Supplementary Notes

Nearly-incompressible transverse isotropy (NITI) of cornea elasticity:

model and experiments with acoustic micro-tapping OCE

John J. Pitre Jr.1*, Mitchell A. Kirby1*, David S. Li1,2, Tueng T. Shen3, Ruikang K. Wang1,3,
Matthew O’Donnell1, and Ivan Pelivanov1

1. University of Washington, Department of Bioengineering, Seattle, Washington, United States

2. University of Washington, Department of Chemical Engineering, Seattle, Washington, United States
3. University of Washington, Department of Ophthalmology, Seattle, Washington, United States

Supplementary Note 1. Table of literature-reported mechanical testing results for ex vivo cornea

The following table summarizes the large variability in reported values of elastic moduli for healthy ex vivo
cornea in a number of mammalian species. We include moduli only for the low-strain region where they
generally are near their lowest value. Best effort was taken to report values for fresh tissue samples. The
reported moduli vary greatly based on the loading method, reconstruction method, and assumed model.
Tissue Loading Condition Loading Reconstruction Assumed Model Young’s Shear Modulus
Type Rate Method Modulus (E) (𝜇 )
Porcine 15mmHg - 140mmHg Dynamic FEM Hyperelastic 300kPa1 -
Inflation/Displacement nonlinear Ogden
material
Porcine 15mmHg - 30mmHg Dynamic FEM Generalized 2.6 MPa2 -
Inflation/Air-puff Maxwell
viscoelastic
Porcine 15mmHg Inflation/Air- Dynamic FEM Hyperelastic .99-1.59 MPa3 -
puff Mooney Rivlin
Porcine Tensile Quasi-static Stress-strain - 1.15-1.93 MPa3 -
Porcine 0mmHg - 40mmHg Dynamic Pressure- Thin shell .1-.3 MPa4 -
Inflation/Air-puff Deformation
Porcine Tensile Quasi-static Stress-strain - 3.193±1.589 -
MPa4
Porcine .75mmHg - 170mmHg Dynamic Pressure- Thin shell .15-.3 MPa5 -
Inflation Deformation
Porcine Tensile Quasi-static Stress-strain - .3-1.1 MPa6 -
Porcine 20mmHg Inflation/Air- Dynamic Modified Isotropic 60 kPa7 -
Puff OCE Rayleigh-Lamb Homogenous
Equation Viscoelastic
Porcine 15mmHg - 30mmHg Dynamic Modified Isotropic 41.8-157 kPa8 -
Inflation/Air-Puff OCE Rayleigh-Lamb Homogenous
Equation Viscoelastic
Porcine Compression Dynamic Transient Transverse Isotropic 5.61±2.27 kPa -
compression Biphasic (compression)9
stress relaxation 1.33± .51 MPa
(tension)9
Porcine Compression Quasi-static Compressive Transverse Isotropic .65-6.32 MPa10 -
strain Biphasic
 Porcine Oscillatory shear Dynamic Shear strain - - 2.5-9 kPa11
(.01-2 Hz)
Porcine 0mmHg - 30mmHg Quasi-static Pressure- Linear elastic 2.29±1.63MPa12 -
Inflation Deformation
Porcine 15mmHg - 35mmHg Dynamic FEM Linear isotropic 25.5 MPa -
Inflation/Air-puff viscoelastic (anterior)
.85 MPa
(posterior)13
Porcine 15mmHg Inflation/Air- Dynamic Group Velocity Homogenous linear 5.9±.6 kPa14 -
Puff OCE isotropic
Porcine Tensile Quasi-static Stress-strain - .8-2.2 MPa15 -
Porcine Tensile Quasi-static Stress-strain - 3.70± .24 MPa16 -
Human Oscillatory shear Dynamic Shear strain - - 3-15 kPa11
(.01-2 Hz)
Human Tensile Quasi-static Stress-strain - .34-4.1 MPa17 -
Human Acoustic Vibration Dynamic FEM Linear isotropic 24.8 kPa -
(50-600Hz) viscoelastic (anterior)
19.8 kPa
(posterior)18
Human 15mmHg - 35mmHg Dynamic FEM Linear isotropic .71MPa13 -
Inflation/Air-puff viscoelastic
Human Oscillatory shear Dynamic Shear strain - - 38.7±8.6 kPa19
(.03 Hz)
Human 1D Shear Quasi-static Shear strain - - 10-90 kPa
(Nasal-
Temporal)
10-150 kPa
(Superior-
Inferior)20
Human Tensile Quasi-static Stress-strain - .8-2.6 MPa15 -
Human Tensile Quasi-static Stress-strain - 3.81± .40 MPa16 -
Human Tensile Quasi-static Stress-strain - 19.1±3.5 MPa21 -
Human .75mmHg- 160mmHg Dynamic Pressure- Thin shell .25-3 MPa22 -
Inflation Deformation
Human .75mmHg- 170mmHg Dynamic Pressure- Thin shell .2-.6 MPa5 -
Inflation Deformation
Bovine 1mmHg - 10mmHg Dynamic FEM Linear isotropic 40-185 kPa23 -
Inflation/Piezo-shaker (300 Hz)
motion MRE
New 15mmHg Inflation/Air- Dynamic FEM Nonlinear 5 MPa24 -
Zealand puff Hyperelastic
White Mooney Rivlin plus
Rabbit a Prony-series
viscoelastic
Rabbit 15mmHg Inflation/Air- Dynamic FEM Homogenous linear 500-800 kPa25 -
Puff OCE isotropic

Supplementary Note 2. Behavior of NITI materials under common mechanical tests

In this study, we show that a nearly incompressible transversely isotropic (NITI) model may explain many of the
discrepancies in reported values of corneal Young’s modulus. In particular, there is a multiple order-of-
magnitude difference between the values reported by tensile/inflation tests and those reported by
shear/transient tests. Tensile tests of corneal strips yield values of 800 kPa – 4.7 MPa,3,4,6,15–17,21 and inflation
tests of corneal trephinates yield values of 100 kPa – 3 MPa.1–5,12,13,22 In contrast, shear (torsional) tests report
shear moduli of 2.5 – 47.3 kPa.11,20,26 This corresponds to a Young’s modulus of 7.5 – 142 kPa if one assumes an
isotropic model. Likewise, transient methods such as optical coherence elastography (OCE) report Young’s
moduli in the range 5.3 – 157 kPa,8,27 again relying on an isotropic model to convert shear modulus to Young’s
modulus.

The NITI model has a key property that can explain both regimes of reported moduli – namely, its shear and
tensile behavior are decoupled. Mechanical tests that probe these behaviors independently will report greatly
different Young’s moduli. In fact, shear tests will report an independent parameter unrelated to Young’s
modulus. The shear behavior is governed by an independent modulus, which we refer to as 𝐺. The tensile
behavior is governed by the modulus 𝜇, which can be related to the Young’s modulus 𝐸𝑇𝐼 = 3𝜇. Here, we
demonstrate that the relevant stress-strain relations for each test type will depend primarily on one elastic
parameter while remaining agnostic to the other. Furthermore, tensile and inflation measurements should yield
similar values, as should shear and transient ones.

Supplementary Fig. S2. Corneal geometry and local coordinate axes used in analyzing deformation for (a)
tensile, torsional, and wave propagation tests and (b) inflation tests.

Hooke’s law for a NITI material. Hooke’s law relates stress 𝜎 and strain 𝜀 in a linearly elastic solid. In its most
general form, it may be written
𝜎𝑖𝑗 = 𝑐𝑖𝑗𝑘𝑙 𝜀𝑘𝑙 , (𝑆2.1)

where summation is implied over repeated indices. In a general anisotropic solid, the fourth-rank stiffness
tensor 𝑐𝑖𝑗𝑘𝑙 contains 21 independent elastic constants. This reduces to 5 constants in a transversely isotropic
solid, 3 constants in a NITI solid, and 2 constants in an isotropic solid. For convenience, Equation S2.1 is often
written in Voigt notation
𝜎𝑥𝑥 𝐶11 𝐶12 𝐶13 𝐶14 𝐶15 𝐶16 𝜀𝑥𝑥
𝜎𝑦𝑦 ∗ 𝐶22 𝐶23 𝐶24 𝐶25 𝐶26 𝜀𝑦𝑦
𝜎𝑧𝑧 ∗ ∗ 𝐶33 𝐶34 𝐶35 𝐶36 𝜀𝑧𝑧
𝜏𝑦𝑧 = ∗ , (𝑆2.2)
∗ ∗ 𝐶44 𝐶45 𝐶46 𝛾𝑦𝑧
𝜏𝑥𝑧 ∗ ∗ ∗ ∗ 𝐶55 𝐶56 𝛾𝑥𝑧
𝜏
[ 𝑥𝑦 ] [ ∗ ∗ ∗ ∗ ∗ 𝐶66 ] [𝛾𝑥𝑦 ]
where 𝜏𝑖𝑗 denotes shear stresses, 𝛾𝑖𝑗 = 2𝜀𝑖𝑗 denotes shear strains, and stars denote symmetric entries (i.e. 𝐶21 =
𝐶12 ). The subscripts 𝑥, 𝑦, and 𝑧 refer to standard Cartesian axes, shown for the cornea in Suppl. Fig. S2a.

To define a material as nearly incompressible, we first consider the case of an isotropic medium. The stiffness
matrix in this case depends on two constants, the Lamé parameters 𝜆 and 𝜇, and the stress-strain relation is

𝜎𝑥𝑥 𝜆 + 2𝜇 𝜆 𝜆 𝜀𝑥𝑥
𝜎𝑦𝑦 𝜆 𝜆 + 2𝜇 𝜆 𝜀𝑦𝑦
𝜎𝑧𝑧 𝜆 𝜆 𝜆 + 2𝜇 𝜀𝑧𝑧
𝜏𝑦𝑧 = 𝛾𝑦𝑧 . (𝑆2.3)
𝜇
𝜏𝑥𝑧 𝜇 𝛾𝑥𝑧
[ 𝜏𝑥𝑦 ] [ 𝜇] [𝛾𝑥𝑦 ]

We then define the mean internal pressure 𝑃 and the dilatation 𝜃 in the medium as
1
𝑃= (𝜎 + 𝜎𝑦𝑦 + 𝜎𝑧𝑧 ), (𝑆2.4)
3 𝑥𝑥
𝜃 = 𝜀𝑥𝑥 + 𝜀𝑦𝑦 + 𝜀𝑧𝑧 . (𝑆2.5)

For a nearly-incompressible material, the dilatation must equal zero, but the internal pressure must remain
finite. Rewriting the stress tensor in terms of the dilatation, we see that
𝜎𝑖𝑗 = (𝜆𝜃)𝛿𝑖𝑗 + 2𝜇𝜀𝑖𝑗 , (𝑆2.6)

and the internal pressure becomes

2
𝑃 = 𝜆𝜃 + 𝜇𝜃. (𝑆2.7)
3
As 𝜃 → 0, the pressure remains finite only if 𝜆 → ∞. Therefore, the internal pressure for a nearly-incompressible
material may be defined
𝑃 = lim 𝜆𝜃 . (𝑆2.8)
𝜆→∞
𝜃→0

A similar analysis helps define conditions for a nearly-incompressible transversely isotropic (NITI) solid.28 Unlike
the isotropic case, a NITI solid cannot be defined precisely by a single condition. Instead, there are a family of
materials which satisfy the incompressibility condition, with constraints on valid stiffness values. For a general TI
solid, the stress-strain relation is defined by five independent constants (with 𝐶12 = 𝐶11 − 2𝐶66 ):

𝜎𝑥𝑥 𝐶11 𝐶12 𝐶13 𝜀𝑥𝑥

𝜎𝑦𝑦 𝐶12 𝐶11 𝐶13 𝜀𝑦𝑦
𝜎𝑧𝑧 𝐶13 𝐶13 𝐶33 𝜀𝑧𝑧
𝜏𝑦𝑧 = 𝛾𝑦𝑧 . (𝑆2.9)
𝐶44
𝜏𝑥𝑧 𝐶44 𝛾𝑥𝑧
𝜏
[ 𝑥𝑦 ] [ 𝛾
[ 𝑥𝑦 ]
𝐶66 ]

For notational convenience, we may rename some of the constants as follows

 𝜎𝑥𝑥 𝜆 + 2𝜇 𝜆 𝐶13 𝜀𝑥𝑥
𝜎𝑦𝑦 𝜆 𝜆 + 2𝜇 𝐶13 𝜀𝑦𝑦
𝜎𝑧𝑧 𝐶13 𝐶13 𝐶33 𝜀𝑧𝑧
𝜏𝑦𝑧 = 𝛾𝑦𝑧 . (𝑆2.10)
𝐺
𝜏𝑥𝑧 𝐺 𝛾𝑥𝑧
[ 𝜏𝑥𝑦 ] [ 𝛾
𝜇 ] 𝑥𝑦 ]
[

As before, the dilatation of a material is defined in terms of the normal strain components:

𝜃 = 𝜀𝑥𝑥 + 𝜀𝑦𝑦 + 𝜀𝑧𝑧 (𝑆2.11)

where the strains are related to the normal stresses 𝜎𝑥𝑥 , 𝜎𝑦𝑦 , and 𝜎𝑧𝑧 and the longitudinal part of the
compliance tensor 𝐶ℓ−1 :

𝜀ℓ = 𝐶ℓ−1 𝜎ℓ , (𝑆2.12)

where

2 2 −𝐶 𝐶
𝐶 𝐶33 − 𝐶13 𝐶13 𝐶13 (𝐶12 − 𝐶11 )
1 11 12 33
𝐶ℓ−1 2
= [ 𝐶13 − 𝐶12 𝐶33 2
𝐶11 𝐶33 − 𝐶13 𝐶13 (𝐶12 − 𝐶11 )] (𝑆2.13)
Δ 2 − 𝐶2
𝐶13 (𝐶12 − 𝐶11 ) 𝐶13 (𝐶12 − 𝐶11 ) 𝐶11 12

2 ]
Δ = det 𝐶ℓ = (𝐶11 − 𝐶12 )[(𝐶11 + 𝐶12 )𝐶33 − 2𝐶13 (𝑆2.14)

As before, we require the normal stresses and the mean internal pressure of the material to describe
deformations of this system. We define new variables

𝑄1 = 𝐶13 − 𝜆 (𝑆2.15)

𝑄2 = 𝐶33 − (𝜆 + 2𝜇) (𝑆2.16)

and write the stresses, internal pressure 𝑃, and dilatation 𝜃 as:

𝜎𝑥𝑥 = 𝜆𝜃 + 2𝜇𝜀𝑥𝑥 + 𝑄1 𝜀𝑧𝑧 (𝑆2.17)

𝜎𝑦𝑦 = 𝜆𝜃 + 2𝜇𝜀𝑦𝑦 + 𝑄1 𝜀𝑧𝑧 (𝑆2.18)

𝜎𝑧𝑧 = 𝜆𝜃 + 2𝜇𝜀𝑥𝑥 + 𝑄1 (𝜀𝑥𝑥 + 𝜀𝑦𝑦 ) + 𝑄2 𝜀𝑧𝑧 (𝑆2.19)

2 1 1
𝑃 = 𝜆𝜃 + 𝜇𝜃 + 𝑄1 𝜃 + (𝑄1 + 𝑄2 )𝜀𝑧𝑧 (𝑆2.20)
3 3 3

2𝜇
𝜃= {(𝑄2 − 𝑄1 + 2𝜇)(𝜎𝑥𝑥 + 𝜎𝑦𝑦 ) − 2(𝑄1 − 𝜇)𝜎𝑧𝑧 } (𝑆2.21)
𝛥

2
Δ = 4𝜇 {(𝜆 + 𝜇)𝑄2 − 𝑄1 (𝑄1 + 2𝜆) + 3𝜇 (𝜆 + 𝜇)} (𝑆2.22)
3
We define a NITI material as one for which the internal pressure remains finite as 𝜆 → ∞ and 𝜃 → 0, while the
parameters 𝜇, 𝑄1 , and 𝑄2 remain finite. Note that we can require 𝐶13 and 𝐶33 to be asymptotically equal to 𝜆 so
that 𝑄1 and 𝑄2 remain finite.

To satisfy incompressibility exactly for any general stress field requires that the coefficients of the stresses equal
zero exactly:

𝑄2 − 𝑄1 + 2𝜇 = 0 (𝑆2.23)

𝑄1 − 𝜇 = 0 (𝑆2.24)

which yields 𝑄1 = 𝜇, 𝑄2 = −𝜇. These values correspond to the case 𝛥 = 0, for which the stiffness matrix is
singular. This means that there is no exact general definition of a NITI material. However, from the definition of
𝜃, we note that 𝜃 → 0 if 𝛥 → ∞, provided that 2𝜇{(𝑄2 − 𝑄1 + 2𝜇)(𝜎𝑥𝑥 + 𝜎𝑦𝑦 ) − 2(𝑄1 − 𝜇)𝜎𝑧𝑧 } remains finite
(it does by definition). We note that in the limit 𝜆 → ∞, this condition is indeed satisfied.

As in the isotropic case, the definition of the internal pressure for a NITI material requires that lim 𝜆𝜃 must
𝜆→∞
remain finite:

2𝜇𝜆
lim 𝜆𝜃 = lim {(𝑄2 − 𝑄1 + 2𝜇)(𝜎𝑥𝑥 + 𝜎𝑦𝑦 ) − 2(𝑄1 − 𝜇)𝜎𝑧𝑧 } . (𝑆2.25)
𝜆→∞ 𝜆→∞ 𝛥

Note that the term in braces is finite, and we may evaluate the limit as follows:

𝜆
lim 𝜆𝜃 = 2𝜇{(𝑄2 − 𝑄1 + 2𝜇)(𝜎11 + 𝜎22 ) − 2(𝑄1 − 𝜇)𝜎33 } ⋅ lim
𝜆→∞ 𝜆→∞ 𝛥

𝜆 𝜆
lim =
𝜆→∞ 𝛥 4𝜇{𝜆(𝑄2 − 2𝑄1 + 3𝜇) + 𝜇𝑄2 − 𝑄12 + 2𝜇2 }

𝜆 1
lim = (𝑆2.26)
𝜆→∞ 𝛥 4𝜇(𝑄2 − 2𝑄1 + 3𝜇)

which is finite provided 𝑄2 − 2𝑄1 + 3𝜇 ≠ 0.

The conditions for a NITI material then may be summarized as:

1. 𝐶13 and 𝐶33 are asymptotically equal to 𝜆 (𝑆2.27𝑎)

2. Δ ≠ 0 (𝐶13 ≠ 𝐶33 ) (𝑆2.27𝑏)
3. 𝑄2 − 2𝑄1 + 3𝜇 ≠ 0 (𝑆2.27𝑐)

Note that these are very loose conditions. For instance, given two constants 𝐴 and 𝐵 that are small relative to 𝜆,
we then note that

𝐶13 = 𝜆 + 𝐴 (𝑆2.28)

𝐶33 = 𝜆 + 𝐵 (𝑆2.29)

satisfy condition (𝑆2.27𝑎).

These give 𝑄1 = 𝐴 and 𝑄2 = 𝐵 + 2𝜇. Of course, there are still values of 𝐴 and 𝐵 that violate the constraints, but
there are at most three such real values of 𝐴 for any given 𝐵:

2
𝛥 = 4𝜇 {(𝜆 + 𝜇)(𝐵 + 2𝜇) − 𝐴(𝐴 + 2𝜆) + 3𝜇 (𝜆 + 𝜇)} , (𝑆2.30)
3

𝐵 − 2𝐴 + 5𝜇 ≠ 0 (𝑆2.31)

So there is an extensive family of NITI materials, but we must be careful of the relative values of the stiffness
entries. This is especially problematic for an automated optimization routine that must avoid these singular
points when inverting the stiffness values from observed mechanical behavior.

The simplest solution is to define a NITI material by taking the longitudinal portion of the stiffness matrix to be
isotropic. In this case, 𝐶13 = 𝜆 and 𝐶33 = 𝜆 + 2𝜇. This is equivalent to 𝑄1 = 𝑄2 = 0. In this case, we have

2
𝛥 = 12𝜇2 (𝜆 + 𝜇) (𝑆2.32)
3

which is clearly nonzero. Additionally, 𝑄2 − 2𝑄1 + 3𝜇 ≠ 0 is obviously satisfied. For a NITI material, Equation
S2.2 therefore reduces to

𝜎𝑥𝑥 𝜆 + 2𝜇 𝜆 𝜆 𝜀𝑥𝑥
𝜎𝑦𝑦 𝜆 𝜆 + 2𝜇 𝜆 𝜀𝑦𝑦
𝜎𝑧𝑧 𝜆 𝜆 𝜆 + 2𝜇 𝜀𝑧𝑧
𝜏𝑦𝑧 = 𝛾𝑦𝑧 . (𝑆2.33)
𝐺
𝜏𝑥𝑧 𝐺 𝛾𝑥𝑧
[ 𝜏𝑥𝑦 ] [ 𝛾
𝜇 ] 𝑥𝑦 ]
[

This approximation corresponds to the case of a weakly anisotropic material, since the only anisotropy is due to
the shear terms 𝐶44 = 𝐺. It was recently demonstrated using Brillouin microscopy that the cornea is only weakly
anisotropic in its longitudinal terms, with stiffness values reported as 𝐶11 = 2.971 GPa, 𝐶33 = 2.662 GPa, and 𝐶13
≈ 2.680 GPa.29 These are clearly not as restrictive as our assumption, but in practice, these terms contribute little
to the behavior we probe (guided waves) since it is dominated by the shear terms. We demonstrate this in
Supplementary Note 4.

In the following sections, we consider the stress-strain behavior of the NITI model (Equation S2.33) under various
corneal mechanical tests.

Tensile testing of corneal strips. In tensile tests, rectangular strips of ex vivo cornea are subjected to uniaxial
tension. The corresponding strain is measured, and the Young’s modulus is quantified as 𝐸 = 𝜎/𝜀. In general, the
orientation of the cornea strip lies in the xy-plane, and the direction of the applied stress aligns with the long
axis of the strip. While anisotropy within the xy-plane has been reported by some studies, the degree of
anisotropy at low IOP (low pre-stress)6,30–33 suggests that the cornea microstructure can be approximated with
the NITI model as symmetric for any direction in the xy-plane. Thus, it is sufficient to consider only one
orientation of the strip. Consider uniaxial loading in the 𝑥 direction. The stress-strain relation becomes
 𝜎𝑥𝑥 𝜆 + 2𝜇 𝜆 𝜆 𝜀𝑥𝑥
0 𝜆 𝜆 + 2𝜇 𝜆 𝜀𝑦𝑦
0 𝜆 𝜆 𝜆 + 2𝜇 𝜀𝑧𝑧
= 𝛾𝑦𝑧 . (𝑆2.34)
0 𝐺
0 𝐺 𝛾𝑥𝑧
[ 0 ] [ 𝛾
𝜇 ] 𝑥𝑦 ]
[

It is clear from this linear system that the shear strains are zero and consequentially any equations with 𝐺
vanish. We are left with
𝜎𝑥𝑥 𝜆 + 2𝜇 𝜆 𝜆 𝜀𝑥𝑥
[ 0 ]=[ 𝜆 𝜆 + 2𝜇 𝜆 ] [𝜀𝑦𝑦 ] . (𝑆2.35)
0 𝜆 𝜆 𝜆 + 2𝜇 𝜀𝑧𝑧
Solving the third row for 𝜀𝑧𝑧 gives
𝜆
𝜀𝑧𝑧 = − (𝜀 + 𝜀𝑦𝑦 ). (𝑆2.36)
𝜆 + 2𝜇 𝑥𝑥
Substituting this into the second row and solving for 𝜀𝑦𝑦 gives

𝜆
𝜀𝑦𝑦 = − 𝜀 . (𝑆2.37)
2(𝜆 + 𝜇) 𝑥𝑥
Finally, substituting into the first row and defining the Young’s modulus as 𝐸 = 𝜎𝑥𝑥 /𝜀𝑥𝑥, we find

𝜇(3𝜆 + 2𝜇)
𝐸= , (𝑆2.38)
𝜆+𝜇
which in the incompressible limit 𝜆 → ∞ gives 𝐸 = 3𝜇. From this, we see that a NITI material under tensile test
will have an apparent Young’s modulus that depends only on 𝜇 and not on 𝐺.

Inflation tests of corneal trephinates. In corneal inflation tests, a circular region of the cornea and sclera (called
a trephinate) is dissected from an ex vivo eye (see for example, Anderson et. al.1). It is clamped above a fluid-
filled chamber and sealed around the scleral rim. The chamber is connected to a water column whose height is
varied to simulate intraocular pressure (IOP). For a given pressure 𝑝, the test measures the rise of the corneal
apex 𝑟. The corneal thickness ℎ, radius of curvature 𝑅, and the contact angle at the clamped edge 𝛾 are also
estimated. Deformation of the cornea is then modeled using the theory of spherical shells.34

The corneal inflation test can be described by the coordinate system shown in Suppl. Fig. S2b. The deformation
of an infinitesimal shell element under pressure is governed by the normal stress resultants 𝑁𝜑 and 𝑁𝜃 , bending
moments 𝑀𝜑 and 𝑀𝜃 , and shear stress resultant 𝑄𝜑 . Here, 𝜑 represents the meridional coordinate (sometimes
denoted by 𝑦 in local element coordinates) and 𝜃 denotes the azimuthal coordinate (sometimes denoted as 𝑥 in
local element coordinates). The resultants and moments are defined:
ℎ
2 𝑧
𝑁𝜃 = ∫ 𝜎𝜃 (1 − ) d𝑧 , (2.39)
−
ℎ 𝑅
2

ℎ/2
𝑧
𝑁𝜑 = ∫ 𝜎𝜑 (1 − ) d𝑧,
−ℎ/2 𝑅
 ℎ
2 𝑧
𝑀𝜃 = ∫ 𝜎𝜃 𝑧 (1 − ) d𝑧 ,
−
ℎ 𝑅
2

ℎ/2
𝑧
𝑀𝜑 = ∫ 𝜎𝜑 𝑧 (1 − ) d𝑧,
−ℎ/2 𝑅
ℎ/2
𝑧
𝑄𝜑 = ∫ 𝜏𝜑𝑧 (1 − ) d𝑧.
−ℎ/2 𝑅

Balancing the forces and moments on the shell element and considering the relationship between strain and the
deformed shape leads to a system of five equations describing the rise of the corneal apex following inflation.1

In assigning a constitutive model to the shell, we apply Hooke’s law in the local coordinate system of the shell
element. As the shell is assumed to be thin relative to its radius of curvature (ℎ ≪ 𝑅), a plane stress
approximation is also applied. This yields the following stress-strain relation for the shell element:
𝜎𝜃 𝜆 + 2𝜇 𝜆 𝜆 𝜀𝜃
𝜎𝜑 𝜆 𝜆 + 2𝜇 𝜆 𝜀𝜑
0 𝜆 𝜆 𝜆 + 2𝜇 𝜀𝑧
(𝑆2.40)
0 = 𝐺 𝛾𝜑𝑧 .
0 𝐺 𝛾𝜃𝑧
[𝜏𝜃𝜑 ] [ 𝛾
𝜇 ] [ 𝜃𝜑 ]
We immediately see, as with the tensile test, that the shear strains vanish and the solution will not depend on 𝐺.
Proceeding as in the tensile test, we solve the third row for 𝜀𝑧 ,
𝜆
𝜀𝑧 = − (𝜀 + 𝜀𝜑 ) (𝑆2.41)
𝜆 + 2𝜇 𝜃
and substitute into rows 1 and 2 to obtain a simplified system

4𝜇(𝜆 + 𝜇) 2𝜆𝜇
𝜎𝜃 = 𝜀𝜃 + 𝜀 , (𝑆2.42)
𝜆 + 2𝜇 𝜆 + 2𝜇 𝜑
2𝜆𝜇 4𝜇(𝜆 + 𝜇)
𝜎𝜑 = 𝜀𝜃 + 𝜀 . (𝑆2.43)
𝜆 + 2𝜇 𝜆 + 2𝜇 𝜑
For convenience, we convert Equations S2.42 and S2.43 from 𝜆-𝜇 (Lamé parameter) form to the Young’s
modulus and Poisson’s ratio form using the following relationships:
𝐸𝜈
𝜆= , (𝑆2.44)
(1 + 𝜈)(1 − 2𝜈)
𝐸
𝜇= . (𝑆2.45)
2(1 + 𝜈)
to obtain
𝐸
𝜎𝜃 = (𝜀 + 𝜈𝜀𝜑 ), (𝑆2.46)
1 − 𝜈2 𝜃
𝐸
𝜎𝜑 = (𝜀 + 𝜈𝜀𝜃 ). (𝑆2.47)
1 − 𝜈2 𝜑
For an incompressible material, 𝜈 = 0.5, and the stresses depend only on the Young’s modulus 𝐸 = 3𝜇. As these
stresses are the only terms that appear in the stress resultants and bending moments, it follows that those
quantities depend only on 𝜇. The apical rise measured in an inflation test depends on these resultants and
moments, and so also depends only on 𝜇. Thus, like the tensile test, an inflation test of a NITI material will
measure 𝜇 and not 𝐺.

Shear torsional test of corneal trephinates. In torsional tests of corneal trephinates, a cylindrical section of ex
vivo cornea (with thickness ℎ and radius 𝑅) is placed between parallel platens and a rotational deformation Θ is
applied to the sample. The torque 𝑇 at the top platen is measured, and Hooke’s law for shear is used to estimate
the shear modulus of the sample. The torque is given by the integral
𝑅 2𝜋
𝑇 = ∫ ∫ 𝜏𝜃𝑧 𝑟 d𝜃d𝑟 . (𝑆2.48)
0 0

The transformation from Cartesian coordinates to cylindrical coordinates gives the shear stress and strain

𝜏𝜃𝑧 = − sin 𝜃 𝜏𝑥𝑧 + cos 𝜃 𝜏𝑦𝑧 , (𝑆2.49)

𝛾𝜃𝑧 = − sin 𝜃 𝛾𝑥𝑧 + cos 𝜃 𝛾𝑦𝑧 . (𝑆2.50)

For the NITI material, we have

𝜏𝑥𝑧 = 𝐺𝛾𝑥𝑧 , (𝑆2.51)

𝜏𝑦𝑧 = 𝐺𝛾𝑦𝑧 . (𝑆2.52)

Substituting equations S2.51 and S2.52. into S2.49, we obtain

𝜏𝜃𝑧 = 𝐺𝛾𝜃𝑧 . (𝑆2.53)

The shear strain and stress can be approximated as

𝑟Θ
𝛾𝜃𝑧 = (𝑆2.54)
ℎ
𝐺𝑟Θ
𝜏𝜃𝑧 = (𝑆2.55)
ℎ
Evaluating the torque integral and solving for 𝐺, we obtain
2ℎ
𝐺=𝑇 . (𝑆2.56)
𝜋Θ𝑅4
Clearly, the shear torsional test provides an estimate of 𝐺 only and is not influenced by 𝜇. Thus, modulus
estimates from shear torsional tests may vary greatly from tensile and inflation measurements.

Dynamic measurements of corneal elasticity. Dynamic or transient elasticity measurements use the
propagation of elastic waves to estimate elastic moduli. They can provide non-destructive in vivo measurements
of corneal elasticity and have the potential to be useful in clinical measurements. Because soft biological tissues
are nearly incompressible, the longitudinal wave speed is large (≈1540 m/s) compared to the shear wave speed
(1-10 m/s), and the shear wave speed provides a full description of tissue elasticity.

Two models of elastic wave propagation have been used to estimate corneal Young’s modulus. The simplest of
these treats the cornea as a semi-infinite isotropic solid.27 This model suffers from serious inaccuracies in
converting the measured group velocity to bulk shear wave speed.35 However, we analyze the assumption here
for completeness. In this case, a Rayleigh wave propagates along the air-cornea interface at constant speed
proportional to the bulk shear wave speed. By measuring the group velocity of the Rayleigh wave 𝑐𝑅 , one can
estimate the Young’s modulus as
𝑐𝑅 2
𝐸 = 3𝜌 ( ) . (𝑆2.57)
0.9553
In a transversely isotropic material, the Rayleigh wave speed can be obtained numerically by employing the
Stroh formalism36–40 or evaluating the Green’s function.41 For materials with 𝐺 < 𝜇, such as we expect for the
cornea, the Rayleigh wave speed is primarily governed by 𝐺 and only slightly influenced by 𝜇 (Supplementary
Note 3 provides detailed analysis of the Rayleigh wave speed in a NITI medium).

The cornea should be more accurately modeled as a bounded material. The bounded geometry gives rise to
dispersive guided waves, whose frequency-wavenumber behavior must be analyzed to quantify elasticity. Partial
wave analysis of the cornea as a flat isotropic plate bounded above by air and below by water leads to a secular
equation that describes guided wave modes.8,42 We detail the mode behavior (Results, Guided wave behavior in
a bounded NITI layer) and the partial wave solution for a NITI material (Supplementary Note 4) elsewhere in this
work.

In both our study and previous OCE studies of porcine cornea, only the A0 mode is analyzed to estimate
elasticity. For a bounded NITI material, the A0 mode is governed primarily by 𝐺 and slightly influenced by 𝜇.
Therefore, in both shear torsional tests and dynamic OCE tests, the observed mechanical behavior is governed
primarily by 𝐺 rather than 𝜇 (as in tensile/inflation tests). This is consistent with shear modulus values reported
by dynamic OCE studies where guided wave behavior has been taken into account through dispersion analysis of
the A0 mode.8,30

Supplementary Note 3. Wave behavior in a bulk NITI medium

Transversely isotropic materials support three bulk waves (quasi-longitudinal, quasi-shear, and shear) whose
wave speeds depend on the propagation angle. The mechanical behavior is symmetric with respect to rotations
about the z-axis, but varies with respect to the angle 𝜃 formed by the propagation direction and the z-axis
(Suppl. Fig. S3.1a).

When the propagation direction lies in the xy-plane (𝜃 = 90°), three pure wave modes exist (one longitudinal and
two orthogonal shear waves). Their wave speeds are

𝐶11
𝑐𝐿 = √ (𝑆3.1)
𝜌

𝑐𝑆1 = √𝐶44 /𝜌, polarized in z,

𝑐𝑆2 = √𝐶66 /𝜌, polarized in the xy-plane,

where 𝜌 is the material density.

For a general propagation angle, the polarization of propagating waves is not fully parallel or orthogonal to the
propagation direction and, therefore, one quasi-longitudinal, one quasi-shear, and one shear wave should be
considered. The following equations can be used to calculate their wave speeds 43:

𝐶11 sin2 𝜃 + 𝐶33 cos2 𝜃 + 𝐶44 + √𝑀(𝜃)

𝑐𝑞𝐿 = √ , (𝑆3.2)
2𝜌

𝐶11 sin2 𝜃 + 𝐶33 cos2 𝜃 + 𝐶44 − √𝑀(𝜃)

𝑐𝑞𝑆 = √ , (𝑆3.3)
2𝜌

𝐶66 sin2 𝜃 + 𝐶44 cos2 𝜃

𝑐𝑆 = √ , (𝑆3.4)
2𝜌

𝑀(𝜃) = [(𝐶11 − 𝐶44 ) sin2 𝜃 + (𝐶44 − 𝐶33 ) cos2 𝜃]2 + (𝐶13 + 𝐶44 )2 sin2 2𝜃 . (𝑆3.5)

For the NITI model, these equations simplify to

𝜆 + 2𝜇 + 𝐺 + √𝑀(𝜃)
𝑐𝑞𝐿 = √ , (𝑆3.6)
2𝜌

𝜆 + 2𝜇 + 𝐺 − √𝑀(𝜃)
𝑐𝑞𝑆 = √ , (𝑆3.7)
2𝜌

𝜇 sin2 𝜃 + 𝐺 cos2 𝜃
𝑐𝑆 = √ , (𝑆3.8)
2𝜌

𝑀(𝜃) = (𝜆 + 2𝜇 − 𝐺)2 cos2 2𝜃 + (𝜆 + 𝐺)2 sin2 2𝜃 . (𝑆3.9)

By evaluating these equations for a range of angles using approximate mechanical properties for cornea (𝜌 =
1000 kg/m3, 𝜇 = 1 MPa, 𝐺 = 20 kPa, and setting 𝜆 so that the average longitudinal wave speed is approximately
1540 m/s), we can gain some intuition of the bulk wave behavior in a NITI material. The quasi-longitudinal wave
speed is nearly constant over all angles, with variations of ± 0.01%. The quasi-shear waves’ speeds show a large
range of variability over angle (Suppl. Fig. S3.1b).
Supplementary Fig. S3.1. (a) Coordinate system, symmetry axis, and propagation angle for a TI material. (b)
Quasi-shear and shear wave speed as a function of propagation angle for a NITI material with properties
representative of cornea (𝜌 = 1000 kg/m3, 𝜇 = 1 MPa, 𝐺 = 20 kPa, 𝜆 = 2.37 GPa).

When the 𝑥𝑦-plane corresponds to a free surface, a Rayleigh wave propagates along the surface with speed 𝑐𝑅 .
The value of 𝑐𝑅 can be found using the Stroh formalism.36–40 Here, we consider the solution of a Rayleigh wave
propagating in the 𝐦 = [1, 0, 0] direction along the free surface defined by normal vector 𝐧 = [0, 0, 1] (because
of symmetry in the 𝑥𝑦-plane, the propagation direction is arbitrary). The Stroh formalism defines the 3×3
matrices

𝑄𝑖𝑘 = 𝑐𝑖𝑗𝑘𝑙 𝑚𝑗 𝑚𝑙 − 𝜌𝑣 2 𝛿𝑖𝑘 (𝑆3.10)

𝑅𝑖𝑘 = 𝑐𝑖𝑗𝑘𝑙 𝑚𝑗 𝑛𝑙 (𝑆3.11)

𝑇𝑖𝑘 = 𝑐𝑖𝑗𝑘𝑙 𝑛𝑗 𝑛𝑙 (𝑆3.12)

where 𝑐𝑖𝑗𝑘𝑙 is the fourth-order stiffness tensor, 𝜌 is the density, and 𝑣 is the wave speed. These matrices are
combined to form the 6×6 Stroh eigenvalue problem,
𝐚 𝐚
𝐍[ ] = 𝑝[ ], (𝑆3.13)
𝐛 𝐛
where

−𝐓 −1 𝐑𝑇 𝐓 −1
𝐍 = [ −1 𝑇 ]. (𝑆3.14)
𝐑𝐓 𝐑 − 𝐐 −𝐓 −1 𝐑𝑇
The eigenvectors contain the polarization vectors 𝐚 and traction vectors 𝐛 of harmonic waves that satisfy the
free surface boundary condition. The displacements and tractions for these solutions are

𝐮 = 𝐚𝑒 𝑖𝑘(𝐦⋅𝐱+𝑝𝐧⋅𝐱−𝑣𝑡) , (𝑆3.15)

𝐭 = 𝑖𝑘𝐛𝑒 𝑖𝑘(𝐦⋅𝐱−𝑣𝑡) . (𝑆3.16)

The Rayleigh wave is formed by a linear combination of wave modes whose amplitude must decay with depth.
For this reason, we only consider solutions where 𝑝 has a positive imaginary part. Below the limiting velocity of
the material, the six eigenvalues occur in complex conjugate pairs, and so only three modes form the Rayleigh
wave. Furthermore, the free surface condition implies that the linear combination of the tractions must be zero,
3

∑ 𝑐𝑖 𝐭 𝑖 = 𝟎. (𝑆3.17)
𝑖=1

This is often rewritten in the following form:

| | | 𝑐1
𝑖𝑘 [𝐛1 𝐛2 𝐛3 ] [𝑐2 ] 𝑒 𝑖𝑘(𝐦⋅𝐱−𝑣𝑡) = 𝐁𝐜 = 𝟎, (𝑆3.18)
| | | 𝑐3
which has a nontrivial solution if and only if the determinant of 𝐁 is zero.

To solve for the Rayleigh wave speed, one must employ an iterative algorithm. At each iteration, a trial wave
speed 𝑣 is considered, and the Stroh eigenvalue problem is solved to obtain the relevant eigenvectors, as
described above. The traction vectors are extracted and used to form the matrix 𝐁 and calculate its
determinant. This process is repeated until the absolute value of the determinant is minimized. The resulting
wave speed corresponds to the Rayleigh wave speed 𝑐𝑅 .

Following this procedure for the same representative NITI material used in the previous section (𝜌 = 1000 kg/m3,
𝜇 = 1 MPa, 𝐺 = 20 kPa, and setting 𝜆 = 2.37 GPa so that the average longitudinal wave speed is approximately
1540 m/s), we found that the Rayleigh wave speed in a cornea-like NITI solid is approximately √𝐺/𝜌. The exact
value varies slightly with the degree of anisotropy 𝐺/𝜇, from approximately 0.9553√𝐺/𝜌 in the isotropic limit
(𝐺 = 𝜇) to √𝐺/𝜌 in the highly anisotropic limit (𝐺 ≪ 𝜇) (Fig. S3.2).

Supplementary Fig. S3.2. Rayleigh wave speed as a function of anisotropy. The Rayleigh wave speed is primarily
governed by 𝐺 in a NITI solid.

Supplementary Note 4. Analytical solution for guided wave modes in a NITI solid

The cornea can be modeled as an infinite layer of thickness ℎ and density 𝜌 bounded above by air and below by
water (Fig. S4.1a). Assuming a NITI material model, the stiffness tensor (Fig. S4.1b) contains material constants
𝜆, 𝜇, and 𝐺. Because acoustic micro-tapping generates a pseudo-line source, we also assume a plane strain
state.

Supplementary Fig. S4.1. (a) Geometry and boundary conditions and (b) stiffness tensor for a NITI plate
bounded above by air and below by water.

Introducing scales

Position: 𝒙~ℎ

Displacement: 𝒖 ~ ℎ

Time: 𝑡 ~ ℎ/√𝜇/𝜌

Frequency: 𝑓 ~ √𝜇/𝜌/ℎ

Wavenumber: 𝑘 ~ ℎ

the dimensionless equations of motion yield the elastodynamic equations for a NITI material

𝑢𝑡𝑡 = 𝛽 2 𝑢𝑥𝑥 + 𝛼 2 𝑢𝑧𝑧 + 𝛾 2 𝑣𝑥𝑧 (𝑆4.1)

𝑣𝑡𝑡 = 𝛼 2 𝑣𝑥𝑥 + 𝛽 2 𝑣𝑧𝑧 + 𝛾 2 𝑢𝑥𝑧 (𝑆4.2)

𝐺
𝛼2 = (𝑆4.3)
𝜇
𝜆 + 2𝜇
𝛽2 = (𝑆4.4)
𝜇
𝜆+𝐺
𝛾2 = (𝑆4.5)
𝜇
Here, 𝒖 = (𝑢, 𝑣) is the dimensionless displacement field with 𝑥 and 𝑧 components, respectively; 𝛼, 𝛽, and 𝛾 are
dimensionless parameters, and subscripts denote partial differentiation. We assume harmonic solutions for the
displacements of the form

𝑢(𝑥, 𝑧, 𝑡) = 𝐴𝑒 𝑖(𝑘𝑥+ℓ𝑧−𝜔𝑡) (𝑆4.6)

𝑣(𝑥, 𝑧, 𝑡) = 𝐵𝑒 𝑖(𝑘𝑥+ℓ𝑧−𝜔𝑡) (𝑆4.7)

where 𝑘 and ℓ are dimensionless angular wavenumbers, 𝜔 is the dimensionless angular frequency, and 𝐴 and 𝐵
are arbitrary constants.

Substituting these solutions into the governing equations yields the homogeneous linear system

𝛼2 2 𝛾2
𝑞𝛽2 + ℓ 𝑘ℓ
𝛽2 𝛽2 𝐴 0
[ ]=[ ] (𝑆4.8)
𝛾2 𝛽 2 𝐵 0
[ 𝑘ℓ 2 𝑞𝛼2 + 2 ℓ2 ]
𝛼 𝛼
where

𝜔2
𝑞𝛼2 = 𝑘 2 − (𝑆4.9)
𝛼2
𝜔2
𝑞𝛽2 = 𝑘 2 − . (𝑆4.10)
𝛽2

This has non-trivial solutions if and only if the determinant of the matrix is zero, resulting in a biquadratic
equation for ℓ with four solutions

𝛼2 2 𝛽2 2 𝛾 4 𝑘2 2
ℓ4 + [ 𝑞 + 𝑞 − ] ℓ + 𝑞𝛼2 𝑞𝛽2 = 0 (𝑆4.11)
𝛽2 𝛼 𝛼2 𝛽 𝛼2 𝛽2

1
ℓ = ±√ [𝜑 ± √𝜑2 − 4𝑞𝛼2 𝑞𝛽2 ] (𝑆4.12)
2

𝛾 4 𝑘2 𝛼2 2 𝛽2 2
𝜑 = 2 2 − 2 𝑞𝛼 − 2 𝑞𝛽 (𝑆4.13)
𝛼 𝛽 𝛽 𝛼

Without loss of generality, we assume 𝐵 = 1 and solve the corresponding coefficient 𝐴 for each solution ℓ. This
yields the following solutions, where the outer and inner ± signs match for each 𝐴 and ℓ,

𝛾2 𝑘
√2 √𝜑 ± √𝜑2 − 4𝑞𝛼2 𝑞𝛽2
𝛼2
𝐴=± − (𝑆4.14)
𝛽2
𝜑 + 2 2 𝑞𝛽2 ± √𝜑2 − 4𝑞𝛼2 𝑞𝛽2
𝛼
[ ]
The full solutions are therefore linear combinations of four partial waves,
4
𝑢(𝑥, 𝑧, 𝑡) = ∑ 𝐶𝑗 𝐴𝑗 𝑒 𝑖ℓ𝑗𝑧 𝑒 𝑖(𝑘𝑥−𝜔𝑡) (𝑆4.15)
𝑗=1

4
𝑣(𝑥, 𝑧, 𝑡) = ∑ 𝐶𝑗 𝑒 𝑖ℓ𝑗 𝑧 𝑒𝑖(𝑘𝑥−𝜔𝑡) (𝑆4.16)
𝑗=1

In the fluid domain, we assume an acoustic material. The dimensionless acoustic wave equation in velocity
potential form is

𝒖̇ 𝑓 = ∇Φ (S4.17)
 𝜌𝑓
𝑝𝑓 = − Φ (𝑆4.18)
𝜌 𝑡
1
ΔΦ − Φ =0 (𝑆4.19)
𝛿 2 𝑡𝑡
2
𝜌(𝑐 𝑓 )
𝛿2 = (𝑆4.20)
𝜇
The general solution for the fluid domain takes the form

Φ = 𝐶5 𝑒 𝜉𝑧 𝑒 𝑖(𝑘𝑥−𝜔𝑡) (𝑆4.21)

𝜔2
𝜉 = √𝑘 2 − (𝑆4.22)
𝛿2

Re(𝜉) > 0 (𝑆4.23)

The constants 𝐶𝑗 are chosen so that the solutions satisfy the non-dimensionalized boundary conditions:

𝜎𝑥𝑧 = 0 at 𝑧 = 1
𝜎𝑧𝑧 = 0 at 𝑧 = 1
𝜎𝑥𝑧 = 0 at 𝑧 = 0 (𝑆4.24)
𝑓
𝜎𝑧𝑧 = 𝜎𝑧𝑧 at 𝑧 = 0
𝑣̇ = 𝑣̇ 𝑓 at 𝑧 = 0
where the superscripts 𝑓 denote quantities in the fluid domain 𝑧 < 0. Substituting the general solutions into the
boundary conditions yields a 5×5 homogeneous system for the coefficients, 𝐌𝐜 = 𝟎. This system has nontrivial
solutions if and only if the determinant of the matrix 𝐌 (equation S4.25) is zero. For a given frequency 𝜔,
wavenumbers 𝑘 can be found that satisfy the dispersion relation using numerical root-finding methods or by
minimizing the absolute value of the determinant
(S4.25)

(ℓ1 𝐴1 + 𝑘)𝑒 𝑖ℓ1 (ℓ2 𝐴2 + 𝑘)𝑒 𝑖ℓ2 (ℓ3 𝐴3 + 𝑘)𝑒 𝑖ℓ3 (ℓ4 𝐴4 + 𝑘)𝑒 𝑖ℓ4 0
[𝑘(𝛾 2 − 𝛼 2 )𝐴1 + 𝛽 2 ℓ1 ]𝑒 𝑖ℓ1 [𝑘(𝛾 2 − 𝛼 2 )𝐴2 + 𝛽 2 ℓ2 ]𝑒 𝑖ℓ2 [𝑘(𝛾 2 − 𝛼 2 )𝐴3 + 𝛽 2 ℓ3 ]𝑒 𝑖ℓ3 [𝑘(𝛾 2 − 𝛼 2 )𝐴4 + 𝛽 2 ℓ4 ]𝑒 𝑖ℓ4 0
𝐌= ℓ1 𝐴1 + 𝑘 ℓ 2 𝐴2 + 𝑘 ℓ 3 𝐴3 + 𝑘 ℓ 4 𝐴4 + 𝑘 0
𝑘(𝛾 − 𝛼 2 )𝐴1 + 𝛽 2 ℓ1
2 𝑘(𝛾 − 𝛼 2 )𝐴2 + 𝛽 2 ℓ2
2 𝑘(𝛾 − 𝛼 2 )𝐴3 + 𝛽 2 ℓ3
2 𝑘(𝛾 − 𝛼 2 )𝐴4 + 𝛽 2 ℓ4
2 𝜔𝜌 𝑓 /𝜌
[ 𝜔 𝜔 𝜔 𝜔 −𝑖𝜉 ]
In Supplementary Note 2, we noted that our definition of a NITI material was isotropic in its longitudinal
terms, but still can accurately approximate guided wave behavior in TI materials that are weakly
anisotropic in their longitudinal terms.29 The NITI dispersion relation S4.25 is a special case of the
general dispersion relation for a TI solid, which may be solved in a similar manner as presented in this
section. Using values for the stiffness coefficients from a recent Brillouin microscopy study29 of the
cornea, we take 𝐶11 = 2.971 GPa, 𝐶33 = 2.662 GPa, and 𝐶13 + 2𝐺 = 2.680 GPa, along with what we
consider to be representative values of 𝐺 and 𝜇 for cornea (𝐺 = 20 kPa and 𝜇 = 1 MPa), and solve both
the general and NITI dispersion relations. For all calculations, we assumed a layer thickness of 0.5 mm.
We compare the A0 mode for these two TI models and an isotropic model (𝐺 = 𝜇 = 20 kPa) to obtain the
following:

Supplementary Fig. S4.2. Comparison of the A0 mode phase velocity spectra for a TI material with weak
longitudinal anisotropy, NITI model, and an isotropic model. Both the NITI and TI models differ greatly
from the isotropic model. Anisotropy in the longitudinal terms of the TI stiffness tensor is a second-
order effect, and the NITI model captures most of the behavior, suggesting that the shear moduli 𝜇 and
𝐺 dominate guided wave behavior.

The change from isotropic to a NITI model is rather large, while the change from NITI to a general TI
model is very small. This shows that variations in 𝐶13 and 𝐶33 are second-order effects when examining
guided shear waves in nearly incompressible TI materials. We argue that it is reasonable to neglect these
terms, replacing them with the NITI model.

Supplementary Note 5. Finite element model of a NITI cornea

To provide an ideal, noise-free comparison to our experimental measurements, we developed a two-

dimensional (plane strain) finite element model of guided wave propagation in a NITI material using
OnScale (OnScale, Redwood City, CA). The model mirrors the configuration of our OCE experiments in
porcine cornea. A thin elastic layer of thickness ℎ = 0.55 mm and density 𝜌 = 1000 kg/m3 is bounded
above by air (free surface condition) and below by a layer of water (Fig. S5a). The elastic layer is
modeled as a NITI material with parameters 𝜆, 𝜇, and 𝐺, while the water layer is modeled as an isotropic
solid with a density 𝜌 𝑓 = 1000 kg/m3, shear wave speed 𝑐𝑠 = 0, and longitudinal wave speed 𝑐𝐿 =
√(𝜆 + 2𝜇)/𝜌.

Supplementary Fig. S5. (a) Finite element model geometry used to simulate guided wave propagation in
a NITI model of the cornea. (b) Acoustic intensity field measured from the air-coupled AμT transducer
shows a nearly Gaussian profile (c) when sampled along a 45° line through the transducer focus.

The bottom and right boundaries of the domain function as absorbing conditions to minimize reflections
as elastic waves leave the computational domain. To ensure stability at the boundary, the water layer
extends along the right edge of the tissue domain and maintains the boundary condition. Symmetry is
assumed with respect to the 𝑦𝑧-plane. A temporally- and spatially-varying pressure load 𝑃(𝑥, 𝑡) is
applied to the top boundary to mimic the AμT push of OCE experiments. The spatial profile is modeled
as a Gaussian with a full-width-at-half-max (FWHM) 𝑑 = 600 μm. This value was obtained by measuring
the ultrasonic field of the AμT transducer with a needle hydrophone (HNC-1000, Onda, Sunnyvale, CA,
USA) in air, sampling along a 45° line passing through the transducer focus (Fig. S5b and S5c). The
temporal push profile is modeled by a super-Gaussian function with FWHM 𝑇 = 100 μs. A small offset 𝑡0
is included to avoid impulsive loading at 𝑡 = 0. The full form of the pressure load is given by

𝑥 2 𝑡 − 𝑡0 4
𝑃(𝑥, 𝑡) = 𝑃0 exp [−4(ln 2) ( ) ] exp [−16(ln 2) ( ) ], (𝑆5.1)
𝑑 𝑇

where the pressure amplitude 𝑃0 is an arbitrary constant (5 kPa in this study).

Biological tissue is nearly incompressible, exhibiting longitudinal wave speeds on the order of 1540 m/s
and shear wave speeds in the range of 1-10 m/s. This corresponds to a Poisson’s ratio of 𝜈 ≈ 0.4999995.
In practice, it is not necessary to directly model this exact value of Poisson’s ratio, and solutions for 𝜈 >
0.4995 converge to the incompressible solution.35 In this study, we enforce this constraint for all models
by requiring that the ratio √(𝜆 + 2𝜇)/𝜇 = 35. In the isotropic limit 𝐺 = 𝜇, this gives a Poisson’s ratio of
𝜈 = 0.4996.

The computational domain was discretized using linear finite elements on a regular rectangular grid with
a minimum of 40 elements per elastic wavelength, as approximated by the push width. Belytchko-
Bindeman hourglass suppression was applied to prevent spurious modes from corrupting the solution.
All simulations used an explicit time-stepping method to generate the full displacement and velocity
fields at each space and time point. However, only the vertical velocity component was used in our
analysis. This is analogous to OCE experiments where only this component is available.

Supplementary Note 6. Finite element model to investigate the effect of corneal curvature

In the analytical solution described in Supplementary Note 4 and the finite element model described in
Supplementary Note 5, we model the cornea as a flat NITI layer bounded above by air and below by
water. In reality, the cornea is not flat, but rather curved, and this curvature could affect guided wave
behavior. In particular, the dispersion relation of guided waves in a curved plate varies from the flat
plate case based on the relationship between the wavelength, the radius of curvature, and the layer
thickness. Krauklis and Molotkov 44 derived dispersion relations for azimuthal modes in a cylindrical shell
and meridian modes in a spherical shell and classified dispersion behavior in two regimes:
𝜆 ℎ
(1) > (𝑆6.1)
𝑅 𝜆
𝜆 ℎ
(2) <
𝑅 𝜆
When condition (1) holds, dispersion is strongly influenced by the curvature of the shell, and dispersion
is stronger for spherical modes than cylindrical modes. Condition (2) corresponds to mild curvature, and
the dispersion is mainly governed by the terms obtained for a flat plate.

The expected elastic wavelength is determined by some combination of the product 𝑐𝑠 𝑇 and 𝑑, where
𝑐𝑠 is the bulk shear wave speed of the medium, 𝑇 is the excitation duration, and 𝑑 is the excitation
width. Generally, when the ratio 𝑐𝑠 𝑇/𝑑 is small, pressure confinement is achieved and the expected
wavelength is on the order of 𝑑. In contrast, when the ratio is large, the wavelength is on the order of
𝑐𝑠 𝑇. For a typical acoustic micro-tapping excitation, we have 𝑇 = 100 μs and 𝑑 = 500 μm. For 𝑐𝑠 = 4 m/s,
this gives 𝑐𝑠 𝑇/𝑑 = 0.8, and the wavelength 𝜆 is on the order of 0.4-0.5 mm. Combining these estimates
(and assuming 𝑅 = 6.5 mm, ℎ = 0.5 mm), we obtain 𝜆/𝑅 ≈ 0.08 and ℎ/𝜆 ≈ 1. Thus, we expect condition
(2) to hold, and curvature can be neglected.

For additional validation, we developed a finite element model in OnScale to estimate the effect of
curvature on the dispersion behavior. This model was similar to that described in Supplementary Note 5,
but featured a few distinct differences in both geometry and post-processing.
Supplementary Fig. S6.1. Curved geometry used to study guided waves in a spherically curved layer. The
full model geometry can be considered a solid of revolution obtained by revolving the domain shown
about the axisymmetric boundary.

The finite element model consisted of a rectangular domain of thickness ℎ = 0.55 mm bent to an outer
radius of curvature of 6.5 mm (Supplementary Fig. S6.1). Here we assume an axisymmetric rather than a
plane strain state. The tissue layer was assumed to be isotropic with density 𝜌 = 1000 kg/m3, shear wave
speed 𝑐𝑆 = 4 m/s, and longitudinal wave speed 𝑐𝐿 = 35𝑐𝑆 . The tissue layer sat atop a simulated water
layer, modeled as an elastic solid with shear wave speed 𝑐𝑆 = 0 m/s and a density and longitudinal wave
speed matched to the tissue layer. Outer edges of the domain were set as absorbing boundaries. The
spatio-temporal push profile was defined as in Supplemental Note 5, with a Gaussian profile in space
and a super-Gaussian profile in time. The domain was discretized with linear finite elements on a regular
conformal grid with at least 40 elements per elastic wavelength.

Solving this model produces a pseudo-point source solution, a numerical approximation to the Green’s
function for this problem. The wave field solution for a line source, such as the one used in our AμT
experiments, can be obtained by convolving the Green’s function with a source distribution. We
numerically evaluated this convolution integral along a linear path on the spherical surface using
trapezoid integration. This yielded the wave field along the entire spherical surface. Supplemental Video
1 shows the results. In general, the wave field is more complex compared to the flat plate plane strain
approximation. However, extracting the wave field from the mid-plane normal to the line source (as
would be measured using OCE), we find that the flat plate and spherical models produce very similar
wave fields (Supplementary Fig. S6.2, a-b). Analyzing the 2D Fourier spectrum of the spherical model
along this mid-plane, we find that the isotropic dispersion relation for a flat plate42 provides a close fit to
the numerical solution (Supplementary Fig. S6.2, c). This is expected, as the radius of curvature of the
cornea is large relative to its thickness.44,45 Thus, it appears reasonable to neglect corneal curvature in
analyzing guided wave behavior in the cornea. While these computations were performed using an
isotropic model, we expect a NITI model to produce similar results.
Supplementary Fig. S6.2. Surface vertical velocity wave fields for a simulated (a) flat plate and (b)
spherically curved layer are very similar. (c) The 2D Fourier spectrum for the spherically curved layer
closely matches the theoretical dispersion curves obtained for a flat plate (yellow markers).

Supplementary Note 7. Fitting experimental data with the NITI model

Quantitative moduli estimates in bounded materials require a method to determine the dispersion
relation most closely matching observed guided wave modes. In a typical dynamic OCE experiment, the
OCT signal phase difference is used to reconstruct the vertical component of the velocity field. We
extract the surface part of this signal to create an XT representation of the wave field and apply a two-
dimensional Fourier transform to produce a frequency-wavenumber (FK) spectrum. This spectrum can
then be compared to a theoretical dispersion relation to obtain modulus estimates.

Our study differs from many dynamic OCE studies in how we analyze the FK spectrum. A common
approach is to algorithmically identify the local peaks of the FK spectrum. They ideally trace the central
portions of the modes and can, therefore, be used to produce “measured” dispersion curves to compare
against theoretical solutions. Typically, these peaks are first converted to the frequency-phase velocity
(FC) domain using the relationship between the phase velocity 𝑐 (in m/s), frequency 𝑓 (in Hz), and
wavenumber 𝑘 (in m-1): 𝑐 = 𝑓/𝑘.

In Figure S7.1, we reproduce FK maps from Figure 4 for PVA phantoms and porcine cornea to illustrate
that this process is not without error. Applying a peak-finding algorithm to the FK spectra (Figures S7.1a,
S7.1d) yields the narrow dispersion curves shown in Figures S7.1b and S7.1e. They are then converted to
phase velocities (Figures S7.1c and S7.1f). The peak-finding algorithm requires tuning various
parameters such as minimum peak heights, minimum peak distances, and minimum peak prominence.
Even with well-chosen parameters, spurious points may be chosen that are not related to the mode (for
example, Figures S7.1e-f). Compared to the full FK spectra, this representation ignores the fact that
these points contain much less energy than the points covered by the best-fit dispersion curves. These
points must either be manually cleaned from the data prior to fitting dispersion curves or weighted
carefully to ensure that they do not introduce fit errors (Note that the fits shown in Figure S7.1 are
based on our FK-domain method and do not use spectral peaks). Another technical challenge with
spectral peaks is that more than one peak may be returned for each frequency. This follows physically
from the dispersion relation, where multiple modes can exist at any given frequency. However, spurious
peaks make it difficult to algorithmically sort modes from noise.
Supplementary Figure S7.1. Typical method of dynamic OCE dispersion analysis shown for PVA phantom
(a-c) and porcine cornea (d-f) data, presented in Figure 4. (a,d) The FK spectra are computed from the
surface XT fields, shown here over a 20 dB dynamic range. (b,e) The local peaks of the FK spectra are
identified algorithmically (white circles). (c,f) These peaks are then converted to the frequency-phase
velocity representation and compared with theoretical solutions to the dispersion relation (yellow and
pink lines).

To avoid these issues, we developed a fitting routine based on the full FK spectra. It finds the dispersion
curve that overlaps with the maximum amount of spectral power. Supplementary equation S4.25 acted
as the forward model for the dispersion curves. A number of physical parameters were considered fixed,
including the corneal density (ρ = 1000 kg/m3), corneal longitudinal wave speed (cL = 1540 m/s), and
mean corneal thickness (measured from B-mode OCT images). The cornea was bounded from below by
water with a density of 1000 kg/m3 and longitudinal wave speed of 1480 m/s. Because we did not
observe the S0 mode in corneal measurements, we extracted only the A0 mode from the forward model
using a mode-tracing routine (similar to Pavlakovic et. al 46).

Fitting the theoretical dispersion relation was performed by maximizing the following objective function:
1 𝜇
Φ(𝜇, 𝐺) = ∑ ∑ 𝑤(𝑓, 𝑘; 𝜇, 𝐺)|𝑣̂(𝑓, 𝑘)|2 − 𝛽 | | (𝑆7.1)
𝑁𝑓 𝜆
𝑓 𝑘

Here, 𝑣̂ is the normalized 2D Fourier spectrum of the measured surface velocity data. The function
𝑤(𝑓, 𝑘; 𝜇, 𝐺) is related to the A0 mode solution for a NITI or isotropic (𝐺 = 𝜇) material. The first term
describes the energy covered by the dispersion curve, normalized by the number of FFT bins included in
the fit (𝑁𝑓 ). The second acts as a regularization term that ensures that the ratio 𝜇/𝜆 remains small, thus
satisfying the nearly-incompressible assumption. In equation S7.1, 𝛽 was set to 1, based on an L-curve
analysis.47 The value of 𝜆 was updated at each iteration according to 𝜆 = 𝜌𝑐𝐿2 − 2𝜇. At a given frequency
𝑓 and given the current parameter set (𝜇, 𝐺), the dispersion relation solver returns the wavenumber
associated with the A0 mode, 𝑘0 . Let 𝛥𝑘 be the sampling interval in the wavenumber. The weight
function 𝑤 is then defined as
2
1 𝑘−𝑘0 (𝑓,𝜇,𝐺)
− ( )
𝑒 2 1.2𝛥𝑘 , |𝑘 − 𝑘0 (𝑓, 𝜇, 𝐺)| ≤ 3𝛥𝑘
𝑤(𝑓, 𝑘; 𝜇, 𝐺) = { (S7.2)
0, otherwise
That is, the weights are assigned at each frequency with a peak value at the wavenumber of the
theoretical dispersion curve and weights decaying (with a Gaussian distribution) as the distance from
the theoretical curve increases.

Note that this optimization is not posed as a least-squares error problem. Rather, we seek to maximize
the spectral energy contained in a small weighted window around the A 0 mode. The dispersion relation
estimates the location of the A0 mode (e.g. the wavenumber of the mode for a given frequency), but
cannot predict the spectral energy in a frequency-wavenumber bin (which depends on the method used
to excite guided waves).

To compare the isotropic and NITI fits in this study, we defined a goodness-of-fit metric based on an
“unconstrained, global optimum” for the objective function, Φmax. At each frequency 𝑓, we solve the
following optimization problem:

̃(𝑓, 𝑘; 𝑘̃ , 𝜇, 𝐺)|𝑣̂(𝑓, 𝑘)|2

𝑘max = argmax ∑ 𝑤 (𝑆7.3)
̃
𝑘 𝑘

1 𝑘−𝑘 ̃ 2
−2(1.2𝛥𝑘)
𝑒 , |𝑘 − 𝑘̃ | ≤ 3𝛥𝑘
̃(𝑓, 𝑘; 𝑘̃ , 𝜇, 𝐺) = {
𝑤 (𝑆7.4)
0, otherwise
1
Φmax (𝜇, 𝐺) = ̃(𝑓, 𝑘; 𝑘max , 𝜇, 𝐺)|𝑣̂(𝑓, 𝑘)|2
∑∑𝑤 (𝑆7.5)
𝑁𝑓
𝑓 𝑘

This is equivalent to finding the Gaussian-weighted window containing the most spectral energy at each
frequency independently and summing those contributions to the total mode energy. Since 𝑘max may
vary at each frequency independently from the dispersion relation, 𝑘max may not follow the mode
shape exactly and may not even be smooth. However, this means that Φmax represents an upper bound
on the value of Φ calculated during fitting. We therefore compare the following goodness-of-fit
measures:
Φiso
𝑔iso = (𝑆7.6)
Φmax
ΦNITI
𝑔NITI = (𝑆7.7)
Φmax

Thus, 𝑔NITI and 𝑔𝑖𝑠𝑜 indicate what portion of the maximum possible mode energy is captured by a given
A0 dispersion curve for NITI and isotropic models, respectively, with values near 1 indicating that the
theoretical dispersion curve captures almost all of the mode’s energy.
References

1. Anderson, K., El-Sheikh, A. & Newson, T. Application of structural analysis to the mechanical
behaviour of the cornea. J. R. Soc. Interface 1, 3–15 (2004).
2. Bekesi, N., De La Hoz, A., Kling, S. & Marcos, S. The Effects of Cross-linking on the Static and
Dynamic Corneal Viscoelastic Properties. Assoc. Res. Vis. Ophthalmol. 520461 (2014).
3. Bekesi, N., Dorronsoro, C., De La Hoz, A. & Marcos, S. Material properties from air puff corneal
deformation by numerical simulations on model corneas. PLoS One 11, (2016).
4. Boschetti, F., Triacca, V., Spinelli, L. & Pandolfi, A. Mechanical characterization of porcine
corneas. J. Biomech. Eng. 134, (2012).
5. Elsheikh, A., Alhasso, D. & Rama, P. Biomechanical properties of human and porcine corneas.
Exp. Eye Res. 86, 783–790 (2008).
6. Elsheikh, A. & Alhasso, D. Mechanical anisotropy of porcine cornea and correlation with stromal
microstructure. Exp. Eye Res. 88, 1084–1091 (2009).
7. Han, Z. et al. Quantitative assessment of corneal viscoelasticity using optical coherence
elastography and a modified Rayleigh – Lamb equation of corneal viscoelasticity equation. J.
Biomed. Opt. 20, 2–5 (2015).
8. Han, Z. et al. Optical coherence elastography assessment of corneal viscoelasticity with a
modified Rayleigh-Lamb wave model. J. Mech. Behav. Biomed. Mater. 66, 87–94 (2017).
9. Hatami-Marbini, H. & Etebu, E. An experimental and theoretical analysis of unconfined
compression of corneal stroma. J. Biomech. 46, 1752–1758 (2013).
10. Hatami-Marbini, H. & Etebu, E. Hydration dependent biomechanical properties of the corneal
stroma. Exp. Eye Res. 116, 47–54 (2013).
11. Hatami-Marbini, H. Viscoelastic shear properties of the corneal stroma. J. Biomech. 47, 723–728
(2014).
12. Kling, S., Ginis, H. & Marcos, S. Corneal biomechanical properties from two-dimensional corneal
flap extensiometry: Application to UV-Riboflavin cross-linking. Investig. Ophthalmol. Vis. Sci. 53,
5010–5015 (2012).
13. Kling, S., Bekesi, N., Dorronsoro, C., Pascual, D. & Marcos, S. Corneal viscoelastic properties from
finite-element analysis of in vivo air-puff deformation. PLoS One 9, (2014).
14. Singh, M. et al. Noncontact Elastic Wave Imaging Optical Coherence Elastography for Evaluating
Changes in Corneal Elasticity Due to Crosslinking. IEEE J. Sel. Top. Quantum Electron. 22, 1–32
(2017).
15. Wollensak, G., Spoerl, E. & Seiler, T. Stress-strain measurements of human and porcine corneas
after riboflavin-ultraviolet-A-induced cross-linking. J. Cataract Refract. Surg. 29, 1780–1785
(2003).
16. Zeng, Y., Yang, J., Huang, K., Lee, Z. & Lee, X. A comparison of biomechanical properties between
human and porcine cornea. J. Biomech. 34, 533–537 (2001).
17. Hoeltzel, D., Altman, P., Buzard, K. & Choe, K. Strip extensiometry for comparison of the
 mechanical response of bovine, rabbit, and human corneas. J. Biomech. Eng. 114, 202-215.
(1992).
18. Kling, S. et al. Numerical model of optical coherence tomographic vibrography imaging to
estimate corneal biomechanical properties. J. R. Soc. Interface 11, 20140920–20140920 (2014).
19. Petsche, S. J., Chernyak, D., Martiz, J., Levenston, M. E. & Pinsky, P. M. Depth-Dependent
Transverse Shear Properties of the Human Corneal Stroma. 53, (2017).
20. Sloan, S. R., Khalifa, Y. M. & Buckley, M. R. The location- and depth-dependent mechanical
response of the human cornea under shear loading. Investig. Ophthalmol. Vis. Sci. 55, 7919–7924
(2014).
21. Bryant, M., Szerenyi, K., Schmotzer, H. & McFonnell, P. Corneal tensile strength in fully healed
radial keratotomy wounds. in Invest Ophthalmol Vis Sc 35(7):3022–31 (1994).
22. Elsheikh, A., Wang, D. & Pye, D. Determination of the modulus of elasticity of the human cornea.
J. Refract. Surg. 23, 808–818 (2007).
23. Litwiller, D. V. et al. MR elastography of the ex vivo bovine globe. J. Magn. Reson. Imaging 32,
44–51 (2010).
24. Bekesi, N., Kochevar, I. E. & Marcos, S. Corneal biomechanical response following collagen cross-
linking with Rose Bengal-green light and riboflavin-UVA. Investig. Ophthalmol. Vis. Sci. 57, 992–
1001 (2016).
25. Singh, M. et al. Quantifying the effects of hydration on corneal stiffness with noncontact optical
coherence elastography. J. Cataract Refract. Surg. 44, 1023–1031 (2018).
26. Petsche, S. J., Chernyak, D., Martiz, J., Levenston, M. E. & Pinsky, P. M. Depth-dependent
transverse shear properties of the human corneal stroma. Investig. Ophthalmol. Vis. Sci. 53, 873–
880 (2012).
27. Singh, M. et al. Assessing the effects of riboflavin/UV-A crosslinking on porcine corneal
mechanical anisotropy with optical coherence elastography. Biomed. Opt. Express 8, 349–366
(2017).
28. O’Donnell, M. & Skovoroda, A. R. Prospects for elasticity reconstruction in the heart. IEEE Trans.
Ultrason. Ferroelectr. Freq. Control 51, 322–328 (2004).
29. Eltony, A. M., Shao, P., & Yun, S.-H. Measuring mechanical anisotropy of the cornea with Brillouin
microscopy. arXiv:2003.04344v1 (2020)
30. Ambroziński, Ł. et al. Acoustic micro-tapping for non-contact 4D imaging of tissue elasticity. Sci.
Rep. 6, 38967 (2016).
31. Nguyen, T. D. & Boyce, B. L. An inverse finite element method for determining the anisotropic
properties of the cornea. Biomech. Model. Mechanobiol. 10, 323–337 (2011).
32. Li, J. et al. Revealing anisotropic properties of cornea at different intraocular pressures using
optical coherence elastography. Opt. Elastography Tissue Biomech. III 9710, 97100T (2016).
33. Singh, M. et al. Investigating Elastic Anisotropy of the Porcine Cornea as a Function of Intraocular
Pressure With Optical Coherence Elastography. J. Refract. Surg. 32, 562–567 (2016).
34. Timoshenko, S. & Woinowsky-Krieger, S. Theory of plates and shells. 2nd ed. McGraw-Hill, New
York. (1959).
35. Pelivanov, I. et al. Does group velocity always reflect elastic modulus in shear wave elastography?
J. Biomed. Opt. 24, 1 (2019).
36. Barnett, D. M. Synthesis of the sextic and the integral formalism for dislocations, Green’s
function and surface waves in anisotropic elastic solids." Phys. Norv 13, (1973).
37. Chadwick, P. & Smith, G. Foundations of the Theory of Surface Waves in Anisotropic Elastic
Materials. In: Advances in Applied Mechanics 17, 303–376 (1977).
38. Ting, T. C. T. The Stroh Formalism. In: Anisotropic Elasticity, Theory and Applications. 155–184
(1996).

39. Tanuma, K. Stroh Formalism and Rayleigh Waves. J Elasticity 89, 5–154 (2007).
40. Cherry, M. R., Sathish, S. & Grandhi, R. A numerical method for predicting Rayleigh surface wave
velocity in anisotropic crystals. J. Comput. Phys. 351, 108–120 (2017).
41. Payton, R. G. Elastic wave propagation in transversely isotropic media. Martinus Nijhoff
Publishers, The Hague, Netherlands (1983).
42. Kirby, M. A. et al. Optical coherence elastography in ophthalmology. J. Biomed. Opt. 22, 1 (2017).
43. Auld, B. Acoustic fields and waves in solids. R.E. Krieger., Malabar, Florida (1990).
44. Krauklis, P. V. & Molotkov, L. A. Low-frequency lamb waves in cylindrical and spherical layers in
an elastic medium. J. Soviet Mathematics 3, 82–90, (1975).
45. Brekhovskikh, L. M. Waves in Layered Media. Academic Press, New York (1976).
46. Pavlakovic, B., Lowe, M., Alleyne, D. & Cawley, P. Disperse: A General Purpose Program for
Creating Dispersion Curves. In: Review of Progress in Quantitative Nondestructive Evaluation.
185–192 (1997).
47. Hansen, P. C. Analysis of discrete ill-posed problems by means of the L-curve. SIAM Review,
34(4), 561–580 (1992).