Protonation and deprotonation of edges in graphene oxide and MXenes as a driving force for actuation in responsive 2D membranes

Abstract

Controlling bending in two-dimensional (2D) materials is essential for the development pf responsive systems and miniaturized actuators. Traditional approaches, particularly for graphene oxide (GO), rely on mismatched thermal expansion between GO and its reduced form. Here, we report a scalable method for assembling anisotropic membranes with chemically distinct top and bottom surfaces, achieved through pH-programmed control of flake protonation. Actuation is driven by edge-to-edge interactions among GO and MXene (Ti₃C₂T_x) flakes, where differential protonation induces localized strain and in-plane flake sliding during thermal dehydration. This gradient in charged and neutral functional groups enables directional bending upon mild heating. Extending this approach to MXenes yields robust, low-dimensional actuators with tunable chemical and mechanical properties. Demonstrated applications include soft robotics and climate-adaptive architecture. Systematic analysis of thermal response, water retention, and fabrication scalability underscores the broad potential of this platform for 2D material-based devices.

A scalable strategy to assemble anisotropic GO and MXene membranes with pH-tuned edge functionalization enables programmable bending via thermo-responsive water release, with applications in soft robotics and adaptive bionic architecture

Introduction

The fundamental working principle for intelligent systems and computing: “Input—Processing—Output” can be reapplied to the construction of functional intelligent materials through “Sensing—Phase Changing—Actuation”. This approach creates materials with dynamic and responsive properties such as two-dimensional (2D) soft actuators. 2D actuators can potentially function in a flat, planar manner which is essential for constructing miniaturized sensors^1,2, microelectromechanical systems^3,4, and the integration of nanotechnology into 4D printable constructs⁵, smart architecture⁶, and interactive textile⁷. However, achieving these dynamic and responsive properties in 2D materials involves addressing several challenges related to understanding the underlying mechanisms, fabrication techniques and scalability.

While ion intercalation or water desorption typically causes uniform expansion or contraction in 2D multilayers, bending requires asymmetric shrinkage or expansion between the top and bottom layers. To date, the development of 2D actuators from 2D materials involved harnessing the distinct swelling behavior of graphene oxide (GO) layers and reduced graphene oxide (rGO) layers^8,9, either in bilayers or through the assembly of the membranes with a skin layer^7,10–12. The proposed mechanism is grounded in the thickness changes occurring within the layers¹³, yet it struggles to fully elucidate the processes of folding and unfolding observed in 2D films. As a result, the underlying mechanisms responsible for this phenomenon remain a topic of ongoing investigation and have yet to be comprehensively understood.

Optimizing 2D membranes for actuator technology requires mastering the lateral expansion/compression of such laminate structures. The general swelling of GO, associated with variations in the interlayer distances between the sheets^8,14 upon release and uptake of water molecules by functional groups on the basal plane^12,15,16, would not provide efficient actuator properties. In this paper, we propose a mechanism responsible for the lateral expansion/contraction of laminated 2D membranes and provide experimental verification of the phenomenon. Specifically, we propose that protonation/deprotonation of the functional groups at the edges of the flakes (upon changes in the pH of the localized water between the layers) results in strong lateral Coulomb repulsion between the flakes, leading to significant changes in the area of the laminates.

In this study, we experimentally demonstrate the temperature-induced actuator action (triggered by uptake and release of water by the charged groups located on the edges) of membranes based on GO and MXene (Ti₃C₂T_x) flakes. In both compounds, the surface charges are regulated by pH-dependent protonation/deprotonation of the functional groups^17–19. Furthermore, we demonstrate that the gradual changes in the lateral expansion properties across the thickness of the membrane, which are required for the actuator function, can be achieved through a self-assembly mechanism. This makes the fabrication of such actuators technologically simple.

At neutral pH, incomplete protonation or deprotonation generates anisotropic membranes with distinct surface charge densities, influencing water binding strength and membrane actuation. Using the bimetallic strip model, we calculated the positive thermal expansion coefficient for protonated and negative one for unprotonated layers in pH-programmed GO, explaining the contraction and retention of layers with varying degree of protonation. Density functional theory (DFT) simulations revealed that edge sites of flakes are significantly more reactive to protons than basal planes, driving lateral shrinking/expansion cycles through changes in electrochemical potential. This pH-assisted programming enables precise control of mechanical and thermal properties in GO and MXene multilayers, creating seamless, robust materials with high actuation amplitude, rapid response, and durability. Such thin, flexible membranes are ideal for applications like breathable bionic architecture for climate control, soft robotics, 4D printing, and beyond.

Results

Self-assembly of GO and Ti₃C₂T_x multilayer membranes

We begin by describing the fabrication of GO and Ti₃C₂T_x membranes with gradient properties across their thickness, which can be programmed during pH-assisted self-assembly (Supplementary Fig. 1). The change in pH affects GO and Ti₃C₂T_x by inducing deprotonation (charging) at basic pH or protonation (neutralization) of their functional groups at acidic pH. Multilayers were constructed using GO flakes (mean size 2.9 μm, Fig. 1a and Supplementary Fig. 2a) and Ti₃C₂T_x flakes (mean size 0.9 μm, Fig. 1g and Supplementary Fig. 2b). Atomic force microscopy analysis (Supplementary Fig. 3) shows that GO flakes exhibit a smooth surface with a thickness of approximately 1.5 nm, typical of single-layer sheets, while MXene flakes display an average thickness of around 3 nm, reflecting their multi-layered structure. The Ti₃C₂T_x dispersions were prepared by selectively etching the aluminum layer from the Ti₃AlC₂ MAX precursor²⁰.

Fig. 1. Morphology and structure of self-assembled multilayer GO and Ti₃C₂T_x membranes.

a SEM image of GO flakes, showing their characteristic sheet-like morphology. Experiments were independently repeated at least three times with consistent results. b Cross-sectional SEM image of the GO membrane prepared at pH 6, showing a densely packed laminar stacking structure, indicative of compact interlayer arrangements. Experiments were independently repeated at least three times with consistent results. c A photograph of the GO membrane prepared by vacuum filtration, illustrating its uniformity and smooth surface. d XRD patterns of GO membranes. The interlayer distance decreases from 8.63 ± 0.04 Å (unprotonated) to 8.33 ± 0.01 Å (protonated), corresponding to two graphene sheets and a water layer²¹. e The subtracted FTIR spectra for GO membranes prepared at pH 2, 6, and 10, highlight pH-dependent changes in hydrophilic functional groups on the GO surfaces. The O-H vibration peaks (3200–3500 cm⁻¹, hydrogen-bonded water), C-O bond intensify with increasing surface charge density. Detailed characterization of FTIR, and Raman spectroscopy for the membranes is provided in Supplementary Fig. 4. f Stress–strain curves of GO membranes. For GO at pH 2, the Young’s modulus is 2.7 GPa with a strain of 3.46%. At pH 6, the Young’s modulus increases to 7.69 GPa, with strain decreasing to 2.73%, and at pH 10, the modulus reaches 13.6 GPa with strain further decreasing to 1.98%. g SEM image of MXene flakes, revealing their exfoliated and sheet-like morphology. Experiments were independently repeated at least three times with consistent results. h Cross-sectional SEM image showing the laminar stacking structure of an MXene membrane prepared by MXene dispersion at pH 6. Experiments were independently repeated at least three times with consistent results. i A photograph of the MXene membrane prepared via vacuum filtration, highlighting its structural integrity and uniformity. j XRD patterns of MXene membranes, the interlayer distance increases from 13.07 ± 0.05 Å to 13.3 ± 0.05 Å with increasing pH. k The subtracted FTIR spectra for MXene membranes prepared at pH 2, 6, and 10. C-O, Ti-O bonds intensify with increasing surface charge density. l Stress–strain curves of MXene membranes. At pH 2, the Young’s modulus is 1.3 GPa with a strain of 1.16%. At pH 6, the Young’s modulus increases to 3.0 GPa, with strain decreasing to 0.72%, and at pH 10, the modulus reaches 5.6 GPa with strain further decreasing to 0.28%. This trend mirrors GO membranes, demonstrating that as pH increases, both materials become stiffer (higher Young’s modulus) and less flexible (lower strain) due to changes in protonation and surface charge density. Source data are provided as a Source Data file.

Aligned GO and MXene multilayers were formed via vacuum filtration of dispersions across pH 2–10. Scanning electron microscopy (`SEM) (Fig. 1b, h) and x-ray diffraction (XRD) (Fig. 1d, j) confirmed well-defined lamellar structures^20,21, with XRD revealing a ~ 0.3 Å interlayer distance increase at higher pH due to electrostatic repulsion^21–23. Fourier transform infrared spectroscopy (FTIR) (Fig. 1e, k) showed intensified O-H, C-O, and Ti-O vibrations in both membranes with increased surface charge density. Stress-strain analysis (Fig. 1f, l) revealed that both membranes became stiffer (higher Young’s modulus) and less flexible (lower strain) as pH increased, attributed to changes in protonation and surface charge density. Supplementary data provides detailed characterization (Supplementary Fig. 4), including water loss trends from thermogravimetric analysis (TGA) (Supplementary Fig. 5).

Self-assembly and actuation properties of bilayer membranes

We first prepared bilayer membranes by sequentially filtering GO dispersions at pH 2 and pH 10 (Fig. 2a). First, a pH 2 GO dispersion was filtered to form a membrane. After removing excess water, a pH 10 GO dispersion was added and filtered, creating in a pH2_bottom/pH10_top bilayer membrane. Using the same method, pH10_bottom/pH2_top and pH2_bottom/pH6_top were also assembled.

Fig. 2. Self-assembly and curvature evaluation of bilayer membranes.

a Schematic illustration of the preparation of bilayer GO membranes by sequential vacuum filtration of pH 2 (dark gray) and pH 10 (brown) GO dispersions. b The snapshots of video recorded bending process of a pH2_bottom/pH10_top bilayer GO actuator under heating (dimensions: 4 ×20 mm). c Bending profiles of membranes fitted with ImageJ software for the curvature analysis. d Schematic diagram and curvature response of bilayer GO membranes with three different structures upon heating from room temperature to 100 °C. In the schematic, dark gray, yellow, and brown represent GO prepared at pH 2, pH 6, and pH 10, respectively. Bilayer membranes were prepared with pH2_bottom/pH10_top, pH2_bottom/pH6_top and pH10_bottom/pH2_top (5 mg each). The pH2_bottom/pH10_top membrane showed the largest curvature (2.91 cm⁻¹). The pH2_bottom/pH6_top membrane exhibited moderate curvature (1 cm⁻¹), and pH10_bottom/pH2_top showed limited response (0.28 cm⁻¹ at 100 °C) due to partial neutralization of the pH 10 layer by protons from the pH 2 dispersion. e Schematic illustration and curvature response of a pH2_bottom/pH10_top bilayer membrane with varying mass percentages of pH10 GO. In the schematic, dark gray and brown represent GO prepared at pH 2 and pH 10, respectively. The total membrane mass was kept at 10 mg. Pure pH 2 GO showed no thermal response, while increasing the pH 10 GO ratio improved curvature, peaking at 50% pH10 GO. At 100% pH10 GO, the membrane still exhibited thermal response with a curvature of 1.11 cm⁻¹ at 100 °C. Source data are provided as a Source Data file.

The membranes were cut into strips measuring 4 mm × 20 mm and heated from room temperature to 100 °C to measure curvature changes (Fig. 2b). pH2_bottom/pH10_top membrane showed the highest curvature (2.91 cm⁻¹ in 22 s), analyzed using the ImageJ (Fig. 2c). The pH2_bottom/pH6_top exhibited moderate curvature (1 cm⁻¹), while the pH10_bottom/pH2_top showed a limited response (0.28 cm⁻¹ at 100 °C) due to partial neutralization when the high-proton-concentration dispersion (pH 2) was filtered through layers formed at pH 10 (Fig. 2d). Adjusting the pH2/pH10 GO mass ratio in the pH2_bottom/pH10_top layers revealed a curvature peak at 50% pH10 GO, with 100% pH10 GO still achieving 1.11 cm⁻¹ at 100 °C (Fig. 2e and Supplementary Fig. 6). A similar bilayer strategy for MXene (pH2_bottom/pH10_top) yielded a curvature of 8.7 cm⁻¹ at 100 °C (Supplementary Fig. 7). Thus, bilayer membranes with charged and uncharged layers exhibited tunable curvature responses to heating.

pH-programmed self-assembly of membranes with chemical and reactivity anisotropy

The pH-programmed protonation drives the self-assembly of anisotropic membranes as illustrated in Fig. 3a-d. During vacuum-assisted sedimentation (Supplementary Fig. 8a, b), uncharged protonated flakes deposit at the bottom and charged deprotonated flakes form the top layers. Potentiometric titration^17,18 data (Fig. 3a)^17,18 shows that at pH 2, the flakes are nearly uncharged, while increasing the pH results to surface charge density reaches −0.16 C·m⁻² for GO flakes and −0.4 C·m^-2 for Ti₃C₂T_x flakes at pH 10. At pH 6–7, approximately 50% of the flakes are charged and 50% are uncharged.

Fig. 3. pH-programmed self-assembly, characterization and curvature evaluation of anisotropic membranes.

a Surface charge density of GO and MXene flakes at different pH, calculated using potentiometric titration. b Schematic diagram of the pH-programmed self-assembly process, where uncharged protonated flakes (gray) deposit first, followed by charged deprotonated flakes (brown). c Thermal actuation performance of GO and MXene at 100 °C. The actuators prepared at different pH. d X-ray photoelectron spectroscopy (XPS) of O1s analysis of GO membranes. The O1s spectrum further supports these observations. At pH2, the relative atomic concentrations of the functional groups are as follows: -OH (2.37 at%), C-O (55.18 at%), C=O (42.45 at%). At pH6, the concentrations are -OH (16.45 at%), C-O (66.55 at%), 531.03 eV C = O (16.99 at%). At pH10, the concentrations are C=O (44.28 at%), C-O (53.47 at%) and -OH (2.25 at%). e XPS of O1s analysis of MXene membranes. At pH2, the relative atomic concentrations of the functional groups are as follows: C-Ti-OH (38.45 at%), C-Ti-O (39.65 at.%), Ti-O (21.90 at%). At pH 6, the concentrations are C-Ti-OH (35.85 at%), C-Ti-O (39.92 at%), Ti-O (24.23 at%). At pH 10, the concentrations are Ti-H₂O (2.57 at%), C-Ti-OH (32.10 at%), C-Ti-O (41.18 at%), Ti-O (24.15 at%). f Differential FTIR spectra of GO and MXene membranes at various pH, highlighting the C=O band at 1600–1700 cm⁻¹. The set-up for in-situ FTIR experiments before and after infrared irradiation is shown in Supplementary Fig. 9. The spectra were obtained by subtracting the room-temperature data from those after 3 min of infrared light irradiation. After infrared irradiation, a decrease in C=O band intensity is observed. Additionally, pH-dependent peak shifts are evident: under acidic conditions (pH 2), the C=O peak exhibits a red shift compared to alkaline conditions (pH 10), indicating protonation of carbonyl groups. g Thermal actuating performance of pH-programmed GO and MXene multilayers with different pH. The pH 2 GO membrane remains thermally unresponsive, while the pH 6 membrane exhibits the largest curvature. The pH 10 membrane bends under heating, but with a smaller curvature compared to pH 6, reflecting reduced thermal asymmetry. Actuator dimensions: 4 ×20 mm. Source data are provided as a Source Data file.

X-ray photoelectron spectroscopy (XPS) analysis (Fig. 3d, e and Supplementary Fig. 8c-f) reveals an increased signal around 534.0 eV in GO membranes prepared at pH 6, compared to those at pH 2 and pH 10. This signal is attributed to bound water associated with charged carboxyl groups in GO. MXene membranes do not exhibit a similar effect, probably due to their lower overall water content. TGA analysis (Supplementary Fig. 5) indicates that MXene membranes content a maximum 4 wt% water, while GO membranes retain approximately 15 wt%. Differential FTIR spectra (Fig. 3f) of both GO and MXene membranes, obtained after 3 min of infrared irradiation, show a decrease in C=O peak intensities and pH-dependent shifts at 1600–1700 cm⁻¹. Under acidic conditions (pH 2), the C=O peak red-shifts compared to alkaline conditions (pH 10), indicating carbonyl protonation. This highlights pH-induced chemical modifications in actuation behavior (setup in Supplementary Fig. 9).

Actuation properties of membranes with chemical and reactivity anisotropy

Upon heating, anisotropic membranes bend toward the charged side, with curvature correlating to the ratio of charged and uncharged flakes (Fig. 3c, g). GO multilayers show a maximum response at pH 6, while Ti₃C₂T_x peaks at pH 7, exhibiting a larger amplitude due to the largest chemical gradient. In contrast, isotropic multilayers at pH 2 show no thermal response. To further examine the effects of pH variation during self-assembly process, GO membranes were prepared by progressively adjusting the pH of the dispersion from 2 to 4, 6, 8, and 10 during filtration (Supplementary Fig. 10). After drying, thermal response tests revealed a curvature of 2.1 cm^-1 at 100 °C, slightly exceeding the 1.9 cm^-1 observed for membranes prepared from a pH 6 dispersion. These results confirm that pH-programmed anisotropy influences the thermal response of multilayer membranes.

Calculation of thermal expansion coefficients using bimetallic strip model, adapted for 2D heterostructures

Using the bimetallic strip (BMS) model, which describes thermal expansion as the change in length or area of materials in response to temperature, we calculated the thermal expansion coefficients of our membranes. Adapted for 2D heterostructures, this model explains lateral expansion and deformation, differing from the volume changes characteristic of 3D hydrogels.

First, we validated its applicability for predicting the bending behavior of GO/rGO membranes with known compositions and used it to analyze the actuating properties of pH-programmed membranes (details in Supplementary Note 1). This approach works well for GO, as it has a fully uncharged counterpart, rGO, obtained through chemical reduction (details provided in Supplementary Note). In contrast, fully unchanging MXenes is not possible due to the intrinsic presence of surface terminations (–OH, –O, and –F), which cannot be completely removed without compromising the material’s structural integrity.

Using the Eq. 1, deriving from the BMS model, we calculated the thermal expansion coefficients (φ) of the charged and uncharged layers through the average curvature (k) at different T and half thickness of a membrane (d). The detailed calculation process is in the Supplementary Note 1, Supplementary Table 1 and Table 2.

$$k=\frac{\mathrm{\Delta }T\cdot \mathrm{\Delta }\phi }{d}$$ 1

As shown in Fig. 4a, the φ values of the charged and uncharged GO layers in the membranes prepared at pH 2 fluctuate around 0, making them relatively insensitive to temperature changes (Fig. 3e). In membranes assembled at pH 10, both the charged and uncharged layers exhibit negative φ, indicating contraction with increasing T. However, the φ of the charged GO layer is more negative than the uncharged GO layer, the GO strip bends in the direction of the charged top layer when T increases. At pH 6 (Fig. 3e), the uncharged GO layers have a positive φ, while the charged layers have a negative φ. The contrast in φ between the layers creates significant chemical anisotropy, resulting in the largest thermal response amplitude at pH 6.

Fig. 4. Mechanistic insights into the bending behavior of pH-programmed anisotropic membranes.

a Thermal expansion coefficient (φ) of charged and uncharged GO layers at pH 2, 6, and 10, respectively. $\Delta T$ denotes the temperature increment. The φ values were calculated using the bimetallic model, where positive φ values indicate thermal expansion, while negative values represent thermal contraction. b Schematic diagram illustrates the bending behavior of an asymmetrical membrane upon heating due to different edge distance changes of uncharged (dark gray) and charged (brown) GO flakes. The d₀ and d₁ denote the distance between the edges of the charged GO flakes before and after heating, respectively. c DFT study of binding strengths with water molecules of protonated and unprotonated layers: charge density difference, Bader charge transfer, and adsorption strength (E_ads) between the water network and GO edges. The charge accumulation and depletion are depicted by green and blue color, respectively. d Thermal response of GO membranes as a function of GO flake size. The flake size was tuned by varying sonication time, with smaller flakes exhibiting a higher edge index (defined as the flake perimeter divided by its equivalent diameter). The inset images show the folded membranes (scale bar: 5 mm). e Effect of different cations on the thermal actuation behavior of GO membranes. Inset images of folded membranes under IR stimulation are shown (scale bar: 5 mm). Source data are provided in the Source Data file.

Mechanistic insights into the actuation behavior of pH-programmed anisotropic membranes

Figure 4b provides a schematic illustration of the actuation mechanisms in 2D structures. In contrast to 3D actuators, relaying on volume change, bending in 2D materials is driven by in-plane shrinkage via edge-to-edge attraction strength between charged flakes. This behavior arises due to the distinct reactivity of edge sites compared to basal planes: exposed and terminated edge sites exhibit greater reactivity toward protons, making them more prone to protonation^17,23. Specifically, –COOH groups (pKa ∼4.6) are predominantly located on the edges of GO flakes, while -OH groups (pKa ∼10) are attached to the basal plane¹⁷.

DFT computations (Fig. 4c) reveal that water molecules exhibit stronger binding with deprotonated edge groups, with an adsorption energy of -0.54 eV. Upon protonation, the adsorption energy of the edge groups for water is reduced to -0.19 eV. Deprotonation of the edge groups introduces two charge transfer channels at the water-GO interface, resulting in a significant charge transfer (0.1223 |e | ) from oxygen to hydrogen atoms compared to the uncharged state (0.0585 |e | ). Further, crystal orbital Hamilton populations (COHP) analysis (Supplementary Fig. 11) also confirms that the charged edge groups on GO (-0.23 eV) have a stronger ability to interact with water than uncharged edge groups on GO (-0.12 eV).

Combining experimental observations with computational data, we conclude that water molecules interact more strongly with charged edges than with uncharged edges. At elevated temperatures, both charged and uncharged layers lose water. However, only charged GO flakes undergo reorganization upon heating. This occurs because the removal of water molecules that screen the charged groups leaves the flakes’ edges in a metastable state, which in turn promotes in-plane sliding and reorganization of the 2D sheets.

This hypothesis is experimentally supported by reducing the size of GO flakes via ultrasonication, which increases the number of edge sites and enhances the thermal responsiveness of GO membranes (Fig. 4d). Furthermore, recognizing that counterions at basic pH can modulate edge-to-edge interactions, we tested membrane responses in the presence of NH₄⁺, Na⁺, K⁺, and Li⁺. At low pH, carboxyl groups (–COOH) are protonated and interact weakly with cations. At neutral or basic pH, deprotonated –COO⁻ groups can participate in ion-pairing, with the interaction strength governed by the Gibbs free energy of binding. These interactions are influenced by ion size, charge density, and hydration energy. As observed in protein systems, cation-binding strength follows the trend: Li⁺ > NH₄⁺, with Na⁺ and K⁺ intermediate in affinity²⁴. GO membranes exhibit a similar trend (Fig. 4e), with Li+ binding most strongly to partially protonated –COO⁻ groups following water release.

Furthermore, we should not exclude the contribution of attractive hydrophobic forces between flakes. In experimental measurements, it was observed that hydrophobic interactions, which contribute to alterations in the water structure when two surfaces approach each other, exhibit greater strength compared to van der Waals attraction^25,26. Such short-range hydrophobic attraction typically diminishes exponentially with a decay length of around 1–2 nm. On the other hand, the long-range hydrophobic attraction extends up to approximately 100 nm, originating from gas bubbles forming spontaneously through water decomposition which is, in our case, more prone to catalytically active edges. These gas bubbles generate an attraction, primarily due to the meniscus force²⁷.

Responsive properties and applications of pH-programmed anisotropic membranes

To assess the responsive properties of our pH-programmed graphene-based actuators, we conducted a series of tests to evaluate their response time, cyclic stability, and long-term stability. To measure the response time, we subjected the actuators to a heating process on a hot plate set at 100 °C for approx. 12 s. Subsequently, we cooled the actuators on an RT glass plate for about 5 s, and they returned to their initial state (Fig. 5a). Curvature changes of GO membranes during folding and unfolding cycles under thermal conditions showed consistent, reversible variations, indicating no hysteresis and stable performance (Supplementary Fig. 12). Figure 5b shows the cyclic stability of the GO actuators, which underwent repeated bending and recovery motions for up to 500 cycles, maintaining a consistent bending curvature. The actuators exhibited stable performance throughout the cycles, highlighting their durability and reliability over extended use. This cyclic stability test further underscores the material’s robustness, demonstrating its potential for long-term practical applications. Furthermore, it is notable that the present research on 2D material-based actuators often overlooks the study of their long-term stability, which is an essential factor affecting their practical application^{1,9,16,28–32}. Here, we tested our actuators under ambient conditions and showed properly repeatable responsiveness even after two years of storage (Fig. 5c). This could be attributed to the fact that our one-step method only uses a single material, GO, with no long-term delamination issues between the layers of the membrane.

Fig. 5. Responsive properties of pH-programmed actuators and their application in smart architecture.

a Response and recovery times of the pH6 GO actuator under thermal stimulation. b Cyclic stability of a pH-programmed GO actuator over 500 thermal actuation cycles, with performance recorded every 50 cycles. c Long-term operational stability of the pH6 GO actuator. d Time-resolved actuation of a flamelike graphene-based window under IR illumination. e, f Temperature-dependent actuation of the flamelike window under daily temperature fluctuations, demonstrating pH-programmed responsiveness. g, h Conceptual implementation of the breathable 2D actuator material in a smart house design, with corresponding IR thermal images. Room temperature is around 22 °C and sunlight exposure raises temperatures up to 35 °C. i Temporal evolution of indoor temperature in buildings equipped with smart versus traditional windows when exposed to direct sunlight. Source data are provided as a Source Data file.

The membrane response was also evaluated under extremely dry conditions. GO membranes prepared at pH 6 were placed in a glove box with 0.01 ppm moisture for 24 h. At 27 °C, the initial bending curvature increased to 3.66 cm⁻¹ due to dehydration and further increased to 3.89 cm⁻¹ at 100 °C (Supplementary Fig. 13a). When returned to ambient conditions, the membrane’s responsiveness was restored (Supplementary Fig. 13b). This demonstrates that our graphene-based actuators have a consistent and reproducible response, facilitating their reliability and suitability for practical applications. In addition, comparing the response performance of our actuators with the reports of one-step preparation of single GO and MXene actuators with structural gradient as listed in Supplementary Table 3, our GO actuator shows a response time of 12 s and a curvature of 1 cm⁻¹, which is comparable to or better than other methods such as drop coating (280 s, 0.3 cm⁻¹)¹⁰ and scraper coating (3 s, 1.3 cm⁻¹)³³. For MXene actuators, our method achieves a response time of 30 s and a curvature of 3 cm⁻¹, similar to cast knife coating (25-45 s, 2.1–3.6 cm⁻¹)⁷.

Furthermore, to demonstrate the scalability of our pH-programmed anisotropic membrane preparation, a large GO membrane (10 cm × 12 cm) was fabricated by sequentially casting pH2 GO dispersion at the bottom, followed by pH10 GO at the top (Supplementary Fig. 14a, b). The thermal actuation performance of this large-area membrane (2.76 cm⁻¹ at 100 °C) was found to be comparable to that of membranes prepared via vacuum filtration (Supplementary Fig. 14c, d), highlighting the efficiency and competitiveness of our pH-programmed self-assembly method.

Obviously, the utility of the low-dimensional actuators extends beyond their traditional applications. These lightweight and robust materials hold significant potential as structural elements in diverse fields such as soft robotics, self-cleaning membranes, and medical manipulators. Here, we explored an application for these actuators, showing their versatility in architectural solutions and the design of smart housing. 2D actuators can adapt to different shapes, conform to irregular surfaces, and exhibit flexibility, making them suitable for complex architectural designs and flexible housing applications. Their 2D nature allows for their integration into lightweight and low-dimensional architectural elements.

Such actuators offer an approach to reimagining and incorporating circular windows in contemporary architectural designs. Also, we observe light patterns that such windows create in the interior. Figure 5d shows the flamelike opening of the circular window upon heating by 850 nm IR light with a power of 1 Wcm⁻². We observe the full opening within 8 s upon heating and closing within 1.5 min upon colling. The recovery time here is longer than the heat dissipation directly on the glass sheet in the previous section due to the substrate underneath also being heated and requiring time for heat dissipation. Furthermore, we assemble the windows at different pH to program the windows to respond to the airflow T changes depending on the times of the day. As we see in Fig. 5e, our windows show different sensitivity to the ambient temperature (~25 °C) depending on the pH value at which they are assembled. At night, in the absence of sunlight the assembled at pH 6 and 8 windows partially open while the assembled at pH 2-4 and 10 windows close. During the day (Fig. 5f), when the sun shines the assembled at pH 6 and 8 windows completely open; the assembled at pH 10 windows partially open, the assembled at pH 4 and 3 windows slightly open, and the assembled at pH 2 one still closes. We see that the pH programmed windows are highly sensitive to even tiny temperature fluctuations during the daytime.

We further explored the applicability of using this adaptive material in the construction of smart architecture. We prepared the building frame by 3D printing and selected traditional glass windows were replaced with smart windows that maintained adequate illumination. As shown in Fig. 5g, h, the smart windows were capable of automatically adjusting their opening and closing based on the ambient temperature, thereby controlling the circulation of indoor air. In addition, we simulated a scenario where people would close the windows during the day when they were outdoors and compared the indoor temperature change between the smart house and the traditional house during the day (Fig. 5i). In the traditional house, the indoor temperature increased by 10 °C in just 10 min under the sun and continued to rise. In contrast, the smart house gradually opens the adaptive windows to keep the indoor air circulation when the environment continues to warm up. In windy conditions, the opened smart windows accelerated indoor heat dissipation, resulting in a drop in indoor temperature, while the traditional house experienced continued warming up during such periods. Consequently, even with prolonged exposure to sunlight, the indoor temperature of the smart house remained relatively stable. Therefore, it is prospective to use lightweight low-dimensional 2D materials for smart and environmentally friendly housing with energy-saving climate-controlled capabilities.

Additionally, pH-programmed memebranes were utilized to design traditional actuator elements, including grippers and walkers. As shown in Fig. 6a, a GO strip can be used as a soft gripper to grab cargo. The 2D membrane stripes grip and lift a polymer cube after 18 s of IR light illumination, released after 7 s when the light is off (Supplementary Movie 1). The weight of our 2D gripper is 1.4 mg and the weight of the polymer cube is 14 mg which is ten times the gripper weight. IR temperature plots (Supplementary Fig. 15) show the initial state of the gripper at 0 s and the state of the cargo be lifted under IR light stimulation at 20 s, from which we can see that the T of the gripper has only increased by about 5 °C, indicating that the 2D membranes are very sensitive to T changes.

Fig. 6. Applications of pH-programmed actuators for soft robotics.

a Sequential images showing the pick-up and release of a foam object by a GO actuator strip (2 cm × 4 mm) under infrared (IR) light stimulation. b Schematic of a GO-based soft gripper with a cross-shaped design, illustrating the actuation-driven gripping of a foam object under sunlight. In the schematic, dark gray and brown regions represent uncharged and charged GO flakes, respectively. c Locomotion sequence of a soft walking robot fabricated from a GO membrane, activated by IR light.

Figure 6b demonstrates that even sunlight can actuate a 2D gripper with a structure, weighing 2.06 mg (Supplementary Fig. 16a and Supplementary Movie 2). The 2D shape ensures high stability of the gripper, allowing it to securely hold cargo during shaking and sharp turns (Supplementary Movie 2). The gripper reliably lifted up to 11 times its own weight (22.5 mg) within 10 s (Supplementary Fig. 16b).

Finally, we assembled a soft walking robot. IR pointer triggers the directional movement of the 2D walker. As shown in Fig. 6c, we put the 2D walker on the experimental table to measure the step length. The step length of the 5 μm-thick walker is about 4 mm (Supplementary Movie 3), providing a comparable crawl distance compared to existing GO-based actuators^{12,16,28,33,34}.

Discussion

The key to our approach lies in the assembly of chemically anisotropic multilayers by modulating the interaction of charged and uncharged flakes with water molecules. Both GO and MXene flakes acquire a net electric charge in response to changes in environmental pH. Upon heating, the charged groups lose water, leaving the edges in a metastable state. The presence of an electric field at the edges triggers in-plane sliding and reorganization of the 2D sheets. The folding/unfolding process, resulting in differences in surface areas between the top and bottom sides of the multilayers, arises from the in-plane motion of the flakes. The distinctive properties of 2D actuators offer several advantages over 3D materials in terms of sensing capabilities and response times. The high surface area-to-volume ratio and the absence of a bulk phase in 2D actuators allow for enhanced sensitivity to environmental changes. The multiple interfaces between the flakes provide additional sites for interaction with the molecular triggers, amplifying the sensitivity of the material to external stimuli. The absence of a bulk phase allows for faster diffusion of trigger molecules throughout the material, facilitating a swift response. This means that even small variations in temperature, pH, light intensity, or chemical composition can be detected and translated into a high amplitude change in shape. By harnessing the thin, flexible, and robust nature of the 2D multiplayer, we explore applications for our actuators such as breathable bionic architecture. The design of flamelike 2D windows that open and close in response to small changes in the ecosystem demonstrates the potential for energy-efficient climate control and innovative functional design in sustainable houses. The breathability of these architectural elements holds promise for the designs of sustainable housing, wearable technologies, and biomedical devices.

Methods

Materials

Aqueous GO dispersion (4 mg mL⁻¹, monolayer content >95%, Graphenea Inc., USA), polyethersulfon membrane filter (PES, pore size 0.03 µm, diameter 47 mm, Sterlitech Corporation, USA), hydrochloric acid (37%, ACS reagent, Sigma-Aldrich), ammonia solution (2.0 M in ethanol, ACS reagent, Sigma-Aldrich), sodium hydroxide (NaOH, ACS reagent, ≥98%, pellets, Sigma-Aldrich), potassium hydroxide (KOH, ACS reagent, ≥85%, pellets, Sigma-Aldrich), lithium hydroxide (LiOH, reagent grade, powder, ≥98%, Sigma-Aldrich), ascorbic acid (ACS reagent, Sigma-Aldrich), lithium fluoride (LiF, Sigma-Aldrich, BioUltra, ≥99.0%), Ti₃AlC₂ MAX powders (MAX, Tongrun Info Technology Co. Ltd, China), UV sensitive Resin (UV wavelength 405 nm) for 3D printing was purchased from ANYCUBIC, all materials were received and used without further purification.

Preparation of Ti₃C₂T_x nanosheets

Ti₃C₂T_x nanosheets were prepared according to ref. ²⁰. 3.0 g of LiF was added to 9.0 M HCl solution (40 ml). After the dissolution of LiF, 1.0 g of Ti₃AlC₂ MAX powder was slowly added into the HF-containing solution, and then the mixture was kept at 35 °C for 24 h. Afterward, the solid residue was washed with 2 M HCl three times. Then, the solid residue was further washed with deionized water several times until the pH value increased to 6. Subsequently, the washed residue was added into 150 mL of deionized water, ultrasonicated for 30 min under N₂ atmosphere, and centrifuged at 1713 g for 20 min. The supernatant was collected as the suspension of Ti₃C₂T_x MXene with a concentration of 5 mg mL⁻¹.

Measurement of the surface charge of GO and MXene flakes by potentiometric titration

The concentration of ionized groups on GO and MXene at various pH levels was determined through a pH titration method. The method is described in detail using GO as an example. Initially, 100 mg of GO was placed in a beaker containing 20 mL of 0.1 M NaOH solution. Subsequently, increments of 0.25 mL of 0.1 mol HCl solution were added, and the pH of the solution was recorded after reaching equilibrium at each step. This procedure was repeated with the same volume of NaOH but without GO. The discrepancy in the volumes of HCl between the two titration curves at the same pH value, denoted as ∆V, provides the concentration of ionized groups per gram of GO at that specific pH. The surface charge of GO can then be calculated using the Eq. 2 derived from ∆V³⁵:

$${\sigma }_s\left(\mathrm{pH}\right)=-F\frac{{\mathrm{\Delta }n}_{\mathrm{sol},\left(\mathrm{pH}\right)}}{m{A}_{\mathrm{sp}}}=-F\frac{c\mathrm{\Delta }V}{m{A}_{\mathrm{sp}}}$$ 2

where: ${\sigma }_s$ is surface charge (C m⁻²); F is Faraday constant (F = 96500 C mol⁻¹); ${\mathrm{\Delta }n}_{\mathrm{sol},\left(\mathrm{pH}\right)}$ is the concentration of ionized species (mol); m is the sample mass (g); $A_{sp}$ is the total surface area per one gram (m² g⁻¹).

Preparation of GO and MXene membranes at different pH

The original aqueous GO and MXene dispersion was added into deionized water to obtain a diluted dispersion (0.5 mg mL⁻¹). The diluted GO and MXene dispersion (0.5 mg mL⁻¹) had a natural pH of 3 and 6, respectively (measured by pH meter, SevenExcellence, Mettler-Toledo (S) Pte. Ltd.). To prepare GO and MXene membranes at different pHs, we first adjusted the pH of GO and MXene dispersions with hydrochloric acid and ammonia solutions, and then applied vacuum-assisted filtration to prepare the membranes. All vacuum filtrations were performed with 47 mm polyethersulfone (PES) membranes from STERLITECH (0.03 nm pore size).

Thermal response characterization

The vacuum-filtered GO and MXene membranes were dried at room temperature (RT) for 24 h and then removed from the PES filter for testing. The thermal response of the 2D actuators with a typical strip shape (length 20 mm and width 4 mm) was investigated quantitatively in the temperature (T) range from room temperature to 100 °C. To compare the actuating performance of different 2D actuators, the curvature analysis function of ImageJ is used to obtain the average curvature of the whole actuator strip by fitting the curved strip. The long-term stability test is conducted by keeping the GO strips in the laboratory at all times at about 25 °C and 60% humidity.

Preparation of GO breathable bionic architecture

We used computer-aided design software (Autodesk AutoCAD) and 3D printing (ANYCUBIC Photon Mono X) to prepare the building frame. The building’s overall dimensions are 13 cm × 6 cm × 6 cm. The rectangular GO smart windows were cut from the freestanding GO membrane with a diameter of 3.6 cm, to a size of 3 cm × 2.5 cm.

Characterization

The microstructure of the membranes was observed with a scanning electron microscope (SEM, Zeiss Sigma 300), membranes were ruptured in liquid nitrogen (for cross-sections) and sputtered with gold (~5 nm) for GO membranes before observation. The XRD patterns were obtained by a thin-film X-ray diffractometer (TF-XRD, Bruker D8 Advance). The zeta potential of GO flakes was measured by Zetasizer (ZSU5700, Malvern Panalytical). TGA was performed by TA Instrument Discovery TGA1-0247 under airflow at a heating rate of 10 °C min⁻¹. The chemical confirmation of membranes was performed by attenuated total reflection-Fourier transform infrared spectroscopy (ATR-FTIR, IRTracer-100, Shimadzu) in a range of 400–4000 cm⁻¹ with 4 cm⁻¹ resolution. The mechanical property measurements were conducted by dynamic mechanical analyzer (DMA 850, TA Instruments, Inc.) with membrane width of 5 mm and strain ramping of 1% min⁻¹. XPS was carried out using a scanning X-ray microprobe (Kratos Axis Ultra DLD) operating at 15 kV with monochromated Al Kα radiation (1486.71 eV). The binding energies were calibrated with reference to the sp3 C1s peak at 284.8 eV.

Computational methods

Spin-polarized density functional theory computations were carried out in the Vienna Ab initio Simulation Package. Perdew−Burke−Ernzerhof functional within the generalized gradient approximation was employed for the exchange-correlation potential³⁶. A kinetic energy cutoff of 400 eV was used for the plane wave expansion of valence electrons. The convergence criteria were set as 1.0 × 10⁻⁵ and 2.0 × 10⁻² eV Å⁻¹ for the energy and atomic force, respectively. 1 × 1 × 1 Gamma-centered mesh was used to sample the Brillouin zone. To eliminate the periodic interactions between the adjacent images, a vacuum space of 15 Å was used in the perpendicular direction of the GO edges. DFT-D3 method with the standard parameters by Grimme et al. was employed to describe the van der Waals interactions³⁷. To clarify the charge effect on the interaction of water with the different GO structures, a linearized Poisson–Boltzmann implicit solvation model was used as implemented in VASPsol³⁸. A dielectric constant of 80 for water and a Debye length of 3.0 Å were used to simulate 1 mol of electrolyte solution of monovalent cations.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

D.V.A. conceived, designed and supervised the study. Q.W. conducted the fabrication and analysis of 2D membranes. X.Y.G. performed the DFT calculations. Q.W., M.L., K.Y., M.S.C., S.C., and M.T. carried out the membrane characterizations. Q.W., X.Y.G., K.S.N., and D.V.A. interpreted the results and wrote the paper. All authors discussed the results.

Peer review

Peer review information

Nature Communications thanks SungWoo Nam, who co-reviewed with Soyeong Kwon, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data supporting the findings of this study are available within the article and its Supplementary Information. Additional raw data is available from the corresponding author. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-63800-9.