Properties from architecture
Everything earlier in this course traced properties to chemistry and microstructure: bonds set stiffness, dislocations set strength, bands set colour. A metamaterial turns that around. Its properties come mainly from the geometry of a designed unit cell, repeated on a scale much smaller than the wavelength or the specimen, so that the same polymer can be stiff or soft, can fatten when stretched, or can block sound at one frequency and pass it at another, depending on how it is arranged. Natural materials discovered this first: bone, wood, honeycomb in a bird's skull and the porous core of a plant stem are all architected for stiffness at low weight.
The word arrived with optics. In 1968 Victor Veselago asked what a material with both negative permittivity and negative permeability would do; around 2000 John Pendry at Imperial College London and David Smith's group, then at the University of California, San Diego, showed how to build one for microwaves from wires and split rings, and the field took off. In parallel, mechanics had its own story: Roderic Lakes made foams that expand sideways when stretched in 1987, and additive manufacturing, covered in Lumenary ME's additive manufacturing, made it possible from the 2010s to print almost any lattice. This lesson takes the mechanical side in depth and the acoustic and optical sides in outline, then asks what has left the laboratory.
Metamaterials, architected materials and effective media
Definition— Architected material and metamaterial
An architected material has a designed internal geometry — a lattice, honeycomb, foam or fibre arrangement — at a scale between the microstructure and the component. A metamaterial is an architected material whose effective properties are unusual or absent in its constituents, such as a negative Poisson's ratio, negative effective mass density or modulus near a resonance, or a negative refractive index. The effective properties are those of the homogeneous medium that reproduces the architected material's behaviour at scales much larger than its cell, obtained by homogenisation: averaging stress and strain (or fields) over a representative cell.
For an isotropic linear-elastic solid, positive-definite strain energy requires . Ordinary materials sit between about 0.2 and 0.5 (rubber near 0.5, steel about 0.3, cork near 0). Materials with are called auxetic, a word coined by Ken Evans in 1991. Anisotropic architected materials are not bound by the isotropic limits: a honeycomb can have in-plane Poisson's ratios well above 1, as the next section shows. Multiscale modelling covers homogenisation in general; here the cell is simple enough to analyse by hand.
The re-entrant honeycomb
A honeycomb cell has inclined struts of length at an angle to the horizontal and vertical struts of length . Regular hexagons have and . If is negative, the inclined struts point inwards and the cell becomes a bow tie: a re-entrant honeycomb.
Theorem— Poisson's ratio of a honeycomb
When the inclined struts deform by bending or by rotating about their joints, a honeycomb loaded along direction 1 (the horizontal) has in-plane Poisson's ratio
which is negative for re-entrant cells (, with ), and .
Proof
Take the hinging picture, which gives the same ratio as bending because in both the vertical struts stay straight and the inclined struts change their angle. The cell's repeat lengths are horizontally and vertically. A small rotation gives strains and . Hence . Loading along 2 runs the same mechanism in reverse, so . If , while the other factors are positive, so .
∎
The program evaluates the Gibson–Ashby beam-bending formulas for , the two in-plane moduli and the relative density, and checks the Poisson's ratio against a finite-difference calculation of the hinge kinematics.
import numpy as np
# A honeycomb cell: inclined struts of length l at angle theta to the horizontal (theta < 0: re-entrant),
# vertical struts of length h, all of thickness t, out-of-plane depth b. Gibson and Ashby's bending model.
def honeycomb(theta_deg, h_over_l, t_over_l):
th = np.radians(theta_deg)
s, c = np.sin(th), np.cos(th)
nu12 = c**2 / ((h_over_l + s) * s) # pull along x (1), measure contraction along y (2)
E1 = t_over_l**3 * c / ((h_over_l + s) * s**2) # E1 / E_solid
E2 = t_over_l**3 * (h_over_l + s) / c**3 # E2 / E_solid
rho = t_over_l * (h_over_l + 2) / (2 * c * (h_over_l + s)) # relative density
return nu12, E1, E2, rho
# An independent check: the cell's repeat lengths are X = 2 l cos(theta) and Y = 2 (h + l sin(theta)).
# Rotating the struts slightly (as a hinged mechanism would) changes both; nu12 = -(dY/Y) / (dX/X).
def nu_from_kinematics(theta_deg, h_over_l, d=1e-6):
th = np.radians(theta_deg)
X = lambda a: 2 * np.cos(a)
Y = lambda a: 2 * (h_over_l + np.sin(a))
eps1 = (X(th + d) - X(th - d)) / X(th)
eps2 = (Y(th + d) - Y(th - d)) / Y(th)
return -eps2 / eps1
print("h/l = 2, t/l = 0.1")
print(" theta nu12 (beam model) nu12 (kinematics) E1/Es E2/Es rel. density")
for theta in (-45, -30, -20, -10, 10, 20, 30, 45):
nu, E1, E2, rho = honeycomb(theta, 2.0, 0.1)
print(f" {theta:4d} {nu:+8.3f} {nu_from_kinematics(theta, 2.0):+8.3f} {E1:.2e} {E2:.2e} {rho:.3f}")
nu, E1, E2, rho = honeycomb(30, 1.0, 0.1)
print(f"\nregular hexagons (theta = 30, h/l = 1): nu12 = {nu:.3f}, E1/Es = {E1:.2e}, relative density {rho:.3f}")The regular honeycomb has exactly, already outside the isotropic range. With , the re-entrant cell at has , and at the ratio reaches ; the conventional cell at reaches . The two independent calculations agree to the printed digits. The moduli are tiny — about of the solid's at a relative density of 0.1 to 0.2 — because the struts bend, and bending stiffness scales as .
What auxetic behaviour buys: an auxetic sheet bent out of plane curves into a dome rather than a saddle, so it conforms to doubly curved surfaces; under a local impact, material flows towards the indentation, raising indentation resistance; and an auxetic tube expands when stretched, useful for stents and for fasteners that grip harder when pulled.
Lattice materials and additive manufacturing
Lattices — periodic networks of struts or shells — are the most practical architected materials, because additive manufacturing can build them in metals (laser powder-bed fusion), polymers (vat photopolymerisation, including continuous "digital light synthesis") and ceramics. Their behaviour depends first on whether the struts bend or stretch.
Gibson and Ashby's analysis of foams gives, for bending-dominated structures, a relative modulus and relative strength , because a strut's bending stiffness grows as its thickness to the fourth power while its mass grows as the square. In a stretch-dominated lattice, where loads are carried axially by triangulated members, both scale linearly, and at best, which is far stiffer at low density. Which one a lattice is follows from Maxwell's counting rule for pin-jointed frames: with struts and joints, in two dimensions and in three. If the frame is a mechanism, and only the joints' stiffness and the struts' bending resist load; Deshpande, Ashby and Fleck showed in 2001 that a periodic lattice needs a nodal connectivity of 6 in two dimensions and 12 in three to be stretch-dominated, which the octet truss achieves.
The block applies both laws to additively manufactured Ti-6Al-4V (rounded typical solid properties), asks what relative density gives a stiffness of 3 GPa — within the range of trabecular bone, the target for orthopaedic implants that should not shield bone from load — and applies Maxwell's rule to four cells.
# Gibson-Ashby scaling for cellular solids and lattices, with the usual prefactors:
# bending-dominated (open-cell foams, low-connectivity lattices): E/Es = (rho/rhos)^2, sigma/sigma_ys = 0.3 (rho/rhos)^1.5
# stretch-dominated (fully triangulated lattices, e.g. the octet truss): E/Es = (rho/rhos)/3, sigma/sigma_ys = (rho/rhos)/3
# Solid: additively manufactured Ti-6Al-4V, rounded typical values.
Es, sys, rhos = 110e3, 900.0, 4430.0 # MPa, MPa, kg/m^3
print("rel. density density bending: E (GPa) strength (MPa) stretching: E (GPa) strength (MPa)")
for r in (0.02, 0.05, 0.10, 0.20, 0.30):
Eb, sb = Es * r**2, sys * 0.3 * r**1.5
Est, sst = Es * r / 3, sys * r / 3
print(f" {r:4.2f} {rhos * r:5.0f} kg/m3 {Eb / 1000:8.3f} {sb:7.1f} {Est / 1000:7.2f} {sst:7.1f}")
# Design question: which relative density gives a stiffness of 3 GPa (in the range of trabecular bone)?
target = 3e3
r_bend = (target / Es) ** 0.5
r_stretch = 3 * target / Es
print(f"\nfor E = 3 GPa: bending-dominated needs rel. density {r_bend:.3f}, stretch-dominated {r_stretch:.3f}")
print(f" -> the stretch-dominated lattice is {r_bend / r_stretch:.1f} times lighter for the same stiffness")
# Maxwell's counting rule for a pin-jointed frame of b struts and j joints: M = b - 2j + 3 in 2D,
# M = b - 3j + 6 in 3D. M < 0: a mechanism, so loads bend the struts; M >= 0: struts can carry load by stretching.
frames = {"square (2D)": (4, 4, 2), "triangle (2D)": (3, 3, 2), "cube (3D)": (12, 8, 3), "octahedron (3D)": (12, 6, 3)}
print()
for name, (b, j, dim) in frames.items():
M = b - 2 * j + 3 if dim == 2 else b - 3 * j + 6
print(f"{name:16s} b = {b:2d}, j = {j}: M = {M:d} -> {'stretch-dominated possible' if M >= 0 else 'mechanism: bending-dominated'}")At 10 per cent relative density (443 kg/m³, lighter than many woods), the ideal stretch-dominated titanium lattice is 3.67 GPa stiff and 30 MPa strong, against 1.10 GPa and 8.5 MPa for a bending-dominated one. For a 3 GPa implant, the stretch-dominated design needs a relative density of 0.082 and the bending-dominated one 0.165, twice the mass — though for bone ingrowth, pore size and permeability matter as much as stiffness, and implant makers often accept bending-dominated cells for that reason. Maxwell's rule confirms the intuition: a square or cube is a mechanism ( and ), a triangle or octahedron is rigid ().
The record-breaking results of the 2010s were at small scales: hollow nickel microlattices of 2011 lighter than 1 mg/cm³; stretch-dominated ceramic and polymer microlattices in 2014 that kept nearly linear stiffness scaling down to very low density; and hollow alumina nanolattices made by two-photon lithography, also in 2014, whose walls were thin enough (tens of nanometres) to bend and recover rather than fracture — a size effect, since flaws shrink with the walls, as Fracture and toughness predicts. Since about 2020, inverse design with machine learning (for example, deep-learning models that output a truss topology for a requested anisotropic stiffness tensor) and topology optimisation have turned lattice design into a computational problem, linked to Lumenary ME's simulation-driven design.
Beyond stiffness and strength, mechanical metamaterials include pentamode materials, easy to shear but hard to compress, which behave like solid "metafluids"; multistable lattices of snapping elements that absorb energy and can be reused; chiral lattices that twist when compressed (2017); kirigami sheets whose cuts make them stretch and morph; and topological mechanical lattices with floppy modes localised on one edge. Most of these are research topics.
Acoustic and elastic metamaterials, in outline
Waves in a periodic medium have band gaps, frequency ranges in which they cannot propagate. In a phononic crystal the gap comes from Bragg scattering and opens near wavelengths comparable to the period, so blocking audible sound with a phononic crystal needs cells tens of centimetres across. In 2000 Ping Sheng's group at the Hong Kong University of Science and Technology showed another mechanism: lead spheres coated in soft silicone rubber, embedded in epoxy, act as local resonators, and near their resonance the composite behaves as if it had a negative effective mass, opening a gap at wavelengths far larger than the cell. The simplest model is a chain of masses joined by springs , each carrying an inner mass on a soft spring . The block computes its band structure and the transmission through 20 cells.
import numpy as np
# A one-dimensional "mass-in-mass" chain: outer masses m1 joined by springs k1 (spacing a), each carrying
# an inner mass m2 on a spring k2 -- the simplest locally resonant acoustic metamaterial.
m1, m2, k1, k2, a = 1.0, 0.5, 1.0, 0.05, 1.0 # arbitrary consistent units
w0 = np.sqrt(k2 / m2) # the inner resonator's own frequency
# Dispersion: for each wavenumber q, solve the 2x2 generalised eigenproblem K(q) u = w^2 M u.
qs = np.linspace(1e-4, np.pi / a, 400)
branches = []
for q in qs:
K = np.array([[2 * k1 * (1 - np.cos(q * a)) + k2, -k2], [-k2, k2]])
Minv_sqrt = np.diag([m1 ** -0.5, m2 ** -0.5])
w2 = np.linalg.eigvalsh(Minv_sqrt @ K @ Minv_sqrt)
branches.append(np.sqrt(w2))
branches = np.array(branches)
gap_lo, gap_hi = branches[:, 0].max(), branches[:, 1].min()
print(f"resonator frequency w0 = {w0:.4f}")
print(f"band gap from {gap_lo:.4f} to {gap_hi:.4f} (expected near w0 to w0*sqrt(1 + m2/m1) = {w0 * np.sqrt(1 + m2 / m1):.4f})")
c_long = a * np.sqrt(k1 / (m1 + m2)) # long-wavelength sound speed of the chain
print(f"wavelength of sound in the chain at w0: {2 * np.pi * c_long / w0:.1f} cells")
# Transmission through a finite chain of N cells driven at one end (light damping), for several frequencies.
def transmission(w, N=20, eta=0.01):
n = 2 * N
Kmat = np.zeros((n, n), dtype=complex)
Mvec = np.zeros(n)
for i in range(N):
o, r = 2 * i, 2 * i + 1 # outer and inner mass of cell i
Mvec[o], Mvec[r] = m1, m2
kk2 = k2 * (1 + 1j * eta)
Kmat[o, o] += kk2; Kmat[r, r] += kk2; Kmat[o, r] -= kk2; Kmat[r, o] -= kk2
if i < N - 1:
kk1 = k1 * (1 + 1j * eta)
p = 2 * (i + 1)
Kmat[o, o] += kk1; Kmat[p, p] += kk1; Kmat[o, p] -= kk1; Kmat[p, o] -= kk1
A = Kmat - w**2 * np.diag(Mvec)
F = np.zeros(n, dtype=complex); F[0] = 1.0 # unit force on the first outer mass
u = np.linalg.solve(A, F)
return abs(u[2 * (N - 1)]) / abs(u[0])
print("\nfrequency |u_out / u_in| through 20 cells")
for w in (0.10, 0.20, 0.28, 0.32, 0.36, 0.40, 0.60):
T = transmission(w)
print(f" {w:5.2f} {T:9.2e} " + ("in the gap" if gap_lo < w < gap_hi else ""))The gap runs from 0.3142 to 0.3873, just below the resonator frequency 0.3162 up to , and at the resonance a wave in the chain is 16.2 cells long: the gap is sub-wavelength, which a Bragg gap cannot be. Through 20 cells, the transmitted amplitude falls to at 0.32 and at 0.36, while outside the gap it is of order 1. The price is bandwidth: a local-resonance gap is narrow, set by the mass ratio, and damping smears it.
In practice, acoustic metamaterials appear in noise-control panels and ducts tuned to particular tones, in vibration isolation for machinery and vehicles, and as acoustic Bragg-reflector stacks in the solidly mounted bulk-acoustic-wave filters used in many phones, phononic structures engineered at the micrometre scale. Proposals to shield buildings from earthquakes with metre-scale resonator arrays have been explored in field tests, but remain far from practice.
Electromagnetic and optical metamaterials, in outline
Electromagnetic metamaterials use sub-wavelength metal or dielectric elements to engineer the effective permittivity and permeability . A lattice of thin wires behaves like a very dilute metal with a low plasma frequency, below which is negative; split-ring resonators, small broken loops, give a magnetic resonance with negative just above it. Combine both in the same band and the refractive index is negative, as Smith's group showed at microwave frequencies in 2000–2001. The block computes the wire medium's plasma frequency from Pendry's 1996 formula, then the phase profile a flat metalens must impose, a modern application of the same idea at optical wavelengths.
import numpy as np
c0 = 2.998e8
# 1. A lattice of thin metal wires behaves like a dilute metal with a low plasma frequency
# (Pendry and colleagues, 1996): omega_p^2 = 2 pi c0^2 / (a^2 ln(a / r)), neglecting losses.
for a, r in ((5e-3, 1e-6), (5e-3, 10e-6), (1e-3, 1e-6)):
wp = np.sqrt(2 * np.pi * c0**2 / (a**2 * np.log(a / r)))
print(f"wires of radius {r * 1e6:3.0f} um, {a * 1e3:2.0f} mm apart: plasma frequency {wp / (2 * np.pi) / 1e9:5.1f} GHz")
a, r = 5e-3, 1e-6
wp = np.sqrt(2 * np.pi * c0**2 / (a**2 * np.log(a / r)))
for f in (4e9, 8e9, 12e9):
eps = 1 - wp**2 / (2 * np.pi * f) ** 2
print(f" at {f / 1e9:4.1f} GHz the effective permittivity is {eps:+6.2f}")
# 2. A metalens: to focus a plane wave of wavelength lam at distance f, the surface must add the phase
# phi(r) = (2 pi / lam) (f - sqrt(r^2 + f^2)), wrapped into [0, 2 pi) by nanopillars of different sizes.
lam, focal, radius = 940e-9, 1e-3, 0.5e-3 # near-infrared, 1 mm focal length, 1 mm aperture
r = np.linspace(0, radius, 6)
phi = 2 * np.pi / lam * (focal - np.sqrt(r**2 + focal**2))
print("\nmetalens, 940 nm light, f = 1 mm:")
for ri, p in zip(r, phi):
print(f" r = {ri * 1e6:5.0f} um: required phase {p:9.1f} rad -> wrapped {p % (2 * np.pi):4.2f} rad")
zones = (np.sqrt(radius**2 + focal**2) - focal) / lam
pillar = 400e-9
print(f" {zones:.0f} full 2-pi zones across the radius; about {np.pi * radius**2 / pillar**2:,.0f} pillars on a 400 nm grid")
print(f" numerical aperture {radius / np.sqrt(radius**2 + focal**2):.2f}")Wires of 1 µm radius spaced 5 mm apart give a plasma frequency of 8.2 GHz, so the medium's permittivity is −3.20 at 4 GHz and positive above 8.2 GHz; a natural metal's plasma frequency is in the ultraviolet. The metalens needs a phase that grows to −789 radians at its edge, which a metasurface wraps into with 126 concentric zones of pillars, about 4.9 million of them on a 400 nm grid — a pattern made by the same lithography as a chip.
That last point explains what has reached products. Metasurfaces — single layers of sub-wavelength pillars that set the local phase, as in the "generalised Snell's law" of Federico Capasso's group (2011) — can be made in semiconductor fabs, and metalenses for near-infrared sensing and time-of-flight depth cameras have been shipped in consumer devices since 2022. Electronically steered metasurface antennas are used in satellite terminals and radar. By contrast, negative-index materials and invisibility "cloaks" remain demonstrations: the 2006 microwave cloak at Duke University worked in two dimensions, in a narrow band, for one polarisation; at optical frequencies, metal losses absorb most of the light, and Pendry's "perfect lens" is limited by the same losses. Lumenary EE's antennas and radio and photonics cover the devices.
What is real, and what remains a laboratory curiosity
A fair accounting, as of late 2026:
| Status | Examples |
|---|---|
| In products | Additively manufactured porous titanium in orthopaedic implants; printed polymer lattices in shoe midsoles, bicycle saddles and helmet liners (sold since about 2017–2020); honeycomb and foam cores in aircraft and packaging (architected materials, decades old); acoustic Bragg reflectors in phone filters; metalenses in sensors; metasurface antennas |
| Being qualified | Lattice heat exchangers and lightweight brackets in aerospace; crash and impact absorbers; vibration-isolation mounts; auxetic textiles and medical devices; machine-learning-designed lattices |
| Laboratory curiosities, so far | Optical invisibility cloaks at human scale; perfect lenses; seismic cloaks for cities; most programmable and topological mechanical metamaterials |
The pattern is that a metamaterial reaches a product when its manufacturing route already exists (lithography, additive manufacturing, honeycomb expansion), when its narrow band or limited strain is acceptable, and when it can be qualified like any other material — tested for fatigue, environmental durability, variability and inspection. The obstacles are rarely the physics and usually manufacturing at scale, defects, cost and certification.
Common misconceptions
Misconception
"A Poisson's ratio below −1 or above 0.5 is impossible." Those bounds hold for isotropic materials. Anisotropic honeycombs and lattices can exceed them in particular directions, as the program's and show, without violating thermodynamics.
Misconception
"Metamaterials beat the laws of physics." They exploit them: negative effective properties exist only in a frequency band near a resonance, are accompanied by dispersion and loss, and respect causality. A "negative mass" is the effective response of a resonating cell, not antimatter.
Misconception
"Lattices are always lighter for the same performance." For stiffness per weight in one load case, a solid plate or a well-designed sandwich panel often beats a lattice; lattices win when multiple functions matter (energy absorption, heat exchange, bone ingrowth, tunable properties) or when the load cases are complex.
Misconception
"Invisibility cloaks are around the corner." Working cloaks are narrowband, polarisation-specific, usually two-dimensional and much larger than what they hide, and broadband cloaking of visible light for a large object faces fundamental limits from causality and loss.
Exercises
1. Metamaterials and architected materials
Warm-upHumanUse Maxwell's rule for a two-dimensional pin-jointed frame, . What is for a single hexagon of six struts joined at six pin joints?
2. Metamaterials and architected materials
CoreHumanUsing the ideal Gibson–Ashby laws, for a stretch-dominated lattice and for a bending-dominated one, how many times stiffer is the stretch-dominated lattice at a relative density of 0.04? Give the nearest integer.
3. Metamaterials and architected materials
CoreHumanWhy can a locally resonant acoustic metamaterial block sound at wavelengths much longer than its cells, while a phononic crystal (Bragg) cannot?
4. Metamaterials and architected materials
CoreHumanAs of 2026, which of these architected materials or metamaterials are in commercial products? Select all that apply.
5. Metamaterials and architected materials
CoreHumanThe lesson's program found for a honeycomb with and . Does this violate the bound ?
6. Metamaterials and architected materials
CoreComputationalWrite honeycomb_poisson(theta_deg, h_over_l) returning . Raise ValueError when is 0 or ±90° (the formula breaks down) or when (the re-entrant struts would cross).
7. Metamaterials and architected materials
StretchComputationalWrite resonant_gap(m1, m2, k1, k2, a=1.0, nq=401) for the lesson's mass-in-mass chain. Sample nq wavenumbers in , solve the eigenproblem at each, and return (lo, hi): the maximum of the lower branch and the minimum of the upper branch. The tests compare with exact band edges.
Further reading and resources
- Foam Structures with a Negative Poisson's Ratio · Roderic Lakes · 1987
- Models for the elastic deformation of honeycombs · I. G. Masters, K. E. Evans · 1996Flexure, stretching and hinging of honeycomb cells, including re-entrant ones.
- Cellular Solids: Structure and Properties (2nd edition) · Lorna J. Gibson, Michael F. Ashby · 1997Cambridge University Press. The source of the scaling laws for foams and honeycombs.
- Foam topology: bending versus stretching dominated architectures · V. S. Deshpande, M. F. Ashby, N. A. Fleck · 2001
- Ultralight Metallic Microlattices · Schaedler et al. · 2011
- Ultralight, ultrastiff mechanical metamaterials · Zheng et al. · 2014
- Strong, lightweight, and recoverable three-dimensional ceramic nanolattices · Lucas R. Meza, Satyajit Das, Julia R. Greer · 2014
- Flexible mechanical metamaterials · Bertoldi, Vitelli, Christensen, van Hecke · 2017
- Three-dimensional mechanical metamaterials with a twist · Tobias Frenzel, Muamer Kadic, Martin Wegener · 2017
- Inverting the structure–property map of truss metamaterials by deep learning · Bastek, Kumar, Telgen, Glaesener, Kochmann · 2022
- Locally Resonant Sonic Materials · Liu, Zhang, Mao, Zhu, Yang, Chan, Sheng · 2000
- Extremely Low Frequency Plasmons in Metallic Mesostructures · Pendry, Holden, Stewart, Youngs · 1996
- Composite Medium with Simultaneously Negative Permeability and Permittivity · Smith, Padilla, Vier, Nemat-Nasser, Schultz · 2000
- Experimental Verification of a Negative Index of Refraction · Shelby, Smith, Schultz · 2001
- Negative Refraction Makes a Perfect Lens · J. B. Pendry · 2000
- Metamaterial Electromagnetic Cloak at Microwave Frequencies · Schurig et al. · 2006
- Light Propagation with Phase Discontinuities: Generalized Laws of Reflection and Refraction · Yu, Genevet, Kats, Aieta, Tetienne, Capasso, Gaburro · 2011