Direct in-situ observation of hydrogen-enhanced diffusive deformation at the crack tip in pure aluminium

Abstract

Reliable service of lightweight high-strength alloys in energy efficient structures is challenged by hydrogen embrittlement. Extensive research efforts have been devoted to uncovering the mechanism underlying the phenomenon for decades, primarily focusing on the interaction between hydrogen and dislocations. Nonetheless, the explicit role of hydrogen in the direct vicinity of a propagating crack tip, wherein the deformation carriers nucleate, remains obscure. Here, by using in-situ high-resolution environmental transmission electron microscopy, we directly capture hydrogen-enhanced diffusive deformation-mediated crack propagation in pure Al, manifesting vacancy formation and atomic diffusion on the crack tip surface. This is corroborated by density functional theory calculations revealing considerably weakened Al-Al bonds at the adsorption sites and lowered energy barriers for vacancy formation and adatom diffusion. This diffusive cracking pathway is applicable under high hydrogen fugacity and low strain rates, which imposes evident effects on bulk materials. The findings point to restraining atomic diffusion to alleviate the hydrogen-assisted cracking in advanced aluminum alloys.

The role of hydrogen at crack tips remains ambiguous. Here, the authors directly observe hydrogen-enhanced diffusive deformation at the atomic level and demonstrate the formation of vacancies as well as atomic diffusion, which contribute to crack propagation.

Introduction

The pledge of carbon neutrality has rendered an increasing demand for materials with high specific strength, such as aluminum, that can effectively reduce the weight of structures¹. However, high-strength materials are generally more sensitive to a failure mode known for decades as hydrogen embrittlement (HE)^2–4 which critically endangers the service reliability of structure alloys in energy-efficient vehicles, hydrogen storage tanks, etc.^5,6 Though the solubilities of H atoms in materials are intrinsically low, with its ultrasmall atom radii, H can readily diffuse and enrich at vulnerable sites such as crack tips or grain boundaries^7,8, leading to the premature fracture at stress/strain levels well below the normal limits of structure materials. Current counter strategies entail nanoprecipitates⁹ and grain boundary engineering¹⁰, which can be crippled by welding. An explicit mechanistic understanding of H-assisted cracking is still indispensable in resolving this issue.

It is generally accepted that H-assisted cracking roots in the complicated effects of H on dislocation activities^8,11 and H-enhanced decohesion of interfaces^7,12. However, peculiar deformation modes other than dislocations may be activated in the highly-strained vicinity of the crack tip, such as phase transformations, twinning, and atomic diffusion^4,13. In addition, driven by the strain concentration at the crack tip, H atoms can enrich at the crack tip and be significantly affected by the cracking process through interplay with the surface facets¹⁴, atomic diffusion¹⁵, dislocation nucleation, and phase transformations^4,16,17. As such, the actual mechanism involving H at the crack tip can be far beyond the simplified description of H-dislocation interactions.

Although progress has been made with atomic-scale direct visualization of the cracking processes, such as dislocation emission and crack-tip blunting^13,18–20, phase transformation and twinning^16,21,22, and cleavage fracture²³, obtaining comparable atomic-resolution evidence in hydrogen-charged Al remains challenging. Hydrogen readily escapes from the thin transmission electron microscopy specimens during testing, and the native oxide on Al easily forms and obscures crack-tip processes. There has been a breakthrough in visualizing hydrogen-initiated bubble nucleation beneath the oxide on Al¹⁵. Direct atomic-resolution imaging of hydrogen-assisted crack-tip processes in Al has been largely inaccessible.

Herein, by using an aberration-corrected environmental transmission electron microscope (ETEM), we were able to drive the H-charging process into the loaded crack tips in pure Al and thereby captured the H-assisted cracking process at atomic resolution in Al. The direct observations and quantitative analysis reveal that H prompts atomic diffusion at the crack tip and effectively attenuates the critical strain required for crack propagation. The enhanced atomic diffusion is further corroborated by density functional theory (DFT) calculations demonstrating that H considerably weakens the interatomic bonds in the vicinity of the crack tip.

Results and discussion

Figure 1a presents a schematic illustration of the in-situ ETEM setup used in this study. A 99.999 wt.% pure Al foil was thinned into electron-transparent regions for observation. Depending on whether hydrogen was introduced into the ETEM chamber, the experiments were conducted either under high vacuum or under a controlled hydrogen environment, allowing a direct comparison of crack-propagation behavior in the two conditions. Upon tensile straining, microcracks nucleated at the notch and subsequently propagated in the electron-transparent region (Fig. 1b), enabling real-time comparison of the crack-tip evolution in vacuum and hydrogen environments.

Fig. 1. Schematic illustration of the in-situ TEM setup for studying the atomic mechanism of hydrogen-induced cracking in aluminum.

a The experimental setup (σ denotes tensile stress, and E-beam is electron beam). b An atomic resolution image of a crack tip. Images from five independent experiments are shown, with similar results obtained in each experiment. Scale bar, 1 nm.

Atomic crack propagation mechanism in vacuum

Figure 2a–f and Supplementary Movie 1 show sequential high-resolution TEM images of a crack propagation in an Al foil in vacuum and corresponding strain evolution derived from geometrical phase analysis on the atomic resolution TEM images. Initially, the crack tip was bounded by {1 1 1} and {2 0 0} surface facets and encompassed 9 {1 1 1} layers at the tip (Fig. 2a). Upon straining, the elastic strain gradually increased at the crack tip, leading to crack propagation by one atomic layer and concomitant reduction of the crack opening (Fig. 2b). This process was apparently mediated by the formation and migration of surface steps at the crack tip (Supplementary Fig. 1). Accompanying this process, dislocation nucleated and climbed by inserting atom columns perpendicular to the loading direction (Fig. 2a–c), which effectively attenuated the elastic strain at the crack tip (Fig. 2d–f). Although the observable region shown in Fig. 2 is subject to geometric and boundary constraints that limit the plastic activity, slip events were occasionally observed to initiate from the crack tip during the experiment, as shown in Supplementary Fig. 2.

Fig. 2. Atomic-scale processes of crack propagation in Al in vacuum a–c, and H₂ environment g–i.

Representative images from three independent experiments are shown, with similar results obtained in each experiment. σ denotes the tensile stress, and T denotes the dislocation. d–f strain evolution corresponding to the regions shown in panels a-c. j–l strain evolution corresponding to the regions shown in panels (g–i). Scale bars in a–f 2 nm, g–l 1 nm.

H-enhanced diffusive deformation in hydrogen environment

By contrast, the crack propagation in an H₂ environment is predominantly mediated by atomic diffusion at the crack tip. Figure 2g–i and Supplementary Movie 2 show sequential atomic resolution snapshots of a crack propagation in Al in the H₂ environment. Figure 2g shows the initial crack profile (marked by the yellow dotted line), with a crack-tip width of six atomic layers. After 0.5 s of continuous tensile loading, five atomic layers disappeared from the crack tip, reducing the tip width to three layers, as delineated by the new crack profile (purple dotted line) in Fig. 2h. Upon further loading, an additional atomic layer was lost at the crack tip which then became passivated and expanded from three to four layers in width (see the blue dotted line in Fig. 2i). Similar step formation and migration events occur more frequently under hydrogen condition (Supplementary Fig. 3). No dislocation nucleation or motion is detected at the crack tip (Fig. 2g–i). Instead, atoms migrate forward along the surface through diffusion, reconstructing the crack tip from a sharp three-layer configuration into a blunted four-layer structure (Fig. 2h, i). Meanwhile, irregular and diffuse atomic contrast appears on both crack surfaces and evolves dynamically during crack advance. These observations collectively confirm that the crack propagation is governed by hydrogen-promoted atomic diffusion¹³ rather than cleavage or dislocation-mediated mechanisms. Further atomic-resolution observations reveal that atoms detached from the crack front migrate directionally toward the crack flanks (Supplementary Fig. 4 and Supplementary Movie 3). Vacancy formation and subsequent step migration at the crack tip surface have been clearly captured in Supplementary Figs. 5 and 6.

Theoretical analysis of hydrogen at the crack tip

Actually, the H-enhanced localized plasticity has been widely reported in literature, with extensive experimental evidence showing prompted dislocation density in front of the crack tip^24,25. By contrast, the roles of H in deformation at the direct vicinity of the crack tip, as observed here, are essentially different, featuring enhanced diffusive rather than displacive deformation. To unravel the origin of the enhanced diffusive deformation, we discussed the effect of hydrogen based on the Coble creep model²⁶ (Supplementary Fig. 7). Accordingly, the strain rate that can be mediated by atomic diffusion on boundaries and/or surfaces is proportional to ${\mathrm{e}}^{-\frac{Q_{\mathrm{f}}+Q_{\mathrm{m}}}{\mathrm{k}T}}$, wherein $Q_{\mathrm{f}}$ and $Q_{\mathrm{m}}$ are the energy expenses for the vacancy formation and adatom migration, respectively.

Therefore, DFT calculations are performed to pinpoint the most probable adsorption and solution sites nearby an Al {1 1 1} surface (see Fig. 3a). The calculations further show that the energy expenses for vacancy formation ($Q_{\mathrm{f}}$) nearby the H lodging sites are marginally lower than those without H. Though tensile strain may retard the surface adatom diffusion on the (1 1 1) facet²⁷, the DFT calculations indicate that the energy barrier for the Al adatom diffusion ($Q_{\mathrm{m}}$) reduces from 0.12 to 0.09 eV in the presence of H (Supplementary Fig. 8). Thereby, the strain rate originating from the H-enhanced diffusive deformation can be up to 10⁴ times higher than that of the H-free scenario (Supplementary Discussion 1). The hydrogen-enhanced vacancy formation at the crack tip is also evidenced by the micro-beam positron annihilation spectroscopy results, which indicate a vacancy concentration of 3.1 × 10⁻⁴ at the hydrogen-charged crack tip, approximately one order of magnitude higher than that under hydrogen-free conditions (3.5 × 10⁻⁵) (Supplementary Fig. 9 and Supplementary Discussion 2). The H-enhanced vacancy formation^28–30 and atomic diffusion offer a rational understanding of the prompted diffusive events as observed in the in-situ experiments (Fig. 2g–i).

Fig. 3. Density function theory calculations on the vacancy formation energy and Al-Al bond strength nearby the adsorbed/solution H atom.

a A schematic illustration of the Al {1 1 1} surface with H atoms adsorbed/soluted at various sites. The red and gray spheres represent H and Al atoms, respectively. b Vacancy formation energies at various sites (V1–V7) are shown in panel (a). c ICOHP values of the Al-Al bonds without (bonds 1–2) and with (bonds 3–8) the presence of H. Integrated crystal orbital Hamilton population (ICOHP) is adopted to quantitatively characterize the strength of bonding interactions between specific pairs of atoms.

From a fundamental perspective³¹, the enhanced atomic activities and cracking should essentially originate from the weakened interatomic bonds. By using the crystal orbital Hamilton population (COHP) analysis (see “Methods” for details), the effects of H on the strength of Al-Al bonds right next to the four probable lodging sites of H were studied (Fig. 3c). It is found that the H adsorption cuts down the bonding contributions between −10 eV and the Fermi level, and anti-bonding contributions appear between −2 eV and the Fermi level (Supplementary Fig. 10). As a result, the integrated COHP value of the in-surface Al-Al bonds drops by −19% to −45%, and that of the Al-Al bonds between the surface and sub-surface layers decreases by −26% to −65%. Consistently, as shown in Fig. 4a, the maximum elastic strain normal to the {1 1 1} plane during the crack propagation is ≈15% in the H-free case, which is roughly equal to the ideal tensile strain along the <1 1 1> direction of Al³²; in contrast, the maximum elastic strain at the tip of the crack is reduced to less than 10% when H is involved.

Fig. 4. The embrittlement effect of hydrogen-enhanced diffusive deformation.

a An evolution of maximum elastic strains at the crack tips is shown in Fig. 2. The yellow and blue data points are measured strain values in cases with and without H, respectively. The abscissa represents the time stamp of the in-situ TEM observation on the crack propagation process. The average value was calculated by selecting ten strain measurement points at the crack tip. These ten points are spatially separated pixels extracted from a single TEM micrograph via geometric phase analysis (GPA), and they represent spatial sampling rather than independent experimental replicates. Data are presented as mean values ± SD. b Correlation between the diffusion activation energy of alloying elements and the I_SSRT of the AlZnMgCu-X alloys (X stands for the additional alloying elements). The red line denotes the linear fitting curve of the data points, with the fitting equation of y = −0.16% kJ⁻¹ mol x + 80.45% and a coefficient of determination R² of 0.76. Source data are provided as a Source Data file.

Comparison with prevalent HE mechanisms

Prevalent theories of HE include: (1) hydrogen-enhanced decohesion (HEDE)^4,8,11,12; (2) hydrogen-enhanced localized plasticity (HELP)^{8,11,17,33,34}, which highlights the interaction between hydrogen and dislocations; and (3) hydrogen-enhanced strain-induced vacancy (HESIV)^8,35, wherein hydrogen stabilizes vacancies and hence prompts fracture. Distinct from the above-mentioned pathways, our observations indicate that hydrogen facilitates the formation of surface vacancies and promotes atomic diffusion at the crack tip, and thereby drives the crack propagation. Given sufficiently low strain rates and high hydrogen fugacity, these processes could suppress dislocation activity at the crack tip, setting it apart from the classical HELP and HESIV mechanisms.

Note that the hydrogen-enhanced diffusive deformation identified in our study is distinct from the HEDE mechanism, which is a brittle cleavage model essentially originating from the hydrogen-induced surface-energy reduction³⁶. In contrast, it critically involves vacancy formation and step migration on the crack tip surface (see Fig. 2, Supplementary Figs. 1 and 3–6). Referring to previous studies¹³, such a step-mediated atomic relocation is essentially a diffusion process, driven by surface curvature and tensile stress. From a mechanistic perspective, the reduction in surface energy originates from hydrogen-induced weakening of interatomic bonds, which in turn lowers the energy barriers for vacancy formation and atomic diffusion. Therefore, the enhanced atomic diffusion here is intrinsically linked to the same bond-weakening effect underlying surface-energy reduction, suggesting that these processes are closely coupled rather than independent. In fact, there have been theoretical studies showing enhanced atomic diffusion by reducing surface energies³⁷. The fact that the reduced surface energy did not induce cleavage fracture in this case may be attributed to the low strain rate that allows the diffusion to carry the deformation at the crack tip.

The newly formed crack surfaces are expected to undergo surface relaxation due to hydrogen-induced weakening of Al-Al bonds, which may also facilitate surface vacancy formation and atomic diffusion³⁸. Surface relaxation and hydrogen-enhanced atomic diffusion are thus closely related through their common origin, but differ in their characteristic spatial scales and roles. Surface relaxation is generally regarded as a localized structural adjustment that largely preserves surface periodicity³⁹, contributing primarily to local structural stabilization and reduction of surface energy. In contrast, the long-range and directional atomic migration observed along the crack surfaces requires sustained atomic transport driven by chemical potential gradients associated primarily with stress concentration and surface curvature at the crack tip⁴⁰. The observed phenomenon at the crack tip, therefore, reflects the coupled occurrence of these two processes, which share a common physical origin but play distinct and complementary roles.

The hydrogen-induced superabundant vacancy formation (SAV)⁴¹ offers a theoretical basis for the mechanism here. By reducing the vacancy formation energy and atomic migration barriers, hydrogen creates a vacancy-rich diffusion-active environment that is further intensified by hydrogen enrichment at the crack tip. Extensive atomistic simulations have shown that hydrogen suppresses dislocation emission⁴ at crack tips and shifts fracture behavior toward a semi-brittle regime where cleavage competes with dislocation activities⁴². Other than cleavage, crack advance may also be directly mediated by atomic dissolution^43,44. Given that adsorbed hydrogen atoms can dislocate the surface atoms, the dissolution-driven crack propagation scenarios share obvious similarities with our in-situ TEM observations. Featuring vacancy formation and atomic diffusion, our findings show a hydrogen-assisted cracking mode mediated by localized diffusive activities of surface atoms at the crack tip.

Applicability analysis and alleviating HE

Since the observed H-enhanced diffusive deformation is localized at the crack tip, the phenomenon may cooperate with many other mechanisms of H-assisted cracking involving H-dislocation interactions far from the crack tip^8,11. Certainly, the predominance of the observed mechanism is conditional, favorable at high hydrogen fugacity and ensuing high vacancy concentrations at the crack tip. Local hydrogen enrichment substantially lowers the critical energy and increases the events of vacancy formation, enabling atoms at the crack front to detach from and migrate along the crack tip surface. Moreover, the kinetic process should only prevail when the crack propagation or strain rate is sufficiently slow; at high strain rates and low hydrogen fugacity (e.g., when the electron-beam intensity is relatively low), deformation near the crack tip is dominated by dislocation activities.

It should be noted that the present experimental configuration, featuring thin-foil samples, quasi-static loading, and elevated local hydrogen fugacity, favors the predominance of the observed diffusion-mediated deformation. Nevertheless, stress-relaxation experiments on bulk samples demonstrate that hydrogen-enhanced diffusive deformation is also operative in bulk Al (Supplementary Fig. 11 and Supplementary Discussion 3). It is expected to contribute locally within the highly stressed crack-tip region, where stress triaxiality promotes vacancy accumulation and hydrogen enrichment by reducing the hydrogen solution energy^45,46.

The mechanism suggests that increasing the energy barriers for atomic diffusion may alleviate the HE in Al. To validate this, we deliberately cast 14 different Al alloys by alloying various elements into an Al-7Zn-2Mg-1Cu (alloy composition, wt.%) high-strength baseline alloy. Then, the HE susceptibilities of these alloys are evaluated by using low-strain-rate tensile tests in a dynamic electrochemical H-charging condition (see “Methods”). Results show that, with increasing diffusion activation energies of the alloying elements, the HE susceptibilities of the Al alloys drop (see Fig. 4b and Supplementary Table 1). Consistently, our recent machine learning results on a much larger data set (containing over 200 different Al alloys) show that the stress corrosion cracking susceptibility of aluminum alloys critically depends on the diffusion activation energies of the alloying elements⁴⁷. Since the effects of H in prompting vacancy formation and adatom diffusion exist in many other metals, especially those with strong electronic interaction with H^48–50, the H-enhanced diffusive deformation is likely not unique to crack tips in Al, offering potentially effective guidance to the alloy design of durable lightweight alloys for hydrogen-involving applications.

Methods

Materials

Pure Al (99.999 wt.%) foil was used for this study. As a key structural material with high specific strength, formability, and recyclability for automotive and aerospace industries, Al alloys suffer from the HE-induced premature failure and catastrophic accidents. Subsequent electrochemical hydrogen charging solutions were prepared using NaCl (purity ≥ 99.5 wt.%) and K₄P₂O₇ (purity ≥ 99.5 wt.%) reagents.

TEM sample preparation

The Al was cut into plates of 0.5 mm in thickness, then mechanically polished with SiO₂ suspension to ≈ 100 μm in thickness, and then milled with Argon ion to electron transparency, as shown in Supplementary Fig. 12. The size information of the TEM sample can be found in the Supplementary Fig. 13.

In-situ TEM experiment procedure

The Thermo Fisher Titan ETEM, equipped with imaging lens spherical aberration correctors, was used for the in-situ TEM tests. A Gatan 654 tensile holder was used to apply strain to the sample. As shown in Supplementary Fig. 12, a hole was milled at the center of the thin foil sample, wherein the sample thickness at the perimeter of the hole is sufficiently small to allow the electron beam to penetrate. As such, upon applying uniaxial tensile loading, a crack was initiated at such thin regions at the perimeter of the hole and subsequently propagated along the radial direction of the hole. The images and videos were recorded with a charge-coupled device at 2 frames per second. The approximate area available for the present high-resolution TEM observation can be found in Supplementary Fig. 14.

Geometric phase analysis (GPA) was performed using the GPA plugin implemented in DigitalMicrograph software. Based on the resulting strain maps, a local region encompassing 3 × 3 atomic sites at the crack tip, corresponding to the maximum strain concentration, was selected, and the average strain within this region was used to characterize the crack-tip strain state.

For experiments performed in hydrogen, gaseous H₂ was introduced into the sample chamber to maintain a partial pressure of ≈ 0.1 Pa. Under the energized electron beam, H₂ molecules can dissociate into atomic hydrogen, which has a low barrier for entering the Al lattice⁵¹. The calculated local hydrogen fugacity at the crack tip reached 8.1 × 10²−5.6 × 10⁴ Pa, ≈ 4–6 orders of magnitude higher than the chamber hydrogen pressure (Supplementary Discussion 4).

DFT calculations

The calculations were performed in the Vienna ab initio simulation package⁵². The core electrons were modeled with the projector-augmented-wave method⁵³. The exchange and correlation function of Perdew-Burke-Ernzerhof was adopted within the framework of the generalized gradient approximation⁵⁴. The plane-wave basis kinetic energy cut-off was set to 400 eV. The Brillouin zones of Al (1 1 1) surface calculations were sampled with a 3 × 3 × 1 Monkhorst-Pack k-point grid. During the geometrical optimizations, the atoms were allowed to fully relax until the forces on each atom were less than 0.02 eV Å⁻¹. The total energy was converged to 10⁻⁶ eV for the self-consistent field calculation. All structures in our calculations are based on an idealized environment without external stress. The supercell of Al(111) surface features a lattice defined by vectors of 17.1, 17.1, and 30.0 Å along the x, y, and z directions, respectively. This structure contains 216 Al atoms arranged over multiple layers, with a large vacuum region of 15 Å in the z-direction to simulate the open environment of the Al (111) surface. The stepped Al(111) surface was constructed by overlaying a single layer of Al atoms, covering half the area of the entire atomic plane, on top of six layers of Al atoms. The supercell of the stepped Al surface is defined by lattice vectors of 19.8, 11.4, and 30.0 Å along the x, y, and z directions, respectively. This structure contains a total of 208 Al atoms and includes a large vacuum region of 15 Å in the z-direction to simulate the open environment of the surface. The formation energy of vacancy was defined by Eq. (1):

$$E_{\mathrm{f}}=E_{\mathrm{v}}+E_{\mathrm{Al}}-E_{\mathrm{total}}$$ 1

Where E_v and E_total are the energies of supercells with or without an Al vacancy, E_Al is the energy of per atom in the Al bulk. The stable adsorption site for the Al atom on the Al (1 1 1) surface was obtained. Then, the climbing-image nudged elastic band (CI-NEB) method⁵⁵ was utilized to identify the most energy-efficient pathway for the Al atom to migrate and to calculate the transition barriers for Al atom diffusion on the Al (1 1 1) surface, both with and without a H atom adsorbed on the Al atom. COHP partitions the band-structure energy into orbital-pair interactions, which is usually plotted alongside the DOS (which just shows where electrons are, but nothing about their bonding character). Otherwise, the integrated COHP hints towards the bond strength. The COHP curves and ICOHPs were analyzed with the LOBSTER software⁵⁶.

HE susceptibility evaluation

Low-strain-rate tensile tests in both dynamic H-charging conditions and dry air were carried out to evaluate the HE susceptibility (I_SSRT) of each alloy. The strain rates were all 1 × 10⁻⁶ s⁻¹. The electrolyte for electrochemical H-charging is 35 g L⁻¹ NaCl + 3 g L⁻¹ K₄P₂O₇. A constant current density of 0.05 A cm⁻² in the H-charging. The HE susceptibility is defined by Eq. (2):

$$I_{\mathrm{SSRT}}=1-\frac{{\sigma }_{\mathrm{UTS}}^{\mathrm{H}}\cdot \left(1+{\epsilon }_{\mathrm{f}}^{\mathrm{H}}\right)}{{\sigma }_{\mathrm{UTS}}^{\mathrm{air}}\cdot \left(1+{\epsilon }_{\mathrm{f}}^{\mathrm{air}}\right)}$$ 2

wherein, ${\sigma }_{\mathrm{UTS}}^{\mathrm{H}}$ and ${\epsilon }_{\mathrm{f}}^{\mathrm{H}}$ are the ultimate tensile strength and elongation of the alloy in the dynamic H-charging condition, respectively; ${\sigma }_{\mathrm{UTS}}^{\mathrm{air}}$ and ${\epsilon }_{\mathrm{f}}^{\mathrm{air}}$ are those of the alloy in dry air. To validate the critical role of atomic diffusion in the HE process, 13 alloying elements (namely Mn, Cr, Zr, Ti, Fe, Si, Co, Ni, Ag, Sn, Li, V, Sc) with different diffusion activation energies were separately added into the same Al-7Zn-2Mg-1Cu (alloy composition, wt.%) baseline alloy by the same amount of 0.1 wt.%. Together with the baseline alloy, these 14 alloys were homogenized in a resistance furnace at 400 °C for 24 h and then at 470 °C for 24 h. The homogenized alloys were then hot-rolled at 400 °C from 15 mm in thickness to 2.5 mm. Specimens for the low-strain-rate tensile tests were directly machined from the hot-rolled plates, subsequently homogenized (at 450 °C for 1 h, then 470 °C for 1 h, then 48o °C for 0.5 h), quenched in water, and finally annealed at 120 °C for 24 h.

Micro-beam positron annihilation spectroscopy measurement

Micro-beam positron annihilation spectroscopy is performed using a new digital positron annihilation lifetime spectrometer developed by the University of Science and Technology of China (see Supplementary Fig. 9a)⁵⁷. Its time resolution is ≈190 ps. High-purity aluminum (99.99 wt.%) was used for the study. The hydrogen charging solution consisted of 35 g L⁻¹ NaCl + 3 g L⁻¹ K₄P₂O₇, with a constant charging current density of 0.15 A cm⁻². The specimens were pre-charged with hydrogen for 6 h before loading, and the hydrogen-charged state was continuously maintained during the subsequent stress relaxation process. Pre-cracks were introduced via tensile loading at a displacement rate of 0.001 mm s⁻¹. Crack initiation was identified by the onset of stress drop in the load-displacement curve, where the loading was terminated. The sample counting rate during positron annihilation spectroscopy measurement is 5.8 × 10⁶. Two specimens were tested for each condition, and the average results from the two specimens were used to compare. After the measurement, all lifetime spectra were analyzed using the LTv9 program based on the Gauss-Newton nonlinear fitting algorithm.

Stress-relaxation experiments

High-purity (99.99 wt.%) aluminum was used as the raw material to conduct stress relaxation experiments. Stress-relaxation experiments were performed in air (hydrogen-free) and electrochemical hydrogen charging (hydrogen-containing) conditions, as shown in Supplementary Fig. 11a. The hydrogen charging solution is 35 g L⁻¹ NaCl + 3 g L⁻¹ K₄P₂O₇, with a constant charging current density of 0.15 A cm⁻². The specimens were pre-charged with hydrogen for 6 h before loading, and the hydrogen-charged state was continuously maintained during the subsequent stress relaxation process.

Reporting summary

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

Author contributions

L.Q., J.X., and L.W. created the experimental protocols. L.J., X.Y., and B.Z. Synthesized the TEM sample and performed the TEM experiment and data analysis under the direction of Y.H. Y.M. performed the atomistic simulations. L.J. X.W., and Y.H. drafted the manuscript under the supervision of J.X. and L.Q. All authors contributed to the discussion of results and revision of the manuscript. These authors contributed equally: L.J. and Y.M.

Peer review

Peer review information

Nature Communications thanks Derek Warner and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Natural Science Foundation of China (No. 52271049, 92570301, 52541105, 52401002, 52301073).

Data availability

The data that support the findings of this study are available from the corresponding authors upon request. 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 https://doi.org/10.1038/s41467-026-77328-z.