![Arxiv:2007.00701V2 [Cond-Mat.Soft] 16 Mar 2021 Nnt E Fcreainfntosta Hrceie the Characterizes Microstructure](http://data.docslib.org/img/f48c257de8447fe61d8eabfaec8cc2a3-1.webp)
arXiv:2007.00701v2 [cond-mat.soft] 16 Mar 2021 nnt e fcreainfntosta hrceie the characterizes microstructure. that an functions composite requires correlation the generally of an composite, given set that infinite the property, us of effective teaches properties an investigations phase of of determination set exact latter [13– The procedures series-expansion 16]. in- tech- exact bounding methods, and 4–6], [7–12] of niques [1, variety schemes a moduli approximation using elastic cluding permeability) constant, fluid dielectric effec- and static (e.g., of properties determination theoretical the tive previous on of focused and have preponderance [2], studies The Rayleigh luminaries Lord [3]. the [1], of Einstein Maxwell some including by work science, to of back dates outstand- and an one is ing media composite multiphase of properties h hoeia rbe fetmtn h effective the estimating of problem theoretical The olclEetv lcrmgei aeCaatrsiso Characteristics Wave Electromagnetic Effective Nonlocal (where fsc xasos hi oe-re rnain il ac yield truncations lower-order for their expansions, such of iae fteeetv yai ilcrccntn tensor constant dielectric dynamic effective limitat long-wavelength the the of extend timates to us enables formalism rpriso iodrdtopaecmoie n metamat and composites two-phase disordered of properties rsrbdlnt cls hs u nig a accelerate can findings our Thus, to composites. th microstructure scales. design the length abru can engineering prescribed that one by indices that composite refractive demonstrate disordered results and Our number wave numbers. low-pass selected wave as a act to to up pa capacity hyperuniform the disordered loss including certain less characteristics, that show generally also are We p media to parts. having hyperuniform without disordered microstructures that closed- composite characteristic our of wave class Thus, effective wide the microstr media. efficiently these composite and com of of accurately simulation models accuracy full-waveform disordered The to and comparison by embodied composite. validated information any is microstructural for the compute st via to that orders expansions all strong-contrast to the of representations ag fwv ubr o hc u xeddhomogenization extended our which for numbers wave of range h rqec n aevco fteicdn aito) We radiation). incident account the exactly of that expansions vector strong-contrast wave nonlocal and frequency the ifiiestof set (infinite ota tcncpuesaildseso elbyn h qua the beyond well dispersion spatial capture can it that so edrv xc olclhmgnzdcntttv relatio constitutive homogenized nonlocal exact derive We ε .INTRODUCTION I. e ( eateto hmsr,PictnUiest,Princeton University, Princeton Chemistry, of Department k ℓ eateto hsc,PictnUiest,Pictn N Princeton, University, Princeton Physics, of Department eateto hsc,PictnUiest,Pictn N Princeton, University, Princeton Physics, of Department I ω , sacaatrsi eeoeet eghsae.Bcueo Because scale). length heterogeneity characteristic a is htapyfrawd ls fmcotutrs hs nonlo These microstructures. of class wide a for apply that ) n rneo nttt o h cec n ehooyo Mater of Technology and Science the for Institute Princeton pitcreainfntos n ec utpescatteri multiple hence and functions) correlation -point rneo nvriy rneo,NwJre 84,UAand USA 08544, Jersey New Princeton, University, Princeton rneo nvriy rneo,NwJre 84,USA 08544, Jersey New Princeton, University, Princeton rga nApidadCmuainlMathematics, Computational and Applied in Program eodteQaittcRegime Quasistatic the Beyond Dtd ac 8 2021) 18, March (Dated: avtr Torquato Salvatore au Kim Jaeuk h orsodn ffciewv speed determine wave can effective one property, corresponding effective the this From 18]. [17, ld h awl-ant 2,2]adquasicrystalline and in- 25] matrix [24, a Maxwell-Garnett in the scatterers clude spherical for devised formulas oeiaintere of theories mogenization incoefficient tion in,eg,nnvrapn pee namti seFig. (see matrix a in inclu- spheres well-defined nonoverlapping e.g., are sions, that scatterers dielectric namely, w-hs ilcrccmoie hc eed nthe on depends which frequency composite, dielectric two-phase a 1 ffciednmcdeeti osattensor constant dielectric dynamic effective al when cable sistatic fnneoatdeeti eair ital l earlier all Virtually realm the behavior. formulas is dielectric range nonresonant spectral of This scale. length erogeneity .Eape fsc oua lsdfr approximation closed-form popular such of Examples ). u ou nti ae stedtriaino the of determination the is paper this in focus Our lesta rnmtwvs“isotropically” waves transmit that filters elbyn h ussai eiefra for regime quasistatic the beyond well s l cuaeycpuemlil scattering multiple capture accurately ill uaecoe-omapoiaeformulas approximate closed-form curate o opeemcotutrlinformation microstructural complete for oso ovninlhmgnzto es- homogenization conventional of ions sfrteeetv lcrmgei wave electromagnetic effective the for ns hnternnyeuiomcounter- nonhyperuniform their than y ∗ ε rln-aeeghrgm 1–2[3,ie,appli- i.e., [19–22][23], regime long-wavelength or uain o oh2 n Dordered 3D and 2D both for putations tclt opstsehbtnvlwave novel exhibit composites rticulate oss alrdsailcreain at correlations spatial tailored possess find We simulations. full-blown erform h icvr fnvlelectromagnetic novel of discovery the rasfo rtpicpe.Ti exact This principles. first from erials nteseta est,wihi easy is which density, spectral the in e ε ( ittcrgm (where regime sistatic omfrua nbeoet predict to one enable formulas form e ω k tycag vranro ag of range narrow a over change ptly ffciewv hrceitc fa of characteristics wave effective e ( I hoyapis .. 0 i.e., applies, theory k | cuedpnetapproximations ucture-dependent copihti akb deriving by task this accomplish ω , rwv vector wave or k e esy054 USA 08544, Jersey New , h atcnegneproperties fast-convergence the f I I ω , o rirr microstructures arbitrary for ) wJre 84,USA 08544, Jersey ew wJre 84,USA 08544, Jersey ew | γ ℓ e pl ovr pca microstructures, special very to apply ) h rpneac fpeiu ho- previous of preponderance The . ≪ a omlsaeresummed are formulas cal gt l resfrthe for orders all to ng ,where 1, opst Media: Composite f ials, ε e ( k k I I ω , ℓ ω fteicdn radiation incident the of | ≤ pl nyi the in only apply ) sacaatrsi het- characteristic a is and k I k | ℓ I . c are e 1 n attenua- and ε e ( k I ω , qua- of ) 2
[18, 26, 27] estimates, among others [17]. (Sec. VIII). This validation means that they can be used In the present investigation, we derive nonlocal homog- to predict accurately the effective wave characteristics enized constitutive relations from Maxwell’s equations to well beyond the quasistatic regime for a wide class of obtain exact expressions for the effective dynamic di- composite microstructures (Sec. IX) without having to electric constant tensor εe(kI ,ω) of a macroscopically perform full-blown simulations. This broad microstruc- anisotropic two-phase medium of arbitrary microstruc- ture class includes particulate media consisting of identi- ture that is valid well beyond the quasistatic regime, i.e., cal or polydisperse particles of general shape (ellipsoids, from the infinite-wavelength limit down to intermediate cubes, cylinders, polyhedra) that may or not overlap, and wavelengths (0 kI ℓ . 1). This task is accomplished cellular networks as well as media without well-defined by extending the≤ | strong-contrast| expansion formalism, inclusions (see Sec. IIIB for details). Thus, our nonlocal which has been used in the past exclusively for the static formulas can be employed to accelerate the discovery of limit [11] and quasistatic regime [31], and establishing novel electromagnetic composites by appropriate tailor- that the resulting homogenized constitutive relations are ing of the spectral densities [35, 36] and then generating nonlocal in space (Sec. III), i.e., the average polariza- the microstructures that achieve them [36]. tion field at position x depends on the average electric Although our strong-contrast formulas for the effective field at other positions around x. (Such nonlocal re- dynamic dielectric constant apply to periodic two-phase lations are well known in the context of crystal optics media, the primary applications that we have in mind are in order to account for “spatial dispersion,” i.e., the de- correlated disordered microstructures because they can pendence of dielectric properties on a wave vector [32].) provide advantages over periodic ones with high crys- The terms of the strong-contrast expansion are explicitly tallographic symmetries [37, 38] which include perfect given in terms of integrals over products of Green’s func- isotropy and robustness against defects [39, 40]. We are tions and the n-point correlation functions of the ran- interested in both “garden-variety” models of disordered dom two-phase medium (defined in Sec. II A) to infinite two-phase media [11] as well as exotic hyperuniform forms order. This representation exactly treats multiple scat- [41–43]. Hyperuniform two-phase systems are character- tering to all orders for the range of wave numbers for ized by an anomalous suppression of volume-fraction fluc- which our extended homogenization theory applies, i.e., tuations in the infinite-wavelength limit [41–43], i.e., the 0 k ℓ . 1. spectral densityχ ˜V (Q) obeys the condition ≤ | I | It is noteworthy that the strong-contrast formalism is lim χ˜V (Q)=0. (1) a significant departure from standard multiple-scattering |Q|→0 theory [17, 20, 33, 34], as highlighted in Sec. III. More- Such two-phase media encompass all periodic systems, over, as we show there, our strong-contrast formalism has many quasiperiodic media and exotic disordered ones; a variety of “tuning knobs” that enable one to obtain dis- see Ref. [43] and references therein. Disordered hyper- tinctly different expansions and approximations designed uniform systems lie between liquids and crystals; they are for different classes of microstructures. like liquids in that they are statistically isotropic without Because of the fast-convergence properties of strong- any Bragg peaks, and yet behave like crystals in the man- contrast expansions, elaborated in Sec. IIIB, their lower- ner in which they suppress the large-scale density fluc- order truncations yield accurate closed-form approximate tuations [41–43]. Hyperuniform systems have attracted formulas for the effective dielectric constant that apply great attention over the last decade because of their deep for a wide class of microstructures over the aforemen- connections to a wide range of topics that arise in physics tioned broad range of incident wavelengths, volume frac- [28, 37, 43–53], materials science [36, 54–57], mathemat- tions, and contrast ratios (Sec. IV). Thus, we are able ics [58–60], and biology [43, 61] as well as for their emerg- to accurately account for multiple scattering in the res- ing technological importance in the case of the disordered onant realm (e.g., Bragg diffraction for periodic media), varieties [40, 43, 52, 54, 62–66]. in contrast to the Maxwell-Garnett and quasicrystalline We apply our strong-contrast formulas to predict the approximations, which are known to break down in this real and imaginary parts of the effective dielectric con- spectral range. These nonlocal strong-contrast formulas stant for model microstructures with typical disorder can be regarded as approximate resummations of the ex- (nonhyperuniform) as well as those with exotic hyperuni- pansions that still accurately capture multiple-scattering form disorder (Sec. V). We are particularly interested in effects to all orders via the nonlocal attenuation func- exploring the dielectric properties of a special class of hy- tion F (Q) (Sec. VI). The key quantity F (Q) is a func- peruniform composites called disordered stealthy hyper- tional of the spectral densityχ ˜V (Q) (Sec. IIB), which is uniform media, which are defined to be those that possess straightforward to determine for general microstructures zero-scattering intensity for a set of wave vectors around either theoretically, computationally or via scattering ex- the origin [36, 44, 67–69]. Such materials have recently periments. We employ precise full-waveform simulation been shown to be endowed with novel optical, acoustic, methods (Sec. VII) to show that these microstructure- mechanical, and transport properties [31, 55, 56, 63, 70– dependent approximations are accurate for both two- 73]. Among other findings, we show that disordered hy- dimensional (2D) and three-dimensional (3D) ordered peruniform media are generally less lossy than their non- and disordered models of particulate composite media hyperuniform counterparts. We also demonstrate that 3
FIG. 1. Left panel: The preponderance of previous theoretical treatments of the effective dynamic dielectric constant are local in nature and restricted to dispersions of well-defined dielectric scatterers (inclusions) in a matrix, as illustrated in the leftmost panel. In contrast to the nonlocal strong-contrast formalism presented here, earlier studies cannot treat the more general two- phase microstructures shown in the middle panel (spinodal decomposition pattern [28]) and rightmost panel (Debye random medium [29, 30]), both of which have “phase-inversion” symmetry [11]. Our formalism can in principle treat two-phase media of arbitrary microstructures. disordered stealthy hyperuniform particulate composites tionally invariant and, in particular, the one-point func- exhibit singular wave characteristics, including the ca- (i) tion is position independent, i.e., S1 (x1)= φi. pacity to act as low-pass filters that transmit waves “isotropically” up to a selected wave number. They also can be engineered to exhibit refractive indices that B. Two-point statistics abruptly change over a narrow range of wave numbers by tuning the spectral density. Our results demonstrate that For statistically homogeneous media, the two-point one can design the effective wave characteristics of a dis- correlation function for phase 2 is simply related to that ordered composite, hyperuniform or not, by engineering for phase 1 via the expression S(2)(r)= S(1)(r) 2φ +1, the microstructure to possess tailored spatial correlations 2 2 − 1 at prescribed length scales. and hence the autocovariance function is given by χ (r) S(1)(r) φ 2 = S(2)(r) φ 2, (4) V ≡ 2 − 1 2 − 2 II. BACKGROUND which we see is the same for phase 1 and phase 2. Thus, χ (r = 0) = φ φ and, assuming the medium possesses A. n-point correlation functions V 1 2 no long-range order, lim|r|→∞ χV (r) = 0. For statisti-
cally homogeneous and isotropic media, χV (r) depends A two-phase random medium is a domain of space only on the magnitude of its argument r = r , and hence Rd V ⊆ that is partitioned into two disjoint regions that make is a radial function. In such cases, its slope| at| the origin up : a phase 1 region of volume fraction φ and a V V1 1 is directly related to the specific surface s (interface area phase 2 region 2 of volume fraction φ2 [11]. The phase per unit volume); i.e., asymptotically, we have [11] indicator functionV (i)(x) of phase i for a given realization I is defined as χ (r)= φ φ β(d)s r + ( r 2), (5) V 1 2 − | | O | | 1, x i, (i)(x)= ∈V (2) where I 0, x / . ∈Vi ® Γ(d/2) (i) β(d)= , (6) The n-point correlation function Sn for phase i is defined 2√πΓ((d + 1)/2) by [11, 74]: and Γ(x) is the gamma function. n The nonnegative spectral densityχ ˜ (Q), which is pro- S(i)(x ,..., x )= (i)(x ) , (3) V n 1 n I j portional to scattering intensity [75], is the Fourier trans- ∞j=1 ∫ r Y form of χV ( ), i.e., where angular brackets denote an ensemble average over (i) Q r −iQ·r r Q realizations. The function Sn (x1,..., xn) has a proba- χ˜V ( )= χV ( ) e d 0, for all , (7) Rd ≥ bilistic interpretation: It gives the probability of finding Z the ends of the vectors x1,..., xn all in phase i. For sta- where Q represents the momentum-transfer wave vector. (i) tistically homogeneous media, Sn (x1,..., xn) is transla- For statistically homogeneous media, the spectral density 4 must obey the following sum rule [76]: hyperuniform two-phase media are a subclass of hyper- uniform systems in whichχ ˜ (Q) is zero for a range of 1 V Q Q r wave vectors around the origin, i.e., d χ˜V ( ) d = χV ( =0)= φ1φ2. (8) (2π) Rd Z χ˜V (Q)=0 for0 Q QU, (14) For isotropic media, the spectral density depends only ≤ | |≤ on Q = Q and, as a consequence of Eq. (5), its large-k where QU is some positive number. behavior| is| controlled by the following power-law form: As in the case of hyperuniform point configurations [41–43, 80], there are three different scaling regimes γ(d) s χ˜ (Q) ,Q , (9) (classes) that describe the associated large-R behaviors V ∼ Qd+1 → ∞ of the volume-fraction variance when the spectral density goes to zero as a power lawχ ˜ (Q) Q α as Q 0: where V ∼ | | | | → R−(d+1), α> 1 (Class I), γ(d)=2d π(d−1)/2 Γ((d + 1)/2) β(d) (10) σ2 (R) R−(d+1) ln R, α =1 (Class II), (15) V ∼ is a d-dimensional constant and β(d) is given by Eq. (6). R−(d+α), 0 <α< 1 (Class III),
where the exponent α is a positive constant. Thus, the C. Packings characteristics length of the representative elementary volume for a hyperuniform medium will depend on the We call a packing in Rd a collection of nonoverlap- hyperuniformity class (scaling). Class I is the strongest ping particles [77]. In the case of a packing of identical form of hyperuniformity, which includes all perfect peri- spheres of radius a at number density ρ, the spectral odic packings as well as some disordered packings, such as disordered stealthy packings described in Sec. V D. densityχ ˜V (Q) is directly related to the structure factor S(Q) of the sphere centers [11, 35]:
E. Popular effective-medium approximations χ˜V (Q)= φ2 α˜2(Q; a) S(Q), (11) where Here we explicitly state the specific functional forms d of an extended Maxwell-Garnett approximation and qua- 1 2πa 2 α˜2(Q; a)= J (Qa), (12) sicrystalline approximation for the effective dynamic di- v (a) Q d/2 1 electric constant ε (k ) of isotropic media composed of Å ã e 1 Jν (x) is the Bessel function of the first kind of order ν, identical spheres of dielectric constant ε2 embedded in a φ2 = ρv1(a) is the packing fraction (fraction of space matrix phase of dielectric constant ε1. In Sec. VIII, we covered by the spheres), and compare the predictions of these formulas to those of our nonlocal approximations. The small-wave-number ex- πd/2ad pansions of these popular approximations are provided in v (a)= (13) 1 Γ(1 + d/2) the Supplementary Material [81]. There we also provide the corresponding asymptotic behaviors of our strong- is the d-dimensional volume of a sphere of radius a. contrast approximations.
D. Hyperuniformity and volume-fraction 1. Maxwell-Garnett approximation fluctuations Maxwell-Garnett approximations (MGAs) [24, 25] are Originally introduced in the context of point configura- derived by substituting the dielectric polarizability of a tions [41], the hyperuniformity concept was generalized single dielectric sphere into the Clausius-Mossotti equa- to treat two-phase media [42]. Here the phase volume tion [25], which consequently ignores the spatial corre- fraction fluctuates within a spherical window of radius R, lations between the particles. In three dimensions, we which can be characterized by the local volume-fraction utilize the following extended MGA that makes use of 2 variance σV (R). This variance is directly related to inte- the exact electric dipole polarizability αe(k1) of a single grals involving either the autocovariance function χV (r) dielectric sphere of radius a [82]:
[78] or the spectral densityχ ˜V (Q) [42]. For typical disordered two-phase media, the variance εe(k1) ε1 3 2 −d − =φ2 αe(k1) /a , (16) σV (R) for large R goes to zero like R [11, 78, 79]. How- εe(k1)+2ε1 ever, for hyperuniform disordered media, σ2 (R) goes to V where zero asymptotically more rapidly than the inverse of the −d ′ ′ window volume, i.e., faster than R , which is equiva- 3i m ψ1(mk1a) ψ1(k1a) ψ1(k1a) ψ1(mk1a) αe(k1)= 3 ′ − ′ , lent to the condition (1) on the spectral density. Stealthy 2k m ψ1(mk1a) ξ (k1a) ξ1(k1a) ψ (mk1a) 1 1 − 1 5
(1) m ε2/ε1, ψ1(x) x j1(x), ξ1(x) xh1 (x), the III. EXACT STRONG-CONTRAST prime≡ symbol (′) denotes≡ the derivative≡ of a function, EXPANSIONS AND NONLOCALITY (1) p h1 (x) is the spherical Hankel function of the first kind of order 1, and j1(x) is the spherical Bessel function of The original strong-contrast expansions for the effec- the first kind of order 1. tive dynamic dielectric constant obtained by Rechtsman The 2D analog of Eq. (16) can be obtained by using and Torquato [31] were derived from homogenized con- the dynamic dielectric polarizability αe of a dielectric stitutive relations that are local in space and hence are cylinder of radius a given in Ref. [83]: strictly valid only in the quasistatic regime. In the present work, we follow the general strong-contrast for- ε (k ) ε φ malism of Torquato [11] that was devised for the purely e 1 − 1 = 2 α (k ) /a2, (17) ε (k )+ ε 2π e 1 static problem [11] and show that it naturally leads to e 1 1 exact homogenized constitutive relations for the aver- aged fields that are nonlocal in space. The crucial con- where sequences of this development are exact expressions for the effective dynamic dielectric constant tensor εe(kI ) 4(ε2 ε1) ′ (1) αe(k1)= − J1(mk1a) J (mk1a) (k1a) for a macroscopically anisotropic medium of arbitrary 2 1 H1 ik1 mε1 microstructure into which a plane wave of wave vec- ′ (1) −1 k ε k mJ1(mk1a) 1 (k1a) , tor I is incident. These expressions for e( I ) are − H valid from the infinite-wavelength limit down to wave- (1) lengths (λ = 2π/ k ) on the order of the heterogeneity where (x) is the Hankel function of the first kind of I ν length scale ℓ. We| then| briefly explain how our theory order νH. Here, only transverse-electric (TE) polarization departs substantially from standard multiple-scattering is considered; i.e., the electric field is perpendicular to theory [17, 20, 33, 34]. We explicitly show they neces- the axis of the cylinder. sarily require complete microstructural information, as As with all MGA theories, formulas (16) and (17) ne- embodied in the infinite set of n-point correlation func- glect spatial correlations between the particles and hence tions (Sec. II A) of the composite. We also describe the are only valid for low inclusion packing fractions. In the fast-convergence properties of strong-contrast expansions static limit, Eqs. (16) and (17) reduce to the Hashin- and their consequences for extracting accurate approxi- Shtrikman estimates εHS; see relation (72). mations for εe(kI ).
2. Quasicrystalline approximations A. Strong-contrast expansions
The quasicrystalline approximation (QCA) for the Consider a macroscopically large ellipsoidal two-phase quasistatic effective dynamic dielectric constant εe(k1) statistically homogeneous but anisotropic composite employs the “effective” Green’s function of spherical scat- specimen in Rd embedded inside an infinitely large refer- ε terers up to the level of the pair correlation function g2(r) ence phase I with a dielectric constant tensor I . The mi- [18, 26, 27]. However, the QCA accounts only for the crostructure is perfectly general, and it is assumed that a structure factor in the infinite-wavelength limit [27] [i.e., characteristic heterogeneity length scale ℓ is much smaller S(0) = 1+ ρ (g (r) 1) dr] and consequently, spatial than the specimen size, i.e., ℓ L. The shape of this 2 ≪ correlations at finite wavelengths− are ignored. The QCA specimen is purposely chosen to be nonspherical since for d = 3 can beR explicitly written as follows [18]: any rigorously correct expression for the effective prop- erty must ultimately be independent of the shape of the −1 composite specimen in the infinite-volume limit. It is as- 2 εe(k1) ε1 2 3 φ2 β21 − = φ2 i φ2 S(0) (k1a) sumed that the applied or incident electric field E0(x) is εe(k1)+2ε1 − (3 a plane wave of an angular frequency ω and wave vector ï ò −1 kI in the reference phase, i.e., 2 3 1+ i (k1a) S(0) β21, (18) × 3(1 β21φ2) ˜ − ) E0(x)= E0 exp(i(kI x ωt)) . (19) ï ò · − where β21 is defined in Eq. (33). Interestingly, the QCA Our interest is the exact expression for the effective dy- predicts that the hyperuniform composites [S(0) = 0] namic dielectric constant tensor εe(kI ,ω). Without loss will be transparent for all wave numbers, which cannot of generality, we assume a linear dispersion relation in be true for stealthy hyperuniform media [cf. Eq. (14)], the reference phase, i.e., kI kI = √εI ω/c, where c since the transparency interval must be finite; see Sec. is the speed of light in vacuum,≡ | and| thus we henceforth VIII. Note that Eq. (18) is the complex conjugate of the do not explicitly indicate the dependence of functions on one given in Ref. [18] so that it is consistent with the ω. The composite is assumed to be nonmagnetic, imply- sign convention for the imaginary part Im[εe] used here. ing that the phase magnetic permeabilities are identical, 6
of Torquato [11] closely but departs from it at certain key steps when establishing the nonlocality of the ho- mogenized constitutive relation. (A detailed derivation is given in the Supplementary Material [81].) For simplic- ity, we take the reference phase I to be phase q (equal to 1 or 2). Under the aforementioned assumptions, the lo- cal electric field E(x) solves the time-harmonic Maxwell equation [31]: ω 2 ∇×∇× E(x) k 2 E(x)= P(x) , (20) − q c where P(x) is the polarization field given by
P(x) [ε(x) ε ] E(x) (21) ≡ − q and
ε(x) = (ε ε ) (p)(x)+ ε (22) p − q I q is the local dielectric constant, and (p)(x) is the indica- tor function for phase p [cf. Eq. (2)].I The vector P(x) is the induced flux field relative to reference phase q due to the presence of phase p, and hence is zero in the reference phase q and nonzero in the “polarized” phase p (p = q). Using the Green’s function formalism, the local electric6 field can be expressed in terms of the following integral equation [11, 31]:
E(x)= E (x)+ G(q)(x x′) P(x′) dx′ , (23) 0 − · Z where the second-rank tensor Green’s function G(q)(r) FIG. 2. (a) Schematic of a large d-dimensional ellipsoidal, associated with the reference phase q is given by [84] macroscopically anisotropic two-phase composite medium em- bedded in an infinite reference phase of dielectric constant ten- G(q)(r)= D(q) δ(r)+ H(q)(r) , (24) sor εI (gray regions) under an applied electric field E0(x) = − E˜ 0 exp(i(kI x ωt)) of a frequency ω and a wave vector kI D(q) · − is a constant second-rank tensor that arises when at infinity. The wavelength λ associated with the applied field one excludes an infinitesimal region around the position can span from the quasistatic regime (2πℓ/λ 1) down to ′ ≪ of the singularity x = x in the Green’s function, and the intermediate-wavelength regime (2πℓ/λ . 1), where ℓ is a H(q) characteristic heterogeneity length scale. (b) After homoge- (r) represents the contribution outside of the “ex- nization, the same ellipsoid can be regarded as a homogeneous clusion” region: specimen with an effective dielectric constant εe(kI ,ω), which d/2 depends on ω and kI . As noted in the main text, we omit (q) π kq (1) (1) ε Hij (r)=i kqr d/2−1(kqr) d/2(kqr) δij the ω dependence of e because (without loss of generally) we 2εq 2πr H − H assume a linear dispersion relation between kI and ω. (1) | | + k År ã (k îr) ˆr ˆr , (25)ó q Hd/2+1 q i j (1) i.e., µ1 = µ2 = µ0, where µ0 is the magnetic perme- where ˆr r/ r is a unit vector directed to r, and ν (x) ability of the vacuum. For simplicity, we assume real- is the Hankel≡ | function| of the first kind of order Hν. The valued, frequency-independent isotropic phase dielectric Fourier transform of Eq. (24) is particularly simple and constants ε1 and ε2. Nonetheless, the composite can be concise: generally lossy (i.e., εe is complex-valued) due to scatter- 1 k 2δ k k ing from the inhomogeneities in the local dielectric con- G˜(q)(k)= q ij − i j . (26) ij 2 2 stant. It is noteworthy that our results can be straight- εq k kq − forwardly extended to phase dielectric constants that are complex-valued (dissipative media), but this is not done Note that Eq. (26) is independent of the shape of the in the present work. exclusion region, which stands in contrast to the shape- (q) Here we present a compact derivation of strong- dependent Fourier transform of Hij (r); see Supplemen- contrast expansions. It follows the general formalism tary Material [81] for details. 7
The use of Eq. (21) and (23) leads to an integral equa- respectively, where D(x) is the displacement field. Im- tion for the generalized cavity intensity field F(x): portantly, among these three cases, the original and mod- ified strong-contrast expansions arise only when the ex- F(x)= E (x)+ H(q)(x x′) P(x′) dx′ , (27) clusion region is taken to be a sphere, as we elaborate 0 − · Zǫ below. where the integral subscript ǫ indicates that the integral We now show that our formalism yields an exact rela- is to be carried out by omitting the exclusion region and tion between the polarization field P(x) and the applied then allowing it to uniformly shrink to zero. Here F(x) field E0(x) that is nonlocal in space. It is more conve- is related directly to E(x) via nient at this stage to utilize a compact linear operator notation, which enables us to express the integral equa- F(x)= I + D(q)[ε(x) ε ] E(x). (28) tion (27) as − q · H Using the definitions¶ (21) and (28), we© obtain a linear F = E0 + P, (35) constitutive relation between P(x) and F(x): where we temporarily drop the superscript q. A combi- P(x)= L(q)(x) F(x) , (29) nation of this equation with Eq. (29) yields the following · integral equation for the polarization field where −1 P = LE + LHP. (36) L(q)(x) = [ε(x) ε ] I + D(q)[ε(x) ε ] . (30) 0 − q · − q The desired nonlocal relation is obtained from Eq. (36) It is noteworthy that one is free to choose any conve- ¶ © by successive substitutions: nient exclusion-region shape, provided that its boundary is sufficiently smooth. The choice of the exclusion-region P = SE0. (37) shape is crucially important because it determines the type of modified electric field that results as well as the where corresponding expansion parameter in the series expan- sion for the effective dynamic dielectric constant tensor. S = [I LH]−1L, (38) For example, when a spheroidal-shaped exclusion region − at a position x in Rd is chosen to be aligned with the is a generalized scattering operator that has superior polarization vector P(x) at that position, we have mathematical properties compared to the scattering op- erator T that arises in standard multiple-scattering the- A∗ D(q) = I, (31) ory [17, 20, 33], as we elaborate below in Remark viii.. εq More explicitly, the nonlocal relation (37) can be ex- where I is the second-rank identity tensor and A∗ [0, 1] pressed as is the depolarization factor for a spheroid [11]∈ with the aforementioned alignment. In the special cases of P(1)= d2 S(1, 2) E0(2) , (39) · a sphere, disk-like limit, and needle-like limit, A∗ = Zǫ 1/d, 1, 0, respectively. Thus, for these three cases, Eq. where boldface numbers 1, 2 are shorthand notations for (30) yields the following shape-dependent tensor: position vectors r1, r2. Ensemble averaging Eq. (39) and invoking statistical homogeneity yields the convolution L(q)(x; A∗)= (p)(x) I I relation ∗ dεqβpq, A =1/d (spherical) ∗ ε (1 ε /ε ), A = 1 (disk-like) , (32) P (1)= d2 S (1 2) E0(2) , (40) q q p h i h i − · × − ∗ ǫ εq(εp/εq 1), A = 0 (needle-like) Z − where the operator S depends on relative positions, i.e., where βpqis the dielectric polarizability defined by S (1, 2)= S (1 h 2i), and angular brackets denote an h i h i − εp εq ensemble average. Formally, the nonlocal relation (40) is βpq − . (33) the same as the one given in Torquato [11] for the static ≡ εp + (d 1)εq − problem, but nonlocality was not explicitly invoked there. Moreover, in these three cases, the generalized cavity in- Taking the Fourier transform of (40) yields a compact tensity field F(x; A∗) reduces to Fourier representation of this nonlocal relation, namely,
P(x) ∗ E(x)+ , A =1/d (spherical) P (k)= S (k) E˜ 0(k) , (41) dεq h i h i · F x ∗ D(x) ( ; A ) , A∗ = 1 (disk-like) → εq where f (k) g f (xg) exp( ik x) dx. From Eq. h i ≡ h i − · E(x) , A∗ = 0 (needle-like), (19), E˜ (k)= E˜ δ(k k ), implying that the wave vec- 0 0 R − q (34) tor k in Eq. (41) must be identical to kq. › 8
L(q) As in the static case [11, 15] and quasistatic regime [31], Note that the support ℓs of the kernel e (r) rel- the ensemble-averaged operator S (r), which is given ative to the incident wavelength λ determines the de- explicitly in the Supplementaryh Materiali [81] in terms gree of spatial dispersion. When λ is finite, the rela- of the n-point correlation functions and products of the tion between P (x) and F (x) in Eq. (43) is nonlo- h i h i tensor H(r), depends on the shape of the macroscopic cal in space. In the regime ℓs λ, the nonlocal rela- ellipsoidal composite specimen (see Fig. 2). This shape- tion (43) can be well approximated≪ by the local relation dependence arises from because H decays like r−d for L(q) ′ ′ P (x) e (x ) dx F (x). Indeed, in the static large r, and hence, S (r) involves conditionally conver- h i ≈ · h i limit, L(q)(r) tends to a Dirac delta function δ(r), expres- gent integrals [11]. Toh i avoid such conditional convergence e R sion (43) becomesî the position-independentó local relation issues, we follow previous strong-contrast formulations by seeking to eliminate the applied field E0 in Eq. (40) in F r favor of the average cavity field ( ) in order to get P = L(q) F (45) a corresponding nonlocal homogenizedh i constitutive rela- h i e · h i tion between P (r) and F (r) or vice versa. Thus, h i h i derived earlier [11]. solving for E0 in Eq. (40) and substituting into the en- The nonlocal constitutive relation in direct space, Eq. semble average of (35) yields (43), can be reduced to a linear product form in Fourier space by taking the Fourier transform of Eq. (43): F = [ S −1 + H] P . (42) h i h i h i L(q) P (kq)= e (kq) F (kq) . (46) Inverting this expression leads to the following nonlocal h i · h i L(q) constitutive relation: The wave-vector-dependentg effectiveg tensor e (kq) is postulated (see discussion in the Supplementary Material P (1)= d2 L(q)(1 2) F (2) , (43) [81]) to be given by h i e − · h i Z −1 L(q) ε I I D(q) ε I e (kq) [ e(kq) εq ] + [ e(kq ) εq ] . L(q) ≡ − · · − where e (r) is a kernel that is derived immediately be- (47) low and explicitly given by ¶ L(q) © This linear fractional form for e (k) is consistent with the one derived for the static limit [11] and for the qua- (q) ′ ′ ′ ′ sistatic regime [31]. Taking the inverse Fourier transform L (r) dr [εe(r ) εqI δ(r )] I δ(r r ) e ≡ − · − of Eq. (47) yields its corresponding direct-space repre- Z h −1 sentation Eq. (44). Taking the Fourier transform of Eq. + D(q) (ε (r r′) ε I δ(r r′)) . (44) · e − − q − (42) yields i We are not aware of any previous work that derives such −1 ˜ F (kq)= S (kq) + H(kq) P (kq) . (48) an exact nonlocal homogenized constitutive relation (43) h i h i · h i from first principles. g ïg ò g
Comparing Eq. (46) to Eq. (48), and specifically choosing a spherical exclusion region, as discussed in Eq. (32), yields the desired exact strong-contrast expansions for general macroscopically anisotropic two-phase media:
∞ φ 2β2 [ε (k ) + (d 1)ε I] [ε (k ) ε I]−1 = φ β I A(p)(k ) β n, (49) p pq e q − q · e q − q p pq − n q pq n=2 X
A(p) where n (kq) is a wave-vector-dependent second-rank tensor that is a functional involving the set of correlation (p) (p) (p) H(q) functions S1 ,S2 , ,Sn (defined in Sec. II A) and products of the second-rank tensor field (r), which is explicitly given as [see··· also Eq. (25)]
2 i 2 (1) kq (1) ikq (1) kq (kqr) (kqr) δij + (kqr) ˆriˆrj , d =2 H(q)(r)= 4 H0 − r H1 4 H2 (50) ij exp(ikq r) 2 2 ( 3 1+ ikqr + (kqr) δij + 3 3ikqr (kqr) ˆriˆrj , d =3 εîq 4πr − ó − − where k k . Specifically, for n = 2 and n 3, these n-point tensors associated with the polarized phase p are, q ≡ | q| ≥ 9 respectively, given by
(p) A H(q) −ikq ·r 2 (kq)=dεq dr (r) e χV (r) , (51) ǫ Z n−2 dεq A(p)(k )=dε − dr dr H(q)(r r ) e−ikq·(r1−r2) H(q)(r r ) e−ikq·(r2−r3) n q q φ 1 ··· n−1 1 − 2 · 2 − 3 · p Zǫ Å H(q)(ãr r ) e−ikq ·(rn−1−rn) ∆(p)(r , , r ) , (52) ··· n−1 − n n 1 ··· n (p) where dr lim + dr and ∆n is a position-dependent determinant involving correlation functions of the ǫ ≡ ǫ→0 |r|>ǫ polarized phase p up to the n-point level: R R (p) (p) S2 (r1, r2) S1 (r1) 0 (p) (p) ··· S3 (r1, r2, r3) S2 (r2, r3) 0 ∆(p)(r , , r )= ··· . (53) n 1 ··· n ...... (p) (p) (p) Sn (r , , r ) S (r , , r ) S (r , r ) 1 ··· n n−1 2 ··· n ··· 2 n−1 n
For macroscopically isotropic media, the effective di- iv. The exact expansions represented by (49) are inde- electric tensor is isotropic, i.e., εe(kq) = εe(kq) I. The pendent of the reference phase q and hence inde- corresponding strong-contrast expansion for εe(kq) is ob- pendent of the wave vector kq. tained by taking the trace of both sides of Eq. (49): v. For d = 2, the strong-contrast expansion applies for φ 2β2 [ε (k ) + (d 1)ε ][ε (k ) ε ]−1 TE polarization only. This implies that the electric p pq e q − q e q − q ∞ field and wave vector are parallel to the plane or =φ β A(p)(k ) β n, (54) transverse to an axis of symmetry in a 3D system p pq − n q pq whose cross sections are identical. n=2 X vi. Formally, the original strong-contrast expansions ε (p) A(p) where εe(kq) = Tr[ e(kq)]/d, An (kq) = Tr n (kq) /d that apply in the quasistatic regime [31] can be for n 2 and Tr[ ] denotes the trace operation. Fur- obtained from the nonlocal strong-contrast expan- thermore,≥ for statistically isotropic media,î the effectiveó sions (49) by simply replacing the exponential func- dielectric constant becomes independent of the direction tions that appear in the expressions for the second- A(p) of the wave vector, i.e., εe(kq)= εe(kq). rank tensors n (kq), defined by Eq. (52), by Remarks: unity. i. Importantly, the strong-contrast expansion (49) is vii. For statistically isotropic media, the effective phase a series representation of a linear fractional trans- speed ce(kq) and attenuation coefficient γe(kq) are formation of the variable εe(kq) (left-hand side of determined by the scalar effective dielectric con- the equation), rather than the effective dielectric stant εe(kq): constant tensor itself. The series expansion in pow- −1 ers of the polarizability βpq of this particular ratio- ce(kq) /c =ne(kq ) = Re 1/ εe(kq) , (55) nal function of εe(kq) has important consequences −1 h i for the predictive power of approximations derived γe(kq) /c =κe(kq ) = Im 1/»εe(kq) , (56) from the expansion, as detailed in Sec. IIIB. h i where ne(kq) and κe(kq) are the» effective refractive ii. The fact that the exact expansion (49), extracted and extinction indices, respectively. The quantity from our nonlocal relation (46), is explicitly given exp( 2πγe/ce) is the factor by which the incident- in terms of integrals over products of the rele- wave− amplitude is attenuated for a period 2π/ω. vant Green’s functions and the n-point correlation functions to infinite order implies that multiple- viii. The strong-contrast formalism is a significant de- scattering to all orders is exactly treated for the parture from perturbative expansions obtained range of wave numbers for which our extended ho- from standard multiple-scattering theory [17, 20, S mogenization theory applies, i.e., 0 kq ℓ . 1. 33, 34]. The operator defined in Eq. (38) is a gen- ≤ | | eralization of the standard scattering operator T = −1 iii. Note that Eq. (49) represents two different series [I VG] V [17, 20, 33], where V [ε(r) εq]I is expansions: one for q = 1 and p = 2 and the other the−scattering potential. Thus, the perturbation≡ − se- for q = 2 and p = 1. ries resulting from T is a weak-contrast expansion 10
with the inherent limitations that it converges only for small contrast ratios (see also Sec. IIIB). By contrast, due to the different possible choices for the exclusion regions and reference phases, there is an infinite variety of series expansions that re- sult from the strong-contrast formalism with gen- erally fast-convergence properties. The particular strong-contrast expansion can be designed for dif- ferent classes of microstructures (see Appendixes A and D). The corresponding strong-contrast “self- energy” Σ is a linear fractional transform of Le −1 −1 −1 [see Eq. (48)], namely, Σ = Le I DLe , which is a generalization of the self-energy− in stan- dard multiple-scattering theory [17, 20, 33, 34]. Thus, it is highly nontrivial to relate diagram- matic expansions of the strong-contrast formalism to those of multiple-scattering theory. An elabo- ration of how strong-contrast expansion generalize those from multiple-scattering theory is presented FIG. 3. Schematic of the optimal multiscale “coated-spheres” in the Supplementary Material [81]. model that realizes the isotropic Hashin-Shtrikman bounds on εe [85]. Each composite sphere is composed of a spherical inclusion of one phase (dispersed phase) that is surrounded by a concentric spherical shell of the other phase such that B. Convergence properties and accuracy of the fraction of space occupied by the dispersed phase is equal truncated series to its overall phase volume fraction. The composite spheres fill all space, implying that their sizes range down to the in- The form of the strong-contrast expansion parameter finitesimally small. When phase 2 is the disconnected inclu- sion (dispersed) phase, this two-phase medium minimizes and βpq in Eq. (49) is a direct consequence of the choice of maximizes the effective static dielectric constant εe for pre- a spherical region excluded from the volume integrals in scribed volume fraction and contrast ratio, when ε2/ε1 > 1 Eq. (49) due to singularities in the Green’s functions and ε2/ε1 < 1, respectively. It has recently been proved that [31]. It is bounded by these highly degenerate optimal Hashin-Shtrikman multiscale distributions of spheres are hyperuniform [57, 86]. 1 ε ε β p − q 1, (57) −d 1 ≤ pq ≡ ε + (d 1)ε ≤ − p − q which implies that the strong-contrast expansion (49) the coated spheres in the d orthogonal directions lead to can converge rapidly, even for infinite contrast ratio oriented coated ellipsoids that are optimal for the macro- εp/εq . Other choices for the shape of the exclu- scopically anisotropic case. The lower bound corresponds → ∞ sion region will lead to different expansion parameters to the case when the high-dielectric-constant phase is that will generally be bounded but can lead to expan- the dispersed, disconnected phase and the upper bound sions with significantly different convergence properties corresponds to the instance in which the high-dielectric- [11]. In Appendix A, we present the corresponding ex- constant phase is the connected matrix. Thus, Torquato pansions for disk-like and needle-like exclusion regions, [11, 89] observed that the strong-contrast expansions (49) which are exceptional cases that lead to slowly converg- in the static limit can be regarded as ones that perturb ing weak-contrast expansions with expansion parameter around such optimal composites, implying that the first (εp εq)/εq, and thus are unbounded when εp/εq . few terms of the expansion can yield accurate approxi- − → ∞ Importantly, in the purely static case, the expan- mations of the effective property for a class of particulate sion (49) becomes identical to one derived by Sen and composites as well as more general microstructures, de- Torquato [15] and its truncation after second-order terms pending on whether the high-dielectric phase percolates A(p) [i.e., setting n = 0 for all n 3] yields the generalized or not. For example, even when ε2/ε1 1, the dis- Hashin-Shtrikman bounds [85]≥ derived by Willis [87] that persed phase 2 can consist of identical or≫ polydisperse are optimal since they are realized by certain statistically particles of general shape (ellipsoids, cubes, cylinders, anisotropic composites in which there is a disconnected, polyhedra) with prescribed orientations that may or not dispersed phase in a connected matrix phase [88]. In overlap, provided that the particles are prevented from the case of an isotropic effective dielectric constant εe, forming large clusters compared to the specimen size. the optimal Hashin-Shtrikman upper and lower bounds Moreover, when ε /ε 1, the matrix phase can be 2 1 ≪ for any phase-contrast ratio ε2/ε1 are exactly realized by a cellular network [56]. Finally, for moderate values of the multiscale “coated-spheres” model, which is depicted the contrast ratio ε2/ε1, even more general microstruc- in Fig. 3 in two dimensions. Affine transformations of tures (e.g., those without well-defined inclusions) can be 11 accurately treated. Importantly, we show that for the dy- the two-point level: namic problem under consideration, the first few terms of the expansion (49) yield accurate approximations of ε −1 e(kq) I I A(p) ε (k ) for a similar wide class of two-phase media (see + dφpβpq (1 φpβpq) 2 (kq) βpq/φp e q εq ≈ − − Sec. VIII). Analogous approximations were derived and applied for the quasistatic regime [31, 36]. î (60)ó ∞ We now show how lower-order truncations of the se- = C(p)(k ) β n, (61) ries (49) can well approximate higher-order functionals n q pq n=0 (i.e., higher-order diagrams) of the exact series to all or- X ders in terms of lower-order diagrams. Such truncations C(p) of strong-contrast expansions are tantamount to approx- where the nth-order functional n (kq) for any n is given A(p) imate but resummations of the strong expansions, which in terms of the volume fraction φp and 2 (kq), which enables multiple-scattering and spatial dispersion effects has the following diagrammatic representation: to be accurately captured to all orders. Solving the left- ε hand side of Eq. (49) for e yields the rational function A(p) 2 (kq)= . (62) in βpq:
ε k Here the solid and wavy lines joining two nodes repre- e( q) I 2 I = + dφp βpq φp(1 φpβpq) sent the spatial correlation via χV (r) and a wave vector- εq − (q) k r h dependent Green’s function H (r) e−i q · between the ∞ −1 A(p)(k ) β n−1 . (58) nodes, respectively. The black node indicates a volume − n q pq integral and carries a factor of dε . The first several func- n=2 q X i C(p) tionals n (kq) are explicitly given as Expanding Eq. (58) in powers of the scalar polarizability C(p) I βpq yields the series 0 (kq)= , C(p) I 1 (kq)= dφp , ε ∞ e(kq) (p) n (p) (p) 2 = B (k ) β , (59) C (kq)= d A (kq)+ φp I , ε n q pq 2 2 q n=0 X C(p) d A(p) 2I 2 3 (kq)= î 2 (kq)+ φp ó , φp (p) where the first several functionals B (kq) are explicitly n (p) d î (p) 2 ó 3 (p) (p) (p) C (k )= A (k )+ φ I . given in terms of A , A ,..., A as 4 q 2 2 q p 0 1 n φp î ó B(p) I Thus, comparing (59) to (61), we see that truncation 0 (kq)= of the expansion (49) for [ε + (d 1)ε ] (ε ε )−1 B(p) I e − q · e − q 1 (kq)=dφp at the two-point level actually translates into approxi- B(p) A(p) 2I mations of the higher-order functionals to all orders in 2 (kq)=d 2 (kq)+ φp terms of the first-order diagram φp and the second-order B(p) d A(p) 2 A(p) diagram (62). This two-point truncation can be thought 3 (kq)= î 2 (kq) + φóp 3 (kq) φp of as an approximate but accurate resummed representa- h 2 A(p) 4I tion of the exact expansion (59). Note that the approxi- +2φp 2 (kq)+ φp mate expansion (61) is exact through second order in βpq. i Clearly, truncation of Eq. (49) at the three-point level (p) d (p) 3 (p) (p) B (kq)= A (kq) +2φp A (kq) A (kq) (see Appendix C) will yield even better approximations 4 φ2 2 2 · 3 p " of the higher-order functionals. 2 2A(p) 2 A(p) +3φp 2 (kq) + φp 4 (kq)
3 A(p) 4 A(p) 6I +2φp 3 (kq)+3φp 2 (kq)+ φp , IV. STRONG-CONTRAST APPROXIMATION # FORMULAS
T n where stands for n successive inner products of a Here we describe lower-order truncations of the strong- T second-rank tensor . contrast expansions that are expected to yield accurate Let us now compare the exact expansion (59) to the closed-form formulas for εe(kq) that apply over a broad one that results when expanding the truncation of the range of wavelengths (kq ℓ . 1), volume fractions and exact expression (49) for [ε + (d 1)ε ] (ε ε )−1 at contrast ratios for a wide class of microstructures. e − q · e − q 12
A. Macroscopically anisotropic media describe below. The direct- and Fourier-space represen- tations of F (Q) are given as For the ensuing treatment, it is convenient to rewrite 2d/2 Γ(d/2) i Q d/2−1 the expansions (49), valid for macroscopically anisotropic F (Q) Q2 Rd ≡− π 4 2πr media in , in the following manner: Zǫ (1) (Qr) e−iQÅ·r χ (ãr) dr (68) ε (k ) + (d 1)ε I d/2−1 V φ 2β2 e q − q × H p pq ε (k ) ε I Γ(d/2) χ˜ (q) e q − q = Q2 V dq . (69) M −2d/2πd+1 q + Q 2 Q2 Z =φ β I A(p)(k ) β n + (k ) , (63) | | − p pq − n q pq RM q n=2 The exponential exp( iQ r) in Eq. (68) arises from X the phase difference associated− · with the incident waves where the Mth-order remainder term is defined as at positions separated by r. In the quasistatic regime, ∞ this phase factor is negligible, and Eq. (68) reduces to (k ) A(p)(k ) β n. (64) the local attenuation function (Q) (derived in Ref. [31] RM q ≡ n q pq F n=M+1 and summarized in the Supplementary Material [81]) be- X cause it is barely different from unity over the correla- Truncating the exact nonlocal expansion (63) at the two- tion length associated with the autocovariance function and three-point levels, i.e., setting 2(kq) = 0 and χ (r). The strong-contrast approximation (67) was pos- (k ) = 0, respectively, yields R V R3 q tulated in Ref. [73] on physical grounds. By contrast, the present work derives it as a consequence of our exact ε (k ) + (d 1)ε I φ 2β2 e q − q =φ β I A(p)(k ) β 2, nonlocal formalism (Sec. III). p pq ε (k ) ε I p pq − 2 q pq e q − q For statistically isotropic media, the effective dielectric (65) constant as well as the attenuation function are indepen- ε k I 2 2 e( q) + (d 1)εq I A(p) 2 dent of the direction of the incident wave vector kq, and φp βpq − =φpβpq 2 (kq) βpq εe(kq ) εqI − thus, they can be considered as functions of the wave − h number, i.e., ε (k )= ε (k ) and F (k )= F (k ). Then, A(p) 3 e q e q q q + 3 (kq) βpq , (66) the real and imaginary parts of Eq. (68) can be simplified i as Compared to the quasistatic approximation [31], these Q2 π/2 nonlocal approximations substantially extend the range π2 0 χ˜V (2Q cos φ) dφ , d =2 Im[F (Q)] = − 2Q (70) of applicable wave number, namely, 0 kq ℓ . 1. Q q χ˜ (q) dq , d =3 ≤ | | ( − 2(2πR)3/2 0 V 2Q2 R ∞ 1 Re[F (Q)] = p.v. dq 2 2 Im[F (q)], B. Macroscopically isotropic media − π 0 q(Q q ) Z − (71) All of the applications considered in this paper, will focus on the case of macroscopically isotropic media, i.e., where Eq. (71) is valid for d = 2, 3, and p.v. stands for they are described by the scalar effective dielectric con- the Cauchy principal value. Following conventional us- age, we say that a composite attenuates waves at a given stant εe(kq) = Tr [εe(kq )]/d but depend on the direction wave number if the imaginary part of the effective dielec- of the wave vector kq. tric constant is positive. Recall that attenuation in the present study occurs only because of multiple-scattering 1. Strong-contrast approximation at the two-point level effects (not absorption). While it is the imaginary part of F (Q) that determines directly the degree of attenua- tion or, equivalently, Im[ε ], we see from Eq. (71) that Solving Eq. (54) for the effective dielectric con- e the real part of F (Q) is directly related to its imaginary stant ε (k ) yields the strong-contrast approximation for e q part. It is for this reason that we refer to the complex macroscopically isotropic media: function F (Q) as the (nonlocal) attenuation function. 2 εe(kq) dβpqφp =1+ (p) εq φ (1 β φ ) β A (k ) p − pq p − pq 2 q 2. Modified strong-contrast approximation at the two-point dβ φ 2 level =1+ pq p , (67) φ (1 β φ )+ (d−1)πβpq F (k ) p − pq p 2d/2Γ(d/2) q Here we extend the validity of the strong-contrast ap- proximation (67) so that it is accurate at larger wave (p) k where βpq is defined in Eq. (33), A2 ( q) numbers and hence better captures spatial dispersion. A(p) ≡ Tr 2 (kq) /d, and F (Q) is what we call the nonlo- This is done by an appropriate rescaling of the wave num- cal attenuation function of a composite for reasons we ber in the reference phase, kq, which we show is tanta- î ó 13 mount to approximately accounting for higher-order con- the scaled strong-contrast approximation to better cap- tributions in the remainder term 2(kq). Given that ture dispersive characteristics for ordered and disordered the strong-contrast expansion forR isotropic media per- models via comparison to finite-difference time-domain turbs around the Hashin-Shtrikman structures (see Fig. (FDTD) simulations; see Figs. 8 and 9, and Sec. VIII in 3) in the static limit, it is natural to use the scaling the Supplementary Material [81]. εHS/εqkq, where εHS is the Hashin-Shtrikman estimate, i.e., p V. MODEL MICROSTRUCTURES dφpβpq εHS εq 1+ , (72) ≡ 1 φpβpq We consider four models of 2D and 3D disordered me- − ï ò dia to understand the effect of microstructure on the which gives the Hashin-Shtrikman lower bound and up- effective dynamic dielectric constant, two of which are per bound if εp > εq and εp < εq, respectively. This nonhyperuniform (overlapping spheres and equilibrium scaling yields the following scaled strong-contrast approx- hard-sphere packings) and two of which are hyperuni- imation for statistically isotropic media: form (hyperuniform polydisperse packings and stealthy 2 hyperuniform packings of identical spheres). The par- εe(kq) dβpqφp =1+ . ticles of dielectric constant ε2 are distributed through- εq (d−1)πβpq εHS φp(1 βpqφp)+ d/2 F kq out a matrix of dielectric constant ε1. We also com- − 2 Γ(d/2) εq (73) pute the spectral density for each model, which is the re- We now show that the scaled approximation» (73) in- quired microstructural information to evaluate the nonlo- deed provides good estimates of leading-order correc- cal strong-contrast approximations discussed in Sec. VB. Representative images of configurations of the four tions of 2(kq) in powers of kq. To do so, we employ the conceptR of the averaged (effective) Green’s function aforementioned models of 2D disordered particulate me- dia are depicted in Fig. 4. Note that the degree of G(q)(q) of an inhomogeneous medium which in prin- volume-fraction fluctuations decreases from the leftmost ciple accounts for the all multiple-scattering events image to the rightmost one. ¨ ∂ 2 −1 (q) ω 2 2 G (q) = q ke(ω) I qq , (74) c − − A. Overlapping spheres where¨ k (ω)∂ ε (ω¶î)ω/c = ε ó(ω) /ε k© is the ef- e ≡ e e q q fective wave number at a frequency ω, and εe(ω) is the Overlapping spheres (also called the fully-penetrable- exact effective dynamicp dielectricp constant, assuming a sphere model) refer to an uncorrelated (Poisson) distri- well-defined homogenization description [17, 90]. Since bution of spheres of radius a throughout a matrix [11]. exact complete microstructural information is, in prin- For such nonhyperuniform models at number density ρ ciple, accounted for with the effective Green’s function in d-dimensional Euclidean space Rd, the autocovariance (74), the exact strong-contrast expansion can be approx- function is known analytically [11]: imately equated to the one truncated at the two-point χ (r) = exp( ρv (r; a)) φ 2, (77) level with an attenuation function given in terms of the V − 2 − 1 effective Green’s function, i.e., where φ = exp( ρ v (a)) is the volume fraction of the 1 − 1 matrix phase (phase 1), v1(a) is given by (13), and 2 2 εe(kq) + (d 1)εq φ β − v2(r; a) represents the union volume of two spheres whose p pq ε (k ) ε e q − q centers are separated by a distance r. In two and three =φ β A(p)(k ) β 2 + (k ) , (75) dimensions, the latter quantity is explicitly given, respec- p pq − 2 q pq R2 q (p) 2 tively, by φpβpq A2 (ke(ω)) βpq . (76) ≈ − 2 2 1/2 (p) 2 Θ(x 1)+ π + x 1 x When the functional form of A (Q) or, equivalently, v2(r; a) − π − , d =2 2 −1 = cosh (x) Θ(1 x) F (Q) is available, it is possible to solve Eq. (76) for εe(kq) v1(a) − 3 − in a self-consistent manner. Instead, we show that by as- 3x i x 2 Θ(x 1)+ 1+ 2 2 Θ(1 x) , d =3 suming ke(ω) εHS/εqkq, which results in Eq. (73), − − − ≈ we obtain good estimates of the leading-order corrections where x r/2a, and Θ(xÄ) (equal to 1ä for x > 0 and p ≡ of 2(kq) in powers of kq. Thus, the scaled approxima- zero otherwise) is the Heaviside step function. For d =2 R tion (73) provides better estimates of the higher-order and d = 3, the particle phase (phase 2) percolates when three-point approximation (given in Appendix C) than φ2 0.68 (Ref. [91]) and φ2 0.29 (Ref. [92]), respec- the unmodified strong-contrast approximation (60). This tively.≈ The corresponding spectral≈ densities are easily can be easily confirmed in the quasistatic regime from the found numerically by performing the Fourier transforms (p) (p) small-kq expansions of A2 (kq) and A3 (kq) given in Ref. indicated in Eq. (7). In this work, we apply this model [31]. We also confirm the improved predictive capacity of for φ2’s well below the percolation thresholds. 14
FIG. 4. Representative images of configurations of the four models of 2D disordered particulate media described in this section. These include (a) overlapping spheres, (b) equilibrium packings, (c) class I hyperuniform polydisperse packings, and (d) stealthy hyperuniform packings. For all models, the volume fraction of the dispersed phase (shown in black) is φ2 = 0.25. Note that (a) and (b) are not hyperuniform. While these models consist of distributions of particles, both overlapping and nonoverlapping, the formulas derived in Sec. IV can be applied to any two-phase microstructure. Indeed, Appendix D describes applications to media with phase-inversion symmetry.
B. Equilibrium packings is repeated over all cells. The final packing fraction is N φ2 = j=1 v1(aj) /VF = ρ v1(a), where ρ is the num- Another disordered nonhyperuniform model we treat ber density of particle centers and a represents the mean is distributions of equilibrium (Gibbs) of identical hard sphereP radius. In the small- Q regime, the spectral den- | | spheres of radius a along the stable fluid branch [11, 93]. sities of the resulting particulate composites exhibit a 4 The structure factors of such packings are well approx- power-law scalingχ ˜V (Q) Q , which are of class I. ∼ | | imated by the Percus-Yevick solution [11, 93], which is analytically solvable for odd values of d. For d = 3, the Percus-Yevick solution gives the following expression for D. Stealthy hyperuniform packings the structure factor S(Q) [11]: Stealthy hyperuniform particle systems, which are also 3 16πa 2 class I, are defined by the spectral density vanishing S(Q)= 1 ρ 6 24a1φ2 12(a1 +2a2)φ2q − q − around the origin, i.e.,χ ˜V (Q) = 0 for 0 < Q QU; | | ≤ n 4 see Eq.(14). We obtain the spectral density from real- + (12a2φ2 +2 a1 + a2φ2)q ] cos(q) izations of disordered stealthy hyperuniform packings for + [24a φ q 2(a +2a φ + 12a φ )q3 sin(q) 1 2 − 1 1 2 2 2 d =2, 3 that are numerically generated via the following −1 two-step procedure. Specifically, we first generate such 24φ (a a q2) , (78) − 2 1 − 2 point configurations consisting of N particles in a fun- o damental cell F under periodic boundary conditions via 2 4 where q = 2Qa, a1 = (1+2φ2) /(1 φ2) , and a2 = the collective-coordinate optimization technique [67–69], 2 4 − (1+0.5φ2) /(1 φ2) . Using this solution in conjunc- which amounts to finding numerically the ground-state − − tion with Eq. (11) yields the corresponding spectral den- configurations for the following potential energy; sityχ ˜V (Q). For d = 2, we obtain the spectral density from Monte Carlo generated disk packings [11]. N 1 Φ r = v˜(Q) S(Q)+ u(rij ) , (79) VF Q i 1,Q < Q Q , Class I hyperuniform packings of spheres with a poly- v˜(Q)= L | |≤ U (80) dispersity in size can be constructed from nonhyperuni- 0, otherwise, form progenitor point patterns via a tessellation-based ® procedure [57, 86]. Specifically, we employ the centers and a soft-core repulsive term [94] of 2D and 3D configurations of identical hard spheres in equilibrium at a packing fraction 0.45 and particle num- (1 r/σ)2, r < σ, u(r)= − (81) ber N = 1000 as the progenitor point patterns. One 0, otherwise. begins with the Voronoi tessellation [11] of these pro- ® genitor point patterns. We then rescale the particle In contrast to the usual collective-coordinate procedure in the jth Voronoi cell Cj without changing its center [67–69], the interaction (79) used here also includes a such that the packing fraction inside this cell is iden- soft-core repulsive energy (81), as done in Ref. [94]. tical to a prescribed value φ2 < 1. The same process Thus, the associated ground-state configurations are still 15 disordered, stealthy and hyperuniform, and their nearest- hyperuniform and hyperuniform media for long and in- neighbor distances are larger than the length scale σ due termediate wavelengths. We also provide plots of both to the soft-core repulsion u(r). Finally, to create pack- the real and imaginary parts of F (Q) for the four models ings, we follow Ref. [70] by circumscribing the points by of disordered two-phase media considered in this work, identical spheres of radius a<σ/2 under the constraint which depends on wave number Q. that they cannot overlap (See the Supplementary Mate- The function F (Q) depends on the microstructure via rial [81] for certain results concerning stealthy “nonhype- the spectral densityχ ˜V (Q). Thus, assuming that the α runiform” packings in which QL > 0.) The parameters latter quantity has the power-law scalingχ ˜V (Q) Q used to generate these disordered stealthy packings are as Q 0, the asymptotic behavior of F (Q) in the∼ long- summarized in the Supplementary Material [81]. wavelength→ limit (Q 0) follows as → Qd, nonhyperuniform Im[F (Q)] , as Q 0, E. Spectral Densities for the Four Models ∼ Qd+α, hyperuniform → ® (82) Here, we plot the spectral densityχ ˜ (Q) for the four V Re[F (Q)] Q2, as Q 0, models at the selected particle-phase volume fraction of ∼ →(83) φ2 = 0.25; see Fig. 5. From the long- to intermediate- wavelength regimes (Qa . 4), their spectral densities where we use Eqs. (68) and (70). Recall that the expo- are considerably different from one another. Overlap- nent α lies in the open interval (0, ) for hyperuniform ping spheres depart the most from hyperuniformity, fol- systems (see Sec. II D). For nonhyperuniform∞ systems lowed by equilibrium packings. Stealthy packings sup- studied here, we take α = 0. The reader is referred to press volume-fraction fluctuations to a greater degree the Supplementary Material [81] for derivations of Eqs. than hyperuniform polydisperse packings over a wider (82) and (83). Importantly, in the quasistatic regime, range of wavelengths. In the small-wavelength regime the imaginary parts of the effective dielectric constant (Qa 4), all four curves tend to collapse onto a single ≫ for both strong-contrast approximations, (67) and (73), curve, reflecting the fact that all four models are com- are determined by the asymptotic behaviors F (Q) indi- posed of spheres of similar sizes. cated in Eq. (82), i.e., 0.8 Im[εe(kq)] Im[F (kq)] Overlapping spheres ∼ d Equilibrium packing kq , nonhyperuniform 0.6 Hyperuniform polydisperse packing d+α , as kq 0. (84) ∼ kq , hyperuniform → ) Stealthy hyperuniform; Q a = 1.5 U ® Q ( 0.4 Thus, hyperuniform media are less lossy than their non- V χ ~ hyperuniform counterparts in the quasistatic regime. 0.2 In the case of stealthy hyperuniform media [i.e., χ˜ (Q) = 0 for 0 Q Im[F (Q)]=0, for 0 Q source (67) agrees with the one obtained from previous simula- (a) Periodic tion results for stealthy hyperuniform “point” scatterers PML PML boundary [63], not the finite-sized particles considered here. condi¡ ons Figure 6 shows F (Q) for the four distinct models of 3 disordered particulate media in R : overlapping spheres, source equilibrium packings, stealthy hyperuniform packings, (b) and hyperuniform polydisperse packings. (Its 2D coun- terpart is provided in the Supplementary Material [81].) Periodic We clearly see that these attenuation functions ex- boundary PML PML hibit common large-Q behaviors, regardless of the mi- condi ons crostructures. From the quasistatic to the intermediate- wavelength regimes (Qa < 1), however, the attenuation characteristics (imaginary parts Im[F (Q)]) are consider- ably different from one model to another. For example, stealthy hyperuniform media are transparent up to a fi- nite wavelength, and hyperuniform polydisperse packings exhibit much less attenuation than nonhyperuniform sys- FIG. 7. Schematic of the general simulation setup for either (a) periodic or (b) nonperiodic composites consisting of N tems. spheres of radius a in a matrix. In both cases, Gaussian pulses 0.08 of electric fields propagate from the planar sources (shown in (a) Overlapping spheres )] 0.06 Equilibrium packing red lines) to the packings (shown in black circles). The wave Q Stealthy hyperuniform; QUa = 1.5 ( number (spectrum) of the pulses spans between min[k1] and F 0.04 Hyperuniform polydisperse packing max[k1]. Periodic boundary conditions are applied along all 0.02 -Re[ directions, except for the propagation direction xˆ. The per- 0 -1 (b) 10 fectly matched layers (PML, shown in blue) of thickness LPML )] 0.08 -4 Q 10 are placed at both ends of the simulation box to absorb any ( -7 F 10 0.04 0.1 1 reflected and transmitted waves. To estimate the effective dielectric constant as described in step 2 below, we consider -Im[ 0 the subregion V (shown in gray) that excludes from the com- 0 1 2 Qa posite within the simulation box two relatively thin slabs of thickness Lboundary along the propagation direction. FIG. 6. The negative values of (a) the real and (b) imaginary parts of the nonlocal attenuation function F (Q) as a function of the dimensionless wave number Qa [defined in Eq. (68)] for 1. For a given medium, we obtain the steady-state the four models of 3D disordered composite media considered in this paper. The inset in (b) is the log-log plot of the larger spatial distributions of electric field Ey(r,ω) and panel. The volume fraction of the dispersed phase for each electric displacement field Dy(r,ω) at a given fre- model is φ2 = 0.25. The first three models consist of identical quency ω (or the corresponding wave number k1). spheres of radius a. For class I hyperuniform polydisperse Specifically, the planar source generates Gaussian particulate media, a is the mean sphere radius. pulses of electric fields that propagate along the xˆ direction with wave number k1 that spans between min[k1] and max[k1]. Using the aforementioned MEEP package [98], we compute time evolution of r VII. SIMULATION PROCEDURE TO electric field Ey( ,t) and electric displacement field D (r,t) for a period of time 6π√ε / c min[k ] , COMPUTE EFFECTIVE DYNAMIC y 1 { 1 } DIELECTRIC CONSTANT where c is the speed of light in vacuum. We then compute the temporal Fourier transforms of these fields inside packings. The values of the simulation In Ref. [73], we established preliminary comparisons of parameters (indicated in Fig. 7) for the 2D and 3D strong-contrast approximation (67) and numerical sim- ordered and disordered models studied in this arti- ulations via the extended version of the fast-Fourier- cle are summarized in the Supplementary Material Transform-based technique [95, 96]. Because of conver- [81]. gence issues, however, in this paper, we employ a more reliable numerical technique, i.e., the FDTD method [97], using an open-source software package [98]. We focus 2. At each value of k , we postprocess E (r,ω) and here on particulate media and take the matrix to be the 1 y D (r,ω) to estimate the effective dielectric con- reference medium (phase 1) and the particles to be the y stant ε (k ,ω) of a single configuration by solving polarized phase (phase 2). ∗ 1 the following self-consistent equation: The general simulation setup is schematically illus- trated in Fig. 7. The simulation procedure for macro- scopically isotropic media is carried out in three steps: ε∗(k1)= ε∗(k1) , (87) 17 where periodic packings (square and simple-cubic lattice pack- ˜ 2 ings), which necessarily depends on the direction of the Dy(ke,ω) ke k ε∗(k1) , ε∗(k1) , incident wave 1. While these periodic packings are ≡ E˜y(ke,ω) ≡ ω/c macroscopically isotropic, due to their cubic symmetry, Å ã they are statistically anisotropic (see Supplementary Ma- ke is a complex-valued effective wave number, and terial [81] for details). For simplicity, we consider only 1 the case where k1 is aligned with one of the minimal lat- D˜ (q,ω) D (r,ω) e−iqxˆ·r dr , y ≡ V y tice vectors, i.e., the Γ-X direction in the first Brillouin | | ZV zone. Such periodic models enable us to validate our sim- 1 −iqxˆ·r E˜ (q,ω) E (r,ω) e dr , ulations because εe(k1) also can be accurately extracted y ≡ V y | | ZV from the lowest two photonic bands that are calculated where V is a rectangular parallelepiped subregion via MPB, an open-source software package [100]. The re- within the composite (shown in gray in Fig. 6) sults from the band-structure calculations and our FDTD that is slightly smaller than the simulation box and simulations show excellent agreement. In particular, our is used to reduce undesired boundary effects. The simulations accurately predict two salient dielectric char- homogenization task is carried out by numerically acteristics that must be exhibited by periodic packings: 2 transparency up to a finite wave number associated with finding the minimizer of ε∗ ε∗ with an initial | − | the edge of the first Brillouin zone (i.e., Im[εe] = 0 guess ε∗ = εHS via the Broyden-Fletcher-Goldfarb- Shanno (BFGS) nonlinear optimization algorithm for 0 k1 . π), and resonancelike attenuation due to Bragg≤ | diffraction| within the photonic band gap [101] [99], where εHS is the Hashin-Shtrikman estimate given by Eq. (72). Details of step 2 are provided (i.e., a peak in Im[εe] or, equivalently, a sharp transition in the Supplementary Material [81]. in Re[εe] [102]). Thus, our numerical homogenization scheme is valid down to intermediate wavelengths (see 3. Steps 1-2 are repeated for a sufficient number of the Supplementary Material [81] for comparison of the configurations for disordered media. Then, we com- band-structure and FDTD computations). pute the effective dielectric constant at a given Importantly, while our strong-contrast approximations k1 by ensemble averaging ε∗(k1), i.e., εe(k1) = Eqs. (67) and (73) account for directionality of the inci- ε∗(k1) . dent waves, the MGA and QCA are independent of the h i direction of k . In Fig. 8, the FDTD simulation results It is important to note that we ensure that the run 1 are compared with the MGAs for d = 2, 3 [Eqs. (17) times employed in step 1 are sufficiently long such that and (16)], QCA (18) for d = 3, as well as the unscaled the computed effective dielectric constants achieve stable and scaled strong-contrast approximations (67) and (73) and accurate steady-state values. We emphasize that the for d = 2, 3. While all approximations agree with the result ε (k ) obtained from Eq. (87) in step 2 is nonlo- ∗ 1 FDTD simulations in the quasistatic regime, the MGA cal in space because it is calculated from the nonlocal and QCA fail to capture properly two key features: no constitutive relation D˜ (k ,ω) = ε (k ) E˜ (k ,ω). For y e ∗ 1 y e loss of energy up to a finite wave number and resonance- periodic media, step 3 is unnecessary because all config- like attenuation in the band gaps. Each strong-contrast urations are identical. approximation captures both of these salient character- istics. However, it is noteworthy that the scaled strong- VIII. COMPARISON OF SIMULATIONS OF contrast approximation [Eq. (73)] agrees very well with εe(k1) TO VARIOUS APPROXIMATIONS the FDTD simulations. For contrast ratios ε2/ε1 < 1, FORMULAS FDTD simulations are also in very good agreement with the predictions of strong-contrast approximations for a In this section, we compare our simulations of the ef- wide range of wave numbers, as detailed in the Supple- mentary Material [81]. fective dynamic dielectric constant εe(k1) for various 2D and 3D ordered and disordered model microstructures to the predictions of the strong-contrast formulas as well as to conventional approximations, such as MGA (17) and B. Disordered nonhyperuniform and hyperuniform QCA (18). Most of these models provide stringent tests media of the predictive power of the approximations at finite wave numbers because they are characterized by non- To test the predictive capacity of approximation for- trivial spatial correlations at intermediate length scales. mulas for εe(k1) for disordered media as measured against simulations, we choose to study two distinctly different models: disordered nonhyperuniform packings A. 2D and 3D periodic media (equilibrium packings) and disordered stealthy hyperuni- form disordered packings [i.e.,χ ˜V (Q)=0 for 0 Qa < We first carry out our FDTD simulations for the ef- 1.5] for both 2D and 3D. Again, we compare our≤ simula- fective dynamic dielectric constant εe(k1) of 2D and 3D tion results to the MGAs [Eqs. (17) and (16)], QCA (18), 18 2 1.6 Simple cubic lattice packing ] ] 3D equilibrium packing 1 1 φ = 0.25, ε /ε = 4 ε ε φ = 0.25, ε /ε = 4 1.8 2 2 1 1.55 2 2 1 )/ 1 ) / 1 k ( k e ( 1.6 e ε 1.5 ε Re[ 1.4 Re[ 1.45 FDTD Simulations FDTD simulations ] 0.15 ] 1 Maxwell-Garnett approx. 1 Maxwell-Garnett approx. ε Quasicrystalline approx. ε Quaiscrystalline approx. )/ 0.2 1 Strong-contrast approx. ) / 0.1 Strong-contrast approx. k 1 ( k e Scaled strong-contrast approx. Scaled strong-contrast approx. ( ε 0.1 e ε 0.05 Im[ 0 Im[ 0 0 1 2 3 0 0.2 0.4 0.6 0.8 1 k1 L k1a (a) (a) ] Square lattice packing 1.6 1 ] ε 1.6 φ = 0.25, ε /ε = 4 3D stealthy hyperuniform packing 2 2 1 1 ε φ = 0.25, ε /ε = 4 ) / 1.55 2 2 1 1 ) / k 1 ( e k 1.4 ( ε e 1.5 ε Re[ Re[ 1.45 1.2 ] FDTD simulations 1 0.15 FDTD simulations Maxwell-Garnett approx. ] ε 1 Maxwell-Garnett approx. 0.2 Strong-contrast approx. ε Quasicrystalline approx. ) / 1 Scaled strong-contrast approx. ) / 0.1 Strong-contrast approx. k 1 ( k e Scaled strong-contrast approx. ( ε 0.1 e ε 0.05 Im[ 0 Im[ 0 0 1 2 3 0 0.2 0.4 0.6 0.8 1 k1L k1a (b) (b) FIG. 8. Comparison of the predictions of the strong-contrast FIG. 9. Comparison of the predictions of the strong-contrast formulas, Eqs. (67) and (73), to the Maxwell-Garnett [Eqs. formulas Eqs. (67) and (73) to the MGA(16) and QCA (18) (17) and (16)] and QCA (18) approximations for the effec- approximations for the effective dynamic dielectric constant tive dynamic dielectric constant εe(k1) as a function of the εe(k1) as a function of the dimensionless wave number k1a of dimensionless wave number k1L of periodic packings to our 3D disordered sphere packings to our corresponding computer corresponding computer simulation results. We consider (a) simulation results. We consider (a) equilibrium packings and 3D simple cubic lattice and (b) 2D square lattice of packing (b) stealthy hyperuniform packings [˜χV (Q) = 0 for 0 Qa < fraction φ2 = 0.25 and contrast ratio ε2/ε1 = 4. Here, k1 is ≤ 1.5] of sphere radius a, packing fraction φ2 = 0.25, and phase- the wave number in the reference (matrix) phase along the contrast ratio ε2/ε1 = 4. Here, k1 is the wave number in the Γ-X direction, and L is the side length of a unit cell. reference (matrix) phase, and the error bars in the FDTD simulations represent the standard errors over independent configurations. strong-contrast approximation (67), and the scaled coun- terpart (73). The conventional approximations fail to capture spatial dispersion effects. Specifically, the MGA neglects any microstructural information, except for the proximation provides excellent estimates of εe(k1) for particle shape, and thus cannot account for long-range both disordered models, even for large wave numbers correlations, such as the lossless property of stealthy hy- (0 k1a 1); see Fig. 9. Moreover, the predictions peruniform media. By contrast, while the QCA formula of both≤ strong-contrast≤ approximations accurately cap- yields better estimates of Im[εe] for nonhyperuniform ture the salient microstructural differences between the systems, it cannot generally capture the correct trans- nonhyperuniform and hyperuniform models because they parency characteristics of hyperuniform systems, e.g., it incorporate spatial correlations at finite wavelengths via incorrectly predicts Im[εe(k1)] = 0 for all wave numbers, the spectral densityχ ˜V (Q). For example, they properly regardless of whether the medium is stealthy hyperuni- predict that stealthy hyperuniform media are lossless up form or nonstealthy hyperuniform; see Fig. 9 (b). to a finite wave number, even if at different cutoff values; On the other hand, the scaled strong-contrast ap- see Eq. (86). Corresponding 2D results are presented in 19 the Supplementary Material [81] because they are quali- 2.1 φ =0.25, ε /ε = 10 ] d=3, 2 2 1 tatively the same as the 3D results. 1 ε 1.8 / Overlapping spheres e ε 1.5 Equilibrium packing Stealthy hyperuniform; Q a = 1.5 IX. PREDICTIONS OF STRONG-CONTRAST U Re[ 1.2 Hyperuniform polydisperse packing APPROXIMATIONS FOR DISORDERED PARTICULATE MEDIA ] 1 ε -1 0.4 10 / 3 e Having established the accuracy of the scaled strong- ε -4 10 contrast approximation (73) for ordered and disordered 0.2 7 -7 media in the previous section, we now apply it to the Im[ 10 0.1 1 four different disordered models discussed in Sec. V in 00 0.5 1 1.5 2 2.5 order to study how εe(k1) varies with the microstructure. k1a We first study how εe(k1) varies with k1 at a fixed con- (a) trast ratio ε2/ε1 = 10 for the four models; see Fig. 10. According to Eq. (84), nonhyperuniform and hyperuni- d=2, φ =0.25, ε /ε =10 form media in the quasistatic regime have the different ] 1.6 2 2 1 d d+α 1 scalings, i.e., Im[εe(k1)] k1 and Im[εe(k1)] k1 , ε ∼ ∼ / respectively, where α> 0 for hyperuniform systems. This e Overlapping disks implies that hyperuniform media are less lossy than their ε Equilibrium packing 1.4 Stealthy hyperuniform; QUa=1.3 nonhyperuniform counterparts as k1 tends to zero, as Re[ Hyperuniform polydisperse packing seen in the insets of Fig. 10. Moreover, beyond the ] quasistatic regime, each model exhibits “effective” trans- 1 0.2 parency for a range of wave numbers that depends on ε -1 / 10 2 the microstructure. For 2D and 3D models, hyperuni- e ε -4 0.1 10 form polydisperse packings tend to be effectively trans- 6 -7 parent for a wide range of wave numbers compared to the Im[ 0 100.01 0.1 nonhyperuniform ones, while the stealthy hyperuniform 0 0.5 1 1.5 2 systems are perfectly transparent for the widest range of k a wave numbers, as established in Eq. (86) and Sec. VIII. 1 For each model, the “effective” transparency spectral (b) range must be accompanied by normal dispersion [i.e., an FIG. 10. Predictions of the scaled strong-contrast approxima- increase in Re[εe(k1)] with k1] [32] because our strong- tion (73) for the effective dynamic dielectric constant εe(k1) contrast approximation is consistent with the Kramers- as a function of the dimensionless wave number k1a of the Kronig relations (see Appendix B). Moreover, we see that four models of disordered media at volume fraction φ2 = 0.25 anomalous dispersion [i.e., a decrease in Re[εe(k1)] with and contrast ratio ε2/ε1 = 10: (a) three dimensions and (b) k1] occurs at wave numbers larger but near the respec- two dimensions. The inset in the lower panel is the log-log tive transition between the effective transparency and plot of the larger panel. appreciable attenuation, which again is dictated by the Kramers-Kronig relations. The specific anomalous dis- persion behavior is microstructure dependent. we then shrink particle radii to attain a packing fraction We now examine how the imaginary part Im[ε ] varies φ2 = 0.25, whose stealthy regime is now QUa 1.33. e ≈ with the contrast ratio ε2/ε1 for the disordered models The coefficients ce(k1) and γe(k1) for these packings with for a given large wave number k1 inside the transparency ε2/ε1 = 4 are estimated from the scaled approximation interval for 2D and 3D stealthy hyperuniform systems. (73); see Fig. 12. It is seen that the waves propagate sig- These results are summarized in Fig. 11. The dispar- nificantly more slowly through the denser medium due to ity in the attenuation characteristics across microstruc- an increase in multiple-scattering events. Moreover, the tures widens significantly as the contrast ratio increases. transparency intervals (wave-number ranges where the Clearly, overlapping spheres are the lossiest systems. Hy- effective attenuation coefficients vanish) are larger for the peruniform polydisperse packings can be nearly as loss- packing with the higher stealthy cutoff value QU a =1.5a less as stealthy hyperuniform ones. (φ2 =0.4), as predicted by Eq. (86). We also study the effect of packing fraction φ2 on the effective phase speed ce(k1) and effective attenuation co- X. DISCUSSION efficient γe(k1), as defined by Eqs. (55) and (56), re- spectively. For concreteness, we focus on 3D stealthy hyperuniform packings. We first generate such packings All previous closed-form homogenization estimates of at a packing fraction φ2 = 0.4 and QUa = 1.5, as de- the effective dynamic dielectric constant apply only scribed in Sec. V D. Without changing particle positions, at long wavelengths (quasistatic regime) and for very 20 ε /ε = 4 φ =0.25 ε ε )1/2 0.85 Stealthy hyperuniform packings, 2 1 d=3, 2 , ( HS / 1 k1a=0.7 0.16 1 Overlapping spheres c Equilibrium packings ) / 0.8 φ =0.40, Q a=1.5 1 2 U k ] Stealthy hyperuniform packings; Q a = 1.5 φ =0.25, U ( 2 Q a=1.33 0.12 e U 1 Hyperuniform polydisperse packings c ε 0.75 / e ε 0.08 0.001 1 c 0.06 Im[ ) / 0 0.04 1 k ( e 0.03 γ 0.5 0.6 0.7 0.8 0 - 0 2 4 6 8 10 0 ε /ε 0 1 2 2 1 k1a (a) 1/2 FIG. 12. Predictions of the scaled strong-contrast approxi- 0.15 d=2, φ =0.25, (ε /ε ) k a=0.6 2 HS 1 1 mation (73) for the effective wave speed ce and the negative of Overlapping disks the attenuation coefficient γe as a function of the dimension- Equilibrium packings less wave number k1a for 3D stealthy hyperuniform sphere ] Stealthy hyperuniform dispersions; QUa = 1.3 1 packings of contrast ratio ε2/ε1 = 4 at two different pack- 0.1 Hyperuniform polydisperse packings ε ing fractions: φ2 = 0.4 with QUa = 1.5 and φ2 = 0.25 with / e QUa 1.33. The inset is a magnification of the lower panel. ε ≈ Im[ 0.05 formation through the spectral densityχ ˜V (Q), which is easily ascertained for general microstructures either theo- 0 retically, computationally, or via scattering experiments. 0 2 4 6 8 10 Depending on whether the high-dielectric phase perco- ε /ε 2 1 lates or not, the wide class of microstructures that we (b) can treat includes particulate media consisting of identi- cal or polydisperse particles of general shape (ellipsoids, FIG. 11. Predictions of the strong-contrast approximation cubes, cylinders, polyhedra) with prescribed orientations (73) for the effective dynamic dielectric constant εe(k1) as that may or not overlap, and cellular networks as well as function of dielectric-contrast ratio ε2/ε1 for the four disor- media without well-defined inclusions (Sec. IIIB). Our dered models, as per Fig. 10, at packing fraction φ2 = 0.25 approximations account for multiple scattering across a and wave number (a) εHS/ε1 k1a = 0.7 in three dimensions range of wave numbers. Fourth, we carry out precise and (b) εHS/ε1 k1a = 0.6 in two dimensions. p full-waveform simulations for various 2D and 3D models p of ordered and disordered media to validate the accu- racy of our nonlocal microstructure-dependent approx- special macroscopically isotropic disordered compos- imations for wave numbers well beyond the quasistatic ite microstructures, namely, nonoverlapping spheres or regime. spheroids in a matrix. In this work, we lay the theoret- Having established the accuracy of the scaled strong- ical foundation that enables us to substantially extend contrast approximation (73), we then apply it to four previous work in both its generality and applicability. models of 2D and 3D disordered media (both nonhyper- First, we derive exact homogenized constitutive relations uniform and hyperuniform) to investigate the effect of mi- for the effective dynamic dielectric constant tensor εe(kq) crostructure on the effective wave characteristics. Among that are nonlocal in space from first principles. Second, other findings, we show that disordered hyperuniform our strong-contrast representation of εe(kq) exactly ac- media are generally less lossy than their nonhyperuni- counts for complete microstructural information (infinite form counterparts. We also find that our scaled formula set of n-point correlation functions) for arbitrary mi- (73) accurately predicts that disordered stealthy hyper- crostructures and hence multiple scattering to all orders uniform media possess a transparency wave number in- −1/2 for the range of wave numbers for which our extended terval [0, 0.5 QU(εHS/εq) ) [cf. (86)], where most non- homogenization theory applies, i.e., 0 kq ℓ . 1 (where hyperuniform disordered media are opaque. Note that, ℓ is a characteristic heterogeneity length≤ | scale).| Third, using multiple-scattering simulations, Leseur et al. [63] we extract from the exact expansions accurate nonlocal were the first to show that stealthy hyperuniform systems closed-form approximate formulas for εe(kq), relations should exhibit a transparency interval, but for “point” (67) and (73), which are resummed representations of scatterers, not the finite-sized scatterers considered here. the exact expansions that incorporate microstructural in- Interestingly, their transparency-interval prediction coin- 21 cides with the one predicted by our less accurate strong- structures calculations. Accordingly, it has been of pri- contrast formula [cf. Eq. (86)]. mary importance to bridge the gap between the treat- The accuracy of our nonlocal closed-form formulas has ments of ordered and disordered to better understand the important practical implications, since one can now use optical properties of correlated media. Thus, our work them to accurately and efficiently predict the effective represents an initial step toward a unified theory to de- wave characteristics well beyond the quasistatic regime of scribe the effective optical properties of both ordered and a wide class of composite microstructures without having disordered microstructures over a wide range of incident to perform computationally expensive full-blown simula- wavelengths. tions. Thus, our nonlocal formulas can be used to ac- There are a variety of directions for future research. celerate the discovery of novel electromagnetic compos- First, our formalism can be straightforwardly extended ites by appropriate tailoring of the spectral densities and to heterogeneous materials composed of more than two then constructing the corresponding microstructures by phases or continuous media. Second, it is also of in- using the Fourier-space inverse methods [36]. For exam- terest to extend our formalism to applications and phe- ple, from our findings in the present study, it is clear that nomena (e.g., magnetic effects) relevant to the smaller stealthy disordered particulate media can be employed as wavelengths noted for metamaterials [103–105]. low-pass filters that transmit waves “isotropically” up to a selected wave number. Moreover, using the spectral densities of the type found by Chen and Torquato [36] Appendix A: Different Expansions as a Result of for stealthy hyperuniform packings (characterized by a Different Exclusion-Region Shapes peak value at Q = QU with intensities that rapidly decay to zero for larger wave numbers) and formula (73), one To get a sense of how the resulting expansions change can design materials with refractive indices that abruptly due to the choice of the exclusion-region shape, we con- change over a narrow range of wave numbers. Of course, sider the aforementioned oriented spheroidal exclusion one could also explore the design space of effective wave region in the two limiting disk-like and needle-like cases. properties of nonhyperuniform disordered composite me- Comparing the expansion parameters in the limit cases dia for potential applications. given in Eq. (32) to the strong-contrast expansion with Previously, disordered media were often described us- a spherical exclusion region given in Eq. (54), one can ing cluster expansions [17, 20, 33], while ordered media obtain the counterparts of Eq. (54) with disk-like and were often studied through dispersion relations and band- needle-like exclusion regions. Specifically, we replace parameters in Eq. (54) according to the following mappings: ε (k ) + (d 1)ε dε (k ) β (ε ε )/(dε ), e q − q e q , disk-like, pq → p − q p ε (k ) ε →ε (k ) ε e q − q e q − q ε (k ) + (d 1)ε dε β (ε ε )/(dε ), e q − q q , needle-like, pq → p − q q ε (k ) ε →ε (k ) ε e q − q e q − q resulting in the following expansions, respectively, ∞ ε ε 2 dε (k ) ε ε ε ε n φ 2 p − q e q =φ p − q A(p)(k ; A∗ = 1) p − q , (A1) p dε ε (k ) ε p dε − n q dε p e q − q p n=2 p Å ã Å ã X Å ã ε ε 2 dε ε ε ∞ ε ε n φ 2 p − q q =φ p − q A(p)(k ; A∗ = 0) p − q . (A2) p dε ε (k ) ε p dε − n q dε q e q − q q n=2 q Å ã Å ã X Å ã A(p) ∗ Here the functionals n (kq ; A ) are identical to Eqs. (51) and (52), except for the exclusion-region shape. In the Supplementary Material [81], we discuss how to obtain the analogs of Eqs. (A1) and (A2) that apply to macroscopically anisotropic media. The corresponding series expansions involve tensorial expansion parameters that have rapid convergence properties for stratified and transversely isotropic media, respectively. Appendix B: Kramers-Kronig Relations mogeneous material implies such analyticity properties, the Kramers-Kronig relations enable one to directly link Kramers-Kronig relations connect the real and imagi- the real part of a response function to its imaginary part nary parts of any complex function that is analytic in the or vice versa, even if the real or imaginary parts are only upper half-plane and meets mild conditions [106, 107]. available in a finite frequency range [106, 107]. Thus, Since causality in a dielectric response function of a ho- when a heterogeneous material can be treated as a ho- 22 mogeneous material with a dynamic effective dielectric ties of the nonlocal attenuation function F (Q) (detailed constant εe(kq), Kramers-Kronig relations immediately in the Supplementary Material [81]) induce εe(kq) to have apply to the exact strong-contrast expansion (63), i.e., the following three properties necessary to satisfy the Kramers-Kronig relations (B1) and (B2): (i) εe(kq ) is ∞ 2 q Im[εe(q)] an analytic function in the upper half-plane of kq, (ii) Re[εe(kq)] = εe( )+ p.v. dq , (B1) 2 2 ∞ π 0 q k ε(kq) εq vanishes like 1/ kq as kq goes to infinity, and Z − q − | | | | ∞ (iii) Re[εe(kq)] and Im[ε(kq)] are even and odd functions 2kq Re[εe(q)] εe( ) Im[εe(kq)] = p.v. dq − ∞ , of k , respectively. Property (i) is valid if a+b F (k ) = 0, 2 2 q q − π 0 q k 6 Z − q which is met for all disordered systems considered here. (B2) The fact that the strong-contrast approximations satisfy Eqs. (B1) and (B2) makes physical sense since F (Q) where we assume a linear dispersion relation in the ref- involves G(q)(x, x′) [cf. Eq. (24)], which is the tempo- erence phase (i.e., kq = √εqω/c) and limω→∞ εe(ω)= εq is real-valued. The Kramers-Kronig relations may or ral Fourier transform of the retarded Green’s function G(q) ′ ′ may not be obeyed when the strong-contrast expansion is (x, t, x ,t ) [106] that accounts for causality. We truncated at the two-point level, yielding Eq. (67). Here also numerically show in the Supplementary Material [81] we analytically show that the effective dielectric constant that our approximations obey Eqs. (B1) and (B2). εe(kq) for isotropic media obtained from either the un- scaled or scaled strong-contrast formulas [see Eqs.(67) and (73)] also satisfies the Kramers-Kronig relations. We begin by rewriting either strong-contrast approxi- Appendix C: Strong-Contrast Approximation at the −1 Three-Point Level mation as εe(kq) εq + [a + b F (kq)] , where a and b are nonzero real numbers.≈ The general analytic proper- Here we present the strong-contrast approximation at the three-point level for a spherical exclusion region. It is (p) obtained from Eq. (54) by setting An = 0 for n 4 and by solving it in ε (k ): ≥ e q ε (k ) dβ φ 2 e q =1+ pq p , (C1) 2 (p) εq φ (1 β φ ) + (d 1)π/[2d/2 Γ(d/2)]β F (k ) β A (k ) p − pq p − pq q − pq 3 q where the local three-point parameter is given as 2 (dεq) 1 A(p)(k ) dx dx Tr H(q)(x x ) e−ikq ·(x1−x2) H(q)(x x ) e−ikq·(x2−x3) 3 q ≡− φ 1 2 d 1 − 2 · 2 − 3 p Zǫ ∆(p)(x , x , x ) î ó (C2) × 3 1 2 3 1 1 1 q q 2 4 2 2 q q 2 = 2d d 1 d 2 2 2 (d 1) kq + q1 q2 d(ˆ1 ˆ2) 1 − φp(2π) q 2 k q 2 k − · − Z 1 q 2 q ( ) − − ∆˜ (p)(q + k , q + k ) , (C3) × 3 1 q 2 q where, due to the statistical homogeneity, ∆˜ (p)(q , q ) dr dr e−iq1·r1 e−iq2·r2 S(p)(r ) S(p)(r ) φ S(p)(r , r ) . 3 1 2 ≡ 1 2 2 1 2 2 − p 3 1 2 Z Note that Eq. (C3) is obtained from Eq. (C2) via Parseval’sî theorem. The static limit of Eq.ó (C2) for statistically isotropic media is given by [11] d 2 dr ds A(p)(0) =(d 1)φ (1 φ )ζ = [d(ˆr ˆs)2 1] S(p)(r,s,t) S(p)(r) S(p)(s) /φ , (C4) 3 − p − p p Ω rd sd · − 3 − 2 2 p d ZZ Å ã where Ω is the surface area of a unit sphere in Rd, t r s and the parameterî ζ lies in the closed intervaló [0, 1]. d ≡ | − | p Appendix D: Strong-Contrast Formula for with phase-inversion symmetry. A two-phase medium Phase-Inversion Symmetric Media possesses phase-inversion symmetry if the morphology To further illustrate the flexibility and power of the strong-contrast formalism, we apply it to treat media 23 of phase 1 at volume fraction φ1 is statistically identi- 30] cal to that of phase 2 in the system when the volume −r/r0 fraction of phase 1 is 1 φ1 [11]. In particular, we fol- χV (r)= φ1φ2e , (D3) low the same procedure− used by Torquato [11] to derive strong-contrast expansions designed to apply to media where a positive quantity r0 represents a characteristic with phase-inversion symmetry in the static limit and length scale. Here, we take r0 = a/2. which can be regarded to be expansions that perturb Figure 13 compares the effective dynamic dielectric around the microstructures corresponding to the “self- constant for 3D Debye random media to that of 3D consistent” formula. All we need to do is add the expan- equilibrium hard spheres as predicted from the strong- sion (54) with p = 2 and q = 1 to that with p = 1 and contrast approximations devised, respectively, for phase- q = 2: inversion symmetric media [Eq. (D2)] and dispersions of particles [Eqs. (67)], which do not have such a symme- try. We see that Debye random media are more lossy εe(ω) + (d 1)ε1 εe(ω) + (d 1)ε2 φ2 − + φ1 − (D1) than equilibrium packings. εe(ω) ε1 εe(ω) ε2 φ = 0.25, ε /ε = 4 − − ] 2 2 1 ∞ (2) (1) 1 An (k1) n−2 An (k2) n−2 ε =2 d β21 + β12 , 1.5 ) / − − φ2 φ1 1 n=2 k ( X ñ ô e ε where the term (2 d) on the right-hand side is ob- 1.45 −1 − −1 Re[ tained from β21 + β12 . Since the exact expansion (54) is independent of choice of reference phase and be- ] 0.09 1 3D Debye random media cause the phase-invesrion symmetry places each phase ε 3D equilibrium packing on the same footing, we can write the effective dielec- ) / 1 0.06 k tric constant independent of the reference phase, i.e., ( e εe(ω) = εe(k1(ω),ω) = εe(k2(ω),ω). Truncating Eq. ε 0.03 (D1) at the two-point level and solving it in εe(ω) yields Im[ the following approximation: 0 0 0.2 0.4 0.6 0.8 1 k1a ε1 + ε2 1 εe(ω)= + d(ε1φ2 + ε2φ1) (D2) 2 2 2(ω) − A FIG. 13. Evaluation of the effective dielectric constant εe + 4 (ω)[d (ω)]ε ε as a function of the dimensionless wave number k1a for 3D A2 − A2 1 2 2 1/2 equilibrium packings and 3D Debye random media of volume + [(ε1 + ε2) 2(ω) d(φ2ε1 + φ1ε2)] , fraction φ2 = 0.25 and contrast ratio ε2/ε1 = 4. Here k1 A − ≡ √ε1ω/c is the wave number in the matrix (reference) phase 1, and a is particle radius. (1) (2) where 2(ω) d 1+ A2 (k2) /φ1 + A2 (k1) /φ2, and we chooseA the≡ physically− meaningful solution from the quadratic equation. Note that in the static limit, 2(ω) ACKNOWLEDGMENTS converges to d 1, and thus Eq. (D2) reduces toA the − well-known self-consistent formula [11]. We thank Z. Ma, M. Klatt, Y. Chen, and L. Dal Ne- Examples of phase-inversion symmetric media include gro for very helpful discussions. The authors gratefully the d-dimensional random checkerboard [11] and what has acknowledge the support of Air Force Office of Scien- been called Debye random media [29, 30, 75]. Here we tific Research Program on Mechanics of Multifunctional apply the approximation (D2) to Debye random media, Materials and Microsystems under Grant No. FA9550- which are defined by their autocovariance function [29, 18-1-0514. ∗ [email protected]; http://torquato.princeton.edu [4] D.A.G. Bruggeman, “Berechnung verschiedener [1] J. C. Maxwell, Treatise on Electricity and Magnetism Physikalischer Konstanten von heterogenen Sub- (Clarendon Press, Oxford, 1873). stanzen,” Ann. Phys. (Berlin) 416, 636–679 (1935). [2] Lord Rayleigh, “On the influence of obstacles arranged [5] H. C. Brinkman, “A calculation of the viscous force ex- in a rectangular order upon the properties of medium,” erted by a flowing fluid on a dense swarm of particles,” Philso. Mag. 34, 481–502 (1892). Appl. Sci. Res. 1, 27–34 (1949). [3] A. Einstein, “Eine neue Bestimmung der [6] B. Budiansky, “On the elastic moduli of some hetero- Molek¨uldimensionen,” Ann. Phys. (Berlin) 324, geneous materials,” J. Mech. Phys. Solids 13, 223–227 289–306 (1906). (1965). 24 [7] S. Prager, “Viscous flow through porous media,” Phys. [29] C. L. Y. Yeong and S. Torquato, “Reconstructing ran- Fluids 4, 1477–1482 (1961). dom media,” Phys. Rev. E 57, 495–506 (1998). [8] Z. Hashin and S. Shtrikman, “A variational approach to [30] Z. Ma and S. Torquato, “Generation and struc- the theory of the effective magnetic permeability of mul- tural characterization of Debye random media,” tiphase materials,” J. Appl. Phys. 33, 3125–3131 (1962). Phys. Rev. E 102, 043310 (2020). [9] M. Beran, “Use of the variational approach to determine [31] M. C. Rechtsman and S. Torquato, “Effective dielectric bounds for the effective permittivity in random media,” tensor for electromagnetic wave propagation in random Nuovo Cimento 38, 771–782 (1965). media,” J. Appl. Phys. 103, 084901 (2008). [10] R. V. Kohn and R. Lipton, “Optimal bounds for the [32] V. M. Agranovich and V. Ginzburg, Crystal Optics with effective energy of a mixture of isotropic, incompressible Spatial Dispersion, and Excitons, 2nd ed. (Springer, elastic materials,” Arch. Ration. Mech. Anal. 102, 331– Berlin, Heidelberg, 1984). 350 (1988). [33] L. Tsang, Scattering of Electromagnetic Waves, edited [11] S. Torquato, Random Heterogeneous Materials: Mi- by J. A. Kong, Wiley Series in Remote Sensing (Wiley, crostructure and Macroscopic Properties (Springer- Chichester, UK, 2001). Verlag, New York, 2002). [34] A. Caz´e and J. C. Schotland, “Diagram- [12] G. W. Milton, The Theory of Composites (Cambridge matic and asymptotic approaches to the ori- University Press, Cambridge, England, 2002). gins of radiative transport theory: Tutorial,” [13] W. F. Brown, “Solid mixture permittivities,” J. Chem. J. Opt. Soc. Am. A 32, 1475 (2015). Phys. 23, 1514–1517 (1955). [35] S. Torquato, “Disordered hyperuniform heterogeneous [14] B. U. Felderhof, G. W. Ford, and E. G. D. Cohen, materials,” J. Phys.: Cond. Mat 28, 414012 (2016). “Cluster expansion for the dielectric constant of a po- [36] D. Chen and S. Torquato, “Designing disordered hype- larizable suspension,” J. Stat. Phys. 28, 135–164 (1982). runiform two-phase materials with novel physical prop- [15] A. K. Sen and S. Torquato, “Effective conductivity of erties,” Acta Materialia 142, 152–161 (2018). anisotropic two-phase composite media,” Phys. Rev. B [37] C. L´opez, “The true value of disorder,” Adv. Opt. 39, 4504–4515 (1989). Mater. 6, 1800439 (2018). [16] S. Torquato, “Exact expression for the effective elastic [38] S. Yu, X. Piao, J. Hong, and N. Park, “Bloch-like wave tensor of disordered composites,” Phys. Rev. Lett. 79, dynamics in disordered potentials based on supersym- 681–684 (1997). metry,” Nat. Mater. 6, 8269 (2015). [17] P. Sheng, Introduction to Wave Scattering, Localiza- [39] M. Florescu, P. J. Steinhardt, and S. Torquato, “Opti- tion and Mesoscopic Phenomena (Academic Press, New cal cavities and waveguides in hyperuniform disordered York, 1995). photonic solids,” Phys. Rev. B 87, 165116 (2013). [18] A. Sihvola, Electromagnetic Mixing Formulas and Applications[40] W. Man, M. Florescu, E. P. Williamson, Y. He, S. R. (IET Digital Library, London, 1999). Hashemizad, B. Y. C. Leung, D. R. Liner, S. Torquato, [19] J. B. Keller, “Stochastic equations and P. M. Chaikin, and P. J. Steinhardt, “Isotropic band wave propagation in random media,” gaps and freeform waveguides observed in hyperuniform Proc. Symp. Appl. Math. 16, 145 (1964). disordered photonic solids,” Proc. Nat. Acad. Sci. 110, [20] U. Frisch, “Wave propagation in random media,” in 15886–15891 (2013). Probabilistic Methods in Applied Mathematics, Vol. 1, [41] S. Torquato and F. H. Stillinger, “Local density fluctua- edited by A. T Bharucha-Reid (Academic Press, New tions, hyperuniform systems, and order metrics,” Phys. York, 1968) 1st ed., pp. 75–198. Rev. E 68, 041113 (2003). [21] L. Tsang and J. A. Kong, “Scattering of electromagnetic [42] C. E. Zachary and S. Torquato, “Hyperuniformity in waves from random media with strong permittivity fluc- point patterns and two-phase heterogeneous media,” J. tuations,” Radio Sci. 16, 303–320 (1981). Stat. Mech.: Theory & Exp. 2009, P12015 (2009). [22] X. Jing, P. Sheng, and M. Zhou, “Acoustic and elec- [43] S. Torquato, “Hyperuniform states of matter,” Phys. tromagnetic quasimodes in dispersed random media,” Rep. 745, 1–95 (2018). Phys. Rev. A 46, 6513–6534 (1992). [44] S. Torquato, G. Zhang, and F. H. Stillinger, “Ensem- [23] A notable exception is the generalized coherent poten- ble theory for stealthy hyperuniform disordered ground tial approximation for particle suspensions presented in states,” Phys. Rev. X 5, 021020 (2015). Ref. [22]. However, this is not a closed-form formula for [45] G. Zhang, F. H. Stillinger, and S. Torquato, “The per- the effective dielectric constant and requires, as input, fect glass paradigm: Disordered hyperuniform glasses certain numerical simulations of the electric fields. down to absolute zero,” Sci. Rep. 6, 36963 (2016). [24] J. C. M. Garnett, “Colours in [46] D. Hexner, P. M. Chaikin, and D. Levine, “Enhanced metal glasses and in metallic films,” hyperuniformity from random reorganization,” Proc. Phil. Trans. R. Soc. A 203, 385–420 (1904). Natl. Acad. Sci. U.S.A. 114, 4294—-4299 (2017). [25] R. Ruppin, “Evaluation of extended Maxwell-Garnett [47] J. Ricouvier, R. Pierrat, R. Carminati, P. Tabeling, theories,” Opt. Commun. 182, 273–279 (2000). and P. Yazhgur, “Optimizing hyperuniformity in self- [26] M. Lax, “Multiple scattering of waves. II. The effective assembled bidisperse emulsions,” Phys. Rev. Lett. 119, field in dense systems,” Phys. Rev. 85, 621–629 (1952). 208001 (2017). [27] C. O. Ao and J. A. Kong, “Analytical approx- [48] E. C. O˘guz, J. E. S. Socolar, P. J. Steinhardt, and imations in multiple scattering of electromag- S. Torquato, “Hyperuniformity of quasicrystals,” Phys. netic waves by aligned dielectric spheroids,” Rev. B 95, 054119 (2017). J. Opt. Soc. Am. A 19, 1145–1156 (2002). [49] S. Yu, X. Piao, and N. Park, “Disordered potential [28] Z. Ma and S. Torquato, “Random scalar fields and hy- landscapes for anomalous delocalization and superdif- peruniformity,” J. Appl. Phys. 121, 244904 (2017). fusion of light,” ACS Photonics 5, 1499–1505 (2018). 25 [50] J. Wang, J. M. Schwarz, and J. D. Paulsen, “Hyper- [67] O. U. Uche, F. H. Stillinger, and S. Torquato, “Con- uniformity with no fine tuning in sheared sedimenting straints on collective density variables: Two dimen- suspensions,” Nature Comm. 9, 1–7 (2018). sions,” Phys. Rev. E 70, 046122 (2004). [51] Q-L. Lei and R. Ni, “Hydrodynamics of random- [68] R. D. Batten, F. H. Stillinger, and S. Torquato, “Clas- organizing hyperuniform fluids,” Proc. Nat. Acad. Sci. sical disordered ground states: Super-ideal gases, and 116, 22983–22989 (2019). stealth and equi-luminous materials,” J. Appl. Phys. [52] S. Gorsky, W. A. Britton, Y. Chen, J. Montaner, 104, 033504 (2008). A. Lenef, M. Raukas, and L. Dal Negro, “Engineered [69] G. Zhang, F. Stillinger, and S. Torquato, “Ground hyperuniformity for directional light extraction,” APL states of stealthy hyperuniform potentials: I. Entrop- Photonics 4, 110801 (2019). ically favored configurations,” Phys. Rev. E 92, 022119 [53] M. A. Klatt, J. Kim, and S. Torquato, “Cloaking the (2015). underlying long-range order of randomly perturbed lat- [70] G. Zhang, F. H. Stillinger, and S. Torquato, “Trans- tices,” Phys. Rev. E 101, 032118 (2020). port, geometrical and topological properties of stealthy [54] T. Ma, H. Guerboukha, M. Girard, A. D. Squires, R. A. disordered hyperuniform two-phase systems,” J. Chem. Lewis, and M. Skorobogatiy, “3D printed hollow-core Phys 145, 244109 (2016). terahertz optical waveguides with hyperuniform dis- [71] R. Degl’Innocenti, Y. D. Shah, L. Masini, A. Ronzani, ordered dielectric reflectors,” Adv. Optical Mater. 4, A. Pitanti, Y. Ren, D. S. Jessop, A. Tredicucci, H. E. 2085–2094 (2016). Beere, and D. A. Ritchie, “Hyperuniform disordered [55] Y. Xu, S. Chen, P.-E. Chen, W. Xu, and Y. Jiao, terahertz quantum cascade laser,” Sci. Rep. 6, 19325 “Microstructure and mechanical properties of hyperuni- (2016). form heterogeneous materials,” Phys. Rev. E 96, 043301 [72] G. Gkantzounis, T. Amoah, and M. Florescu, “Hyper- (2017). uniform disordered phononic structures,” Phys. Rev. B [56] S. Torquato and D. Chen, “Multifunctional hyperuni- 95, 094120 (2017). form cellular networks: optimality, anisotropy and dis- [73] J. Kim and S. Torquato, “Multifunctional compos- order,” Multifunctional Materials 1, 015001 (2018). ites for elastic and electromagnetic wave propagation,” [57] J. Kim and S. Torquato, “New tessellation-based pro- Proc. Nat. Acad. Sci. 117, 8764–8774 (2020). cedure to design perfectly hyperuniform disordered dis- [74] S. Torquato and G. Stell, “Microstructure of two-phase persions for materials discovery,” Acta Materialia 168, random media: I. The n-point probability functions,” 143–151 (2019). J. Chem. Phys. 77, 2071–2077 (1982). [58] S. Ghosh and J. L. Lebowitz, “Generalized stealthy [75] P. Debye, H. R. Anderson, and H. Brumberger, “Scat- hyperuniform processes: Maximal rigidity and the tering by an inhomogeneous solid. II. The correlation bounded holes conjecture,” Comm. Math. Phys. 363, function and its applications,” J. Appl. Phys. 28, 679– 97–110 (2018). 683 (1957). [59] J. S. Brauchart, P. J. Grabner, and W. Kusner, “Hy- [76] S. Torquato, “Predicting transport characteristics of hy- peruniform point sets on the sphere: deterministic as- peruniform porous media via rigorous microstructure- pects,” Constructive approximation 50, 45–61 (2019). property relations,” Adv. Water Resour. 140, 103565 [60] S. Torquato, G. Zhang, and M. de Courcy-Ireland, (2020). “Hidden multiscale order in the primes,” J. Phys. A: [77] S. Torquato, “Perspective: Basic understanding of con- Math. & Theoretical 52, 135002 (2019). densed phases of matter via packing models,” J. Chem. [61] Y. Jiao, T. Lau, H. Hatzikirou, M. Meyer-Hermann, Phys. 149, 020901 (2018). J. C. Corbo, and S. Torquato, “Avian photoreceptor [78] B. L. Lu and S. Torquato, “Local volume fraction fluc- patterns represent a disordered hyperuniform solution tuations in heterogeneous media,” J. Chem. Phys. 93, to a multiscale packing problem,” Phys. Rev. E 89, 3452–3459 (1990). 022721 (2014). [79] J. Quintanilla and S. Torquato, “Local volume fraction [62] M. Florescu, S. Torquato, and P. J. Steinhardt, “De- fluctuations in random media,” J. Chem. Phys. 106, signer disordered materials with large complete pho- 2741–2751 (1997). tonic band gaps,” Proc. Nat. Acad. Sci. 106, 20658– [80] C. E. Zachary and S. Torquato, “Anomalous local co- 20663 (2009). ordination, density fluctuations, and void statistics in [63] O. Leseur, R. Pierrat, and R. Carminati, “High-density disordered hyperuniform many-particle ground states,” hyperuniform materials can be transparent,” Optica 3, Phys. Rev. E 83, 051133 (2011). 763–767 (2016). [81] See Supplementary Material at the arXiv article for [64] L. S. Froufe-P´erez, M. Engel, J. Jos´e S´aenz, and formulas of the local strong-contrast expansions; de- F. Scheffold, “Transport Phase Diagram and Anderson tailed derivations of nonlocal strong-contrast expan- Localization in Hyperuniform Disordered Photonic Ma- sions; properties of the nonlocal attenuation function terials,” Proc. Nat. Acad. Sci. 114, 9570—-9574 (2017). F (Q); comparisons of the local and nonlocal attenua- [65] M. A. Klatt and S. Torquato, “Characterization of max- tion functions; derivations of the Kramers-Kronig re- imally random jammed sphere packings. III. Transport lations; comparisons of our approximations with other and electromagnetic properties via correlation func- popular effective-medium approximations; simulation tions,” Phys. Rev. E 97, 012118 (2018). details; results for 2D systems; and the comparison of [66] H. Zhang, W. Wu, and Y. Hao, “Luneburg lens from our formalism with multiple-scattering theory. hyperuniform disordered composite materials,” in 2018 [82] William T. Doyle, “Optical properties of a suspension IEEE International Symposium on Antennas and Prop- of metal spheres,” Phys. Rev. B 39, 9852–9858 (1989). agation & USNC/URSI National Radio Science Meeting [83] M. Silveirinha and N. Engheta, “Design of (2018) pp. 2281–2282. matched zero-index metamaterials using non- 26 magnetic inclusions in epsilon-near-zero media,” Mech. Eng. 157, 69–94 (1998). Phys. Rev. B 75, 075119 (2007). [96] D. J. Eyre and G. W. Milton, “A fast numerical scheme [84] The Green’s function in this work (24) differs from the for computing the response of composites using grid re- one given in Ref. [31] by a multiplicative factor (ω/c)2. finement,” Euro. Phys. J. 6, 41–47 (1999). Thus, Eq. (24) converges to its static counterpart in the [97] A. Taflove, S. G. Johnson, and A. Oskooi, Advances in static limit (i.e., ω 0). FDTD Computational Electrodynamics: Photonics and [85] Z. Hashin and S.→ Shtrikman, “A variational ap- Nanotechnology (Artech House, Boston, 2013). proach to the elastic behavior of multiphase materials,” [98] A. F. Oskooi, D. Roundy, M. Ibanescu, P. Bermel, J. Mech. Phys. Solids 4, 286–295 (1963). J. D. Joannopoulos, and S. G. Johnson, “Meep: [86] J. Kim and S. Torquato, “Methodology to construct A flexible free-software package for electro- large realizations of perfectly hyperuniform disordered magnetic simulations by the FDTD method,” packings,” Phys. Rev. E 99, 052141 (2019). Comput. Phys. Commun. 181, 687–702 (2010). [87] J. R. Willis, “Variational and related methods for the [99] D. C. Liu and J. Nocedal, “On the limited memory overall properties of composites,” Adv. Appl. Mech. 21, BFGS method for large scale optimization,” Math. Pro- 1–78 (1981). gramming 45, 503–528 (1989). [88] G. W. Milton, “Bounds on the complex permittivity of [100] S. G. Johnson and J. D. Joannopoulos, “Block-iterative a two-component composite material,” J. Appl. Phys. frequency-domain methods for Maxwell’s equations in a 52, 5286–5293 (1981). planewave basis,” Opt. Express 8, 173–190 (2001). [89] S. Torquato, “Effective electrical conductivity of two- [101] Here, the large value of the imaginary part Im[εe] is due phase disordered composite media,” J. Appl. Phys. 58, to a small penetration depth of evanescent waves. 3790–3797 (1985). [102] The Kramers-Kronig relations (B1) and (B2) dictate [90] R. Carminati and J. J. S´aenz, “Density of States and that a resonance phenomenon in the dielectric response, Extinction Mean Free Path of Waves in Random Media: that is, a sharp peak in Im[εe], must correspond to a Dispersion Relations and Sum Rules,” Phys, Rev. Lett. sharp transition in Re[εe], and vice versa. 102, 093902 (2009). [103] S. Linden, C. Enkrich, M. Wegener, J. Zhou, [91] J. Quintanilla, S. Torquato, and R. M. Ziff, “Efficient T. Koschny, and C. M. Soukoulis, “Magnetic measurement of the percolation threshold for fully pen- Response of Metamaterials at 100 Terahertz,” etrable discs,” J. Phys. A: Math. & Gen. 33, L399–L407 Science 306, 1351–1353 (2004). (2000). [104] D. R. Smith, J. B. Pendry, and M. C. K. Wilt- [92] M. D. Rintoul and S. Torquato, “Precise determina- shire, “Metamaterials and Negative Refractive Index,” tion of the critical threshold and exponents in a three- Science 305, 788–792 (2004). dimensional continuum percolation model,” J. Phys. A: [105] Y. Wang, T. Sun, T. Paudel, Y. Zhang, Z. Ren, and Math. Gen. 30, L585–L592 (1997). K. Kempa, “Metamaterial-Plasmonic Absorber Struc- [93] J. P. Hansen and I. R. McDonald, Theory of Simple ture for High Efficiency Amorphous Silicon Solar Cells,” Liquids, 4th ed. (Academic Press, New York, 2013). Nano Lett. 12, 440–445 (2012). [94] G. Zhang, F. H. Stillinger, and S. Torquato, “Can ex- [106] J. D. Jackson, Classical Electrodynamics (Wiley, New otic disordered “stealthy” particle configurations toler- York, 1990). ate arbitrarily large holes?” Soft Matter 13, 6197–6207 [107] G. W. Milton, D. J. Eyre, and J. V. Mantese, “Finite (2017). Frequency Range Kramers-Kronig Relations: Bounds [95] H. Moulinec and P. Suquet, “A numerical method for on the Dispersion,” Phys. Rev. Lett. 79, 3062–3065 computing the overall response of nonlinear composites (1997). with complex microstructure,” Comput. Meth. Appl.