Analytical Bounds for Composite Homogenization: A Reference
Contents
This post collects every major analytical and semi-analytical model for the effective elastic moduli of two-phase composites in one place. It is intended as a reference hub: other posts on this site link here for derivations and DOI-linked primary sources. For numerical homogenization (FFT-based), see the FFT Homogenization post.
Overview
The table below summarises all models covered. They are grouped by the strength of their theoretical foundation.
| Model | Type | Key assumption |
|---|---|---|
| Voigt | Exact upper bound | Uniform strain |
| Reuss | Exact lower bound | Uniform stress |
| Hashin-Shtrikman | Tightest variational bounds | Statistically isotropic, spherical inclusions |
| Dilute | Estimate, $v_f \to 0$ | No inclusion interactions |
| Mori-Tanaka | Estimate (= HS lower) | Mean-field; inclusions see matrix average strain |
| Self-consistent | Estimate | Inclusion embedded in unknown effective medium |
| Halpin-Tsai | Semi-empirical | Fitted reinforcement parameter $\xi$ |
| Ponte Castañeda-Willis | Higher-order bounds | Spatial pair-distribution of inclusions |
The Two-Phase RVE
The reference problem is a single spherical inclusion (fiber/particle) of radius $a$ embedded in an infinite matrix. The surrounding matrix is taken as the Representative Volume Element (RVE). This is the exact geometry for which the Eshelby, HS, MT and self-consistent solutions are derived.
The volume fraction of the inclusion phase is
\[v_f = \frac{V_{\rm inclusion}}{V_{\rm RVE}}\]and the matrix occupies $v_m = 1 - v_f$. All models express the effective elastic moduli as functions of $v_f$, the phase moduli $E_f, E_m$, and the phase Poisson ratios $\nu_f, \nu_m$.
Phase bulk and shear moduli:
\[K_i = \frac{E_i}{3(1-2\nu_i)}, \qquad G_i = \frac{E_i}{2(1+\nu_i)} \tag{1}\]Voigt and Reuss Bounds
The Voigt and Reuss bounds are the widest possible rigorous bounds — valid for any microstructure geometry.
Longitudinal modulus E1
Under a strain applied along the fiber axis (direction 1), iso-strain (Voigt) is exact for an ideally bonded continuous fiber:
\[E_1^{\rm V} = v_f E_f + (1 - v_f) E_m \tag{2}\]The iso-stress (Reuss) lower bound:
\[\frac{1}{E_1^{\rm R}} = \frac{v_f}{E_f} + \frac{1 - v_f}{E_m} \tag{3}\]For continuous fibers, the rule of mixtures (Voigt) is the practically correct prediction for $E_1$.
Transverse modulus E2
Applied transversely (direction 2), the same mathematical expressions bound $E_2$:
\[E_2^{\rm V} = v_f E_f + (1 - v_f) E_m \tag{4}\] \[\frac{1}{E_2^{\rm R}} = \frac{v_f}{E_f} + \frac{1 - v_f}{E_m} \tag{5}\]These bounds are very wide for typical fiber/matrix stiffness ratios. For a tight estimate of $E_2$, use the Hashin-Shtrikman bounds below.
References: Voigt [1]; Reuss [2]
See also: Interactive comparison chart — all models on one plot.
Hashin-Shtrikman Bounds
The Hashin-Shtrikman (HS) bounds [4] are the tightest possible bounds for a statistically isotropic microstructure of spherical inclusions, given only $v_f$ and the phase moduli. They are derived by a variational principle using a polarisation field.
For a 3D isotropic two-phase composite with spherical inclusions, the bounds on bulk modulus $K$ and shear modulus $G$ are:
HS lower bounds (assuming $K_f > K_m$, $G_f > G_m$):
\[K^{-} = K_m + \frac{v_f}{\dfrac{1}{K_f - K_m} + \dfrac{3(1-v_f)}{3K_m + 4G_m}} \tag{6}\] \[G^{-} = G_m + \frac{v_f}{\dfrac{1}{G_f - G_m} + \dfrac{6(1-v_f)(K_m + 2G_m)}{5G_m(3K_m + 4G_m)}} \tag{7}\]HS upper bounds:
\[K^{+} = K_f + \frac{1-v_f}{\dfrac{1}{K_m - K_f} + \dfrac{3v_f}{3K_f + 4G_f}} \tag{8}\] \[G^{+} = G_f + \frac{1-v_f}{\dfrac{1}{G_m - G_f} + \dfrac{6v_f(K_f + 2G_f)}{5G_f(3K_f + 4G_f)}} \tag{9}\]Effective Young’s modulus from $K^{\pm}$ and $G^{\pm}$:
\[E^{\pm} = \frac{9K^{\pm}G^{\pm}}{3K^{\pm} + G^{\pm}} \tag{10}\]References: Hashin & Shtrikman [4]; Hill [5]
The Eshelby Tensor
Eshelby’s Inclusion Problem
Eshelby [3] showed that an ellipsoidal inclusion that would undergo a stress-free eigenstrain $\boldsymbol{\varepsilon}^*$ if unconstrained instead develops a uniform constrained strain when embedded in an infinite matrix:
\[\varepsilon^c_{ij} = S_{ijkl}\, \varepsilon^*_{kl} \tag{11}\]The fourth-order tensor $\mathbb{S}$ depends only on the shape of the inclusion and the Poisson ratio $\nu_m$ of the surrounding matrix — not on the moduli contrast. This makes it the universal kernel in every inclusion-based homogenisation model.
Components for a Sphere
For a spherical inclusion embedded in an isotropic matrix with Poisson ratio $\nu_m$, all non-zero independent Eshelby tensor components have closed-form expressions:
\[S_{1111} = S_{2222} = S_{3333} = \frac{7 - 5\nu_m}{15(1-\nu_m)} \tag{12}\] \[S_{1122} = S_{2233} = S_{1133} = \frac{5\nu_m - 1}{15(1-\nu_m)} \tag{13}\] \[S_{1212} = S_{2323} = S_{1313} = \frac{4 - 5\nu_m}{15(1-\nu_m)} \tag{14}\]All other components are zero. The hydrostatic and deviatoric projections are:
\[S_{\rm hyd} = S_{iikk}/9 = \frac{1+\nu_m}{3(1-\nu_m)} \tag{15}\] \[S_{\rm dev} = S_{1212} = \frac{4 - 5\nu_m}{15(1-\nu_m)} \tag{16}\]From these components the HS scalar factors follow: $\beta_K = 3K_m/(3K_m + 4G_m)$ and $\beta_G = G_m(9K_m + 8G_m)/[6G_m(K_m + 2G_m)]$, which appear in the HS bound denominators (eqs. 6–9) and in the dilute-limit expressions (eqs. 19–20).
Full derivation for finite aspect ratio (prolate spheroid, $e = 1$ to $\infty$): see the Eshelby Tensor post.
References: Eshelby [3]; Mura [9]
Mori-Tanaka
The Mori-Tanaka (MT) model [6] places each inclusion in the average strain of the matrix rather than the far-field applied strain, accounting for inclusion-inclusion interactions in a mean-field sense.
For spherical inclusions with $K_f > K_m$, $G_f > G_m$:
\[K^{\rm MT} = K_m + \frac{v_f}{\dfrac{1}{K_f - K_m} + \dfrac{3(1-v_f)}{3K_m + 4G_m}} \tag{17}\] \[G^{\rm MT} = G_m + \frac{v_f}{\dfrac{1}{G_f - G_m} + \dfrac{6(1-v_f)(K_m + 2G_m)}{5G_m(3K_m + 4G_m)}} \tag{18}\]MT = HS lower bound: Equations (17)–(18) are algebraically identical to the HS lower bounds (6)–(7). Benveniste [10] proved this rigorously: when inclusions are stiffer than the matrix, the Mori-Tanaka estimate coincides exactly with the HS lower bound.
References: Mori & Tanaka [6]; Benveniste [10]
Dilute Limit
When $v_f \to 0$, each inclusion is embedded in the unperturbed matrix strain field. The effective bulk and shear moduli are:
\[K^{\rm dil} = K_m + v_f \frac{(K_f - K_m)(3K_m + 4G_m)}{3K_f + 4G_m} \tag{19}\] \[G^{\rm dil} = G_m + v_f \frac{5G_m(G_f - G_m)(3K_m + 4G_m)}{G_m(9K_m + 8G_m) + 6G_f(K_m + 2G_m)} \tag{20}\]These expressions are valid for $v_f \lesssim 0.1$. At higher volume fractions, inclusion interactions are significant and the dilute estimate overpredicts $E_{\rm eff}$.
Self-Consistent Scheme
Hill [7] proposed embedding the inclusion in the unknown effective medium rather than the matrix. The effective moduli satisfy the implicit equations:
\[K^{\rm SC} = \left[\frac{v_f K_f}{K_f + \frac{4}{3}G^{\rm SC}} + \frac{v_m K_m}{K_m + \frac{4}{3}G^{\rm SC}}\right] \left[\frac{v_f}{K_f + \frac{4}{3}G^{\rm SC}} + \frac{v_m}{K_m + \frac{4}{3}G^{\rm SC}}\right]^{-1} \tag{21}\] \[G^{\rm SC} = \left[\frac{v_f G_f}{G_f + \beta^{\rm SC}} + \frac{v_m G_m}{G_m + \beta^{\rm SC}}\right] \left[\frac{v_f}{G_f + \beta^{\rm SC}} + \frac{v_m}{G_m + \beta^{\rm SC}}\right]^{-1} \tag{22}\]where
\[\beta^{\rm SC} = \frac{G^{\rm SC}(9K^{\rm SC} + 8G^{\rm SC})}{6(K^{\rm SC} + 2G^{\rm SC})} \tag{23}\]These equations are solved iteratively (typically 20–100 iterations from an MT initial guess). The self-consistent scheme is symmetric in the two phases and is most appropriate for polycrystal-like microstructures where neither phase forms a continuous matrix. It can violate the HS bounds at extreme volume fractions or high phase contrast ($E_f/E_m \gtrsim 100$).
Reference: Hill [7]
Halpin-Tsai
The Halpin-Tsai equations [8] are semi-empirical, with a reinforcement parameter $\xi$ fitted to geometry:
\[E^{\rm HT} = E_m \frac{1 + \xi\,\eta\, v_f}{1 - \eta\, v_f}, \qquad \eta = \frac{E_f/E_m - 1}{E_f/E_m + \xi} \tag{24}\]Recommended values of $\xi$:
| Geometry | Property | $\xi$ |
|---|---|---|
| Spherical inclusion | $E$ | $2$ |
| Circular fiber (infinite) | $E_2$ | $2$ |
| Short fiber (aspect ratio $l/d$) | $E_2$ | $2l/d$ |
| Any fiber | $G_{12}$ | $1$ (use $G_f/G_m$) |
| Any fiber | $E_1$ | $\to\infty$ (recovers Voigt) |
The Halpin-Tsai formula has no rigorous micromechanical derivation; it is accurate to ~5% for $v_f \lesssim 0.6$ at typical glass/epoxy contrast ($E_f/E_m \approx 20$).
Reference: Halpin & Kardos [8]
Higher-Order: Ponte Castaneda-Willis
The Ponte Castañeda-Willis (PCW) bounds [12] tighten the HS bounds by incorporating the spatial distribution of inclusions through a second morphological descriptor: the distribution tensor $\mathbb{P}^d$, which characterises the shape of the “security zone” — the excluded volume around each inclusion.
For a composite with spherical inclusions and a spherical spatial distribution (inclusions arranged with isotropic pair statistics, non-overlapping), $\mathbb{P}^d$ takes the same isotropic form as the Eshelby polarisation tensor evaluated at the matrix moduli. The PCW bounds on bulk and shear modulus are:
\[K^{\rm PCW,\pm} = K_{\rm ref}^{\pm} + \frac{v_f (K_f - K_{\rm ref}^{\pm}) (K_m - K_{\rm ref}^{\pm})}{v_f (K_m - K_{\rm ref}^{\pm}) + v_m (K_f - K_{\rm ref}^{\pm}) + \beta_K^{\pm}(K_f - K_{\rm ref}^{\pm})(K_m - K_{\rm ref}^{\pm}) / \delta_K^{\pm}} \tag{25}\]where the reference moduli $K_{\rm ref}^{\pm}$ and $G_{\rm ref}^{\pm}$ are the HS reference moduli (stiffer phase for upper, softer for lower), and the scalar factors for spherical geometry are:
\[\beta_K = \frac{3K_{\rm ref} + 4G_{\rm ref}}{3(K_{\rm ref} + \tfrac{4}{3}G_{\rm ref})} = 1, \qquad \delta_K = K_{\rm ref} + \tfrac{4}{3}G_{\rm ref} \tag{26}\]For spherical inclusions with a spherical spatial distribution, the PCW lower bound equals the HS lower bound (PCW lower = HS lower = MT). The PCW upper bound uses $K_{\rm ref} = K_f$, $G_{\rm ref} = G_f$, with “inclusions” = matrix phase at volume fraction $(1-v_f)$:
\[K^{\rm PCW,-} = K^{\rm HS,-} \tag{27}\] \[K^{\rm PCW,+} = K^{\rm HS,+} - \frac{(1-v_f)^2(K_m - K_f)^2 / \alpha_f}{\alpha_f + (1-v_f)(K_m - K_f)}, \qquad \alpha_f = K_f + \tfrac{4}{3}G_f \tag{28}\] \[G^{\rm PCW,+} = G^{\rm HS,+} - \frac{(1-v_f)^2(G_m - G_f)^2 \beta_f}{[1 + (1-v_f)(G_m - G_f)\beta_f]^2}, \qquad \beta_f = \frac{6(K_f+2G_f)}{5G_f(3K_f+4G_f)} \tag{29}\]The PCW upper bound is strictly tighter than the HS upper bound for $0 < v_f < 1$:
\[K^{\rm PCW,+} \leq K^{\rm HS,+}, \qquad G^{\rm PCW,+} \leq G^{\rm HS,+} \tag{30}\]The correction vanishes at $v_f = 0$ and $v_f = 1$, reaching its maximum at intermediate $v_f$ (roughly 0.3–0.6).
Reference: Ponte Castañeda & Willis [12]
Deep dive: How Microstructure Arrangement Affects Composite Bounds: A PCW Investigation — interactive 3D scenes showing isotropic, layered, columnar and periodic microstructures with live PCW bound charts for varying $w$.
Interactive Comparison
All models are computed in the browser. Adjust phase properties; charts update live. The self-consistent result is obtained by iterating eqs. (21)–(23) to convergence (tol = 10⁻⁸). The PCW upper bound (eqs. 28–29) is tighter than the HS upper; the PCW lower = HS lower (eq. 27).
Python Reference Implementation
import numpy as np
def phase_KG(E, nu):
K = E / (3*(1 - 2*nu))
G = E / (2*(1 + nu))
return K, G
def E_from_KG(K, G):
return 9*K*G / (3*K + G)
def hs_bounds(Ef, Em, nuf, num, vf):
"""Hashin-Shtrikman bounds for 3D isotropic composite (spherical inclusions)."""
Kf, Gf = phase_KG(Ef, nuf)
Km, Gm = phase_KG(Em, num)
vm = 1 - vf
K_lo = Km + vf / (1/(Kf-Km) + 3*vm/(3*Km+4*Gm))
K_hi = Kf + vm / (1/(Km-Kf) + 3*vf/(3*Kf+4*Gf))
G_lo = Gm + vf / (1/(Gf-Gm) + 6*vm*(Km+2*Gm)/(5*Gm*(3*Km+4*Gm)))
G_hi = Gf + vm / (1/(Gm-Gf) + 6*vf*(Kf+2*Gf)/(5*Gf*(3*Kf+4*Gf)))
return (E_from_KG(K_lo, G_lo), E_from_KG(K_hi, G_hi),
K_lo, K_hi, G_lo, G_hi)
def dilute(Ef, Em, nuf, num, vf):
"""Dilute limit (valid for vf < 0.1)."""
Kf, Gf = phase_KG(Ef, nuf)
Km, Gm = phase_KG(Em, num)
K = Km + vf*(Kf-Km)*(3*Km+4*Gm) / (3*Kf+4*Gm)
G = Gm + vf*5*Gm*(Gf-Gm)*(3*Km+4*Gm) / (Gm*(9*Km+8*Gm) + 6*Gf*(Km+2*Gm))
return E_from_KG(K, G)
def self_consistent(Ef, Em, nuf, num, vf, tol=1e-8, max_iter=300):
"""
Self-consistent scheme (Hill 1965).
Fixed-point iteration starting from Mori-Tanaka initial guess.
"""
Kf, Gf = phase_KG(Ef, nuf)
Km, Gm = phase_KG(Em, num)
vm = 1 - vf
# Initial guess: HS lower (= MT)
K = Km + vf / (1/(Kf-Km) + 3*vm/(3*Km+4*Gm))
G = Gm + vf / (1/(Gf-Gm) + 6*vm*(Km+2*Gm)/(5*Gm*(3*Km+4*Gm)))
for _ in range(max_iter):
alpha = 4*G/3
beta = G*(9*K + 8*G) / (6*(K + 2*G))
K_new = (vf*Kf/(Kf+alpha) + vm*Km/(Km+alpha)) / (vf/(Kf+alpha) + vm/(Km+alpha))
G_new = (vf*Gf/(Gf+beta) + vm*Gm/(Gm+beta)) / (vf/(Gf+beta) + vm/(Gm+beta))
if abs(K_new - K) < tol and abs(G_new - G) < tol:
K, G = K_new, G_new
break
K, G = K_new, G_new
return E_from_KG(K, G)
def pcw_upper(Ef, Em, nuf, num, vf):
"""
Ponte Castaneda-Willis upper bound (PCW 1995) for spherical inclusions
with a spherical (isotropic) spatial distribution.
PCW lower = HS lower (no correction); PCW upper is tighter than HS upper.
"""
Kf, Gf = phase_KG(Ef, nuf)
Km, Gm = phase_KG(Em, num)
vm = 1 - vf
# Bulk modulus upper bound (ref = fiber; eq. 28)
K_hs_hi = Kf + vm / (1/(Km-Kf) + 3*vf/(3*Kf+4*Gf))
alphaF = Kf + 4*Gf/3
num_cK = vm*vm*(Km-Kf)**2 / alphaF
den_cK = alphaF + vm*(Km-Kf)
K_pcw = K_hs_hi - num_cK / den_cK
# Shear modulus upper bound (ref = fiber; eq. 29)
betaF = 6*(Kf + 2*Gf) / (5*Gf*(3*Kf + 4*Gf))
G_hs_hi = Gf + vm / (1/(Gm-Gf) + 6*vf*(Kf+2*Gf)/(5*Gf*(3*Kf+4*Gf)))
dGmf = Gm - Gf
num_cG = vm*vm*dGmf**2 * betaF
den_cG = (1 + vm*dGmf*betaF)**2 / betaF
G_pcw = G_hs_hi - num_cG / den_cG
return E_from_KG(K_pcw, G_pcw), K_pcw, G_pcw
def halpin_tsai(Ef, Em, vf, xi=2):
"""Halpin-Tsai (xi=2 for sphere/circular fiber; xi=2*l/d for short fiber)."""
eta = (Ef/Em - 1) / (Ef/Em + xi)
return Em * (1 + xi*eta*vf) / (1 - eta*vf)
# Example: glass spheres (Ef=74 GPa, nuf=0.22) in epoxy (Em=3.5 GPa, num=0.35)
vf = np.linspace(0, 0.7, 200)
Elo, Ehi, *_ = hs_bounds(74e3, 3.5e3, 0.22, 0.35, vf)
Esc = np.array([self_consistent(74e3, 3.5e3, 0.22, 0.35, v) for v in vf])
Eht = halpin_tsai(74e3, 3.5e3, vf)
Epcw = np.array([pcw_upper(74e3, 3.5e3, 0.22, 0.35, v)[0] for v in vf])
idx = 60 # vf ~ 0.30
print(f"At vf=0.30: HS [{Elo[idx]:.0f}, {Ehi[idx]:.0f}] MPa")
print(f" PCW [{Elo[idx]:.0f}, {Epcw[idx]:.0f}] MPa (lower unchanged)")
print(f" SC = {Esc[idx]:.0f} MPa")
print(f" HT = {Eht[idx]:.0f} MPa")
Conclusion and Recommendations
Which method should you use?
| Situation | Recommended model | Reason |
|---|---|---|
| Need a guaranteed bound, any microstructure | Voigt + Reuss | Widest but always valid |
| Statistically isotropic matrix-inclusion composite | Hashin-Shtrikman lower bound | Tightest rigorous bound; equals MT |
| Engineering estimate, moderate $v_f$ | Halpin-Tsai ($\xi=2$) | Simple, calibrated, accurate to ~5% |
| Symmetric two-phase (neither phase dominant) | Self-consistent (Hill) | Treats both phases equally |
| Very low $v_f$ ($< 0.10$) | Dilute limit | Exact in dilute regime |
| Tightest upper bound, isotropic distribution | Ponte Castañeda-Willis upper | Strictly tighter than HS upper; PCW lower = HS lower |
| Any geometry, full-field accuracy | FFT homogenization | Numerical, exact for periodic cell |
Key results from the charts
- HS lower = Mori-Tanaka when inclusions are stiffer than the matrix — a single curve covers both.
- Self-consistent sits close to the HS lower bound at low $v_f$ but rises above it at moderate $v_f$, and can exit the HS band at very high volume fractions.
- Halpin-Tsai closely tracks the HS band centre for typical glass/epoxy contrast ($E_f/E_m \approx 20$). At higher contrast it tends toward the HS upper bound.
- PCW upper bound is visibly tighter than HS upper at intermediate $v_f$ (0.3–0.6). The PCW lower bound equals HS lower for an isotropic spatial distribution.
- Dilute limit diverges from MT/HS above $v_f \approx 0.10$ — do not use it outside this range.
- For a continuous fiber composite, $E_1$ is predicted almost exactly by Voigt (rule of mixtures) regardless of model. Model choice matters almost exclusively for transverse properties.
Practical guidance
Halpin-Tsai is the simplest engineering estimate; Mori-Tanaka / HS lower gives a tighter rigorous bound. FFT homogenization gives full-field accuracy for periodic RVEs. The self-consistent scheme is standard for polycrystal plasticity but rarely appropriate for matrix-inclusion composites.
References
[1] Voigt, W. (1889). Über die Beziehung zwischen den beiden Elasticitätsconstanten isotroper Körper. Annalen der Physik, 274(12), 573–587.
[2] Reuss, A. (1929). Berechnung der Fließgrenze von Mischkristallen auf Grund der Plastizitätsbedingung für Einkristalle. ZAMM, 9(1), 49–58. doi:10.1002/zamm.19290090104
[3] Eshelby, J. D. (1957). The determination of the elastic field of an ellipsoidal inclusion, and related problems. Proceedings of the Royal Society A, 241(1226), 376–396. doi:10.1098/rspa.1957.0133
[4] Hashin, Z. & Shtrikman, S. (1963). A variational approach to the theory of the elastic behaviour of multiphase materials. Journal of the Mechanics and Physics of Solids, 11(2), 127–140. doi:10.1016/0022-5096(63)90060-7
[5] Hill, R. (1963). Elastic properties of reinforced solids: some theoretical principles. Journal of the Mechanics and Physics of Solids, 11(5), 357–372. doi:10.1016/0022-5096(63)90036-X
[6] Mori, T. & Tanaka, K. (1973). Average stress in matrix and average elastic energy of materials with misfitting inclusions. Acta Metallurgica, 21(5), 571–574. doi:10.1016/0001-6160(73)90064-3
[7] Hill, R. (1965). A self-consistent mechanics of composite materials. Journal of the Mechanics and Physics of Solids, 13(4), 213–222. doi:10.1016/0022-5096(65)90011-6
[8] Halpin, J. C. & Kardos, J. L. (1976). The Halpin-Tsai equations: A review. Polymer Engineering & Science, 16(5), 344–352. doi:10.1002/pen.760160512
[9] Mura, T. (1987). Micromechanics of Defects in Solids (2nd ed.). Martinus Nijhoff.
[10] Benveniste, Y. (1987). A new approach to the application of Mori-Tanaka’s theory in composite materials. Mechanics of Materials, 6(2), 147–157. doi:10.1016/0167-6636(87)90005-6
[11] Christensen, R. M. (1979). Mechanics of Composite Materials. Wiley.
[12] Ponte Castañeda, P. & Willis, J. R. (1995). The effect of spatial distribution on the effective behavior of composite materials and cracked media. Journal of the Mechanics and Physics of Solids, 43(12), 1919–1951. doi:10.1016/0022-5096(95)00052-6
[13] Moulinec, H. & Suquet, P. (1994). A fast numerical method for computing the linear and nonlinear mechanical properties of composites. Comptes Rendus de l’Académie des Sciences, Série II, 318, 1417–1423.
[14] Moulinec, H. & Suquet, P. (1998). A numerical method for computing the overall response of nonlinear composites with complex microstructures. Computer Methods in Applied Mechanics and Engineering, 157(1–2), 69–94. doi:10.1016/S0045-7825(97)00218-1
Basem Rajjoub