the TEPA/HMDI system, we performed system-specific fine-tuning on the MACE-MPA-0 (medium) foundation model. The training data set was generated from NVT ensemble MD simulations of eight systems at 298.15 K and 1 atm, using the original MACE-MPA-0 model in LAMMPS with a 0.5 fs time step. Trajectories were recorded every 100 steps, with system details summarized in Table . For each system, 2000 frames were generated. Subsequently, energies and forces were calculated via single-point DFT calculations. The final data set comprised 16,000 frames, from which 4000 and 1000 frames were randomly selected for training and testing, respectively.

3. Summary of the Datasets Used for Fine-Tuning the Pretrained MACE-MPA-0 Model, Utilizing Randomly Selected Configurations.

System	Number of molecules
H2O	32
C16	6
TEPA/H2O	1/32
HMDI/H2O	1/32
TEPA/C16	1/6
HMDI/C16	1/6
TEPA	1
HMDI	1

Fine-tuning was performed using the mace_run_train workflow with the MACE-MPA-0 (medium) model as the foundation model. Both energies and forces were used as training targets, with a higher weight assigned to forces (energy weight = 1, force weight = 10) to ensure accurate representation of atomic interactions and dynamical behavior. Root mean square force scaling was applied during the training. A learning rate of 1 × 10^–4 was used, with a batch size of 2 for both training and validation.

To assess the sensitivity of the fine-tuned model to the relative weighting of energy and force terms in the loss function, seven different combinations of energy and force weights were systematically tested. For each weighting scheme, the MAE and RMSE values of both energies and forces were evaluated, as summarized in Table . Among the tested combinations, the model trained with an energy weight of 1 and a force weight of 10 yielded the best MAE and RMSE for both energies and forces, indicating the best overall balance between energetic accuracy and force fidelity. These energy and force errors are illustrated in Figure , with additional localized plots for energy provided in Figures S16 and S17. Consequently, this weighting scheme was selected for the final fine-tuned force field and was used in all subsequent calculations of interaction energies.

4. Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) of Energies and Forces for Fine-Tuned MACE Models Trained with Different Energy and Force Weight Combinations.

No.	Wnergy weight	Force weight	Energy MAE (eV)	Energy RMSE (eV)	Force MAE (eV·Å–1)	Force RMSE (eV·Å–1)
1	1	1	0.0411	0.0536	0.0099	0.0161
2	1	10	0.0361	0.0541	0.0077	0.0124
3	1	100	0.0637	0.0845	0.0075	0.0121
4	1	300	0.1527	0.1911	0.0075	0.0120
5	10	1	0.0389	0.0538	0.0118	0.0192
6	100	1	0.0358	0.0493	0.0176	0.0275
7	300	1	0.0410	0.0563	0.0226	0.0343

4.3. Quantum Mechanical Calculations

All quantum mechanical calculations were conducted using the Gaussian 16 package at 298.15 K and 1 atm. The M06-2X method and the def2-TZVP basis set were used to obtain geometry optimization, transition-state searching, intrinsic reaction coordinate (IRC), and vibrational frequencies. The minima and transition states were confirmed through calculations of the vibrational frequencies, where transition states were identified by the presence of a single imaginary frequency. The unscaled vibrational frequencies were used to estimate the zero-point energy corrections. The effect of solvent was accounted for using the implicit solvation model based on density (SMD). Calculations were performed with dielectric environments corresponding to water (dielectric constant ε = 78.36) and hexadecane (ε = 2.04) to represent polar and nonpolar solvents, respectively. For stochastic packing of water solvation shells around some of the transition-state structures, Orca 6.0.0 software package was used. GaussView 6.0 was used for molecular model construction.

4.4. Experimental Setup

A custom microfluidic platform was constructed using two clean glass slips (75 × 25 × 1 mm) stacked to form a controlled microchannel; the experimental setup is shown in Figure S9. For enhanced visualization, TEPA was labeled with 0.5 wt % FITC prior to introduction into the system (wt % denotes weight percentage, defined as by 100 × m _solute/m _solution). A 1 μL portion of reactant solutions (TEPA and HMDI) was carefully introduced from opposite ends of the microchannel, allowing capillary forces to facilitate their transport and subsequent merger at the center, as shown in Figure a. The entire polymerization process was monitored in real-time using a Zeiss SteREO Discovery. V12 microscope (magnification: 100×) under controlled ambient conditions (25 °C).

4.5. Interfacial Tension

The interfacial tensions of TEPA/water, TEPA/hexadecane, and water/hexadecane systems were determined using a CA100 contact angle measurement gauge (Guangdong Beidou Precision Instrument Co., Ltd.) employing the pendant drop method at 20 °C, as shown in Figure m–o. Image analysis was performed using ImageJ software with five independent measurements collected for each system to ensure statistical reliability.

4.6. Water-Assisted Interfacial Polymerization Kinetics

To investigate the role of water in the IP process, experiments were conducted in a controlled atmosphere glovebox, where the water content was maintained at < 0.001 ppm. Prior to experimentation, TEPA was rigorously dried under a continuous flow of high-purity nitrogen gas (99.999%) for 72 h. Two experimental conditions were established: 99 wt % TEPA in deionized (DI) water and 50 wt % HMDI in liquid hexadecane as the experimental group and 100 wt % TEPA and 50 wt % HMDI in liquid hexadecane as the control group.

The data that support the findings of this study are available within the article.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acscatal.5c08183.

• Computational models and energetics; molecular simulations and diffusion analysis; water-assisted reaction and solvation mechanisms; experimental setup and characterizations; membrane structural properties; machine learning force field benchmarking; DFT energies; and Cartesian coordinates (PDF)

#.

B.L. and Y.Z. contributed equally to this work.

Biyuan Liu: writingoriginal draft, methodology, and data curation. Yonglin Zhang: data curation and experiment. Ying Zhao: data curation and experiment. Prabhat Prakash: review and editing, and data curation. Liyuan Huai: writingreview and editing, and data curation. Hancheng Ma: experiment. Zhengtang Luo: supervision and resources. William A. Goddard III and Jinglei Yang: writingreview and editing, supervision, resources, project administration, investigation, funding acquisition, formal analysis, and conceptualization.

The authors declare no competing financial interest.