Quantum Transport:

Atom to Transistor

Prof. Supriyo Datta

ECE 659
Purdue University

02.14.2003

Lecture 14: Basis Functions, Density Matrix I

Ref. Chapter 4.3

Network for Computational Nanotechnology

 00:00
Retouch on Concepts

• Summary of Equations: • Recall the concept of a Hilbert Space, a function

space conceptually similar to a vector space.
EΦ = HopΦ Basis functions in Hilbert space are like the unit
v v
Φ( r ) = ∑ φmu m (r )
vectors xˆ , yˆ, zˆ in vector space.
• Proper use of basis functions facilitate matrix
m
v v reduction and so are very useful computationally.
Φ(r ) = ∑ φi′u′i (r ) • Computationally we often use a non -orthogonal
i basis set (for example recall H 2). Conceptually this
and can cause many problems The usual procedure is

v
to transform the oblique sets to orthogonal sets.
u ′i = ∑ Cmi um ( r ) • From now on it will be assumed that we are
working with orthogonal basis sets:
m

{Φ} = [C ]{Φ′}
v *
[ A′] = C + AC ∫ u nu m = δ mn
d r
 04:08
Basis Transformations

• Basis transformations are defined

by the relation: • To show this, we have:

v v * v v
u′i = ∑ Cmiu m (r ) Aij′ = ∫ dr u′i (r ) Aop u′j ( r )
m
Where:
v * * v
v v
Φ (r ) = ∑ φm u m (r ) = ∑ φi′ui′ (rv )
= ∫ d r ∑ C miu m ( r ) Aop ∑ C u (rv )
m nj n
m
n
i
So to transform a vector to a new
basis set we’ll have:
= ∑ ∑ C Am nCnj
*
mi
m n

{Φ} = [C ]{Φ′} But Cmi = C

*
( )+
im

• In matrix notation, basis

transformations are defined by:
∴ A′ij = ∑ ∑ C +
m n
( ) im [
Amn Cnj = C + AC ]
ij

[ A′] = [C ]+ A[C ]
Thus, we see that C can be used to transform
matrix operators as well as vectors.
 Hamiltonian Matrix 11:10
Transformation

• First, recall the definition of a unitary
transformation:
- Vector length is preserved
- Proviso on unitary matrix: C +C= CC + = I
• Consider the process of finding the eigenvalues
and eigenvectors of a Hamiltonian matrix.
• One way to visualize this process
is to consider [H] [D]
as a basis transformation from the
real space basis to the eigenvector
basis. Formally this is expressed as:
[D]= V+ [H] [V]

• In matlab, we invoke:
[V,D]= eig (H)
where D is a diagonal matrix with the
eigenvalues on the diagonal and V is a square
matrix with the eigenvectors as its columns.

1 2 n

V=
 Ordering of the 18:16
Hamiltonian Matrix

• How do we know it is D= V+HV and not D=VHV+?

• Look at the organization of old and new basis sets in V, H and D…

new basis old basis new basis

V= old basis H= old basis D= new basis

• Rules of matrix multiplication require that the columns of one matrix match
the rows of next matrix.
• So by observation it must be D= V+HV

new basis old basis new basis

old basis
new basis = x old basis x old basis
new basis

D
Quantum Transport – Prof. Datta
H
 22:38
Electrons in a Box

• As an example, we will calculate the • Remember, by the method of finite

electron density of the ‘electrons in a differences (using box boundary conditions).
box’
+ 2t 0 -t 0
Electrons in a box H= -t 0 +2t 0
O
O O
100Å Box with
• We want to find the electron density
100 pt. lattice
H
n( x ) = ∑φ
2
α ( x)
x occ.α
1 2 3K 100
x
Note: ‘occ. a’ refers to the sum over all
E occupied states. It is very important to not that
here the states are either full or empty.
(ev)

x
eigenvalue #
 25:48
Electron Density

• We can redefine n(x) by applying the • So the electron density is:

Fermi function to all states. Where the Fermi
n (x ) = ∑ φα ( x) f α
2
function provides the “degree of occupation”
between 0 and 1 of a given state at a known α

= ∑ φα ( x) f α φα ( x)
electrochemical potential µ. *

α
• We can also re-write this as:

µ n(x ) = ∑ ∑ φα ( x) ραβ φβ ( x)
Occupied and unoccupied levels at *

α β
where
E f α if α = β
(ev) unoccupied ραβ =
0 if α ≠ β
µ • Note: ραβ forms a diagonal matrix
occupied called the density matrix.

eigenvalue #
 31:00
Density Matrix

• Generalizing we write: • In other words:

ρ~( x, x′ ) = ∑ ∑ φα ( x)ρ αβ φβ* ( x′) Φ α (x ) = (V )α , x
α β
Where n(x) is the diagonal of ρ~( x, x′)
• This relation can be seen to represent a
Φ*β ( x′) = V + ( ) β , x′ = Vx*′, β
• Summarizing,
unitary transformation from the eigenvector
basis to real space. Note: Φα ( x) are
given by columns of V:
ρ~ = Vρ V +
ρ = V + ρ~V

a
- ρ~ is in real space
ρ is in the eigenstate space
KK
-

φα ( x ) = x = V x ,α -The diagonal elements of ρ are

~
equal to the electron density n(x).

Equilibrium Density Matrix
• So, in the eigenstate basis or
“space” ? is a diagonal matrix with
elements
1 matrix calculated? For a diagonal matrix it is
fα =
1 + e ( Eα − µ ) / k B T
= f 0 (Eα − µ )
• Now in general let us denote the
density matrix for any space as ?,
where ? is given by:
ρ = f0 ([H ]− µ[I ])
36:30
• What is meant by f 0 ([H ] − µ [I ]) ?
More Generally, how is the ‘function’ of a
simply the ‘function’ operated on all elements.
How about matrices with off diagonal
elements?
• Example: Given [H] with off diagonal
elements calculate (sin[H]). To do this we
must first diagonalize [H], then operate sin ()
upon the diagonalized form of [ H ] , and then
finally transform [H] back into its original
space.
 42:00
Equilibrium Density Matrix

Example: continued. • Note: In matlab matrix functions and element

(1) Diagonalize [H] by element functions are differentiated by
addition of an ‘m’ for matrix functions. e.g. :
H1 sin () : represents element by element operation
sinm (): represents matrix operations.
H2
O • Finally we see the expression for ? in real
space is:
(2) Operate sin () f 0 (E1 − µ )

sin(H 1) ρ~ = V f 0 ( E2 − µ ) V+
sin(H 2) O
f 0 (E n − µ )
O
(3) Transform back to original • Interestingly, ? is only diagonal in the
space eigenvector basis. Off diagonal elements of ? in
alternate basis sets are used in some
sin(H 1) calculations, but more often than not only the
V sin(H 2) V+ diagonal elements of ? (which in any space

O
provide electron density) are of interest.