When self-assembly meets interfacial polymerization

Interfacial polymerization (IP) and self-assembly are two thermodynamically different processes involving an interface in their systems. When the two systems are incorporated, the interface will exhibit extraordinary characteristics and generate structural and morphological transformation. In this work, an ultrapermeable polyamide (PA) reverse osmosis (RO) membrane with crumpled surface morphology and enlarged free volume was fabricated via IP reaction with the introduction of self-assembled surfactant micellar system. The mechanisms of the formation of crumpled nanostructures were elucidated via multiscale simulations. The electrostatic interactions among m-phenylenediamine (MPD) molecules, surfactant monolayer and micelles, lead to disruption of the monolayer at the interface, which in turn shapes the initial pattern formation of the PA layer. The interfacial instability brought about by these molecular interactions promotes the formation of crumpled PA layer with larger effective surface area, facilitating the enhanced water transport. This work provides valuable insights into the mechanisms of the IP process and is fundamental for exploring high-performance desalination membranes.

The data were collected 5 times and processed by computer. Each sample was repeated independently for 3 times.

Micelle size distribution measurement
The size distribution of micelles in the surfactant solution were evaluated by dynamic light scattering (DLS; ELSZ-1000ZA, Otsuka Electronics, Japan).

Positron annihilation lifetime spectroscopy (PALS)
Free-volume cavity and size distribution of the PA membranes were investigated by PALS (PALS-200A, Toray Research Center, Japan). The radioactive source 22 Na (18 μCi) was sandwiched between two Kapton foils with a time resolution of 278-280 ps at room temperature.
PALS data were recorded with total counts of 5 million. In this work, the free-volume voids are assumed as spherical, and the average vacancy volume size (Va) of the free-volume voids can be estimated by the following equation (56) where Ii (%) is the o-Ps intensity of the estimated pick-off lifetime τi, and Vf represents the mean free volume (Å 3 ) of the cavity in the PA active layer calculated by the mean free volume radius r.
The Doppler energy spectroscopy (DBES) spectra were determined using PALS with a variable monoenergy slow positron beam (0-5 k eV) and recorded using an HP Ge detector (EG&G Ortec, R&D Center for Membrane Technology Chung Yuan University, Taiwan). The positron source is of a 50 mCi of 22 Na radioisotope beam. The S parameter was reported from the DBES measurement. The S parameter, which was from the o-Ps 2g pick-off annihilation in free volume, yielded information about the depth profile of the free volume (Å to nm) in the polyamide layer (25,60).

Evaluation of membrane filtration property
Pure water permeance and solute rejection (2000 ppm NaCl) evaluation tests were conducted by using a lab-scale cross-flow filtration setup with a 7.4-cm 2 effective permeating area under an operational pressure of 15.0 bar. Each sample was evaluated independently for at least 3 times (especially, the PA-CIP and PA-MCIP membranes were repeated for 5 times). All the samples were pre-compacted at 15 bar for 4-6 h. The pure water permeance (PWP), P (L m -2 h -1 bar -1 ), was calculated using Equation S4: where ∆ is the collected permeate volume (L) during a certain time (∆t, h); ∆ is the applied pressure (bar); and is the effective permeating area (m 2 ).
The solute rejection ratio (R) was calculated using Equation S5: where and are solute concentration of the feed and permeate solution, respectively, which were analyzed using a conductivity meter (Ultrameter II TM 4P, Myron L Company, Japan).
For the tests of neutral molecules, the concentrations of neutral molecules in the feed solution and permeate solution were analyzed by the total organic carbon analyzer (TOC-VCSN, Shimadzu Company, Japan).

Crosslinking degree of PA TFC membranes
The crosslinking degree of the TFC membranes was investigated by calculating the atomic concentrations at the membrane surface using Equation S6 (61,62).
where the values of m and n were calculated from the O/N ratio from the XPS data using Equations S7 and S8:

Thermodynamic properties of the interface during the diffusion of MPD
The diffusion-reaction IP process was decoupled in MD simulations, and for simplicity, we investigated the interfaces of the three systems with 3 models containing no TMC molecules. MD simulations were carried out to reveal the interfacial properties during the diffusion process of MPD introduced to avoid the interaction between the n-hexane molecules and the periodic image of the bottom layer of water molecules in the surface) with the interface parallel to the XY plane (Fig. S3).
In particular, a monolayer comprised of 42 uniformly distributed SDS molecules was placed at the water/n-hexane interface in both MLIP and MCIP systems. In addition, an SDS micelle containing 50 SDS molecules with a radius of 15 Å underneath the SDS monolayer with a distance of 25 Å were constructed in MCIP system. The numbers of SDS molecules in the monolayer and in the micelle, as well as the distance between the monolayer and the micelle, is consistent with literature (7,25,52). By appropriately setting the positions of all molecules, a geometry optimization process was first performed, followed by a dynamic run for 20 ps with NVT (constant number of molecules, volume and temperature) ensemble at 298 K. To make sure all systems have reached equilibrium, another run for 20 ps with NVE (constant number of molecules, volume and energy) ensemble followed by 1 ns with NVT ensemble were applied, and the energy as well as temperature of all systems reached the steady values. The simulation results of the final 20 ps run in NVT were used for data analysis. The water/n-hexane interfaces in the three MD systems were determined at around 60 Å in the Z direction with two different definitions. For CIP system, the interface was determined by the minimum density of O atoms in water molecules. For MLIP and MCIP systems, the crest of S atoms concentration could be considered as the interface due to the presence of the monolayer.
The diffusion coefficients (D, m 2 s -1 ) of MPD and SDS molecules in different systems were analyzed from the slope of mean square displacement (MSD) curves by Einstein relationship (53): where N is the total number of targeted molecules (i.e., MPD or SDS in this work), r (0) and r (t) represent the initial position (m) and position at t (m) of the targeted molecule.

Interactions between MPD and SDS molecules
As the interactions between SDS and MPD molecules include electrostatic interactions and hydrogen bonding, we used adsorption/binding energy to represent their interactions. Specifically, the adsorption energy of a single molecule of MPD onto SDS monolayer and/or the SDS micelle was obtained using Adsorption Locator module in Materials Studio 2020. The models in both MLIP and MCIP systems were constructed without water and n-hexane, as presented in

Effect of heating on thermodynamic properties of the interface
The heat released during IP process has been measured with a microfluidic methodology and the temperature generated near the reaction zone was reported to achieve as high as 80 °C in MPD-TMC based IP system (63). In this work, heating effect was explored by simply investigating the interfacial properties of water/n-hexane interface with and without SDS. Therefore, no MPD and TMC molecules were included in this part. The construction of SDS monolayer in MLIP as well as SDS micelle in MCIP system, and MD simulations were performed similarly as described in Supplementary Section 2.1.1 and Fig. S3, except that the MD models were constructed without MPD molecules. To clarify the heating effects, the monomer-free CIP, MLIP and MCIP models were simulated with NVT thermodynamic ensemble at 298 K and 318 K, respectively. The configurations and the thermodynamic properties of the interfaces were captured and analyzed after 20 ps when the systems reached the steady state.

Dissipative particle dynamics (DPD) simulations
DPD method is a mesoscopic simulation method and can provide thermodynamic properties of the system with a larger spatial scale and longer time scale, compared to MD simulation method.
The theories of DPD simulations can be found in the literature (39,64). In this work, DPD simulations were carried out to investigate the IP process of the MPD and TMC monomers at the water/n-hexane interface [13] . Five different components were modeled in the IP system: water, MPD, SDS in water phase, and n-hexane, TMC in organic phase. The coarse-graining of all molecules into DPD beads (Fig. S5) and the repulsion parameters between DPD beads (Table S4) were calculated from Flory-Huggins parameters as reported previously (54). In CIP system, a cubic simulation box of 100 × 100 × 100 Å 3 was constructed and divided into two slabs (100 × 100 × 50 Å 3 ) representing the oil phase and the water phase, respectively. In the upper slab, a layer of TMC beads mixed with n-hexane beads (n-hexane/TMC = 0.925: 0.075) were introduced as the oil phase.
In the water phase, water and MPD beads (water/MPD = 0.925: 0.075) were added. In particular, the SDS monolayer (composed of 102 SDS molecules in the middle of the simulation box, slab dimensions: 100 × 100 × 20 Å 3 ) was included perpendicular to Z-axis in MLIP and MCIP box (lattice parameters 100 × 100 × 120 Å 3 ). Moreover, an SDS micelle was contained in water phase in MCIP system, with a radius of 15 Å. The density of the whole system is 3.0 in reduced units.
Each slab was independently equilibrated before the system was assembled for simulations. After geometry optimization, an initial period of 10 000 steps was carried out for equilibrating the system, and the simulation productions were performed for an additional 50 000 steps (~10 ns).
The cross-sectional configurations of IP interface were extracted on the last 20 000 (~ 4 ns) steps at every 500 steps.
In most cases, DPD methods can provide qualitative description rather than quantitative analysis of the system, since the chemical bond formation could not be depicted (54). Therefore, the amide bond formation is represented by the close contact of the amino beads in MPD molecules and benzoyl chloride beads in TMC molecules.   to the values reported by previous studies (67,68). However, the CMC of the SDS in the TEA-CSA aqueous solution will dramatically decrease from ~ 0.23 wt% to ~ 0.01 wt% (Fig. S8c). As shown in Fig. S8d, the CMC of the SDS gradually decreases with increasing the concentration of the resultant TEA-CSA, which may be due to the counter-ion effect (55,69). The counter-ion effect weakens the electrostatic repulsion between the negative-charge SDS molecules, reducing the free energy of micellization (Δ 0 <0). As a result, more SDS micelles with lower energy will generate and modulate the interfacial polymerization compared with those in the system without the counter ions. In the case of CSA, CSA can react with TEA to form the organic salt, which is believed to help increase the porosity of the PA selective layer (70). In addition to this effect, TEA-CSA organic salts can affect the CMC of SDS (referring to the effect of TEA).

Supplementary Figures
Notably, TEA molecules don't participate in the reaction with TMC molecules during the interfacial polymerization due to lacking an active amino group in its structure (Fig. S8e). In addition, the XPS results indicate there is no obvious S element signal on the front and rear sides of the exfoliated PA active layer (Fig. S8f), confirming our hypothesis that CSA molecules also are not involved in the reaction with TMC molecules during the interfacial polymerization.      In a previous study, the potential of mean force with the MPD molecule at different locations along the direction perpendicular to the water/n-hexane interface was reported to be roughly linear near the interface (44), i.e., the potential of mean force with the MPD molecule from aqueous solution (~ 5 Å from the interface) into the water/n-hexane interface is linearly decreased from 0 to -4 kT, and linearly increased to ~ 5 kT when it gradually penetrates into the n-hexane phase from the interface until it reaches a certain distance ( ~ 7 Å). The relationship between the depth into the n-hexane phase and the potential energy of MPD could be applied to SDS head group which is also insoluble in n-hexane and exhibits similar polar characteristics (compared to the non-polar nhexane) with MPD.  It can be observed that at equilibrium state, the region between n-hexane surface and first layer of water molecules has a small density of sites (vacuum gap between them as shown in Fig. S18b) and does not appear to be molecular sharp, because the low-density region is sufficiently large, ~ 10 Å, for the system to accommodate voids. The voids between water and n-hexane phases were also found in the literature (71). Heating is likely to accelerate the separation of the two immiscible liquids, since the two bare interfaces are repelled further from each other and a larger separation voids is generated between them (Fig. S18c). The water surface area and the n-hexane surface area are both larger than an equivalent planar system, due to surface roughness brought about by thermal fluctuations. Moreover, the voids between the two immiscible liquids can also be found at the existence of MPD molecules after equilibration (Fig. S3a), emphasizing the strong repulsion (interfacial tension) between water and n-hexane phases at molecular scale.
Despite large voids between water and n-hexane phases, there are small quantities of n-hexane molecules penetrated into water phase or water molecules buried in the n-hexane phase. The sites where close contact of n-hexane and water molecules occur constitute the nano-interfaces for initiating IP reaction because the aqueous reactive monomer (e.g., MPD in this work) can only react with TMC when the two monomers are in contact with each other. The distribution of voids and nano-interfaces are considered to be responsible for the ridge-and-valley surface structure of the CIP membranes, as the free energy cost of cavity formation necessary to solvate small molecules with size of the order of water (or n-hexane), i.e., MPD in the IP process, is therefore reduced at the interface. The interfacial width, or the region of voids and nano-interfaces, is most possible to be related with the roughness of the PA layer in CIP. The addition of a monolayer of SDS at the interface has substantially reduce the repulsion (interfacial tension) between water and n-hexane phases (Fig. S19), even when there are MPD molecules in the aqueous phase (equilibrated MLIP box in Fig. S3b), thus a smooth and stable PA layer is formed in MLIP.
In order to account for the partial molar volumes of mixing, NPT ensemble was also used to perform simulations at constant pressure. The volume changes for the two immiscible liquids are very small in comparison to the volume fluctuations presented in Fig. S18. Therefore, other influential factors are not explored further in this work.  The IP reaction is exothermic and vigorous. The heat release from the IP reaction of MPD and TMC is reported to be 7.0 × 10 -7 kcal mol -1 ps -1 with an interfacial area of 16.77 nm 2 , estimated from the bonding energy between MPD and TMC (44). To mimic the heating effect on the interface, we simplified this process by excluding MPD and TMC from the simulation systems and increased the temperature of the water/n-hexane/SDS system from 298 K to 318 K. Different from CIP systsem, where nanovoids and nano-interfaces of the water/n-hexane interface contribute to the roughness of the PA layer and are obviously responsive to heating (Fig. 3e and S18), the monolayers in both MLIP and MCIP systems are relatively stable (Figs. S19 and S20).
Interestingly, the diffusion of SDS is slightly enhanced by the heating, as indicated in the MSD curves in Fig. 20d, while the average concentration distribution of SDS along Z-axis (Fig. 20e, f) keeps almost unchanged for the two systems before and after heating. To verify the exact movement of SDS molecules, we checked the velocity profiles of the monolayers in MLIP and MCIP, respectively (Fig. S21). It is found that the heating seems to accelerate the translation and rotation of SDS molecules along the directions of X and Y-axes (parallel to the interface), especially at the water/monolayer and monolayer/n-hexane interfaces (Fig. S21a, b for MLIP; Fig. S21d, e for MCIP, respectively). However, the movement of SDS molecules along the Z-axis in an up-down motion (penetrating into n-hexane or dissolving in water) is less accelarated by heating (Fig. S21c,   f). It can be explained that the heating may not be sufficient for SDS to overcome the energy penalty from water phase to n-hexane phase (26), or the heat dissipation is hindered by the monolayer (44).
From this aspect, heating during the IP process, seems to be of insiginificant importance (or at most the secondary importance) to the thermodynamic states of the interfaces for MLIP and MCIP in this work. Therefore, the driving force of the fluctuations in the interface is primarily attributed to the electrostatic interactions among different species (i.e., MPD, monolayer and micelle) in MCIP system.          The optimal condition of the IP process is selected as follows: 0.25 wt% of MPD in aqueous phase and 0.15 wt% of TMC in n-hexane phase.  The performances of PA-TFC membranes from the three approaches by using 1, 6hexanediamine as aqueous monomer are compared in Fig. S36. As a control, the PA TFC membrane formed via CIP had a water permeance of 0.9 L·m -2 ·h -1 ·bar -1 , accompanied with a NaCl rejection ratio of 50.8 %. The MLIP PA TFC membrane showed an improved NaCl rejection ratio of 72.2 % and a decreased water permeance of 0.6 L·m -2 ·h -1 ·bar -1 , the decrease in water permeance is attributed to the reduced effective surface area resulted from the smooth surface structures. In contrast, an enhancement in water permeance (1.2 L·m -2 ·h -1 ·bar -1 ) was observed for MCIP PA membranes, without sacrificing the NaCl rejection (67.6 %). The improvement in water permeance is attributed to the increased effective surface area resulted from the crumpled surface structures, as confirmed by experimental observations (Fig. 4). The change trend of the surficial morphology is consistent with our view of CIP, MLIP, and MCIP we made in the main text. The performances of PA-TFC membranes from the three approaches by using toluene as organic phase solvent are compared in Fig. S37. The PA-TFC membranes formed via CIP and MLIP, respectively, showed similar desalination performances and surficial morphologies, which may be due to the relatively strong-polarity toluene with benzene ring structure facilitating MPD transport across the toluene-water interface, compared to the weak-polarity n-hexane containing aliphatic chain structure. The facilitating transport of MPD across interface induced by toluene could weaken the facilitating transport induced by SDS monolayer. In contrast, The PA-TFC membrane of MCIP exhibited an improved water permeance along with a slightly decreased NaCl rejection ratio, which is attributed to the increased effective surface area resulted from the crumpled surface structures. These crumpled surface structures can be derived from the thermodynamic fluctuations of the toluene-water interface induced by the micelles in the bulk aqueous solution, in turn shaping the initial pattern formation of the PA layer. In a word, SDS monolayer at the organicwater interface is easily influenced by organic phase, while SDS micelles in the bulk aqueous solution is more likely to maintain its own physicochemical properties facing with changes in the organic phase. As a result, the MCIP method has a wider range of applications compared to the MLIP method. The performances of PA-TFC membranes from the three approaches were investigated by choosing cationic CTAB (Fig. S38 a) as surfactant. As shown in Fig. S38  when the CTAB concentration is lower than its CMC, the positive-charged CTAB is adsorbed on the surface of the negative-charged PK due to electrostatic attraction, which further led the MPD aqueous solution containing CTAB to be closely distributed on the surface structure of PK, resulting in the growth of PA along the surface of PK; when the CTAB concentration is higher than its CMC, the CTAB molecules adsorbed on PK surface repelled CTAB micelles and MPD molecules in aqueous solution, causing instability of reaction system and inducing PA to form fold morphology.
In a word, CTAB monolayer and CTAB micelles can effectively regulate the IP reaction, endowing the PA nanofilms with different surficial structures and different desalination performances in the MLIP and MCIP processes, which verified that the effectiveness of MLIP and MCIP in regulating the IP reaction is not limited to the system of SDS surfactant. The performances of PA-TFC membranes from the three approaches were investigated by choosing nonionic Tween 80 (Fig. S39 a) as surfactant. As shown in Fig. S39    The separation performance and FESEM images of PA membrane prepared without TEA-CSA additives, with 2 wt% TEA, with 4 wt% CSA, and with 2 wt% TEA and 4 wt% CSA were presented in Fig. S40. Compared with PA membrane without any additions, PA membrane only with 4 wt% CSA showed non-selective separation performance due to many obvious surficial defects (Figs. S40 a and S40 c). CSA is a strong organic acid, and it can affect the pH in the aqueous amine solution, resulting in the protonation of amine groups in MPD, making it difficult for MPD molecules to diffuse into the organic phase, thus generate non-selective defect in the active layer (74,75). PA membrane only with 2 wt% TEA exhibited slightly enhanced water permeance with negligible decreased salt rejection, and similar surficial FESEM image with those of PA membrane without any additions. These results impliedly demonstrated that excessive TEA (relative to 0.25 wt% MPD) as a catalyst may hinder the diffusion of MPD into the n-hexane phase, inducing some tiny defects. PA membrane with TEA-CSA organic salts displayed slightly increased water permeance without compromising salt rejection, compared to that of PA membrane without any additions, which can be attributed to the rougher surface morphology (Fig. S40 e). Those results demonstrated TEA and TEA-CSA organic salts, except for CSA additions, would not exert a notable effect on the separation performance and surface morphology. Therefore, TEA-CSA organic salts, which were used into the CIP, MLIP, and MCIP process, serve mainly to affect the CMC of SDS surfactant, without directly regulating the IP process.