WorksPhysics and Astronomy
Precipitation strengthening as a collective multi-dislocation phenomenon
Abstract
Precipitation strengthening is a cornerstone of physical metallurgy, delivering otherwise unattainable combinations of strength and ductility. The approach relies on nanoscale precipitates that impede the motion of dislocations, the primary carriers of metal plasticity. Historically, precipitation strengthening has been rationalized via two idealized, limiting mechanisms: dislocations either cut through or bow around precipitates. However, under many technologically relevant conditions, multiple dislocations can interact with a precipitate at the same time, potentially giving rise to distinct precipitate-dislocation dynamics that in situ experiments are yet to fully resolve. Here, using large-scale atomistic simulations (100 million atoms) that capture these dynamics, we demonstrate that in this regime, the cutting-versus-bowing dichotomy is incomplete: strengthening arises as an emergent phenomenon driven by concurrent multi-dislocation interactions. These interactions simultaneously induce dislocation accumulation at interfaces, storage within precipitates, and precipitate-mediated multiplication inside the matrix. Our findings establish a distinct mechanistic regime of precipitation strengthening beyond traditional models and provide a quantitative foundation for predicting its behavior.
Introduction
Metals are strengthened by precipitating nanoscale secondary-phase particles into a supersaturated matrix, a process known as precipitation strengthening1–4. Widely regarded as a central nanoscale design strategy in modern metallurgy, this approach predates nanoscience by many decades, with its empirical discovery5 in 1906 preceding the experimental tools needed to directly visualize nanoscale precipitates. Precipitates strengthen metals by impeding the motion of dislocations, curvilinear defects that mediate plastic deformation. This basic principle suggests that precipitates that more effectively obstruct dislocation motion should produce greater strengthening. Yet a quantitative connection between atomic-scale dislocation-precipitate interactions and macroscopic strength remains unresolved6.
Established theories of precipitation strengthening7–10 rest on an idealized premise: an encountering dislocation either shears through a precipitate (cutting) or bypasses it (Orowan looping), depending on which path is energetically favored. This binary framework, illustrated in Figure 1a, has inspired numerous analytical models6,7,11. While these formulations capture specific limiting regimes1,9,12–15, they often deviate substantially from empirical observations. As shown in Figure 1b, comparisons with experimental data often reveal pronounced quantitative discrepancies, particularly near the peak strength. To reconcile such discrepancies, phenomenological fitting coefficients are introduced6,9,13,16, a remedy that suggests the idealized framework omits some of the underlying physics. Furthermore, while classical theories predict an abrupt transition from shearing to bypassing, experiments consistently reveal a broad, gradual crossover (Figure 1b). Although precipitate size polydispersity is often invoked to reconcile this inconsistency13,17–19, this explanation presumes that the underlying transition is itself sharp, an assumption that has not been tested. These discrepancies may become increasingly consequential in heavily worked or additively manufactured metals, where high dislocation densities are likely to introduce interactions beyond those represented by the idealized single-dislocation model in Figure 1a. Collectively, these discrepancies underscore a gap in our mechanistic understanding of precipitation strengthening — a framework that, despite its limitations, continues to direct modern alloy design and processing strategies.

Figure 1. From the classical single-dislocation picture to multi-dislocation precipitate strengthening. a) Schematic of the classical single-dislocation picture: a dislocation either shears a precipitate or bypasses by bowing around it and leaving behind a closed loop. b) Experimental strengthening data for Al-Sc13 and Fe-Cu20 compared with single-dislocation model predictions (described in Supplementary Information Note 1). The solid line is drawn as a guide to the eye. c) Number of dislocations within a precipitate interaction region as a function of precipitate radius () for a dislocation density of . Over much of the precipitate size range relevant to engineering alloys21–26, , indicating that multi-dislocation interactions beyond the single-dislocation picture of a) are likely. d) Representative atomistic configuration of Fe containing spherical, monodisperse Cr precipitates (, volume fraction ) during uniaxial compression at room temperature. e) Precipitation strengthening , as a function of at , where is the increase in flow stress relative to the matrix reference. The dashed line shows the corresponding classical single-dislocation prediction; the grey arrow indicates that it continues to increase monotonically over the precipitate-size range shown. Solid line shows the prediction of partitioning-based strengthening relation (eq. (2)). Error bars denote correlation-corrected standard errors of the mean steady-state flow stress.
Here, we show that precipitation strengthening can arise from collective interactions between multiple dislocations and precipitates, a setting not captured by the single-dislocation model. Using ultra-large-scale molecular dynamics (MD) simulations ( atoms) that resolve these processes in atomistic detail, we directly observe how multiple dislocations engage with precipitates during plastic deformation. In this regime, strengthening response cannot be reduced to a binary choice between shearing and bypass. Instead, it arises from the coupled development of dislocation accumulation at precipitate interfaces, dislocation storage within precipitates, and precipitate-mediated dislocation multiplication in the surrounding matrix. These processes occur concurrently and together control the precipitate contribution to strength, providing a microscopic basis for multi-dislocation precipitation hardening that extends beyond the traditional shearing-versus-bypass picture.
Results
Multi-dislocation interaction with precipitates
As summarized in Supplementary Information Note 1, classical models of precipitation strengthening are formulated in terms of isolated interactions between single dislocations and precipitates. A simple geometric estimate of the number of dislocations () within a precipitate’s interaction region (see Methods section “Estimate of dislocations per precipitate”) shows that this regime is not representative of many commercial alloys. When , individual precipitates are likely to be encountered by multiple dislocations simultaneously rather than one dislocation at a time. Figure 1c illustrates this condition for a dislocation density of , which can be reached during cold working[27]. This multi-dislocation regime spans much of the range of precipitate sizes and volume fractions used in practical alloy design and yet it falls outside the single-dislocation picture of Figure 1a. This disconnect motivates our explicit examination of multi-dislocation interactions with precipitates.
To directly probe this multi-dislocation regime, we carried out large-scale MD simulations of body-centered cubic (BCC) Fe single crystals containing coherent Cr precipitates, compressed along the [001] direction. In the series of simulations reported here, both the precipitate radius () and volume fraction () were varied; however, all precipitates within a given sample had the same radius, yielding a monodisperse microstructure. Elemental BCC Cr is elastically stiffer than the BCC Fe matrix, and because Cr atoms are only slightly larger than Fe atoms, this combination provides a well-defined mechanical contrast for precipitation strengthening while retaining precipitate coherency. All samples were deformed at room temperature over a range of strain rates. Figure 1d shows a representative snapshot of these samples.
As deformation proceeds, dislocations multiply and organize into a dense network that spans the microstructure[28, 29, 30] (see Supplementary Information Movie 1 for the full temporal evolution of dislocation networks). In the representative configuration shown in Figure 1d, multiple dislocations impinge simultaneously on the same precipitate, forming transient multi-dislocation interaction zones (see Supplementary Information Note 2 for additional representative configurations). These configurations, which correspond to in Figure 1c, demonstrate that multi-dislocation interactions characterize the deformation under these conditions, rather than isolated single-dislocation events (see also Supplementary Information Note 13).
Across all samples, the stress-strain response is qualitatively similar: an initial linear elastic regime, followed by yielding as dislocations begin to move and multiply, and eventual convergence to steady-state flow at constant stress. Previously reported for pure metals[28] and solid solutions[29, 30], this steady-flow regime coincides with saturation of the dislocation density, at which point the dislocation network attains a statistically stationary configuration. Here, we further observe that dislocation-precipitate interactions also become statistically stationary. Representative examples of stress-strain response and evolution of dislocation density are provided in Supplementary Information Notes 3-4.
The strengthening contribution of the precipitates is reflected in the flow stress. We express this contribution as the precipitation strengthening , defined as the increase in flow stress relative to that of the matrix under the same deformation conditions. Figure 1e shows as a function of precipitate radius at . Consistent with the experimental trends in Figure 1b, varies non-monotonically with : it increases with , reaches a maximum at nm, and then decreases for larger precipitates. In contrast, the classical models (dashed line in Figure 1e) predict strengthening to increase monotonically up to the cutting-to-bypass transition at nm. Notably, the crossover from increasing to decreasing is gradual rather than abrupt, despite the monodisperse precipitate microstructure. This behavior contrasts with the sharp shearing-to-bypass transition expected from classical models (Figure 1b) which is often attributed to precipitate-size polydispersity13,17–19. Instead, the smooth crossover emerges intrinsically from multi-dislocation interactions with precipitates. The same behavior is observed across all volume fractions investigated (see Supplementary Information Figure 4e).
Dislocation partitioning by precipitates
Building on the observation that deformation involves concurrent interactions between multiple dislocations and a single precipitate, we now examine how dislocations partition between the matrix, the precipitates, and their interfaces. Figure 2a presents a magnified view of a precipitate of initial radius nm after plastic straining of the sample to . During deformation, the precipitate does not retain its initial spherical shape; instead, it evolves into an oblate ellipsoid with its minor axis aligned with the loading direction ([001]). This shape change is entirely due to dislocation shearing, as the precipitate remains coherent throughout deformation, with no evidence of phase transformation or deformation twinning at any stage (see Supplementary Information Note 5). The axial strain of the precipitate along the loading direction in Figure 2a is (see Supplementary Information Note 6), which is notably lower than the applied strain of . This strain mismatch indicates limited dislocation transmission across the precipitate-matrix interface: although dislocations intermittently penetrate and shear the precipitate, a substantial fraction of them are arrested at the interface, while others are redirected through the surrounding matrix due to the elastic and lattice mismatch between Fe and Cr9.
Figure 2. Dislocation partitioning between matrix, precipitate, and interface. a) A typical precipitate after uniaxial compression to , illustrating precipitate shape change under loading. The dashed outline shows expected precipitate shape if it were to deform affinely with the matrix. b) Composition-based voxel segmentation used to define matrix (), interface (), and the precipitate (). c) Dislocation density in the matrix, the precipitates, and their interfaces for nm and . is the dislocation density in pure Fe deformed under the same condition. d) Dislocation densities in the matrix, at the interface, and within the precipitate as functions of precipitate radius (for ). e) Volume fractions of the matrix, interface, and within the precipitate as functions of (for ). Error bars (smaller than the marker size and therefore not visible) indicate correlation-corrected standard errors of the mean of steady-state dislocation densities and microstructure fractions.
To quantify how the competition between interfacial transmission and arrest redistributes dislocations, we resolve the dislocation density separately in the matrix, precipitates, and their interfaces (Figure 2b; see Methods section “Composition-based voxel segmentation of dislocation” for details). Figure 2c shows that the interfaces carry the highest dislocation density, indicating dislocation accumulation near the precipitates[31]. By contrast, the precipitates exhibit a substantially lower dislocation density than the matrix, consistent with incomplete dislocation transmission into the precipitate and limited storage within it. More importantly, the dislocation density in the matrix itself exceeds that of a precipitate-free matrix deformed under exactly the same conditions, indicating precipitate-mediated excess dislocation in the matrix[32, 33]. Multi-dislocation interactions at the precipitates therefore give rise to three concurrent processes: interfacial dislocation accumulation, dislocation storage within precipitates, and precipitate-assisted dislocation multiplication into the surrounding matrix. All three take place at every precipitate and together constitute the microscopic basis of precipitation strengthening under these conditions.
Figure 2d shows how the matrix dislocation density varies with precipitate size . Across the full range of radii, remains higher than that of the precipitate-free matrix (). This substantially elevated matrix density arises because precipitates impose local constraints on dislocation motion (observed in Supplementary Information Movie 1), promoting line bowing9, line-length increase11, junction formation[34], and the emission of additional dislocation segments into the surrounding matrix[32, 33]. At a fixed precipitate volume fraction, varies non-monotonically with precipitate size: it increases with , reaches a maximum, and then decreases for larger precipitates. This trend reflects a balance between individual obstacle strength and dislocation-precipitate interaction frequency. Large precipitates are stronger obstacles, but at fixed they are fewer in number and more widely spaced than small precipitates, reducing the frequency of encounters with moving dislocations. Small precipitates, by contrast, are more numerous but are more readily sheared, making them weaker obstacles. The maximum in therefore occurs at nm when precipitates are large enough to obstruct dislocation motion but still numerous enough to interact frequently with the evolving dislocation network.
Superimposed on this elevated matrix-density background, both and vary monotonically with precipitate size. For small precipitates, interfacial accumulation is limited () while transmission into the precipitate is substantial (). As increases, rises while falls. Both trends have the same physical origin: larger precipitates offer greater intrinsic resistance to dislocation slip, causing more dislocations to be arrested at the interface and fewer transmit into the precipitate. Those arrested at the interface build up a back-stress that, together with the precipitate-matrix stiffness contrast, resists further dislocations approaching. Note, however, that transmission never vanishes. Even for the largest precipitates, , at odds with complete obstacle bypass, e.g. Orowan looping (Figure 1a). Dislocation transmission and interfacial accumulation therefore remain concurrent across the full range of precipitate sizes and volume fractions studied here, rather than separating into distinct cutting and bypass regimes[35, 36] (see Supplementary Information Figure 9).
Precipitate size controls not only how dislocations partition between the matrix, interface, and precipitate, but also the volume fractions occupied by these three regions. Figure 2e shows the volume fractions of the matrix (), interface (), and precipitates () as functions of . For small precipitates, the interfacial region occupies a disproportionately large fraction of the microstructure. This arises partly from the geometric increase in the surface-to-volume ratio with decreasing , and partly from shear-induced interface broadening and associated diffusion[37], which affects smaller and more shearable precipitates more strongly (see Supplementary Information Figures 10e and 10f). As a result, can become comparable to, or even exceed, . As increases, however, the interfacial region occupies a progressively smaller fraction of the microstructure, becoming nearly negligible at the largest radii (see Supplementary Information Figure 10 for other volume fractions).
Taken together, these observations clarify how precipitate size controls multi-dislocation strengthening response (Figure 1e). At small , precipitates offer little resistance to gliding dislocations, which readily transmit across the interfaces to shear the precipitates. Accordingly, barely exceeds . At large , rises well above and exerts a substantial back-stress, but strengthening is again limited because the interfacial regions occupy only a small fraction of the material volume. In the intermediate range of precipitate sizes, the precipitates become more resistant while their interfaces still occupy appreciable fractions of the microstructure, thus allowing precipitate shear resistance and interfacial back-stress to contribute concurrently. This same intermediate range also maximizes precipitate-mediated dislocation multiplication in the matrix ( in Figure 2d), which adds a third concurrent contribution to strengthening. The combined effect of these contributions therefore places the strengthening maximum at nm, which is well below the classical model prediction (Figure 1e). And because the dislocation partitioning and interior volume fractions all evolve continuously with (Figure 2d and Figure 2e), this maximum is smooth, with no need to invoke precipitate-size polydispersity.
A partitioning-based strengthening relation
Having identified three distinct regions involved in precipitation strengthening, we now relate them quantitatively to the strengthening response. Consistent with the classical Taylor hardening equation[38], which relates flow stress to the square root of dislocation density, Supplementary Information Figure 11a shows that precipitation strengthening closely co-varies with the total dislocation density. However, this Taylor-like relation becomes increasingly inaccurate as the precipitate volume fraction increases (Supplementary Information Figure 11b and 11c), indicating that total dislocation density alone does not fully capture this strengthening response. We therefore posit that, in multi-dislocation regime, plastic strength depends not only on how many dislocations are stored in the material, but
also on where they are stored. Recognizing that dislocations stored in the matrix, within precipitates, and at the interfaces obstruct dislocation motion to different degrees, we introduce a region-resolved Taylor-type relation for the stress contribution, in which is the stress needed to move dislocations through region against the dislocations already stored there:
Here, is the dislocation density measured in region (see Methods section “Composition-based voxel segmentation of dislocation”), is the Voigt-averaged shear modulus, is the Burgers vector, is the Taylor coefficient associated with that region[34], and is the standard Taylor factor converting stress resolved on the glide planes to the applied flow stress for uniaxial deformation along the .
In a heterogeneous microstructure containing a matrix, interfaces, and precipitate interiors, the overall resistance to dislocation motion is limited by the least resistive deformation region. This is consistent with the observation in Figure 2a that these regions do not satisfy an iso-strain condition: the deformation accommodated by the matrix () and that by precipitates () are measurably different. The three regions can therefore be reasonably assumed to serve as parallel deformation channels: if one region offers lower resistance and more readily accommodates plastic flow, dislocations will preferentially propagate through it, thereby limiting the net strengthening[39]. This behavior can be expressed by the harmonic combination of the three resistances[40, 41] [16]:
Here, , , and are the volume fractions of the matrix, interface, and precipitate regions, respectively, as defined in Figure 2e, and is the friction stress due to the lattice (Peierls) resistance to dislocation motion. The strengthening, , is then obtained by subtracting the flow stress of the precipitate-free reference material from the predicted obtained from eq. (2).
With a single set of fitted Taylor coefficients , eq. (2) captures strengthening measured in all 60 Fe-Cr samples over the entire range of , , and strain rates (Figure 3). The same coefficients, with no readjustment, produce the solid curve in Figure 1e. The agreement of this region-resolved Taylor model with the measurements is considerably closer than that of the conventional Taylor model relating flow stress to the total dislocation density (see Supplementary Information Figures 11b and 11c). Note also that the fitted parameters take physically sensible values. All three Taylor coefficients, , , and , fall within the range reported for BCC metals[34], and the friction stress exhibits an approximately linear dependence on the strain rate (inset of Figure 3), consistent with behavior previously observed at high strain rates[42].
Figure 3. Figure 3: A partitioning-based strengthening relation. Strengthening from eq. (2) compared with measurements across different , , and strain rates. Unless otherwise noted, the data correspond to the Fe-Cr system. There are 78 data points, each of which represents a sample with atoms. The dashed line denotes perfect agreement. The inset shows the fitted intrinsic friction stress as a function of strain rate for the Fe-Cr system.
So far, eq. (2) has been tested on Fe-Cr alone, where dislocations accumulate preferentially at the interfaces and in the matrix. A more demanding test is a system in which dislocations distribute differently among the three regions. One way to produce such a system is to invert the stiffness contrast, i.e., embedding softer Fe precipitates in a stiffer Cr matrix (see Supplementary Information Note 9). In these samples, referred to here as Cr-Fe systems, remains greater than over the entire range of precipitate radii. The soft precipitates pose a low barrier to entry, so dislocations transmit into them freely and accumulate there, and the stored density raises even though the precipitates themselves are easily sheared. Strengthening in these alloys therefore comes from storage within the precipitates, and the response no longer exhibits the maximum observed in Figure 1e (see Supplementary Information Figure 12b). Equation 2 captures this qualitative difference dependence on (Figure 3). It does so as well for the Cu-Co alloy (see Supplementary Information Note 10), which differs from Fe-Cr in crystal structure, chemistry, and dislocation character. The partitioning depicted in Figure 2 and the strengthening relation built on it therefore carry over to multi-dislocation precipitation strengthening across crystal structures, chemistries, and phase contrasts involving coherent precipitates.
Strengthening maps from stiffness mismatch
The comparison between Fe-Cr and Cr-Fe points to the stiffness contrast between precipitate and matrix as the property that sets the dislocation partitioning: reversing its sign moved dislocations from the matrix into the precipitates. Classical models treat this contrast through the shear modulus mismatch , with strengthening scaling as , so that soft and stiff precipitates of equal strengthen equally9. The dislocation partitioning picture predicts otherwise, since the sign of decides where dislocations are stored and the three regions do not resist dislocation motion equally (eq. (1)).
Testing this systematically in real alloys is difficult because stiffness cannot be varied alone: changing the precipitate chemistry changes its lattice parameter and interface energy along with it, which both affect strengthening. We circumvent this with computational alchemy. Starting from a BCC matrix, we construct precipitate elements whose stiffness is dialed continuously from much softer to much stiffer than the matrix while their lattice parameter stays matched to it, so the precipitates remain coherent and coherency strains play no role (see Methods section “Alchemical model element construction”). This is the continuous counterpart of choosing a chemical element for its stiffness, free of the discrete values the periodic table offers and of the solubility limits that decide which element pairs form coherent precipitates at all. Although the alchemical elements represent no real material, the strengthening map they produce spans the full range between soft and stiff precipitates.
Figure 4a shows that strengthening increases with on both sides of the origin , as classical models predict, but its magnitude is not symmetric and depends on the sign of .
Figure 4. Strengthening maps from alchemical modulus mismatch. a) Strengthening as a function of precipitate modulus mismatch and radius for alchemical precipitates at fixed . The response is asymmetric about , unlike the classical prediction, which scales as . Soft and stiff precipitates with comparable therefore do not strengthen equivalently, as highlighted by the dashed boxes (). Fe-Cr, Cr-Fe, and Fe-Cu samples () at the same conditions lie on the alchemical strengthening map at comparable modulus mismatch. b) Steady-state dislocation densities in the matrix, interface, and precipitate regions as functions of , for representative alchemical systems with .
For example, at , stiff precipitates () strengthen almost twice as much as soft precipitates (), and the shape of the curve differs on the two sides of the origin. Figure 4b shows the dislocation partitioning behind this asymmetry. Stiff precipitates raise the dislocation density in the matrix and at the interfaces alike, with the precipitates density well below both. Soft precipitates invert this ordering: the highest dislocation density is now inside the precipitates while the matrix is left with the lowest density of all. Equation (1) connects this redistribution to strength. The highest densities now sit inside the soft precipitates, where the low shear modulus makes stored dislocations weak obstacles, while the matrix, which occupies most of the volume and controls the flow stress through eq. (2), holds a lower density than it does with stiff precipitates. An obstacle strength that depends on alone, as in the classical models, carries no information about where dislocations are stored and therefore cannot express this redistribution or the asymmetry it produces9,13,16.
The strengthening of the Fe-Cr, Cr-Fe, and Fe-Cu samples (all coherent) reported above falls close to the curve computed for the alchemical alloys at comparable (Figure 4a), so a real alloy can be located on the map by its stiffness contrast. Figure 4a can thus serve as a practical guide for selecting stiffness contrast.
Discussion and Conclusion
A key finding of this work is that precipitates actively alter the dislocation density within the matrix itself[43, 44]. While dislocation storage within precipitates and accumulation at interfaces have direct analogs in classical shearing and bypassing, precipitate-mediated matrix multiplication is not represented in single-dislocation descriptions. This mechanism has been postulated and observed previously[32, 33, 43] but here we relate it to strengthening and show that it becomes consequential under multi-dislocation conditions. Incorporating multi-dislocation precipitation strengthening into mesoscale descriptions[45] [6] therefore requires accounting for the way precipitates reshape the surrounding dislocation network through coupled shearing, bypassing, and multiplication, in addition to the resistance associated with cutting and bypassing. Existing single-dislocation theories do not represent these concurrent mechanisms, while mesoscale approaches such as discrete dislocation dynamics will require further development to capture them quantitatively.
The comparison between measurements and the classical models in Figure 1e and Supplementary Information Figure 14c illustrates how neglecting multi-dislocation effects can lead to substantial differences in the predicted strengthening response. Similarly, while precipitate-size polydispersity typically broadens experimental strengthening curves13,17,18, it is not a prerequisite for a smooth strengthening peak. Here, a broad crossover emerges intrinsically from the co-evolution of interfacial dislocation accumulation, precipitate storage, and precipitate-mediated dislocation multiplication, even within strictly monodisperse arrays (Figure 1e). The sign dependence of strengthening on the stiffness contrast (Figure 4) reaches into the choice of precipitate chemistry itself. The modulus-mismatch term in the alloy-design models used in practice9,13,16 ranks a soft and a stiff candidate precipitate of equal as equivalent, whereas Figure 4 shows the stiff one strengthening almost twice as much, because the two place dislocations in different regions of the microstructure. Screening precipitate phases by modulus mismatch alone therefore misranks candidates, and always in the same direction: it overestimates soft precipitates relative to stiff ones.
The observations presented here pertain most closely to deformation regimes characterized by coherent precipitates, high strain rates, and high dislocation densities, where multiple dislocations interact with a single precipitate simultaneously. Such conditions are particularly relevant to additively manufactured alloys and materials subjected to extreme conditions. Alternative conditions — such as strongly planar slip, semi-coherent or incoherent particles[46], deformation twinning[47], deformation-induced structural transformations[48], or elevated temperatures — can activate additional processes (extended pile-ups or climb-assisted relaxation, for example). These processes may introduce, suppress, or redistribute deformation pathways and thereby alter the relative importance of the active plasticity channels (Supplementary Information Notes 9-11). The Fe-Cu system offers one example (Supplementary Information Note 11): deformation-induced transformation of the Cu precipitates adds a fourth channel alongside the three identified here, and classical models, which lack it, overestimate strengthening in the size range where it activates (Figure 1b). Within its applicability regime, eq. (2) still carries a limitation of a different kind: it relates strengthening to a measured dislocation partitioning, and is in that sense descriptive. A predictive theory would have to anticipate the partitioning from the precipitate’s own properties, and Figure 4 is a first step in that direction for the stiffness contrast.
The agreement between the simulated alloys (Fe-Cr, Cr-Fe, and Fe-Cu) and alchemical systems in Figure 4a suggests that such simulation-derived strengthening maps can guide the choice of precipitate chemistry. Extending them to real alloys demands chemical accuracy and sample size at once, and rapid advances in quantum-accurate machine-learning interatomic potentials[49, 50] and increasingly powerful high-performance computing capabilities are now enabling large-scale MD simulations to serve as controlled computational experiments for quantitatively assessing strengthening trends. The present work illustrates what such experiments can deliver: collective mechanisms that are difficult to isolate experimentally, and with them a pathway toward mechanistically informed descriptions of strengthening in precipitate-containing alloys and other heterogeneous microstructures.
Methods
Estimate of dislocations per precipitate
To obtain an order-of-magnitude estimate of the number of dislocations () that interact with an individual precipitate (Figure 1c), we start from the bulk dislocation density . For a randomly oriented plane, the corresponding effective areal dislocation density is[51]
which gives the expected number of dislocation lines intersecting that plane for a spatially homogeneous dislocation distribution.
We next estimate a characteristic interaction area associated with a spherical precipitate. For a precipitate of radius , the mean intersection radius with a randomly oriented plane is6
Dislocations, however, do not need to pass through this strict geometric intersection region to interact with the precipitate. Because the elastic interaction decays approximately as with distance from the precipitate center[52], dislocations outside the geometric intersection can still experience appreciable precipitate-induced forces. At a distance equal to twice the geometric intersection radius, the interaction strength is reduced to roughly of its interfacial value. While diminished, this interaction remains non-negligible, implying that dislocations far beyond the strict geometric intersection can still experience appreciable precipitate-induced forces. To provide a conservative and geometrically transparent estimate, we define a characteristic interaction radius
which captures both the geometric intersection and a physically meaningful elastic interaction range. Here, is used only to define the geometric estimate and does not imply that dislocation-precipitate interactions vanish beyond it.
Under these assumptions, the expected number of dislocations interacting with a precipitate is obtained by multiplying the effective areal dislocation density by the circular interaction area:
In Figure 1c, this expression was evaluated using a representative dislocation density of . Cold-worked metals are commonly reported[27] to exhibit ex situ dislocation densities on the order of . Such measurements reflect partially relaxed microstructures following deformation, during which processes such as annihilation, junction rearrangement, and recovery reduce the instantaneous dislocation density. During active plastic deformation, transient dislocation densities can exceed the relaxed value[53]. Indeed, substantially higher densities have been measured in heavily deformed engineering alloys, including in cold-rolled ODS-Eurofer steel and in deformation-processed martensitic steel[54, 55]. The choice of therefore represents a reasonable estimate of the instantaneous dislocation density under sustained deformation.
The applicability of this estimate to the samples deformed in this work is discussed in Supplementary Information Note 13.
Mechanical deformation
Molecular dynamics simulations of uniaxial compression were performed on BCC Fe containing spherical Cr precipitates using a timestep of 5 fs. Atomic interactions between elements were described using an embedded-atom method (EAM) potential[56] that accurately reproduces the Fe-Cr phase separation and associated thermodynamic and elastic properties. Simulation cells had initial dimensions of , oriented along the [100], [010], and [001] crystallographic axes, respectively, where is the Fe lattice parameter. Each sample contained approximately 100 million atoms. Periodic boundary conditions were applied in all three dimensions.
Spherical regions of radius were placed at random positions within the simulation cell subject to a non-overlap constraint. The number of spheres was chosen to achieve the target precipitate volume fraction . All atoms within each sphere were then replaced by Cr atoms, producing coherent spherical BCC Cr precipitates. After precipitate construction, 100 vacancy-type prismatic dislocation loops of hexagonal shape and 10 nm diameter were introduced in each sample. The loop orientations were chosen so that all four Burgers vector variants were represented in equal proportion. This procedure yielded an initial dislocation density of . Note that the shapes and density of initial sources affect only the transient plastic response whereas steady-state plastic flow does not depend on the exact configuration of initial dislocation sources[28, 57, 58, 59, 60].
Following loop creation, samples were equilibrated for 100 ps at 300 K using a Langevin thermostat (damping parameter 10 ps) and maintained at zero pressure using an isothermal-isobaric (NPH) barostat (damping parameter 10 ps; chain length 3), allowing only isotropic cell fluctuations. Samples were then deformed in uniaxial compression along [001] at a constant true strain rate[28, 60, 61]. During deformation, the stresses in three directions orthogonal to the loading axis ([100] and [010]) were maintained at zero using an NPH barostat acting on those directions (damping parameter 10 ps; chain length 3), while the temperature was held at 300 K using the Langevin thermostat (damping parameter 10 ps).
Most simulations were performed at to a total true strain of 0.6 (simulation time of 6 ns). For , deformation to a true strain of 0.6 required 0.6 ns. For , configurations first deformed at to a true strain of 0.2 were used as initial states; the strain rate was then reduced to and deformation was continued to a true strain of 0.6 (simulation time of 40 ns). See Supplementary Information Note 12 for a comparison of strengthening across strain rates.
Reported steady-state flow stresses are obtained by averaging flow stress over the true-strain interval ; error bars denote autocorrelation-corrected standard errors of the mean. Within this interval, the dislocation networks remain statistically stationary (Supplementary Information Note 4), making the steady-state flow stress a robust atomistic measure of material strength[28, 60]. This steady-state flow stress differs from the yield strength measured in experiments and predicted by the classical models (Figure 1b and Figure 1e): the yield strength marks the onset of plastic flow, whereas the steady-state flow stress is the stress required to sustain plastic flow against a fully developed dislocation network. Both quantities, however, increase with the resistance that precipitates offer to dislocation motion, and the precipitation strengthening isolates this precipitate contribution by subtracting the response of the precipitate-free matrix deformed under identical conditions. The comparison in Figure 1b and Figure 1e therefore concerns the dependence of on precipitate size, i.e., the position of the strengthening maximum and the breadth of the crossover between the shearing and bypass regimes, which is set by the dislocation-precipitate interactions and should therefore hold for both measures of strength.
All simulations were carried out using the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS)[62] and visualizations were done using Open Visualization Tool (OVITO)[63].
Composition-based voxel segmentation of dislocation
Dislocations were identified using the Dislocation Extraction Algorithm (DXA)[64]. To quantify spatial variations in dislocation density, each snapshot was discretized into cubic voxels of size . A voxel size of exceeds twice the BCC Fe lattice parameter and encompasses interactions up to the sixth nearest-neighbor shell, providing adequate atomic sampling ( atoms) while maintaining the spatial resolution required to resolve precipitates as small as . The local Cr atomic fraction was computed for each voxel and used to classify voxels as matrix (), interface (), or precipitate (). The microstructure fractions of interface () and precipitate () were computed as the fraction of voxels belonging to each region. Importantly, differs from the nominal precipitate volume fraction : quantifies the volume fraction of precipitate regions at steady state as evaluated by our composition-based voxel segmentation method, whereas denotes the initial prescribed precipitate volume fraction based on the ratio of Cr to Fe atoms.
For each region, the associated dislocation density was calculated by summing the total dislocation line length contained within voxels of that region[45] and normalizing by the corresponding region volume. Reported steady-state dislocation properties and region fractions are averaged over the true-strain interval and error bars denote autocorrelation-corrected standard errors of the mean.
Note that the precipitate and interface regions can be smaller than the characteristic forest spacing implied by eq. (1). Accordingly, the computed dislocation densities and using the aforementioned methodology should not be interpreted as local forest densities within an individual precipitate or interface region. Rather, they represent ensemble-averaged dislocation densities over the full population of precipitate and interface regions.
Precipitate strain evaluation
The precipitate strain at the end of deformation was computed from precipitate shapes obtained by cluster analysis. For each cluster (i.e., precipitate), we calculated its gyration tensor . The precipitate extent along the loading direction was defined from the projection of onto the unit vector [65].
The true strain of a precipitate along was then defined as,
where is the corresponding extent measured from the undeformed configuration of the same precipitate. The reported precipitate strain was obtained by averaging over all precipitates in the sample.
Partitioning-based strengthening relation
The strengthening predicted by the partitioning-based relation, shown on the x-axis of Figure 3, was evaluated using eq. (2). The required dislocation densities and volume fractions, , , , , , and , were obtained directly from the MD simulations using the composition-resolved dislocation segmentation procedure described in Composition-based voxel segmentation of dislocation.
The region-resolved stress contributions, , , and , were computed using eq. (1). For each material system, the shear modulus and Burgers vector of the matrix and precipitate regions were computed from the corresponding perfect crystals at 0 K. The interfacial shear modulus and Burgers vector were then taken as the arithmetic averages of the corresponding matrix and precipitate values.
For each material system, the region-resolved Taylor coefficients , , and were obtained by calibrating eq. (2) against the flow stresses measured from the corresponding MD simulations. Once calibrated, the same set of coefficients was used for all microstructures within a given material system.
Alchemical model element construction
Alchemical model element construction
To probe the effect of modulus mismatch on precipitation strengthening, we adopted the computational alchemy methodology of ref. [66]. The Zhou et al. [67] EAM parameterization for Ta was used as the reference matrix potential. To generate model precipitate elements with controlled stiffness, all 11 energy-dimensional parameters of the Ta potential were scaled by a factor , producing an element with stiffness times that of the matrix Ta. This procedure allowed us to construct alchemical matrix-precipitate alloy models in which the modulus mismatch could be systematically varied while retaining the same underlying crystal structure and lattice compatibility.
Elastic modulus is not the only material property affected by the rescaling of energy-dimensional parameters. Cohesive energy, defect energetics, Peierls barriers to dislocation motion, and melting point are all altered as well[66]. This is because, like the elastic response, they arise from the same underlying interatomic energy landscape and therefore naturally covary with stiffness as that landscape changes. Indeed, within the same computational-alchemy framework, the unstable stacking-fault energy was found to vary nearly linearly with stiffness[66]. We find a similar nearly linear dependence of melting point on stiffness (Supplementary Information Figure 17a). In real alloys, the same physical coupling persists: changes in stiffness are accompanied by changes in other properties governed by the underlying energy landscape. Thus, we deliberately retain this coupled variation as part of the strengthening response associated with changing precipitate stiffness in our alchemical models, rather than artificially constraining the elastic modulus to vary independently of other energy-derived properties.
Seven values of from 0.1 to 3.0 were considered. All values produced coherent BCC precipitates, except for , for which some precipitates became semi-coherent or incoherent during deformation. This loss of coherency is consistent with the low melting point of the () precipitates ( K), which is close to the simulation temperature (300 K). However, the strengthening response in Figure 4a remains asymmetric even when the case is excluded and when the simulations are repeated at 50 K, indicating that the asymmetry is not simply a consequence of differences in homologous temperature (see Supplementary Information Note 14). The modulus mismatch reported in Figure 4 was defined as , where and are the Voigt-averaged shear moduli of the corresponding model precipitate and Ta matrix elements[68].
Using Ta as the matrix element and model alchemical elements as the precipitate elements, we performed MD simulations of mechanical deformation at 300 K and a strain rate of s. Each system contained approximately 100 million atoms and was constructed with a lattice constant of 3.303 Å. The simulations followed the same procedure described in Mechanical deformation. Dislocation partitioning was analyzed using the procedure described in Composition-based voxel segmentation of dislocation, except that a voxel size of 7 Å × 7 Å × 7 Å was used.
Data availability
The raw simulation data have not been deposited in a public repository because of their large size, which spans multiple terabytes. Processed data sufficient to reproduce all figures and results in the paper and Supplementary Information will be made available upon publication in a public repository.
Code availability
LAMMPS software[62] used in all molecular dynamics simulations reported in this work is available at https://www.lammps.org. Our figure style is implemented in LovelyPlots[69] under the paper style. Code to reproduce all figures in the paper from processed data will be made available upon publication in a public repository. Any custom code that is not currently available can be subsequently added upon reasonable request to the corresponding author.
Acknowledgments
This material is based upon work supported by the Air Force Office of Scientific Research (AFOSR) under award number FA9550-25-1-0199, through the Young Investigator Program. This work is also supported by the MathWorks Ignition Fund. M.I. and V.V.B. acknowledge funding support from the Laboratory Directed Research and Development Program. All authors acknowledge computational time allocation on Toulumne supercomputer from the Computational Grand Challenge program at Lawrence Livermore National Laboratory. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory, USA under Contract DE-AC52-07NA27344.
Author contributions statement
M.I., V.V.B., and R.F. conceived the project. M.I. performed molecular dynamics simulations and analyzed the results. J.O. performed melting-point calculations. All authors contributed to the interpretation of the results, prepared, reviewed, and edited the manuscript. R.F. and V.V.B. supervised the project.
Competing interests statement
The authors declare no competing interests.
References
References
- [27]G. K. Williamson and R. E. Smallman, “III. Dislocation densities in some annealed and cold-worked metals from measurements on the X-ray debye-scherrer spectrum”, The Philosophical Magazine: A Journal of Theoretical Experimental and Applied Physics (1, 1956), DOI: 10.1080/14786435608238074.DOI
- [28]Luis A. Zepeda-Ruiz, Alexander Stukowski, Tomas Oppelstrup, Nicolas Bertin, Nathan R. Barton, Rodrigo Freitas, and Vasily V. Bulatov, “Atomistic insights into metal hardening”, Nature Materials (2021), DOI: 10.1038/s41563-020-00815-1.
- [29]Mahmudul Islam, Killian Sheriff, and Rodrigo Freitas, “Dislocation-mediated short-range order evolution during thermomechanical processing”, Acta Materialia (2025), DOI: https://doi.org/10.1016/j.actamat.2025.121838.
- [30]Mahmudul Islam, Killian Sheriff, Yifan Cao, and Rodrigo Freitas, “Nonequilibrium chemical short-range order in metallic alloys”, Nature Communications (8, 2025), DOI: 10.1038/s41467-025-64733-z.DOI
- [31]Hao Zhou, Xiaolei Wu, David Srolovitz, and Yuntian Zhu, “Designing heterostructured materials”, Nature Materials (13, 2026), DOI: 10.1038/s41563-025-02444-y.DOI
- [32]Shenyou Peng, Yujie Wei, and Huajian Gao, “Nanoscale precipitates as sustainable dislocation sources for enhanced ductility and high strength”, Proceedings of the National Academy of Sciences (10, 2020), DOI: 10.1073/pnas.1914615117.DOI
- [33]Shenyou Peng, Zhili Wang, Jia Li, Qihong Fang, and Yujie Wei, “Beyond Orowan hardening: Mapping the four distinct mechanisms associated with dislocation-precipitate interaction”, International Journal of Plasticity (2023), DOI: https://doi.org/10.1016/j.ijplas.2023.103710.
- [34]Vasily V. Bulatov, Nicolas Bertin, Sylvie Aubry, Luis A. Zepeda-Ruiz, Xinran Zhou, Aoyan Liang, Tomas Oppelstrup, Babak Sadigh, Athanasios Arsenlis, Ryan Austin, Robert Carson, and Nathan R. Barton, “Network aspects of single crystal plasticity”, Journal of Materials Research and Technology (1, 2025), DOI: 10.1016/j.jmrt.2025.04.014.DOI
- [35]Sabyasachi Chatterjee, Yang Li, and Giacomo Po, “A discrete dislocation dynamics study of precipitate bypass mechanisms in nickel-based superalloys”, International Journal of Plasticity (1, 2021), DOI: 10.1016/j.ijplas.2021.103062.DOI
- [36]Colin A. Stewart, Edwin A. Antillon, Markus Sudmanns, Jaafar A. El-Awady, Keith E. Knipling, Patrick G. Callahan, David J. Rowenhorst, and Richard W. Fonda, “Predicting yield stress in a nano-precipitate strengthened austenitic steel by integrating multi length-scale simulations and experiments”, Acta Materialia (15, 2024), DOI: 10.1016/j.actamat.2024.119918.DOI
- [37]L. Kovarik, R. R. Unoic, Ju Li, P. Sarosi, C. Shen, Y. Wang, and M. J. Mills, “Micotwinning and other shearing mechanisms at intermediate temperatures in Ni-based superalloys”, Progress in Materials Science (1, 2009), DOI: 10.1016/j.pmatsci.2009.03.010.DOI
- [38]Geoffrey Ingram Taylor, “The mechanism of plastic deformation of crystals. Part I.—Theoretical”, Proceedings of the Royal Society of London. Series A, Containing Papers of a Mathematical and Physical Character (2, 1934), DOI: 10.1098/rspa.1934.0106.DOI
- [39]Yuntian Zhu and Xiaolei Wu, “Heterostructured materials”, Progress in Materials Science (1, 2023), DOI: 10.1016/j.pmatsci.2022.101019.DOI
- [40]R.C. Picu and R. Li, “On the superposition of flow stress contributions at finite temperatures and in the athermal limit”, Acta Materialia (2010), DOI: 10.1016/j.actamat.2010.06.020.DOI
- [41]U Lagerpusch, V Mohles, D Baither, B Anczykowski, and E Nembach, “Double strengthening of copper by dissolved gold-atoms and by incoherent SiO2-particles: how do the two strengthening contributions superimpose?”, Acta Materialia (2000), DOI: https://doi.org/10.1016/S1359-6454(00)00172-5.
- [42]Haidong Fan, Qingyuan Wang, Jaafar A. El-Awady, Dierk Raabe, and Michael Zaiser, “Strain rate dependency of dislocation plasticity”, Nature Communications (23, 2021), DOI: 10.1038/s41467-021-21939-1.DOI
- [43]M. F. Ashby, “The deformation of plastically non-homogeneous materials”, The Philosophical Magazine: A Journal of Theoretical Experimental and Applied Physics (1, 1970), DOI: 10.1080/14786437008238426.DOI
- [44]Sylvain Queyreau, Ghiath Monnet, and Benoit Devincre, “Orowan strengthening and forest hardening superposition examined by dislocation dynamics simulations”, Acta Materialia (1, 2010), DOI: 10.1016/j.actamat.2010.06.028.DOI
- [45]Nicolas Bertin, “Connecting discrete and continuum dislocation mechanics: A non-singular spectral framework”, International Journal of Plasticity (1, 2019), DOI: 10.1016/j.ijplas.2018.12.006.DOI
- [46]Mingjun Yang, Andrey Orekhov, Zhi-Yi Hu, Man Feng, Shenbao Jin, Gang Sha, Kai Li, Vahid Samaee, Min Song, Yong Du, Gustaaf Van Tendeloo, and Dominique Schryvers, “Shearing and rotation of β″ and β′ precipitates in an Al-Mg-Si alloy under tensile deformation: In-situ and ex-situ studies”, Acta Materialia (1, 2021), DOI: 10.1016/j.actamat.2021.117310.DOI
- [47]Li Zhong, Yin Zhang, Xiang Wang, Ting Zhu, and Scott X. Mao, “Atomic-scale observation of nucleation- and growth-controlled deformation twinning in body-centered cubic nanocrystals”, Nature Communications (16, 2024), DOI: 10.1038/s41467-024-44837-8.DOI
- [48]Laure Bourgeois, Yong Zhang, Zeshong Zhang, Yiqiang Chen, and Nikhil V. Medhekar, “Transforming solid-state precipitates via excess vacancies”, Nature Communications (6, 2020), DOI: 10.1038/s41467-020-15087-1.DOI
- [49]Tsz Wai Ko and Shyue Ping Ong, “Recent advances and outstanding challenges for machine learning interatomic potentials”, Nature Computational Science (2023), DOI: 10.1038/s43588-023-00561-9.DOI
- [50]Killian Sheriff, Daniel Xiao, Yifan Cao, Lewis R. Owen, and Rodrigo Freitas, Machine learning potentials for modeling alloys across compositions, 2025, DOI: 10.48550/arXiv.2506.12592.arxiv.org/abs/2506.12592
- [51]G. Schoeck, “Correlation between Dislocation Length and Density”, Journal of Applied Physics (1962), DOI: 10.1063/1.1728821.DOI
- [52]John Douglas Eshelby, “The determination of the elastic field of an ellipsoidal inclusion, and related problems”, Proceedings of the Royal Society of London. A. Mathematical and Physical Sciences (1957), DOI: 10.1098/rspa.1957.0133.
- [53]Zhening Yang, Rhys Thomas, Jack Donoghue, Albert Smith, Michael Preuss, and Ali Gholinia, “In situ and correlative study of dislocation density and deformation mechanisms in Inconel 690”, Materials Characterization (2025), DOI: 10.1016/j.matchar.2025.115699.DOI
- [54]R. A. Renzetti, H. R. Z. Sandim, R. E. Bolmaro, P. A. Suzuki, and A. Möslang, “X-ray evaluation of dislocation density in ODS-Eurofer steel”, Materials Science and Engineering: A (2012), DOI: 10.1016/j.msea.2011.11.051.DOI
- [55]B. B. He, B. Hu, H. W. Yen, G. J. Cheng, Z. K. Wang, H. W. Luo, and M. X. Huang, “High dislocation density-induced large ductility in deformed and partitioned steels”, Science (2017), DOI: 10.1126/science.aan0177.DOI
- [56]S. M. Eich, D. Beinke, and G. Schmitz, “Embedded-atom potential for an accurate thermodynamic description of the iron–chromium system”, Computational Materials Science (15, 2015), DOI: 10.1016/j.commatsci.2015.03.047.DOI
- [57]Nicolas Bertin, Robert Carson, Vasily V. Bulatov, Jonathan Lind, and Matthew Nelms, “Crystal plasticity model of BCC metals from large-scale MD simulations”, Acta Materialia (1, 2023), DOI: 10.1016/j.actamat.2023.119336.
- [58]Vasily Bulatov and Wei Cai, Computer Simulations of Dislocations, OUP Oxford, 2, 2006, ISBN: 978-0-19-852614-8.DOI
- [59]F. C. Frank and W. T. Read, “Multiplication Processes for Slow Moving Dislocations”, Physical Review (15, 1950), DOI: 10.1103/PhysRev.79.722.DOI
- [60]Luis A. Zepeda-Ruiz, Alexander Stukowski, Tomas Oppelstrup, and Vasily V. Bulatov, “Probing the limits of metal plasticity with molecular dynamics simulations”, Nature (2017), DOI: 10.1038/nature23472.
- [61]Xinran Zhou, Jaime Marian, Fei Zhou, and Vasily V. Bulatov, “Probing multi-dimensional composition spaces in search of strong metallic alloys”, npj Computational Materials (31, 2026), DOI: 10.1038/s41524-026-01975-5.DOI
- [62]Aidan P. Thompson, H. Metin Aktulga, Richard Berger, Dan S. Bolintineanu, W. Michael Brown, Paul S. Crozier, Pieter J. in ’t Veld, Axel Kohlmeyer, Stan G. Moore, Trung Dac Nguyen, Ray Shan, Mark J. Stevens, Julien Tranchida, Christian Trott, and Steven J. Plimpton, “LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales”, Computer Physics Communications (1, 2022), DOI: 10.1016/j.cpc.2021.108171.DOI
- [63]Alexander Stukowski, “Visualization and analysis of atomistic simulation data with OVITO–the Open Visualization Tool”, Modelling and Simulation in Materials Science and Engineering (2009), ISSN: 0965-0393, DOI: 10.1088/0965-0393/18/1/015012.DOI
- [64]Alexander Stukowski and Karsten Albe, “Extracting dislocations and non-dislocation crystal defects from atomistic simulation data”, Modelling and Simulation in Materials Science and Engineering (2010), ISSN: 0965-0393, DOI: 10.1088/0965-0393/18/8/085001.DOI
- [65]Wayne L. Mattice and U. Suter, Conformational theory of large molecules : the rotational isomeric state model in macromolecular systems, Wiley, 1994.
- [66]Aoyan Liang, Nicolas Bertin, Xinran Zhou, Sylvie Aubry, and Vasily V. Bulatov, “Computational alchemy clarifies origins of alloy strengthening”, npj Computational Materials (18, 2025), DOI: 10.1038/s41524-025-01910-0.DOI
- [67]X. W. Zhou, R. A. Johnson, and H. N. G. Wadley, “Misfit-energy-increasing dislocations in vapor-deposited CoFe/NiFe multilayers”, Phys. Rev. B (2004), DOI: 10.1103/PhysRevB.69.144113.DOI
- [68]James G. Berryman, “Bounds and estimates for elastic constants of random polycrystals of laminates”, International Journal of Solids and Structures (2005), DOI: https://doi.org/10.1016/j.ijsolstr.2004.12.015.
- [69]Killian Sherriff, LovelyPlots, a collection of matplotlib stylesheets for scientific figures, 2022, DOI: 10.5281/zenodo.6903936.


