Quantitative tests revealing hydrogen-enhanced dislocation motion in α-iron

Hydrogen embrittlement jeopardizes the use of high-strength steels in critical load-bearing applications. However, uncertainty regarding how hydrogen affects dislocation motion, owing to the lack of quantitative experimental evidence, hinders our understanding of hydrogen embrittlement. Here, by studying the well-controlled, cyclic, bow-out motions of individual screw dislocations in α-iron, we find that the critical stress for initiating dislocation motion in a 2 Pa electron-beam-excited H2 atmosphere is 27–43% lower than that in a vacuum environment, proving that hydrogen enhances screw dislocation motion. Moreover, we find that aside from vacuum degassing, cyclic loading and unloading facilitates the de-trapping of hydrogen, allowing the dislocation to regain its hydrogen-free behaviour. These findings at the individual dislocation level can inform hydrogen embrittlement modelling and guide the design of hydrogen-resistant steels. Screw dislocations in α-iron move more easily in the presence of hydrogen, as evidenced by real-time imaging using quantitative transmission electron microscopy.

Hydrogen embrittlement jeopardizes the use of high-strength steels in critical load-bearing applications. However, uncertainty regarding how hydrogen affects dislocation motion, owing to the lack of quantitative experimental evidence, hinders our understanding of hydrogen embrittlement. Here, by studying the well-controlled, cyclic, bow-out motions of individual screw dislocations in α-iron, we find that the critical stress for initiating dislocation motion in a 2 Pa electron-beam-excited H 2 atmosphere is 27-43% lower than that in a vacuum environment, proving that hydrogen enhances screw dislocation motion. Moreover, we find that aside from vacuum degassing, cyclic loading and unloading facilitates the de-trapping of hydrogen, allowing the dislocation to regain its hydrogen-free behaviour. These findings at the individual dislocation level can inform hydrogen embrittlement modelling and guide the design of hydrogen-resistant steels.
In a hydrogen-based economy, the use of steels, which consist mainly of iron and often serve vital roles in power plants, vehicles, buildings or critical infrastructures, is at risk owing to hydrogen embrittlement-a sudden and often catastrophic deterioration of the material's load-bearing capacity. Although multiple hydrogen embrittlement mechanisms [1][2][3][4][5] have been proposed, a consensus has yet to be reached on the effects of these embrittlement modes. Also, direct experimental proof of the influence of hydrogen at the single dislocation level is missing. Challenges arise from the formidable complexity of the hydrogen embrittlement phenomenon, owing to the often intricate interplay of several mechanisms. Consequently, some reported macroscopic observations are contradictory. At the macroscale, hydrogen has been shown to cause both hardening and softening in pure iron 6,7 , while in regions below the fracture surface, hydrogen has been found to either enhance 8,9 or reduce 10,11 plastic activity. At the micro-and nanoscales, the main debate revolves around the effect of hydrogen on the energetics and kinetics of dislocations [12][13][14][15][16][17] . According to the classic Cottrell and Snoek theories, diffusible interstitial atoms including hydrogen are expected to form an atmosphere around the dislocation, causing a drag force acting against dislocation motion 18,19 . Interestingly, Birnbaum, Robertson and coworkers observed the opposite, namely, the mobility of dislocations was increased in several metallic materials when exposed to hydrogen [20][21][22][23][24] . They attributed this to elastic shielding of the dislocation's stress field by the hydrogen atmosphere that was assumed to surround the dislocation core. The hydrogen-enhanced dislocation mobility was proposed as the 'root cause' of hydrogen embrittlement, serving as core evidence for the hydrogen-enhanced localized plasticity (HELP) mechanism 2 . However, these results have been challenged by recent atomistic simulations, which did not reproduce hydrogen-enhanced dislocation mobility but instead showed a hydrogen-induced drag effect in aluminium 25 and α-iron 26 . On the experimental side, we confirmed some of these theoretical hypotheses, Article https://doi.org/10.1038/s41563-023-01537-w applied to all cycles, resulting in a normal stress oscillation between the valley stress (σ min ) of ~21 MPa and the peak stress (σ max ) of ~211 MPa. The bow-out motion was stable, fully reversible and reproducible during the first test session in vacuum ( Supplementary Fig. 4a,b). Subsequently, we exposed the sample to ~2 Pa H 2 under electron illumination for about 2 h. Previous experiments have shown that when the 2 Pa H 2 is excited by the high-voltage electron beam, the equivalent fugacity can reach hundreds of megapascals 33 . Using Sieverts' law 34 and 500 MPa hydrogen fugacity, the equilibrium lattice concentration of hydrogen in iron was estimated to be ~2 atomic ppm. Although the concentration of hydrogen at the dislocation core could be much higher, the dislocation configuration was observed to remain unchanged throughout the hydrogenation process ( Supplementary Fig. 4b,c).
After hydrogenation, we applied the same load-time function for the ensuing load cycles (renumbered N = 1, 2, …, 12) (Supplementary Video 2). Before we compared the movements of dislocation 1 without and with hydrogen exposure, we confirmed again that the bow-out motion was stable and fully reversible and the microstructure inside the pillar remained unchanged after a total of 12 loading cycles under 2 Pa H 2 atmosphere, as shown in Supplementary Fig. 4c,d, ensuring a reliable comparison. Then we observed that in 2 Pa H 2 , the bow-out movement of dislocation 1 showed an increased amplitude, as indicated by comparison of its configuration at σ max in vacuum and in the hydrogenated state (Fig. 2b). To further quantify the hydrogenation-induced change in dislocation motion, we performed a frame-by-frame correlation between loading time or stress and the position of dislocation 1 in the video recorded during the whole in situ test. In each cycle, the movement of dislocation 1 became activated only when the applied stress was increased above a critical value referred to as the dislocation activation namely, a hydrogen-induced pinning effect of dislocations 25 , for the case of aluminium, which has a face-centred-cubic structure and contains narrow stacking faults. For metals with a body-centred-cubic (bcc) structure, such as α-iron, tungsten and so on, the motion of screw dislocation via kink-pair formation and propagation is the rate-controlling process of deformation at room temperature 27,28 . However, direct experimental observations that clearly reveal the hydrogen effect on screw dislocation motion at the single-defect scale are still lacking despite earlier models and atomistic simulations 15,16,29,30 . In the classic works by Birnbaum, Robertson and co-workers, their experimental set-up was designed to change the hydrogen atmosphere during or after dislocation glide in an environmental transmission electron microscope (ETEM) [20][21][22][23][24] . However, there are a few unclear issues about the experimental set-up they used for revealing the hydrogen effect on dislocation mobility. First, before hydrogen introduction, the observed dislocations were usually at rest [21][22][23] , thus making it impossible to observe any hydrogen-related drag or pinning effects. Second, when thousands of pascals of H 2 gas were flooded into or pumped out of the ETEM chamber within a few seconds, this might have altered either the local distribution of internal stresses or the external load, which had been assumed to remain constant. These effects might have triggered dislocation motion due to their high sensitivity to stress. Finally, as the dislocations move by dragging pinning points on the foil surfaces 20,23,24 , the velocity of a dislocation would be largely determined by the pinning effect at both dislocation ends. However, dislocation pinning can be altered by possible hydrogen-induced reduction of the surface oxides or removal of organic contaminants from the surface under electron beam illumination 31 . These considerations fuel our motivation to perform well-controlled experiments that are capable of reducing the impact of these factors substantially, particularly related to uncontrolled surface pinning of dislocations.
In this Article, we designed a fully quantitative in situ ETEM mechanical testing protocol, as illustrated in Fig. 1, that enables the observation and comparison of bow-out motion of the same dislocation segments under a vacuum and in a hydrogen-containing environment. To avoid the thin-foil problems described above, we prepared submicrometre-sized free-standing cylindrical α-iron pillars, which were subjected to cyclic stresses with the aim to remove most of the pre-existing dislocations (referred to as the mechanical annealing effect 32 ) while retaining a few individually isolated screw dislocations that showed well-controlled bow-out motion between the pinning points ( In the bow-out motion, the dislocation moves with increasing applied stress by gradually changing from a straight to a bowed-out configuration, as shown in Fig. 1c. Upon unloading, the dislocation segment springs back to its original equilibrium shape and position, indicating that the bow-out motion is reversible and no new pinning in the temporarily sheared region had taken place. This controlled dislocation motion was exactly repeated in a series of testing cycles by maintaining a constant low magnitude of cyclic stress (locally depending on bow-out stress), thus allowing for quantifying the variation in dislocation responses, for example, the stress required for activating the movement of each individual dislocation and its glide distance, upon varied environmental conditions in a highly controlled manner, as shown in Fig. 1d. Compared with previous experiments, our experimental set-up reduces the influence of the possible variation of surface and boundary conditions and inhomogeneous stresses on dislocation motion. Article https://doi.org/10.1038/s41563-023-01537-w stress (σ c ). Then the dislocation continued to glide, until it reached the maximum projected bow-out displacement (δ max ) at the peak stress σ max , as shown in Fig. 2c. The measured δ max and σ c in each cycle (Fig. 2d,e) confirmed that the bow-out of dislocation 1 after hydrogenation can reach a larger distance than that under vacuum, and the average δ max increased by ~65% from 10.8 ± 0.6 nm in vacuum to 17 . 6). Therefore, our experimental results prove that the screw dislocation moves easier (with 27-28% lower stress and 64-65% larger bow-out distance under the same applied load) after exposure to 2 Pa H 2 excited by an electron beam. Evidently, hydrogen substantially enhances the screw dislocation motion in α-iron.
In addition, we noticed that for the reference experiments described above, if the dislocation completely moves out of the hydrogen atmosphere, it is expected to move with a δ max value close to that in vacuum during its ensuing motion. The observed increase of δ max in almost all cycles after hydrogenation thus indicates that the hydrogen atmosphere moves together with the dislocation during its bow-out motion. Therefore, the enhanced dislocation mobility should originate from direct interaction between the dislocation and the hydrogen atoms whose diffusion matches the dislocation velocity.
It is worth noting that our experiments prove the phenomenon of hydrogen-enhanced screw dislocation motion only for the case of low hydrogen concentrations. With increasing hydrogen concentration, it is possible that the hydrogen-drag effect might become important 16 . Indeed, our previous experimental work has shown hydrogen-inhibited dislocation motion in face-centred-cubic aluminium 25 , due to hydrogenated vacancies rather than interstitial hydrogen atoms. Such hydrogenated vacancies, of which the formation could be enhanced by hydrogen during plastic deformation 4 , have also been reported to impose large resistance to motion of dislocations in bcc metals 5,35 . These experimental results and simulations indicate that the hydrogen-dislocation interaction varies with the material, dislocation type, hydrogen and vacancy concentrations. In our study, the drag effect caused by hydrogenated vacancies is not dominant in the α-iron pillars. This implies that superabundant vacancies had not The tests in 2 Pa H 2 were started after the pillar had been exposed to the 2 Pa H 2 atmosphere for ~2 h. Scale bars, 100 nm.
Article https://doi.org/10.1038/s41563-023-01537-w formed under our experimental conditions (for example, in the early stage of plastic deformation). Next, we studied whether the dislocation response could be reversed after switching the hydrogen environment back to vacuum. In the loading cycles in 2 Pa H 2 that followed N = 12, the configuration of dislocation 1 changed due to an accidental depinning event. Hence, we had to increase the stress amplitude of the following loading session, to introduce another similar dislocation (marked as 2 in Fig. 3a). This dislocation showed stable bow-out motion (Supplementary Video 5) under cyclic loading between σ min (~35 MPa) and σ max (~353 MPa). The stress-induced change in shapes and positions of dislocation 2 in cycles before and after degassing are shown in Fig. 3b-d. Throughout all the 23 load cycles during the testing session immediately before degassing, the stress-driven dislocation motion was almost the same, with the average δ max = 6.2 ± 0.5 nm and σ c = 182.3 ± 20.8 MPa. Then, the hydrogen inlet was closed. Within only 2 min, the specimen chamber of the ETEM was pumped to high vacuum (<5 × 10 −4 Pa). In the following 3 h, the sample stayed under vacuum for further degassing. Such a long degassing time was sufficient for hydrogen in the sample to reach a new equilibrium distribution. After degassing, we applied the same cyclic load. In the first cycle, the dislocation moved to almost the same δ max as that in the cycles before degassing. But in the ensuing 8 load cycles, the average δ max of dislocation 2 gradually decreased, from 6.2 nm to 1.8 nm (Fig. 3b-f and Supplementary Video 6). Meanwhile, the measured σ c after degassing remained at the first reload cycle and then gradually increased until reaching a plateau value of 321.2 ± 7.1 MPa after the 9th cycle, showing an ~43% lower activation stress σ c in the hydrogenated state than the hydrogen-free state (Fig. 3e,f). These results show the reversible nature of the hydrogen-dislocation interaction, consistent with the recovery of conventional plastic flow observed in bulk metals with low hydrogen concentrations after degassing 6 .
We note that the recovery of the dislocation response after switching back to vacuum was unexpectedly slow, as even after ~3 h of degassing in vacuum, it still took ~20 s, or ~10 load cycles, for dislocation 2 to gradually recover its stable cyclic motion under vacuum. With the high diffusivity of hydrogen in α-iron (D H , of the order of 10 −9 to 10 −8 m 2 s −1 at room temperature 36 ), the time for hydrogen to diffuse out of the pillar was estimated to be less than 1 s according to λ 2 /4D H (with λ ≈ 1,000 nm being the characteristic length scale). This means that the time lapse (2 s) during each individual load cycle was sufficient for hydrogen to redistribute and reach a new equilibrium distribution inside the pillar. Given that the full restoration of dislocation behaviour after switching from hydrogen back to vacuum requires multiple load cycles, hydrogen traps should play a critical role in the overall recovery kinetics. Even after hours of ageing in vacuum, the traps could still retain a certain amount of hydrogen, which was released only slowly during the ensuing load cycles, so that its effect on dislocation motion remained active until the hydrogen stored in these traps was gradually decreased below a critical level. Considering that no strong hydrogen traps, such as grain boundaries or precipitates, were present in our Article https://doi.org/10.1038/s41563-023-01537-w sample, the delayed restoration of the dislocation behaviour should originate from other traps, such as dislocations and vacancies 37,38 . The fact that the hydrogen-induced enhancement in dislocation motion gradually diminishes with successive oscillating motion of dislocations suggests that cyclic loading and unloading facilitates degassing of hydrogen from those weak hydrogen traps, allowing the dislocation to regain its hydrogen-free behaviour. A detailed discussion of this mechanically-assisted degassing effect is provided in Supplementary Note 4. Our experimental results on hydrogen-enhanced dislocation motion reveal two major effects, that is, the decrease of activation stress and the increase of bow-out displacement. The former is obtained from direct measurement of a stress corresponding to the initiation of motion from a nearly straight screw dislocation, which is usually controlled by the nucleation and propagation of kink pairs. The latter is from direct ETEM observation and can be dictated by a competition between the hydrogen-affected dislocation line tension and kink-pair activity. To investigate the underlying mechanisms of hydrogen-induced reduction in activation stress, we performed atomistic simulations using the nudged elastic band (NEB) method 39,40 . A series of NEB calculations allow us to determine the minimum energy paths (MEPs) and associated activation energies of dislocation motion with and without hydrogen atoms at the dislocation core, respectively. The NEB results in Fig. 4a reveal the representative MEPs of screw dislocation glide under a resolved shear stress τ yz = 24 MPa (Supplementary Note 5). Similar to the previous NEB results in bcc metals without hydrogen 41,42 , it is the correlated kink nucleation that controls the activation process of screw dislocation glide. In particular, the hydrogen atom provides a favourable site for kink nucleation (Fig. 4b,c), leading to a lower energy barrier (from energy barrier without hydrogen E = 0.47 eV to energy barrier with hydrogen E H = 0.42 eV) for screw dislocation glide. In addition, as shown in Supplementary Fig. 7, the decrease of energy barrier (ΔE = E − E H = 0.05 eV) is not sensitive to the range of resolved shear stresses applied (τ yz = 0-58 MPa).
We further evaluated the binding energy difference of hydrogen at various positions along the core of a screw dislocation line via tracking the intermediate atomic configurations of the kink nucleation process around the hydrogen atom, as shown in Fig. 4f,g. The interaction energy between Fe and H is mainly governed by the embedding energy, which is a function of local electron density at the interstitial site 43 , that is, a polyhedron consisting of six nearest iron atoms surrounding a hydrogen atom. Our atomistic calculations (Supplementary Note 6) show that the normalized electron density (ρ H = 45.4) surrounding a hydrogen atom arising from six nearest iron atoms in the initial configuration is larger than that (ρ H = 44.7) in the saddle-point configuration. Consequently, the embedding energy function F H decreases approximately from 1.39 eV to 1.32 eV with decreasing ρ H . This results in the reduction of kink-pair nucleation barrier. In other words, the hydrogen atoms promote kink nucleation and decrease the critical stress of dislocation glide. This effect is also supported by molecular dynamics simulations. Article https://doi.org/10.1038/s41563-023-01537-w In Fig. 4h, we compare the molecular-dynamics-simulated stress-strain curves of screw dislocation glide with and without hydrogen at room temperature. With randomly distributed hydrogen atoms around the dislocation core (0.4 hydrogen atom per nanometre), the critical shear stress of dislocation glide decreases from 32 MPa to 22 MPa. This molecular dynamics result agrees qualitatively with our experimental measurements, namely, the activation shear stress (τ c ) of dislocation 1 reduces from 49.4 MPa to 35.6 MPa, as calculated with the Schmid factor of the most favoured {112}<111> slip system. In non-hydride-forming metals, the HELP mechanism is often considered as a key factor contributing to hydrogen embrittlement, and it is supported by metallographic features stemming from enhanced dislocation activity beneath either intergranular or transgranular fracture surfaces in iron and steels 8,44 . One critical element of the HELP mechanism is the hydrogen-enhanced dislocation activity (for example, mobility) in the early and intermediate stages of plastic deformation, before the formation of microvoids. This is usually interpreted as a result of the elastic shielding effect by hydrogen, which has been proposed on the basis of experimental observations of the hydrogen-induced reduction in spacings among piled-up edge dislocations 45 . In this Article, we provide compelling experimental evidence of hydrogen-enhanced screw dislocation motion in α-iron. The hydrogen-induced decrease of dislocation activation stress in the experiment is consistent with the decrease of kink-pair nucleation barrier caused by hydrogen in atomistic simulation. These results indicate that the hydrogen effect on kink-pair nucleation could be the underlying mechanism of enhanced dislocation motion when the dislocation line is almost straight. However, the primary factor controlling the increased extent of dislocation bow-out by hydrogen remains uncertain. This is because hydrogen could not only lower the kink-pair nucleation barrier on a bow-out dislocation but also reduce the dislocation line tension to lower the dislocation bow-out stress, leading to an increased extent of dislocation bow-out. These uncertainties warrant further study in the future.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41563-023-01537-w.

Sample preparation
Rectangular lamellae with dimensions of ~30 × 20 × 5 µm 3 were cut from a one-side polished single-crystalline α-iron disk (in which the purity of the used iron was ~99.99%) using a focused ion beam (FIB) instrument FEI Helios NanoLab 600, operated at 30 keV. The so-prepared sample was then transferred and mounted to a mechanical testing rig using the built-in FIB lift-out system. The cylindrical pillars with diameters of ~300 nm and an aspect ratio (length/diameter) of ~3 were fabricated on the lift-out lamellae using the same FIB instrument. To minimize any potentially harmful side effects stemming from the Ga + -beam-induced irradiation damage and geometrical tapering of the sample longitudinal shape, the milling current in the final step was reduced to values below 20 pA. A platinum layer of ~150 nm thickness (as shown in Fig. 2a) was formed on the top of the pillar to protect it from non-uniform contact with the diamond punch during cyclic compression tests, thereby avoiding contact-induced dislocation emission. This experimental design provided well-controlled and repeatable contact boundary conditions.

In situ mechanical tests
All mechanical in situ tests were carried out inside of a Hitachi H-9500 ETEM using a Hysitron PI95 H1H sample holder. The holder was equipped with a flat diamond punch that was driven by a microelectromechanical systems transducer. The transducer operated at a force resolution of ~300 nN and a displacement resolution of ~2 nm. Both the mechanical loading direction and the electron-beam direction were oriented along the <100> crystallographic axis of the α-iron sample. Monotonic uniaxial compression tests were performed under displacement control mode at a strain rate of ~5 × 10 −3 s −1 . The cyclic compression and relaxation tests were conducted in load control mode, and the period of each cycle was 2 s (1.0 s loading + 1.0 s unloading). The applied stress that was imposed during each cycle oscillated from a valley stress to a peak stress to carefully and step-wise manipulate the movement of the mobile dislocations. Hydrogenation of the pillar was conducted under an electron-beam intensity of ~1.2 nA µm −2 in a 2 Pa H 2 environment. Degassing lasted over a period of 3 h under a high vacuum (<5 × 10 −4 Pa) under switched-off electron-beam conditions. Real-time videos of the in situ mechanical tests were recorded using a Gatan 832 charge-coupled device camera, operated at an image acquisition rate of 10 frames per second.

Atomistic simulation
Supplementary Fig. 8 shows the simulation cell containing a right-handed ½<111>{110} screw dislocation at its fully relaxed state. Relaxation was conducted by the conjugate gradient method. The simulation cell has dimensions of 5.6 × 5.6 × 47.5 nm 3 and contains a total number of ~80,000 atoms. A periodic boundary condition was imposed along the dislocation line (z-[111] direction), while the surfaces along the x-[211] and y-[011] directions were maintained traction-free. A 0.6-nm-thick layer was fixed at the top and bottom surfaces of the simulation cell in the y direction, respectively. The shear load was exerted by imposing a displacement-controlled boundary condition on the top slab, while the bottom slab was held fixed. We performed NEB calculations with an embedded atom method potential of FeH 38 using the atomistic simulation package LAMMPS (version 29 OCT 2020) 46 . Molecular dynamics simulations were also performed to understand the effect of hydrogen on the critical stress for dislocation motion at room temperature with the same simulation set-up. A constant shear strain rate of 1 × 10 8 s −1 was exerted in the canonical ensemble. The hydrogen atoms were randomly introduced into the region around the dislocation core. The hydrogen concentration at the dislocation line is around 0.4 nm −1 . To obtain the average virial stress, we performed 20 independent sets of molecular dynamics simulations using the canonical ensemble with random initial velocities. The visualization tool OVITO (version 3.3.0) 47 was used to perform common neighbour analysis to clearly show the screw dislocation.

Data availability
The data that support the findings of this study are available from the corresponding authors upon reasonable request.