WorksPhysics and Astronomy
On representation of macroscopic crack in periodic fine-scale discrete mechanical models
Abstract
In multiscale modeling of heterogeneous softening materials, boundary conditions (BC) in the fine-scale model strongly influence the strain localization pattern and the macroscopic response. For rectilinear models (e.g., squares or cubes), standard Periodic BCs produce artificially ductile behavior with excessive energy dissipation when the localization band inclination does not match the periodicity directions. Recently proposed Tessellation and Percolation-path-aligned BCs promise to address this by adapting the periodicity frame to align with the evolving localization bands. Alternatively, spherical/circular models provide an orientation independent response by design. Unfortunately, the standard Periodic BCs do not allow development of proper localization band crossing spherical model's boundaries. A recently proposed modification addresses this by adding a displacement jump to the spherical periodic BCs. This study evaluates the applicability of these novel BCs to a mesoscale discrete particle model of concrete. Two-dimensional square and circular models under uniaxial tension with different loading directions are analyzed, with the selected approaches extended to three-dimensional cube models. Results show that Percolation-path-aligned BCs exhibit major shortcomings: they can lead to multiple localization bands due to uneven straining of the two boundary sections and their weakly constrained section can be prone to spurious strain localization. In contrast, Tessellation BCs consistently yield a well-defined localization band, whose length is determined solely by the model geometry, making it straightforward to account for in post-processing. Periodic boundary conditions augmented with a displacement jump applied to a circular model sometimes incorrect produce crack patterns similar to those under the standard Periodic BCs.
Introduction
The computational analysis of large-scale structures made of complex heterogeneous materials can be significantly facilitated by multiscale modeling techniques. Instead of explicitly resolving the entire microstructure or mesostructure, these methods utilize a representative volume element (RVE), a small fine-scale domain whose response adequately characterizes the behavior of the bulk material. While the existence of an RVE is well established for elastic and hardening materials, its applicability to softening materials remains controversial. The main reason is the strain localization, which violates the assumptions of classical homogenization and gives rise to a deterministic size effect [19].
Despite these theoretical concerns, multiscale approaches have been successfully applied to materials exhibiting strain softening. This is demonstrated by numerous studies employing both continuous fine-scale models [1, 7, 6, 37, 32, 23] and discrete models [13, 14, 29, 30, 39]. These approaches use the homogenized response of a periodic fine-scale model as the constitutive model for the coarse scale, treating the fine-scale model as though it were a classical RVE. An unresolved aspect of multiscale analysis of softening materials concerns the choice of the fine-scale model size, as conventional RVE-based principles cannot be employed. In Ref. [18] this is addressed through a coupled-volume approach in which the size of the fine-scale model is prescribed by the size of the associated coarse-scale finite element. Another approach by Unger [38] decomposes the displacement field of the fine-scale model into a homogeneous part, fluctuations, and a cracking part based on additional degrees of freedom accounting for crack opening and sliding and extends the Hill-Mandel lemma so that the fine-scale model is not required to have the same size as the corresponding macroscopic integration point. References [35, 27] use similar concepts.
Independent of the issues concerning fine-scale model size and representativeness, the influence of the imposed boundary conditions (BCs) presents an important challenge. They influence the onset, evolution, and final pattern of the strain localization, affecting the homogenized response. In specific fine-scale models with a low number of possible localization patterns predetermined by weak zones in which fracture appears and propagates, traditional boundary conditions can yield satisfactory results. Examples include a masonry fine-scale model with periodic boundary conditions [24] and a concrete fine-scale model with cohesive bands at predefined position constrained by minimal kinematic boundary conditions [27]. However, for fine-scale models with complex internal and boundary structures that allow diverse localization patterns to develop, the periodic boundary conditions, commonly adopted in homogenization simulations, lead to artificially increased ductility of the homogenized response whenever the band’s normal direction is not parallel to the periodicity direction. This has motivated developments of alternative definitions of boundary conditions which have been investigated in several studies employing continuous fine-scale models [8, 21, 27].
In this work, the following three novel approaches are investigated. (i) The Percolation-path-aligned BCs introduced by Coenen, Kouznetsova, and Geers [8] and (ii) the Tessellation BCs developed by Goldmann, Brummund, and Ulbricht [21] are suitable for cuboidal fine-scale models. The former is based on rotation of the periodicity frame, while the latter is based on its shifting. Due to the cuboidal shape, its fracture response is inherently orientation dependent, even if the boundary conditions allow arbitrary strain localization. As a remedy, a circular/spherical fine-scale model can be used instead [20]. However, traditional periodic boundary conditions forbid fully developed strain localization in circular models, as there is no unique periodicity direction. (iii) To solve this, a modification to the Periodic BCs was proposed by Hofman, Ke, and Meer [22], introducing an unknown displacement jump between two halves of the circle, based on the approach described in Ref. [27].
An essential input for all of these methods is the orientation of the developing localization band. Since generally this orientation is not known a priori, it must usually be determined as part of the nonlinear solution procedure. Here, the problem is simplified by considering fine-scale models loaded by uniaxial tension with fixed direction, and the localization band is assumed to develop in a normal direction to the load.
The novel approaches, which were originally proposed for continuous two-dimensional fine-scale models, are herein implemented for discrete mechanical models. A 2D numerical study is performed to compare results obtained using the novel BCs with those calculated with traditional BCs, namely the Periodic and Minimal kinematic BCs. The differences between square and circular model geometries are also discussed. Fundamental issues of the Percolation-aligned and the periodic BCs modified with displacement jump are revealed. Their causes are discussed, with a modified variant of the former approach introduced to demonstrate the mechanism. Following the results of the 2D study, the Percolation-path-aligned and the Tessellation BCs approaches are extended to 3D. This extension exposes several implementation challenges, which are described in the Appendices. Finally, a 3D numerical study analogous to the 2D study is conducted.
Homogenization of heterogeneous Cauchy continuum
The asymptotic expansion homogenization, pioneered by Bensoussan, Lions, and Papanicolaou [2] and Sánchez-Palencia [31], introduces two different length scales. is the structural size, often taken as the maximum distance of two points within the domain; is the size of the periodic fine-scale model. The homogenization explores the asymptotic limit for which the ratio between and (also known as separation of scales constant) approaches zero: . There are also two different reference systems: the coarse-scale one denoted and the fine-scale one . The spatial derivative changes to and all variables become dependent on both macro and micro reference systems .
All variables are, in the limit , expanded into infinite series with terms with increasing power of . For example, the displacement field becomes
The strain tensor is the symmetric part of the displacement gradient . Combining the two-scale differentiation with the expanded displacements yields
The expansion of the stress field follows . Defining the constitutive function and the state variables , the first term reads
The balance equation in the two scale expanded framework reads
The terms containing the lowest power of provide the following balance equation . Its solution under the -periodicity of the displacement field gives , , and constantness of in reference system. We can therefore write the next stress term and proceed to the next balance equation.
These equations describe the fine-scale problem to be solved numerically after adding a constraint defined by the periodic boundary conditions. This fine-scale model is the standard mechanical model loaded by the negative coarse-scale strain tensor as eigenstrain. The solution to this problem are fluctuating displacements emerging from the heterogeneity. They must be zero on average. The periodicity of the fine-scale model solution can be enforced either “strongly”, if the boundary nodes have periodic pairs, or by using a weak periodicity scheme in the case of non-periodic boundary meshes [25].
The next balance equation gives a macroscopic formulation corresponding to Cauchy continuum. It also provides the macroscopic stress tensor definition
where is the domain of the fine-scale problem, is its volume, and is the boundary. The unit outward normal will be denoted .
This article is focused to the fine-scale model only, therefore the notation is simplified by dropping the indices. The fine-scale fluctuation displacement will be written as , the fine-scale strain will be represented by plain , and the fine-scale stress will become . The coarse-scale strain will be denoted .
Periodic boundary conditions
The fine-scale model’s boundary is partitioned into two segments: and . The periodic boundary conditions (PBCs) can be applied if each point is mapped to a point by a bijective mapping [21]
Here, the dA symbols refer to the boundary segments associated with the corresponding boundary nodes. For every corresponding pair of boundary nodes, the fine-scale displacement vectors must coincide.
In addition, rigid-body translation of the entire fine-scale domain must be eliminated, for instance by enforcing zero average displacement or by fixing the translations of a single arbitrarily selected node.
Instead of imposing the macroscopic strain by eigenstrain, one can change equation Eq. (8) so that the periodic boundary node pairs must move relative to each other by vector . The updated equation Eq. (8) then reads
This approach is equivalent to periodic BCs with eigenstrains, and is sometimes also called periodic boundary conditions. All the equations hereinafter that refer to some form of periodic boundary conditions might be actually implemented as Eq. (8) with eigenstrains or Eq. (9) without them.
Minimal kinematic boundary conditions
The Minimal kinematic boundary conditions (hereafter referred to as Minimal BCs where appropriate) are derived from the requirement that the average strain over the fine-scale domain should be zero [26]
Defining the fine-scale strain as the symmetric part of the displacement gradient , Equation (10) can be transformed, through application of the divergence theorem, into an integral over the fine-scale domain boundary
When applied to a rectangular or cuboid domain, each pair of opposite sides and must obey the following conditions
for ( is the number of dimensions). In addition, using the same approach as in the periodic boundary conditions, the rigid body translation of the whole fine-scale domain needs to be prevented.
Percolation-path-aligned boundary conditions
The Percolation-path-aligned BCs were developed in Ref. [8]. They will be denoted as Aligned BCs hereinafter. In this formulation, the periodicity relations of the boundary sections and are applied in a coordinate system aligned with the localization pattern. Rather than using the original and directions, the periodic constraints are imposed along the rotated axes and . Assuming that, for each node , a corresponding node exists on the opposite boundary in the direction, and that each node has a counterpart in the direction, the periodic relations illustrated in Figure 1a can be established. The coordinate system is rotated by an angle corresponding to the localization-band orientation. The projection of the boundary along the direction, subsequently denoted as the crack direction, is further divided into two types of segments: and , for which the outward normal vectors of the + and - sections are oriented oppositely, and and , where they are not.
Figure 1. a) Projection of the boundaries of the fine-scale model in the through-crack direction and parallel-to-crack direction in the Percolation-path-aligned BC formulation; b) Mapping offset and the associated parallel-to-crack projection in the Tessellation BCs for crack angle ; red and blue lines indicate crack with the same orientation angle but different position within the fine-scale model
In the through-crack direction , the periodic constraints (14) are imposed only on the boundary segments . For the segments , only the weak condition defined by Eq. (14b) is imposed. According to [8], this avoids overconstraining the fine-scale model. Along the direction , parallel to the localization band, periodicity is enforced on all corresponding boundary segments regardless of their normal vectors.
for
for
for
In practice, the assumption that every node possesses a corresponding periodic image is generally not satisfied. As a result, additional procedures, described in Appendix § 6, are required to formulate the constraints for arbitrary nodal arrangements. Furthermore, rigid-body motion of the fine-scale model must be eliminated, which is achieved by constraining the translational degrees of freedom of a randomly selected node.
The extension to three dimensions is straightforward. However, while both directions and are uniquely defined by the crack normal in two dimensions, in 3D, on the other-hand, a crack with normal only defines the direction; the other orthonormal directions must be chosen from an infinite number of options. The procedure for identifying optimal parallel-to-crack directions used in this work is detailed in Appendix § 7.
Tessellation boundary conditions
The Tessellation boundary conditions were introduced by Goldmann, Brummund, and Ulbricht [21]. Unlike Percolation-path-aligned BCs, the Tessellation BCs ensure gapless tiling of a space by the deformed fine-scale domains. This is achieved by shifting adjacent domains by an offset , enabling the localization band to continue seamlessly across domain boundaries while avoiding the artificial crack-pattern repetition imposed by standard periodic BCs. This is depicted in Figure 1b. For a square two-dimensional domain, the offset or , is defined by the localization-band inclination angle and model size .
As in the case of the Percolation-path-aligned BCs, the periodic relations are formulated in a crack-aligned coordinate system. For , the boundary segment is partitioned into two parts and the periodic pairs are defined as shown in Figure 1b: is shifted by and by . Then periodic boundary conditions are prescribed between the positive (+) and negative (−) sides.
As in the case of the Percolation-path-aligned BCs, the shifted boundary typically does not have node pairs in the inclined periodicity direction. The displacement periodicity on non-periodic nodes is again defined by the techniques described in Appendix § 6. As before, rigid-body translation of the entire fine-scale model is eliminated by fixing the translations of a randomly selected node.
When extended to 3D, the shifting is performed in two orthogonal directions, alongside two of the three opposing cube face pairs. The pair, in which a face’s normal has the lowest direction difference to the crack normal (evaluated via the vector product), is prescribed non-shifted periodic boundary conditions. In the other two directions, the offsets are calculated analogously to Eq. (17) and the shifted periodic constraints are constructed using the procedure described in Appendix § 6.
Circular fine-scale model
The circular fine-scale model, discussed, for example, in Ref. [20], is intrinsically orientation independent because of its shape. To apply the periodic boundary conditions, which ensure consistency between the scales, the fine-scale displacements of pairs of points with anti-periodic normals are coupled analogously to the square model (Eq. (8) with eigenstrain or Eq. (9)).
Because these boundary conditions prevent the localization band from crossing the boundary, a modification was developed that introduces a discontinuity in the periodicity constraints in [22]. In this scheme, upon strain localization onset, an unknown discontinuity is added to the constraint ((8) or (9)), defined as the relative displacement of the two nodes of the opening node pair. The opening node pair is determined by the crack normal and is defined as the pair of nodes located on opposite sides of the intersection between the crack surface and the boundary, as shown in Figure 7c. In 2D, either of the two intersections may be chosen arbitrarily. The addition of the jump allows for the separation of the two halves of the circular model, as illustrated in Figure 2b), provided that the localization band crosses the boundary at the opening node pair. To allow unloading of the bulk material with positive , a correction strain term is defined as
Figure 2. a) Opposing (periodic) point pair on circular fine-scale model boundary; b) the same fine-scale model with the unknown jump calculated as difference in displacement of the opening node pair , defined by the crack normal .
Figure 7. a) Square model loaded by uniaxial tensile strain under angle ; b) circle model loaded identically; c) detail of the opening node pair
where is the area of the circular model with radius , is the effective crack length. These augmented periodic boundary conditions, hereafter denoted Embedded-crack boundary conditions, take the form of the following boundary pair constraint
where the left-hand-side equation is updated Eq. (8) for which the imposed eigenstrain changes to . The right-hand-side equation represents the second approach previously given by Eq. (9).
In the original formulation, the displacement jump is introduced into the boundary conditions only at the onset of localization, thereby requiring its explicit identification. In the present work, for simplicity, the jump is included from the beginning of the simulations, with negligible influence on the elastic response or the onset of softening. Furthermore, in contrast to Ref. [22], where unbreakable inclusions restricted certain crack orientations, the present approach allows the localization band to form between any boundary nodes. Consequently, the distinction between effective and average crack normals, as introduced in Ref. [22], is not required here.
Modeling framework
The discrete fine-scale models, representing concrete at the mesoscale, are mechanical systems comprising interconnected rigid particles with constitutive behavior defined on inter-particle contacts [5]. The process of generating a geometry of the model inside the prescribed shape (here a square, circle, or cube) starts with random placement of circular/spherical inclusions with diameters distributed according to the Fuller curve with exponent 0.5. The placement starts with the largest inclusions, acceptance of each new one is conditioned by no overlap with previously approved inclusions. The weighted Delaunay triangulation and power tessellation on these inclusions then provides the convex polyhedral particles (rigid bodies) with movements governed by translational, , and rotational, , degrees of freedom defined in the centers of the inclusions. The contact between the rigid particles is characterized by the area, length, and a local coordinate system . One such contact is sketched in Figure 3. Such a way of creating the model internal structure has been pioneered by Bolander and Saito [4] using the Voronoi tessellation.
Figure 3. Two neighboring rigid particles and planar contact face with local orthonormal reference system.
A single integration point per contact is defined at the centroid of the contact face. Assuming small displacements and rotations, the rigid body kinematic equation provides strain measures () at the integration point evaluated as discrete displacement jump between the rigid bodies divided by contact length. The vectorial constitutive model formulated in the local reference system then provides traction at the integration point. The constitutive model is a modified version of material behavior originally introduced in Ref. [9] and later refined in the Lattice-Discrete Particle Model (LDPM) articles [10, 40, 36, 3]. The exact constitutive formulation used here can be found in Refs. [11, 15, 28]. The fundamental equations are completed by balance equations for linear and angular momentum for each rigid body.
Constant material parameters are used for all simulations, with values: MPa, , MPa, Jm. The material parameters are homogeneous throughout the model geometry and the heterogeneity of the stress field is a consequence of heterogeneous internal structure only.
Boundary geometry
The square model geometries were generated in two variants that differed in the tessellation at the boundaries. The smooth surface (S) variant featured auxiliary nodes located irregularly at the boundaries. These nodes are placed in a periodic manner on the opposite sides. The normal of the boundary is constant for each side of the square domain. One such internal structure is shown in Figure 4a. The smooth surface results in a boundary or wall effect [12]: the boundary regions exhibit different stiffness compared to the bulk of the fine-scale model due to the structured tessellation. In contrast, models with rough surface (R) eliminate this issue by keeping the tessellation in the boundary region statistically identical to the bulk. In the latter case, the placement of the particles in the domain allows inclusions protruding from the square domain, but restricts overlapping with the previously placed inclusions as well as with their periodic images. The power tessellation is performed on the original inclusions and their periodic images. This results in a periodic internal structure that is not affected by the boundaries [12, 14]. An example of such rough surface model is shown in Figure 4b.
Figure 4. 2D fine-scale model of a) square shape with smooth surface (S); b) square shape with rough surface (R); c) circular shape with smooth surface (C); blue lines depict inter-particle elements and black lines depict contact faces.
The boundary nodes on the rough surface are often poorly connected to the remaining structure and become unstable unless standard Periodic boundary conditions are applied. The strong periodicity in the rotated frame suffices as well, but the Minimal kinematic BCs and the boundary part of the Aligned BCs produce a singular stiffness matrix or excessive displacements of these nodes even in elastic regimes. Additionally, the projections of such boundaries lead to overlap or change of order of the boundary nodes, further preventing the use of the Aligned BCs with these models. In contrast, the Tessellation BCs can still be applied, as they enforce strong periodicity for the whole boundary. Although shifting the boundary can cause occasional overlap and change in order of the boundary nodes, particularly for large , this effect did not significantly influence the results for the geometries considered in the present study.
Homogenization of discrete systems
Rezakhani and Cusatis [29] and Rezakhani, Zhou, and Cusatis [30] were the first to develop homogenization techniques for discrete systems. Their work was further extended in Refs. [14, 13]. The major difference from standard continuum models is the presence of rotational degrees of freedom. Nevertheless, the coarse scale still homogenizes to the Cauchy continuum under the assumption of realistic bending stiffness [13], while at the fine-scale the standard discrete model is used. This result follows the pioneering work of Forest, Pradel, and Sab [16], which showed that the Cosserat continuum at a fine scale results in a Cauchy continuum on a coarse scale under the same assumption. The symmetric coarse-scale strain tensor is still driving the fine-scale problem. For truly periodic boundary conditions (Eq. (8)), it serves as an eigen-strain tensor projected on individual discrete contacts. In the local coordinate system of each element given by the unit vectors ( in 3D), this eigen-strain reads
The expression for the coarse-scale stress tensor given in Eq. (6) can be rewritten as
where the outer summation visits all discrete contacts within the fine-scale model.
The boundary conditions described in Sect. 2 are directly applicable for translational degrees of freedom. However, the BCs variants need to be extended to account for rotations as well. The periodic boundary conditions simply connect all the rotations of each dependent pair [13].
This strong rotational periodicity on appropriate pairs of boundary nodes is then enforced whenever strong periodicity is prescribed. In situations when only weak periodicity applies (Minimal kinematic BCs and in Percolation-path-aligned BCs) the rotations are not subjected to any constraint. Unlike translations, the average fine-scale rotation does not need to vanish [16, 13].
Numerical results
The numerical studies are conducted predominantly in two dimensions. Only selected cases are extended to three dimensions. All the simulations are performed in open-source software called Open Academic Solver (OAS, https://gitlab.com/kelidas/OAS).
Elastic properties of 2D fine-scale models
The effective macroscopic elastic properties of the fine-scale models are known to be dependent on the boundary conditions, particularly for models with low number of grains, as documented, e.g., in Ref. [26]. With an increasing number of grains, achieved here by enlarging the domain while keeping the same grain size distribution, it is expected that the elastic properties converge to a single set of values for given microscopic material parameters.
To verify this behavior for the discrete models at hand, a following analysis was performed: square models with smooth (S) and rough (R) surface were generated with side lengths of 200, 400, 600, and 1000 mm. For each size and boundary type, 10 geometries were created. In addition, for each size, 10 circular (C) models with the same area as the corresponding square models were generated. Then, for each model, the effective Young’s modulus and Poisson’s ratio were calculated as follows: the model, under purely elastic conditions, was subjected to a set of independent macroscopic strain states (two unit axial strains and unit shear) and for each of them the homogenized stress response was calculated. These stress–strain pairs were used to assemble the homogenized matrix of elastic constants. Assuming an isotropic macroscopic response, the effective Young’s modulus and Poisson’s ratio are determined by least-squares fitting the theoretical isotropic matrix to the numerically calculated matrix of material constants.
The effective parameters were calculated for the S and C models using the Periodic and the Minimal BCs. For the models with rough surface, the Minimal boundary conditions cannot be used, as discussed in § 3.1. The mean values and standard deviations of coarse-scale elastic constants, computed from 10 realizations for each model size and type of boundary conditions, are shown in Figure 5. As expected, the small model sizes with Minimal kinematic BCs exhibit lower stiffness than those with Periodic BCs. This difference decreases as the model size increases. Circular and periodic boundary models show reduced variability in elastic parameters in all sizes. For small model sizes, the advantage of the R (and also C) models is evident, as they suffer the stiffness overestimation to a much lesser degree compared to the S model. This is surprising for C models as they also have structured smooth boundary geometry.
Figure 5. Mean and standard deviation of effective Young’s modulus and Poisson’s ratio for different sizes of 2D fine-scale models with smooth (S), rough (R) and circular (C) surface and Periodic and Minimal BCs
Before conducting the fracture analysis, the novel boundary conditions in the elastic regime were verified. In the original formulation of the Embedded-crack BCs, here applied to the circular models, the displacement jump is introduced only at the strain localization onset. Therefore, in elasticity, they coincide with Periodic BCs and their influence on elastic response needs not be investigated. The Aligned and Tessellation BCs were applied to the sets of S and R models of size 200 mm at different angles from 0 to . The effective elastic parameters were calculated and normalized by the reference values calculated with the Periodic BCs. The results are shown in Figure 6.
Figure 6. Effective Young’s modulus and Poisson’s ratio normalized by their reference values dependent on the rotation of the reference frame for models with smooth (S) and rough (R) surface and Aligned and Tessellation BCs
The Tessellation BCs on S models influence the effective elastic properties minimally compared to Periodic ones. The Aligned BCs, which can only be applied to S models, show lower stiffness for values of between 0 and and and . This is likely related to the lower number of constraints at these values caused by projection of the borders, compared to the Periodic BCs. At , the number of constraints is very close to the Periodic BCs and the elastic constants almost coincide. The Tessellation BCs applied to the R models show lower stiffness for most of the directions. This is also caused by the lower number of integration zones (and, therefore, constraints) due to the projection of the nonuniform boundary geometry. The lower values at the 0, and angles, where the Tessellation BCs should equal Periodic BCs, arise from numerical inaccuracies in the Lagrange multiplier implementation and ambiguous corner node assignment to one or the other side of the rectangular domain.
In all cases, the maximum difference is close to 2.5%. Therefore, all the boundary conditions are considered to be valid for linear elasticity.
Fracture in 2D fine-scale models
In this section, the strain localization under different types of periodic boundary conditions is analyzed. Twenty periodic square fine-scale models with a side length of mm were generated. The individual realizations differ in the spatial distribution of the virtual circular aggregates and in the associated tessellation. In all cases, the square domain was oriented so that its sides were parallel to the and axes.
The models were subjected to loading through projection of the coarse-scale strain tensor describing uniaxial tension in different directions (Figure 7a). The normal of the assumed crack, which defines the Percolation-path-aligned and Tessellation BCs, was taken to be parallel to the tensile loading direction. The principal strain , the only nonzero component of , was increased from zero to in 73 pseudo-time steps.
The orientation of the applied load, defined by the angle , was varied from 0 to in 31 steps. These limits reflect that the problem is periodic in the statistical sense. The homogenized coarse-scale stress and the total internal energy were recorded and subsequently averaged over the 20 geometries. Average stress-strain curves for two selected loading directions are presented in Figure 11. On the horizontal axis the normal strain in the loading direction (i.e., the principal strain ) is plotted, while the vertical axis shows the normal stress in the same direction: and with denoting the loading direction unit vector. The reference curve corresponds to the average response obtained with the Periodic BCs at . For , as well as for and , the localization band can develop perpendicular to the loading direction without violating periodicity of the Periodic BCs. In the absence of boundary-condition-induced artifacts, all other loading angles and boundary-condition formulations would be expected to produce identical results.
Figure 11. Average stress-strain responses obtained from 20 two-dimensional square fine-scale models with side length mm subjected to uniaxial tension. Results are shown for loading angles (left) and (right). The reference response corresponds to the models with Periodic BCs loaded at .
The boundary-condition formulations are compared using three quantities: the average total energy , the average peak stress , and the average strain corresponding to the peak stress. Because the simulations are typically terminated in state not corresponding to the stress-free crack, the reported energy includes both dissipated and residual strain-energy contributions, with the latter typically representing only a minor fraction. Relative energy values as function of the loading angle, normalized by the average energy obtained from the reference simulations, are reported in Figure 8. Figure 9 shows the peak stresses and corresponding strains, also normalized by the reference model results.
Figure 8. Average relative total energy for different BC types and crack angle from 20 square 2D fine-scale square models with mm: left-hand side: Smooth surface; right-hand side: Rough surface
Figure 9. Average relative peak stress (left-hand side) and normalized strain at peak stress (right-hand side) for different BC types and varying crack angle from 20 two-dimensional square fine-scale models of size
The results obtained with the Periodic BCs are consistent with observations reported in previous studies [22, 33]. For loading directions other than , and , increase in the total energy is observed. This stems from the periodicity constraints, which force the formation of repeating crack patterns composed of multiple localization bands that are not orthogonal to the straining direction. As the total crack length increases, additional energy is required for their development, resulting in a more ductile response of the fine-scale models. The effect reaches its maximum at and , where the total energy exceeds that of the reference case by more than a factor of two. The localization patterns exhibit spurious branching and incorrect inclination for most load orientations, as illustrated for in Figure 10a.

Figure 10. Crack patterns in 2D square fine-scale model with size mm loaded in uniaxial tension under angle , with (a) Periodic, (b) Aligned, (c) Tessellation, and (d) Minimal boundary conditions. Coloring represents the crack opening, ranging from 0 (blue) to 0.19 mm (red).
As noted in Refs. [8, 21], Minimal kinematic BCs are unsuitable for problems involving strain localization. Owing to their weak kinematic constraints, localization tends to develop spuriously along the boundaries of the fine-scale domain, where detachment of a small number of boundary nodes is energetically more favorable than the formation of a localization band across the domain. Figure 10d illustrates a typical example of this behavior. The short localization band leads to low energy dissipation and low peak stress and corresponding strain compared to the reference values, as shown in Figure 8 and Figure 9.
The Percolation-path-aligned BCs were designed to accommodate a single localization band regardless of the loading orientation , which should result in an energy dissipation nearly independent of and close to the reference value. Contrary to this expectation, significantly higher energy values were obtained for . Moreover, the onset of localization exhibits a pronounced dependence on the loading orientation, reflected in the peak stresses and corresponding strains shown in Figure 9. Both quantities are markedly lower than those of the reference solution, as also demonstrated by the stress-strain responses in Figure 11. The cause is the faster crack opening in region with strong periodicity compared to the region with weak periodicity. The opening is driven by releasing the strain energy from unloading parts, but the size of the unloading parts differ and, hence, the energy flux. In the region, the size of the material domain measured in the direction is , but in the region it is an average value much smaller than . The model either opens a second crack in the region or accounts for uneven crack opening by large shear strains. Which of these situations occurs depends on the internal heterogeneity, the loading angle, and the material parameters. In the models studied, the second crack often occurred for . When using a lower value of the elastic parameter , no secondary cracks were detected at any angle.
To substantiate the proposed explanation of the orientation dependence observed for the Aligned BCs, an alternative formulation referred to as Aligned-weak BCs was considered. In this approach, the weak condition associated with is applied along the entire boundary, meaning that the periodicity in the direction is enforced only through a single integral constraint. This eliminates the occurrence of multiple localization bands and the corresponding energy increase. It, however, leads to spurious localization similar to that observed for the Minimal kinematic BCs when is not close to , as the constraints imposed on are insufficient to prevent it.
By construction, the Tesselation BCs give rise to a single localization band for all values of . Owing to the shifted periodic constraints, this band may appear segmented as it exits and re-enters the fine-scale domain, as illustrated in Figure 10c. The length of the localization band varies with its inclination and reaches a maximum for . The energy dissipation is proportional to the relative length of the crack . For square domains, it reads [34]
where is the factor introduced in Ref. [34] to account for this geometric effect. Figure 8 demonstrates close agreement between this factor and the normalized dissipated energy under the Tessellation BCs. Theoretically, the influence of the square domain could be corrected for during post-processing using the factor. The Tessellation BCs can be viewed as true periodic boundary conditions in a rotated frame of reference, where the original square domain of size is reshaped to a rectangular one with dimensions perpendicularly and parallel to the crack normal.
The circular fine-scale model, considered only in the smooth surface variant, is generated in 10 different geometries inside a circle of diameter mm (its area is equal to the area of the square models). The same analysis is performed, consisting of an increasing uniaxial tension load in a set of directions defined by angle . Standard periodic boundary conditions prevent a localization pattern from crossing the boundary, resulting in the average total energy at the end of the load being higher compared to the reference value obtained in square models, as evidenced in Figure 12. The same reasoning as before is valid also here: the straining acts over regions of different lengths. Therefore, the energy flux available for crack opening is the largest in the center of the circle and zero at the boundary. The uneven opening of the crack is compensated for by opening the second crack. Its inability to separate the model into two parts is also manifested in the nonzero residual stress seen in the strain-stress diagram in Figure 12. The resulting crack patterns for two different load angles are visualized in Figure 13a,Figure 13c, showing that the main crack suffers from uneven opening and another crack might be opened.
Figure 12. Left-hand side: Relative energy averaged over 10 circular models with Periodic BCs compared to 2 different circular models with the Embedded-crack BCs; right-hand side: Stress-strain diagram showing the difference in response of two circular models to the same load, depending on whether the localization happens between the opening nodes (Geo 2) or not (Geo 1).

Figure 13. Crack patterns in 2D circular fine-scale model of size mm, loaded by tension under different angles and boundary conditions: a) Periodic; b) Embedded-crack; c) Periodic; d) Embedded-crack. Coloring represents the crack opening, ranging from 0 (blue) to 0.27 mm (red).
When Embedded-crack BCs are applied, the crack can develop evenly across the whole fine-scale model under any loading angle. However, the correct failure mode occurs only if the emerging crack localizes between the opening node pair , defined by the vector . This situation is shown in Figure 13b, while the crack developed in the model of identical internal structure and loading direction with the standard Periodic BCs is shown in Figure 13a. When the same model is loaded at different angles , the standard Periodic BCs might lead to a crack pattern located far from the center (Figure 13c). Consequently, also Embedded-crack BCs produce problematic results as the opening node pair is not connected to the main crack (Figure 13d). Whether one or the other scenario occurs is a random process depending on the internal geometry of the model. The consequences of these scenarios on the coarse-scale mechanical response are shown in Figure 12. The total energy and stress is close to the average reference solution when the localization band occurs between the opening node pair, or close to the average solution with Periodic BCs, when not. Two different internal structures (Geo 1 & 2) show that the solution randomly oscillates between these scenarios depending on angle . The stress-strain response for one selected angle on the right-hand side shows that correct localization scenario (Geo 2) provides results similar to reference ones, while incorrect scenario (Geo 1) results are similar results with standard Periodic BCs.
Fracture in 3D fine-scale models
All studies of periodic boundary conditions allowing strain localization were done in two dimensions only. Here, the Tessellation and Aligned BC types are implemented in 3D cube models, with the accompanying challenges described in Appendices § 6 and § 7. Similarly to the 2D study, 10 random fine-scale geometries are generated within a cube of size . Only smooth surfaces are considered; models with rough surfaces pose additional challenges for extending the model to 3D, and the Minimal and Aligned BCs remain inapplicable to rough surfaces also in 3D.
Cubic models are loaded by projection of the uniaxial tensile strain tensor in different directions. The normal of the assumed crack surface is considered identical to the direction of the tensile load , which is given by the azimuth angle measured from the axis and the polar angle measured from the axis. The maximum value of the only nonzero principal strain component is reached in 50 pseudo-time increments. The angles and vary between 0 and , with 85 loading orientations, shown by small circles in Figure 14, Figure 15 and Figure 16.
Figure 14. Left-hand side: factor expressing theoretical area of inclined crack in a cube of size 1; right-hand side: average relative energy computed by cubic fine-scale model of size mm with Periodic BCs
Figure 15. Average relative energy computed by cubic fine-scale model of size mm with Periodic BCs: Tessellation BCs (left-hand side) and Aligned BCs (right-hand side)
Figure 16. Average relative strain at peak stress computed by cubic fine-scale model of size : left-hand side: Periodic BCs (left-hand side) and Aligned BCs (right-hand side)
For the 3D cube, a scaling factor analogous to that in Eq. (24) can be derived, resulting in (23). The inverse of this factor gives the area of a single planar crack in a unit cube. The inverses of the three expressions in Eq. (25) correspond to the area of the projections of the three orthogonal cube faces onto the crack plane defined by the azimuth and polar angles. The largest of three values, corresponding to the smallest area, represents the actual area of the crack. The boundaries separating the three cases are obtained by equating the expressions for each pair of cases.
The maximum relative crack area is found for and . The energy dissipation of a single crack should be proportional to the area of the crack; therefore, we expect the models to give energy dissipation according to factor.
Figure 14 shows the total energy for the Periodic BCs normalized by the energy of the reference case obtained for and under the same boundary conditions. The values between the circles, where the results are extracted from the mechanical models, are obtained by linear interpolation. Similarly to the 2D case, energy provided by the standard Periodic BCs increases for all crack orientations other than those perpendicular to the cube sides. An example of a resulting crack pattern with multiple opened cracks driving the increase in energy is shown in Figure 18a. The highest energies are about 2.5 larger than in the reference case. The values along the edges , , and of the plot correspond to a loading in which the crack plane is perpendicular to two opposing sides of the cube. These cases should reproduce the behavior of two-dimensional square fine-scale model shown in Figure 8, and indeed they do. In terms of peak stress, the results are similar to those in 2D, with the peak values being close to the reference solution. However, the strain value corresponding to the peak stress tends to be significantly higher for load orientations close to the diagonal case of and , as seen in Figure 16.
Figure 18. Crack patterns in cubic fine-scale model of size loaded by tension: a) , , Periodic BCs; b) , , Aligned BCs; c) , , Tessellation BCs
The Aligned BCs often lead to energy values closer to those from the reference case, however, for some orientations they exceed it significantly. The reason is the same as in two dimensions; the uneven crack opening due to different strain energy available for cracking results in secondary cracks in the region. In other cases, the total energy decreases below the reference value, pointing to spurious localization occurring in the weakly periodic boundary segment. An example of such a crack pattern is shown in Figure 18b. Once again, the results of the 2D-like loading seen at the edges , , and match the results of 2D simulations (Figure 8), apart from the central values, equivalent to case in 2D, where in 3D spurious localization lowers below the reference value. The peak stress is close to the reference value for all loading orientations, except for the cases of spurious localization discussed earlier. This contrasts with the 2D case, where most orientations yielded substantially lower values (Figure 9). The likely cause is the stabilization of the model by the third direction of constraints. Probably for the same reason, the strain values corresponding to the peak stress state are also below the reference value only in spurious localization cases and otherwise are higher, as seen in Figure 16. The Aligned BCs also lead to a higher variance in response across the different geometries in the model set compared to other BC variants. This can be seen in the stress-strain diagrams in Figure 17, where the standard deviation is plotted around the mean response of 10 internal structures. Diagrams for two loading directions are shown: the one on the left-hand side resulted for Aligned BCs in a single localization band, while the other one shows a case of spurious localization.
Figure 17. Average stress-strain response and its standard deviation computed from over 10 random geometries of cubic fine-scale model with size under uniaxial tensile load direction , (left-hand side) and , (right-hand side)
The Tessellation boundary conditions show promising results also in the three-dimensional case. Once again, they produce a single crack for all load orientations. An example of a single crack, leaving and re-entering the cube model, resulting in three crack segments is shown in Figure 18c. As seen in Figure 15, the energy depends on the orientation of the crack in a way consistent with the factor, with a maximum value close to the orientation of the diagonal crack at and . The results do not match the factor as well as in the 2D case, which is likely caused by the discretization of the boundary conditions and the small size of the model. In terms of peak stress and the corresponding strain values, the results are similar to those of the Aligned BCs shown earlier, but there is no spurious localization. The variance of the response across different internal structures is similar to the Periodic BCs, see Figure 17.
The results of the Minimal kinematic BCs and the weaker version of Aligned BCs are not presented. They are expected to also suffer from spurious localization in three dimensions. Similarly, the defects of the Embedded-crack BCs constraining the circular model are expected to be manifested also in spherical models.
Constraint complexity and computational cost
The computational cost associated with the investigated boundary conditions can be divided into two components: constraint generation and their influence on the iterative equilibrium calculation.
The standard periodic constraints are directly defined by the model geometry and require no additional preprocessing. The generation of the Aligned and Tessellation BCs can be costly, especially in 3D, depending on the number of boundary nodes and the projection tessellation cell size (Appendix § 6). The Tessellation BCs are advantageous in this regard because one of the boundary projections retains the predefined periodic constraints. In our case with predefined crack normal, the constraints are generated only once and prior to the simulation run. Moreover, the associated computational cost is negligible compared to the cost of the simulation itself.
In the square and cubical models, the constraints are imposed in the form of Lagrange multipliers. The standard periodic BCs represent the highest number of constraints, while the other variants result in the same or lower number, equal to the number of the integration zones. The maximum differences in the number of constraints are roughly around 20%, therefore, the computational cost per iteration is comparable among all variants. The total computational cost therefore primarily changes with the number of iterations required by the NewtonRaphson solver. Figure 19 shows that only the Tessellation BCs lead to identical number of iterations irrespectively of the loading direction thanks to the unobstructed development of single straight localization band.
Figure 19. Average total iterations during uniaxial loading simulation; left-hand side: smooth-surface square models; right-hand side: circular models
For the circular model, introducing a displacement jump into the periodic boundary conditions does not increase the number of constraints and any change in per-iteration cost is negligible. Figure 19 shows that no significant difference between the BC variants was found in terms of average number of total iteration over the model sets.
Conclusions
Three recently proposed periodic boundary condition (BC) formulations, designed to mitigate the load-orientation dependence of the homogenized response of softening fine-scale models, were implemented for discrete mechanical models. The formulations of the Percolation-path-aligned BC and Tessellation BC variants were extended to three-dimensions, with the implementation procedures presented in detail.
An analysis of the macroscopic response and localization patterns of square, cubic, and circular fine-scale discrete particle models loaded in uniaxial tension with various forms of boundary conditions was performed, including novel BC variants and traditional BC formulations. The following conclusions are drawn.
Standard Periodic boundary conditions (BCs) produce a strongly orientation dependent softening response of the fine-scale model. The Periodic BCs lead to correctly developed localization band only for a few load orientations in the square and cube models, in circular models they forbid proper band development for all load orientations. The Minimal kinematic [26] cannot be used with softening models at all due to the spurious localization into few boundary elements.
The Percolation-path-aligned BCs [8], which rotate the periodicity frame with the localization band, have been applied to the square and cube models. It was shown that for a certain range of load orientations, the difference in straining between its weakly and strongly periodic segments leads to multiple localization bands, which cause overestimation of dissipated energy and overly ductile behavior. This was evidenced by the introduction of the weak modification, which replaces the strongly periodic segment with a the weakly periodic one. This variant does not suffer from the aforementioned issue, however, because the weakly constrained segment can be prone to spurious localization observed with Minimal kinematic BCs, this variant is also not viable. While both these issues might not manifest for different geometries or material parameters, they are inherent to these boundary conditions.
The Tessellation BCs [21], which shift the periodicity frame so that it can accommodate inclined localization bands without periodic repetitions, have proven to offer a consistent response when applied to square models with uniform and periodic boundaries, as well as to cubic models. A single localization band, which may exit and reenter the model, develops for any load orientation. The dissipated energy in two dimensions is proportional to the crack area, which depends solely on the geometry of the domain and can be easily calculated, allowing for potential post-processing corrections. In three dimensions, the proportionality is weaker but still reasonable. The Tessellation BCs were found to offer a significant improvement over standard Periodic BCs.
The Embedded-crack periodic BCs [22], which introduce a displacement jump in the boundary conditions for circular domains, allow a full localization band separating two halves of the model. However, this is conditioned by the initiation of the crack in the central part of the circle. The crack may then connect with the displacement jump at the boundary. When the crack initiates outside the domain center, these boundary conditions produce multiple localization bands and dissipate an excessive amount of energy.
The present conclusions apply to the discrete models considered here, which assume homogeneous material parameters and monotonic uniaxial tensile loading. In other types of models, such as continuum models with a heterogeneous distribution of material parameters, additional complications might appear (for example, stiff inclusions combined with shifted periodicity might prevent strain localization [17]. Furthermore, the applicability of the studied boundary conditions in more complex loading scenarios is uncertain. Further work would require a robust technique to determine the crack normal at the onset of localization, which sets the rotated periodicity frame and therefore constraints further crack propagation.
Discretization of Boundary Projections and Constraints
Definition
The periodicity of the fine-scale model boundary nodes does not conform to the rotated periodic frame that must be imposed in many of the boundary conditions studied. To accommodate the rotated periodic frame, both and are partitioned into zones. Each such zone contains exactly one node on one side (either + or −) and at least one node on the opposite side. This ensures that all nodes are constrained and that the number of constraints is as close as possible to the number of constraints in the periodic model.
Algorithmically, the zones are created by the following procedure. Firstly, groups of + and − nodes are found. Each node on the + side finds the closest (in the projection) node on the − side, denoted the central node (possibly more + nodes are attached to the same central node). Then, each − node that was not previously connected to a + node finds the closest + node. This can lead to clusters with multiple + and − nodes. To separate such a group, the links between the central node and the + nodes are kept only if the + node is not connected to any other − node. In case no such + node exists, only the shortest link is kept.
With the groups completed, the projection area is divided into a number of cells. In 2D models, these cells are small line sections of equal size, in 3D models small rectangular cells are used. Then each cell is assigned to the closest node in the projection (not distinguishing the + and − sides). Cells that belong to a single connected group of nodes then constitute the integration zone .
The strong periodicity of translations and rotations in zone is prescribed by the integral condition
The integral is evaluated numerically. In 2D, linear interpolation of displacements between nodes is assumed, and integration points lie in the center between two nodes or nodes and the zone boundary. This is depicted in Figure 20a. A simpler variant is used in 3D models. The integration points lie in the centers of the cells and their displacement value is taken from the closest node. This approximate constraint was verified for elastic material behavior, where it yields practically identical results as the standard periodic models aligned with the periodic frame.
Figure 20. a) Boundary nodes on + (green) and - (blue) side of 1D projection in 2D models, connected into groups (dashed lines), around which the integration zones (thick black lines) are built from cells (thin gray lines). The linear displacement field and are numerically integrated in the integration points (red); b) the same principle applied in a 2D projection of a 3D model, integration points are hidden to improve image clarity
Note that if only one zone is considered for the whole boundary, the integral periodic boundary conditions become the Minimal kinematic BCs, except of course the rotations that are left unconstrained under Minimal kinematic BCs.
The constraints are prescribed in the form of Lagrange multipliers, and a direct solver is used to ensure that they are exactly satisfied.
Choice of Directions Parallel to the Crack in 3D
Three projection directions must be selected for the Aligned BCs when applied to the cubic domain, so that it corresponds to the Periodic BCs when crack normal is perpendicular to the cube sides. The first () is given by the loading direction, the other two ( and ) might be constrained to create orthonormal system with the first one (), but there are still infinitely many options for them.
In this work, and are selected such that the two opposing sides are projected to a line along each of them. Therefore, these sides are omitted when defining the constraints in the and directions. The total number of constraints decreases and becomes similar to that of periodic BCs. Such directions can be obtained as a cross product of the crack normal and the model’s side normals, here coincident with the unit vectors of a Cartesian basis , , and . This yields three directions , , . These projections are shown in Figure 21. Two of these three parallel-to-crack directions need to be chosen so that the total number of projections is three. The most convenient pair was chosen to minimize the difference in the density of nodes on the and parts of the boundary. The uneven densities of the nodes degrade the quality of the periodic constraint imposed on the opposite boundary parts. Technically, this is done by dropping the vector that is closest to some of the cube side normals. Note that this means that and are generally only orthogonal with and not with each other, however, this was found to not be an issue. Additionally, in case of crack normal perpendicular to the cube sides, , and correspond to , , and (not necessarily in this order), leading to Aligned BCs corresponding with Periodic BCs.
Figure 21. Boundary nodes a cubical fine-scale model; a) 3D view with a possible crack location and shape in Aligned BCs; b) through-crack projection along the vector, with black and white nodes lying in the boundary section and the rest in the section; c), d), e) parallel to crack projections along , , vectors
Acknowledgement
The authors gratefully acknowledge financial support from the Czech Science Foundation under project number GA24-11845S. The simulations were performed on computational servers acquired within project INODIN (Innovative methods of materials diagnostics and monitoring of engineering infrastructure to increase its durability and service time – CZ.02.01.01/00/23_020/0008487) co-funded by European Union. The authors also acknowledge inspiring discussions with Milan Jirásek and Martin Doškář from Czech Technical University.
Data availability
The input files for OAS solver and results presented in this article, together with Python scripts used for generation of the Aligned and Tessellation BCs, are available at https://doi.org/10.5281/zenodo.20415173.
References
- [1]Ted Belytschko and Jeong-Hoon Song. “Coarse-graining of multiscale crack propagation”. In: International Journal for Numerical Methods in Engineering 81.5 (2010), pp. 537–563. DOI: 10.1002/nme.2694.DOI
- [2]Alain Bensoussan, Jacques-Louis Lions, and George Papanicolaou. Asymptotic Analysis for Periodic Structures. Vol. 5. Studies in Mathematics and its Applications. Amsterdam: North-Holland, 1978.
- [3]T. Bhaduri, S. Gomaa, and M. Alnaggar. “Coupled Experimental and Computational Investigation of the Interplay Between Discrete and Continuous Reinforcement in Ultra-High Performance Concrete Beams. II: Mesoscale Modeling”. In: J Eng Mech 147 (9 2021), p. 04021050. ISSN: 0733-9399. DOI: 10.1061/(ASCE)EM.1943-7889.0001941.
- [4]J. E. Bolander and S. Saito. “Fracture analysis using spring network with random geometry”. In: Engineering Fracture Mechanics 61(5-6) (1998), pp. 569–591.
- [5]John E. Bolander et al. “Discrete mechanical models of concrete fracture”. In: Engineering Fracture Mechanics 257 (2021), p. 108030. ISSN: 0013-7944. DOI: 10.1016/j.engfracmech.2021.108030.DOI
- [6]E. Bosco, V. G. Kouznetsova, and M. G. D. Geers. “Multi-scale computational homogenization-localization for propagating discontinuities using X-FEM”. In: International Journal for Numerical Methods in Engineering 102.3-4 (2015), pp. 496–527. DOI: 10.1002/nme.4838.
- [7]E. Bosco et al. “A multiscale framework for localizing microstructures towards the onset of macroscopic discontinuity”. In: Computational Mechanics 54 (2014), pp. 299–319. ISSN: 1432-0924. DOI: 10.1007/s00466-014-0986-4.DOI
- [8]Erica Coenen, Varvara Kouznetsova, and Marc Geers. “Novel boundary conditions for strain localization analyses in microstructural volume elements”. In: International Journal for Numerical Methods in Engineering 90 (2012), pp. 1–21. DOI: 10.1002/nme.3298.DOI
- [9]Gianluca Cusatis and Luigi Cedolin. “Two-scale study of concrete fracturing behavior”. In: Engineering Fracture Mechanics 74.1 (2007). Fracture of Concrete Materials and Structures, pp. 3–17. ISSN: 0013-7944. DOI: https://doi.org/10.1016/j.engfracmech.2006.01.021. URL: https://www.sciencedirect.com/science/article/pii/S0013794406000440.
- [10]Gianluca Cusatis, Daniele Pelessone, and Andrea Mencarelli. “Lattice discrete particle model (LDPM) for failure behavior of concrete. I: Theory”. In: Cement and Concrete Composites 33.9 (2011), pp. 881–890. DOI: 10.1016/j.cemconcomp.2011.02.011.
- [11]Jan Eliáš. “Adaptive Technique for Discrete Models of Fracture”. In: International Journal of Solids and Structures 100–101 (2016), pp. 376–387. ISSN: 0020-7683. DOI: 10.1016/j.ijsolstr.2016.09.008.DOI
- [12]Jan Eliáš. “Boundary Layer Effect on Behavior of Discrete Models”. In: Materials 10.2 (2017). ISSN: 1996-1944. DOI: 10.3390/ma10020157. URL: https://www.mdpi.com/1996-1944/10/2/157.DOI
- [13]Jan Eliáš and Gianluca Cusatis. “Do Discrete Fine-Scale Mechanical Models with Rotational Degrees of Freedom Homogenize Into a Cosserat or a Cauchy Continuum?” In: Journal of the Mechanics and Physics of Solids 207 (2026), p. 106422. ISSN: 0022-5096. DOI: 10.1016/j.jmps.2025.106422.DOI
- [14]Jan Eliáš and Gianluca Cusatis. “Homogenization of discrete mesoscale model of concrete for coupled mass transport and mechanics by asymptotic expansion”. In: Journal of the Mechanics and Physics of Solids 167 (2022), p. 105010. ISSN: 0022-5096. DOI: 10.1016/j.jmps.2022.105010.DOI
- [15]Jan Eliáš and Gianluca Cusatis. “Macroscopic Stress, Couple Stress and Flux Tensors Derived Through Energetic Equivalence from Microscopic Continuous and Discrete Heterogeneous Finite Representative Volumes”. In: Computer Methods in Applied Mechanics and Engineering 436 (2025), p. 117688. ISSN: 0045-7825. DOI: 10.1016/j.cma.2024.117688.DOI
- [16]Samuel Forest, Francis Pradel, and Karam Sab. “Asymptotic analysis of heterogeneous Cosserat media”. In: International Journal of Solids and Structures 38.26 (2001), pp. 4585–4608. ISSN: 0020-7683. DOI: 10.1016/S0020-7683(00)00295-X.DOI
- [17]Erik Giesen Loo and Frans P. van der Meer. “Stress-controlled weakly periodic boundary conditions: Axial stress under varying orientations”. In: International Journal for Numerical Methods in Engineering 121.19 (2020), pp. 4458–4470. DOI: https://doi.org/10.1002/nme.6441. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/nme.6441. URL: https://onlinelibrary.wiley.com/doi/abs/10.1002/nme.6441.
- [18]I.M. Gitman, H. Askes, and L.J. Sluys. “Coupled-volume multi-scale modelling of quasi-brittle material”. In: European Journal of Mechanics - A/Solids 27.3 (2008), pp. 302–327. ISSN: 0997-7538. DOI: https://doi.org/10.1016/j.euromechsol.2007.10.004. URL: https://www.sciencedirect.com/science/article/pii/S0997753807000976.
- [19]I.M. Gitman, H. Askes, and L.J. Sluys. “Representative volume: Existence and size determination”. In: Engineering Fracture Mechanics 74.16 (2007), pp. 2518–2534. ISSN: 0013-7944. DOI: https://doi.org/10.1016/j.engfracmech.2006.12.021. URL: https://www.sciencedirect.com/science/article/pii/S0013794406004772.
- [20]R. Glüge, M. Weber, and A. Bertram. “Comparison of spherical and cubical statistical volume elements with respect to convergence, anisotropy, and localization behavior”. In: Computational Materials Science 63 (2012), pp. 91–104. ISSN: 0927-0256. DOI: https://doi.org/10.1016/j.commatsci.2012.05.063. URL: https://www.sciencedirect.com/science/article/pii/S0927025612003448.
- [21]Joseph Goldmann, Jörg Brummund, and Volker Ulbricht. “On boundary conditions for homogenization of volume elements undergoing localization”. In: International Journal for Numerical Methods in Engineering 113.1 (2018), pp. 1–21. DOI: https://doi.org/10.1002/nme.5597. URL: https://onlinelibrary.wiley.com/doi/abs/10.1002/nme.5597.
- [22]Pieter Hofman, Lu Ke, and Frans P. van der Meer. “Circular representative volume elements for strain localization problems”. In: International Journal for Numerical Methods in Engineering 124.4 (2023), pp. 784–807. DOI: https://doi.org/10.1002/nme.7142. URL: https://onlinelibrary.wiley.com/doi/abs/10.1002/nme.7142.
- [23]Lu Ke and Frans P van der Meer. “A computational homogenization framework with enhanced localization criterion for macroscopic cohesive failure in heterogeneous materials”. In: Computer Methods in Applied Mechanics and Engineering 2 (2022), pp. 1–28. DOI: 10.46298/jtcam.7707.DOI
- [24]Emma La Malfa Ribolla, Giuseppe Giambanco, and Antonino Spada. “Mesoscopic aspects of the computational homogenization with meshless modeling for masonry material”. In: International Journal for Numerical Methods in Engineering 121.16 (2020), pp. 3610–3635. DOI: https://doi.org/10.1002/nme.6373. URL: https://onlinelibrary.wiley.com/doi/abs/10.1002/nme.6373.
- [25]F. Larsson et al. “Computational homogenization based on a weak format of micro-periodicity for RVE-problems”. In: Computer Methods in Applied Mechanics and Engineering 200.1 (2011), pp. 11–26. ISSN: 0045-7825. DOI: https://doi.org/10.1016/j.cma.2010.06.023. URL: https://www.sciencedirect.com/science/article/pii/S0045782510001908.
- [26]Sinisa Mesarovic and Jagan Padbidri. “Minimal kinematic boundary conditions for simulations of disordered microstructures”. In: Philosophical Magazine 85 (Jan. 2005), pp. 65–78. DOI: 10.1080/14786430412331313321.DOI
- [27]J. Oliver et al. “Continuum approach to computational multiscale modeling of propagating fracture”. In: Computer Methods in Applied Mechanics and Engineering 294 (2015), pp. 384–427. ISSN: 0045-7825. DOI: 10.1016/j.cma.2015.05.012.DOI
- [28]Jan Raisinger et al. “Simulating heterogeneity within elastic and inelastic discrete mechanical models”. In: Engineering Fracture Mechanics 326 (2025), p. 111362. ISSN: 0013-7944. DOI: https://doi.org/10.1016/j.engfracmech.2025.111362. URL: https://www.sciencedirect.com/science/article/pii/S0013794425005636.
- [29]Roozbeh Rezakhani and Gianluca Cusatis. “Asymptotic expansion homogenization of discrete fine-scale models with rotational degrees of freedom for the simulation of quasi-brittle materials”. In: Journal of the Mechanics and Physics of Solids 88 (2016), pp. 320–345. ISSN: 0022-5096. DOI: 10.1016/j.jmps.2016.01.001.DOI
- [30]Roozbeh Rezakhani, Xinwei Zhou, and Gianluca Cusatis. “Adaptive multiscale homogenization of the lattice discrete particle model for the analysis of damage and fracture in concrete”. In: International Journal of Solids and Structures 125 (2017), pp. 50–67. ISSN: 0020-7683. DOI: h10.1016/j.ijsolstr.2017.07.016.arxiv.org/abs/1702.00695
- [31]Enrique Sánchez-Palencia. Non-Homogeneous Media and Vibration Theory. Vol. 127. Lecture Notes in Physics. Berlin: Springer-Verlag, 1980.
- [32]Harpreet Singh and Puneet Mahajan. “Strain localization in reduced-order asymptotic homogenization”. In: Mathematics and Mechanics of Solids 25 (4 2019), pp. 913–936. ISSN: 1081-2865. DOI: 10.1177/1081286519892655.DOI
- [33]Jan Stransky and Milan Jirásek. “Calibration of particle-based models using cells with periodic boundary conditions”. In: Oct. 2011.
- [34]Erik Svenning, Fredrik Larsson, and Martin Fagerström. “Two-scale modeling of fracturing solids using a smeared macro-to-micro discontinuity transition”. In: Comput. Mech. 60.4 (Oct. 2017), pp. 627–641. ISSN: 0178-7675. DOI: 10.1007/s00466-017-1426-z. URL: https://doi.org/10.1007/s00466-017-1426-z.
- [35]S. Toro et al. “A two-scale failure model for heterogeneous materials: numerical implementation based on the finite element method”. In: International Journal for Numerical Methods in Engineering 97.5 (2014), pp. 313–351. DOI: 10.1002/nme.4576.DOI
- [36]Matthew Troemner, Erol Lale, and Gianluca Cusatis. “Lattice discrete particle model simulations of energetic size effect and its implications for shear design specifications of reinforced concrete squat walls”. In: Engineering Structures 322 (2025), p. 119085. ISSN: 0141-0296. DOI: 10.1016/j.engstruct.2024.119085.DOI
- [37]Sergio Turteltaub et al. “Multiscale analysis of mixed-mode fracture and effective traction-separation relations for composite materials”. In: Journal of the Mechanics and Physics of Solids 117 (2018), pp. 88–109. ISSN: 0022-5096. DOI: 10.1016/j.jmps.2018.04.009.DOI
- [38]Jörg F. Unger. “An FE2-FX1 approach for multiscale localization phenomena”. In: Journal of the Mechanics and Physics of Solids 61.4 (2013), pp. 928–948. ISSN: 0022-5096. DOI: 10.1016/j.jmps.2012.12.010.DOI
- [39]Yingbo Zhu, John Brigham, and Alessandro Fascetti. “Data-driven multiscale lattice discrete particle model for digital twin modeling of concrete structures”. In: Computer Methods in Applied Mechanics and Engineering 445 (2025), p. 118183. ISSN: 0045-7825. DOI: 10.1016/j.cma.2025.118183.DOI
- [40]Zhongfeng Zhu et al. “Lattice Discrete Particle Modeling of concrete under cyclic tension–compression with multi-axial confinement”. In: Construction and Building Materials 352 (2022), p. 128985. ISSN: 0950-0618. DOI: 10.1016/j.conbuildmat.2022.128985.DOI










