Hongkong

Chapters
coauthored with:
Maxence Cassier
Ornella Mattei
Moti Milgrom
Aaron Welters
With tremendous
help from
Nelson Beebe.
Its available !
http://www.math.utah.edu/~milton/
Rewriting some of the linear
equations of Physics.
Constitutive Law:
Differential Constraints:
Key Identity:
Time Harmonic Equations:
Acoustics:
Elastodynamics:
Maxwell:
Thermoacoustics: (s( = 0 )
Key Identity:
Acoustics in the time domain:
Key Identity:
Elastodynamics in the time domain:
Key identity:
Elastodynamics in a moving frame: Galilean transformation
Matrix in the constitutive law now has Willis type couplings:

Yellow=Compliant, Blue=Stiff
Red=Rubber, Black=Lead
Time harmonic acceleration with no strain
gives stress: Example of a Willis material
Electric dipole array
generates
polarization field
Force dipole array

generates
stress field
Piezoelectricity in the time domain:
Key identity:
Biot equations in the time domain ( s = 0 )
Key identity:
Thermal Conduction and Diffusion:
Key Identity:
Thermoelasticity in the time domain:
Key Identity:
Maxwell’s equations in the time domain:
Differential constraints
Key Identity:
h2I=2m
A=¹
Key Identity:
(¹
h = 1)
Key identity still holds, and the above equation reduces to:
Minimization principles for SchrÄ
odinger's equation
with complex energies
h2I=2m
A=¹
Minimize over Ã 0
where
subject to suitable boundary conditions on Ã 0

The Desymmetrization of SchrÄ
odinger's equation
h2I=2m
A=¹
Replace with:
Let ¤ denote appropriate symmetrization operator:
Advantage: Can solve iteratively using FFT, and

the FFT operations only need be done on , i.e.
only on two electron co-ordinates not all n electrons.
Are there other boundary field equalities or inequalities
that use partial information about what is inside the body?
Following the ideas of Straley, Milgrom and Shtrikman

suppose there is a matrix such that
In two dimensions suppose
Following ideas of Keller, Dykhne and Mendelson, we have

the boundary field equality
Due to the fact that the equations are satisfied with

To establish the inequality one needs to pick a
such that the volume average of
is non-negative for any C-periodic function
satisfying the appropriate differential constraints
which we write as
Find such that is positive

semidefinite for all
Then we have the boundary field inequality:

New Methods
for Imaging
Key idea: making a direct link between
Dirichlet-to-Neumann maps for bodies
and effective tensors for composites.
Abstract Theory of Composites
Hilbert Space
Operator L:H!H
Given
Solve
With
Then defines :U !U
Example: Conducting Composites
- Periodic fields that are square integrable over
the unit cell
- Constant vector fields
- Gradients of periodic potentials

- Fields with zero divergence and zero
average value
- Total electric field
- Total current field
L= - Local conductivity
- Effective conductivity
Variational principles if L is self-adjoint and
positive definite:
Leading to the elementary bounds:

Formula for the effective operator
where is the projection onto .

Leads to series expansions:
We want to reformulate it as a problem in
the abstract theory of composites, so we
can apply the machinery of the theory of
composites.
The abstract problem in composites consists in
finding for a given field e(x) in
(with associated boundary potential V0 (x) ) the
fields which solve:
j0 (x) + J(x) = ¾(x)[e(x) + E(x)]
with
0
j (x) 2 U; J(x) 2 J ; E(x) 2 E
which is exactly the conductivity problem we would
solve for the Dirichlet problem.
Furthermore if we knew the effective operator
:U !U
Then we have
0
j = L¤e
and the boundary values of n ¢ j0 (x) allow us
to determine the Dirichlet-to-Neumann map
assuming the fields in have been numerically
calculated
Analyticity properties of effective tensors as
functions of the moduli of the component materials
(Bergman, Milton, Golden and Papanicolaou)
extend to the Dirichlet-to-Neumann map
T
Pn
¾(x) = R(x) [ i=1 Âi (x)¾i]R(x)
The Dirichlet-to-Neumann map is a Herglotz function

of the matrices ¾1; ¾2; : : : ¾n in the domain where
these have positive definite imaginary parts, modulo
a rotation in the complex plane.
Easiest to prove using an approach of Bruno:
The truncated series expansion,
m
¼
with L = is a polynomial in the matrix
elements of ¾1; ¾2; : : : ¾n and hence
will be an analytic function of them in the
domain of convergence of the series
One obtains integral representation formulas for

and hence for the Dirichlet to Neumann map.
Time Harmonic Equations:
Acoustics:
Elastodynamics:
Maxwell:
In all these examples Z has positive semidefinite
imaginary part (that often can be made positive
definite by a slight rotation in the complex plane)
The analog of the variational principles of

Cherkaev and Gibiansky are then the variational
principles of
Milton, Seppecher, and Bouchitte (2009)

Milton and Willis (2010)
For acoustics, electromagnetism, elastodynamics

Minimization variational principles for
electromagnetism at fixed, possibly complex,
frequency in lossy materials.
Maxwell’s equations: (here we assume real !)
Let
For electromagnetism, acoustics and elastodynamics,
the Dirichlet-to-Neumann map is a Herglotz function
of the matrices Z1; Z2; : : : ; Zn of the component
materials
For electromagnetism an alternative and rigorous

proof was obtained with Maxence Cassier
and Aaron Welters (in the new book)
Some inverse problems for two-component bodies
Electromagnetism:
Look for special complex frequencies where
Extrapolate (using representation formulas or bounds)

measurements at different frequencies, or transient
responses, to the neighborhood of these special
frequencies
Quasistatic Elastodynamics:
Extrapolate to frequencies where
Quasistatic Electromagnetism:
Extrapolate to ratios "1 ="2 close to 1

Rigorous Upper and Lower Bounds on the Stress Relaxation in
cylindrical composites in antiplane elasticity
Note collapse of bounds: they can be used in an inverse

way to almost exactly recover the volume fraction from
measurements at this particular time. Analogously for
Dirichlet to Neumann map, one expects to recover
geometry
With Ornella Mattei in new book,

Generalizing the concept of function to
Superfunctions!
Adding resistor networks
Multiplying resistor networks
Substitution of networks
We should consider a resistor network in conjunction
with its batteries
Space H Space V
Combined Space K = H © V
Incidence Matrices:
Two natural subspaces:

J the null space of M T (current vectors)
E the range of M (potential drops)
These are orthogonal spaces and K = E © J
Other spaces:
Divide the bonds in H into n groups (representing
the different impedances).
Define Pi as the space of vectors P with elements Pj

that are zero if bond j is not in group i.
The projection ¤i onto the space Pi is diagonal and

has elements
f¤i gjk
Thus
This is an orthogonal subspace collection Y(n)
Y(n) subspace collection:
Z(n) subspace collection:
s
Superfunction F (n): Y(n) subspace collection
with
Subspace collections need not have

orthogonal subspaces
U V Y(2)
Z(2) J
P2
J
P1 E
P2
E P1
Key: Allow VI
nonorthogonal
s
Subspace E F (1)
collections. P1
J
Then we have a
whole algebra. VO
Application: Accelerating some Fast Fourier Transform
Methods in two-component composites
The effective conductivity ¾¤ is an analytic function

of the component conductivities ¾1 and ¾2
With ¾2 = 1, ¾¤ (¾1 ) has the properties of a
Stieltjes function:
Bergman 1978 (pioneer, but faulty arguments)

Milton 1981 (limit of resistor networks)
Golden and Papanicolaou 1983 (rigorous proof)
Numerical scheme of Moulinec and Suquet (1994)
Numerical scheme of Eyre and Milton (1999)
Ideal scheme:
But we want to do this transformation at the

level of the subspace collection, to recover the fields
At a discrete level
Problem: this substitution shortens the branch cut

instead of lengthening it.
Solution:
Substitute non-orthogonal subspace collections
Model example: a square array of
squares at 25% volume fraction
Obnosov’s exact formula

Thank you!
Thank you!
Thank you!
Thank you!
Thank you!

Hongkong

Uploaded by

Copyright:

Available Formats

Hongkong

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hongkong

Uploaded by

Copyright:

Available Formats

Chapters

Matrix in the constitutive law now has Willis type couplings:

Force dipole array

subject to suitable boundary conditions on Ã 0

Let ¤ denote appropriate symmetrization operator:

Advantage: Can solve iteratively using FFT, and

Following the ideas of Straley, Milgrom and Shtrikman

Following ideas of Keller, Dykhne and Mendelson, we have

Due to the fact that the equations are satisfied with

Find such that is positive

Then we have the boundary field inequality:

- Gradients of periodic potentials

Leading to the elementary bounds:

where is the projection onto .

The Dirichlet-to-Neumann map is a Herglotz function

One obtains integral representation formulas for

The analog of the variational principles of

Milton, Seppecher, and Bouchitte (2009)

For acoustics, electromagnetism, elastodynamics

Maxwell’s equations: (here we assume real !)

For electromagnetism an alternative and rigorous

Look for special complex frequencies where

Extrapolate (using representation formulas or bounds)

Extrapolate to frequencies where

Extrapolate to ratios "1 ="2 close to 1

Note collapse of bounds: they can be used in an inverse

With Ornella Mattei in new book,

Two natural subspaces:

Define Pi as the space of vectors P with elements Pj

The projection ¤i onto the space Pi is diagonal and

Z(n) subspace collection:

Subspace collections need not have

The effective conductivity ¾¤ is an analytic function

Bergman 1978 (pioneer, but faulty arguments)

But we want to do this transformation at the

Problem: this substitution shortens the branch cut

Obnosov’s exact formula

You might also like