Lecture

13: Strain part 2

GEOS 655 Tectonic Geodesy
Jeff Freymueller
 Strain and RotaAon Tensors
•  We described the deformaAon as the sum of two
tensors, a strain tensor and a rotaAon tensor.

1 ⎛ ∂ui ∂u j ⎞ 1 ⎛ ∂ui ∂u j ⎞
( ) ( )
ui x 0 + dx = ui x 0 + ⎜⎜ + ⎟⎟ dx j + ⎜⎜
2 ⎝ ∂x j ∂ x i ⎠
− ⎟⎟ dx j
2 ⎝ ∂x j ∂ x i ⎠

( ) ( )
ui x 0 + dx = ui x 0 + εij dx j + ω ij dx j

⎡ ∂u1 ⎡ ∂u1 1 ⎛ ∂u1 ∂u2 ⎞ 1 ⎛ ∂u1 ∂u3 ⎞⎤ ⎡ 1 ⎛ ∂u1 ∂u2 ⎞ 1 ⎛ ∂u1 ∂u3 ⎞⎤
∂u1 ∂u1 ⎤ ⎢ ⎜ + ⎟ ⎜ + ⎟⎥ ⎢ 0 ⎜ − ⎟ ⎜ − ⎟⎥
⎢ ⎥ ∂x1 2 ⎝ ∂x 2 ∂x1 ⎠ 2 ⎝ ∂x 3 ∂x1 ⎠⎥ ⎢ 2 ⎝ ∂x 2 ∂x1 ⎠ 2 ⎝ ∂x 3 ∂x1 ⎠⎥
x1
⎢ ∂€ ∂x 2 ∂x 3 ⎥ ⎢
⎢∂u2 ∂u2 ∂u2 ⎥ ⎢ 1 ⎛ ∂u1 ∂u2 ⎞ ∂u2 1 ⎛ ∂u2 ∂u3 ⎞⎥ ⎢1 ⎛ ∂u2 ∂u1 ⎞ 1 ⎛ ∂u2 ∂u3 ⎞⎥
=⎢ ⎜ + ⎟ ⎜ + ⎟⎥ + ⎢ ⎜ − ⎟ 0 ⎜ − ⎟⎥
⎢ ∂x1 ∂x 2 ∂x 3 ⎥ ⎢ 2 ⎝ ∂x 2 ∂x1 ⎠ ∂x 2 2 ⎝ ∂x 3 ∂x 2 ⎠⎥ ⎢2 ⎝ ∂x1 ∂x 2 ⎠ 2 ⎝ ∂x 3 ∂x 2 ⎠⎥
⎢∂u ∂u3 ∂u3 ⎥ ⎢ 1 ⎛ ∂u ∂u ⎞
⎢ 3 ⎥ 1 3 1 ⎛ ∂u2 ∂u3 ⎞ ∂u3 ⎥ ⎢1 ⎛ ∂u3 ∂u1 ⎞ 1 ⎛ ∂u3 ∂u2 ⎞ ⎥
∂x 3 ⎦ ⎢ 2 ⎜⎝ ∂x + ∂x ⎟⎠ ⎜ + ⎟ ⎥ ⎢2 ⎜⎝ ∂x − ∂x ⎟⎠ ⎜ − ⎟ 0
⎣ ∂x1 ∂x 2
⎣ 3 1 2 ⎝ ∂x 3 ∂x 2 ⎠ ∂x 3 ⎦ ⎣ 1 3 2 ⎝ ∂ x 2 ∂x 3 ⎠ ⎥
⎦
symmetric, strain anti-symmetric, rotation

€
 OK, So What is a Tensor, Anyway
•  Tensor, not tonsure! è
•  Examples of tensors of various ranks:
–  Rank 0: scalar
–  Rank 1: vector
–  Rank 2: matrix
–  A tensor of rank N+1 is like a set of tensors of rank N, like
you can think of a matrix as a set of column vectors
•  The stress tensor was actually the first tensor -- the
mathemaAcs was developed to deal with stress.
•  The mathemaAcal definiAon is based on
transformaAon properAes.
And WriAng Them in Index NotaAon
•  We can write the strain tensor
⎡ε11 ε12 ε13 ⎤
⎛
1 ∂u ∂u ⎞ ⎢ ⎥
εij = ⎜⎜ i + j ⎟⎟ = ⎢ε21 ε22 ε23 ⎥
2 ⎝ ∂x j ∂x i ⎠
⎢⎣ε31 ε32 ε33 ⎥⎦

•  The strain tensor is symmetric, so only 6 components

are independent. Symmetry requires that:
€
ε21 = ε12
ε31 = ε13
ε23 = ε32
•  The rotaAon tensor is anA-symmetric and has only 3
independent components: ⎛
1 ∂u ∂u ⎞
⎡ 0
⎢
ω12 ω13 ⎤
⎥
€ ω ij = ⎜⎜ i − j ⎟⎟ = ⎢−ω12 0 ω 23 ⎥
2 ⎝ ∂x j ∂x i ⎠
⎢⎣−ω13 −ω 23 0 ⎥⎦
 EsAmaAng Strain and RotaAon
from GPS Data
•  This is preYy easy. We can esAmate all of the components of
the strain and rotaAon tensors directly from the GPS data.
–  We can write equaAons in terms of the 6 independent strain tensor
components and 3 independent rotaAon tensor components
–  Or we can write equaAons in terms of the 9 components of the
displacement gradient tensor
•  Write the moAons (relaAve to a reference site or reference
locaAon) in terms of distance from reference site:

( ) ( )
ui x 0 + dx − ui x 0 = εij dx j + ω ij dx j
•  Here x0 is the posiAon of the reference locaAon, and dx is the
vector from reference locaAon to where we have data

€
 The EquaAons in 2D
•  The equaAons below are wriYen out for the
2D case (velocity and strain/rotaAon rate):
v x = Vx + ε˙xxΔx + ε˙xy Δy + ω˙ Δy
v y = Vy + ε˙xyΔx + ε˙yy Δy − ω˙ Δx
–  For 2D strain, we have 4 parameters (3 strain
rates, 1 rotaAon rate) and we need ≥2 sites with
€horizontal data.
–  For 3D, we would have 9 parameters and would
require ≥3 sites with 3D data.
VelociAes RelaAve to Eurasia

Chen et al. (2004), JGR

 Displacement and Strain
•  Displacements (or rates) are a combinaAon of
rigid body translaAon, rotaAon and internal
deformaAon

⎡ v east ⎤ ⎡v e,body ⎤ ⎡ ε˙11 1

2 (ε˙12 + ω˙ )⎤⎡x⎤
⎢ ⎥=⎢ ⎥ + ⎢1 ⎥⎢ ⎥
⎣v north ⎦ ⎣v n,body ⎦ ⎣ 2 (ε˙12 − ω˙ ) ε˙22 ⎦⎣y⎦
ε = strain tensor components (x, y) = position
ω = rotation v = velocity
Uniform Strain Rate
ε! 0 −1 0 −1 0
1
= ε! + ε!
0 ε! 0 1 0 0
2 2

2
 Deforming Block Model
•  Based on GPS data from 44 sites
•  Four blocks moving relaAve to each other
on major faults, plus uniform strain
–  Block moAons are predominantly strike-slip
•  Models with spaAal variaAons in strain do
not fit significantly beYer than uniform
strain
•  Models with all slip concentrated on a few
faults fit worse than the deforming block
model.
 4.4+/-1.1

7.4+/-0.7

5.9+/-0.7
Chen et al. Deforming Block Model
 RotaAon Tensor
⎡ 1 ⎛ ∂u1 ∂u2 ⎞ 1 ⎛ ∂u1 ∂u3 ⎞⎤
⎢ 0 ⎜ − ⎟ ⎜ − ⎟⎥
⎢ 2 ⎝ ∂x 2 ∂x1 ⎠ 2 ⎝ ∂x 3 ∂x1 ⎠⎥
⎢1 ⎛ ∂u2 ∂u1 ⎞ 1 ⎛ ∂u2 ∂u3 ⎞⎥
⎢ ⎜ − ⎟ 0 ⎜ − ⎟⎥
⎢2 ⎝ ∂x1 ∂x 2 ⎠ 2 ⎝ ∂x 3 ∂x 2 ⎠⎥
⎢1 ⎛ ∂u3 ∂u1 ⎞ 1 ⎛ ∂u3 ∂u2 ⎞ ⎥
⎢2 ⎜⎝ ∂x − ∂x ⎟⎠ ⎜ − ⎟ 0 ⎥
⎣ 1 3 2 ⎝ ∂x 2 ∂ x 3 ⎠ ⎦
⎡ 1 ⎛ ∂u1 ∂u2 ⎞ 1 ⎛ ∂u1 ∂u3 ⎞⎤
⎢ 0 ⎜ − ⎟ ⎜ − ⎟⎥
⎢ 2 ⎝ ∂x 2 ∂x1 ⎠ 2 ⎝ ∂x 3 ∂x1 ⎠⎥
In terms of our earlier ⎢1 ⎛ ∂u2 ∂u1 ⎞ 1 ⎛ ∂u2 ∂u3 ⎞⎥
€ ⎢ ⎜
2 x
−
x
⎟ 0 ⎜ − ⎟⎥ = (β ij ) x − x
2 ⎝ ∂x 3 ∂x 2 ⎠⎥
coordinate rotation matrix: ⎢ ⎝ ∂ 1 ∂ 2 ⎠
⎢1 ⎛ ∂u3 ∂u1 ⎞ 1 ⎛ ∂u3 ∂u2 ⎞ ⎥
⎢2 ⎜⎝ ∂x − ∂x ⎟⎠ ⎜ − ⎟ 0 ⎥
⎣ 1 3 2 ⎝ ∂x 2 ∂ x 3 ⎠ ⎦
 RotaAon as a Vector
•  We can represent the rotaAon tensor as a vector.
Why?
–  RotaAon can be described by a vector; think angular
velocity vector
–  With only three independent components, the number of
terms adds up
•  We can write a cross product operator using the
“permutaAon tensor” (actually a tensor of rank 3,
like a matrix with 3 dimensions): Ωk = − 12 eijkωij
⎧ e = e = e =1 ⎫
⎪⎪ 123 231 312 ⎪⎪
eijk = ⎨ e213 = e321 = e132 = −1 ⎬
⎪ ⎪
⎪⎩ otherwise := 0 ⎪⎭
 More on the PermutaAon Tensor
•  We can write the vector cross product in terms of this
permutaAon tensor

(a × b) k = eijk aib j
•  Remember this rule?

–  The permutaAon tensor is like a short-hand for that rule

€ •  The permutaAon tensor can be wriYen in terms of the
Kronecker delta (this is someAmes useful):

eijk eist = δ jsδkt − δ jtδks

 Ω and ωij
•  We can write Ω in terms of u:
Ωk = − 12 eijkω ij = − 12 ekijω ij
⎛ ∂ u ∂ u ⎞
i j
Ωk = − 12 ⎜⎜ 12 ekij − 12 ekij ⎟⎟
⎝ ∂x j ∂x i ⎠
⎛ ∂ui 1 ∂u j ⎞
1 1
Ωk = 2 ⎜⎜ 2 ekji + 2 ekij ⎟⎟ Notice the permutation
⎝ ∂x j ∂x i ⎠
⎛ ∂u ⎞
Ωk = 2 ⎜⎜ekji i ⎟⎟
1
These two terms are equal. Why?
⎝ ∂x j ⎠
Ω= 1
2 (∇ × u) ∂ ˆ ∂ ˆ ∂ ˆ
It is the curl of u ∇= i+ j+ k
∂x ∂y ∂z

€
 ωij and Ω
•  We can turn this around and write ωij in terms of Ω. Start
with:

Ωk = − 12 eijkω ij emnkΩk = − 12 emnkeijkω ij

•  MulAply both sides by emnk emnkΩk = − 12 (δmiδnj − δmjδni )ω ij
emnkΩk = − 12 (δmiδnjω ij − δmjδniω ij )
emnkΩk = − 12 (ω mn − ω nm )
emnkΩk = − 12 (ω mn + ω mn )
emnkΩk = −ω mn
ω ij = −eijk Ωk
 One Final ManipulaAon
•  Let’s go back to the original term in the Taylor Series
expansion:

u(rot
i
)
= ω ij dx j = −eijk Ωk dx j = eikj Ωk dx j
u(rot ) = Ω × dx
–  This makes it a bit more clear that this is a rotaAon.

€
 Strain Tensor
Axial Strains Shear Strains

⎡ ∂u1 1 ⎛ ∂u1 ∂u2 ⎞ 1 ⎛ ∂u1 ∂u3 ⎞⎤

⎢ ⎜ + ⎟ ⎜ + ⎟⎥
⎢ ∂x1 2 ⎝ ∂x 2 ∂x1 ⎠ 2 ⎝ ∂x 3 ∂x1 ⎠⎥
⎢1 ⎛ ∂u1 ∂u2 ⎞ ∂u2 1 ⎛ ∂u2 ∂u3 ⎞⎥
⎢ ⎜ + ⎟ ⎜ + ⎟⎥
⎢2 ⎝ ∂x 2 ∂x1 ⎠ ∂x 2 2 ⎝ ∂x 3 ∂x 2 ⎠⎥
⎢1 ⎛ ∂u1 ∂u3 ⎞ 1 ⎛ ∂u2 ∂u3 ⎞ ∂u3 ⎥
⎢2 ⎜⎝ ∂x + ∂x ⎟⎠ ⎜ + ⎟
2 ⎝ ∂x 3 ∂ x 2 ⎠ ∂x 3 ⎥
⎣ 3 1 ⎦

€
Examples of Strains

uniaxial
strains

An example
Including shear strain
 Principal Axes of Strain
•  For any strain, there is a
ε1 coordinate system where
ε2 the strains are uniaxial – no
shear strain.
•  Called the principal axes of
the strain tensor.
€ –  Strains in those direcAons are
€ the principal strains.
•  Principal axes are the
eigenvectors of the strain
tensor and principal strains
are the eigenvalues.
 Strain in a ParAcular DirecAon
•  Let’s look at how we can use the strain tensor to get a line
length change in an arbitrary direcAon. Start with a simple
example in 2D:
⎡ε11 0⎤
⎢ ⎥ ε11 > 0
⎣ 0 0⎦
•  If we start with a circle, it will be deformed into an ellipse.

€
 Strain in a ParAcular DirecAon
ΔL
⎡ε11 0⎤
⎢ ⎥ ε11 > 0 ux
⎣ 0 0⎦ r

ux θ
x
r

ux = xε11 = ( r ⋅ cosθ )ε11

uy = 0
ΔL = ux cos θ = rε11 cos2 θ
ΔL /L = ε11 cos 2 θ
 Strain in a ParAcular DirecAon
•  For a strain only in the x direcAon,
⎡ε11 0⎤ 2
⎢ ε
⎥ 11 > 0 ΔL /L = ε11 cos θ
⎣ 0 0⎦
•  If we do the same for a strain in only in the y
direcAon, we get something similar,
€
⎡0 0 ⎤ 2
⎢ ⎥ ε22 > 0 ΔL /L = ε22 sin θ
⎣0 ε22 ⎦
•  In general, we get an equaAon for an ellipse:
2 2
€ rre = ε11 cos θ + ε22 sin θ + ε12 sin2θ
 Suppose we have mulAple lines
•  If we measure line
length changes for
several lines of different
orientaAons, θi θi
•  If we have lines of three
or more different
orientaAons, we can
esAmate the three
horizontal strains.
2 2
(err ) i = ε11 cos θ i + ε22 sin θ i + ε12 sin2θ i
 Suppose we have mulAple lines
•  This is the basis for
esAmaAng strain from
EDM networks
•  The lines don’t have to
have the same origin. I
just drew it that way for
convenience.
•  The more lines of different
orientaAons, the beYer
the esAmate becomes
2 2
(err ) i = ε11 cos θ i + ε22 sin θ i + ε12 sin2θ i
Chen and Freymueller (2002)
•  Evaluated strain rates
for four different near-
fault EDM networks
along the San Andreas
fault
•  All sites were so close
to the fault we assume
uniform strain within
the network
EsAmated Strain Rate Tensors
EsAmaAng Strain from Line Lengths
•  Obviously, what we looked at is a 2D problem. We assumed
the state of plane strain, in which all deformaAon is
horizontal.
•  Line lengths or line length changes are invariant under
rotaAons, so we can learn nothing about the rotaAon tensor
from them
–  But we also can esAmate strain without worrying about rotaAon
•  We have to assume uniform strain, which might be the wrong
assumpAon.
•  An alternaAve might be to assume simple shear (no axial
strains), such as for glacial flow or some strike-slip fault
moAon problems.
 Strain and Angle Changes
•  If we strain an object, lines within the object will, in
general, rotate:

•  But a uniform strain (uniform scaling up or down in

size) does not rotate any lines. So that means we
can’t esAmate all three components of the strain
tensor from angle changes.
 EsAmaAng Strain from Angle
Changes
•  We can’t esAmate the dilataAon (Δ). Rewrite our
three horizontal strains this way:
Δ = ε11 + ε22 Dilatation

γ1 = ε11 − ε22 Pure shear with N-S contraction, E-W extension

Pure shear with NW-SE contraction, NE-SW extension

γ 2 = 2ε12 (or simple shear with displacement along x-axis)

•  The gammas are called the engineering shear strains

(the epsilons are the tensor strains). The factor of 2
is from historical pracAce.
€
–  Older papers ooen use the gammas. Newer work usually
uses tensor strains. Be careful of the factor of 2!
 IllustraAons of Engineering Shears
γ1 = ε11 − ε22 > 0 γ 2 = 2ε12 > 0

€ €
 What Can Be Determined From the
Data?
•  There is a general lesson here:
–  You might want to know all components of the strain and
rotaAon tensors.
–  When your data only determine some parts of your
parameters, it makes sense to re-parameterize the problem
so that you directly esAmate what your data can determine
•  The alternaAve is to put possibly arbitrary constraints on your
parameters
–  The discussion of what you can learn about a plate rotaAon
from a single site was another example.
–  What we are doing is breaking up our parameters into two
sets (determined and impossible to determine) that are
(ideally) orthogonal in some way.
 More examples of this
•  If you have GPS displacement/velocity data, you can
determine all components of the displacement
gradient tensor (or strain + rotaAon)
•  If you have line length changes, you can determine
all components of strain, but nothing about rotaAon
–  So don’t try to esAmate the displacement gradient tensor;
esAmate the strain tensor instead
•  If you have angle changes, you can’t determine
rotaAon or dilataAon
–  So don’t try to esAmate the strain tensor; esAmate the
gammas instead.
 Axis of Maximum Shear
•  One other thing you can get out of the
gammas is the axis of maximum shear
tan2Θ s =
γ1
γ2
Θ s = 12 tan−1 (γ1 γ 2 )
γ = γ12 + γ 22 Magnitude of maximum shear

–  If γ1 = 0, the maximum shear is along the x-axis.

– €If γ2 = 0, the maximum shear is 45° from the x-axis.
 (QuanAtaAve) RotaAon of a Line
Segment
•  Define the rotaAon vector of the
x+dx
ξ+dξ line by the equaAon at leo. In
index notaAon it is:
θ
eijk dx j dξ k
Θi =
dx n dx n
•  Because ξi = xi + ui,
∂ξ k ⎛ ∂uk ⎞
x ξ dξ k = dx m = ⎜δkm + ⎟ dx m
∂x m ⎝ ∂x m ⎠
€ •  Note: curly d and straight d are
Define the rotation vector Θ not the same (parAal vs. total
derivaAves).
dx × dξ €
Θ= •  Also the magnitudes of dx and dξ
dx ⋅ dξ are equal for a pure rotaAon.
Θ = sin θ ≈ θ
 Total vs. Partial Derivative
•  There is a subtlety to partial derivatives that you might or
might not know.
•  Assume a function of space and time, f(x,y,t), where the
variables x and y also depend on time.
–  Example: A function that depends on the angle of the sun, which
depends on your position (as a function of time if you are
moving), and with time directly
–  The function may have a direct dependence on time, and an
indirect dependence that comes from x(t) and y(t).
•  The total derivative is

df ∂f ∂f dx ∂f dy
= + +
dt ∂t ∂x dt ∂y dt
 (QuanAtaAve) RotaAon of a Line
Segment
•  Define the rotaAon vector of the
x+dx
ξ+dξ line by the equaAon at leo. In
index notaAon it is:
θ
eijk dx j dξ k
Θi =
dx n dx n
•  Because ξi = xi + ui,
∂ξ k ⎛ ∂uk ⎞
x ξ dξ k = dx m = ⎜δkm + ⎟ dx m
∂x m ⎝ ∂x m ⎠
€ •  Note: curly d and straight d are
Define the rotation vector Θ not the same (parAal vs. total
derivaAves).
dx × dξ €
Θ= •  Also the magnitudes of dx and dξ
dx ⋅ dξ are equal for a pure rotaAon.
Θ = sin θ ≈ θ
 RotaAon of Line Segment
•  SubsAtuAng this in gives:
1 ⎡ ⎛ ∂uk ⎞⎤
Θi = ⎢eijk dx j ⎜δkm + ⎟⎥dx m
dx n dx n ⎣ ⎝ ∂x m ⎠⎦
1 ⎡ ⎛ ∂uk ⎞⎤
Θi = ⎢eijk dx j ⎜δkm dx m + dx m ⎟⎥
dx n dx n ⎣ ⎝ ∂x m ⎠⎦
1 ⎡ ∂uk ⎤
Θi = ⎢eijk dx j dx k + eijk dx j dx m ⎥
dx n dx n ⎣ ∂x m ⎦
This term is the velocity
This term is just x x x = 0 gradient tensor, or strain
dx j dx m + rotation
Θi =
dx n dx n
[0 + eijk (εkm + ω km )]
These are just unit
vectors
Θ i = eijk (εkm + ω km ) dxˆ j dxˆ m
 What Does This Mean?
•  Here’s the equaAon:

Θ i = eijk (εkm + ω km ) dxˆ j dxˆ m

•  This equaAon tells us that a line segment can rotate because
of either a strain or a rotaAon.
–  The rotaAon part is obvious
–  The strain part we saw earlier (graphically). Depending on the values
€ of the strain components, the rotaAon might be zero or non-zero.
–  Also, the two terms might cancel out!