5.

73 Lecture #1
Handouts: 1. administrative structure 2. narrative 3. Last years lecture titles (certain to be modified) 4. Gaussian and FT 5. PS #1 Due 9/13

1-1

Read CTDL, pages 9-39, 50-56, 60-85 Administrative Structure

20%

In-class 5 minute Quizzes. Exercise concepts immediately after they are introduced. ~10 Problem Sets

40%

Difficult, mostly computer based problems group consultation encouraged TAs (grade problem sets) some help with computer programs

I WILL DEAL WITH THE QM, NOT THE COMPUTER PROGRAMS! Optional Recitation! R. Field - answer questions about Problem Sets How about: Wednesdays, 5:00PM

40%

Take home Exam no group consultation about methods of solution, OK for clarification of meaning of the questions. CTDL Handouts Lecture Notes - formal, elegant, analytic - other texts and Herschbach - provide tools for solving increasingly complex problems

NO PHILOSOPHY, NO PREACHING TO THE CONVERTED

5.73 Lecture #1
Course Outline increasingly complex, mostly time-independent problems *

1-2

1D in (x) picture
spectrum {En } potential V( x)

I(t)

V(x)

central problem in Physical Chemistry until recently

control

femtochemistry: wavepackets exploring V(x), information about V(x) from timing experiments. How is a wavepacket encoded for xc, x, pc, p?(c = center) evaluate integrals stationary phase interpret information contained in (x) with respect to expectation values and transition probabilities

Confidence to draw cartoons of (x), even for problems you have solved once but no longer remember the details.

(x)

EHIGHER ELOW x

(x)

5.73 Lecture #1
* Matrix Picture

1-3

(x) replaced by collections of numbers called matrix elements tools: perturbation theory * * small distortions from exactly solved problems f(quantum numbers) F(potential parameters)

representation of spectrum

representation of potential e.g. Dunham Expansion

Vibration-Rotation Energy Levels:

1 m EvJ = Ylm v + [ J ( J + 1)] 2 l, m R Re V( ) = a n n Re n=0

Linear Algebra: Diagonalization Eigenvalues and Eigenvectors How to set up and read a matrix. Density Matrices: specify general state of system () vs. an operator (Op) that corresponds to a specific type of measurement, populations and coherences. * 3D Central Force - 1 particle

radial, angular factorization specific universal, exactly soluble

ANGULAR MOMENTUM map one problem surprisingly onto others symmetry classification of operators matrix elements reduced matrix elements

5.73 Lecture #1
* Many Particle Systems many electron atoms

1-4

Slater determinants satisfy antisymmetrization requirement for Fermions Matrix elements of Slater determinantal wavefunctions orbitals configurations states (terms) spectroscopic constants for many electron systems orbital integrals * Many-Boson systems: coupled vibrations: Intramolecualr Vibrational Redistribution (IVR)

* Periodic Lattices -band structure of metals

Some warm-up exercises Hamiltonian

special QM prescription

p2 H=T+V= + V(x) 2m h px = i x h2 2 H= + V(x) 2 2m x

Schr. Eqn.

(H E) = 0

1. Free particle V(x) = const. = V0

Schr. Eqn.

h2 d2 + V 0 E = 0 2m dx 2

5.73 Lecture #1
d2 2m = ( E V 0 ) 2 dx 2 h
call this k
2

1-5

k real if E V0 k imaginary if E < V0 general solution

(classically forbidden region)

(x) = A e ikx + Be ikx

Complex Numbers: i2 = 1

z = x + iy, z* = x iy e ikx = cos kx i sin kx

What is k?
1/2

2m k = ( E V 0 ) 2 h

but

What happens when we apply p to e ikx ?

pe ikx =
hk = p

h d ikx e = hke ikx i dx

eigenvalue

eigenfunction

of p

a number, not an operator This suggests (based on what we know from classical mechanics about momentum) that if k > 0 something is moving to right (+x direction) and if k < 0 moving to left How do we really know that something is moving? We need to resort to time dependent Schr. Eqn.

5.73 Lecture #1

1-6

k is wave vector (or wave number). Why is it called wave vector?

in 3 - D get e

rr ik r

r where k points in direction of motion periodic k = 2 k=

wavelength

e i(kx+2) = e ikx e ik(x+ ) e ikx

advance x by one full oscillation cycle

k is # of waves per 2 unit length

is probability amplitude probability distribution

= Ae ikx + Be ikx

travels to left?

* =|A|2 +|B 2 +A * Be 2ikx | +AB * e 2ikx

simplify:

2 i Im (A * B) = A * B AB *

2 Re(A * B) = A * B + AB *

x = Re( x) + i Im( x)

travels to right?

e i = cos i sin

5.73 Lecture #1

1-7

* =| A |2 + | B |2 +2 Re( A * B)cos 2 kx + 2 Im( A * B)sin 2 kx

constant (delocalized particle) wiggly - only present if both A and B are nonzero standing wave, real not complex or imaginary

A,B determined by specific boundary conditions.

Cant really see any motion unless we go to time dependent Schr. Eq. Need superposition of +k and k parts to get wiggles. Wiggles = superpositon of waves with different values of k = another kind of superposition (wave packet) Motion becomes really clear when we do two things: * time dependent (x,t) * create localized states called wavepackets by superimposing several eikx with different |k|s. [NEXT LECTURE: CTDL, pages 21-24, 28-31 (motion, infinite box, -function potential, start wavepackets).]

Dave Lahr to talk here about use of computers.

5.73 Lecture #2

2-1

Last Time:

free particle V(x)=V0 = Aeikx + Beikx

general solution

A,B are complex constants, determined by boundary conditions

k= p
h

(from e ikx , eigenfunction of p, and the real number, p, is the eigenvalue)

1/2

2m k = (E V0 ) 2 h

for E V0

probability distribution

P( x ) = * = |14243 + 2 Re( A * B) cos 2 kx24444444kx A 4+ | B |2 14444444+ 2 Im( A * B)sin 23 |2 4

const. wiggly

only get wiggly stuff when 2 or more different values of k are superimposed. In this special case we had +k and k. TODAY

and 1. infinite box 2. (x) well 3. (x) barrier

and

5.73 Lecture #2
What do we know about (x) for physically realistic V(x)?
( ) = ? * ( x) ( x) for all x?

2-2

Continuity of and d /dx ?

* ( x) ( x)dx ?

Computationally convenient potentials have steps and flat regions.

infinite step

finite step

infinitely high but infinitely thin step, -function

continuous d d 2 , not continuous for infinite step, and not for -function dx dx 2 d is continuous for finite step dx

More warm up exercises 1. Infinite box V(x) 0 L x

( x) = Ae ikx + Be ikx = C cos kx + D sin kx

[C=A+B, D=iA iB]

(0) = 0 C = 0 (L) = 0 kL = n n = 1, 2 ,

( why not n

= 0?)

5.73 Lecture #2
recall
2m n 2 2 = 2 V0 = 0 L h2 Insert kL = n boundary condition. k 2 = ( E V0 ) En = n2
2 h2 2 2 h 2 = n 2 mL 8mL2

2-3
here. En is integer multiple of common factor, E1. Important for # of bound levels wavepackets! n = 0 would be empty box
because
L 2 0 sin ( nx)dx = L / 2

E1

normalization (P=1 for 1 particle in well) L |D|= (2 / L)1/2 1 =| D |2 dx sin 2 ( nx)

0

n ( x) = (2 / L )1/ 2 sin( nx)

D = (2 / L)1/2 e i {
arbitrary phase factor

cartoons of n(x): what happens to {n} and {En} if we move well: left or right in x? up or down in E?

Infinite well was easy: 2 boundary conditions plus normalization requirement. Generalize to stepwise constant potentials: in each V(x)=constant region, need to know 2 complex coefficients and, if the particle is confined within a finite range of x, there is quantization of energy. * boundary and joining conditions * normalization * overall phase arbitrariness So next step is to deal with case where boundary conditions are not so obvious. (x) well and barrier.

5.73 Lecture #2
0 V(x) x

2-4

V(x) = a (x)

a>0

a has units Energy x Length (because, as we will see, (x) has units of reciprocal length)

= 0 everywhere except V(0) = a strength of the -function well

2m d2 2 = ( E + a( x ) 1 24 4 3 2 dx E V( x ) h

Schrdinger Equation Integrate:

+ 2 ma 2 mE dx = lim dx 2 ( x) + 2 ( x) ( x) lim h 0 0 dx2 h d d = LHS = size of discontinuity dx x =+ dx x =

+

d2

in

d at x = 0 dx

RHS = 0

2ma h2

(0)

because 2mE (0) h2 is finite and integral over region of length 2 0.

because, by the definition of a fn

(x)(x)dx = (0)
or, more generally

(x a)(x)dx = (a)

5.73 Lecture #2

2-5

Since the potential has even symmetry wrt x x, (x) must be even or odd (not a mixture) with respect to x x, thus (x) = (x). If (x) is even, there must be a cusp in (x) at x = 0 (x) is

(x)
0 OR

continuous

BUT NOT

(x) is not continuous at x = 0 So what happens when (x) is an odd function?

d(+) d() 2ma = 2 (0) dx dx h

The new boundary condition

since there is + reflection symmetry for an even (x)

d(+) d() = dx dx
ma d() = m 2 (0) dx h

Now find the eigenfunctions and eigenvalues. Standard procedure: divide space into regions and match and d/dx across boundaries.

5.73 Lecture #2

2-6

Region I 0 V(x) |x|>0 Let E < 0

E= |E|

Region II x

L = I = A L e + x + BL e x R = II = A R e + x + BR e x | E | 2m = h2
1/ 2

(8 unknowns, because A and B can be complex numbers)

(THIS IS WHAT WE DO WHEN k WOULD BE IMAGINARY)

(+)=0 ()=0 L()=R(+)0

AR = 0 BL = 0 AL = BR A arbitrary phase normalization

unknowns determined (2) (2) (2) (1) (1) (8)

L = Ae x R = Ae x

Done!

d R ( + ) ma = Ae 0 = 2 (0) A h dx
= ma h2

required discontinuity in d/dx at x = 0.

d L () + ma = + Ae +0 = 2 A ) (0 h dx

again

ma = 2 h

5.73 Lecture #2
Only one acceptable value of one value of E < 0

2-7

ma 2 h 2 ma 2 = 2 |E |= = = E 2 h 2m 2h
E= ma 2h 2

Actually, the above solution was specifically for an even (x). What about odd (x)? No calculation is needed. Why? Normalization of
1 = | |2 dx

R = Ae max/ h
0

1 = 2 | A |2 e ma A = 2 h

2 ma h 2 x

h2 dx = 2 | A |2 2 ma
see Gaussian Handout

1/ 2

ma = 2 h

1/2

/h e ma|x|

only one bound level, regardless of magnitude of a

large a, narrower and taller

There is a continuum of s possible for E > 0. Since the particle is free for E > 0, specific form of must reflect specific problem: e.g., particle probability incident from x < 0 region. It is even more interesting to turn this into the simplest of all barrier scattering problems. See Non-Lecture pp. 2-8, 9, 10.

5.73 Lecture #2

2-8

Nonlecture
Consider instead scattering off V(x) = + (x) a>0

V(x) = +(x) x
2mE k= 2 h
1/2

0
L = A L e ikx + BL e ikx R = A R e ikx + BR e ikx

In this problem we have flux entering exclusively from left. The entering probability flux is |AL|2. Two things can happen: 1. 2. transmit through barrier reflect at barrier
2

|AR|2 |BL|2

There is no way that BR can become different from 0. Why? Our goal is to determine A R and BL vs. E
2 2

L(0) = R(0)

continuity of

A L + BL = AR + B R

but BR = 0

A L + BL = AR

2ma d R (+0) d L (0) dx = + h 2 (0) dx 2ma R(0) ikA R (ikA L ikBL ) = 2 A R h AR = AL + BL 2ma ik( A L + BL ) ik(A L BL ) = 2 ( A L + BL ) h
L(0)

2ikBL =

2ma ( A L + BL ) h2 2ma 2ma BL 2ik 2 = 2 A L h h h2 2ma ikh 2 AL = 1 2ik 2 = BL 2ma ma h

5.73 Lecture #2

2-9

+1 =

ikh 2 ma

B A R = A L + BL = A L L + BL = BL + BL = BL ( + 1) BL

ikh 2 A R = BL ma
Transmission is T= AR AL Reflection is
What is T(E), R(E)?

= AL/BL

2 2 2 2

R=

BL AL

AR

= BL
*

k 2h 4 m a
2 2

= BL

2mE h 4 h
2

m a

2 2

= BL

2h 2 E ma 2

ikh 2 ikh 2 AL AL 1 1 = BL BL ma ma AL k 2h4 2h 2 E + ma 2 2 = 2 2 +1 = m a ma 2 BL

1 2

2h 2 E ma 2 R(E) = 2 = + 1 2h E + ma 2 ma 2

decreasing to zero as E increases

ma 2 2h 2 E T(E) = 2 = 2 + 1 . 2h E + ma 2 2h E R(E) + T(E) = 1

increasing to one as E increases

5.73 Lecture #2
Note that: R(E) starts at 1 at E = 0 and goes to 0 at E

2 - 10

T(E) starts at 0 and increases monotonically to 1 as E increases. Note also that, at E =

ma 2 2h 2

R as E approaches ma2/2h2 from above and then changes sign as E passes through ma2/2h2 !

This is the energy of the bound state in the (x)-function well problem.

See CTDL Chapter 1 Problem #3b (page 87) for a related problem

5.73 Lecture #3
Reading Chapter 1, CTDL, pages 9-39, 50-56, 60-85
h2 En = n 2 2 8mL
d d 2 , dx dx 2

3-1

Last time: 1.

1-D infinite box continuity of

(x),

n = (2 / L)1/2 sin(nx)

confinement quantization 2. function well one bound level

E= ma 2 2h 2
1/2

ma = 2 h

e ma|x|/h

(what happens to as a increases?)

Why do we know there is only one bound level? What do we know about (p) ? How does this depend on a ?

what about <p>?

TODAY and WEDNESDAY: 1. motion time dependent Schr. Eq. 2. motion of constant phase point on (x,t) -- phase velocity 3. motion of |(x,t)|2 requires non-sharp E 4. encode (x,t) for x0, x, p0, p 5. p0, p from |g(k)| 6. x0, x from stationary phase argument 7. moving, spreading wavepacket |(x,t)|2 8. group velocity phase velocity -- see CTDL, pages 28-31

revised 9/4/02 10:35AM

5.73 Lecture #3
1. Motion time dependent Schr. Eq.
ih = H t

3-2

TDSE

if V(x) is time independent, then

n (x, t) = n (x)e iE n t/h

satisifies TDSE?

( x,0) = an n (x)

can use this form of to describe time dependence of any non-eigenstate initial preparation: e.g. wavepackets superposition of eigenstates n = En/h

( x, t ) = an n (x)e i n t

go back to free particle to really see motion of QM systems

| k | ( x) = Ae ikx + Be ikx E k V0 = p2 h2 k 2 = 2m 2m

hk 2 0 2m add a phase factor which expresses the arbitrariness of the zero of energy : k = ( E k V0 ) h = | k | ( x, t ) = e i k t Ae ikx + Be ikx e iV0 t / h = Ae i ( kx k t ) + Be i ( kx + k t ) e iV0 t / h
argument phase constant

WHAT ABOUT ARBITRARY ZERO OF E?

2. How does point of

move?

const = kx k t t x (t) = + k + x (0) k

moves in +x direction if k > 0

revised 9/4/02 10:35AM

5.73 Lecture #3
v = = k 2m dt k dx
hk 2

3-3
phase velocity v = hk p v = = 2m 2m 2
(half as fast as we naively expect)

first term in (x,t) moves to +x (right), second to x (left).

But if we treat the e iV0t/h = e i 0t term explicitly, we get v = k + 0 ! Any velocity we want! IS THIS A PROBLEM? WHY NOT? k

(compare v for a +k, k pair of free particle states)

3.

But what about the probability distribution, P (x,t)?

P(x, t) = * (x, t)(x, t) =|A|2 +|B|2 +2 Re(A * B)cos 2kx +2 Im(A * B)sin 2kx
no time dependence! lose all t-dependence because cross terms (+k, k) still belong to same Ek! The wiggles in * are standing waves, not traveling waves. No ambiguity about V0 either?
* pdx < p >= ? * dx
p = hk
2 2 |A | |B| 2 2 |A | +|B|

What is the expectation value of p

This is an interesting result that suggests something that is always true and a very useful inspection tool. Whenever the wavefunction is pure real or pure imaginary, p = 0.

SO HOW DO WE ENCODE (x,t) for both spatial localization and temporal motion? need several k components, not just +k, k

revised 9/4/02 10:35AM

5.73 Lecture #3
**4. Recipe for encoding Gaussian Wavepacket for x0, x, p0, p

3-4

Start with (x,0) and later build in correct e i k t dependence for each k component.

THIS IS GOING TO BE SHOWN TO BE the that yields * as a Gaussian prob. distrib. in x and p

(x, 0) =

a1/2 (2)3/4

a 2 /4 ( k k ) 2 0 e e ikx dk FT of a g(k) Gaussian

*form of a Gaussian in k * superposition of many eikx ASIDE see Gaussian Handout a Gaussian prob. distrib. in x

WHAT KIND OF FUNCTION IS THE SQUARE OF A GAUSSIAN? WHAT IS ITS WIDTH?

G( x;x 0 , x ) (2 )

1/ 2

1 ( xx 0 ) e x

[2(x)2 ]

G is normalized

G(x; x 0 , x)dx = 1
( x)2 is called the variance in x
2 1/2

x = x0 x2 = ( x)2 + x2 0

[x

revised 9/4/02 10:35AM

5.73 Lecture #3
4A. k0, k. Now what can we say about g(k) in (x,0) above?

3-5

a G( k; k 0 , k ) = (2) 1/2 1/2 g(k) 2

a2 1 4 2(x)2 a 1 1/2 x 2

g( k) = e (a

/4 )( k k 0 ) 2

compared to G( x; x0 , x) = (2 ) 1/2
k = k0 k = 21/2 a

1 ( x x 0 ) 2 e x

[2( x) ]
2

you can verify by doing relevant integrals

So we already know, by inspection (rather than integration), the k0, k parts for (x,0).

revised 9/4/02 10:35AM

5.73 Lecture #3
4B. What about x 0, x for G(k;k0,k)?

3-6

To do this, perform the FT implicit in defn. of (x,0) [CTDL, pages 61-62]

2 a1/ 2 (a 2 4 )( k k 0 ) +ikx (x, 0) = dk e (2)3/ 4

complete the squares in the exponent (because Gaussian integrals are easy)

a2 a / 4 (k k 0 ) + ikx = 4

2ix (k k 0 ) a 2

x2 + ik 0 x 2 a
k-independent take outside integral

k-dependent change variables and evaluate Gaussian integral

result

2 (x, 0) = 2 a

1/ 4

x = x0 = 0

2 2 e ik 0 x e x /a FT of a Gaussian is a Gaussian!

21/ 2 a

x = 2 1/ 2 a

ALL OF THE k0, k (for * , x = a / 2, k = , and xk = 1 / 2) INFORMATION IS HIDDEN

1 a

x k = 1 x p = h minimum uncertainty!

revised 9/4/02 10:35AM

5.73 Lecture #3
This wavepacket: 1. 2. minimum uncertainty centered at x 0 = 0

3 - 7

5,6. How to build a w.p. (not necessarily Gaussian) centered at arbitrary x0 with arbitrary x? * Start again with a new g(k)
(x, 0) = a1/2 (2)
3/4

g(k) e i(k)e ikx dk

complex g(k) written in amplitude, argument form

(a 5. let g(k) be sharply peaked near k = k0 . [It could be e and then we already know k 0 and k = 21 / 2 / a . ]
2

/4)(k k 0 ) 2

6. Thus we really only need to look at (k) near k 0 in order to find info
about x and x . This is a very important simplification (or focussing of attention)!

Stationary phase argument

**

Expand in (k) in power series in (k-k 0)

(k) (k 0 ) + (k k 0 )
0

d dk k = k 0

revised 9/4/02 10:35AM

5.73 Lecture #3

3-8

Thus the exponential in integral becomes

d i ( k k 0 ) + kx dk e i 0 e
very wiggly function of x except at a special region of x

Now what we want to know is the value of x (for k near k 0) where the phase factor becomes independent of k. This is because, when we integrate over k, if the wiggly factor in the integrand stops wiggling, the integral accumulates to its final value near this value of k!
plot I(k) vs. k I( k ) = (integrand )dk
k

The value of the integral evaluated at this special value of x (that we do not yet know) x = x0(t) is ~g(k0)k where k is the change in k required to cause the phase factor to change by .

MOST IMPORTANT IDEA IN THE ENTIRE LECTURE!

Solve for value of x where the phase factor stops changing, i.e. d d (k k 0 ) dk + kx = 0 dk 144 2444 4 3
phase factor

stationary phase requirement

want

d +x=0 dk

revised 9/4/02 10:35AM

5.73 Lecture #3

3 - 9

If we let

d x 0 , then the phase factor is stationary when x is near x0 dk k =k 0

a1/2 (x, 0) = (2)3/4

a 2 /4 ( k k 0 ) 2 i ( k k ) x ikx 0 0

e dk

shifts to be sharply peaked at any x0

= e ik ( x x 0 )e ik 0 x 0
( x, x 0 )

This ||2 is localized at x0, k0, and has widths x, k, k = ? x = ? (easy: by inspection) (must perform Fourier transform)

This prescription does not permit free specification of x. x must still be x = 21/2a if |g(k)| is a Gaussian [shortcut: xk = 1]. [N.B. We are talking about the shape of (x,0), not the QM x and p associated with a particular .]

revised 9/4/02 10:35AM

5.73 Lecture #3

3 - 10

I(k) g( k 0 )

I(k)

x = x 0 + x

x = x0

0
x = x 0 x

g(k)

k0

Integral accumulates near k = k0 but only when x x0.

revised 9/4/02 10:35AM

5.73 Lecture #3
7. Now we are ready to let (x,t) evolve in time

3 - 11

a1/ 2 i( k k 0 )x 0 ikx i k t (x, t) = e e dk |g(k)|e 3/ 4 (2) E k h 2 k 2 hk 2 k = = = h 2mh 2m

See CTDL, page 64 for evaluation of of (x,t) and (x, t) . | ( x , t ) |2 .
2

SPECIAL CHOICE OF ZERO OF E, V0 = 0

dk

integral and simplification

Arbitrary choice of zero of E drops out of

2

x t 608 7 2 hk 0 t 2a x m 1/ 2 2 2 1/ 2 1 + 4 h t 2 = 2 exp x, t 2 2 4 2 2 a 4 4h t 14444 m a 24444 3 a + t-dependent m2 normalization factor 144 2444 4 3

( )

( )

spreading

Complicated:

* x depends on t, reaching minimum value when t = 0 * x0t, the center of the wavepacket, moves as hk x 0t = 0 t m {
v group v phase = hk 0 ! 2m

revised 9/4/02 10:35AM

5.73 Lecture #3

3 - 12

*|g(k)|, which is independent of time, contains all info about p0, p. Therefore these quantities do not evolve in time for a free wavepacket. They do evolve if V(x) is not constant. Think about chopping up the Fourier transform of |(x,t)|2 into pieces corresponding to different values of p. If there is no force acting on the wavepacket, the <p> for each piece of the original |(x,t)|2 remains constant.

Summary We know how to encode a wavepacket for p 0, p, x 0 (and since x is an explicit function of time, we can let (x,t) evolve until it has the desired x and then shift x0t back to the desired location where x has the now specified value).

We also know how to inspect an arbitrary Gaussian (x,t) to reveal its x0t, x, p0t, p without evaluating any integrals.

revised 9/4/02 10:35AM

5.73 Lecture #4
Lecture #4: Stationary Phase and Gaussian Wavepackets

4-1

Last time: tdSE motion, motion requires non-sharp E phase velocity began Gaussian Wavepacket goal: x, x, p = hk, p = hk by construction or inspection (x,t) is a complex function of real variables. Difficult to visualize. What are we trying to do here? techniques for solving series of increasingly complex problems illustrate philosophical points along the way to solving problems.
So far: free particle infinite well very artificial - function * nothing particle-like * nothing molecule-like * no spectra

Minimum Uncertainty (Gaussian) Wavepacket -- QM version of particle. We are going to construct a (x,t) for which | (x,t)|2 is a Gaussian in x and the FT of (x,t), gives (k,t), for which |(k,t)|2 is a Gaussian in k. center of wavepacket follows Newtons Laws extra stuff: spreading interference tunneling Today: (improved repeat of material in pages 34 through 31
infer k by comparing g(k) to std. G(x; x0 , x) g( k) =| g( k) | e i ( k ) d dk x0
k =k 0 2

for k near k0

STATIONARY PHASE moving, spreading wavepacket how is it possible that the center of the wavepacket moves at a different velocity than its center k - component

( x, t)

vG v

revised 9/4/02 2:33PM

5.73 Lecture #4
Here is a normalized Gaussian (see Gaussian Handout) 1 ( x x 0 ) 2 G( x; x 0 , x ) = (2)1/2 e 2(x)2 x

4-2

normalized G(x; x , x )dx = 1 0 center x = x0 by construction 2 1/2 std. dev. x x 2 x

Now compare this special form against

a1/2 (x, 0) = (2)3/4

2 ( ) e 4 244 e ikx dk 1 4 3 { free particle

a 2 4 ( k k 0 ) g(k)

F.T. of a Gaussian in k

a Gaussian in k, but what width and k? by analogy

G(k; k 0 , k ) = (2)

1/2

a g(k) 21/2 123

1 k

1 a2 = 4 2(k)2 21/2 k = a

by analogy with G(x;x0,x)

So casual inspection of this form of (x,0) gives us k and k. Not quite so easy to get x and x. If we actually carry out the F.T. specified in the definition of (x,0) above (see bottom of page 34), we get

2 ( x, 0) = 2 a x = x0 = 0 x = 2
1/ 2

1/ 4

2 2 e ik 0 x e x a

1 1 2 = 2 a 2(x) x = a 21/2

a,

21/ 2 previously k = k 0 , k = ; a

revised 9/4/02 2:33PM

5.73 Lecture #4

4-3

But the square of a Gaussian is a Gaussian and its x or k is a factor of 2-1/2 smaller than the original value.

x for ( x, 0) is 2 -1/ 2 a, x for ( x, 0) 2 is

a . 2

21/ 2 1 k for ( k , 0) is , k for ( k , 0) 2 is . a a a1 1 See CTDL, p. 231 [x,k are defined xk = = rigorously in contrast to treatment on p. 23.] 2a 2
This is a very special Gaussian wavepacket * minimum uncertainty * x0 = 0 What about more general Gaussian wavepackets.? g(k) is a complex function of k sharply peaked near k = k0
g(k) = |g(k) e i(k) |

amplitude, argument form

If |g(k)| is sharply peaked near k = k0, then the only relevant part of (k) is the part for k near k0

d Expand (k) = ( k 0 ) + ( k k 0 ) + 13 2 dk k = k
0
0

higher terms neglected

a1/ 2 ( x, 0) = | g( k ) | e i( k )e ikx dk 3 / 4 1442443 (2 )

g(k) e
i 0 d i ( k k 0 ) dk e
k=k 0

+ kx

We want to cook (x,0) so that it is localized near x = x0. In order for this to happen, the factor ( k k 0 ) d dk

k=k0

+ kx

, must be indpendent of k near k = k0. Stationary Phase!

revised 9/4/02 2:33PM

5.73 Lecture #4
How does integral of a wiggly function accumulate?

4-4

e.g.,

k I(k) = e ik x dk

I(k) k

but if phase factor stops wiggling near k = k0

F(k0)k k0 where k is range of k over which the phase factor changes by . So, arrange for phase factor to become stationary near k = k0
d d (k k 0 ) dk + kx dk d 0= +x satisfied if dk 0=

d dk

k = k0

x0

revised 9/4/02 2:33PM

5.73 Lecture #4

4 - 5

Thus
d dk k = k 0

2 a1/2 ( a 2 /4 )( k k 0 ) i ( k k 0 ) x 0 ikx i 0 (x, 0) = e3 14 314 3/4 e e 4244 e 4244 dk (2) g(k) ik ( x x ) ik x

insertion of e i( k k 0 ) x 0 phase factor to center w.p. at x0 .

( x x 0 ) shifts to any

0 0

(stops wiggling only when x x0)

desired x0

Now put in time-dependence by adding

i k t

factor

E k h2 k 2 1 k = = h 2m h hk 2 k = 2m

a1/ 2 i k k 0 )x 0 ikx i k t ( x, t ) = g( k ) e ( e e dk 3 / 4 1442443 { (2 ) eigenstate of H

g( k )

revised 9/4/02 2:33PM

5.73 Lecture #4
This FT is evaluated and simplified in CTDL, page 64

4-6

hk 0 2 2 2a x t 2 2 1/ 2 1 + 4h t exp 2 = 2 m (x, t) 2 2 m 2a 4 a 2 14243 a 4 + 4h t 2 time dependent 144 2444 4 m 3 normalization

( )

Gaussian with time dependent width and center position

Maximum of Gaussian occurs when numerator of exp [ ] is 0. MOTION:

0=x vG =

hk 0 t m

x 0 (t) =

hk 0 t m

hk p d x 0 (t) = 0 = 0 = vclassical dt m m

This is 2 larger than v. p Classically expect free particle to move at constant v = m WIDTH:
compare coefficient of ( x - x 0 (t))2 in exp [
1/ 2
]

to standard G(x;x 0 , x) in handout

a 4 + 4h 2 t 2 / m 2 x = 4a 2

a + 2 {
minimum width at t=0

width increases linearly in t at long time (quadratically at early time).

ht ma {

1 2( x)
2

=
4

2a 2 a + 4 h2 t 2 m2

x and x are time dependent, but what about k and k? recall original definition of (x,0) (page 4-2), where (x,0) is written as the FT of a Gaussian in k

g(k, t) = e i k t g(k, 0)

revised 9/4/02 2:33PM

5.73 Lecture #4
k = k0 ( k, t) has 1 time independent k = a
2

4-7

We know free particle must have time independent k0 and k (no forces divide w.p. into k slices)
4h 2 t 2 1 xk = 1 + 2 4 2 m a
1/ 2

minimum uncertainty at t = 0 (and linearly increasing at long t).

For free particle, build w.p. with any desired x0, k0, k starting from
( x, t ) =

g(k)e ikx e i k t dk k = d dk k = k 0 vG =
1/ 2

hk 2 2m

find x 0 from

x 0 (t ) = x 0 + vG t a 4h 2 t 2 x = 1 + 2 4 2 m a

hk 0 m

if we want a value of x other than a/2 at t = 0, replace x by x = x + such that when the w.p. reaches x0 at t = 0 it has the desired width.

Could have started with (k,0) =

and then encoded k 0 in g(x) thru d dx

Gaussian in x

g(x) e ikx dx {{ inverse

F.T.

x=x0

= +k 0

where ( x) is the argument of g( x) = g( x) e i( x )

For next class read C-TDL pages 103-107, 1468-1476.

revised 9/4/02 2:33PM

5.73 Lecture #5
Continuum Normalization
Last time: Gaussian Wavepackets
How to encode x in

5-1

g( k )e ikx dk = (x)

or k in

g( x )e ikx dx = (k)

stationary phase: good for cooking or inspecting wiggly functions and for crudely evaluating integrals of wiggly integrands. vgroup vphase Today: Normalization of eigenfunctions which belong to continuously (as opposed to discretely) variable eigenvalues.

convenience of ortho-normal basis sets we often talk about density of states, but in order to do that we need to define state computation of absolute probabilities cannot depend on how we choose to define state.

1. Identities for -functions. 2. k, p,E for eigenfunctions corresponding to continuously variable eigenvalues. 3. finite box with countable discrete states taken to the limit L . Normalization independent quantity: # states # particles x is the argument of the delta-function. So if we integrate over a region of and x, we have the absolute probability. 4. two examples predissociation rate and smoothly varying spectral density.

revised 9/11/02 10:58 AM

5.73 Lecture #5

5-2

In Quantum Mechanics, there are two very different classes of systems. * SPATIALLY CONFINED: what is E good for? E quantized can count states, easy to compute density of states dn = E dE
* can normalize to 1 = E Edx

T: classical period of oscillation * # of encounters /sec: 1 T

* fraction of time in region of length L:

L /v T

* SPATIALLY UNCONFINED: E continuously variable dn cant count states, so how to compute ? dE ** can ask what is the absolute probability of finding the system between E, E + dE and x, x + dx For confined systems, we can express ortho-normalization in terms of Kronecker-

ij =

* j dx i

ij = 0 ij = 1

ij i=j

orthogonal normalized

For unconfined systems, we are going to ortho-normalize states to Dirac -functions In order to do this we need to know better what a --function is and what some of its mathematical properties are.

One of several equivalent definition of - function: 1 ( x x) = (x, x) = e iu(x x )du. 2 What is it good for?

(x, x)(x)dx = (x).

shifts a function evaluated at x to the same function evaluated at x.
revised 9/11/02 10:58 AM

5.73 Lecture #5
Prove some useful Identities

5-3

We do this so that we will be able to transform between k, p, and E (where E = f(k)) normalization schemes. 1.

(ax, ax ) =

1 ( x, x ) a

e.g., (p p ) = ( h( k k )) =

1
h

( k k )

nonlecture proof

(ax, ax ) =

1 iu ( ax ax ) du e 2

change variables

v = au dv = a du
1 1 1 e iv ( x x )dv = ( x, x ) 2 a a but, since (ax, ax ) = (ax ax) = (ax ax) = ([a](x x)) 1 ( is an even function), (ax,ax) = ( x, x) |a| (ax, ax ) =

2.

g(x) = i {
( )

dg(x i ) 1 (x, x i ) dx

provided that

zeros of g(x)

dg(x i ) 0 dx

expand g(x) in the region near each 0 of g(x), i.e., x near x i g(x) dg dx
x =x i

(x xi ).

If there is only 1 zero, then identity #1 above gives the required result. It is clear that (g(x)) will only be nonzero when g(x) = 0. Otherwise we need to carry out the sum in identity #2.
revised 9/11/02 10:58 AM

5.73 Lecture #5
EXAMPLES A. g(x) = (xa)(xb) This has zeroes at x = a, and x = b.
You should show that ( g( x)) = 1 [(x, a) + (x, b) ]. ab

5-4

B. (E1/2 , E1/2 )
has one zero at E = E, expand g(E) about E = E, thus for E near E g(E) = E1/2 E1/2 1 g(E) E 1/2 (E E). 2 you should show that E1/2 , E1/2 = 2 E1/2 (E, E) This is useful because k E
1/2

m (E E) = 2 2 h (E V0 )

1/2

( k k )

Another property of - functions: ( x, x) is an even function: expect

d ( x, x) dx

d ( x, x) (x, x) to be an odd function: dx d This is useful because ( x, x) is capable of picking dx df out evaluated at x. dx

Non-lecture:
Use definition of derivative to prove that
(x, x)f(x)dx = f (x )
[ (x + , x ) (x, x )] d (x, x ) = lim dx 0

(x + , x)f(x)dx = f(x ) (x, x)f(x)dx =

lim
0

f(x )

[(x + , x) (x, x) ] f ( x)dx = lim f ( x ) f ( x) = f ( x)

revised 9/11/02 10:58 AM

5.73 Lecture #5
k *

5-5

Our goal is to create ortho-normalized s that look like eikx: normalized to a -function in k
1 ix(k k ) * (k , k) = k,k k,k dx dx e 2

std. defn. of function in k

k,k (2) 1/2 e ikx

for V(x) = constant.

k,k is said to be normalized to (k, k ).

What is the probability of finding the system, which is described by k,k , to be located between 0 x L ?

L 0

* k,k k,k dx =

1 L L dx = = P k (L) 0 2 2

probability grows without limit as L

But, more interestingly, what is the probability of finding a system in a k-normalized state within a region of length equal to one de Broglie ?
= h /p = 2 k Pk () = 1 = 2 k

k normalized states (for V(x) = constant) have: 1/k particle per of x

or 1 particle per unit length 2

revised 9/11/02 10:58 AM

5.73 Lecture #5
What about ordinary space normalization?

5-6

k = N k e ikx k = N k e ik x
2 i( k k)x * dx k k dx =|N k | e
0 if k k if k = k
THIS IS THE PROBLEM! Cant specify Nk.

GENERALIZE
(k, k )
* 1 iu(k k ) du = k,k k, k dx e 2
, thus ( k , k ) = 1 2

1/2 ikx e where k,k (2)

e i( k k ) x dx

notation (k , k ) = (k k ) = (k k , 0) when (k,k) is multiplied onto f(k) and integrated over all k, we get f(k)

(k , k )f(k)dk = f( k )

(k,k) is zero when k k and is one when k = k

1 with eigenvalue k : k k , k 0 is eigenfunction of k = 0 k , k 0 = k 0 k , k 0 i x x, x 0 is eigenfunction of x = x with eigenvalue x 0

revised 9/11/02 10:58 AM

5.73 Lecture #5
Other normalization schemes for free particle p *
p,p = N p e ipx/h what is value of N p ?

5-7

(p, p ) = N * N p exp ix ( p / h p / h ) dx p = N * N p 2(p / h, p / h) p 14 244 4 3

1 using (p,p ) identity #1 h 1/2 ipx/h
1

p,p = (2h)

1 h particle per = p p

L * 0 p,p p,pdx = 2h

1 particle per unit * length h

E *

E = N E e ikx e ikx

2m( E V0 ) k= h2

1/ 2

degenerate pair of states

you show that

+ E,E E,E

m = 2E 2 h 2 m = 2E 2 h 2

1/4

2mE 1/2
cos 2 x
h
2mE 1/2
sin 2 x

h

1/4

+ E, E is orthogonal to E, E

+ +* E, E E, Edx

1 1 1 from E, E = = + another E 2E 2E

probability for E - normalized state per

* Volume of N-dimensional phase space occupied by a p normalized state is hN

revised 9/11/02 10:58 AM

5.73 Lecture #5
Thus there are 1 particles per for a E - normalized state. E
1/ 2
or 2 m particles per unit length
h 2E

5-8

( )

So we have assembled all the basic stuff we will need, at least for V(x) = constant problems. Now use it to examine a problem we understand perfectly.
En
-L/2 L/2

2 = L

1/2

sin 2mE n 2 cos h

1/2

x n odd

n even

2mE n kn = h2

1/2

h2 2 En = n 2 8mL
1 particle per box of length L 1 particle per unit length L 0 as L particle per L

4 normalization schemes (k, p, E, box): each gives different #/L or #/ . Why - because each scheme defines state differently.
# particles # states must be independent of However, expect that normalization scheme x
k, p, E or box

Why? Because the probability of finding a system between x, x + dx AND , + d is observable. We have completely specified what counts as an observation.

revised 9/11/02 10:58 AM

5.73 Lecture #5
Normalization-Independent Quantity for general V(x):
dn d L 2 13 lim 1 L/2 # states # particles * L L / 2 dx = x

5-9

density of states (# states per unit

The infallible way to get the invariant reference density is to box normalize (so that one can count states) and then take limit L . Why? Because most realistic potentials become smooth and flat at large enough x.

V(x)

x=L

x_(E) - inner turning point Procedure:

1. Box normalize E n 2. Compute

(E is quantized)

dn from E(n) dE 1 dn dn but 3. take limit L remains finite dE L dE

example:
0 L
L

2 E = L

1/2

2mE 1/2 n sin x h2

* E n E n dx = 1 by construction (for box normalization)

revised 9/11/02 10:58 AM

5.73 Lecture #5
h2 En = n 8mL2
2

5 - 10
dE 2nh 2 = dn 8mL2 8mE n L n= h2
2 1/ 2

E (E) =

dn 2L m 1/ 2 = dE h 2E

( )

REFERENCE DENSITY
1/ 2 2 m 1/ 2 dn 1 L P (x, x + x; E, E + E ) 2m * lim E E dx = = 2 = 04244 h 2E h E L dE L x E 4 3 {14 3 1 24 # # indep. of L E L

THIS SECTION TO BE REPLACED

dn p dn k dn E for E , for p , for k dE dp dk

1/2

2m h 2E

= lim

above derived reference density

dn E 1 L * dx dE 144E44 L 0 2 E 3

m 2 2h E

1/2

derived earlier

dnE = 2( E E ) = 2 dE

2 because each E is doubly degenerate

1 L * 1 0 p pdx = h L dn p 1 2m 1/ 2 = dp h h 2 E dn p 2m 1/ 2 2m p = E = p =E dp

revised 9/11/02 10:58 AM

5.73 Lecture #5
k 1 L

5 - 11

* k k dx =

1 2 dn k 2m k = 2 = dk h k E

dn d E 2E / E p p / E k k / E

see the pattern?

These results are only valid for problem. They illustrate all continuum normalization problems where it is desired to calculate probabilities.

2 Schematic Examples * Bound free transition probabilities * Constant spectral density across a dissociation or ionization limit.

revised 9/11/02 10:58 AM

5.73 Lecture #5
Bound-Free Transition (predissociation)

5 - 12

V(x)

bound (box normalized discrete energy levels)

repulsive (continuum of Elevels, cant really box normalize)

xstationary
phase

at t = 0 system is prepared in (x,0) = bound(x) Fermis Golden Rule:

Rate = boundfree = n (E) E = E dE 2 free* bound 2 E (E)H E dx E (E) h

derive this key quantity by box normalizing 1 dn repulsive state and taking lim L L dE

Then compute the H integral using two box normalized functions.

Constant spectral density on both sides of a bound/free limit

I()

( )

v = 0

Intensity( ) ~ smooth function of , no E discontinuity at onset of continuum

revised 9/11/02 10:58 AM

5.73 Lecture #6

6-1

Last time: Normalization of eigenfunctions which belong to continuously variable eigenvalues.

1. 2. 3. identities k , p , E , box trick using box normalization

# states # particles x
L 1/L

for box normalization

4.

dn often needed - alternate method via JWKB next lecture dE

1. 2.

V(x) = x linear potential solve in momentum representation, (p), and take F.T. to (x) Airy functions Semi-classical (JWKB) approx. for (x)
* * p( x) = [(E V( x))2 m ] ( x) = p( x)
1/2 1/2

envelope

i exp h

variable

p( x )dx

* visualize y(x) as plane wave with x-dependent wave vector * useful for evaluating stationary phase integrals (localization, causality) **** splicing across classical (E > V)|| forbidden (E < V) ] Next lecture

WKB Quantization Condition

x (E) x _+(E) p(x )dx = 2 (n + 1 2) n = 0,1,

updated 9/12/02 11:39 AM

5.73 Lecture #6
Linear Potential. V(x) = x
2 = p + x H 2m coordinate representation momentum representation x x p p h p x ih i x p note [ x, p] = ih in both representations - prove this? 2 2 2 = h d + x = p + ih d H H 2 m dx2 2m dp 2nd order 1st order - easier!

6-2

Solve in momentum representation (a sometimes useful trick) Schr. Eq.

d(p) i = E p 2 2m (p) dp h
(p) = Ne ap+bp
3

solution

gives p 2 times ( p )

gives constant times ( p )

plug into Schr. Eq. and identify, term-by-term, to get a =

iE h i b= 6h m
easy?

i (p) = N exp Ep p 3 / 6m h

* (p) (p) = 1!

N = 1!
updated 9/12/02 11:39 AM

5.73 Lecture #6

6-3

Now p is an observable, so it must be real. Thus (p) is defined for all (real) p and is oscillatory in p for all p. NEVER exponentially increasing or decreasing! IT IS STRANGE THAT (p) does not distinguish between classically allowed and forbidden regions. IS THIS REALLY STRANGE? If we allow p to be imaginary in order to deal with classically forbidden regions, (p) becomes an increasing or decreasing exponential.

If we insist on working in the (x) picture, we must perform a Fourier Transform

(x) = N (x) = N

e ipx / h (p) dp i exp h p(x E) + p 3 / 6m dp 144 2444 4 3 odd function of p: O(p)

= cos + isin { {
even odd

sin O (p) dp = 0
since O(p) is odd wrt p p.

(x) =

(x E)p + p 3 / 6m cos dp h

Ai ( z) = 1/2
Surprise!

cos s3 3 + sz ds

This is a named (Airy) and tabulated integral

* numerical tables for x near turning point i.e., x E/ analytic a s y m p t o t i c functions for x far from turning point
useful for deriving f(q.n.) and for matching across boundaries. * zeroes of Airy functions [Ai(zi)=0] and of derivatives of Airy

functions [Ai(zi)=0] are tabulated. (Useful for matching across center of potentials with definite even or odd symmetry.) [Two kinds of Airy functions, Ai and Bi.]

updated 9/12/02 11:39 AM

5.73 Lecture #6
s3 Ai(z) = 1/2 0 cos + sz ds 3
s p(2mha)1/3 (if > 0) for our specific problem (x E) 1/3 2m h 2 z Turning point

6-4

V(x) = x

>0

x+(E)

At turning point E = V(x + ) = x + x + ( E ) = E /

Problems with linear potentials:

boundary conditions

V(x)

+x

When there is symmetry (or 1/2 symmetry) we need to know where the zeroes of (x) { and
for even functions

d /dx 123 4 4

are located.
tables of zeroes of Ai(z) and Ai (z) "z n " " z " n

for odd functions or barrier

See Handout.

When there is no symmetry, must match Ai (or, more precisely, a linear combination of Ai and Bi) and Ai across boundaries, but we do not have to actually look at the Airy function itself near the joining point.
updated 9/12/02 11:39 AM

5.73 Lecture #6

6-5

This is not as bad as it seems because we are usually far from turning point at internal joining point and can use analytic asymptotic x expressions for Ai(z). 2 linear potentials of different |slope|. For > 0 there are 2 cases (classical and nonclassical) x (i)
z << 0 , E > V(x) classically allowed region 2 Ai(z) -1/2 ( z )1/ 4 sin ( z )3/2 + / 4 { { { positive 3 x is in here phase shift
* oscillatory, but wave vector, k, varies with x * Ai vanishes as x because of (-z) -1/4 factor * Bi is needed for case where Airy function must vanish as x + in classical region

asymptotic form for z << 0.

3/ 2 -1/2 /2 113 e (224 z /4 1 /3) z3 Ai(z) 2 4 positive decreasing exponential well behaved * not oscillatory, monotonic * Ai vanishes as x + * Bi vanishes as x in forbidden region

(ii) z >> 0 , E < V( x) forbidden region

asymptotic form for z >> 0.

Cartoon
need numerical tables to define (x) in this region

The two asymptotic forms of Ai(z) are not normalized, but their amplitudes (& phases) are matched. Links B.C. at x + to B. C. at x .

updated 9/12/02 11:39 AM

5.73 Lecture #6
NonLecture
OTHER CASE: < 0 z (||x + E) 2m|| 1/3 h2 ||

6-6

<0 x

for this case, need Bi(z) instead of Ai(z)

2 3/2 Bi ( z) 1/2 2 |z| 1/4 exp z 3 2 3/2 Bi ( z) 1/2 |z| 1/4 cos z + 4 3

( forbidden region, z << 0.) (allowed region, z >> 0.)

What is so great about V(x) =x? (x) is ugly need lookup tables, complicated solutions! Ai(z) turns out to be key to generalization of quantization of all (well behaved) V(x)! There are semi-classical JWKB (x)s These blow up near turning points (i.e. on both sides). The Ai(z)s permit matching of JWKB (x)s across the large gap where JWKB is invalid, ill-defined.

(JEFFREYS) WENTZEL KRAMERS BRILLOUIN

JWKB provides a way to get n(x) and En without solving differential equations or performing a FT.
But actually, the differential equations are easy to solve numerically. The reason we care about JWKB is that it provides a basis for: * physical interpretation (semi-classical) * RKR inversion from EvJ VJ(R). * semi-classical quantization. * the link to classical mechanics is essential to wavepacket pictures.
updated 9/12/02 11:39 AM

5.73 Lecture #6

6-7

(generalize on eikx for free particle by letting k =p(x)/h depend explicitly on x (why does this not violate [x,p]=ih ?) i x k(x) and p(x) are classical mechanical 1/2 JWKB = p(x) p( x )dx functions of x, not QM operators. exp 1 24 4 3 h c
classical envelope

p(x) = [2m(E V(x)]1/2

phase factor: adjustable to satisfy boundary conditions

p(x) is pure real (classically allowed) or pure imaginary (classically forbidden). p(x) is not Q.M. momentum. It is a classically motivated function of x which has the form of the classical mechanical momentum and has the property that the = h varies with x in a p reasonable way.

p( x)

1/2

is probability amplitude envelope because probability 1 so amplitude v 1 v

i exp h

p( x )dx is generalization of e ipx/h to non - constant V(x).

* *

h p(x) gives easily identifiable stationary phase region for many wiggly integrands. (Both s have same at x s.p. ) node spacing (x) =

Long Nonlecture derivation/motivation.

updated 9/12/02 11:39 AM

5.73 Lecture #6
i x Try (x) = N(x) exp c p( x )dx h

6-8

plug into Schr. Eq. and get a new differential equation that N(x) must satisfy
d 2 2m + ( E V(x)) = 0 dx 2 h 2 d2 1 2 2 + 2 p(x) = 0 dx h

**

* derived in box below

2ip( x) ip ( x) i 0 = N N N exp h h h

p( x ) dx

This is a new Schr. Eq. for N(x). Now make an approximation, to be tested later, that N is negligible everywhere. This is based on the hope that a slowly varying V(x) will lead to a slowly varying N(x).

i i x d = N p(x) exp c p( x )dx h h dx d2 i ip i i i x = N N p Np N Np exp c p( x )dx h h h h h dx 2 2i ip = N N p N h h p2 i x N exp c p( x )dx h h2

d 2 p2 i x 0 = 2 + 2 N exp c p( x )dx h h dx 2ip(x) i x ip 0 = N N exp c p( x )dx N h h h

updated 9/12/02 11:39 AM

5.73 Lecture #6

6-9

so, if we neglect N , we get 2pN + pN = 0 if p 0, then 2p d Np1/2 dx

1/2 1/2

) = Np1/2 + 1 p1/2 pN )=0

2

1 1/2 N + p p N = 0 2

d Np1/2 dx

N(x)p1/2 (x) = constant

N(x) = cp(x)1/2
OK, now we have a form for N(x) which we can use to tell us what conditions must be satisfied for N(x) to be negligible everywhere.

N = cp 1/2 1 dp 1/2 dp = p 3/2 2 dx dx dp dV 1 = p m dx dx m dV dp 1/2 = p 5/2 2 dx dx

2 d 2 p 1/2 m dV 5 7/2 m dV 5/2 m d V = p p dx + p 2 2 2 dx 2 { dx dx 2

p(x) = [2m( E V(x))]

1/2

ignore
updated 9/12/02 11:39 AM

5.73 Lecture #6
dV 2 5 N = c m 2 p 9/2 dx 4

6 - 10

But we have made several assumptions about N

* N << * N <<

2ip icm 3/2 dV p N = + h h dx ip icm 3/2 dV N= p h h dx

p2 c * N << 2 N = 2 p +3/2 h h
all of this is satisfied if

5 mh dV 3 p << 1 4 i dx
Is this the JWKB validity condition? Spirit of JWKB: if initial JWKB approximation is not sufficiently accurate, iterate:

p(x) 0 (x) 0 (x) p1 (x) p1 (x) 1 (x)

(ordinary JWKB) (first order JWKB)

h2 d20 e .g. + 2 0 = 0 p1 ( x) = 2 h dx2 0 ( x) dx i x 1/2 1 ( x) = p1 ( x) exp p1 ( x ) dx h c d20

2 p1

1/2

see ** Eq. on p. 6-8

iterative improvement of accuracy

p1(x) is not smaller than p0(x). It has more of the correct wiggles in it.
END OF NONLECTURE

updated 9/12/02 11:39 AM

5.73 Lecture #6
Resume Lecture

6 - 11

(x) p(x) { provided that

1/ 2

envelope

i exp h

p(x ) dx
adjustable phase shift.

d2V is negligible dx 2

AND

dp d m dV << 1 same as (x) < p( x) or << 1 3 dx dx dx |p| 14 244 4 3

h

required for N (x ) to be negligible

Next need to work out connection of JWKB(x) across region where JWKB approx. breaks down (at turning points!).
d at turning point because p(x) 0 dx

BUT ALL IS NOT LOST near enough to a turning point all potentials V(x) look like V(x)=x! Now our job is to show that asymptotic AIRY and JWKB are identical for a small region not too close and not too far on both sides of each turning point.

THIS PERMITS ACCURATE SPLICING OF (x) ACROSS TURNING POINT REGION!

updated 9/12/02 11:39 AM

5.73 Lecture #6

6 - 12

V(x) E

II
V(x) = x >0
approx. linear V(x)

a-AIRY

a-AIRY

JWKB

JWKB

a
Region I E > V(x) classical
2 I aAIRY ~ 1/2 (z)1/ 4 sin (z)3/2 + 3 z= (x E) 2m h2
1/3

x - E at turning point E = V(a) = a so = (x a)

2m z = (x a) 2 h
Region II

1/3

<< 0

when x << a

Region I/II splice using a-Airy.

E < V(x) forbidden region, z >> 0

II aAIRY ~
1/2

z 1/ 4 e 2/3z

3/ 2

Now consider JWKB for a linear potential and show that it is identical to a-Airy!
JWKB ~ c p(x)
1/2

i x exp p( x )dx h a

both c+ and c additive terms could be present

p(x) [2m( E V(x))]

1/2

updated 9/12/02 11:39 AM

5.73 Lecture #6
x < a x > a classical , forbidden , p is real , JWKM p is imaginary , JWKB oscillates is exponential

6 - 13

pretend V(x) looks linear near x = a

(l-JWKB )
linear

p(x) = [2m(a x )]

1/2

a p( x)dx = (2m )
= (2m )

1/2 1/2 x a (a x) dx 1/2

2 (a x )3/2 3 a

2 = (2m )1/2 (a x )3/2 3

Region I

I JWKB (x) ~ p(x) l = p(x) 1

h

1/ 2

[Ae

+ Be i

]
2 ( a x) 3 / 2 3

1/ 2

C sin( + )
1/ 2

Define the JWKB phase factor, (x): =

2m p(x ) dx = 2 h

Now compare (x) to z(x)

2m but, earlier, z = (x a) 2 h
1/2

1/3

2 = (z)3/2 3
for exponential factor for pre-exponential factor

p = (2m h)1/3 ( z)1/2 p = (2m h)1/6 ( z)1/ 4

Thus, putting all of the pieces together

updated 9/12/02 11:39 AM

5.73 Lecture #6
|p| 6 74 4 8 6444 7444 8 4 4 2 I lJWKB = (2mh)1/6 (z)1/ 4 C sin (z)3/2 3 I = aAIRY
If C = (2m h)1/6 1/2
1/ 2

6 - 14

= 4
I I lJWKB exactly splices onto aAIRY with a /4 phase factor (shifted from what the argument of sine would have been if one had started the phase integral at x = a

Similar result in Region II

f (x) II + Be +f (x) JWKB ~ Ae

at x +
II lJWKB

f(x) = A(2m)
1/ 4

B = 0 (x a)
1/ 4

2m 1/2 2 exp 2 (x a)3/2 3 h

II which is equal to aAIRY if A = (2mh)+1/6 1/2 2

I II II Final step: I JWKB aAIRY , JWKB aAIRY

require A = C/2 perfect match on opposite sides of turning point.

Ai(z) valid in region where JWKB is invalid.

updated 9/12/02 11:39 AM

5.73 Lecture #7
JWKB QUANTIZATION CONDITION

7-1

Last time:
1. V( x ) = x i Ep p 3 6m (p) = N exp h ( x ) = Ai(z) * zeroes of Ai, Ai * tables of Ai (and Bi) * asymptotic forms far from turning points

2.

Semi-Classical Approximation for (x)

* p(x) = [( E V(x))2m ]
1/2

classical wavefunction

i x 1/2 exp * (x) = p(x) p(x )dx h 243 envelope 14c wiggly- variable k ( x )

modifies classical to make it QM wavefunction

adjustable phase for boundary conditions

without differential equation qualitative behavior of integrals (stationary phase) d * validity: << 1 valid not too near turning point. dx
[One reason for using semi - classical wavefunctions is that we often need to evaluate integrals of the type exp i O p jdx. If O p is a slow function of x, the phase factor is
*

i d [pj ( x) pi ( x)]dx. Take dx = 0 to find x s.p.. x is range about x s.p. over which h phase changes by / 2. Integral is equal to I( x s.p. )x.]

Logical Structure of pages 6-11 to 6-14 (not covered in lecture): 1. JWKB not valid (it blows up) near turning point cant match s on either side of turning point. Near a turning point, x(E), every well-behaved V(x) looks linear
V(x) V( x + (E)) + dV (x x + ) dx x=x +

2.

first term in a Taylor series.

This makes it possible to use Airy functions for any V(x) near turning point.
updated 9/16/02 11:29 AM

5.73 Lecture #7
3.

7-2

asymptotic-Airy functions have matched amplitudes (and phase) across validity gap straddling the turning point. JWKB for a linear V(x) is identical to asymptotic-Airy!

4.

TODAY 1. Summary of regions of validity for Airy, a-Airy, l-JWKB, JWKB on both sides of turning point. This seems complicated, but it leads to a result that will be exceptionally useful! WKB quantization condition: energy levels without wavefunctions! compute dnE/dE (for box normalization can then convert to any other kind of normalization) trivial solution of Harmonic Oscillator Ev = h (v+1/2) v = 0, 1, 2

2. 3.

4.

Non-lecture (from pages 6-12 to 6-14)

1/12 1/ 2

classical

2 m a AIRY = 1/12 2 h a AIRY = lJWKB = lJWKB =

2 2m (a x)1/ 4 sin 2 3 h

(a x)3/ 2 +

forbidden

1/12 2m 2 h 2 C

1/12

2 2m 1/ 2 x a )1/ 4 exp 2 ( x a )3/ 2 ( 3 h

1/ 4

classical

(a x )

2 2m 1/ 2 sin 2 (a x )3/ 2 + 3 h 2 2m 1/ 2 exp 2 ( x a )3/ 2 3 h

forbidden

(x a )

1/ 4

C, D, and are determined by matching. These Airy functions are not normalized, but each pair has correct relative amplitude on opposite sides of turning point. l-JWKB has same functional form as a-Airy. This permits us to link pairs of JWKB functions across invalid region and then use JWKB to extend (x) into regions further from turning point where linear approximation to V(x) is no longer valid.
updated 9/16/02 11:29 AM

5.73 Lecture #7
Regions of Validity Near Turning Point E = V(x(E))

7-3

I CLASSICAL

II FORBIDDEN

(x)

x+ (E)

| |
OK at t.p.
LINEAR V(x)
AIRY but not too OK at edges far of Airy aAIRY region OK far enough JWKB from t.p. OK at onset of JWKB region

Common Validity of l-JWKB and a-AIRY ASYMPTOTIC AIRY

LINEAR V(x) EXACT JWKB

lJWKB

| |

LINEAR V(x) JWKB

Common region of validity for a-AIRY and l-JWKB same functional form, specify amplitude and phase for JWKB(x) valid far from turning point for exact V(x)!

updated 9/16/02 11:29 AM

5.73 Lecture #7
Quantization of E in Arbitrary Shaped Wells I

7-4
II
V ( x ) E + ( x b) V ( x+ ) = E x+ ( E ) = b >0

V( x ) E ( x a ) V( x ) = E x ( E ) = a

III

>0

a
x (E) = a

b
x + (E) = b

Already know how to splice across I, II and II, III but how do we match s in a < x < b region?
C 1/ 2 h x p(x )dx p(x) e 2
1 a

Region I

I JWKB (x) =

x<a

(forbidden region) (real, no oscillations)

Note carefully that argument of exp goes to as x , thus I() 0. Note also that (I/C) increases monotonically as x increases up to x = a.

When you are doing matching for the first time, it is very important to verify that the phase of varies with x in the way you want it to.
updated 9/16/02 11:29 AM

5.73 Lecture #7
Region II 1 x IIa JWKB (x) = C p(x) 1/ 2 sin a p(x )dx + h 4

7-5
a<x<b

The first zero is located at an accumulated phase of (3/4) inside x=a because (3/4 + 1/4) = and sin = 0. It does not matter that IIa is invalid near x = a, x = b Note that phase increases as x increases - as it must. The /4 is the extra phase required by the AIRY splice across I,II. It reflects the tunneling of (x) into the forbidden region. PHASE starts at /4 in classical region and always increases as one moves (further into classical region) away from turning point. NEVER FORGET!

Region III

C 1/ 2 h b p(x ) dx III JWKB (x) = p(x) e 2

x

x>b

Note that phase advances (i.e. the phase integral gets more positive) as x .

III JWKB decreases monotonically to 0 as x +.

Region II again

1 b IIb JWKB (x) = C p(x) 1/ 2 sin x p(x )dx + h

note: argument of sine starts at /4 and increases as one goes from x = b inward. In other words, opposite to IIa , the argument decreases from left to right!
But IIa (x) = IIb (x) for all a < x < b !

2 ways to satisfy this requirement

1.

sin( (x) ) = sin[( (x)) + (2n + 1)] AND C = C { 123 4

argument of IIa argument of IIb

[sin = sin(),

sin( + (2n + 1)) = sin ,

sin = sin( + (2n + 1))]

updated 9/16/02 11:29 AM

5.73 Lecture #7
2.
sin( (x)) = sin[ (x) + 2n ] if C = C

7-6

now look at what the 2 cases require for the arguments 1. C = C

1 xpdx + = 1 bpdx + + (2n + 1) h a h x 4 4
IIa (x)

IIb (x) + (2n + 1)

1 x b a + x pdx = (2n + 1) 4 4 h

ap(x)dx = h[2n + 1 2]
2.
C = C get
b

Quantization.

ap(x)dx = h[2n 1 2]
a p( x)dx = h( n + 1 2) ** WKB quantization
b

combine the two:

n = 0,1, 2, C = C(1)n

Most important result of this lecture.

condition.

n is # of internal nodes because argument always starts at /4 and increases inward to (n + 3/4) at other turning point. inner t.p. outer t.p.

for n = 0 n = 1

s i n ( /4) sin(3 / 4 ) s i n ( /4) sin(7 / 4 ) etc.

NO NODE! 1 node.

Node count tells what level it is. pdx at arbitrary Eprobe tells how many levels there are at E Eprobe!

updated 9/16/02 11:29 AM

5.73 Lecture #7
Density of States
n(E) =
dn dE

7-7
dn h dE is the classical period of oscillation.

but p E ( x ) 0

2 x + (E) 1 x (E) p E (x)dx 2 h dn 2 dx dx x dp = p E ( x + ) + p E ( x ) + x + E dx dE dE h dE dE dn 2 x + d = x [2m(E V(x))]1/ 2 dx dE dE h

(must take derivatives of limits of integration as well as integrand)

x+ dn 2 1 = (2m) [2m(E V(x))]1/ 2 dx dE h 2 x

you show that, for harmonic oscillator

V(x) = 1 2 kx 2
1/2

(k m) that dn 1 = dE h

independent of E, thus period of h.o. is independent of E.

Non-lecture
for general box normalization

x+ = LLL:

can still use this to compute

dn because dE

dx + = 0 (even though p E ( x + ) 0). dE

location of right hand turning point is independent of E.

Can always use WKB quantization to compute density of box normalized Es, provided that E > V(x) everywhere except the 2 turning points.

updated 9/16/02 11:29 AM

5.73 Lecture #7

7-8

Use WKB to solve a few standard problems. Since WKB is semi-classical, we expect it to work in the n limit. Could be some errors for a few of the lowest-n Ens. Harmonic Oscillator
1/ 2

V(x) = kx2/2

(k is force constant, not wave vector)

1 p( x ) = 2m E kx 2 2 At turning points, V ( x tp ) = E and p( x tp ) = 0, thus, at turning points x = [2E n k ] because E n = 1 2 kx 2

x + =[ 2E n k ]
1/ 2 1/ 2
1/ 2

V(x)

h( n + 1 / 2) =

=[ 2E n k ]

[ (

2m E n kx 2 2

)]

1/2

dx

Non-lecture: Dwight Integral Table 350.01

t a 2 x2

xt a 2 1 tdx = 2 + 2 sin (x / a) here t = 0 at both x + and x-

1/2

I = (2mk 2)1/2

[2E

[2E n k ]1/ 2
n k] 1/ 2

2E n k x 2

1/2

dx

2E I = (2mk 2)1/2 n
sin 1 1 sin 1 (1) k

1/2

1/2 1/2
m
E 2 2 = m
E I= ) (
))
n n ((
k

use the nonlecture result:

m h( n + 1 / 2) = k

En

1/2 k ( n + 1 / 2) En = h m 4 4 12 3

updated 9/16/02 11:29 AM

5.73 Lecture #7

7-9

I suggest you apply WKB Quantization Condition to the following problems: See Shankar pages 454-457.
Vee quartic
l = 0, H atom

V(x) = a|x|
V(x) = bx 4 V(x) = cx 1 1 V(x) = kx 2 2

E n ( n + 1 / 2)2/3 E n ( n + 1 / 2)4/3 E n n 2

harmonic

E n ( n + 1 / 2)1

What does this tell you about the relationship between the exponents m and in Vm xm and En n? power of x in V(x) 1 1 2 4 power of n in E(n) 2 2/3 1 4/3

Validity limits of WKB? * *

splicing of IIa , IIb ?
JWKB is bad when

d2V dx 2

cant be too large near the splice region ( changes by more than itself for x = )

d >1 dx ~

near turning points and near the minimum of V(x)

* cant use WKB QC if there are more than 2 turning points d2V d near bottom of well 2 is not small and dx > 1 * dx (near both turning points). However, most wells look harmonic near minimum and WKB gives exact result for harmonic oscillator - should be more OK than one has any right to expect. * semi-classical: should be good in high-n limit. If exact En has same form as WKB QC at low-n, WKB En is valid for all n. H.O., Morse Oscillator
updated 9/16/02 11:29 AM

5.73 Lecture #8

Rydberg Klein Rees

8-1

Last time: WKB quantization condition for bound eigenstates of almost general V(x) Connection into bound region from left and right
x + (E) x (E) pE (x )dx = 2 (n + 1/ 2)

pE (x) = [ 2m(E V(x))]

1/2

En without n !

But where do we get V(x)? Certainly not from femtochemistry!

From FREQUENCY DOMAIN SPECTROSCOPY

E v,J V(x)

timing of w.p. as it moves on V(x)

RKR method

Next time: Numerical Integration of 1-D Schr. Eq. see handouts Then begin working toward matrix picture Need background in Ch.2 of CTDL pages 94-121 soon, pages 121-144 by next week Postulates and theorems not to be covered except as needed for solving problems.

Today:

E v,J spectroscopic notation A(E, J) =

E V(x)

dV Equilibrium: = 0 xe dx

xe

A A 1 1 , x + (E) x (E) and x + (E) x (E) E J

WKB QC applied to

A A , G( v),B(v) used to determine x (E). E J

Long Range Theory: Ultra Cold Collisions: Atom in Molecule

updated 9/18/02 8:57 AM

5.73 Lecture #8

Rydberg Klein Rees

8-2

Someday you will discover that the energy levels of a diatomic molecule are given by

E evJ / hc = T e + electronic

= e + Y 00 + e (v + 1 / 2) e x e (v + 1 / 2)2 +
G(v) B(v)

vibration

G(v)

rotation

F v (J)

cm-1

+ [Be e (v + 1 / 2)+]J(J + 1) DJ 2 (J + 1)2

RKR requires only G(v) and B(v) to get VJ(x)

h 2 J(J + 1) where V J (x) = U(x) + J-dependent bare 2x 2

effective potential potential centrifugal barrier (actually rotational kinetic energy)

x R Re
= m1m2 m1 + m2

We are going to derive V 0 (x) directly from G(v), B(v) data. This is the only direct spectrum to potential inversion method! WKB quantization is the basis for this.

x (E

x + (E v )
v

pE v ( x)dx = (h / 2)(v + 1 / 2)

v = 0,1,# of nodes

In this equation, what we know (Ev) and what we want (V(x) and x at turning points) are hopelessly mixed up. There is a trick!

A(E, J)

x + (E,J)

x (E,J)

[E V J (x)]dx
x(E) x+(E) area at E
updated 9/18/02 8:57 AM

5.73 Lecture #8

Rydberg Klein Rees

8-3

but, still, we know neither V J (x) nor x (E, J)! !

A A and are numerically evaluable J E integrals (via WKB QC) involving only E v,J info data input here A A 2 . independently, and determine J E 1 1 2 eqs. in 2 [x + (E, J) x (E, J)] and x (E, J) x (E, J) unknowns give +

Roadmap: 1 . Show that

turning points

Do #2 first because it is so easy

A x + (E,J) = E E x (E,J)
= x +(E,J) 1dx

h 2 J(J + 1) E U( x ) dx 2x 2 + 0 + 0
x (E, J) E

x (E,J)

contributions from

are zero because integrand is 0 at turning points

A = x + (E, J) x (E, J) E

A x + (E,J) = J x (E,J) J

h 2 J(J + 1) E U( x ) 2 dx 2x

h 2 x + (E,J) 2J + 1 dx + 020 = 1+ 3 2 x (E,J) x 2 integrand = 0 at x

h2(2J + 1) 1 A 1 =+ x (E, J) x (E, J) J 2 +
So, if we can evaluate these derivatives from EvJ data, we have VJ(x)!
updated 9/18/02 8:57 AM

5.73 Lecture #8

Rydberg Klein Rees

8-4

some clever manipulations to put A(E,J) into convenient form (see nonlecture notes on pages 8-5,6,7)
A(E, J) =
x + (E,J)

x (E,J)

[E V J (x)]dx
1/2

2h 2
A(E, J) = 2

E E
vJ v(E24 {

min ,J) 14 3
data
data
v(E,J)

1/2

dv

skipped steps are shown on pages 8-5, 6, 7.

this integral could be evaluated at any E, but we really only A A and . Evaluate these derivatives at J = 0. want
E J
2h2 A = 2 E
1/ 2 1 / 2 1 v( E, J ) v( E min , J ) [E EvJ ] dv + 0 + 0 2

1 Y v(E min , J) = 00 2 e

integrand = 0 at upper limit

lower limit independent of E

defined so that G(vmin) = 0

G(v) = Y 00 + e ( v + 1 / 2) 0 = G(v min ) = Y 00 + e ( v min + 1 / 2) Y 00 = v min + 1 / 2 e Y 1 v min = 00 e 2
for J = 0 E = G(v) v,J

[vmin is slightly different from 1/2]

A 2h 2 = E

1/ 2

data

v(E) 1/ 2 1/ 2Y 00 / e [ E G(v)] dv 2f(E)

data

evaluate this integral numerically at any E. [Singularity at upper limit fixed by change of variable: Zeleznik JCP 42, 2836 (1965).]
updated 9/18/02 8:57 AM

5.73 Lecture #8

Rydberg Klein Rees

8-5

2h 2 v(E) A E = [E G(v)]1/ 2 dv + 0 + 0 J J=0 1/ 2Y 00 / e J E = BJ J(J + 1)

E = Bv (2J + 1) J

E = Bv J J=0

2h 2 A = J J=0

1/ 2

data h2 1/ 2 [E G(v)] Bvdv 2g(E) 2 1/ 2Y 00 / e data

v(E)

(again, a nonfatal singularity at upper limit)

f(E) and g(E) are Klein action integrals which are jointly determined by
empirical G(v) and B(v) functions.

Nonlecture derivation of this useful form of 2h 2 A(E,J) = 2

1/ 2

v(E

v(E ,J)
min

[ E EvJ ]1/ 2 dv ,J )

Begin here:

A(E, J) = x +( E,J ) [ E V J (x )]dx

x (E,J)

integral identity

2 x a 1/2 ba= dx a b x

let

b= E a = VJ ( x ) x = E vJ

x - a E vJ VJ ( x ) = so that b- x E E vJ

x (E,J) A(E, J) = x +(E,J) [ b a]dx

Now insert the integral identity

updated 9/18/02 8:57 AM

5.73 Lecture #8
x (E,J)

Rydberg Klein Rees

8-6
put in values of a, b, and x

+ 2 b x a 1/ 2
A(E, J) = dx dx a b x x (E,J)
1/ 2 + E E V (x ) 2 J = dE dx vJ vJ
E E
vJ
x (E,J) V J (x) x (E,J)

reverse order of integration and recognize WKB QC in disguise

1/2 x + (E,J) E V J (x ) 2 vJ = dx dE vJ E E vJ x (E,J) V J (x) E

numerator of dx integral is QC insert QC and then integrate by parts. denominator is independent of x, so insert QC
+ + [E V(x)]1/2 dx = (2)1/2 p(x)dx x (E,J) x

x (E,J)

= (2)1/2

h (v + 1/ 2) 2

v(E , J) + 1/ 2 1/2 h 2 dE A(E, J) = (2 ) 1/2 vJ 2 vJ E min E E

**

** integrate by parts

)1/2 1/2 f = 2(E E ) vJ g = [ v(E , J) + 1/ 2] vJ

f = E E vJ g = dv , dE
2h 2 + 1 24 4 3
1/ 2

(not a typo because variable is EvJ not E)

which is known from E vJ

A(E, J) =

E=E fg E=E min

=0 at both limits

E dv 2( E E )1/ 2 dE E min dE

(caution: f and g here are not Kleins action integrals)

updated 9/18/02 8:57 AM

5.73 Lecture #8
dv = dv dE dE

Rydberg Klein Rees

8-7

** change variables from dE to dv

v(E,J)

limits of integration become finished:

2h 2 A(E, J) = 2
1/ 2

v( E min ,J )

[E EvJ ]1/ 2 dv v( E min ,J )

v(E,J)

we have two independent evaluations of f(E) and g(E) one leads to

x + (E, 0) x (E, 0) = 2f(E) 1 1 = 2g(E) x + (E, 0) x (E, 0)

1/2 pair of turning f(E) x (E, 0) = f(E) / g(E) + f(E)2 points

from quadratic formula

so we get a pair of turning points at each E. Not restricted to Es with integer vs!

V(x)

connect the dots!

Robert LeRoy: modern, n-th generation RKR program at http://theochem.uwaterloo.ca/~leroy/ Download program, instructions, and sample data.

RKR does not work for polyatomic molecules because E V Q r ~ does not determine the multicomponent vector P

( )

updated 9/18/02 8:57 AM

5.73 Lecture #8
Long-Range Molecules

Rydberg Klein Rees W.C. Stwalley CPL 6, 241 (1970)

(weak perturbation of atomic properties)

8 - 8

R.J. Bernstein & R.LeRoy JCP 52, 3869 (1970)

hard turning point

soft turning point

What does (x) look like at very high v? * lots of nodes (v nodes) * small lobe at inner turning point. Why? * large lobe at outer turning point. Why?
H int: Force = dV(x) dx

at sufficiently large v, it is certain that (x) is dominated by o u t e r-most lobe and any expectation value of a function of x, such as V(x), will be dominated by the outer turning point region. Since the vibrational Schrdinger equation contains V(x), it is evident that Ev at high v should be determined primarily by the long range part of V(x) (and insensitive to details near x e and at the inner turning point).

What do we know about covalent bonding?

ATOMIC ORBITAL OVERLAP IS REQUIRED!
NO OVERLAP at large x, V(x) determined by properties of isolated
a t o m s: dipole moment, polarizability return to this later when we
do perturbation theory.

It is always possible to predict the longest range term in V(x) = C n x n where the longest range term is the one with SMALLEST n.

updated 9/18/02 8:57 AM

5.73 Lecture #8

Rydberg Klein Rees

8-9
x+(vD) =

Quick review of the Long-Range Theory x(vD) x(v)

v E vD E v

x+(v)

vD is noninteger v of level at top of potential v is binding energy of v-th level

xe at J = 0
V0 (x ) = U(x ) = Cnx n U( ) = 0 E v D U( x e ) = De x + (v ) = ( Cn E v ) x + ( vD ) =
1/n

at long range (large x)

[E

= V( x + (v)) = Cn x n +

binding energy: v = E v D E v = Cnx n +

How many levels are there in potential?

+ D h ( vD + 1 / 2) = 2 x (v D )

x (v

) =

p D (x )dx .

Now we do not know v D , Cn , or D, but we do know n and know that E v will be primarily determined by long-range part of V(x) near v D . So, for any Ev we expect that it will be possible to derive a relationship between

( v D v)

# of levels below highest bound level

and

(E v

Ev

binding energy

by some clever tricks you may discover on Problem Sets #4 and 5, we find
n2 v D v = a n v2n

Tells us how to plot Ev vs. v to extrapolate to vD and then to obtain accurate value of De from a linear plot near dissociation.

updated 9/18/02 8:57 AM

5.73 Lecture #8
Power n=1 2 3 3

Rydberg Klein Rees

8 - 10

of longest range term in V(x): charge - charge charge - dipole charge - induced dipole dipole - dipole (also transition dipoles) 5 dipole - induced dipole 5 quadrupole - quadrupole 6 induced dipole - induced dipole

+, point charges (e.g. H atom) + H + H Na 2 S + Na 2P

( )

( )

I( P3/ 2 ) + I( 2 P3/ 2 )
2

Na 2 S + Na 2 S

( )

( )

not only is the limiting n known, but also Cn is known because it is calculable from a measurable property of the free atom. Many molecular states are described at long range by the same Cn 's ! Ultra-cold collisions now used to determine V(x) to very large x. Now best route to the properties of separated atoms! Mostly, long-range theory has been used as a guide to extrapolation to accurate dissociation energy (relevant to H ). Now Bose condensates. f Molecule trapping.
x 1 and x 2 potentials have number of bound levels.
x 3 , x 5 , x 6 potentials

have finite number, and the number of levels breaks off more abruptly as n increases.

high n

low n

i.e., # of bound levels

action integral affected more by wider classical x region than by deeper E binding region because p (EV(x))1 / 2

updated 9/18/02 8:57 AM

5.73 Lecture #8

Rydberg Klein Rees

8 - 11

This means (equation at bottom of 8-9) that if we plot (given that we can predict n with certainty) correct D: STRAIGHT LINE wrong D: guess for Ev D too high

vD

wrong D: guess for Ev D too low

(Ev
n2 2n

n2 E v 2n

guessed

known

n 3 5 6 7

it is possible to determine D and v D very accurately much better than Birge Sponer plot,which
is valid only for a Morse potential

1/6 3/10
1/3 5/14

Birge-Sponer: G vs. v G(v+1/2)=G(v+1) G(v)

long linear extrapolation to vD

v vD G( v) = e (v + 1 / 2) e x e (v + 1 / 2) 2 G(v + 1 / 2) = G(v + 1) G(v) = e e x e (2v + 2) decreasing to 0 as v increases Morse Potential when G(v + 1 / 2) = 0, e = e x e (2v + 2) 2 e V0 ( x ) = D[1 e Ax ] vD = 1 v D is noninteger # of bound vibrational levels 2 e x e 0 for Morse
D = G( v D ) = 1 2 e e xe 4 e xe e e xe ( v D + 1) e 2 4 2
updated 9/18/02 8:57 AM

= ( v D + 1)

But Morse inevitably has incorrect long-range form

5.73 Lecture #8

Rydberg Klein Rees

8 - 12

Which is longer range? Morse or Cnxn? Take ratio of binding energy at large x.
lim lim C n x n C n x n = x D 1 e Ax 2 D x De 2Ax 2De Ax

lim C n x n e 2Ax x D 2De Ax lim C n n Ax = x e x 2D =

dominant term

This means that Morse binding energy gets small faster than Cnxn for any n.

Morse

Cnxn

G(v+1) G(v) will get small faster for Morse. Plot G(v + 1/2) vs. v.

G(v+1) G(v)

Morse vD

True vD

linear Birge-Sponer

Dissociation energy usually underestimated by linear Birge-Sponer extrapolation. Long-range plot of correct power of E v D E v gives more accurate dissociation energy.
updated 9/18/02 8:57 AM

5.73 Lecture #9
Numerov-Cooley Method : 1-D Schr. Eq.
Last time: Rydberg, Klein, Rees Method and Long-Range Model G(v), B(v) rotation-vibration constants VJ(x) potential energy curve x = R Re

9-1

Ev,J, v,J, all conceivable experiments

wp( x, t ) = ai i e iEi t / h
i
initial preparation of wp: a i =

*i [wp(x,0)]dx

determined by VJ(x) free evolution of wp

Method: A(E,J) = area of V(x) below E:

used WKB QC

V(x)

obtained x(E,J) Today: What do we do when we have VJ(x) (especially when V(x) is not suited for WKB)? Solve Schr. Eq. numerically! No models 15 digit reproducibility cheap This is the final tool we will develop for use in the Schrdinger representation. To summarize the classes of 1D problem we have solved: * piecewise constant potentials (matrix approach for joining at boundaries) * Airy functions (linear potential and joining JWKB across turning point) * JWKB (quantization condition and semi-classical wavefunctions) * numerical integration (today)

updated 9/18/02 8:54 AM

5.73 Lecture #9
Numerical Integration of the 1-D Schrdinger Equation

9-2

widely used incredibly accurate no restrictions on V(x) or on EV(x) [e.g. nonclassical region, near turning points, double minimum potential, kinks in V(x).] For most 1-D problems, where all one cares about is a set of {Ei, i}, where i is defined on a grid of points xi, one uses Numerov-Cooley See 1. Cooley, Math. Comput. 15, 363 (1961). 2. Press et. al., Numerical Recipes, Chapters 16 and 17 Handouts 1. Classic unpublished paper by Zare and Cashion with listing of Fortran program (now see LeRoy web site) 2. Tests of N-C vs. other methods by Tellinghuisen

Basic Idea: grid method

* solve differential equation by starting at some xi and propagating trial solution from one grid point to the next

* apply (x) = 0 BCs at x = 0 and by two different tricks and then force agreement at some intermediate point by adjusting E.

updated 9/18/02 8:54 AM

5.73 Lecture #9
Eulers Method h want (x) at a series of grid points x0, x1, xn=x0+nh call these i = (xi) x0 x1 h

9-3
h x2

Need a generating function f(xn, n)

n+1 = n + hf ( x n , n )

increment in x xn+1 xn = h [NOT Plancks constant]

prescription for going n n + 1 must depend on both xn and n. xn samples potential, n samples previous value of .

(n is a number, not the entire wavefunction.) For the Euler method, the generating function is simply:

f (x n , n ) =

d n n+1 n n+1 = dx x n x n+1 x n h

The value of this derivative actually comes from the differential equation that must satisfy, not from prior knowledge of (x) (which we do not yet have!)

For the Schrdinger Eqn.

d 2 2 2 = 2 ( E U ( x )) dx h

All constants absorbed in V(x). V(x) Must be in units of 2.

d2 = V(x)(x) dx 2 i d i+1 h dx x i

V ( x ) C[U ( x ) E ]

U ( x ) is potential. h is increment of distance, in . E and U(x) are in cm-1 units (E / hc)

m1m 2 m1 + m 2

C = 10 16 (8 2c h ) = 0.0593203146 A

(amu ,

12

d2 i i i1 h = i+1 2 h h dx x
i

A =

= h 2 [ i+1 2 i + i1 ]
updated 9/18/02 8:54 AM

5.73 Lecture #9

9-4

Schr. Eq. tells us the rule for propagating . Employing Eulers method (h is not Plancks constant): h 2 i+1 2 i + i1 = V i i

i+1 2 i + i1 = h 2 V i i i+1 = 2 i i1 + h 2 V i i
a recursion relationship. Need both i and i1 to get i+1.

in order to get things started we need two values of starting at either edge of the region where is defined and starts out very small. See Press et. al. handout for discussion of nth-order Runge-Kutta method. The generator is chosen more cleverly than in the Euler method so that stepping errors are minimized by taking more derivatives at intermediate points in the xi, xi+1 interval.
Cooley specifies

y i +1 = 2 y i y i 1 + h 2 Vi i y i = 1 ( h 2 12)Vi i (and vice versa)

* use i to get yi * use i and yi (and yi1 ) to get yi+1 * use yi+1 to get i+1

The result is that the error in y i+1 is on the order of h6 i V i smaller error if h is smaller 240

(much better than Euler)

updated 9/18/02 8:54 AM

5.73 Lecture #9
So what do we do?

9-5

De

R(E) Re x R Re

R+(E)

e.g., VMORSE (R ) = De 1 e (R R e ) V ( ) = 0 , V ( R e ) = De

De

at R = 0 R=

x = Re

(Re) = 0 boundary conditions () = 0

2 boundary conditions handled differently because we want to define a finite # of equally spaced grid points (not actually necessary see Press: variable grid spacing which is needed to sample infinite range of x with a finite number of grid points)

* at R = 0

0 0

(required)

1 = 1020 (arbitrarily chosen small number to be corrected later upon normalization)

use this to start the integration outward. If we have made a wrong choice for 1, this can be corrected merely by dividing all i i 1 by an i-independent correction factor.
updated 9/18/02 8:54 AM

5.73 Lecture #9

9-6

At large R (the classically forbidden region), choose n at the last grid point, xn, to be small and use WKB only once to compute the next to last grid point. We do this because we have no reason to go to x .

n = 1030

(the final grid point)

n 1 e = R (V )1/2 n e n n
recall JWKB =| p |
1/ 2

R n1 (Vn1 )1/2

The next to final grid point [This is the only place WKB enters into this problem!]

1 x | p| dx h R+ (E )

1 p n ~ Vn/ 2

numerator denominator

1 xn 1 p n1 exp p n1 dx h R+ (E ) 1 xn 1/ 2 pn exp p n dx h R+ (E )
1/ 2

pre-exponential factors are approximately equal integrals in exponential factors are evaluated as summations in n1 /n, the common terms in the summations in the exponential factors cancel

Once n-1 is generated from n by JWKB, return to Cooleys method of numerical integration for all successive grid points.

So now we propagate one from i = 0 out toward right and the other one from i = n in toward the left. The shooting method.

i = 0

n
updated 9/18/02 8:54 AM

5.73 Lecture #9

9-7

Stop the inward propagation of when a point is reached where, for the first time, m m+1 . Since |i| is exponentially increasing from 1030 at i=n until it reaches its first maximum inside the classically allowed region, this outer lobe of is also the most important feature of (because most of the probability resides in it). outermost lobe of maximum of || at
( x m+1 )
xm+1

V(x) asymptotic approach to 0

Use outermost lobe because this is the global maximum of (x), this minimizes the problem of precision being limited by finite number of significant figures in the computer.

Set value of m = 1.0 by renormalizing both functions

* from n, n 1, m (from the right) * from i = 0, 1, m (from the left) M replace each i by i m

for all i down to m. M replace each i by i m for all i up to m.

i = i m

= 1 m

must be continuous, even at the joining grid point, m.

The renormalized s are denoted by .

updated 9/18/02 8:54 AM

5.73 Lecture #9

9-8

This ensures that (x) is continuous everywhere and that it satisfies grid form of Schr. Eq. everywhere except i = m

0 = ( y i+1 + 2y i y i1 ) + h 2 V i i
In order to satisfy Schr. Eq. for i = m, it is necessary to adjust E. The above equation can be viewed as a nonlinear requirement on E. At the crucial grid point i = m, define an error function, F(E).

E E E E E F( E ) = y m+1 + 2y m y m1 + h 2 V m m
where we want to search for zeroes of F(E).

Assume that F(E) can be expanded about E1 (E1 is the initial, randomly chosen value of E.)

F( E ) = F( E1 ) +
Call this E2

dF (E E1 ) + discard higher terms dE E1

and solve for the value of E where F(E) = 0.

dF 0 = F ( E1 ) + ( E 2 E1 ) dE E1

F( E1 ) E2 = + E1 (dF dE )E
1
Correction to E1

This gives an estimate of where the zero of F(E) nearest E1 is located.

updated 9/18/02 8:54 AM

5.73 Lecture #9
Usual approach: compute F( E1 + ) F( E1 ) dF = dE E1

9-9

Once the derivative is known, use it to compute correction to E1 (assuming linearity).

Newton-Raphson method for solving nonlinear equation

E 2 = E1 +

F( E1 ) (dF dE )E

Iterate until the correction, , to E is smaller than a pre-set convergence criterion .

Now we have an eigenfunction of H and eigenvalue, E.

Normalize E by dividing by * dx
2 * dx = i h i=0
box normalized:

1/2

= NE

integral evaluated by summation over grid points.

E (xi ) =

2 j h j

1/2

real for bound 1-D function

This procedure has been used and tested by many workers. A good version, Level 7.1 (schrq. f), is obtainable at Robert LeRoys web site:

http://theochem.uwaterloo.ca/~leroy/

I will assign some problems based on Numerov-Cooley method for integrating the 1-D Schr. Eq.

updated 9/18/02 8:54 AM

5.73 Lecture #10

Matrix Mechanics should have read CDTL pages 94-121 read CTDL pages 121-144 ASAP Last time:

10 - 1

Numerov-Cooley Integration of 1-D Schr. Eqn. Defined on a Grid. 2-sided boundary conditions nonlinear system - iterate to eigenenergies (Newton-Raphson) {Ei, i(x)} V(x)

So far focussed on (x) and Schr. Eq. as differential equation. Variety of methods

Often we want to evaluate integrals of the form

overlap of special on standard functions {} expectation values transition moments

* ( x) i ( x) dx = ai

a is mixing coefficient {} is complete set of basis functions

OR
*i xn jdx (xn )ij

There are going to be elegant tricks for evaluating these integrals and relating one integral to others that are already known. Also selection rules for knowing automatically which integrals are zero: symmetry, commutation rules Today: begin matrix mechanics - deal with matrices composed of these integrals focus on manipulating these matrices rather than solving a differential equation - find eigenvalues and eigenvectors of matrices instead (COMPUTER DIAGONALIZATION)

H I G * Perturbation Theory: tricks to find approximate eigenvalues of infinite matrices H L * Wigner-Eckart Theorem and 3-j coefficients: use symmetry to identify and interrelate values of nonzero integrals I G H * Density Matrices: information about state of system as separate from measurement operators T S

updated 9/18/02 8:55 AM

5.73 Lecture #10

First Goal:

10 - 2

Dirac notation as convenient NOTATIONAL simplification It is actually a new abstract picture (vector spaces) but we will stress the utility ( | relationships) rather than the philosophy!

Find equivalent matrix form of standard (x) concepts and methods. 1. Orthonormality 2. completeness (expand )

*i j dx = ij
(x) is an arbitrary function

A. Always possible to expand (x) uniquely in a COMPLETE BASIS SET {}

(x) =

a ii (x)
i
mixing coefficient how to get it? * left multiply by j

*
B.

a i = *dx i

(expand B)

Always possible to expand B in {} since we can write in terms of {}. So simplify the question we are asking to Bi = b jj What are the b j ? Multiply by * j j

{ }

b j = *B i dx B ji j

B i = B ji j
j

note counter-intuitive pattern of indices. We will return to this.

* The effect of any operator on i is to give a linear combination of js.

updated 9/18/02 8:55 AM

5.73 Lecture #10

3. Products of Operators
(AB)i = A(Bi ) = A Bji j
j

10 - 3

can move numbers (but not operators) around freely note repeated index = B A = B A
ji j ji kj k j j k

= A kjB ji k = (AB)ki k
j,k k

* Thus product of 2 operators follows the rules of matrix multiplication:

AB acts like A B

Recall rules for matrix multiplication:

indices are A row,column

must match # of columns on left to # of rows on right

order matters!

(N N ) (1 N )
" row vector"

(N N ) ( N 1)
" column vector"

(N N ) (1 1) (N N )

a matrix a number a matrix!

( N 1)
column vector

(1 N )
row vector

Need a notation that accomplishes all of this memorably and compactly.

updated 9/18/02 8:55 AM

5.73 Lecture #10

Diracs bra and ket notation Heisenbergs matrix mechanics ket
a1 a is a column matrix, i.e. a vector 2 M aN

10 - 4

contains all of the mixing coefficients for expressed in some basis set. [These are projections onto unit vectors in N-dimensional vector space.] Must be clear what state is being expanded in what basis
ai

(x) = *dx i (x) i

* 1 dx * expressed in basis * * a column of complex #s 2 dx = * nothing here is a function of x M * dx N bookkeeping device (RARE)
OR, a pure state in its own basis 0 1 2 = 0 one 1, all others 0 M 0

[ i

bra

is a row matrix

( b 1,b 2 b N )

contains all mixing coefficients for * in * basis set * (x ) =

[ i * dx ]*i (x )
i

{ }

(This is * of (x) above)

The * stuff is needed to make sure = 1 even though i is complex.

updated 9/18/02 8:55 AM

5.73 Lecture #10

The symbol a|b, a braket, is defined in the sense of product of (1 N) (N 1) matrices a 1 1 matrix: a number! Box Normalization in both and | pictures

10 - 5

1 = * dx

= *dx i i * = j * dx * j
j

( i

expand both in basis

1 = * dx = 1=

j * dx ( *i dx ) *j idx
i, j 2

*j dx
j

c.c.

ij p forces 2 sums to collapse into 1

real, positive #s

We have proved that sum of |mixing coefficients|2 = 1. These are called mixing fractions or fractional character.

now in

picture

* 1dx = 1 * dx 2 * dx L * dx 2 1444444 444444 2 3 M 3 row vector: bra 1 24 4 column vector ket

*dx j

same result

[CTDL talks about dual vector spaces best to walk before you run. Always translate into picture to be sure you understand the notation.]
updated 9/18/02 8:55 AM

5.73 Lecture #10

Any symbol is a complex number. Any symbol | | is a square matrix.

10 - 6

again = ( 1

1 2 ) 2 M

= i i = = 1
i

1 1 0 L 0 what is 1 1 = (1 0 0) = 0 0 M O M 0 1 1 0 what is i i = 0 1 i O
unit or identity matrix = 1

1 unit matrix

a shorthand for specifying only the important part of an infinite matrix

completeness or closure involves insertion of 1 between any two symbols.

updated 9/18/02 8:55 AM

5.73 Lecture #10

10 - 7

Use 1 to evaluate matrix elements of product of 2 operators, AB (we know how to do this in picture). square matrix i-th 0 : i A j = (0 1 0)(A) 1 j-th position picks : out j-th column of A i-th 0
A1 j = (0 1 0) A 2 j = A ij M

i AB j = i A k k B j
k

= A ik Bkj = (AB)ij a number

k

In Heisenberg picture, how do we get exact equivalent of (x)? basis set (x,x0) for all x0 this is a complete basis (eigenbasis for ^, eigenvalue x0) - perfect localization at any x0 x x is the same thing as (x) x is continuously variable (x) overlap of state vector with (x) a complex number. (x) is a complex function of a real variable.
i.e.,

( x, x ) * ( x ) dx = ( x)

updated 9/18/02 8:55 AM

5.73 Lecture #10

other relationships

10 - 8

1. All observable quantities are represented by a Hemitian operator (Why because expectation values are always real). Definition of Hermitian operator.

A ij = A* ji
2. Change of basis set

or

A = A

Easy to prove that if all expectation values of A are real, then A = A and vice-versa

A A u

{} to {u}

Aij i A j = i 1A1 j

=
1st index is u

k,l

i
u

uk u k A u l ul j

S* S ki ik

Slj

S is Frequently used to denote overlap integral u j-th column of S

k,l

Sik AulSlj = S Au S k

ij

Aij

A = S Au S

a special kind of transformation (unitary) (different from usual T1AT similarity transformation)

updated 9/18/02 8:55 AM

5.73 Lecture #10 What kind of matrix is S?

10 - 9

Sl j = u l | j

* Sl j = u l | j * = j |u l Sj l

means take complex conjugate and interchange indices. Using the definitions of S and S :

SljS = u l | j j | u k jk SljS = jk

u l | j j | u k = u l | 1 | u k

= u l | u k = lk = 1 lk

SS = 1

OR

S = S1 Unitary

a very special and convenient property.

updated 9/18/02 8:55 AM

5.73 Lecture #10

10 - 10

Unitary transformations preserve both normalization and orthogonality.

A = SA uS SAS = SSA uSS = A u A u = SAS Take matrix element of both sides of equation: A u = u i | A | u j = (SAS ) ij ij =

k ,l

S ik k | A | l Slj

uj =

l S lj

j - th column of S

u via S ,S : u j is j- th column of S

updated 9/18/02 8:55 AM

5.73 Lecture #10

10 - 11

Thus,
0 S1 j M S 2j = 1 = M M M 0 Snj

j-th

uj

Alternatively, A = p|A|q = (S Au S) pq pq =
q =

mn

Spm um |A|u n Snq

u n Snq

q - th column of S

q
q-th

0 S1q M S 2q = 1 = M M S nq u 0

updated 9/18/02 8:55 AM

5.73 Lecture #10

Commutation Rules
[A, B] = AB BA e.g. [ x, p] = ih = ih
If A and B are Hermitian, is AB Hermitian?
Hermitian A and B

10 - 12

means ( xp - px ) = x

h d h d +x i dx i dx

( AB)ij = A ik Bkj = A* B* = B* A* = (BA )* ki jk jk ki ji

k
but this is not what we need to say AB is Hermitian: That would be:

(AB)ij = (AB)* ji
AB is Hermitian only if [A,B] = 0 However, 1 [AB + BA] is Hermitian if A and B are Hermitian. 2

This is the foolproof way to construct a new Hermitian operator out of simpler Hermitian operators. Standard prescription for the Correspondence Principle.

updated 9/18/02 8:55 AM

5.73 Lecture #11

11 - 1

Eigenvalues, Eigenvectors, and Discrete Variable Representation (DVR) should have read CDTL pages 94-144 Last time:

bra

(a

* 1

* aN

in {} basis set a1 0 a M 2 i = 1 = M M M 0 a N a j= j i

ket

a1 M aN N N matrix

(complex ) #
1 1 1 = 0
at end of lecture

k
k

i AB j = i A k k B j 1 24 4 3 k
1

= A ik Bkj = (AB)ij
k

updated 10/1/02 11:00 AM

5.73 Lecture #11

11 - 2

What is the connection between the Schrdinger and Heisenberg representations?

i ( x) = x i x 0 = ( x, x 0 )
eigenfunction of x with eigenvalue x0

Using this formulation for i(x), you can go freely (and rigorously) between the Schrdinger and Heisenberg approaches.

1=

k k =

x x dx

Today:

eigenvalues of a matrix what are they? how do we get them? (secular equation). Why do we need them?

eigenvectors how do we get them? Arbitrary V(x) in Harmonic Oscillator Basis Set (DVR)

updated 10/1/02 11:00 AM

5.73 Lecture #11

Schr. Eq. is an eigenvalue equation = a in matrix language
A i = ai i 0 a1 a2 A= O aN 0 1 0 1 = 0 M

11 - 3

satisfies A 1 = a 1 1

but that is the eigen-basis representation a special representation! What about an arbitrary representation? Call it the representation.
N A c i i = a ci i * * * * * * i =1 i A as transformation on each i

Eigenvalue equation

N unknown coefficients {c i }

How to determine {c i } and a ? Secular Eqn. derive it.

first, left multiply by j

i = 1 to N

A c
i

ji i N

=a

c
i ji

j i =a

c
i

i ij

0=

c [A
i i =1

a ij

one equation

N unknowns

next, multiply original equation by k

0=

c [A
i i =1

ki

a ik

another equation

etc. for all

N linear homogeneous equations in N unknowns Condition that a nontrivial (i.e. not all 0s) solution exists is that determinant of coefficients = 0.

updated 10/1/02 11:00 AM

5.73 Lecture #11

11 - 4
A12 A 22 a O A1N

A11 a 0= A 21

A NN a
Nth order equation as many as N different values of a satisfy this equation (if fewer than N, some values of a are degenerate). Does everyone know how to expand a determinant? {ai} are the eigenvalues of A (same as what we would have obtained by solving differential operator eigenvalue equation)

If we know the eigenvalues, then we can find the N { i

} such that

i =

cj j

expand the eigenbasis in (computationally convenient) basis

i A j = a j ij 0 a1 A = 0 O aN A 1 1 0 = a1 0 M

But we generally start with A in nondiagonal form

1. 2. 3. transform to diagonal form by T A T = A the diagonal elements are eigenvalues the diagonalizing transformation is composed of eigenvectors, column by column of T .

computer

updated 10/1/02 11:00 AM

5.73 Lecture #11

Hermitian Matrices
A = A A = A* ji ij

11 - 5

(can use this property to show that all expectation values of A are real) These matrices can be diagonalized (i.e. the set of all eigenvalues can be found) by a unitary transformation.

T 1 = T

unitary matrix

0 a1 a2 A diagonalization T A T = O diagonal not diagonal 0 aN

T A TT j
diagonal A Eigenvector expressed in basis set

updated 10/1/02 11:00 AM

5.73 Lecture #11

eigenvector

11 - 6

0 0 T1i M M T T i = T 1 = 2 i = i = 1 i - th position M M M T i-th Ni position 0 0

i-th column of T

suppose we apply

A i

0 M = A 1 M i-th 0

0 M = ai M 0

0 M = ai 1 M 0

RECAPITULATE: Start with arbitrary basis set Construct A : Not Diagonal, but basis set was computationally convenient. a1 0 0 = A O Find T (computer) that causes T A T = aN 0

Eigenstates (eigenkets) are columns of T in basis set.

Columns of T are the linear combination of eigenvectors that correspond to each basis state. Useful for bright state calculations.

updated 10/1/02 11:00 AM

5.73 Lecture #11

Can now solve many difficult appearing problems!

11 - 7

Start with a matrix representation of any operator that is expressable as a function of a matrix.

e.g.

e iH( t t 0 )/h propagator

f ( x) potential curve

prescription example

f ( x ) = Tf (T xT)T 123

x1 T xT = 0

diagonalize x so f( ) is applied to each diagonal element

x2

0 O xN

f (x1 ) 0 f (x 2 ) f (T xT) = O 0 f (x N )
Then perform inverse transformation T f(T x T)T undiagonalizes matrix, to give matrix representation of desired function of a matrix. Show that this actually is valid for simple example

f(x) = x N f(x) = T TxT TxT L TxT T

[(
[

= T T x T T = xN

1 N

)(

) (

)]

apply prescription

get expected result

general proof for arbitrary f(x) expand in power series. Use previous result for each integer power.

updated 10/1/02 11:00 AM

5.73 Lecture #11

11 - 8

John Light: Discrete Variable Representation (DVR) General V(x) evaluated in Harmonic Oscillator Basis Set. we did not do H-O yet, but the general formula for all of the nonzero matrix elements of x is:
h n x n +1 = 2
1/2

( n + 1)1/2

= (k )

1/2

(infinite dimension matrix)

1/2 h x= 2

0 1 0 M M

1 0 2 M M

0 2 0 3 M

L L 3 0 4

L 0 0 L 0 0 0 = 0 0 4 0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

x 2 = h 2

1 0 2 0 0 0 0 0 0

0 3 0 6 0 0 0 0 0

2 0 5 0 12 0 0 0 0

0 6 0 7 0 0 0 0

0 0 0 0 0 0 12 0 0 0 0 0 0 0 0 9 0 0 0 0 11 0 0 0 13 0 0 0 15 O 0 0 O O 0 0 0 0

[CARTOON]

etc. matrix multiplication to get matrix for f(x) diagonalize e.g., 1000 1000 (truncated) x matrix that was expressed in harmonic oscillator basis set.

updated 10/1/02 11:00 AM

5.73 Lecture #11

x1 0 TxT = 0 0 0 0 0 x2 0 0 0 O 0 0 0 x 1000 x

11 - 9
diagonalized-x basis {xi} are eigenvalues. They have no obvious physical significance.

0 0 0 V( x 1 ) 0 0 V( x 2 ) 0 V( x ) x = 0 O 0 0 0 0 0 V( x 1000 ) full V( x ) H O = TV( x ) x T = complicated matrix

H= p2 + V(x) 2

next transform back from x-basis to H-O basis set

1000

T was determined when x was diagonalized

1000 HO

need matrix for p2, get it from p (the general formula for all non-zero matrix elements of p)

updated 10/1/02 11:00 AM

5.73 Lecture #11

h() n p n + 1 = i ( n + 1)1/2 2 0 1 0 0 2 1/2 1 h() p = i 2 0 2 0 0 O 0 0 O 0 1 0 2 0 0 3 0 O h() 2 0 5 0 p2 = 2 0 O O O 0 O O O
if H= p2 1 2 + kx 2 2
1/2

11 - 10

0 0 0 0 0 O O O O 0 0 O O 0

same structure as x

1 k = 1 2 2 2

2 h 0 H= 4 0 0

0 0 0 1 / 2 0 0 0 0 0 6 0 0 = h 0 3 / 2 0 10 0 0 0 5/ 2 0 0 0 14 0 0 O 0

but for arbitrary V(x), express H in HO basis set,

2 pHO H HO = + V( x ) HO 2

E1 0 0 eigenvalues obtained by S H HOS = 0 O 0 0 0 EN

TV(x)xT

columns of S are eigenvectors in HO basis set!

updated 10/1/02 11:00 AM

5.73 Lecture #11

1. Express matrix of x in H-O basis (automatic; easy to program a computer to do this), get xHO. 2. Diagonalize xHO. Get xx and T. 3. Trivial to write V(x)x as V(xi) = V(x)x in x basis 4. Transform V(x)x back to V(x)HO 5. Diagonalize HHO. Solve many V(x) problems in this basis set.

11 - 11

1000 1000 T matrix diagonalizes x 1000 xis Save the T and the {xi} for future use on all V(x) problems. To verify convergence, repeat for new x matrix of dimension 1100 x 1100. There will be no resemblance between {x i}1000 and {x i}1100 . If the lowest eigenvalues of H (i.e. the ones you care about) do not change (by measurement accuracy), converged! Next: Matrix solution of HO (no wave functions at all) Start from Commutation Rule!

Then Perturbation Theory.

updated 10/1/02 11:00 AM

5.73 Lecture #12

Matrix Solution of Harmonic Oscillator
Last time:
a1 0 0 * T A T= 0 O 0 0 0 aN

12 - 1

T1i * eigenbasis i = M T Ni

i th column of T

* matrix of function of matrix is given by Tf T xT T * Discrete Variable Representation: Matrix representation for any 1- D problem

TODAY: Derive all matrix elements of x, p, H from [x,p] commutation rule and definition of H. Example of how one can get matrix results entirely from commutation rule definitions (e.g. of an angular momentum: J2, Jx, Jy, Jz and Wigner-Eckart Theorem) NO WAVEFUNCTIONS, NO INTEGRALS, ALL MAGIC! Outline of steps: 1. Assumptions

p 2 kx 2 + 2m 2 * eigen basis exists for H * H= * [x, p] = ih * x and p are Hermitian (real expectation values)

2. 3. 4. 5.

xnm and pnm in terms of (EnEm) xnm in terms of pnm Block Diagonalize x, p, H Lowest quantum number must exist (call it 0) explicit values for
x 01
2

6. Recursion relationship for xnn1 and pnn1 7. Magnitudes and phases for xnn1 and pnn1 8. Possibility of noncommunicating blocks along diagonal of H, x, p eliminated See CTDL pages 488-500 for similar treatment. IN MORE You will never use this methodology - only the results! ELEGANT NOTATION

and p 01

updated 9/20/02 11:44 AM

5.73 Lecture #12

1. recall assumptions 2. x and p matrix elements derived from Comm. Rules
p2 1 2 1 1 + kx = x,p2 = [ x,H] = x, (p[x,p] + [x,p]p) 2m 2m 2 2m

12 - 2

[ ]

** [ x,p] = ih

[ x, H ] =

p ih 2ih = p 2m m

m p = [ x,H] ih take matrix elements of both sides, insert completeness between x and H
m p nm = ( x nl H lm H nl x lm ) ih l

similarly, starting from [p, H] = p, 1 kx 2 = ihkx 2

x nm = i ( p nl H lm H nl plm ) kh l

but we know that some basis set must exist where H is diagonal. Use it implicitly: replace Hlm by Emml
m p nm = ( x nm E m E n x nm ) ih m p nm = x nm ( E m E n ) ih

pnn = 0 (but, in addition, if H has a degenerate eigenvalue, then pnm = 0 if En = Em) similarly for
x nm = i p nm ( E m E n ) hk

xnn = 0 (and xnm = 0 if En = Em)

updated 9/20/02 11:44 AM

5.73 Lecture #12

3. solve for xnm in terms of pnm
multiply the x nm equation by p nm multiply the p nm equation by x nm
The LHSs of both resulting equations are equal

12 - 3

equate RHS:

i 2 m 2 x nm ( E m E n ) = p nm ( E m E n ) ih hk

* If En = Em (degeneracy) then we already know that xnm = 0, pnm = 0

1 2 * If En Em x 2 = p nm nm km

x nm = i(km)1/ 2 p nm
THERE IS A PHASE AMBIGUITY HERE!

m x nm ( E m E n ) ih m i( km )1/2 p nm ( E m E n ) plug in new result for xnm p nm = ih Either

earlier we derived

p nm =

(OK to divide thru by pnm )

* *

p nm 0 AND E m E n = h( k m ) OR p nm = 0 x nm = 0

1/ 2

h !!

The only non-zero off-diagonal matrix elements of x and p involve eigenfunctions of H that have energies differing by exactly h! A selection rule! The only nonzero matrix elements of x and p are those where indices differ by 1.

updated 9/20/02 11:44 AM

5.73 Lecture #12

4. x, p, H are block diagonal ized

12 - 4

In what sense? There is a set of eigenstates of H that have energies that fall (1) onto the comb of evenly spaced E n E (1) = n(h ) + 1 n could be another set

E (2) = n( h ) + 2 n

where 2 1 nh
all n

but within each set, there must be a lowest energy level

Set I

E(1) is lowest 0
(1) E1 = E(1) + h 0

II

etc.

Set II

E (2) is lowest 0 (2) E1 = E(2) + h 0

etc.

interleaved but noncommunicating?

Since x and p have nonzero elements only within communicating sets for H, thus x, p, H are block diagonalized into sets I, II, etc.

I 0 0 0 0 II 0 0 H, x, p = 0 0 III 0 0 0 0 O
We will eventually show that all of these blocks along the diagonal are identical (and that each energy level is nondegenerate). If x, p are block diagonal, then x2, p2 are similarly block diagonal.
updated 9/20/02 11:44 AM

5.73 Lecture #12

5. A lowest index must exist within each block. Call it 0.

12 - 5

[x,p] = ih is a diagonal matrix (x nl plm p nl x lm ) = ih nm

l

ih = ( x nn+1p n+1n p nn+1x n+1n ) + ( x nn1p n1n p nn1x n1n )

must be equal

These are the only surviving nonzero terms in the sum over l!

but there must be a lowest Ei because E = T + V and T 0, E Vmin

let n = 0 be lowest index p0,1 = x0,1 = 0

x 01p10 p 01x10 = ih
used Hermiticity here

x,p are Hermitian A = A thus x 01p* p 01x* = ih 01 01

previously x nm = i( km )1/2 p nm

(note that the same symbol is used for mass and basis state index)

we must make phase choices so that x and p are Hermitian

updated 9/20/02 11:44 AM

5.73 Lecture #12

12 - 6

phase ambiguity: we can specify absolute phase of x or p BUT NOT BOTH because that would affect value of [x,p] BY CONVENTION: matrix elements of x are REAL p are IMAGINARY tryx = + i ( km) 1/2 p and eliminate p by plugging this into 01 01 01
* * x01 p01 p01 x01 = ih

get

x 01 p 01

2 2

1/ 2 h = ( km ) 2

+1/ 2 h km ) ( 2

If we had chosen x 01 = i( km )1/ 2 p 01 we would have obtained x 2 = h ( km )1/ 2 which is impossible! 01 2

two things that must be checked for self-consistency of seemingly arbitrary phase choices at every opportunity: * Hermiticity

*
6. Recursion Relation for x ii+1
2

start again with general equation derived in #3 above using the phase choice that worked in #5 above

x nn +1 = i( km )1/ 2 p nn +1
index going up Hermiticity
1/2 * x* p n+1n n+1n = i( km )

c.c. of both sides index going down

x n +1n = i( km )1/ 2 p n +1n

updated 9/20/02 11:44 AM

5.73 Lecture #12

x nn1 = i( km )1/2 p nn1

12 - 7

now the arbitrary part of the phase ambiguity in the relationship between x and p is eliminated Apply this to the general term in [x,p] lots of algebra NONLECTURE : from four terms in [x,p] = ih
( km )1/2 * x nn+1p n+1n = x nn+1p* = x nn+1 x nn+1 nn+1 i = x nn+1
2

(+i(km)1/2 )
)

( km )1/2 2 1/2 p nn+1x n+1n = x nn+1 x* nn+1 = x nn+1 +i( km ) i

( km )1/2 * x nn1p n1n = x nn1p* x nn1 nn1 = x nn1 + i = x nn1

2

(i(km)1/2 )
)

( km )1/2 2 1/2 p nn1x n1n = x nn1 x* nn1 = x nn1 i( km ) i

combine 4 terms in [x,p] = ih to get

ih = 2i( km )1/2 x nn+1 x nn1 h(km)1/2 2 + x nn1 2 2 2 h but x 01 = x10 = ( km )1/2 2 x nn+1 =
2

]
recursion relation
h

each step up produces another additive term:

( km) 1/2

thus

h 2 x nn+1 = ( n + 1) ( km )1/2 2 h 2 p nn+1 = ( n + 1) ( km )+1/2 2

general result

updated 9/20/02 11:44 AM

5.73 Lecture #12

7. Magnitudes and Phases for xnn1 and pnn1 verify phase consistency and hermiticity for x and p
in # 3 we derived x nn1 = i( km )1/2 p nn1

12 - 8

one self1/2 consistent set is h 1/2 x nn+1 = +( n + 1) = +x n+1n 2( km )1/2 x real 1/2 and h 1/2 = +x nn1 positive x nn1 = +( n ) 2 km 1/2 ( ) AND
p imaginary with sign flip for up vs. down

Note that nonzero matrix elements of x and p are always [larger quantum number]1/2 This is the usual phase convention Must be careful about phase choices because one never really looks at wavefunctions, operators, or integrals 8. Possible existence of noncommunicating blocks along diagonal of H, x, p

1/2 1/2 p nn+1 = i( n + 1)1/2 h( km ) = p n+1n 2 1/2 1/2 1/2 h( km ) = p n1n p nn1 = +i( n ) 2

you show that H nm

k = ( n + 1 / 2)h m

1/2

nm

note that x 2 and p2 have nonzero n = 2 elements but 2 1 kx 2 + p has cancelling contributions in n = 2 locations 2 2m
This result implies * all of the possibly independent blocks in x, p, H are identical * i = (1/2)h for all i * degeneracy of all En? all En must have same degeneracy, but cant prove that it is 1.

updated 9/20/02 11:44 AM

5.73 Lecture #13

End Matrix Solution of H-O, a + a Operators p2 1 2 starting from H = + kx and [x,p] = ih 2m 2 showed p nm = m x nm ( E m E n ) ih i x nm = p nm ( E m E n ) hk x nm x nn = 0, p nn = 0 and = 0 if E n = E m p nm
x2 = nm 1 2 p nm km

13 - 1

1.

2.

3.

E m E n = ih

= ( k m )1/2

the only non-zero x and p elements are between states whose Es differ by h 4. 5.
(i) combs of connected states, block diag. of H, x, p, x2, p2 E n = hn + i

lowest index must exist because lowest E must exist. Call this index 0 2 h x 01 = (km)1/2 2 from phase choice 2 h +1/2 p 01 = (km) x 01 = +i(km)1/2 p 01 2 Today

6.

Recursion Relationship

x nn+1

in terms of x nn-1

2 2 2

general matrix elements x nn+1 , p nn+1

7. 8. general x and p elements
only blocks correspond to i = 1 h 2

Dimensionless x, p, H and a (annihilation) and a (creation) operators

updated 9/27/02 9:22 AM

5.73 Lecture #13

13 - 2

phase ambiguity: we can specify absolute phase of x or p BUT NOT BOTH because that would affect value of [x,p] BY CONVENTION: matrix elements of x are REAL p are IMAGINARY try

x 01 = +i( km )1/2 p 01 and plug this into x 01p* p 01x* = ih 01 01

get

x 01 p 01

2 2

1/ 2 h = ( km ) 2

+1/ 2 h km ) ( 2

If we had chosen x 01 = i( km )1/ 2 p 01 we would have obtained x 2 = h ( km )1/ 2 which is impossible! 01 2

check for self-consistency of seemingly arbitrary phase choices at every opportunity: * Hermiticity

*
6. Recursion Relation for x ii+1
2

start again with gerand equation derived in #3 above using the phase choice that worked above 1/2 x nn+1 = i(km) p nn+1 going up Hermiticity
1/2 * x* p n+1n n+1n = i( km )

c.c. of both sides going down

x n +1n = i( km )1/ 2 p n +1n

updated 9/27/02 9:22 AM

5.73 Lecture #13

x nn1 = i( km )1/2 p nn1

13 - 3

now the arbitrary part of the phase ambiguity in the relationship between x and p is eliminated Apply this to the general term in [x,p] algebra NONLECTURE : from terms in [x,p] = ih
( km )1/2 * x nn+1p n+1n = x nn+1p* = x nn+1 x nn+1 nn+1 i = x nn+1
2

(+i(km)1/2 )
)

( km )1/2 2 1/2 p nn+1x n+1n = x nn+1 x* nn+1 = x nn+1 +i( km ) i

( km )1/2 * x nn1p n1n = x nn1p* x nn1 nn1 = x nn1 + i = x nn1

2

(i(km)1/2 )
)

( km )1/2 2 1/2 p nn1x n1n = x nn1 x* nn1 = x nn1 i( km ) i

ih = 2i( km )1/2 x nn+1 x nn1 h(km)1/2 2 + x nn1 2 2 2 h but x 01 = x10 = ( km )1/2 2 x nn+1 =
2

]
recursion relation

thus

h 2 x nn+1 = ( n + 1) ( km )1/2 2 h 2 p nn+1 = ( n + 1) ( km )+1/2 2

general result

updated 9/27/02 9:22 AM

5.73 Lecture #13

7. Magnitudes and Phases for xnn1 and pnn1 verify phase consistency and Hermiticity for x and p
in # 3 we derived x nn1 = i( km )1/2 p nn1

13 - 4

one self1/2 consistent set is h 1/2 x nn+1 = +( n + 1) = +x n+1n 2( km )1/2 x real 1/2 and h 1/2 = +x nn1 positive x nn1 = +( n ) 2 km 1/2 ( )
1/2 1/2 p nn+1 = i( n + 1)1/2 h( km ) p imaginary = p n+1n 2 w/sign flip for up vs. down 1/2 1/2 1/2 h( km ) = p n1n p nn1 = +i( n ) 2

( km)1/2 = m

AND

Phase is a recurrent problem in matrix mechanics because we never look at This is the usual phase convention wavefunctions or evaluate integrals explicitly.

8. Possible existence of noncommunicating blocks along diagonal of H, x, p

1/2 k you show that H nm = ( n + 1 / 2)h nm m note that x 2 and p2 have non - zero n = 2 elements but 2 1 p kx 2 + has cancelling contributions in n = 2 locations 2 2m

This result implies * all of the possibly independent blocks in x, p, H are identical * i = (1/2)h for all i * degeneracy of all En? all same, but cant prove that it is 1.

updated 9/27/02 9:22 AM

5.73 Lecture #13

Creation and Annihilation Operators (CTDL pages 488-508) * * * * *

13 - 5

Dimensionless operators simple operator algebra rather than complicated real algebra matrices arranged according to selection rules matrix elements calculated by extremely simple rules automatic generation of any basis function by repeated operations on lowest (nodeless) basis state

get rid of system-specific factors of k, , and also h.

= ( k /m)

1/2

dimensionless

m x x ~ h regular 1/2 p p (hm )

~

1/2

h 2 x2 = x m ~ p2 = hm p2
~

We choose these factors to make everything come out dimensionless.

1 1 2 2 H= x +p H= ~ ~ ~ h 2

1 2 p2 1 = h x2 + p2 H = kx + 2 2m 2 ~ ~

1/2 1 m 1 x, p = [x, p] = h (ih) = i ~ ~ h hm

from results for x, p, H

x mn = 2 1/ 2 ( n + 1) mn +1 + n1/ 2 mn 1 ~
1/ 2

p mn = 2 1/ 2
~

[ i[( n + 1)

1/ 2

mn +1 n1/ 2 mn 1

] ]

square root of larger q.n.

H mn = ( n + 1 / 2) mn ~

note the negative sign

diagonal

Kronecker - s specify selection rules for nonzero matrix elements

now define something new more!

a, a to clean things up even

updated 9/27/02 9:22 AM

5.73 Lecture #13

a = 2 1/2 x + i p ~ ~ a = 2 1/2 x i p ~ ~

13 - 6

( ) p = 2 1/2 i(a a ) ~
x = 2 1/2 a + a ~

Lets examine the matrix elements of a and a

a mn = 2 1/2 x + 2 1/2 i p mn ~mn ~ 1 1 1 1 = ( n + 1)1/2 mn+1 ( n + 1)1/2 mn+1 + n1/2 mn1 + n1/2 mn1 2 2 2 2
group according to selection rule

x ~
cancel

ip
~

x ~
add

ip
~

a mn = n1/2 mn1

first index is one smaller than second column

a mn = m a n

row similarly

n1/2 n 1

a is lowering or annihilation operator

a = ( n + 1)1/2 mn+1 mn

first index is one larger than second a is a creation operator

updated 9/27/02 9:22 AM

5.73 Lecture #13

0 0 0 1 11/ 2 1 2 3 4

13 - 7
0 11/ 2 0 0 a = 0 0 0 0 0 0 0 0 0 41/ 2 0

a =2 0 3 0 4 0

0 0 21/ 2 0 0

0 0 0 31/ 2 0

0 0 0 0 41/ 2

0 0 0 0 0

0 21/ 2 0 0 0

0 0 31/ 2 0 0

square root of integers always only one step below main diagonal. a, a are obviously not Hermitian

square root of integers always only one step above main diagonal

e.g. 3 a 2 = 31/ 2 a raises

What is so great about a, a?
a n = n1/2 n 1 a n = ( n + 1)1/2 n + 1 annihilates 1 quantum creates 1 quantum

e.g. 3 a 4 = 4 1/ 2 a lowers

n n = [ n!]1/2 a 0

( )

generate any state from lowest one 0

needed to normalize. Each application of a gives next integer. Do it n times, get n!

more tricks: look at aa and aa is aa Hermitian? [(AB) = BA] definition of hermitian

(aa )

= aa = aa

aa and aa are Hermitian to what observable quantity do they correspond? We will see that one of these is the number operator.

updated 9/27/02 9:22 AM

5.73 Lecture #13

1 2 1 2 x + i p x i p = x + i p x i x p+ p ~ ~ ~ ~ ~~ ~ ~ ~ ~ 2 2

13 - 8

aa =

1 1 = x 2 + p 2 i[ x , p] = x 2 + p 2 + 1 ~ ~ ~ ~ 2 ~ ~ { 2 i 1 similarly a a = x 2 + p 2 1 2~ ~

1 H = a a + aa ~ 2 H = a a + 1 / 2
~

)
)

and

[a,a ] = 1 simple form for H

number operator + 1/2

H = h H = h a a + 1 / 2 ~ a a n = n n

# of quanta

a a is "number operator"

[aa

n = (n + 1) n

not as useful

What have we done? We have exposed all of the symmetry and universality of the HO basis set. We can now trivially work out what the matrix for any XnPm operator looks like and organize it according to selection rules.

updated 9/27/02 9:22 AM

5.73 Lecture #13

What about X3?
m X = h
3 ~ 3/2

13 - 9
When you multiply this out, preserve the order of a and a factors.
3

X3 ~
3

X3 = 2 3/2 a + a

n =

( )( ) = (2 3/2 )a 3 + (aaa + aaa + aaa ) + (aaa + aaa + aaa ) + a

3, 1, +1, +3

(# of minus # of non-)

Simplify each group using commutation properties so that it has form a a a n or a a a n

[ ]

[ ]
1/2

1/2

n n 1

( n + 1)

n n +1

NONLECTURE: Simplify the n = 1 terms.

aaa = aaaaa4 +44 = aaa a 14 a a 3 2 aa

aaa = aa aaa4 +44 = aaa + a 14 a aaa 2 3

[ ]

a ,a a=a a a,a =a

add and subtract term needed to reverse order

[aaa + aaa + aaa ] = 3aaa

NONLECTURE: Simplify the n = +1 terms.

[ ]

try to put everthing into aaa order

aaa = aaa a44 +44a = aaa + a 1 aa 2 aa 3 a aa = aaaa44 +44 = aaa + a 1 a a a aa 2 3

[a,a ]a =a [ ]

[aaa + aaa + a a a] = 3a a a + 3a
updated 9/27/02 9:22 AM

a a,a =a

5.73 Lecture #13

n = 3
a3 n = [( n + 1)( n + 2)( n + 3)] a 3 n = [ n( n 1)( n 2)]
1/2

13 - 10
n3
1/2

n+3

n = 1

[aa a

[a

aa + aa a + aaa n = 3 n 3/ 2 n 1 + a aa
1/ 2

] ( ) + a a a ] n = 3[(n + 1)
= 3(n + 1) = 3(n + 1)

n n + 1 + (n + 1)

1/ 2

n +1

1/ 2 3/ 2

(n + 1) n + 1
n +1

no need to do matrix multiplication. Just play with a, a and commutation rule and aa number operator

Second Quantization
same as |n+3n|
mn+3

h x3 = mn 2m

3/2

(( n + 1)( n + 2)( n + 3))1/2

|n+1n|
1/2

+ mn+1 3( n + 1)3/2 + mn1 3n 3/2

+ mn3 ( n( n 1)( n 2))

|nn3|

|nn1|

simple! x3 is arranged into four separate terms, each with its own explicit selection rule.

V(x) =

1 2 kx + ax 3 2bx 4 1 +4 4 3 2 anharmonic terms perturbation theory

2

IR transition intensities n x n + 1

* Survival and transfer probabilities of initially prepared pure harmonic oscillator non-eigenstate in anharmonic potential. * Expectation values of any function of x and p.

updated 9/27/02 9:22 AM

5.73 Lecture #13

13 - 11

Universitality: all k,m (system-specific) constants are removed until we put them back in at the end of the calculation.

e.g., What is x
2
h h

= x2 x

]
2

1 ( )2 x= x = a +a m ~ m 2 x
2

[ (a + a ) 2m
h

a + a

]?

pure numbers in [ ]

updated 9/27/02 9:22 AM

5.73 Lecture #14

Perturbation Theory I
(See CTDL 1095-1107, 1110-1119)

14 - 1

Last time: derivation of all matrix elements for Harmonic-Oscillator: x, p, H selection rules scaling dimensionless quantities
n xij n xii i n /2

i j n in steps of 2

(e.g. x 3 : n = 3, 1)

m 1/2 x= x ~ h p = (hm )1/2 p

~

H= ~

1 H h

annihilation creation number commutator

a = 2 1/ 2 x + i p ~
~

a = 2 1/ 2 x i p ~
~

a a( not aa )
[a,a] = +1

a n = n1/ 2 n 1

a n = (n + 1) a a n = n n

1/ 2

n +1

modified 9/30/02 10:13 AM

5.73 Lecture #14

14 - 2

a little more:
a 01 = 11/2

0 0 a = 0 0 0

1 0 0 0 0

0 2 0 0 0

0 0 0 0 3 0 0 O 0 0 0 1/ 2 ( n + 1)! 1! L L 0

(one step to right of main diagonal)

0 L ( n!)1/ 2 L 0 0 L a n = L L L 0 L
n selection rule for a ij

0 L 1/ 2 ( n + q )! q!

(n steps to right)

j -i = n j - i = n

selection rule

for a ij

n = [ n!]
m

1 / 2

(a )

0
1/ 2

[(a ) (a) ]
n

jk

( k!) ( j!) = j, k n + m 123 ( k n )! ( j m )! 4 4 selection rule

Selection rules are obtained simply by counting the numbers of a and a and taking the difference.

The actual value of the matrix element depends on the order in which individual a and a factors are arranged, but the selection rule does not. Lots of nice tricks and shortcuts using a, a and aa

modified 9/30/02 10:13 AM

5.73 Lecture #14

14 - 3

One of the places where these tricks come in handy is perturbation theory. We already have: 1. WKB: local solution, local k(x), stationary phase 2. NumerovCooley: exact solution - no restrictions 3. Discrete Variable Representation: exact solution, as linear combination of H-O.

Why perturbation theory?

replace exact H which is usually of dimension by Heff which is of finite

dimension. Truncate infinite matrix so that any eigenvalue and eigenfunction can be computed with error < some preset tolerance. Fit model that is physical (because it makes localization and coupling mechanisms explicit) yet parametrically parsimonious

derive explicit functional relationship between the n-dependent observable and n

e.g. E n = e ( n + 1 / 2) e x e ( n + 1 / 2)2 + e y e ( n + 1 / 2)3 hc establish relationship between a molecular constant (e, exe, ) and the 3 parameters that define V(x) e.g. ex e ax

There are 2 kinds of garden variety perturbation theory: 1. Nondegenerate (Rayleigh-Schrdinger) P.T. simple formulas 2. Quasi-Degenerate P.T. matrix Heff finite Heff is corrected for out-of-block perturbers by van Vleck or contact transformation ~4 Lectures Derive Perturbation Theory Formulas correct E n and n directly for neglected terms in exact H correct all other observables indirectly through corrected

* *

modified 9/30/02 10:13 AM

5.73 Lecture #14

Formal treatment
E n = 0 E(0) + 1E(1) + 2 E(2) n n n
(0) (1) n = 0 n + 1 n

14 - 4
usually stop at 2 usually stop at 1 (because all observables involve ) order-sorting is MURKY

H = 0 H(0) + 1H(1)

is an order-sorting parameter with no physical significance. Set = 1 after all is done. = 0 1 is like turning on the effect of H(1). Equations must be valid for 0 1. Plug 3 equations into Schr. Equation Hn = Enn and collect terms according to order of . 0 terms
(0) (0) H(0) n = E (0) n n
(0) left multiply by m

H(0) = E(0) mn mn n
(0) requires that H(0) be diagonal in n (0) and eigenfunctions eigenvalues E n

{ }

{ } of H
( 0) n

(0)

CALLED BASIS FUNCTIONS

CALLED ZERO ORDER MODEL

modified 9/30/02 10:13 AM

5.73 Lecture #14

So we choose H(0) to be the part of H for which: * it is easy to write a complete set of eigenfunctions and eigenvalues * it is easy to evaluate matrix elements of common perturbation terms in this basis set * sometimes choice of basis set is based on convenience rather than goodness doesnt matter as long as the basis is complete. 1 V(x) = kx 2 examples: Harmonic Oscillator 2 2 V(x) = D 1 e ax Morse Oscillator

14 - 5

Quartic Oscillatr n-fold hindered rotor

V(x) = bx 4
V n ( ) = V 0 2 (1 cos n ) n

Now return to the Schr. Eq. and examine the 1 and 2 terms.
1 terms
(0) (1) (0) (1) H(1) n + H(0) n = E (1) n + E (0) n n n

(0) multiply by n

from H operating to left

(0) (1) (0) (1) H (1) +E (0) n n =E (1) +E (0) n n nn n n n

(could also require

same get rid of them

(0) (1) n n = 0 we do require this later

modified 9/30/02 10:13 AM

5.73 Lecture #14

14 - 6

H(1) = E (1) nn n

1st-order correction to E is just expectation value of perturbation term in H: H(1).

(0) return to 1 equation and this time multiply by m

(0) (1) (0) (1) H(1) + E(0) m n = 0 + E(0) m n mn m n (0) (1) H(1) = m n E(0) E(0) mn n m (0) m (1) n =

H(1) mn E(0) E(0) n m

(0) (0) completeness of (0) : k k k (1) n =

{ }
(0) k

(0) k

(1) n

but we know this

(1) (0) n = k E (0) E (0) k n k

H(1) kn

* index of (1) matches 1st index in denominator n * n = k is problematic. Insist exclude k = n. k * we could have demanded (0) (n1) = 0 n

modified 9/30/02 10:13 AM

5.73 Lecture #14

terms
2

14 - 7

most important in real problems although excluded from many text books.
(1) (1) (0) H(1) n = E (1) n + E (2) n n n (0) multiply by n (0) (1) n n = 0

(0) (1) n H(1) n = 0 + E(2) n

completeness

( ( ( ( n0 ) H(1) k0 ) k0 ) n1) = E (n2 )

H(1) n,k

(0) k E n E (0) k

H(1) k,n

(1) 2 H k,n (2) E n = (0) k E n E (0) k

always first we have derived all needed formulas

matrix element squared over energy difference in energy denominator

(0) (1) E (0) , E (1) , E (2) ; n , n n n n

Examples 1 V(x) = kx 2 + ax3 2

V(x)

(a < 0)

H(0) =

1 2 p2 kx + 2 2m

H(1) = ax3

(actually ax3 term with a < 0 makes all potentials unbound. How can we pretend that this catastrophe does not affect the results from perturbation theory?)
modified 9/30/02 10:13 AM

5.73 Lecture #14

need matrix elements of x3 two ways to do this * matrix multiplication * a,a tricks
h x = m
3 3/2

14 - 8

x3 = x ijx jk x kl il
j,k

h x = ~ m
3 3

3/2

2 1/2 a + a

)]

h = 2 m

3/2

[a + (a aa + aa a + aaa ) + (aa a

+ a aa + a a a + a 3

each group in ( ) has their own v selection rule (see pages 13-8 and 9): simplify using [a,a] = 1 Goal is to manipulate each mixed a,a term so that the number operator appears at the far right and exploit aa n = n n Only nonzero elements:
a3 n3n = [ n( n 1)( n 2 )]
1/2 1/2

a3 = [( n + 3)( n + 2)( n + 1)] n+3n

square root of larger q.n.

(a aa + aa a + aaa ) = 3aa a

because a aa = aa a + [a ,a ]a = aa a a

(aa a )

aaa = aa a + a[a ,a ] = aa a + a
n 1n

= n 3/ 2

modified 9/30/02 10:13 AM

5.73 Lecture #14

14 - 9

(aaa + aaa + aaa) = 3aaa + 3a [3aaa + 3a ]n+1n = 3n(n + 1)1/2 + 3(n + 1)1/2 = 3(n + 1)3/2
So we have worked out all x3 matrix elements leave the rest to P.S. #5. Property other than En?
(0) (1) Use n = n + n

e.g. transition probability (electric dipole allowed vibrational transitions)

P nn x nn for H - O x nn
2 2

h = 1/2 n > n > ,n < +1 2(km) m (only n = 1 transitions)

for a perturbed HO, e.g. H(1) = ax3

n = n =

0 (n )

+ +

H (1) nk
( E k0) 0) (n +3

0 (n )

k E (0 ) n (1 ) H nn +3

(k0)
(1 ) H nn +1 (0) H (1) 1 (0) H (1) 3 (0) nn n +1 + n 1 + nn n 3 + h + h +3h

3h

modified 9/30/02 10:13 AM

5.73 Lecture #14

14 - 10

1st index

n + 3 n + 4, n + 2 n + 1 n + 2, n n X n + 1, n 1 n 1 n, n 2 n 3 n 2, n 4
h nxn +7 = 2( km )1/2 m a4 2 x nn+7 7 11 n 7 m
7/2

Allowed 2nd Indices

For matrix elements of X.

cubic anharmonicity of V(x) can give rise to n = 7, 5, 4, 3, 2, 1, 0 transition

a2 ( n + 7)! 2 n! ( 3h ) 7/2 n
1/2

other less extreme n transitions go as lower powers of

1 and n

modified 9/30/02 10:13 AM

5.73 Lecture #15

Perturbation Theory II
(See CTDL 1095-1104, 1110-1119) Last time:
(0) (0) H(0) n = E (0) n n

15 - 1

{ }{ }

H(0) is diagonal (0) n , E(0) are n basis functions and zero - order energies

E (1) = H(1) n nn

expectation value of perturbation operator

E (2) n

= (0) (0) k En Ek
1st index

H(1) nk

sum excludes k = n matrix element vs. energy denominator

E n = E(0) + E(1) + E(2) n n n H(1) (1) (0) n = (0) nk (0) k k En Ek

sum excludes k = n

mixing coefficient, order sorting parameter, convergence criterion

Today:

1. cubic anharmonic perturbation x3 vs. a,a ax3 x and Y00 contributions 2. nonlecture Morse oscillator pert. theory for ax3 3. transition probabilities orders and convergence of p.t. Mechanical and electronic anharmonicities.

modified 9/30/02 10:12 AM

5.73 Lecture #15

Example 1. H = p2 1 2 + kx 2m 2
H(0)
V(x)

15 - 2
+ax 3
H(1)

(a < 0)
unphysical
x

need matrix elements of x3 one (longer) way

x3 = x ijx jk x kl il
j,k

4 different selection rules: l i = 3, 1, 1, 3 i=3 i i +1, i + 1 i + 2, i + 2 i + 3

one path

[(i + 1)(i + 2)(i + 3)]1/ 2

l

i=1

i i + 1, i + 1 i + 2, i + 2 i + 1 i i 1, i 1 i, i i + 1 i i + 1, i + 1 i, i i + 1
There are three 3-step paths from i to i + 1. Add them.

[(i + 1)(i + 2)(i + 2)]1/ 2 + [(i)(i)(i + 1)]1/ 2 + [(i + 1)(i + 1)(i + 1)]1/ 2
algebraically complicated

other (shorter) alternative: a, a, and aa

h x = m
3

3/ 2

h x = ~ m
3

3/ 2

2 1/ 2 a + a
3

)]

(a + a ) 3 (a + a ) = a3 + [aaa + aaa + aaa ] + [aaa + aaa + aaa] + a

h = 2m
four terms, four different selection rules.

3/ 2

modified 9/30/02 10:12 AM

5.73 Lecture #15

15 - 3

Use simple a,a algebra to work out all matrix elements and selection rules by inspection.
recall: a n = ( n + 1)1/2 n + 1 , a n = n1/2 n 1 , aa n = n n a,a = 1 prescription for

[ ]

aa = 1 + aa
1/2

permuting a thru a

n = 3 a 3 n3,n = [( n 2 )( n 1)( n )] 1/2 n = +3 a3 n+3,n = [( n + 3)( n + 2 )( n + 1)] n = 1

[aaa + aaa + aaa ]n1,n

goal is to rearrange each product so that it has number operator at right

aaa = aaa a aaa = aaa + a aaa = aaa 3aaa + 0

n = 1 n = +1

[ ]n1,n = 3(aaa )

[aaa + aaa + a a a]

n1,n

= n 1 3a aa n = 3n 3/2

( )

aaa = aaa + a = aaa + 2a aaa = aaa + a a a a = a a a 3aaa + 3a

3 n + 1 aaa + a n = 3 n( n + 1)1/ 2 + ( n + 1)1/ 2 = 3( n + 1)3/ 2

all done not necessary to massage the algebra as it would have been for x3 by direct x multiplication!
modified 9/30/02 10:12 AM

5.73 Lecture #15

Now do the perturbation theory:
E n = E (0) + E (1) + E (2) = h ( n + 1 / 2) + 0 + (0) n n n k E n E (0) k 3 x nn = 0 H(1) nk
2

15 - 4

H(1) nk k = n3 k = n 1 k = n+3 k = n+3

2 2 2 2

E(0) E(0) n k +3h +1h 1h 3h

h 3 a ( n 2)( n 1)( n ) 2m h 3 3 a 9n 2m h 3 a 9( n + 1)3 2m h 3 a ( n + 3)( n + 2)( n + 1) 2m

h a2 2m = h
all of the constants
3

E (2) n

( n 2)( n 1)( n) ( n + 3)( n + 2)( n + 1) 9 n 3 9( n + 1)3 + 3 3 1 1

2 nearly cancelling pairs

E(n2 ) =

a 2h 2 2 30(n + 1 / 2 ) 3.5 3 4 8m

algebra

E(n2 ) =

a 2h 2 15 7 2 n + 1 / 2) + 3 4 4 ( 16 m

(m3 4 = mk 2 )

all levels shifted down regardless of sign of a cant measure sign of cubic anharmonicity constant, a, from vibrational structure alone
E n = h ( n + 1 / 2) h 15 a 2 h 7 a 2h ( v + 1 / 2 )2 h 3 4 4 m 3 4 16 m h e x e hY 00 E n = h Y 00 + e ( v + 1 / 2) e x e ( v + 1 / 2)2 + e y e ( v + 1 / 2)3

]
modified 9/30/02 10:12 AM

ax3 makes contributions exclusively to Y00 and exe

5.73 Lecture #15

Nonlecture Morse Oscillator via perturbation theory
V( x ) = D 1 e

15 - 5

E n = h ( n + 1 / 2) ( n + 1 / 2)2 x

x 2

by WKB or DVR

known in advance compare to pert. theory applied to Taylor series expansion of V(x)

Our initial goal is to re-express the Morse potential in terms of and x rather than D and . Then we will expand VMORSE in a Taylor series and look at the coefficient of the x3 term. First we must take derivatives of Ev with respect to v n + 1/2
at dissociation, dE v = 0 = h( 2( n + 1 /2)x) dv = nD + 1 /2 2x

at dissociation asymptote
2 D = E n D = h x 4 x 2 2 x
nD + 1 / 2

( nD + 1 / 2 )2

D=h

2 4 x

now expand V(x)

V( 0 ) = 0 h 2 +2 e x 2 e 2x , V (0) = 0 V (x) = 4 x 2 2 h 2 h 2 2 x 2 2 x 2 h 2 e + 4 e , V (0) = 2 = V (x) = 4 x 4 x 2 x 3h 2 3 h 2 +2 3e x 8 3e 2x , V (0) = V (x) = 4 x 2 x

[ [ [

] ]

but
1/2 h 2 2 2m x = V (0) k = m = h 2 x 3/2 2 3 h 2m x V (0) = 2 x h 1 2 V(x) = kx + ax3 thus V (x) = 6a 2 1 h 2 2m x 3/2 1 4 m 3 x a= a2 = 4 x h 2 h 2

now we can eliminate from higher derivatives (at x = 0). This is to be compared to V(0) for the cubic anharmonic potential.

modified 9/30/02 10:12 AM

5.73 Lecture #15

a 2h m 3 4 15 a 2 h from pert. theory (#15 - 4) x = 4 m 3 4 x = 2

15 - 6

same functional form but different numerical factor (2 vs. 3.75)

One reason that the result from second-order perturbation theory applied directly to V(x) = kx2/2 + ax3 and the term-by-term comparison of the power series expansion of the Morse oscillator are not identical is that contributions are neglected from higher derivatives of the Morse potential to the (n + 1/2)2 term in the energy level expression. In particular
h 2 4 4 E (1) = V (0) x 4 4! = 7 / 2 x 24 n x h 2 n x4 n = 4( n + 1 / 2)2 + 2 2m

contributes in first order of perturbation theory to the (n + 1/2)2 term in En.

E (1) = n 7 7 x ( n + 1 / 2 )2 + x 12 24

Example 2

Use perturbation theory to compute some property other than Energy

(0) (1) need n = n + n to calculate matrix elements of the operator in question,

for example, transition probability, x: for electric dipole transitions, transition probability is P n n x nn 2
For H - O n n 1 only h 2 x nn+1 = ( n + 1) 2m

for perturbed H-O

H(1) = ax3

Standard result. Now allow for mechanical and electronic anharmonicity.

H(1) (0) (0) n = n + (0) kn (0) k k En Ek (1) H(1) H(1) H(1) (0) (0) (0) H (0) (0) n = n + nn+3 n+3 + nn+1 n+1 + + nn1 n1 + nn3 n3 3h h h 3h

modified 9/30/02 10:12 AM

5.73 Lecture #15

initial state
(0) n
(0) (1) n + n

15 - 7
effect of x anharmonic final state

n+4 n+3 n+2 n+1 n+1 n n1 n2

n+7 , n+5, n+4, n+3, n+1 n+5, n+3, n+2, n+1, n3 n+4, n+2, n+1, n, n2 n+3, n+1, n, n1, n3 n+2, n, n1, n2, n4 n+1, n1, n2, n3, n5 n-1, n-3, n-4, n-5, n-7
2

n n1

n3

n4

Many paths which interfere constructively and destructively in x nn

n = n + 7, n + 5, n + 4, n + 3, n + 2, n + 1, n, n 1, n 2, n 3, n 4, n 5, n 7
only paths for H-O!

The transition strengths may be divided into 3 classes 1. direct: n n 1 2. one anharmonic step n n + 4, n + 2, n, n 2, n 4 3. 2 anharmonic steps n n + 7, n + 5, n + 3, n + 1, n 1, n 3, n 5, n 7 Work thru the n = 7 path
h nxn +7 = 2m
3/2+3/2+1/2

x3 n,n+3 x3 n+4,n+7
h3a 4 n 7 x nn+7 4 7 7 11 3 2 m
2

a 3) n + 4) ( n + 5)( n + 6)( n + 7) + 1)( + 2 ( n 44n 22 )( n +3 ( 1 4 444 123 144 2444 4 3 ( 3h ) x n,n +3 x n +3,n +4 x n +4,n +7
2

1/2

x n+3,n+4

modified 9/30/02 10:12 AM

5.73 Lecture #15

* you show that the single-step anharmonic terms go as

15 - 8

3/2+1/2 a h x nn+4 [(n + 1)(n + 2)(n + 3)(n + 4)]1/2 2m ( 3h )

x nn+4
*

h2a 2 n 4 2 4 4 6 3 2 m

Direct term

h1 x nn+1 ( n + 1) 32m1 1
2
hn 3a 2 each higher order term gets smaller by a factor 32 23 m 3 5 which is a very small dimensionless factor. RAPID CONVERGENCE OF PERTURBATION THEORY!

What about Quartic perturbing term bx4?

Note that E (1) = n bx 4 n 0 and is directly sensitive to sign of b!

modified 9/30/02 10:12 AM

5.73 Lecture #16

Last time 1 2 3 1. V(x) = 2 kx + ax

16 - 1

Perturbation Theory III

cubic anharmonic oscillator algebra with x 3 vs. operator with a,a ax 3 x & Y00

cant know sign of a from vibrational information alone. [Can know it if rotationvibration interaction is included.] 2.
Morse Oscillator V(x) = D 1 e x

* * *

D, , x, m 3h 2 3 d 3V morse d 3V = 6a = = 2 x dx3 x=0 dx3 a 2h 15 a 2 h x = 2 3 4 direct from Morse vs. 4 m 3 4 m 1 2 from pert. theory on kx + ax 3 2
a 2h x = 2 3 4 m 15 a 2 h from pert. theory (#15 - 4) x = 4 m 3 4

same functional form

Today:

1. 2. 3.

4.

Effect of cubic anharmonicity on transition probability orders of pert. theory, convergence [last class: #15-6,7,8]. Use of harmonic oscillator basis sets in wavepacket calculations. What happens when H(0) has degenerate E(0) s? n Diagonalize block which contains (near) degeneracies. Perturbations accidental and systematic. 2 coupled non-identical harmonic oscillators: polyads.

modified 10/1/02 10:06 AM

5.73 Lecture #16

16 - 2

One reason that the result from second-order perturbation theory applied directly to V(x) = kx2/2 + ax3 and the term-by-term comparison of the power series expansion of the Morse oscillator are not identical is that contributions are neglected from higher derivatives of the Morse potential to the (n + 1/2)2 term in the energy level expression. In particular
h 2 4 4 4 (1) E n = V (0) x 4! = 7 / 2 x 24 x 2 h n x4 n = 4( n + 1 / 2 ) 2 + 2 2m

contributes in first order of perturbation theory to the (n + 1/2)2 term in En.

E (1) = n 7 7 x ( n + 1 / 2 )2 + x 12 24

Example 2

Compute some property other than Energy (repeat of pages 15-6, 7, 8)

(0) (1) need n = n + n

transition probability: for electric dipole transitions P n n x nn 2

For H - O n n 1 only h 2 x nn+1 = ( n + 1) 2m

for perturbed H-O

H(1) = ax3

H(1) (0) (0) n = n + (0) kn (0) k k En Ek (1) H(1) H(1) H(1) (0) (0) (0) H (0) (0) n = n + nn+3 n+3 + nn+1 n+1 + + nn1 n1 + nn3 n3 3h h h 3h

modified 10/1/02 10:06 AM

5.73 Lecture #16

initial state
(0) n
(0) (1) n + n

16 - 3
effect of x anharmonic final state

n+4 n+3 n+2 n+1 n+1 n n1 n2

n+7 , n+5, n+4, n+3, n+1 n+5, n+3, n+2, n+1, n3 n+4, n+2, n+1, n, n2 n+3, n+1, n, n1, n3 n+2, n, n1, n2, n4 n+1, n1, n2, n3, n5 n-1, n-3, n-4, n-5, n-7
2

n n1

n3

n4

Many paths which interfere constructively and destructively in x nn

n = n + 7, n + 5, n + 4, n + 3, n + 2, n + 1, n, n 1, n 2, n 3, n 4, n 5, n 7
only paths for H-O!

The transition strengths may be divided into 3 classes 1. direct: n n 1 2. one anharmonic step n n + 4, n + 2, n, n 2, n 4 3. 2 anharmonic steps n n + 7, n + 5, n + 3, n + 1, n 1, n 3, n 5, n 7 Work thru the n = 7 path
h nxn +7 = 2m
3/2+3/2+1/2

x3 n,n+3 x3 n+4,n+7
h3a 4 n 7 x nn+7 4 7 7 11 3 2 m
2

a 3) n + 4) ( n + 5)( n + 6)( n + 7) + 1)( + 2 ( n 44n 22 )( n +3 ( 1 4 444 123 144 2444 4 3 ( 3h ) x n,n +3 x n +3,n +4 x n +4,n +7
2

1/2

x n+3,n+4

modified 10/1/02 10:06 AM

5.73 Lecture #16

* you show that the single-step anharmonic terms go as

16 - 4

3/2+1/2 a h x nn+4 [(n + 1)(n + 2)(n + 3)(n + 4)]1/2 2m ( 3h )

x nn+4
*

h2a 2 n 4 2 4 4 6 3 2 m

Direct term

h1 x nn+1 ( n + 1) 32m1 1
2
hn 3a 2 each higher order term gets smaller by a factor 32 23 m 3 5 which is a very small dimensionless factor. RAPID CONVERGENCE OF PERTURBATION THEORY!

What about Quartic perturbing term bx4?

Note that E (1) = n bx 4 n 0 and is directly sensitive to sign of b!

modified 10/1/02 10:06 AM

5.73 Lecture #16

2. What about wave packet calculations?
( n expressed as superposition of k0 ) terms

16 - 5

( ( x,0 ) expanded as superposition of k0 ) terms (usually only

one term, called the bright state). But we must also expand
( k0 ) as a superposition of eigenbasis, k , terms.

( x ,t ) oscillates at e iE n t h
E n = E (0) + E (1) + E (2) n n n

(0) A state which is initially in a pure n will dephase, then exhibit partial recurrences at

m2

t=

m2

but

* not perfect since E n E m h (n m) not quite integer multiples!

* time of 1st recurrence will depend on E ! because E n+1 E n1 decreases as n increases. 2

modified 10/1/02 10:06 AM

5.73 Lecture #16

Degenerate and Near Degenerate E (0) n

16 - 6

* * *

Ordinary nondegenerate p.t. treats H as if it can be diagonalized by simple algebra. (0) CTDL, pages 1104-1107 find linear combination of degenerate n for which H(1) lifts degeneracy. This problem is usually treated in an abstract way by people who never actually use perturbation theory!

Whenever

1 block of H = H(0) + H(1) E (0) E (0) n k

H(1) nk

must diagonalize the n,k 2 2

accidental degeneracy spectroscopic perturbations systematic degeneracy 2-D isotropic H-O, polyads quasi-degeneracy safe chunk of H effects of remote states Van Vleck Pert. Theroy - next time

Continuum Philosophy:

En particular class of experiments does not look at all Ens - only a given E range and only a given E resolution!

Want a model that replaces dimension H by simpler finite one that does really well for the class of states sampled by particular experiment.

NMR IR UV

nuclear spins (hyperfine) vibr. and rotation electronic

don't care about excited vib. or electronic don't care about Zeeman don't care about Zeeman

modified 10/1/02 10:06 AM

5.73 Lecture #16

quasidegenerate block sampled by our specific experiment

16 - 7

H=

NN

quasidegenerate blocks sampled by other experiments

each finite block along the diagonal is an Heffective fit model. We want these fit models to be as accurate and physically realistic as possible. * * fold important out-of-block effects into N N block 2 stripes of H diagonalize augmented N N block - refine parameters that define the block against observed energy levels.
next time review V-V transformation

4. Best to illustrate with an example 2 coupled harmonic oscillators: Fermi Resonance [approx. integer ratios between characteristic frequencies of subsystems]
2 p2 1 1 2 p H = 1 + k1 x1 + 2 + k 2 x 2 + k122 x1 x 2 2 2 2m 2 2m 2
) (n0n2 = (n11) ( x1 ) (n0 ) ( x2 ) 1 2

why not k12x1x2?

(0) H1

H(0) 2

E(0) = h1( n1 + 1 / 2) n
1

E(0) = h 2 ( n 2 + 1 / 2) n
2

E (0) = h[1( n + 1 / 2) + 2 ( n + 1 / 2)] nm

let 1 = 2 2

(m1 = m 2 , k1 = 4k 2 )
modified 10/1/02 10:06 AM

systematic degeneracies

5.73 Lecture #16

H
(1)

16 - 8
3/ 2

k 122 x 1 x 2 2

h = k 122 2m

2 1 2 1

1/ 2

[(a

+ a a 2 + a 2 + a 2a + a a 2 1 2 2 2 2

)(

)]

aa + a a = 2a a + 1

H(1) nm;kl
nk m l 2 +2 0 2 +2 0 H (1)

H(1) = (constants)

a1a 2 2 a1a2 2

1 1

[(n + 1)(m + 2)(m + 1)]1/2

[(n + 1)(m)(m 1)]1/2

6 types of terms

a1 2a a 2 + 1 2
a1 a 2 2 a1 a2 2

)
)

1 +1 +1 +1

[(n + 1)(2m + 1) ]

2 1/ 2

[(n )(m + 2)(m + 1)]1/2

[(n )(m)(m 1)]1/2

a1 2a a 2 + 1 2

(n)(2m + 1)2

1/ 2

Seems complicated but all we need to do is look for systematic near degeneracies Recall 1 = 2 2
List of Pol y ads by Membership (n1 , n2) (0,0) (0,1) (1,0), (0,2) (1,2), (0,3) (2,0), (1,2), (0,4) E(0)/h 2 [2(n1 + 1/2) + (n2 + 1/2)] degeneracy 1 1 2 2 3 3 4 4 etc. 1+ 1/2 = 1 + 3/2 = 3 + 1/2 = 3 + 3/2 = 3/2 5/2 7/2 9/2; 11/2 13/3 15/2 17/2 19/2 1 + 7/2 = 9/2 0 1 2 3 4 5 6 7 8 P = 2n1 + n2

modified 10/1/02 10:06 AM

5.73 Lecture #16

General P block:
3 + (2 n1 + n2 ) = P + 3 /2 2 # of terms in P block depends on whether P is even or odd
( EP0) h 2 =

16 - 9

P+2 states 2 P +1 states 2

P P even P n1 = , n2 = 0 , n1 = 1, 2 , (0 , P) 2 2 odd P P 1 , n2 = 1 , (0 , P) n1 = 2
not 0 because P = 2n1 + n2 is odd

H(1) = a1a2 + a1 a 2 + a1a 2 + a1 a2 + a1 2a a 2 + 1 + a1 2a a 2 + 1 3/2 3/2 1/2 1 2 2 2 2 2 2 3/2 h m 1 2 k122 2 P= 0 0 4 +4 2 +2

) (

inside polyad

between polyad blocks

POLYAD

H(0) P = h 2

0 0 0 P + 3 / 2 0 P+3/2 0 0 0 0 O 0 0 0 0 P + 3 / 2

n, m

P ,0 2
H (1) P = stuff

P 1, 2 2

M
1, P 2 0, P

0 P ( 2 1) 2 sym 0 0 sym 0 0 0 0

P ,0 2

P 1, 2 2

1/ 2

P 2,4 2

0 P 1/ 2 1( 3 4 ) 2 0 sym 0

0 0 (even P) 0 0 L 0 1/ 2 0 [(1)(P )(P 1)] sym 0

L0, P

Note that all matrix elements may be written in terms of a general formula computer decides membership in polyad and sets up matrix

modified 10/1/02 10:06 AM

5.73 Lecture #16

16 - 10

So now we have listed ALL of the connections of P = 6 to all other blocks! So we use these results to add some correction terms to the P = 6 block according to the formula suggested by Van Vleck.

H(2) P nm

= (0) (0) P En + Em E(0) k 2

H(1) H(1) nk km

for our case*, the denominator is h 2 [ P P ]

For this particular example there are no cases where there are nonzero elements for n m (many other problems exist where there are nonzero n m terms)

3 4 8 =5 2 30 2 2 4 (2) 22 h 2 H 6 3 3 1 2 2 3 = 14 h m 1 2 k122 2 06 dimensionless

50 75 4 36 + = 8 3 4 4 2

105 81 162 12 60 + = 2 4 4 2 2

169 56 197 = 2 4 2

Computers can easily set these things up. Could add additional perturbation terms such as diagonal anharmonicities that cause 1 : 2 2 : 1 resonance to detune.

modified 10/1/02 10:06 AM

5.73 Lecture #16

For concreteness, look at P = 6 polyad (3,0), (2,2), (1,4), (0,6)

16 - 11

H(1) 6 stuff

30 22 30 0 (3 2 1)1/2 22 sym 0 14 0 sym 06 0 0

14 0 (2 4 3)1/2 0 sym

06 0 0 (1 5 6)1/2 0

now what are all of the out of block elements of x 1 x 2 that affect the P = 6 block? 2
H (1)/stuff P = 2 P=6~P=4

E(0) E(0) P P2
+2h 2 +2h 2 +2h 2 2h 2 2h 2 2h 2 2h 2 +4h 2 +4h 2 4h 2 4h 2 4h 2 4h 2

a1 2a a 2 + 1 2

)
)

3,0 ~ 2,0 2,2 ~ 1,2 1,4 ~ 0,4 0,6 ~ 3,0 ~ 4,0 2,2 ~ 3,2 1,4 ~ 2,4 0,6 ~ 1,6 3,0 ~ 2,2 ~ 1,0 1,4 ~ 0,2 0,6 ~

31/2 2 1/2 5 11/2 9 41/2 31/2 5 21/2 9 21/2 (2 1)1/2 11/2 (4 3)1/2 [421]1/2 [343]1/2 [265]1/2 [187]1/2

P = +2

a1 2a a 2 + 1 2

P = 4

a1a 2 2

11/2 13

P = +4

a1 a2 2

3,0 ~ 4,2 2,2 ~ 3,4 1,4 ~ 2,6 0,6 ~ 1,8

H eff = H(0) + H(1) + H(2) P=6 6 6 6

h 2 ( 6 + 3 / 2 )

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
modified 10/1/02 10:06 AM

5.73 Lecture #17 Perturbation Theory IV

17 - 1

Last time: Transition probabilities in the presence of cubic anharmonicity

direct
(0) (0) n n1

n = 1

singly forbidden

(0) (1) (0) n n + n

n = 4, 2, 0

doubly forbidden

(1) (0) (1) (0) n + n n + n

n = 7, 5, 3, 1

Extra basis states mixed in by ax3 anharmonicity denoted by underline. Today **

(0) brief remarks about ( x, 0) = n in the presence of anharmonic mixing: * partial recurrences depend on d * rate of dephasing of recurrences depends on dn , the fractional change in the average frequency.

** **

x-k relationships Quasi degeneracy

when H(1) nk E (0) E (0) n k 1 must diagonalize

**

coupled oscillator example: POLYADS, IVR

Possibility: Intramolecular Vibrational Redistribution in Acetylene

modified 9/30/02 10:12 AM

5.73 Lecture #17

What about Quartic perturbing term bx4? special hint
Note that E(1) = n bx 4 n 0 and is n directly sensitive to sign of b!

17 - 2

What about wave packet calculations?

(0) n expressed as superposition of k basis state terms (0) ( x,0) expanded as superposition of k terms

(perturbed eigenstates)

(state prepared at t = 0) (evolving prepared state)

( x, t ) oscillates at e iE n t h
E n = E (0) + E (1) + E (2) n n n
(0) A state which is initially in a pure n will dephase, then exhibit partial recurrences at
m2 n t
not quite integer multiples

m2 where m is an integer but recurrence is not perfect since E n Em h ( n m) t recurrence =

E n + E n+1 2 decreases as n increases because

* *

time of 1st recurrence will depend on E! successive recurrences will occur with larger phase error for n,n1 vs. n+1,n

1st recurrence phase discrepancy is 2nd recurrence phase discrepancy is 2 etc.

modified 9/30/02 10:12 AM

5.73 Lecture #17

17 - 3

On pages 16-5 through 16-11 I worked out how a block of Heff is corrected so that out-of-block off-diagonal matrix elements can be safely ignored. These corrections come in two forms:

(i)

Second-order perturbation theory corrections to diagonal matrix elements. One example is the xk relationships by which the xij vibrational anharmonicity constants are evaluated in terms of third and fourth derivatives of V(Q). Van Vleck transformation of quasi-degenerate or resonant blocks of Heff. Something analogous to second-order perturbation theory is used to fold out-of-block off-diagonal matrix elements into polyad blocks along the diagonal of H. These corrections occur both on and off the diagonal within these quasi-degenerate polyad blocks. Resonance is not accidental. Once it appears it affects larger and larger groups of near-degenerate basis states.

(ii)

Consider the following 3 3 example of a Van Vleck transformed Heff:

2 0 5 H = 5 2 0.5 2 0.5 20

This H has a 2 2 quasidegenerate block and both members of this block interact weakly with a nonquasidegenerate remote state.

modified 9/30/02 10:12 AM

5.73 Lecture #17

17 - 4

0 0 0 H( 0 ) = 0 2 0 0 0 20 2 0 5 H(1) = 5 0 0.5 2 0.5 0 22 20 1 H( 2 ) = 19 0 (2)(0.5) 0+2 20 2 0.5 2 18 0 0 2 2 0.5 2 + 20 18 0

H H H = ( 0) En + E in-block k E (k0 ) out-of-block 2

( 2) nn

(1) nk

(1) kn ( 0) n

modified 9/30/02 10:12 AM

5.73 Lecture #17

17 - 5

See Ian M. Mills Vibration-Rotation Structure in Asymmetric- and SymmetricTop Molecules, pages 115-140 in Molecular Spectroscopy: Modern Research, Volume I, K. Narahari Rao, and C. Weldon Mathews, Academic, 1972. Second-order perturbation theory is used to derive the famous x-k relationships, namely the relationship between the second, third, and fourth derivatives of V(Q) and the normal mode i and xij molecular constants.

G(V ) =

(v
i i

+ 1 /2) +

i,j i j rst

xij ( vi + 1 /2)( v j + 1 /2) 3V qr q s qt qrq sqt

1 V (Q) = 2 1 + 24

2V

1 2 qr + 2 6 qr

rstu

4V qr q s qt qu qrq sqt qu

[unrestricted sums: get several identical partial derivatives.] Lengthy derivations:

1 V 1 x ii = 4 16 q i 16 s
4
direct firstorder contribution from quartic force constant

3V 2 q i q s

(8 2 3 2 ) i s 2 2 s (4 i s )

second-order summation over cubic force constants

modified 9/30/02 10:12 AM

5.73 Lecture #17

17 - 6

1 4V 1 xij = 4 qi2 q 2 4 j
first-order from quartic force constant

3V 3V q 2 q 2 i t q j qt
second-order sum over vi= vj = 0 terms for cubic force constants

1 2

2 3 V 2 t t i2 2 j qi q j qt

ijt

second-order sum over all vi = 1, vj = 1 terms for cubic force constants

ijt = i + j + t i j t i + j t i j + t
ijt is Resonance denominator. When

)(

)(

)(

i = j + t or or j = i + t t = i + j

perturbation theory blows up. Must go to Heff polyads and diagonalize.

modified 9/30/02 10:12 AM

5.73 Lecture #17

Degenerate and Near Degenerate E (0) n

17 - 7

* * *

Ordinary nondegenerate p.t. treats H as if it can be diagonalized by simple algebra. (0) CTDL, pages 1104-1107 find linear combination of degenerate n for which H(1) lifts degeneracy. This problem is usually treated in an abstract way by people who never actually use perturbation theory!

Whenever

1 block of H = H(0) + H(1) E (0) E (0) n k

H(1) nk

must diagonalize the n,k 2 2

accidental degeneracy spectroscopic perturbations systematic degeneracy 2-D isotropic H-O, polyads quasi-degeneracy safe chunk of H effects of remote states Van Vleck Pert. Theroy - next time

Continuum Philosophy:

En particular class of experiments does not look at all Ens - only a given E range and only a given E resolution!

Want a model that replaces dimension H by simpler finite one that does really well for the class of states sampled by particular experiment.

NMR IR UV

nuclear spins (hyperfine) vibr. and rotation electronic

don't care about excited vib. or electronic don't care about Zeeman don't care about Zeeman

modified 9/30/02 10:12 AM

5.73 Lecture #17

quasidegenerate block sampled by our specific experiment

17 - 8

H=

NN

quasidegenerate blocks sampled by other experiments

each finite block along the diagonal is an Heffective fit model. We want these fit models to be as accurate and physically realistic as possible. fold important out-of-block effects into N N block 2 stripes of H diagonalize augmented N N block - refine parameters that define the block against observed energy levels. (0) (0) next time review V-V = n ( x1 ) n ( x 2 ) 1 2 * *
transformation

(0) n n

1 2

4. Best to illustrate with an example 2 coupled harmonic oscillators: Fermi Resonance [approx. integer ratios between characteristic frequencies of subsystems]
2 p2 1 1 2 p H = 1 + k1 x1 + 2 + k 2 x 2 + k122 x1 x 2 2 2 2m 2 2m 2
2

why not k12x1x2?

n n (0) 1 2 n1 n 2

(0) (0) (0) = n ( x1 ) n ( x 2 )

= (n11) ( x1 ) (n0 ) ( x2 ) 2

(0) H1

H(0) 2

E(0) = h1( n1 + 1 / 2) n
1

E(0) = h 2 ( n 2 + 1 / 2) n
2

E (0) = h[1( n + 1 / 2) + 2 ( n + 1 / 2)] nm

let 1 = 2 2

(m1 = m 2 , k1 = 4k 2 )
modified 9/30/02 10:12 AM

systematic degeneracies

5.73 Lecture #17

h 3/2 1 H = = k122 2m 2 1 2 aa + a a = 2a a + 1
(1)

17 - 9
1/2

k122 x1 x 2 2

[(a + a )(a
1 1

2 2

+ a2 + aa + aa 2

)]

H(1) nm;kl
nk ml 2 +2 0 2 +2 0 H (1)

H(1) = (constants)

a1a 2 2 a1a2 2

1 1

[(n + 1)(m + 2)(m + 1)]1/2

[(n + 1)(m)(m 1)]1/2

6 types of terms

a1 2a a 2 + 1 2
a1 a 2 2 a1 a2 2

)
)

1 +1 +1 +1

[(n + 1)(2m + 1)]1/2 [(n )(m + 2)(m + 1)]1/2

[(n )(m)(m 1)]1/2

a1 2a a 2 + 1 2

[(n )(2m + 1)]1/2

Seems complicated but all we need to do is look for systematic near degeneracies Recall 1 = 2 2
List of Polyads by Membership (n1 , n2) (0,0) (0,1) (1,0), (0,2) (1,2), (0,3) (2,0), (1,2), (0,4) degeneracy 1 1 2 2 3 3 4 4 etc. 1+ 1/2 = 1 + 3/2 = 3 + 1/2 = 3 + 3/2 = 3/2 5/2 7/2 9/2; 11/2 13/3 15/2 17/2 19/2 1 + 7/2 = 9/2 0 1 2 3 4 5 6 7 8 E(0)/h 2 P = 2n1 + n2

modified 9/30/02 10:12 AM

5.73 Lecture #17

General P block:
3 E (0) h 2 = + (2n1 + n 2 ) = P + 3 / 2 p 2 # of terms in P block depends on whether P is even or odd

17 - 10

P+2 states 2 P +1 states 2

even P odd P

n = P , n = 0 , n = P 2 1,2 ,(0,2P ) 1 2 2 1 2 P 1 n = , n 2 = 1 ,(0,2P 1) 1 2

H(1) = a1a2 + a1 a 2 + a1a 2 + a1 a2 + a1 2a a 2 + 1 + a1 2a a 2 + 1 3/2 3/2 1/2 1 2 2 2 2 2 2 3/2 h m 1 2 k122 2 P= 0 0 4 +4 2 +2

) (

inside polyad

between polyad blocks

POLYAD

H(0) P = h 2

0 0 0 P + 3 / 2 0 P+3/2 0 0 0 0 O 0 0 0 0 P + 3 / 2

n m

P ,0 2

P 1, 2 2 0

P 2, 4 2 0 0 0 L 1/ 2 0 [(1)(2 P)(2 P 1)] 0 0

1/ 2 P P ,0 0 2 (2 1) 2 P H(1) 1, 2 P = 2 0 sym stuff M sym M 0 0 0, P 0

P 2 1 (3 4) 0 sym

1/ 2

Note that all matrix elements may be written in terms of a general formula computer decides membership in polyad and sets up matrix

modified 9/30/02 10:12 AM

5.73 Lecture #17

17 - 11

So now we have listed ALL of the connections of P = 6 to all other blocks! So we use these results to add some correction terms to the P = 6 block according to the formula suggested by Van Vleck.

H(2) P nm

= (0) (0) P En + Em E(0) k 2

H(1) H(1) nk km

for our case*, the denominator is h 2 [ P P ]

For this particular example there are no cases where there are nonzero elements for n m (many other problems exist where there are nonzero n m terms)

3 4 8 =5 2 30 2 2 4 (2) 22 h 2 H 6 3 3 1 2 2 3 = 14 h m 1 2 k122 2 06 dimensionless

50 75 4 36 + = 8 3 4 4 2

105 81 162 12 60 + = 2 4 4 2 2

169 56 197 = 2 4 2

Computers can easily set these things up. Could add additional perturbation terms such as diagonal anharmonicities that cause 1 : 2 2 : 1 resonance to detune.

modified 9/30/02 10:12 AM

5.73 Lecture #17

For concreteness, look at P = 6 polyad (3,0), (2,2), (1,4), (0,6)

17 - 12

H(1) 6 stuff

30 22 30 0 (3 2 1)1/2 22 sym 0 14 0 sym 06 0 0

14 0 (2 4 3)1/2 0 sym

06 0 0 (1 5 6)1/2 0

now what are all of the out of block elements that affect the P = 6 block?
H (1)/stuff P = 2 P=6~P=4

E(0) E(0) P P2
+2h 2 +2h 2 +2h 2 2h 2 2h 2 2h 2 2h 2 +4h 2 +4h 2 4h 2 4h 2 4h 2 4h 2

a1 2a a 2 + 1 2

)
)

3,0 ~ 2,0 2,2 ~ 1,2 1,4 ~ 0,4 0,6 ~ 3,0 ~ 4,0 2,2 ~ 3,2 1,4 ~ 2,4 0,6 ~ 1,6 3,0 ~ 2,2 ~ 1,0 1,4 ~ 0,2 0,6 ~

31/2 2 1/2 5 11/2 9 41/2 31/2 5 21/2 9 21/2 (2 1)1/2 11/2 (4 3)1/2 [421]1/2 [343]1/2 [265]1/2 [187]1/2

P = +2

a1 2a a 2 + 1 2

P = 4

a1a 2 2

11/2 13

P = +4

a1 a2 2

3,0 ~ 4,2 2,2 ~ 3,4 1,4 ~ 2,6 0,6 ~ 1,8

H eff = H(0) + H(1) + H(2) P=6 6 6 6

h 2 ( 6 + 3 / 2 )

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
modified 9/30/02 10:12 AM

5.73 Lecture #18

Variational Method
(See CTDL 1148-1155, [Variational Method] 252-263, 295-307[Density Matrices]) Last time:

18 - 1

Quasi-Degeneracy Diagonalize a part of infinite H * sub-matrix : H(0) + H(1) * corrections for effects of out-of-block elements: H(2) (the Van Vleck transformation) *diagonalize Heff =H(0) + H(1) + H(2) coupled H-Os 2 : 1 (122) Fermi resonance example: polyads 1. 2. 3. 4. 5. 6. Perturbation Theory vs. Variational Method Variational Theorem Stupid nonlinear variation Linear Variation new kind of secular Equation Linear combined with nonlinear variation Strategies for criteria of goodness various kinds of variational calculations

1. Perturbation Theory vs. Variational Method Perturbation Theory in effect uses basis set goals: parametrically parsimonious fit model, Heff fit parameters (molecular constants) parameters that define V(x) H(1) errors less than this mixing order - sorting (0) nk (0) < 1 En Ek angle times the previous order nonzero correction term (n is in-block, k is out-of block) because diagonalization is order (within block).

Variational Method best possible estimate for lowest few En, n (and properties derivable from these) using finite basis set and exact form of H.

modified 10/9/02 10:21 AM

5.73 Lecture #18

18 - 2

Vast majority of computer time in Chemistry is spent in variational calculations Goal is numbers. Insight is secondary. Ab Initio vs. semi-empirical or fitting [intentionally bad basis set: Hckel, tight binding qualitative behavior obtained by a fit to a few microscopiclike control parameters] 2. Variational Theorem
not necessarily normalized any observable

If is approximation to eigenfunction of A belonging to lowest eigenvalue a 0 , then

A a0

the variational Theorem

PROOF: eigenbasis (which we do not know but know it must exist)

A n = an n

expand in eigenbasis of A, exploiting completeness

= n n
n
completeness

A = n n A n n = n a n
n, n
an nn

eigenbasis

= n n = n
n

A = n 2 n
n

an n

all terms in both sums are 0

subtract a 0 from both sides

a0 = n

(a n a 0 ) n n
n 2

0 both sums are 0

again, all terms in

modified 10/9/02 10:21 AM

5.73 Lecture #18

a0. QED

18 - 3

because, by definition of a0, an a0 for all n and all terms in sum are 0.
but useless because we can' t know a n or n It is possible to perform a variational calculation for any A, not limited to H.

3. Stupid Nonlinear Variation Use the wrong functional form or the wrong variational criterion to get poor results illustrates that the variational function must have sufficient flexibility and the variational criterion must be as it is specified in the variational theorem, as opposed to a clever shortcut. The H atom Schr. Eq. (l = 0)

H=

1 1 2 r 2 r 2 r r
T

1 r
V

1s (r ) = r 1s = 1/ 2 e r and we know E1s = 1 / 2 au but try r = 3 2

[1 au = 219475 cm ]
1

1/ 2

(r )e r

normalized for all

is a scale factor that controls overall size of (r) [actually this is the form of 2p(r)] which is necessarily orthogonal to 1s! STUPID!

((0) = 0

but

1s (0) = 1/ 2

)
skipped a lot of algebra

H 4 2 3 = = 3 8
minimize :

d = 0 min = 3 / 2 min = 3 / 8 au d
1 c .f . the true values: E1s = 1 /2 au , E 2s = au 8
modified 10/9/02 10:21 AM

FAILURE!

5.73 Lecture #18

Try somethng clever (but lazy): What is the value of that maximizes 1s?
for the best variational = 3 / 2, C1s ( = 3 / 2) 1s = 0.9775 if we maximize C1s wrt. : = 5 / 3 C1s = 0.9826

18 - 4

better?

but = 0.370 results, a poorer bound than = 3/2 = 0.375

* need flexibility in * can improve on t d by employing an alternative variational strategy d

This was stupid anyway because we would never use the variational method when we already know the answer! 4. Linear Variation Secular Equation

= cn n
n=1

KEY TOPIC for this lecture

n H n = H nn n n = Snn

overlap integrals (non-orthogonal basis sets are often convenient)

m, m

n, n

cn cn Hnn
cm cm Smm
rearrange this equation

c m c m Smm = c n c n H nn
m, m n, n

to find minimum value of , take c j for each j and require that

= 0 for each j c j

linear variation!

because we are seeking to minimize with respect to each cj. Find the global minimum of the (c1,c2,cN) hypersurface.

the only terms that survive

are those that include c j . c j

modified 10/9/02 10:21 AM

5.73 Lecture #18

c m Smj + S jm = c n H jn + H nj
m n

18 - 5

if { n } are real 0 = c n H jn S jn
n=1 N

Sij = S ji , H ij = H ji

These are all of the surviving terms (i.e. those that include j). Each j term appears twice in both sums, once as a bra and once as a ket.

one such equation for each j (same set of unknown {cn}) N linear homogeneous equations in N unknown cns Non trivial {cn} only if |H S| = 0 (Not same form as |H 1E| = 0) The result is N special values of that satisfy this equation. CTDL show: all N n values are upper bounds to the lowest N Ens (provided that and all {n}s are othogonal! they belong to different How to solve |H S| = 0 values of En) 1. Diagonalize S
U SU = S Sij = si ij

(orthogonalize {} basis)
2. Normalize S

( S)

1/2

1/2 = SS

()

1 S = TST

3 diagonal matrices

where T = US

1/2

(S )
1/2

s1/2 0 0 1 = S1/2 = 0 s2 1/2 0 0 0 O

unitary This is not an orthogonal transformation, but it does not destroy orthogonality because each function is only being multiplied by a constant.

modified 10/9/02 10:21 AM

5.73 Lecture #18

3. Transform H to orthonormalized basis set

18 - 6

H = S 1/2 U HU S 1/2
T new secular equation T

U diagonalizes S not H

H S = 0 H

but

S=

1
diagonalized by usual procedure!

1=0

usual H

5. Combine Linear and Nonlinear Variation typically done in ab initio electronic structure calculations Basis set:

n ( n r )
n

linear variation where n is a radial scale factor

nonlinear variation

= cn n ( n r )

Snn ( n , n ), Hnn ( n , n )
0. 1. 2. pick arbitrary set of { i }

calculate all H ij i , j & Sij i , j Solve H - S = 0

a. b. c. d. SS (S)
1/ 2

diagonalize S

(orthogonalize)

(normalize)

HH diagonalize H

nonlinear variation begins find global minimum of lowest with respect to each i

modified 10/9/02 10:21 AM

5.73 Lecture #18

3. 4. 5. change 1 from (0) (1) = (0) + 1 1 1 recalculate all integrals in H and S involving 1 Solve H - S = 0 to obtain a new set of {i}. Pick lowest i. 6.
7.

18 - 7

calculate

new lowest old lowest = lowest (0) 1 1 (1) 1

repeat #3 6 for each i (always looking only at lowest i) This defines a gradient on a multidimensional (1,N) surface. We seek the minimum of this hypersurface. Take a step in direction of steepest descent by an amount determined by |/steepest| (small slope, small step; large slope, large step). This completes 1st iteration. All values of {i}are improved.

8.

Return to #3, iterate #3-7 until convergence is obtained.

Nonlinear variations are much slower than linear variations. Typically use ENORMOUS LINEAR {} basis set. Contract this basis set by optimizing nonlinear parameters (exponential scale factors) in a SMALL BASIS SET to match the lowest {}s that had initially been expressed in large basis set.

modified 10/9/02 10:21 AM

5.73 Lecture #18

6. Alternative Strategies

18 - 8

* rigorous variational minimization of Elowest: ab initio * constrain variational function to be orthogonal to specific subset of functions e.g. orthogonal to ground state to get variational convergence. Applies only to higher members of specific symmetry class or orthogonal to core: frozen-core approximation. Pseudopotentials (use some observed energy levels to determine Zeff(r) of frozen core)

* least squares fitting minimize differences between a set of measured energy levels (or other properties) and a set of computed variational eigen-energies (or other properties computed from variational wavefunctions).

{observed E n } {parameters in Heff }

molecular constants experimental s in finite variational basis set * semi-empirical model
replace exact H by a grossly simplified form and restrict basis set to a simple form too. Then adjust parameters in H to match some observed pattern of energy splittings. Use parameters to predict unobserved properties or use values of fit parameters to build insight.

modified 10/9/02 10:21 AM

5.73 Lecture #19

19 - 1

Density Matrices I (See CTDL pp. 252-263, 295-307**, 153-163, 199-202, 290-294) Last time: Variational Method Linear variation: 0 = H S 0 = H d = cn n =0 dc n n [Variation vs. pert. theory TODAY phase ambiguity but for all observables each state always appears as a bra and a ket. what is needed to encode motion in the probability density? A superposition of eigenstates belonging to several different values of E. coherent superposition vs. statistical mixture: think about polarized light no phase ambiguity coherences in off-diagonal position populations along diagonal

A = Tr ( A) = Tr ( A )

Quantum Beats prepared state detection D (detect or destroy coherences)

(t )
t

d i = [ , ] + t h dt d = [ (t ), ] ih * state: dt

* evolution: H * detection: D

modified 10/9/02 10:20 AM

5.73 Lecture #19

Let us define a quantity called Density Matrix
a pure state

19 - 2

can be any sort of QM wavefunction * eigenstate of H * coherent superposition of several eigenstates of H

but cannot represent a statistical (i.e. incoherent) mixture of several different s However, can represent a statistical (i.e. equilibrium) mixture of states p k k k = p k k
k k

pk =1
Example * one beam of linarly polarized light, with the polarization axis (-field) y * 45
e = 2 1/2 e x + e y

x two beams of linearly polarized light, 50% along x, 50% along y.

These 2 cases seem to be identical if you make 2 measurements with analyzer polarizers along x then y. But they are different with respect to 2 measurements with analyzer polarizers along 2 1/2 e x + e y then 2 1/ 2 e x e y . In the statistical mixture, it does not matter how the analyzer is oriented.

modified 10/9/02 10:20 AM

5.73 Lecture #19

What are the properties of ?

19 - 3

1. for a pure state is Hermitian with positive elements along diagonal and other elements off-diagonal.

nm = n m
cn
c* m

= cn n

nm = c n c* m
but

can expand | in any basis set but H eigenbasis is most useful.

( )nm

* = mn = [ m n

= m n = n m = nm
=

nn = n n = c n c* = c n 0 n
positive along diagonal 2. 2 2 Example Coherent Superposition vs. Statistical Mixture

1 = 2 1/ 2 1

cs =

1 1 1 1 (1 1) = 1 1 1 2 1 2

Trace = 1

Now consider a statistical mixture state.

modified 10/9/02 10:20 AM

5.73 Lecture #19

sm =
1 1 1 0 0 (1 0) + 2 1 (0 1) 2 1 1 0 trace = 1 = 0 1 2

19 - 4

The difference is in the off-diagonal positions of diagonal elements populations off-diagonal elements coherences
Expectation values of A in terms of

(statistical mixture states have strictly diagonal )

A = A = |k k A j j|
j,k

= j k A kj
j,k j

= ( A) jj Trace( A)
Could have arranged the factors A kj j k = (A )kk = Trace(A) j,k k
A = Trace( A ) = Trace( A)
So describes state of system, A describes a measurement to be made on the system simple prescription for calculating A The separation between initial preparation, evolution, and measurement of a specific observable is very convenient and instructive.

jk

modified 10/9/02 10:20 AM

5.73 Lecture #19

Example: Quantum Beats Preparation, evolution, detection magically prepare some coherent superposition state (t)

19 - 5

( t ) = N a n n e
n

iE n t h

( t ) = (t) (t)

Several eigenstates of H. Evolve freely without any time-dependent intervention

2 N = a n n

1/2

normalization

Case (1): Detection: only one of the eigenstates, 1, in the superposition is capable of giving fluorescence that our detector can see.

Thus

1 0 L D = 1 1 = 0 0 0 M 0 0

a projection operator
(designed to project out only |1 part of state vector or 11 part of .

a 2 a a* e ( E1 E 2 )t h L 1 1 2 2 a2 = N2 2 a3 O

12 = 1 2 12 = N 2a1e E1 t ha* e +iE 2 t h 2

D picks out only 1st row of .

a 1a * e i12 t 2 0 M stuff L 0 0 M M

a 2 1 2 D t = Trace(D) = N Trace 0 M = N2 a 1
2

no time dependence!

modified 10/9/02 10:20 AM

5.73 Lecture #19

19 - 6

case (2): a particular linear combination of eigenstates is bright: the initial state 21/2(1 + 2) has D = 1. a projection operator. 1 How much of the original D= 1 + 2 1 + 2 state is present in the 2 evolved state?

)(

1 1 1 + 2 2 + 1 2 2 1 0 0 L 0 0 0 L 0 1 0 0 0 0 0 1 0 0 0 = 0 0 0 0 + 0 0 0 0 + 0 2 M 0 0 0 M 0 0 0 M 1 1 0 L 1 1 1 0 0 = 2 0 0 0 0 M 0 0 0 =

+ 2 1 1 0 0 0 0 0 0 0

L 0 0 0 1 0 0 + 0 0 0 M 0

0 0 0 0

L 0 0 0

1 2 why do we need to look at Trace(D ) = N Trace only the 1,2 block of ? 2 1 * (D )11 = N 2 a1 2 + a1a 2 e +i( E1 E 2 )t h 2 1 (D )22 = N 2 a 2 2 + a1a* e i( E1 E 2 )t h 2 2 1 2 2 * Trace(D ) = N 2 a1 + a 2 + 2 Re a1 a 2 e +i12 t 2

1 -1 0 1 -1 1 0 -1/ 2 (1 2 ), then D = 2 0 0 0 if the bright state had been 2 0 0 0

0 0 0 0

]]

beat note at 12 [if the bright state had been 21/2(12), then Tr(D ) would be the same except for 2Re[ ] ]
2 2

If a1 = a2

(and a1, a2 real), Trace(D) = N 2 a1 2 [1 cos 12t ]

QUANTUM BEAT! 100% modulation!

(N2 = 1/2)

modified 10/9/02 10:20 AM

5.73 Lecture #19

19 - 7

So we see that the same (x,t) or (t) can look simple or complicated depending on the nature of the measurement operator! The measurement operator is designed to be sensitive only to specific coherences (i.e. locations in ) which oscillate at ij. THIS IS THE REASON WHY WE SEPARATE PREPARATION AND OBSERVATION SO CLEANLY.
Time evolution of nm and A

Start with the time-dependent Schrdinger equation:

H = ih

H = ih t H = ih t
n

for time-independent H we know ( t ) = a n n e iE n t h 1. (t)

(t) = (t) (t) nn (t) = n (t) (t) n = a n

2
a time independent population in state n.

i E E t h nm (t) = a n a* e ( n m ) = a n a* e i nm t a coherence which m m 2. t

oscillates at nm (eigenstate energy differences /h)

Recall ih

= H t

A A = A + + A t t t t 1 A 1 = + A H H A + t ih ih i A Heisenberg Equation = [ H , A] + of Motion h t

Note that nothing has been assumed about the time-dependence of H.

modified 10/9/02 10:20 AM

5.73 Lecture #19

19 - 8

If A commutes with H (regardless of whether H is time-dependent), there is no dynamics as far as observable A is concerned. However, if A does not commute with H, there can be dynamics of A even if both A and H are time-independent. Similarly can derive ih = [H( t), ] evolution of under H(t). This is a matrix equation. t It specifies the time dependence of each If H is element of . time dependent Summarize

A = Tr(A) = Tr(A)
info about quantity being measured info about state on which measurement is to be made

ih

= [H, ] t
state

time evolution

each expressed independently in time evolution of : H the form of matrices which can be observable quantity: A easily read (or designed!).
NMR pulse gymnastics statistical mixture states - use same machinery BUT add the independent k matrices with weights pk that correspond to their fractional populations.

initial state:

is Hermitian so can be diagonalized by TT. However, if is time-dependent, T would have to be time-dependent. This transformation gives a representation without any coherences in even if we started with a coherent superposition state. No problem because this transformation will undiagonalize H, thereby reintroducing time dependencies.
modified 10/9/02 10:20 AM

5.73 Lecture #20

Density Matrices II

20 - 1

Read CTDL, pages 643-652. Last time: , , = coherent superposition vs. statistical mixture populations along diagonal, coherences off diagonal A = Trace(A) = Trace(A )

Today:

Quantum Beats prepared state detection as projection operator D What part of D samples a specific off-diagonal element of ? Optimize magnitude of beats [partial traces] system consisting of 2 parts e.g. coupled oscillators motion in state-space vs. motion in coordinate space.

The material on pages 202, 3, 5, and 7 is an exact duplication of pages 195, 6, 7 and 8.

revised October 21, 2002

5.73 Lecture #20

Example: Quantum Beats Preparation, evolution, detection magically prepare some coherent superposition state (t)

20 - 2

( t ) = N a n n e
n

iE n t h

( t ) = (t) (t)

Several eigenstates of H. Evolve freely without any time-dependent intervention

2 N = a n n

1/2

normalization

Case (1): Detection: only one of the eigenstates, 1, in the superposition is capable of giving fluorescence that our detector can see.

Thus

1 0 L D = 1 1 = 0 0 0 M 0 0

a projection operator
(designed to project out only |1 part of state vector or 11 part of .

a 2 a a* e ( E1 E 2 )t h L 1 1 2 2 a2 = N2 2 a3 O

12 = 1 2 12 = N 2a1e E1 t ha* e +iE 2 t h 2

D picks out only 1st row of .

a 1a * e i12 t 2 0 M stuff L 0 0 M M

a 2 1 2 D t = Trace(D) = N Trace 0 M = N2 a 1
2

no time dependence!

revised October 21, 2002

5.73 Lecture #20

20 - 3

case (2): a particular linear combination of eigenstates is bright: the initial state 21/2(1 + 2) has D = 1. a projection operator. 1 How much of the original D= 1 + 2 1 + 2 state is present in the 2 evolved state?

)(

1 1 1 + 2 2 + 1 2 2 1 0 0 L 0 0 0 L 0 1 0 0 0 0 0 1 0 0 0 = 0 0 0 0 + 0 0 0 0 + 0 2 M 0 0 0 M 0 0 0 M 1 1 0 L 1 1 1 0 0 = 2 0 0 0 0 M 0 0 0 =

+ 2 1 1 0 0 0 0 0 0 0

L 0 0 0 1 0 0 + 0 0 0 M 0

0 0 0 0

L 0 0 0

1 2 why do we need to look at Trace(D ) = N Trace only the 1,2 block of ? 2 1 * (D )11 = N 2 a1 2 + a1a 2 e +i( E1 E 2 )t h 2 1 (D )22 = N 2 a 2 2 + a1a* e i( E1 E 2 )t h 2 2 1 2 2 * Trace(D ) = N 2 a1 + a 2 + 2 Re a1 a 2 e +i12 t 2

1 -1 0 1 -1 1 0 -1/ 2 (1 2 ), then D = 2 0 0 0 if the bright state had been 2 0 0 0

0 0 0 0

]]

beat note at 12 [if the bright state had been 21/2(12), then Tr(D ) would be the same except for 2Re[ ] ]
2 2

If a1 = a2

(and a1, a2 real), Trace(D) = N 2 a1 2 [1 cos 12t ]

QUANTUM BEAT! 100% modulation!

(N2 = 1/2)

revised October 21, 2002

5.73 Lecture #20

if a1 = a 2
Dt
2 2

20 - 4

a1, a 2 real

2 N 2 a1

bright state 2 1/2 ( 1 + 2 ) 1 1 0 D = 1 1 0 0 0 0

0
2 12
Dt

2 N 2 a1

bright state 2 1/2 ( 1 2 ) 1 1 0 D = 1 1 0 0 0 0

0
2 12
2

Usually a1 t = e t / - exponential decay: beats superposed on decay

what happens if a1 a 2 ? 1 2 , 2 try a1 + a 2 = 1

2 2

as mixing fractions

2a1a 2 = 2 1 2

1/2

()
max. beat amplitude when a1 = a 2
2 2

2a1a 2

2 1/ 2

revised October 21, 2002

5.73 Lecture #20

20 - 5

So we see that the same (x,t) or (t) can look simple or complicated depending on the nature of the measurement operator! The measurement operator is designed to be sensitive only to specific coherences (i.e. locations in ) which oscillate at ij. THIS IS THE REASON WHY WE SEPARATE PREPARATION AND OBSERVATION SO CLEANLY.
Time evolution of nm and A

Start with the time-dependent Schrdinger equation:

H = ih

H = ih t H = ih t
n

for time-independent H we know ( t ) = a n n e iE n t h 1. (t)

(t) = (t) (t) nn (t) = n (t) (t) n = a n

2
a time independent population in state n.

i E E t h nm (t) = a n a* e ( n m ) = a n a* e i nm t a coherence which m m 2. t

oscillates at nm (eigenstate energy differences /h)

Recall ih

= H t

A A = A + + A t t t t 1 A 1 = + A H H A + t ih ih i A Heisenberg Equation = [ H , A] + of Motion h t

This is a scalar equation, not a matrix equation. It tells us about the motion of the center of a wavepacket. Note that nothing has been assumed about the time-dependence of H.
revised October 21, 2002

5.73 Lecture #20

Nonlecture

20 - 6

= [ ]= + t t t t 1 1 = + ih ih 1 = [ ] ih ih = [, ] t
no requirement that H be independent of t. But if H is independent of t, then take matrix elements of both sides of equation.

ih jk = j H H k = E j jk jk E k = E j E k jk jk = i E j E k jk h

jk (t) =

i E j E k t e h

jk (0)

Time evolution of all coherences in the absence of external manipulation!

revised October 21, 2002

5.73 Lecture #20

20 - 7

If A commutes with H (regardless of whether H is time-dependent), there is no dynamics as far as observable A is concerned. However, if A does not commute with H, there can be dynamics of A even if both A and H are time-independent. Similarly (as on page 20 - 6) can derive ih = [H( t ), ] evolution of under H(t). t This is a matrix equation. If H is It specifies the time time dependence of each dependent element of . Summarize

A = Tr(A) = Tr(A)
info about quantity being measured info about state on which measurement is to be made

ih

= [H, ] t
state

time evolution

each expressed independently in time evolution of : H the form of matrices which can be observable quantity: A easily read (or designed!).
NMR pulse gymnastics statistical mixture states - use same machinery BUT add the independent k matrices with weights pk that correspond to their fractional populations.
is Hermitian so can be diagonalized by TT = . However, if is time-dependent, T would have to be time-dependent. This transformation gives a representation without any coherences in even if we started with a coherent superposition state. No problem because this transformation will undiagonalize H, thereby reintroducing time dependencies.
revised October 21, 2002

initial state:

5.73 Lecture #20

Systems consisting of 2 parts: method of partial traces e.g. coupled harmonic oscillators direct product representation

20 - 8

2 recall anharmonically coupled oscillators, k122 , q1q2 , (q1, q2 ) = (q ) (q ) v1 1 v2 2

( x1, x 2 ) = 1,n1 ( x1 ) 2,n 2 ( x 2 ) = (1) (2) has 4 indices n n ;n n = n1 n 2 n 2 n1

1 2 1 2

n1, n 2

We might want to measure expectation value of operator that operates on both systems 1 and 2: A(1,2)

A = Trace( A) = ( A)n n ;n n 1 2 1 2
n1 ,n 2
Alternatively, we might want to measure expectation value of an operator that operates only on system 1: B(1). To use Trace( B) method, need concept of partial traces and need to formally extend B to act as dummy operator on system 2.

B(1) = B(1)

1(2)

Several types of initial preparation are possible: 1. pure state of 1 2 (a tensor product state) 2. statistical mixture in 1, pure state in 2. 3. statistical mixture in both. Entanglement! Handout from 10/11/02. Science. Several types of observation are possible: 1. separate observation of subsystem 1 or 2 2. simultaneous measurement of both systems
revised October 21, 2002

5.73 Lecture #20

CTDL use this definition of B(1) (page 306) to prove that B(1) = Tr ( (1)B(1))

20 - 9
calculated as if system 1 were isolated from system 2

for coupled HO system

operator of type (1,2) a1 a1a a 2 2
type (1) a1 a1

(a correlated property of two parts of the system)

or (2) a a 2 2 or (1 + 2)

(a1a1 + a2a2 )

t = 0 wavepacket is located at turning point of v2 = 5 in oscillator #2 and at x1 = 0 for oscillator #1

( x1, x 2 , t = 0) = a n 2 0, n 2
n 2 =0
(0 )

(0)

suppose we have 1 = 22 P = 2 n1 + n 2 polyads. and only the 0,P state is " bright" (i.e. excitation is initially in oscillator #2)
eigenvectors of H(1,2) expressed in HO Basis set

P/2 need to write 0, P (0) as b n n, P 2n n=0

TH P T columns of T
The initial state is a coherent superposition of several polyads. Motion occurs in both coordinate space and state space. Each kind of motion is sampled by a different class of diagnostic.
revised October 21, 2002

5.73 Lecture #20

20 - 10
-iE p,n t h

so that we can use E p,n in e to express ( x1, x 2 , t )

get motion of w.p. on

V(x2)

get motion of pieces of state vector within each Polyad P. Could want expectation values of quantities like
N1 = a1 a1 state space N2 = a2 a2 2N1 ( t ) + N2 ( t ) = P

x1 = 2 1/2 a1 + a1 coordinate space x1 x 2 == 2 3/2 a1 + a1 a2 + a 2 2

)(

)2

revised October 21, 2002

5.73 Lecture #21

3D-Central Force Problems I
Read: C-TDL, pages 643-660 for next lecture.

21 - 1

All 2-Body, 3-D problems can be reduced to * a 2-D angular part that is exactly and universally soluble * a 1-D radial part that is system-specific and soluble by 1-D techniques in which you are now expert Next 3 lectures:
Correspondence Principle Commutation Rules all matrix elements

Roadmap 1. define radial momentum pr = r-1(q p - ih) 2. define orbital angular momentum L = q p
general definition of angular momentum and of vector operators

also L L = ihL and L i ,L j = ih ijk L k k

3. separate p2 into radial and angular terms: p2 = p2 + r 2 L2 r 4. find Complete Set of Commuting Observables (CSCO) useful for blockdiagonalizing H H,L2 = [H, L i ] = L2 ,L i = 0 H,L2 ,L i CSCO

L, M L universal basis set

p2 h 2 l(l + 1) 5. separate radial H l (r) = r + V(r) + 2 part of H: 2r 2
1-D Schrd Eq.

effective radial potential

V l (r)

6. ALL Matrix Elements of Angular Momentum Components Derived from Commutation Rules. 7. Spherical Tensor Classification of all operators.

8. Wigner-Eckart Theorem all angular matrix elements of all operators. I hate differential operators. Replace them using exclusively simple Commutation Rule based Operator Algebra.

revised October 21, 2002 @ 10:15 AM

5.73 Lecture #21

21 - 2
r r q1 q 2

Lots of derivations based on classical VECTOR ANALYSIS much will be set aside as NONLECTURE Central Force Problems: 2 bodies where interaction force is along the vector 1
r r1 r q1

CM r q
r r2

12

r q cm
r q2

r r r q 2 = q1 + q12 r r r q12 = q 2 q1 = ( x 2 x1 ) + ( y 2 y1 ) + k (z 2 z1 ) i j
r 2 2 2 1/2 r q12 = ( x 2 x1 ) + ( y 2 y1 ) + (z 2 z1 )

origin also C.M. Coordinate system r r r r1 = q1 q cm r r v r2 = q 2 q cm

[ r1 [ r2

r = m2 M

] r = m1 M ]

H = H translation + H center of mass

motion of fictitious free translation particle of mass of C of M of system of mass = m1m2 M = m1 + m2 m1 + m2 in coordinate system with origin at C of M (CTDL page 713)

LAB

Htranslation

2 Ptrans = + V 2( m1 + m2 ) constant

free translation of system with respect to lab (not interesting) motion of particle of mass with respect to origin at c. of m.

BODY

1 2 HC . M . = P + V( r) 2 cm free{ rotation

(no , dependence)

2 GOAL IS TO SIMPLIFY Pcm because that is only place where , degrees of freedom appear.

revised October 21, 2002 @ 10:15 AM

r 1. Define Radial Component of P cm

5.73 Lecture #21

21 - 3

Correspondence Principle * classical mechanics * Cartesian Coordinates * symmetrize to avoid failure to satisfy Commutation Rules

** verify that all three derived operators, p, pr and L are Hermitian satisfy [q,p]=ih Purpose of this lecture is to walk you through the standard vector analysis and Quantum Mechanics Correspondence Principle procedures

r q + + kz ix jy r p x + y + kp z ip jp r x2 + y2 + z2

= [q q ]1/2 =|q| r r find radial (i.e. along q) part of p

1/2

r r project p onto q q p = q p cos

r q

r p

qp cos(q, p) = { qp

qp qp = r qp

radial component of p is r r obtained by projecting p onto q

p r = p cos = p

r r so from standard vector analysis we get p r = r 1q p

revised October 21, 2002 @ 10:15 AM

5.73 Lecture #21

21 - 4

This is a trial form for pr, but it is necessary, according to Correspondence Principle, to symmetrize it.
pr = 1 1 r (q p + p q ) + (q p + p q )r 1 4

arrange terms in all possible orders! NONLECTURE (except for Eq. (4)) SIMPLIFY ABOVE Definition to pr = r 1(q p ih)

(r is not a vector)

[q, p] is a vector commutator be careful r r [q, p] = [x,px ] + [y,py ] + [z,pz ] = 3ih r r p q = q p [q, p]
pr = r r r r 1 1 r (2q p [q, p]) + (2q p [q, p])r 1 4

r r

(1)

1 = r 1 4q p r 1 2q p + 2q pr 1 6ihr 1 2 3 4 1444 444 add and subtract 2r 1qp 3 1 = r 1q p ihr 1 + q p,r 1 2 2

(2)

(3)

1 LEMMA: need more general Commutation Rule for which q p,r is a special case

r r r 1st simplify: [ f(r),q p] = q [ f(r), p] + [ f(r), q ] p

revised October 21, 2002 @ 10:15 AM

5.73 Lecture #21

but, from 1-D, we could have shown

21 - 5

[f (x), p] = f (x) h x h x (f (x)) i i

=

[f (x), p] = ih x

h [f (x) f f] = ihf (x) if

for 1 - D

Thus, in 3-D, the chain rule gives

f r f r f r r +k +j f (r ), p] = ih i [ r z r y r x
evaluate these first
1/2 r 2 = x + y 2 + z2 = x x2 + y2 + z2 x x r r etc. for & y z

1/2

= x /r

r r f q f x y z Thus [f (r ), p] = ih i + j + k = ih r r r r r r r r f f x 2 + y 2 + z2 f (r ), q p] = q [f (r ), p] = ih [ = ih r r r r

[f(r),q p] = ih

f r r

this is a scalar, not a vector, equation

(4)

But we wanted to evaluate the commutation rule for f(r) = r1

[
RESUME HERE

plug this result into (3)

pr = r 1q p

1 r , q p = ih r = ihr 1 r r
1

(5)

3 1 ihr 1 + ihr 1 2 2

)
(6)

pr = r 1(q p ih)

This is the compact but non-symmetric result we got starting with a carefully symmetrized starting point as required by Correspondence Principle.
revised October 21, 2002 @ 10:15 AM

5.73 Lecture #21

* This result is identical to result obtained from standard vector analysis IN THE LIMIT OF h 0. show [r,pr] = ih show pr is Hermitian

21 - 6

Still must do 2 things:

[r, pr ] = [r, r 1(q p ih)]

= r 1[r, q p]

= r 1[r, q p] r 1[r, ih] + r, r 1 (q p ih) Use Eq. (4)

[r, q p] = ihr
* *

[r, p r ] = ih

we do not need to confirm that pr is Hermitian because it was constructed from a symmetrized form which is guaranteed to be Hermitian.

Correspondence Principle! 2. Verify that Classical Definition of Angular Momentum is Appropriate for QM.

k i j r r r L=qp= x y z px p y pz

(7)

We will see that this definition of an angular momentum is acceptable as far as the correspondence principle is concerned, but it is not sufficiently general.

NONLECTURE

r What about symmetrizing L?

L x = yp z zp y = p z y p y z r r = ( p q )x p q = L

PRODUCTS OF NON-CONJUGATE QUANTITIES

revised October 21, 2002 @ 10:15 AM

5.73 Lecture #21

r q p p q = 2L
r But is L Hermitian as defined? BE CAREFUL:

21 - 7
symmetrization is impossible! antisymmetrization is unnecessary!

qp+pq = 0

(q p) p q !

go back to definition of vector cross product

L x = ypz zpy L = py p z = pz y py z = ypz zpy = L x x z y

(p,q are Hermitian)

r L is Hermitian as defined.

RESUME 3A. Separate p2 into radial and angular terms.

GOAL: vector analysis
2 p2 = pr + r 2 L2 r r r p = p|| + p

r (|| and with respect to q)

p

(8)

r p

r q

p||

Classically

r r r p = r 2 q(q p) q (q p) r r component component in q,p r r

|| to q

r L 678

(9)

plane which is to q (is the sign correct?)

r2 is needed in both terms to remain dimensionally correct.

revised October 21, 2002 @ 10:15 AM

5.73 Lecture #21

talk through this vector identity
1st term (p|| ): q p = q p cos r r q / q = unit vector along q

21 - 8

2nd term (p ):

q p points up out of paper

finger palm

thumb

thumb }

q p is in plane of paper in opposite direction of p , { finger hence minus sign.

Is it necessary to symmetrize Eq. (9)? NONLECTURE Examine Eq. (9) for QM consistency x component

p x = r 2 x xp x + yp y + zp z yL z zL y but px = r

yL z zL y = y xp y yp x + z( xp z zp x )
2

[(

[(x

) (

+ y + z p x + ( xy yx )p y + ( xz zx)p z = p x

)]

similarly for py , p z
r Symmetrize? No, because 2 parts of p are already shown to be Hermitian.
RESUME

revised October 21, 2002 @ 10:15 AM

5.73 Lecture #21

3B. Evaluate pp r p2 = pr 2 [q(q p) q (q p)]

21 - 9
(10)

[goal is p2 = pr2 + r2L2 ]

r commute p through r-2 to be able to take advantage of classical vector triple product

NONLECTURE
r [p,r2 ] = ihi x r2 + j y r2 + k z r2 r = 2ihr 4q

f Recall [f(x),p x ] = ih x 2 r 1 2x because r = 2r 3 = 2r 3 = 2x r 4 2 r x x

r r r thus pr 2 = r 2 p + 2ihr 2 q

( ) r r p2 = r 2 (p + 2ihr 2 q )[q(q p) q (q p)]

(11) (12)

get 4 terms
p2 = r 2 (p q )(q p) r 2 p [q (q p)] + r 2 (2ih)r 2 (q q )(q p) r 2 (2ih)r 2 q [q (q p)]

II

III

IV

rpr + ih I = r 2 (q p 3 ih)(q p) 2 1 r (q p ih)(q p) = r pr (q p) III = r 2 ( 2 ih)(q p)

II = r 2 (p q) (q p) = r 2 L2 = r 2 L2 0 IV = r 4 ( 2 ih)(q q) (q p)

( )

p2 = r 1pr (rpr + ih) + r 2 L2 = r 1 [rpr ih]pr + r 1pr ih + r 2 L2 = p2 + r 2 L2 r

(13)

rpr-[r,pr]

revised October 21, 2002 @ 10:15 AM

5.73 Lecture #21

RESUME
2 This p2 = pr + r 2 L2 equation

21 - 10

is a very useful and simple form for p2 separated into additive radial and angular terms! If H can be separated into additive terms, then the eigenvectors can be factored. SUMMARY
pr = r 1(q p ih)
2 p2 = pr + r 2 L2

radial momentum separation of radial and angular terms

p2 L2 H= r + + V(r) 2 2r 2 eventually h 2 l(l + 1) + V(r) V l (r) = 2r 2

Next: properties of L i , L2 CSCO

revised October 21, 2002 @ 10:15 AM

5.73 Lecture #22

3D-Central Force Problems II
Last time:

22 - 1

[x,p] = ih vector commutation rules: generalize from 1-D to 3-D conjugate position and momentum components in Cartesian coordinates Correspondence Principle Recipe Cartesian and vector analysis Symmetrize (make it Hermitian) classical in h 0 limit

Derived key results:

f [f (x), px ] = ih x

f (r ) , q p = ih

*pr = r 1 (q p ih) *p2 = p2 + r 2 L2 r

*L = q p

f df r r based on and r = x 2 + y2 + z2 r dr x

1/2

(came from symmetrization in Cartesian coordinates)

operator algebra gave simple separation of variables not necessary (or possible) to symmetrize Vl (r ) radial effective potential
We do not yet know anything about L2 nd Li.

L2 *H = + + V(r) 2 2 r2
p2 r
TODAY

[purpose is mostly to practice [,] and angular momentum algebra]

Obtain angular Momentum Commutation Rules Block diagonalize H ijk Levi-Civita Antisymmetric Tensor useful in derivations, vector commutators, and remembering stuff. Next Lecture: Begin derivation of all angular momentum matrix elements starting from Commutation Rule definitions of angular momentum.

revised 21 October, 2002 @ 10:19 AM

5.73 Lecture #22

GOALS

22 - 2
any scalar function of scalar r. difficult - need ijk
2 j

1. 2. 3. 4. 5.

[Li , f (r )] = 0 [Li , pr ] = 0
i i 2 r 2

[L , p ] = 0 [L , L ] = 0 ( but [L , L ] 0!)
i

C.S.C.O.

H , L2 , Li block diagonalize H

These 1-4 are chosen to show that all terms in H commute with L2 and Li

1.

[Lz , f(r)] = [xpy ypx , f(r)] = x[py , f ] + [x, f ]py y[px , f ] [y, f ]px
[ x, f ] = 0,
i [y, f ] = 0 because [q, f(r)] = 0 + 0 j + 0 k r

recall [f (r ), px ] = ih

f r f x = ih r x r r [Lz,f (r )] = ih f x y y x = 0 r r r

2.

[Lz , pr ] = [Lz , r 1 (q p ih)] = [Lz , r 1q p] 0 1 = [L z , r ]q p + r 1 [L z , q p] r r [Lz , q p] = q [Lz , p] + [Lz , q] p two vector commutators on RHS
r Note that q is not f(r )!
need to define special notational trick to evaluate these DIFFICULT COMMUTATORS

revised 21 October, 2002 @ 10:19 AM

5.73 Lecture #22

Levi-Civita Symbol cyclic order adjacent interchange 2 repeated indices

22 - 3
ijk xyz = yzx = zxy = +1 yxz = zyx = xzy = 1 xxy = etc.= 0

I claim L i , p j = ih ijk pk . This will become the definition of a vector operator k with respect to L.

Nonlecture: Verify claim for 1 of 3 3 = 9 possible cases let i = x, j = y

[Lx ,py ] = [ypz zpy ,py ] = [ypz ,py ] + 0 0 = [y,py ]pz + y[pz ,py ]
= ihpz

Now check this using ijk

[L ,p ] = ih
x y k

xyk

pk = ih xyx px + xyy py + xyzpz = ihpz .

OK

All other 8 cases go similarly Other important Commutation Rules

[Li ,p j ] = ih ijkpk k [Li ,q j ] = ih ijkq k k [Li ,L j ] = ih ijk Lk k

general definition of a vector operator general definition of an angular momentum. Works even for spin where q p definition is inapplicable

All angular momentum matrix elements will be derived from these commutation rules. FOR THE READER: VERIFY ONE COMPONENT OF EACH OF THE THREE ABOVE COMMUTATORS

revised 21 October, 2002 @ 10:19 AM

5.73 Lecture #22

22 - 4

[Li ,L j ] = ih ijk Lk is identical to k

(

L L = ihL r r expect 0! because vector cross product A B = | A || B | sin e AB

k i j L L L L + L L L L + k L L L L L L = Lx L y Lz = i y z j z y x z y x z x x y L L L x y z jL = ih x + y + kLz = ihL iL
This vector cross product definition of L is more general than q p because there is no way to define spin in q p form but S S = ihS is quite meaningful. Can one generalize that, if L L = ihL (instead of 0), and the [Li, Lj] and [Li, pj] commutation rules have similar forms, that L p = ihp? NO! Check for yourself! 2. Continued.

r r [L ,p ] = r q [L ,p] + r [L ,q] p r L , p] = ih ( + + k )p i j [
1 1 z r z z i ixk iyk izk

vector commutators
k

sum of 3 terms

r q [L i , p] = ih x ixk + y iyk + z izk pk

k

only one of these terms is nonzero

= ih ijk q jpk
j, k

(1)

revised 21 October, 2002 @ 10:19 AM

5.73 Lecture #22

r r and the other term [L i , q ] p

22 - 5

r [L ,q] = ih
i k i k

[i

ixk

+ iyk + k izk qk j

r [L ,q] p = ih [

ixk

qk px + iyk qk py + izk qk pz

]
(2)

= ih ijk qk p j = ih ikjq jpk

j, k k, j

(kj labels permuted) (k

= ih ijk q jpk
k, j

switch order of j and k

putting Eqs. (1) and (2) together

q [L i ,p] + [L i ,q ] p = ih ijk q jp k ijk q jp k = 0!

j,k
Elegance and power of ijk notation! We have shown that: * [Li,pr] = 0 for all i * easy now to show [Li,pr2] = 0

Finally L i ,L2 = L i ,L2 = L j L i ,L j + L i ,L j L j j

] j[

] j( [

] [

] ]

= L j ih ijk L k + ih ijk L k L j k k j
same trick: permute j k indices in second term ijk = ikj

ih ijk L j L k k

=0
But be careful: L i ,L2 = L j L i ,L j + L i ,L j L j = ih L j ijk L k + ijk L k L j j
k k

] [

because this is a sum only over k, so cant combine and cancel terms.
revised 21 October, 2002 @ 10:19 AM

5.73 Lecture #22

for i = x, j = y

22 - 6

[L ,L ] = ih[L L
x 2 y

y z

+ Lz L y 0!

so we have shown

[L2 ,Li ] = 0 [L2 ,f(r)] = 0

[Li ,f(r)] = 0

[L2 ,pr ] = 0
[Li ,pr ] = 0

L2, Li, H all commute Complete Set of Mutually Commuting Operators

eigenfunction of L2 with eigenvalue h 2 L( L + 1)

So what does this tell us about L H L = ?

BLOCK DIAGONALIZATION OF H!

Basis functions

= ( r ) L2 , L z = nLM L
radial special angular universal

eigenfunctions of Lz eigenfunctions of L2 which radial eigenfunction?

Next time I will show, starting from

[L ,L ] = ih
i j k

ijk

Lk

, that

* *

L2 nLM L = h 2L (L + 1) nLM L L z nLM L = hM L nLM L

L = 0,1, M L = L, L + 1, + L

also derive all L x and L y matrix elements in nLM L basis set.

revised 21 October, 2002 @ 10:19 AM

5.73 Lecture #23

Angular Momentum Matrix Elements LAST TIME:

23 - 1

* all [, ]=0 Commutation Rules needed to block diagonalize p 2 L2 H= r + + V (r) in nLM L basis set 2 2 2r * ijk Levi-Civita antisymmetric tensor useful properties * Commutation Rule DEFINITIONS of Angular Momentum and Vector Operators [Li , L j ] = ih ijk L k
k

[Li , Vj ] = ih ijk Vk
k

Classification of operators: universality of matrix elements.

TODAY: Obtain all angular momentum matrix elements from the commutation rule definition of an angular momentum, without ever looking at a differential operator or a wavefuncton. Possibilities for phase inconsistencies. [Similar derivation for angular parts of matrix ( elements of all spherical tensor operators, Tq k ).]

1. 2. 3. 4

Define Components of Angular Momentum using a Commutation Rule. Define eigenbasis for J2 and Jz show 2 raising and lowering operators (like a, a and x i p ) ~
~

J| gives eigenfunction of Jz belonging to h eigenvalue and eigenfunction of J2 belonging to eigenvalue 5. Must be at least one MAX and one MIN such that J(J+|MAX) = 0 J+(J|MIN) = 0
n n n This leads to max = h , = h2 + 1 . 2 22

6.

Obtain all matrix elements of Jx, Jy, J, but there remains a phase ambiguity Standard phase choice: Condon and Shortley
revised 4 November, 2002

7.

5.73 Lecture #23

23 - 2

1. Commutation Rule J i , J j = ih ijk J k k This is a general definition of angular momentum (call it J, L, S, anything!). Each angular momentum generates a state space. 2. eigenfunctions of J2 and Jz exist J 2 =
J z =

(Hermitian operators. Guaranteed by symmetrization.)

but what are the values of ,? J2 and Jz are Hermitian, therefore , are real 3. find upper and lower bounds for in terms of : 2
( J 2 J 2 ) = 2 z

Want to show that this is 0.

but J 2 = J 2 + J 2 + J 2 x y z J2 J2 = J2 + J2 z x y

2 = J 2 + J 2 x y
completeness

2 =

[ J
x

J x

+ J y J y

Hermitian: 2 =

J x = J x * J x
2

+ J y

0 and MAX 1/ 2 , MIN 1/ 2

Thus 2 0 and 2 0

revised 4 November, 2002

5.73 Lecture #23

4. Raising/Lowering Operators

23 - 3
J = J ) +

J J x iJ y

[J z , J

(not Hermitian: ] = [J , J ] i[J , J ]

z x z y

( just like a,a )

= ihJ y i( ihJ x ) = h J x iJ y = hJ

J z J = J J z hJ

right multiply by

J z ( J ) = J ( J z ) hJ = J hJ = ( h)( J

(J

) is an eigenfunction of J z belonging to eigenvalue h.

Thus J raises or lowers J z eigenvalue in steps of h.

Similar exercise for J 2 , J to get effect of J on eigenvalue of J 2

[J

, J = J2 , J x i J2 , J y = 0

(J ) belongs to same eigenvalue of J 2 as

J has no effect on .

J 2 ( J ) = J J 2 = ( J

] [

] [

(We already know that [J

)

, Ji = 0

] )

revised 4 November, 2002

5.73 Lecture #23

* * * upper and lower bounds on are 1/2 J raises/lowers by steps of h Since J x =
1 1 (J + + J ) and J y = (J + J ), 2 2i

23 - 4

The only nonzero matrix elements of Ji in the | basis set are those where = 0, h and = 0. As for derivation of Harmonic Oscillator matrix elements, we are not assured that all differ in steps of h. Divide basis states into sets related by integer steps of h in .
2 5. For each set, there are MIN and MAX :

Thus, for each set

J + MAX = 0 J MIN = 0

but

J J + = J x iJ y J x + iJ y = J 2 + J 2 + iJ x J y iJ y J x x y = J2 + J2 + i J x , J y x y = J 2 + J 2 + i (ihJ z ) x y = J 2 + J 2 hJ z x y

)(

but

J 2 + J 2 = J 2 J 2 , thus x y z J J + = J 2 J 2 hJ z z

0 = J J + MAX = J 2 J 2 hJ z MAX z
2 MAX

( = (

hMAX MAX

= 2 + h MAX MAX
Similarly for MIN J + J MIN = 0
revised 4 November, 2002

5.73 Lecture #23

23 - 5

J + J = J 2 J 2 + hJ z z

= 2 h MIN MIN
subtract 2 equations for 0 = 2 2 + h( MAX + MIN ) MAX MIN 0 = ( MAX + MIN )( MAX MIN + h)
Thus MAX = MIN
(impossible)

now factor this equation

OR MAX = MIN h

Thus for each set of , goes from MAX to MIN in steps of h

MAX = MIN + nh = MAX + nh

MAX =

n h 2

Thus is either integer or half integer or both! Thus there will at worst be only two non-communicating sets of because if were both integer and 1/2-integer, each would form a set of -values, the members of which would be separated in steps of h. Now, to specify allowed values of :

2 MAX

+ h MAX
n j 2

nn n n h + h h = h2 = +1 2 2 22

let

MAX = hj MIN = hj = h j ( j + 1)
2

j either integer or half integer or both

revised 4 November, 2002

5.73 Lecture #23

rename our basis states
J 2 jm = h2 j ( j + 1) jm J z jm = hm jm

23 - 6

valid for all operators that satisfy [A i , A j ] = ih ijk A k

k

OK to define an ama basis set for any angular momentum operator defined as above.

6. Jx, Jy, J matrix elements recall page 23-3, but in new notation

jm 1 = N J jm

(J raises / lowers m by 1)
) ( N J
jm

normalization factor (to be determined below)

1 = jm 1 jm 1 = ( N J jm
* N = N

)
!

J = J m
1 = N
2

jm J m J jm
2 x 2 y

J m J = ( J xmiJ y )( J x iJ y ) = J + J i[ J x , J y ] = J 2 J 2 i(ihJ z ) = J 2 J 2mhJ z z z = J 2 J z ( J z h)

use this to evaluate matrix elements of JmJ

1 = N N =

[h

j ( j + 1) h2 ( m( m 1))]
1 / 2

[ j( j + 1) m(m 1)] h

e i
arbitrary phase factor from taking square root

1/ 2 J jm = h[ j ( j + 1) m( m 1)] jm 1 e i

Usual phase choice is = 0 for all j, m: the Condon and Shortley phase choice

(sometimes = 2

so be careful)
revised 4 November, 2002

5.73 Lecture #23

std. phase choice: = 0
1/ 2 j m J jm = h j j m m 1[ j ( j + 1) m( m 1)]

23 - 7

since J x =

1 (J + + J ) 2

(or h

( ) jj m m 1[ j ( j + 1) m m ]
h

1/ 2

remember matrix elements of x and p in harmonic oscillator basis set?

1/ 2

j m J x jm =

j j m m +1[ j ( j + 1) m( m + 1)]

1/ 2 + m m 1[ j ( j + 1) m( m 1)]

}
1/ 2

Jy =

1 (J + J ) 2i

two sign surprises

j m J y jm = i j j m m +1[ j ( j + 1) m( m + 1)] 2

1/ 2 m m 1[ j ( j + 1) m( m 1)]

This phase choice leaves all matrix elements of J 2 , J x and J real and positive, but those of J y imaginary

[if use

= + 2 , this gives J y real and J x , J imaginary]

Summary

j m J 2 jm = j j m m h2 j( j + 1) r jm J jm = khm r h 1/2 jm 1 J jm = m ij i j( j + 1) m ( m 1) 2 1 1 iJ x + y = i ( J + + J ) + ( J + J ) jJ j 2 2i 1 1 i = J + ij + J + ij i 2 2

) [

revised 4 November, 2002

5.73 Lecture #24

Last time:
starting with [ J i , J j ] = ih ijk J k
k

24 - 1
J Matrices
DEFINITION !

J 2 jm = h 2 j( j + 1) jm J z jm = hm jm J = J x iJ y J jm = h[ j( j + 1) m(m 1)]
1/ 2

jm 1

nonzero matrix elements and Condon Shortley phase choice j m J 2 jm = h 2 j( j + 1) j j m m j m J z jm = hm j j m m j m J jm = h[ j( j + 1) mm ] j j m m 1

1/ 2

(J , J , J , J , J , J ) all stay within j

2 z x y +

all matrix elements of J 2 , J z, J x , J are real and positive (only those of J y are imaginary)
TODAY: 1. What do the matrices look like for J = 0, 1 / 2, 1? 2. many operators are expressed as an angular momentum times a constant Zeeman example density matrix r 3. other operators involve things like q or products of two angular momenta

Stark effect

Wigner-Eckart Theorem * classify operators by commutation rule * matrix elements in convenient basis sets * transform between inconvenient and convenient basis sets.

revised November 15, 2001

5.73 Lecture #24

24 - 2

[p , p ] = 0
x y

A student in 1999 suggested that he could find f(x,y) such that

2 f 2 f xy yx Thus [p x , p y ] 0!

This is possible, but f(x,y) would have to have a form that excludes it as an acceptable (x,y). Typically, the f(x,y) will have to be discontinuous or have discontinuous first derivatives. For all well behaved V(x,y), (x,y) will have continuous first derivatives. The f(x,y) used to prove a commutation rule must be acceptable as a quantum mechanical wavefunction, (x,y). This is a good thing because (see Angular Momentum Handout)

e ia p x
e ia p x
h

x1 = x1 + a

generates a linear translation of + a in x direction.

linear translations commute (but rotations do not)

This is the basis for (or a consequence of )

[p , p ] = 0 [J , J ] = ih
i j i j k

ijk

Jk

revised November 15, 2001

5.73 Lecture #24

Nonlecture prepare (excite) evolve detect
e.g.

24 - 3

e iH t

basis set 0 , 1 , 2

excite:

0 0 = 1 1

1 0 0

1 1 0 0 0 = 1 0 0 1

The excitation matrix E creates equal amplitudes in two excited eigenstates:

2 1

(0) = E 0 0 E
evolve: If we are in the eigenbasis of H
iE t a ae a b = be iE b t iE c t c ce
h

e iH t

(translation in time)

but otherwise, need [Te iT (t ) = Te iT

HTt / h

T ] = U(t, 0)

HT t

TE 0 0 ETe + iT

HT t

= U(t, 0)(0)U (t, 0)

(0) in eigenbasis of H

detect:

the detection matrix

D t = Trace(D)
Building Blocks

revised November 15, 2001

5.73 Lecture #24

Many QM operators have the form f ( J) e.g. Zeeman effect r Others have the form f (q) e.g. Stark effect r

24 - 4

r H Zeeman = B J r r HStark = eE q

(B is magnetic field)
(E is electric field)
r

Others have the form of e.g. spin - orbit

f ( J1 , J 2 ) H SO = aL S

We are going to want to be able to write matrix representations of these operators.

Let us begin by writing matrices for J2, Jz, Jx, Jy, J+, J .

j=0

only basis state is 1 1 matrix

jm = 00

J 2 00 =

same for all components j = 1/2

11 1 1 and 22 2 2
J
2 (1 / 2 )

2 2 matrices
J2 11 11 1 1 = h2 +1 22 22 2 2
m = +1/2

3 1 = h2 4 0 1 1 h 2 0

0 1 0 1 1 0
e.g.

J(z1/ 2 ) = J+
(1 / 2 )

0 1 1 0 = 0 0 0 0 0 1 0 1 = 0 0 1 0
m = 1/2

1 J(+ / 2 )

11 =0 22 1 1 11 =h 2 2 22
revised November 15, 2001

0 = h 0

1 J(+ / 2 )

5.73 Lecture #24

0 J(1/ 2 ) = h 1 J(x1/ 2 ) = J(y1/ 2 ) = 0 0 1 0 1 h 0 = 0 2 i i 0

24 - 5

1 1 0 J + + J ) = h ( 2 2 1

1 i 0 J + J ) = h ( 2i 2 1
2 2 x 2 y 2 z

verify that J = J + J + J 0 J = 4 1
2 x

(J(1/ 2) ) 2 = 3h
0 1 0 1

1 0 4 0 1
2

1 0 0 1 i 0 0 i

1 h 2 1 = 0 4 0 i h 2 1 = 0 4 0

0 J = 4 i
2 y

An amazing amount of insight gained from this complete set of 2 2 matrices

1 0 I= 0 1 0 x = 1 0 y = i = 1 z 0 1 (1/ 2) Jx 0 i (1/ 2) Jy 0 0 (1/ 2) Jz 1

CTDL, pages 417-454 1. Pauli Matrices 2. Diagonalization of 2 2 3. Geometric interpretation of 2 2 in terms of fictitious spin 1/2 4. spin 1/2 5. magnetic resonance

What is [x, y] = ?

3 matrices with eigenvalues 1

revised November 15, 2001

5.73 Lecture #24

m11 arbitrary M = m21

24 - 6

m12 m11 + m22 m m22 m + m21 m m21 I + 11 z + 12 x + i 12 y = m22 2 2 2 2

r r M = a 0I + a
scalar part of M vector part of M

1 Tr (M) 2 r 1 a = Tr(M ) 2 1 a x = Tr (M x ) 2 1 a y = Tr (M y ) 2 1 a z = Tr (M z ) 2 a0 =

Center of Gravity M

Information in 2 2 is repackaged into a 3 component vector. Visualization of dynamics!

This provides a basis for taking apart the dynamics of an arbitrary 2 2 into dynamics of x, y, z fictitious spin-1/2 components. Beat the S = 1/2 Zeeman problem to death and use it as basis for understanding dynamics of any 2 2 space.

revised November 15, 2001

5.73 Lecture #24

J=1 A set of 3 3 matrices

24 - 7

J2

(1)

1 = 2 h2 0 0 1 = h 0 0 0 0 0

0 1 0

0 0 1 0 0 1 1 0 0 0 0 1 1 0 1 i 0 i 0 1 0 0 0 0 0 1 0 0 i 0
0 1 0 1 0 0 0 1 0 = 0 0 0 0 0 0

J (z1)

J (+1)

0 = 21/ 2 h 0 0 0 = 2 h 1 0
1/ 2

(1)

J(+1) 11 = 0

J (x1)

0 = 2 1/ 2 h 1 0 0 = 2 1/ 2 h i 0

J (y1)

For 2 2 problem (e.g. J = 1/2), needed 4 independent 2 2 matrices (because there are 4 elements in a 2 2 matrix) to represent arbitrary 2 2 matrix.
for 3 3 problem, need 9 independent 3 3 matrices

(x, y,z,

x 2 , y 2 , z 2 , xy, xz, yz) (because there are 9 elements in a 3 3 matrix)

actually scalar (I), vector, tensor s+p+d

[for 2 2 it was s + p = 4]. Can you write out each of the J(3/2) matrices (16 4 4 matrices)?
revised November 15, 2001

5.73 Lecture #24

24 - 8

9 3 3 basis matrices is not nearly so nice as the 4 basis matrices for 2 2 problem. But this turns out to be what is needed to understand and picture spin = 1 systems. similarly for j = 3/2, 2, etc. There are 2 lovely consequences of being able to take an arbitrary matrix and rewrite it as sum of J matrices. 1. If M is the matrix of an operator a term in the Hamiltonian then it is clear that this operator may be re-expressed as a sum of operators, each of which behaves exactly like a (combination of) component(s) of J evaluated in the jm basis set.

M ( j ) = a0 I + ali J i J j +
i

i, j , k

c3ijk J i J j J k

+
( basis for classification of operators into Tmk )

and Wigner Eckart Theorem for evaluation of matrix elements. r 2. especially for 2 level systems, if M = and a is defined from M as on page 24-6, then we have a vector picture to understand preparation, evolution, detection

r a

/2 pulse
evolution of vector, fictitious B-fields

y
pulse
revised November 15, 2001

5.73 Lecture #24

24 - 9

Now lets do some J = 1 examples Zeeman effect for an

l

= 1, (p orbital) state
r r r L=qp r r r L q p r r L L
field strength

( up out of page) (down into page) r

r L L

e-

current on a circular wire

classical energy

r r ( L) = B L E = B L = Bzk z z
field exclusively along z

for L = 1 system:

H Zeeman

1 = Bz h 0 0

0 0 0

0 0 1

case (1) Let (0) = LML = 11

1 = = 0 (1 0 0 1 0) = 0 0 0 0 0 0 0 0

ELM L = E11 = Trace(H) 1 = hBz Tr 0 0 = hBz 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1

revised November 15, 2001

5.73 Lecture #24

What about?

0 0 0 = 0 1 0 0 0 0 E10 = Trace (H) = 0

24 - 10

E11 = + Bz h

no motion of E

( ) 1/ 2 case (2) Let 0 = 2 ( 11 + 10 ) (t ) = 2 1/ 2 [ 11 e iE t + 10 e i 0 t

11
h

]
h h

1 ( 11 11 + 10 10 + 11 10 e iE11 t + 10 11 e + iE11 t 2 1 e i11 t 0 1 1 0 (t ) = e i11 t 2 0 0 0

(t ) =

nothing at location of coherence in

1 0 0 H(B || +z) = Bz h 0 0 0 0 0 1 1 E (t ) = H = Trace(H) = Bz h(1) 2

no motion of E

Looked at 2 cases:
1. pure state 11 , B || z E = Bz h 2. mixed state 1 E = Bz h 2 2 -1/2 ( 11 + 10 ), B || z

always mixed state gives time independent E NMR: oscillating Bx, By, cw Bz

revised November 15, 2001

5.73 Lecture #24

Stark Effect: Electric field r r r classical E (qe q p + ) z z
so we will need matrix elements of x, y, z in jm basis set. How?
Based on

24 - 11

[ z, L ] = ih
j k

zjk

qk

vector operator definition later

Other angular momenta 1. l electron orbital ang. mom. 2. s electron spin 3. I nuclear spin These separate angular momenta interact with each other spin - orbit: (r ) s

Zeeman: hyperfine:

- Bz (L z + g sSz + g II z ) a IS
l

coupled and uncoupled basis sets:

m sms jlsm j
l

revised November 15, 2001

5.73 Lecture #24 case (3)

24 - 12

same (0) = 2 1/ 2 ( 11 + 10
r but H is for B || x

one at coherence in

0 1 0 H = Bx h2 1/ 2 1 0 1 0 1 0
Something subtle is intentionally wrong here. Can you find it?
1 E(t ) = Tr(H) = Bx h[e + i 11t + e i 11t + 0] 2 = Bx h cos 11t
E (t )
Bx h

revised November 15, 2001

5.73 Lecture #25 HSO + HZeeman Coupled vs. Uncoupled Basis Sets
Last time:
2 matrices for J , J+, J, Jz, Jx, Jy in jmj basis for J = 0, 1/2, 1 Pauli spin 1/2 matrices r r arbitrary 2 2 M = a0 I + a1

25 - 1

When M is visualization of fictitious vector in fictitious B-field When M is a term in H idea that arbitrary operator can be decomposed as sum of Ji. types of operators aJ e.g. magnetic moment (a is a known constant or a function of r) r q how to evaluate matrix elements (e.g. Stark Effect) J1 J 2 e.g. Spin - Orbit

TODAY: 1. 2. 3. 4. 5. HSO + HZeeman as illustrative Dimension of basis sets JLSMJ and LMLSMS is same matrix elements of HSO in both basis sets matrix elements of HZeeman in both basis sets ladders and orthogonality for transformation between basis sets. Necessary to be able to evaluate matrix elements of HZeeman in coupled basis. Why? Because coupled basis set does not explicitly give effects of Lz or Sz.

revised 4 November, 2002

5.73 Lecture #25

25 - 2

Suppose we have 2 kinds of angular momenta, which can be coupled to each other to form a total angular momentum. r L orbital r S spin operate on different coordinates or in different vector spaces r r r J = L+ S total The components of L,S, and J each follow the standard angular momentum commutation rule, but r r [L,S] = 0 , [Ji,L j] = ih ijkLk
ijk Sk . k These commutation rules specify that L and S act like vectors wrt J but as scalars wrt to each other. r J jm j r L lm r S sms
i j

[J ,S ] = i

Coupled jlsm j vs. uncoupled lm sms representations.

l

* * *

n SO H = (r )l s h l s will give a factor of anomalous g - value of e 1. H Zeeman = B (l + 2s ) ( )(l + 2s ) z z z z z 0

l

matrix elements of certain operators are more convenient in one basis set than the other a unitary transformation between basis sets must exist limiting cases for energy level patterns l and s will each give a factor of h

and 0 are in rad / s * evaluate matrix elements in both basis sets

nl
l

* look at energy levels in high field Bz >> n limit * look at energy levels in low field Bz << n limit
l

revised 4 November, 2002

5.73 Lecture #25

lower case for 1e atom angular momenta Notation: upper case for many - e angular momenta

25 - 3

two different CSCOs

a) H elect , J 2 , J z , L2 , S2 nJLSM J b) H elect ,

L

coupled basis

, L z , S2 , S z

nLM L SM S

(can be factored) recall tensor product t uncoupled basis states and entanglement (explicitly factored)

2. Coupled and Uncoupled Basis Sets Have Same Dimension

COUPLED r r r J = L+S LS J L+S

each J has 2 J + 1 M J s

J L

J = L+S L + S 1 L+S2

2( L + S ) + 1 2( L + S 1) + 1 2( L + S 2 ) + 1 LLLL 2( L S ) + 1

every J contributes 2L + 1 to sum

If L > S , there are 2S + 1 terms in sum

(2S + 1)(2L + 1) + 2[S + (S 1) + L (S)] = (2S + 1)(2L + 1) =0 total dimension of basis set for specified L,S
UNCOUPLED LM L SM S 1 31 3 2 2
2 L +1 2 S +1

total dimension (2 L + 1)(2 S + 1) again

term for term correspondence between 2 basis sets a transformation must exist: Coupled basis state in terms of uncoupled basis states:

JLSM J = a ML LM L SMS = M J M L ML 1442443 4 4

constraint

Trade J, MJ for ML, MS, but MJ = ML + MS.

revised 4 November, 2002

5.73 Lecture #25

Uncoupled basis state in terms of coupled basis states: L +S OR

25 - 4

LM J SMS =

J = L S

n
h

b J JLSM J = M L + MS 14 244 4 3
constraint

3.

Matrix elements of HSO =

A. Coupled Representation

r r r J=L+S

J 2 = L2 + S2 + 2L S J 2 L2 S2 LS = 2
(useful trick!)

J LS M L S JLSM J = J

2 J (J + 1) L( L + 1) S(S + 1) J J L L S S M J M J

)[

a purely diagonal matrix. B. Uncoupled Representation

L S = L zS z +
diagonal

1 L+ S + L S + 2
off-diagonal

L M S MS L S LM LSMS = h 2 L LS S L
cant change L

1 1/ 2 [M LMS ML ML MS MS ] + [L (L + 1) M M L ] L 2
1/ 2 S S

cant change S

[S(S + 1) M M ]

ML ML 1 MS MS m1

M L = MS = 0, 1

Nonlecture notes for evaluated matrices

S = 1 / 2, L = 0,1, 2
2

S, 2P, 2D states

revised 4 November, 2002

5.73 Lecture #25

2 S +1 2

25 - 5
= =
h

LJ

NONLECTURE for HSO : COUPLED BASIS

HSO COUPLED HSO COUPLED 2
h

S1/ 2 P

ns (0) np
2 0 0 0 0 0 0 2 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1

J = 1/2

J = 3/2

L ( 2S1/ 2 ) 0 ( 2 P1/ 2 ) 1 ( 2 P3 / 2 ) 1 ( 2 D3 / 2 ) 2 ( 2 D5 / 2 ) 2
2

J J(J+1) L(L+1) S(S+1) = 1/2 3 / 4 0 3 / 4 0 1 / 2 3 / 4 2 3 / 4 2 3 / 2 15 / 4 2 3 / 4 +1 3 / 2 15 / 4 6 3 / 4 3 5 / 2 35 / 4 6 3 / 4 +2

J =3/2

HSO COUPLED

nd

3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 2 0 0 0 0 0 0 2 0 0 0 0 0 0 2 0 0 0 0 0 0 2 0 0 0 0 0 0 2
J =5/2

center of gravity rule: trace of matrix = 0 (obeyed for all scalar terms in H)

revised 4 November, 2002

5.73 Lecture #25

2S +1 2 2

25 - 6
= h ns (1 2 0) = (0) = h np

NONLECTURE for HSO : UNCOUPLED BASIS

S P

HSO UNCOUPLED HSO UNCOUPLED

ML 1 1 0 0 1 1

MS 1 / 2 1 / 2 1 / 2 0 1/2 0 1 / 2 0 1/2 0 1 / 2 0 0 1 / 2 21/2 0 0 0 0 21/2 0 0 0 0 0 0 0 0 21/2 0 0 0 0 21/2 1 / 2 0 0 0 1 2 0 0 0

Each box is for one value of MJ = ML + MS.

D
MS

HSO UNCOUPLED = h nd

ML 2 2 1 1 0 0 1 1 2 2

1 / 2 1 1 / 2 0 1 / 2 0 1 / 2 0 1 / 2 0 1 / 2 0 1 / 2 0 1 / 2 0 1 / 2 0 1 / 2 0

0 1 1 0 0 0 0 0 0 0

0 1 1/2 0 0 0 0 0 0 0

0 0 0 1 / 2
1 /2

0 0

0 0 0 0 0 0
1 /2

0 0 0 0 0

0 0 0 0 0
1 /2

0 0 0 0 0 0 0 1 1 0

(3 / 2)
0 0 0 0 0

(3 / 2)
0 0 0 0 0 0

(3 / 2)
0 0 0

(3 / 2)

0 0 1/2 1 0

1 /2

1 / 2 0 0 0

0 0 0 0 0 0 0 0 0 1

revised 4 November, 2002

5.73 Lecture #25

4. Matrix Elements of H Zeeman = Bz (L z + 2Sz ) A. very easy in uncoupled representation
H Zeeman = Bz L M S MS L z + 2Sz LM LSMS uncoupled L = Bz h(M L + 2MS ) L LS S ML ML MS MS

25 - 7

strictly diagonal B. coupled representation

Lz + 2S z = J z + S z
easy hard no clue!

cant evaluate matrix elements in coupled representation without a new trick 5. If we wish to work in coupled representation, because it diagonalizes HSO, need to find transformation

JLSM J = a ML LM LSMS = M J M L
ML

lengthy procedure:

J = L + S

and orthogonality

Always start with an extreme ML, MS basis state, where we are assured of a trivial correspondence between basis sets:

M L = L,

M S = S,

M J = M L + M S = L + S, J = L + S

J = L + S LSM J = L + S = LM L = L SM S = S

coupled

uncoupled

revised 4 November, 2002

5.73 Lecture #25

MJ 6J8 7 68 7 J | L + S LS L + S = (L + S ) LM L = L SMS = S

25 - 8

( L + S )( L + S + 1) ( L + S )( L + S 1)

1/ 2

L + S LS L + S 1 = L( L + 1) L( L 1)

[ ] +[S(S + 1) S(S 1)]

1 /2

LL 1SS LLSS 1

1 /2

Thus we have derived a specific linear combination of 2 uncoupled basis states. There is only one other orthogonal linear combination belonging to the same value of ML + MS = MJ: it must belong to the L + S 1 LS L + S 1 basis state. lower J NONLECTURE Work this out for 2P
JLSM J = 3 / 2 1 1 / 2 3 / 2 = LM LSMS = 1 1 1 / 2 1 / 2 \ 21/ 2 1 0 1 / 2 1 / 2 + 1 1 1 / 2 1 / 2 JLSM J 1 = 31/ 2

now use orthogonality:

J 1LSM J 1 = 1 / 2 1 1 / 2 1 / 2 = 1 0 1 / 2 1 / 2 + 21/ 2 1 1 1 / 2 1 / 2 . 31/ 2

Continue laddering down to get all 4 J = 3/2 and all 2 J = 1/2 basis states.
2 3 / 2 1 1 / 2 1 / 2 = 3 1 1 / 2 1 1 / 2 1 / 2 = 3
1 /2

1 1 0 1 / 2 1 / 2 + 3 2 1 0 1 / 2 1 / 2 + 3

1 /2

1 1 1 / 2 1 / 2

3 / 2 1 1 / 2 3 / 2 = 1 1 1 / 2 1 / 2
1 /2 1 /2

1 1 1 / 2 1 / 2

You work out the transformation for 2D! Next step will be to evaluate HSO + HZeeman in both coupled and uncoupled basis sets and look for limiting behavior.

revised 4 November, 2002

5.73 Lecture #26 JLSMJ and LMLSMS HSO + HZeeman in

Last time:
HSO = l s HZeeman = Bz LZ + 2S Z JLSM J LM L SM S

26 - 1

OK to set up H in either basis problem about HZeeman in Coupled Basis need to work out explicit transformation between basis sets to evaluate matrix elements of HZeeman in coupled basis.

Today: 1. Ladders and Orthogonality method for JLSMJ LMLSMS (coupleduncoupled) transformation, term by term. 2. 3. evaluate HZeeman in coupled basis for 2P state. Correlation Diagram, Noncrossing Rule * simple patterns without calculations * guidance for intermediate case War between two limits
( * one term creates Eij0 ) 0 ( * other term causes Hij1) 0 The two terms play opposite roles in the two basis sets.

4.

Stepwise picture of level structure working out from 2 opposite limits * strong spin-orbit, weak Zeeman * strong Zeeman, weak spin-orbit Distortions from limiting patterns (via 2nd-order nondegenerate perturbation theory) give the other (pattern distorting) parameter. How does a zero-order picture identify the picture defining and the picture destroying parameters.

updated November 4, 2002

5.73 Lecture #26

26 - 2

HZeeman = Bz Lz + 2S z = Bz J z + S z

OK Not OK for coupled basis

to evaluate matrix elements of Lz or S z in JLSM J need to work out JLSM J =

ML

a m LM L SM S = M J M L
L

ladders and orthogonality

J = L + S

begin with " extreme" J = L + S M J = J ; M L = L, M S = S basis states where there is a 1:1 correspondence
JLSM J = J = LM L = L SM S = S
=L+S

J L + S L S L + S = L + S LL SS

[( L + S)( L + S + 1) ( L + S)( L + S 1)]1/ 2 L + S L S L + S 1 = [ L( L + 1) L( L 1)]1/ 2 LL 1 SS + [S( S + 1) S( S 1)]1/ 2 LL SS 1

L + S L S L + S 1 =

[2L ]1/2 LL 1
1 /2

SS + [2S ]

1 /2

LL SS 1
1 /2

[2(L + S )]

1 /2

for 2P L = 1, S = 1/2

L L + S L S L + S 1 = L + S
2 = 3
1/2

S LL 1 SS + L + S
1 + 3
1/2

LL SS 1

3 2 1 12 12

1 1 1 0 2 2

11

1 1 2 2

orthogonality: there are only 2 MJ = 1/2 possibilities, J = 3/2 and J = 1/2 for specified L,S
updated November 4, 2002

5.73 Lecture #26

1 11 1 1 = 3 2 22
1/ 2

26 - 3
1/ 2

11 2 10 + 3 22

11 1 2 1 2

(I always choose sign in front of smaller coefficient - arbitrary phase choice)

Nonlecture: Summary for 2 P 11 3 13 1 = 11 2 22 22 3 11 2 1 = 3 2 22

1/2

(coupled uncoupled)

11 1 10 + 3 22

1/2

(2 /3)1/2 1 1 11 = 1/2 2 2 (1 /3) u (1 /3)1/2 1 1 11 = 1/2 2 2 (2 /3) u

1 11 1 1 = 3 2 22

1/2

11 2 10 + 3 22

1/2

continue down to MJ = 1/2, MJ = 3/2 (or start at MJ = 3/2 and ladder up.

Now work in Coupled Representation for HSO + HZeeman HSO + HZeeman = n l s BzJ z BzS z
l

diagonal, easy
2 h J (J + 1) L( L + 1) S(S + 1) l s = recall 2 J z = hM J

off diagonal in J (cant be off diagonal in L,S,MJ - WHY?)

For 2P, there are 2 2 2 MJ = 1/2 and MJ = 1/2 blocks.

J Evaluate

L S

MJ

3 /2 1 1 /2 1 /2 S z 3 /2 1 1 /2 1 /2

by inserting the above transformation into the uncoupled basis set.

L ML S MS 2 1 1 0 1 / 2 1 / 2 S z 1 0 1 / 2 1 / 2 + 1 1 1 / 2 1 / 2 S z 1 1 1 / 2 1 / 2 = 3 3 2 1 1 1 1 h 3 2 + 3 2 = h 6

THERE ARE NO OFF-DIAGONAL MATRIX ELEMENTS OF HZeeman IN THE UNCOUPLED BASIS SET.
updated November 4, 2002

5.73 Lecture #26

2 3 / 2 1 1 / 2 1 / 2 Sz 1 / 2 1 1 / 2 1 / 2 = 3

1 /2

1 2 + 2 3

1 /2

1 2 = 2 3

26 - 4
1 /2

etc. for Sz (2P)

1 / 2 1/ 2 1 / 6 2 / 3 sym 1 / 6 1/ 2 1 / 6 +2 / 3 sym 1/6 1 / 2

Basis State
3/2 1 3/2 1 1/2 1 1/2 3 /2

Sz =h

1/2 1/2 1/2 1/2 3 / 2 1 1 / 2 1 / 2 1 / 2 1 1 / 2 1 / 2 3 / 2 1 1 / 2 3 / 2

(trace Sz = 0) 2 2 2s and 2 1 1s
If we had worked in uncoupled representation, HZeeman would have been fully diagonal but HSO 1) Zeeman produces E(ij0 ) 0 across which HSO has H(ij 0. would be 2 2 2's and 2 1 1's. H

If we work in coupled representation, HSO is diagonal but H Zeeman is off - diagonal. SO Zeeman (0) (1) has Hij 0. H produces Eij 0 across which H

WAR:

HSO tries to force system to coupled representation HZeeman tries to force system to uncoupled representation 2 Limiting Cases: weak and strong spin-orbit.

updated November 4, 2002

5.73 Lecture #26

extreme level patterns
Bz 0, n >> 0
l

26 - 5
Bz >> 0, n 0
l

COUPLED

(connect same MJ states)

UNCOUPLED

(M
2

,MS

P3 / 2

1 / 2 np

2Bz
Bz

(1, 1/2) (0, 1/2)

0
Bz
2

(1, 1/2), (1, 1/2) (0, 1/2) (1, 1/2)

P/2 1

np

2Bz

correlation diagram: noncrossing rule for MJ: why? states of same MJ do not cross

In coupled (strong spin - orbit) picture HZeeman is H(1) In uncoupled picture (strong field) HSO is H(1)
2

P matrices for HSO + HZeeman

coupled
MJ = 3 / 2 MJ = 1 / 2 / 2 2Bz
2 j = 3 / 2 / 2 Bz 3 sym j =1/ 2 2 Bz 3 1 Bz 3
1 /2

uncoupled
2 2Bz

ML,MS =

(1,1 / 2) 2 2 (0,1 / 2) sym B

1/2 z

CONTINUED AT BOTTOM OF PAGE 26-6

updated November 4, 2002

5.73 Lecture #26

Spin orbit only z 0 (J, MJ) MJ

26 - 6
Zeeman only (ML, MS) 2B 3/2 B 1/2 (0, 1/2) (1, 1/2)

(3/2, 3/2) (3/2, 1/2) (3/2, +1/2) (3/2, +3/2)

/2

1/2 +3/2 1/2

(1, 1/2) (1, 1/2)

(1/2, 1/2) (1/2, +1/2)

1/2

B (0, 1/2)

(1, 1/2)

There are only repulsions (shown by vertical arrows) between same-MJ pairs. coupled uncoupled
1/ 2

M J =1 / 2

( Bz )2 Bz B 9 = z 2 + 4 2 16 4 4

E = same as coupled (even though matrices are different)

M J = 1 / 2 j = 3 / 2 / 2 + 2Bz 3 sym j =1/ 2

2 1/ 2 Bz 3 1 + Bz 3

B z 2 1/2 sym 2

M J = 1/2

B 9 1 = + z 2 + Bz 4 2 16 4
2 + 2Bz

( )

B z 4

1 /2

E = same as coupled

E MJ = 3 / 2

2 + 2Bz
updated November 4, 2002

5.73 Lecture #26

Energy eigenvalues come out to be identical (as they must) in both representations.

26 - 7

Coupled picture is good for 2nd order perturbation theory in the weak field (|| << nl) limit.

Zeeman splitting (and tuning rates,

coupled picture H
( 0)

dE , as B z is varied) dB z

first order correction

second order correction H Zeeman ij

2

=H

SO

E (1) = H Zee
2z MJ = 3/2 +2/3 z MJ = 1/2

( 2)

E (i 0 ) E (j0 )

4 27

( Bz )2

}
MJ = 1/2 2/3 z MJ = 3/2 2z

4 + 27

( Bz )2

MJ = 0
matrix elements

1/3 z MJ = 1/2
4 27

( Bz )2

}
MJ = 1/2 1/3 z

4 27

( Bz )2

updated November 4, 2002

5.73 Lecture #26

r r Uncoupled picture is good for strong field limit where Bz causes L + S to uncouple r from J and into the laboratory (Paschen - Back limit).
ZERO-ORDER (ML, MS) (1, 1/2) FIRST-ORDER SECOND-ORDER MJ (3/2) +/2

26 - 8

2z

(1/2) z (0,1/2) + /2z

2

(1,1/2) (1,1/2) /2

(1/2) + /2z
2

2/2z (1/2) z (0,1/2) 2/2z (1/2) 2z (1,1/2) +/2 (3/2)

Regular Zeeman Pattern.

Small distortions from equal intervals, from which may be determined.

Extra distortions. Repulsions between SameMJ components

This is a regular Zeeman pattern, but with small distortions (shown as vertical arrors on expanded scale) from equal intervals, from which may be determined.
updated November 4, 2002

(3/2,3/2) / 2 2 0 2 / 2 0 (3/2,1/2) 3 sym (1/2,1/2) h (3/2,-1/2) (1/2,1/2) (3/2,3/2)

2 0 3 1 0 3 2 / 2 + 0 3 sym

1/ 2

21/ 2 0 3 1 + 0 3

/ 2 + 2 0

Bz 0

(ML,MS)

UNCOUPLED REPRESENTATION

(1,1/2) (1,1/2) (0,1/2)(0,-1/2)(1,1/2) (1,1/2) (1,1/2) / 2 2 0 1/ 2 (1,1/2) / 2 2 sym 0 (0,1/2) 1/ 2 h 0 2 (0,-1/2) sym 2 (1,1/2) / 2 + 2 0 (1,1/2)

(3/2,3/2)

/2-20
2 4 2 0 0 + 3 27

(1,1/2)

/220
2 2 0

(3/2,1/2) (1/2,1/2) (3/2,1/2) (1/2,1/2) (3/2,3/2)

/2

2 8 2B z + 3 27

(1,1/2)

2 +

2 2
2 B z

1 4 2 0 0 3 27
2 4 2 0 0 + 3 27

1 8 2B z 3 27
2 8 2B z + 3 27 1 8 2B z 3 27

(0,1/2)

2 2 0
2 2 0 2 2 0

+
+
+

2 2
2 B z

/2+

(0,1/2)

+0 +

2 2
2 B z

1 4 2 0 0 3 27

(1,1/2)

2 2
2 B z

/ 2 + 2 0

+2

(1,1/2)

/2+20

+2

J = 3/2

J = 1/2

Bz

Bz

Note that every eigenstate has a different Zeeman tuning rate. The gross energy level pattern and the Zeeman tuning rates tell which limiting case one is near and which state is which. The details of the nonlinear Zeeman tuning of one assigned level determine and .

5.73 Lecture #27 Wigner-Eckart Theorem

CTDL, pages 999 - 1085, esp. 1048-1053 Last lecture on 1e Angular Part Next: 2 lectures on 1e radial part Many-e problems What do we know about 1 particle angular momentum?
1. JM Basis set

27 - 1

[J , J ] = i
i j

ijk

Jk

definition all matrix elements in JM J basis set.

2.

J = J1 + J 2 Coupling of 2 angular momenta coupled uncoupled basis sets transformation via J = L + S plus orthogonality. Also more general methods. HSO + H Zeeman example * easy vs. hard basis sets * limiting cases, correlation diagram * pert. theory patterns at both limits plus distortion

TODAY: 1. Define Scalar, Vector, Tensor Operators via Commutation Rules. Classification of an operator tells us how it transforms under coordinate rotation. 2. Statement of the Wigner-Eckart Theorem 3. Derive some matrix elements from Commutation Rules Scalar Vector S V J = M = 0, M independent J = 0,1, M = 0, 1, explicit M dependences of matrix elements

These commutation rule derivations of matrix elements are tedious. There is a more direct but abstract derivation via rotation matrices. The goal here is to learn how to use 3-j coefficients.

updated November 4, 2002

5.73 Lecture #27

27 - 2

Classification of Operators via Commutation Rules with CLASSIFYING ANGULAR MOMENTUM Like c o m p o n e n t s ( ) =0 scalar "constant" 0 J = 0 = 0z v e c t o r 3 components 1 J = 1
+1 21/ 2 ( x + iy) 1 +2 1/ 2 ( x iy)

tensor

(2 + 1) components [ is "rank"]

2nd 3rd

2 3

2 3

+2, , 2

Spherical Tensor Components [CTDL, page 1089 #8] Definition:

[J , T ] = [J , T ] =
z
( )

h h

[( + 1) ( 1)]1/ 2 T(1)

( )

( T )

( This classification is useful for matrix elements of T ) in JM basis set.

Example: J = L + S

1.

[L , S] = 0

r r

common sense ?2.

(vector analysis)

3. 4. [Because L S is composed of products of components of two vectors, it could act as a second rank tensor. But it does not !]

L & S act as scalar operators with respect to each other. r r L and S act as vectors wrt J r r L S acts as scalar wrt J r r L S gives components of a vector wrt J .

[Nonlecture: given 1 and 2, prove 3] Once operators are classified (classifications of same operator are different wrt different angular momenta), Wigner-Eckart Theorem provides angular factor of all matrix elements in any basis set!
( J NJM T ) NJM = AJ[ M ] M,M+ 14 2 ) NJ NJ4 ( 44 T M 3 123 reduced matrix element

specifies everything else

vector coupling coefficient

no M , M, no

redundantusually omitted

updated November 4, 2002

5.73 Lecture #27

* * triangle rule

rank of tensor like an angular momentum

27 - 3

( J J J + : selection rule for T) , J = , ( 1), 0.

reduced matrix element contains all radial dependence when there is no radial factor in the operator, then the J,J dependence can often be evaluated as well vector coupling coefficients are tabulated lots of convenient symmetry properties: A.R. Edmonds, Ang. Mom. in Q.M., Princeton Univ. Press (1974).

Nonlecture: L & S act as vectors wrt J but scalars wrt each other

[L,S] = 0 scalars wrt each other [J, L] = [L + S, L] = [L, L] vector wrt. J if L is an angular momentum

1/ 2 1/ 2 1/ 2 J z, 2 [L x + iL y ] = 2 ih[L y iL x ] = 2 h[L x + iL y ] ( = + hT+1)[L ] 1

( T+1)[ L] = 2 1/ 2 [L x + iL y ] 1

( components of L satisfy the T1) definition

This notation means: construct

(1) an operator classified as T+1 r out of components of L.

[J ,T
z

(1) +1

etc.

[J, S] = [S, S]
J=L+S

( [L ]] = h( +1)T+11)[L ]

S is vector wrt J

Show that LS acts as scalar wrt J

[J , L S ] = [J

, LxS x + LyS y + LzSz i J y , LxS x + LyS y + LzS z

= four terms

] [

[J , L S] = [L x , L x S x + L yS y + L zSz ] + [S x , L x S x + L yS y + L zSz ]
= ih(L z S y L yS z ) + ih(L yS z L z S y ) iih( L z S x L x S z ) iih( L x S z + L z S x )] =0

i[L y , L x S x + L yS y + L z S z ] i[S y , L x S x + L yS y + L z S z ]

[J z , L S] = [L z , L x S x + L yS y + L zSz ] + [Sz , L x S x + L yS y + L zSz ]

= ih(L yS x L x S y ) + ih(L x S y L yS x ) =0

( L S acts as T00 )

This notation means: construct an operator classified as T0(0) out of the

( T [ A, B] = (1) T [ A]T )[B] components of A k ( 0) 0 k ( ) k k =

and B.
updated November 4, 2002

5.73 Lecture #27

27 - 4

Vector Coupling Coefficients all serve Clebsch - Gordan Coefficients same function 3 - J coefficients all related to what you already know how to obtain by ladders and orthogonality for

JJ1 J2 M =

M 2 = M M1 M2 = J 2

+ J2

J1 M1 J2 M2 J1 M1 J2 M2 JJ1 J2 M 144 2444 4 3

v.c. coefficient

completeness

p. 46 Edmonds (1974) general formula

J J2 3 J: 1 M1 M 2
1 J3 J1 J 2 M 3 (2J 3 + 1) 2 (J1M1J 2M 2 J1J 2 J 3 M 3 ) = (1) M 3

Constraint: M1 + M2 + M3 = 0 This constraint is imposed in (|) notation but not

in | notation.]

W-E Theorem is an extension of V-C idea because we think of operators as like angular momenta and we couple them to angular momenta to make new angular momentum eigenstates. What is so great about W-E Theorem? vast reduction of independent matrix elements e.g. J = 10, = 1 (vector operator) possible values of J limited to 9, 10, 11 by triangle rule
( J M T1) JM

Total # of M. E. J = 9 (29+1)(210+1) 10 (210+1)(210+1) 11 (211+1)(210+1)

399 441 483 1323

#R.M.E. 1 1 1 3

9 T(1) 10 10 T(1) 10

11 T(1) 10

updated November 4, 2002

5.73 Lecture #27

CTD-L, pages 1048-1053 Outline proof of various parts of W-E Theorem Scalar Operators S [Ji,S] = 0
1. 2. 3.

27 - 5

Definition (for all i)

] M = 0 selection rule from [J , S] = 0 M - independence from [J , S] = 0

J = 0 selection rule from J 2 , S = 0
z

1.

show J = 0:

J M S JM = 0 if J J

[J , S] = 0
2

0 = J M ( J2S SJ2 ) JM =

[J (J + 1) J (J + 1)] J M S JM

direction of operation by J2

either J = J or J M S JM = 0

(only J = 0 matrix elements of S can be nonzero)

JM S JM = 0 if M M

2.

show M = 0:

[J , S] = 0 0 = JM ( J S SJ ) JM
z z z

(M M )

JM S JM

either M = M or JM S JM = 0
3 show M - Independence of matrix elements

direction of operation by S

[J , S] = 0

uses J = M = 0 for S

0 = JM J S SJ JM = sJM JM J JM sJM JM J JM = sJM sJM JM J JM

we already know that S is diagonal in M.

[Should skip pages 27-6,7,8 and go directly to recursion relationship on page 27-10.]

updated November 4, 2002

5.73 Lecture #27

Thus either sJM = sJM or JM J JM = 0 Thus sJM is independent of M Putting all results for S together J M S JM = J J M M J S J
Vector Operators V

27 - 6

[J , V ] = i
i j

ijkVk
k

1. 2. 3.

r M selection rules from Jz , V

J selection rules from J2 , J2 , V

[ [

]r

]]

M - dependence of matrix elements of V from double commutator

1.

M selection rules
a.

[J , V ] = 0 0 = J M ( J V
z z z

Vz Jz JM =

(M M )

J M Vz JM

either M = M or ME = 0

b.

[J , V ] = [J , V ] i[J , V ]
z z x z y

= ih V y i Vx

))

J M Jz V V Jz JM
h h

= hV = h J M V JM

(M M )

J M V JM

= h J M V JM =0

( M M m 1)

J M V JM ME =0

M = M 1 or

updated November 4, 2002

5.73 Lecture #27

Thus we have selection rules: Vz acts like Jz on M V acts like J on M

27 - 7

2.

M selection rules
need to use a result that requires lengthy derivation 2 2 2 2 2 J , J , V = 2h J V 2( J V )J + VJ see proof in Condon and Shortley, pages 59 - 60

[ [

]]

Take J M

JM Matrix elements of both sides of above Eq.

LHS = J M J2 ( J2V ) J2VJ2 J2VJ2 + VJ2J2 JM =

h
4

r RHS = 2h 4 J ( J + 1) + J ( J + 1) J M V JM 4h 4 J M ( J V )J JM

r [(J (J + 1)) 2J (J + 1)J (J + 1) + J (J + 1) ] J M V JM

2 2 2

scalar

J M ( J V )J JM =

J M

J M ( J V ) J M J M J JM
J = J , M = M

= J J V J J M J JM 14 244 4 3
J =J

= J J V J JM J JM J J
two cases for overall matrix element A. J J B. J = J

updated November 4, 2002

5.73 Lecture #27

27 - 8

A.

J J
r RHS = 2h 4 J ( J + 1) + J ( J + 1) J M V JM LHS =
h
4

[J (J + 1)
2

2J ( J + 1)J ( J + 1) + J 2 ( J + 1)

] J M V JM

2 2 0 = LHS RHS = algebra = h 4 J M V JM [( J J ) 1][( J + J + 1) 1]

ME = 0 unless J = J 1 or J = J

r J = 1 selection rule for V

(J = J is impossible except for J = J = 0, but this violates J J assumption)

B.

J = J
LHS = 0 0 = RHS = 4h 2

J ( J + 1) JM V JM J J V J JM J JM

J J V J r r JM V JM = JM J JM 2 ( h J J + ) 14241 4 4 3
C 0 (J )

A WONDERFUL AND MEMORABLE RESULT. It says that all J = 0 matrix r r elements of V are corresponding matrix element of J! A simplified form of W - E Theorem for vector operators.

updated November 4, 2002

5.73 Lecture #27

27 - 9

Lots of (NONLECTURE) algebra needed to generate all J = 1 matrix r elements of V. SUMMARY OF C.R. RESULTS: Wigner-Eckart Theorem for Vector Operator

J = 0 JM Vz JM = Co ( J)M

special case: These are exactly the same form as corresponding 1/ 2 JM 1 V JM = Co ( J)[ J( J + 1) M( M 1)] matrix element of Ji. J + 1M 1 V JM = mC + ( J)[( J M + 2)( J M + 1)] J + 1M Vz JM = + C + ( J)[( J + M + 1)( J M + 1)]
1/ 2 1/ 2

J = +1

J = 1

J 1M 1 V JM = C( J)[( J m M)( J M + 1)] J 1M Vz JM = + C( J)[( J M)( J + M)]

1/ 2

1/ 2

only Co(J), C+(J), C(J) : 3 unknown J-dependent constants for each J. NONLECTURE (to end of notes). Example of how recursion relationships (reduced matrix elements) are derived for each possible J. r J = 1 matrix elements of V
M = +1 using J+ , V+ = 0 M selection rule for V+ is M = +1 M selection rule for J+ V+ is M = +2
CR = 0 = J + 1M + 1 J+ V+ V+ J+ JM 1 0 = J + 1M + 1 J+ J + 1M J + 1M V+ JM 1 J + 1M + 1 V+ JM JM J+ JM 1

(arrow denotes J+ operates to right)

expand using completeness

(J+ operates to left)

J + 1M V+ JM 1
h

[(J + M)(J M + 1)]

JM J + JM 1

=
1/ 2

J + 1M + 1 V+ JM
h

[(J + M + 2)(J M + 1)]1/ 2

J + 1M + 1 J + J + 1M

The matrix elements in the denominator are to be replaced by their values, and a common factor is cancelled
updated November 4, 2002

5.73 Lecture #27

multiply both sides by [ J + M + 1]
1 / 2

27 - 10
to display symmetry

J + 1M V+ JM 1 J + 1M + 1 V+ JM = C+ ( J ) ( J + M )1/ 2 ( J + M + 1)1/ 2 ( J + M + 1)1/ 2 ( J + M + 2)1/ 2

M M + 1 recursion relationship

ratio is independent of M

C+ ( J ) J + 1 V J J + 1M + 1 V+ JM = C+ [( J + M + 1)( J + M + 2)]
1/ 2

sign chosen so that Vz matrix elements will be +C+(J)

Remaining to do for J = 1 matrix elements J , V+ = 2hVz gives Vz matrix element when we take M = 0 A. matrix element of both sides J , Vz = hV gives V matrix element when we take M = 1 B. matrix element of both sides

[ [

A.

[J , V ] = 2
+

Vz

M = 0 selection rule for both sides

RHS = 2h J + 1M Vz JM LHS = J + 1M J V+ V+ J JM = J + 1M J J + 1M + 1 J + 1M + 1 V+ JM J + 1M V+ JM 1 JM 1 J JM =
h

[(J + 1)(J + 2) M (M + 1)] [ ]

1/ 2

1/ 2

J + 1M + 1 V+ JM

h J ( J + 1) M ( M 1)

J + 1M V+ JM 1

rearrange this and use V+ recursion rule from above

LHS = hC+ ( J )[( J + M + 1)( J M + 1)]

[( J + M ) ( J + M + 2)]1 1/ 2 = 2 hC+ ( J )[( J + M + 1)( J M + 1)]

1/ 2

updated November 4, 2002

5.73 Lecture #27

RHS = LHS

27 - 11
1/ 2

J + 1M Vz JM = C+ ( J )[( J + M + 1)( J M + 1)]

B.

[J , V ] =
z

take J + 1M 1L JM

RHS = h J + 1M 1 V JM LHS = J + 1M 1 J J + 1M J + 1M Vz JM J + 1M 1 Vz JM 1 JM 1 J JM = hC+ ( J )[( J M + 2)( J M + 1)]

1/ 2
1/ 2

J + 1M 1 V JM = + C+ ( J )[( J M + 2)( J M + 1)]

VERY COMPLICATED AND TEDIOUS ALGEBRA

updated November 4, 2002

5.73 Lecture #28

Hydrogen Radial Wavefunctions

28 - 1

The Hydrogen atom is special because it has electronic states and properties that scale with n and l in a simple and global way. This is structure that is more than a collection of unrelated facts. H serves as our model for electronic structure of manyelectron atoms, molecules, and possibly solids. By showing how E, r (size and shapes), nl|r|nl (general matrix element) scale with n and l, it tells us the kind of behavior to look for in more complex systems. * * as a perturbation on H (quantum defects) as a hint of relationships useful for extrapolation, assignment, for recognizing when something behaves differently from naive expectations.

TODAY

1. 2. 3. 4.

Simplified Radial Equation Boundary conditions at r 0 and r qualitative features of Rnl(r) n-scaling of r 5. mathematical form of Rnl(r) 6. regular and irregular Coulomb functions

updated September 19,

5.73 Lecture #28

For any central force problem

28 - 2

p2 l2 H= r + 2 + V (r ) 2 2r

We know that H , l 2 , l z commute, so spherical harmonics, Ylm ( , ) , are eigenfunctions of H with eigenvalues h 2 l( l + 1). ( r, , ) = R( r )Ylm ( , ) trial form for separation of p2 l2 + V ( r ) Ylm ( , ) R ( r ) = E H = r + 2 2 2r Operate on the Ylm ( , ) angular wavefunction and move it through to left. p 2 h 2 l ( l + 1) = Y m , r + + V ( r ) R ( r ) = E H ) l ( 2 1442 2444 2r4 3 Vl ( r )

so we can take Y (, ) out of the Schrdinger Equation and we are left with a 1-D radial equation where the only trace of the angular part is the l -dependence of V (r), the effective potential energy function.
m
l

Since the differential equation depends on l, R(r) must also depend on l, thus Rn (r) is the radial part of , and it will generally be an explicit function of two quantum numbers, n and l.
l

Usually n specifies the number of radial nodes and l the number of angular nodes, but a special numbering convention for Hydrogen (and hydrogenic ions) causes a slight distortion of this rule.

updated September 19,

5.73 Lecture #28

28 - 3

The radial equation, when the explicit differential operator form of Pr2 is derived and inserted, has the form h2 1 d 2 2 r dr 2 Tr r + h2 l(l + 1) + V( r ) 2 2r Rnl ( r ) = E nl Rnl ( r ) Vl
1 u ( r) r nl

It is customary to simplify this equation by replacing Rnl ( r ) by R nl ( r ) =

* * *
insert 1 r

1 u ( r) r nl
equation looks simpler volume element looks simpler behavior as r 0 seems more familiar
u n (r ) in place of Rn (r ) and then multiply through on left by r
l l

h 2 d2 + 2 2 dr

h l l

+ 1)

2r 2

+ V (r ) E n u n (r ) = 0
l l

looks like ordinary 1-D Schrdinger Equation. Boundary condition:

u n (r ) 0
l

as

r0

WHY? Because for all l > 0, Vl(0) .

exactly as if V (r ) = r 0, but of course r < 0 is impossible, so we had better be careful about behavior of u n (r ) and Rn (r ) as r 0
l l

updated September 19,

5.73 Lecture #28

28 - 4

Note also that d 3r = r 2 sin drdd

* * Rn Rn r 2 dr = un (r )un (r )dr
l l l l

r2 cancelled. So volume element looks just as in 1-D problem

Return to special situation as r 0. Why do we care? It turns out that s-orbitals have Rns(0) 0 and that in ESR one measures Fermi-contact hyperfine structure which is the spin-density at specific nuclei. It is a direct measure of the ns atomic orbital character in each molecular orbital! CTDL, p. 781 What is the worst possible divergence of Rnl(r) as r 0? For r 0, Rnl(r) will be dominated by rs where |s| is as small as possible. This is the most strongly divergent part of Rnl(r), which is all we need to be concerned with as r 0. Let Rnl ~ Crs, where this is a good approximation at r 0. Plug this definition into Schrdinger Equation

d2 dr
2

rRnl ( r ) =
h

d2 dr
2 2

Cr s +1 = (s + 1)(s)Cr s 1

Tr =

1 d 2 r dr 2
h

HRnl ( r ) =

C(s + 1)(s) r 1 r

s 2

h l l

( + 1)
2

Cr s 2 + V( r )Cr s E nl Cr s = 0

*

if V( r )

As r 0 V( r ) rarely diverges more rapidly than 1 / r , thus V(r)C r s gives r s-1 .

Then, in the limit r 0, the coefficients of the rs2 term (i.e. the most rapidly divergent term) must be = 0

(s + 1)s + l(l + 1) = 0
*

This excludes the stronger divergence of the centrifugal barrier term in Vl(r).
updated September 19,

5.73 Lecture #28

28 - 5
or s = (l + 1)
l l

satisfied if s =

s(s + 1) = ( l 1)( l 1 + 1) = ( l + 1)( l ) =

In other words R nl ( r ) r l
well behaved at r 0

verify second possibility:

+ 1)
as r 0

OR (if s = (l + 1))

1 r l +1

disaster even if l = 0

Actually both of these possibilities satisfy the differential equation for 1 V( r ) = (known as the Coulomb or H atom Hamiltonian), but r the one that diverges as r 0 cannot satisfy the r 0 boundary condition for the H atom.
** Regular and Irregular Coulomb wavefunctions we will return to these later in the context of Quantum Defect Theory. So for now we insist that
Rn (r ) r
l l

as

r0

Rns (0) 0 special situation for R Rnl>0(0) = 0 unl (0) = 0 for all l

ns(r)

(no special case for uns(r))

updated September 19,

5.73 Lecture #28

28 - 6

For Hydrogen V (r ) = +
l

h l l

( + 1) e 2 2r 2 r

q2 e 4
2

=
V (r )
l

mem p me + m p

me

0 0

V0 (r )

V (r )
l

What do we know from our study of 1-D problems? WKB

envelope p 1/ 2 ~

h = 2 2p(r )

ASK QUESTIONS shape of un (r )

l

# of nodes, placement of nodes, degeneracy, behavior at inner and outer turning points, location of inner and outer turning points

1st lobe, last lobe

h of action acquired 2 (associated with tunneling into nonclassical region)? inner vs. outer part of u nl ( r ) where is the extra
recall

r > (E)

r < (E)

p( r )dr =

h 2

( n + 1 /2)

updated September 19,

5.73 Lecture #28

28 - 7

Find that

n At turning point Vl ( r ) = E nl =
h l l

E nl =
2

e4 me 2 h2
h l l
2

( + 1)

2 2r n2 solve for r as function of n and l

e r

= n2

( + 1) r e 2 2 2r2

r = a0 n 2 n(n 2 l(l + 1))

1/ 2

2r2 = h2l(l + 1) r e 2 2 2 n Use Quadratic formula to find r(n)

h
2

l(l + 1) 1/ 2 2 = a0 n 1 1 n2

a0 =

Bohr radius e 2m e when l << n, where are r+ and r ?

Use this equation for the turning points to construct qualitatively correct cartoons of Rnl(r) in crucial regions. surprising systematic degeneracy
etc. 3s 2s 3p 2p 3d

En

1s

Because of pattern, we use n to label degenerate groups n2 hence n is not # of radial nodes. En =
l

updated September 19,

5.73 Lecture #28

orbital 1s 2s 2p 3s 3p 3d # of radial nodes 0 1 0 2 1 0
# radial nodes = (n 1) l # angular nodal surfaces total # nodes n 1
l

28 - 8

(because it is lowest solution to

= 1 equation)

n 1 2 3 n

degeneracy 1 1 + (2 l + 1) = 4 1+3+5=9

n2

n - scaling of r
two limits: determined near

<0
inner turning point

vs.

>0
outer turning point
Bohr model rn = a0 n 2
l

~ n 3
(see argument on next page)

r a0 n 2

Expectation values of r vs. transition moments and off--diagonal matrix elements of r. Stationary phase.

updated September 19,

5.73 Lecture #28

inner region

28 - 9

3% of IP
109737cm 1 = 3000cm 1 36 36

n=6

En =

n=1

p2 T= > IP in the inner region. 2

variation of T n = 6 to n = < 3% variation of p < 1.5%
deBroglie ~ h independent of n ! Because p is large and p fractional change of p vs. n is negligible. location of innermost node

e comes into core region fast and leaves fast t same for all n fraction of time inside core?
time inside 2( v ) = h one period E n E n +1
1

E n = n2

2 h E n + 2 E n 2 = 3 /v p n 2 p m 4 m = 2 3 h pn 2 n3 n-independent probability of finding e inside core n3 !

updated September 19,

5.73 Lecture #28

fraction of time inside n 3 ~

28 - 10

amplitude of n n 3/ 2 inside core region

l

Basis of all Rydberg scaling 1st node does not shift with n inner lobe Astonishingly important! amplitude in first lobe scales as n 3/2 all n , n matrix elements of r where < 0 scale as ( nn ) Some matrix elements scale this way even when > 0.
3/2

McQuarrie, page 223 (n l 1)! 2 Rn (r ) = 3 2n (n + l )! na0

1/ 2
l l

+ 3/ 2

2r r e r na0 L2 ++1 n 1 na0

l l

normalization exponential 0 as r Regular and Irregular Coulomb functions (E 0) un (r) rR(r) r0

l

associated Laguerre functions (polynominals)

regular

f(E,l,r)

r +1
l

u(,l,r)sin v(,l,r)ei, which is an increasing exponential except when is a positive integer. Need some other way to satisfy r boundary condition when is not an integer

irregular

g(E,l,r)

r( )
l

u(,l,r)cos + v(,l,r)ei(+1/2), which blows up. * u(,l,r) is an increasing exponential as r * v(,l,r) is a decreasing exponential as r (see Gallagher, page 16)
updated September 19,

5.73 Lecture #28

T.F. Gallagher, Rydberg Atoms, page 25
r 1 2 [3n l(l + 1)] 2 n2 [5n2 + 1 3l(l + 1)] 2 1 n2 1 n (l + 1 / 2 )
3

28 - 11

r2 1r 1 r2

1 r3

1 n (l + 1)(l + 1 / 2)l
3

1 r4

3n 2 l(l + 1) 2 n 5 (l + 3 / 2)(l + 1)(l + 1 / 2)l(l 1 / 2) 35n 4 5n 2 [6l(l + 1) 5] + 3(l + 2)(l + 1)l(l 1) 8n 7 (l + 5 / 2)(l + 2)(l + 3 / 2)(l + 1)(l + 1 / 2)l(l 1 / 2)(l 1)(l 3 / 2)

1 r6

Note!

all r

< 1 >0

scale as n 3! scale as n 2 !

updated September 19,

5.73 Lecture #29

Begin Many-e Atoms: Quantum Defect Theory

29 - 1

See MQDT Primer by Stephen Ross, pages 73-110 in Half Collision Resonance Phenomena in Molecules (AIP Conf. Proc. #225, M. Garci-Sucre, G. Raseev, and S.C. Ross) 1991.

Last Time:

e2 + * turning points of V (r ) = r
l

h l l

( + 1) r2

l(l + 1) 1/ 2 l(l + 1) 2 r (n, l ) = a0 n 1 1 for l << n a0 n 1 1 m n2 2n2

2

envelope
l l

un (r ) pn (r )1/ 2

* n (r)rRn (r) dominated by small lobe (n-independent nodal position) at inner turning point, amplitude scales as n3/2, and large lobe at outer
l l

turnin point (essentially all of the probability.

* *

En = IP
l

nodes:

n2 n l 1 radial nodes l angular nodes n 1 total nodes 2 gives spacing between radial nodes
r < 1 +1 = 1 n 3 n 2 n 2 (see below) (H - atom energy levels)

expectation value scaling

geometric mean of expectation values of r for off - diagonal matrix elements r

nl

n2
1/2

( r+ nl )

( r+ n l )

1/2

] [(n )
?)

2 1/2

( n 2 )1/2 ]

= ( nn ) n 2

( when is r n r

updated September 19,

5.73 Lecture #29

TODAY

29 - 2

1. Many-e atom treated as core plus outer e that sees shielded core as Z(r). 2.
l

-dependent energy shifts n-independent quantum defects E nl = IP

3. energy shifts are actually phase shifts in unl(r) relative to unl(r) for H-atom 4. Rigorous QDT

( n nl )2

A. regular and irregular Coulomb functions f,g satisfy Hydrogen-like Schr. Eq. OUTSIDE core B. Boundary conditions at r
noninteger values of = require mixture of f and g E nl find = n l satisfies r boundary condition number of members in series of with integer spacings, constant quantum defect s
1/2

C. l is a phase shift repeated patterns in each integer region of

D. Multi-channel QDT matrices

e colliding with core can also transfer energy and angular momentum to core-e * channels rather than eigenstates * focus on dynamics, but in a black box way. Dynamics happens within a restricted region of space. This region of space is always sampled, regardless of E, in the same way. Everything is determined by the boundary conditions for the outgoing wave. SCATTERING THEORYrather than EFFECTIVE HAMILTONIAN MODEL. The goal here is to extract from a complicated many-body problem some regular features that will help in assigning and modeling experimental data.

updated September 19,

5.73 Lecture #29

1. Many-e Atom r0

29 - 3

e
outer electron (valence, Rydberg) Z

+Z

outside core e sees Z = +1 inside core e sees Z(r)

1 rcore

Shell Structure
ensures n3 rather than n2

2.

-dependent sampling of core high l : see Z (r ) ~ +1

low l : see Z (r ) = Z eff >> 1
l

as r 0, H(1) diverges faster than e 2 r because Z(0) > 1.

energy stabilization

e2 r

E (n1) = nl H(1) nl
l

H=H
H(1) r
l l

( 0)

[ Z(r ) 1]e 2

+H

( ) 1

c n3
l

c En = 2 3 n n (n
so far we have focussed on Enl 3. What does Z(r) do to unl(r)?

2 ) expand in Taylor series

2
l

<< n

call this , effective principal qn

* outside core sees same Vl ( r ) as H * must be same as u nl ( r ) for H except for phase shift inward (why inward?) * all the unique stuff is inside core causes the phase shift. - nodal structure inside core is invariant wrt n or E

updated September 19,

5.73 Lecture #29

Mulliken: ontology recapitulates phylogeny intra-core nodal structure is n-independent nodal structure encodes all enucleus dynamics! 4. Do all of this more rigorously: QDT * * * *

29 - 4

regular Rydberg series, one for each l These are what n-scaling of inner lobe amplitude and of all matrix elements we will obtain. large quantum defects for small l entire Rydberg series and associated ionization continuum (e ejected in l-partial wave) is a single entity
follow Ross but not using atomic units

redefine 0 of E

En =

n2

n = 1, 2, for H
1 / 2

E n = n

E u and use (effective principal quantum number) rather than E as a label for nl ( r ) generalize to noninteger n for non - hydrogen:

1/2

Schrdinger Equation for H (the Coulomb Equation)

h d 2 2 dr
2 2

l l

+ 1) e E u ( , r ) = 0 2 r r
2
l

as variable rather than as q.n.

2 well known solutions: 2nd order differential equation - two linearly independent solutions (at each l,)

E=

fl ( , r ) 0 as r 0

" regular" " irregular"

g l ( , r ) as r 0

of no use for Hydrogen, but it turns out that we need both f and g to satisfy r boundary condition.

updated September 19,

5.73 Lecture #29

for H, we have no use for gl(,r) because it cannot satisfy boundary conditions as r 0

29 - 5

A. For many-e atoms, beyond some critical r0, Schr. Eq. is identical to that of H the only difference is that we must treat the r 0 boundary condition differently
Universal boundary conditions are r , u l ( , r ) 0 for E < 0, r , asymptotic forms for f and g are

f ( , r ) C(r ) sin[ ( l )]e r /

l

g ( , r ) C(r ) cos[ ( l )]e r /

l

C(r ) 0 as r but C(r )e r / as r , so the only way to satisfy the r boundary condition for a pure u ( , r ) = f ( , r ) is for ( l ) = integer
l l

B. But we might want to use a mixture of fl and gl to deal with non-integer ( l), as we will need for many-electron atoms.

H Na Z = +11 core

Na ul(,r) emerges from core with extra phase sucking in of hydrogenic function

* invariance of intra-core nodal structure amount of phase shift should be independent of n. [We expect this to be true.]
u ( , r) = f ( , r ) (1 2 )
l l

1/ 2

g ( , r)
l

**

* mixing of 2 types of function is required in order to have noninteger , yet still satisfy ul(,r)0 as r boundary condition

updated September 19,

5.73 Lecture #29

TRICK:

29 - 6
plug this into * * equation = sin ( l ) on page 29 - 5

cos( l ) (1 2 )
l l

1/2

= u ( , r) (CORE) = f ( , r) cos g ( , r) sin 14243

l l l

)]

other e

plug in asymptoptic forms for f,g

sin[ ( l )] cos + cos[ ( l )] sin

f ( , r) g (,r) { } 0 as r is required. How?
l

)}C (r )e
l

r/

} = 0 : sin[ ( l)] cos( ) = cos[ ( l)]sin( )

sin( sin[ ( l )] = cos[ ( l )] cos(
l
l

tan[ ( l )] = tan( let =

) ) )
l

tan = tan( ) = tan( + n)

integer
l

constraint on . What are all of the values of which are consistent with this constraint?

tan[ ( l )] = tan( + n)
thus + n = ( l)
but = ( n + l )
l

= n ( l) = ( n + l) n = n + l, l = n

n + l n

integer

= n l

is smaller than integer n by constant term l.

Implies existence of infinite series of values of for which 0 as r .

updated September 19,

5.73 Lecture #29

specifying one -independent value of l!

29 - 7

Get this infinite series of s, increasing in steps of 1, simply by

All of the -dependence (E-dependence) of l(,r) is explicitly built into fl(,r) and gl(,r). l describes the relative amounts of fl and gl in . This f,g mixing is determined when the e leaves the core with the precise phase shift so that 0 at r .
C. How can we show that l is a phase shift?

The asymptotic form of is sin[ ( l )] cos( ) + cos[ ( l )]sin(

l

)}C(r )er /

sin ( l ) +
l

but f ( , r ) sin[ ( l )]C(r )e r /

so this modified function is identical to the regular [i.e. fl(,r)] Coulomb function but with a l phase shift.

{ [

use double angle formula sin A cos B + sin B cos A = sin( A + B)

l

]}

C(r )e r /

If l > 0, this corresponds to an advance of the phase of ul(,r) relative to that for H. As expected, is sucked into core by an amount l [+ an arbitrary number of 2s] by the Z(r) core. l is the phase shift that occurs inside the core. It is the boundary condition at r = 0 shifted out to r = r0. On the other hand, the r b.c. is satisfied by = n l where n is integer. the quantum defect!

updated September 19,

5.73 Lecture #29

29 - 8

n+1

exact replica of lower-n pattern

s >> p > d > f 0

everything is repeated in each integer region of , not E, is the way to look at Rydberg patterns Finding the way to see a pattern is ALWAYS the route to both assignment and insight
l decreases as l increases because of the expected behavior of Z eff ( r ) as sampled in the presence of a centrifugal barrier
l l

( + 1)
2r 2

updated September 19,

5.73 Lecture #29

D. inter-series interactions? Suppose you have B 2s22p1
B + (2 s 2 p)

29 - 9

autoionization
B + (2 s 2 )

perturbation

B 2s2 2 p 2P

Separate series converging to 2 series limits perturbations autoionization

updated September 19,

5.73 Lecture #29

Described by a Multichannel Quantum Defect Theory
Replace s, p , d etc.

29 - 10

by 3 3 matrices, one for each symmetry

[2s ( S) ] [2s2 p( P ) [2s2 p( P )

2 1 1l 3 1

2l

3l

] 1]
1

l l l

2 2

more complicated many-electron coupling problem subject of next few lectures.

Overall symmetry: H is totally symmetric. off-diagonal elements describe inter-channel interactions (exchange of angular momentum between Rydberg e and core es.) describe what happens in a collision of e with ion-core. Does it change the state of the ion? Does it change the kinetic energy and/or angular momentum of the e? Unified picture of scattering at negative E (bound states) and at positive E.

Next few lectures: states of many-electron atoms How to calculate matrix elements of many-electron (many Fermion) systems.

updated September 19,

5.73 Lecture #30

30 - 1

Matrix Elements of Many-Electron Wavefunctions

Last time:
= noninteger principal quantum number En , f ( , r) solutions to Schrd. Eq. outside sphere of radius r0 g ( , r)
l l l

1 /2

need both f and g to satisfy boundary condition for E < 0 as r

=n
l l

is phase shift of f ( , r )
l

infinite set of integer-spaced -values that satisfy r boundary condition Wave emerges from core with -independent phase. Core transforms wave with correct r 0 limiting behavior into one that exits imaginary sphere of radius r0, which contains the core region, with l phase shift. Core sampled by set of different ls. Today: Wavefunctions and Energy States of many-electron atoms 1. orbitals configurations L - S states 2. electrons are Fermions must be antisymmetrized: KEY PROBLEM
3. Slater determinants are antisymmetric wrt all e , e permutations i j A. Normalization

B. C.

Matrix Elements of one - e Operators: e.g. HSO = Matrix Elements of two - e Operators: e.g. He = (a very bad perturbation)
l

a ri

( )

li

si

e 2 rij

i> j

next few lectures

4.

H eff in terms of n ,

orbital Slater-Condon 1/ rij energy

F k ,G k 123 , n parameters
l

spinorbit

Page 31-9 is an example of what we will be able to do. * Interpretable trends: Periodic Table * Atomic energy levels: mysterious code no atom-to-atom relationships evident without magic decoder ring.
updated September 19,

5.73 Lecture #30

Many-electron H

30 - 2
N

e2 H = h(i) + + a ri l i s i i=1 12 3 1= 1 rij 24444 4 4 i > j4444i 3 (0)

N H H (1)
sum of hydrogenic 1- e terms: Z 2 n2 = n

( )

(unshielded orbital energies)

How do we set up matrix representation of this H? H(0) defines basis set (complete, orthonormal, )

(0)

i =1

h(i) =
N

i =1

[ (r ) sm (i) ]
i i s

spatial part

spin part

spin-orbital

the i(ri)s could be hydrogenic or shielded-core Rydberg-like orbitals.

[h(1) + h(2)] (1) (2) = [h(1)(1)](2) + [h(2)(2)](1)

a constant with respect to h(1). H as sum, E as sum, as product Electronic Configuration: list of orbital occupancies e.g. C 1s22s22p2 six e not sufficient to specify state of system
several L , S terms arise from this configuration: e.g. p2 1D, 3P , 1S r r r r L li S si
i i

we know that

L , Lz , S

e , S z commute with h(i) +

rij

so we can use these to block diagonalize H.

Note that although

li

does not commute with e 2 rij , this is not

a problem for s2 and siz because h(i) and rij do not involve spin. i rij destroys l i but not L!
updated September 19,

5.73 Lecture #30

How do we get eigenstates of L2 , L z , S2 , S z ?

30 - 3

either:

I.

Method of ML, MS boxes Advanced Inorganic Which LS terms exist, not the specific linear combinations of spin-orbital products that correspond to these terms. Angular momentum coupling techniques 3-j ladders plus orthogonality projection operators

II.

We will return to this problem and approach it both ways. One rigorous symmetry must be imposed: Pauli Exclusion Principle: electrons are Fermions and therefore any acceptable wavefunction must be antisymmetric with respect to permutation of ANY pair of e electrons

e.g.

1, 2 = u1(1) u2 ( 2)
orbitals

P12 1, 2 = u1( 2) u2 (1) 2,1

e are indistinguishable, [ H , Pij ] = 0
2 all ' s must belong to + or eigenvalue of Pij (note that Pij = I)

Boson Fermion

( integer spin) (1/2 integer spin)

= 21/2 1, 2 2,1

P12 = 21/2 2,1 1, 2 =

fermions

bosons

updated September 19,

5.73 Lecture #30

generalize to 3 e? 3! combinations needed! Horrible s have N! terms (each a product of N spin-orbitals) matrix elements have (N!)2 additive terms! TRICK! Slater Determinants e

30 - 4

u1(1) u2 (1) = u1(1)u2 ( 2) u1( 2)u2 (1) u1( 2) u2 ( 2)

orbital

row is label for e column is label for spin-orbital

you show that 3 3 Slater determinant gives 6 additive product terms Determinants N N * * * * N! terms in expansion of determinant determinant changes sign upon permutation of ANY two rows [es] or columns [spin-orbitals] determinant is zero if any two rows or columns are identical. determinant may be uniquely specified by main diagonal

MUST SPECIFY IN ADVANCE A STANDARD ORDER IN WHICH THE SPIN-ORBITALS ARE TO BE LISTED ALONG MAIN DIAGONAL

e.g.

s, s, p1, p1, p0, p0, p - 1, p - 1,

[or for pN, suppress p in notation: 110 ML = 2, MS = 1/2]

Need a fancy notation to demonstrate how Slater determinants are to be manipulated in evaluating matrix elements. This notation is to be forgotten as soon as it has served its immediate purpose here.

updated September 19,

5.73 Lecture #30

30 - 5
# of binary permutations

( N !)1/2 (1) p[ u1(1) uN ( N ) ] =

N! different s

is ONE prescription for rearranging the orbitals from the initially specified order is product of several Pijs or, more useful for proving theorems, a product of N factors Pi which tell whether the i-th electron is to be left in the i-th spin-orbital or transferrred to some unspecified spin-orbital

u1(1) uN ( N ) =
A. Normalization
Verify that ( N!)
1 / 2

Piui (i )

i =1

is correct normalization factor

,

N N = (N!)1

(1)p + p[ u1 (1) uN (N) ][ u1 (1) uN (N) ].

Now rearrange into products of one - e overlap integrals, N N = (N!)1

,

(1)p + p

i= 1

Piu i (i) Piu i (i) .

ui are orthonormal u(i) u(j) has no meaning because bra and ket must be associated with same e

The only nonzero legal terms in

are those where EACH Pi = Pi

otherwise there will be AT LEAST 2 mis - matched bra - kets ui ( k ) u j ( k ) u j ( l ) ui ( l ) =0 =0

(Here the electron names match in each bra-ket, but the spin-orbital quantum numbers do not match.)

updated September 19,

5.73 Lecture #30

30 - 6

Thus it is necessary that = , p = p , (1) p + p = +1 and N N

( N !)1 =

u1 (1) u1 (1) uN (N ) uN (N ) =1 =1

each term in sum over gives + 1, but there are N possibilities for P1 , N 1 possibilities for P2

N ! possibilities for sum over

N N = ( N !)
Thus the assumed (N!)
1/2

normalization factor is correct.

1=1
r r L = li
i

B. Matrix elements of one-electron operators

F=

f ri

( )

e.g.

( N !)1/ 2 (1) p a1 (1)

1/ 2

a N (N ) bN (N )

B ( N !)

(1) p b1 (1)

A F B

( N !)1 (1) p+ p[ a1(1) ]f ( ri )[ b1(1) ] = ( N !)1 (1) p+ p [ P1a1(1) P1b1(1) ] =

i ,, i ,,

Pi a i (i ) f ri Pibi (i )

( )

P a ( N ) P b ( N ) N N N N

Product of N orbital matrix element factors in each term of sum. Of these, N1 are orbital overlap integrals and only one involves the one-e operator.
updated September 19,

5.73 Lecture #30

30 - 7

SELECTION RULE A F B = 0 if A and B differ by more than one spin - orbital

(at least one of the orbital overlap integrals would be zero)

two cases remain: 1. differ by one spin-orbital

A = u1(1) a k ( k)uN ( N ) the mismatched orbitals B = u1(1) bk ( k)uN ( N ) are in the same position
use ui to denote common spin-orbitals use ak, bk 0 to denote unique spin-orbitals for this choice, all N Pi factors of each must be identical to all N factors of additional requirement: must bring mismatched orbitals into i-th position so that they match up with the f(ri) operator to give a k (i ) f ( ri ) bk (i ) ANY OTHER ARRANGEMENT GIVES

a ( l ) bk ( l ) u (i ) f r ui (i ) = 0 1k 4 44 1i442i443 42 3
=0 0
(N 1)! ways of arranging the e in the other N 1 matched orbitals and there are N identical terms (in which the e is in the privileged location) in the sum over i

( )

1 A F B = ( N!) ( N 1)! N ak f bk

updated September 19,

5.73 Lecture #30

30 - 8

If the order of spin-orbitals in A or B must be arranged away from the standard order in order to match the positions of ak and bk, then we get an additional factor of (1)p where p is the number of binary permutations

A F B
i.e.

( 1) p ak f bk =

for difference of one spin-orbital

A = 12 5 7 B = 12 3 5 = 12 5 3

A F B = 7 F 3
2. A = B Differ by zero spin-orbitals

A F B

( N!)1 =
i ,

[ P a (i ) f ( r ) P a (i ) ]
i i i i i
all other factors are =1

N ! identical terms from sum over again ( N 1)! N

A F B =

a i f ri a i

( )

Normalization * 1 e Operator F

comes out almost the same as naive expectation WITHOUT need for antisymmetrization!

Examples of f3:

= 31 2 L z = h(3 + 1 2) L z S z = h2

( 3 + 12 1) 2
=
h

J + 31 2 = L + 31 2 + S + 31 2

[0 + 101/2 3 2 2

+ 101/2 31 1 + 0 + 0 + 0 ]

next time G(i,j)

updated September 19,

5.73 Lecture #31

MATRIX ELEMENTS OF F(i) AND G(i,j)
Last time:

31 - 1

orbitals configurations states (terms) Fermions: Slater Determinants: Pauli Exclusion Principle Notation for Slater Determinant: main diagonal . 1. SLATER DETERMINANTAL MATRIX ELEMENTS A. Normalization

TODAY:

B. F(i) One - e operator

e.g.

H SO = a(ri )l i si
i

C. G(i, j) Two - e operator e.g.

H = e 2 rij
e i> j

Recall: specify standard order (because Determinant changes sign upon binary permutation) Goal: make inconvenience of Slater determinants almost vanish matrix elements will be almost what you expect for simple non-antisymmetrized products of spin-orbitals. pages 31-2,3,4 are repeat of 30-6, 7,8
A. Normalization: N = ( N!)
1 / 2

( 1) p[ u1 (1) uN ( N )

verify that (N!)1/2 is correct factor N N = (N!) 1

,

(1)p + p u1 (1) u N (N) u1 (1) u N (N)

] [

rearrange into products of one - e - overlap integrals: = (N!) 1

ui are othonormal u(i) u(j) has no meaning because bra and ket must be associated with SAME e only nonzero LEGAL terms in
,

(1)p + p Pi ui Piui
i =1

* *

are those where EACH Pi = Pi otherwise get AT LEAST

2 MISMATCHED bra - kets ui ( k ) u j ( k ) u j ( l ) ui ( l ) =0 =0

(Here the electron names match in each bra-ket but the spin-orbitals do not match.) Think of a one- or two-e operator as a scheme for dealing with or hiding the small number of mismatched spin-orbitals.
updated September 19,

5.73 Lecture #31

31 - 2

Thus it is necessary that = , p = p , (1) p + p = +1 and N N

( N !)1 =

u1 (1) u1 (1) uN (N ) uN (N ) =1 =1

each term in sum over gives + 1, but there are N possibilities for P1 , N 1 possibilities for P2

N ! possibilities for sum over

N N = ( N !)
Thus the assumed (N!)
1/2

normalization factor is correct.

1=1
r r L = li
i

B. Matrix elements of one-electron operators

F=

f ri

( )

e.g.

( N !)1/ 2 (1) p a1 (1)

1/ 2

a N (N ) bN (N )

B ( N !)

(1) p b1 (1)

A F B

( N !)1 (1) p+ p[ a1(1) ]f ( ri )[ b1(1) ] = ( N !)1 (1) p+ p [ P1a1(1) P1b1(1) ] =

i ,, i ,,

Pi a i (i ) f ri Pibi (i )

( )

P a ( N ) P b ( N ) N N N N

Product of N orbital matrix element factors in each term of sum. Of these, N1 are orbital overlap integrals and only one involves the one-e operator.
updated September 19,

5.73 Lecture #31

31 - 3

SELECTION RULE A F B = 0 if A and B differ by more than one spin - orbital

(at least one of the orbital overlap integrals would be zero)

two cases remain: 1. differ by one spin-orbital

A = u1(1) a k ( k)uN ( N ) the mismatched orbitals B = u1(1) bk ( k)uN ( N ) are in the same position
use ui to denote common spin-orbitals use ak, bk 0 to denote unique spin-orbitals for this choice, all N Pi factors of each must be identical to all N factors of additional requirement: must bring mismatched orbitals into i-th position so that they match up with the f(ri) operator to give a k (i ) f ( ri ) bk (i ) ANY OTHER ARRANGEMENT GIVES

a ( l ) bk ( l ) u (i ) f r ui (i ) = 0 1k 4 44 1i442i443 42 3
=0 0
(N 1)! ways of arranging the e in the other N 1 matched orbitals and there are N identical terms (in which the e is in the privileged location) in the sum over i

( )

1 A F B = ( N!) ( N 1)! N ak f bk

updated September 19,

5.73 Lecture #31

31 - 4

If the order of spin-orbitals in A or B must be arranged away from the standard order in order to match the positions of ak and bk, then we get an additional factor of (1)p where p is the number of binary permutations

A F B
i.e.

( 1) p ak f bk =

for difference of one spin-orbital

A = 12 5 7 B = 12 3 5 = 12 5 3

A F B = 7 F 3
2. A = B Differ by zero spin-orbitals

A F B

( N!)1 =
i ,

[ P a (i ) f ( r ) P a (i ) ]
i i i i i
all other factors are =1

N ! identical terms from sum over again ( N 1)! N

A F B =

a i f ri a i

( )

Normalization * 1 e Operator F

comes out almost the same as naive expectation WITHOUT need for antisymmetrization!

Examples of f3:

= 31 2 L z = h(3 + 1 2) L z S z = h2

( 3 + 12 1) 2
=
h

J + 31 2 = L + 31 2 + S + 31 2

[0 + 101/2 3 2 2

+ 101/2 31 1 + 0 + 0 + 0 ]

now G(i,j)
updated September 19,

5.73 Lecture #31

31 - 5

C. G(i,j) : 4 cases 1. differ by more than 2 spin-orbitals: Matrix Element 0 2. differ by 2 spin-orbitals: one pair of nonzero matrix elements 3. differ by 1 spin-orbital: sum over pairs of nonzero matrix elements 4. expectation value : differ by 0 spin-orbitals: double sum over pairs of matrix elements 1. is obvious only way to make up for orbital mismatch is to hide the mismatched orbitals in |g(i,j)| (rather than in an overlap integral). But one can only hide 2-mis-matched pairs in, e.g.

ai a j g(i, j ) bi b j A G(i, j ) b = 0
if A, B differ by more than 2 pairs of spin-orbitals

2. differ by two pairs of spin-orbitals

A = u1 (1) a1 (i ) a 2 ( j) uN (N ) B = (1) p u1 (1) b1 (i ) b2 ( j) uN (N )

permutations needed to put b1 and b2 in the i and j positions
1 A G B = ( N!) i> j

( 1) p + p [orthogonality integrals]

[ P a (i )
i 1

Pj a2 ( j ) g(i, j ) Pib1 (i ) Pjb2 ( j )

are (N 2)! ways of permuting the N 2 matched uk functions that are not filled with e i and j. Moreover these permutations must involve Pk = Pk (all k i,j). also N(N 1) identical terms in sum over i > j

updated September 19,

5.73 Lecture #31

Thus there are ( N 2)!( N 1) N = N! identical terms in But there are still two possibilities:
1. = p = p and Pi = Pi, Pj = Pj 2. same as except for i, j pair where
i > j ,

31 - 6
sums.

THIS MEANS WE PUT THE j-th e in Pj where WE PUT THE i-th e in Pi

Pi = Pj Pj = Pi

(1) p+ p = 1
the 2 s differ by one binary permutation

THUS: A G B = a1 (1)a2 (2) g(1, 2) b1 (1)b2 (2) a1 (1)a2 (2) g(1, 2) b2 (1)b1 (2)
For A,B different by 2 spin-orbitals # of permutations needed to make B match A no sign ambiguity if standard order is initially specified

3. A,B differ by only one pair of spin-orbitals You work this out
(a,b matched) (a matched with un, un matched with b)
n

A G B =
n

[ a(1)u (2) g(1, 2) b(1)u (2)

n

a(1)un (2) g(1, 2) un (1)b(2)

not arbitrary

4. differ by zero spin-orbitals : expectation value

A GA =

n>m

[ u (1)u
n

(2) g(1, 2) un (1)um (2) un (1)um (2) g(1, 2) um (1)un (2)

DIRECT EXCHANGE unexpected: consequence of antisymmetrization

what we would expect without antisymmetrization

2 (1 )g(1 ,2 )(m) n

updated September 19,

5.73 Lecture #31

31 - 7

The ONLY real surprise that results from the antisymmetrization requirement for two-electron operators is one extra term (and some signs) that has no counterpart if antisymmetrization had been ignored.

SUMMARY * antisymmetrize Slater determinants * matrix elements are hardly more complicated than those of simple spinorbital products signs due to permutation [Standard order] extra terms in G(i,j) Do some examples for p2 1. 2. What L,S terms belong to p2 (Lecture #32: method of crossing out microstates) What is the correct linear combination of Slater determinants that corresponds to a specific L-S term in either the JLSMJ or the LMLSMS basis set ladders plus orthogonality (Lecture #32) L2 and S2 matrices 3-j coefficients
e 2 rij F k (nl, n l ), G k (nl, n ) Slater Condon parameters relative energies of L - S terms expressed in terms of F k and G k ' s

3.

4.

Matrix elements of HSO (NLS) (NLS) nl full HSO in terms of nl

updated September 19,

5.73 Lecture #31

EXAMPLES:

31 - 8

Slater : p2 11 1D
Lz = 11 Lz 11
=
h

M L = 2, M S = 0

[1 + 1] = 2h

Sz

1 1 = h + = 0h 2 2
lz

L2z

tricky! L2z =
i
The only 2 terms in sum are 1,2 & 2,1

+
i j

lz lz i j

Easier to do this by applying L z = l zi

i

1e operator

2e operator

twice

L2z =

[12 + 12 ] +
i j 2 2 2 2

11 l l 11 11 l l 11 zi zj zi zj
from spin-mismatch 2 h

= 2 h + h [1 + 1 0 0] = 4
2 Sz =

as expected

1 1 1 1 + + h 2 0 0 = 0 h 2 as expected 4 4 4 4

L2 L2z = L2x + L2y = L2 =

1 L+ L + L L+ 2

1 L+ L + L L+ + L2 z 2 can you show L2 = h2 6 for 11 of p2 ?

updated September 19,

5.73 Lecture #31

31 - 9

Patterns of Lowest-Lying States: Aufbau for adults!

Atom lowest config. L-S terms Lowest Term

C 1s22s22p2
1

N 1s22s22p3
4

O 1s22s22p4
1

S, 1D, 3P
3

S, 2P, 2D
4

S, 1D, 3P
3

P0
resultant configuration

S3/2

P2

(regular)
excitation

(no fine structure)

(inverted)

Transitions to 2p 2s 2s2p3 lowest 3s 2p 2s22p3s configurations l = 1 3d 2p 2s22p3d

5,3

2s2p4 2s22p23s 2s22p23d

2,4

2s2p5 2s22p33s 2s22p33d

P [ P2 ] 5,3 S, 3,1P, 3,1D [5 S 2 ]
3 1,3

[lowest L-S-J term of each

configuration]
characteristic transition (2p 2s) lowest L-S states of the two relevant configuration
3 3

S, 3,1P, 3,1D [5 S 2 ] 1,3 P 3 [ P0 ] 1,3 F, 1,3P [3 F 2 ]

2,4

F, 2,4P, 2D 2G, 2S

P, 2D, 2S [4P5/2] 2,4 P, 2D, 2S [4P1/2] [4F3/2]

5,3

D, 3,1F, 3,1P, 1,3G, 1,3S [5D0]

regular
S 1
4

inverted
3

5/2

S2
4

P 3/2
1/2 1/2 Pi 3/2 5/2

Pi 1
2

inverted
3

P 1
0

Pi 1
2

3/2

2p(C) < 2p(N) < 2p(O)

updated September 19,

5.73 Lecture #32

Last time: Matrix elements of Slater determinantal wavefunctions Normalization: (N!)1/2 F(i): selection rule (s-o 1), sign depending on order G(i,j): selection rule (s-o 2), two terms with opposite signs

32 - 1

TODAY: Configuration which L-S terms? L-S basis states matrix elements

Method of crossing out M L , MS boxes ladders plus orthogonality Many worked out examples will not be covered in lecture.
KEY IDEAS: * 1/rij destroys spin-orbital labels as good quantum numbers. * Configuration splits into widely spaced L-S-J terms. * 1/rij is a scalar operator with respect to L, S, and J thus matrix elements i > j are independent of M , M , and M . L S J * Configuration generates all possible ML, MS components of each L-S term. * It cant matter which ML, MS component we use to evaluate the 1/rij matrix elements * Method of microstates and boxes: Book-keeping which L-S states are present, organize the algebra to find eigenstates of L2 and S2, basis for sum rule method (next lecture). Longer term goals: represent electronic structure in terms of properties of atomic orbitals 1. Configuration L,S terms 2. Correct linear combination of Slater determinants for each L,S term: several methods 3. 1/rij matrix elements Fk, Gk Slater-Condon parameters, Slater sum rule trick 4. HSO * (NLS) coupling constant for each L-S term in an electronic configuration * (NLS) nl one spin-orbit orbital integral for entire configuration * full HSO matrix in terms of nl 5. Stark, Zeeman, optical transitions r (matrix elements of r, g - values) 6. transition strengths nl r n l + 1 There are a vastly smaller number of orbital parameters than the number of electronic states. The periodic table provides a basis for rationalization of orbital parameters (dependence on atomic number and on number of electrons.)
updated September 19,

5.73 Lecture #32

Which L-S terms belong to (nf)2
* shorthand notation for spin - orbitals nlml / e.g. 4f3 , could suppress 4 and f ( main diagonal for Slater determinant, for simple product of spin - orbitals) * standard order (to get signs internally consistent) 3 32 2 - 3 - 3 is my standard order for f

32 - 2

( 2 l + 1)( 2s + 1) = (7 )( 2) = 14 spin orbitals

* which Slater determinants are nonzero and distinct (i.e., not identical when spin - orbitals are permuted to a different ordering)?

f2 - take any 2 s-os and list in standard order

2 0 is OK but 0 2 is not in standard order and 2 2 0
How many nonzero and distinct Slater determinants are there for f2?

14 spin - orbitals 14 13 = 91 2 identical electrons 2

p general ( nl) : nl

Slater determinants!

1 2( 2 l + 1) p ! p!

[2(2l + 1)]!

put p indistinguishable e and 2(2l+1)-p holes into 2(2l+1) boxes

subshell : one such factor for each subshell

How to generate all 91 linear combinations of Slater determinants that correspond to the 91 possible LMLSMS basis states that arise from f2? Next lecture. * ladders plus orthogonality * construct and diagonalize L2 and S2 matrices * projection operators * 3-j, 6-j, 9-j coefficients

all of these are labor intensive

updated September 19,

5.73 Lecture #32

Sometimes all we want to know is which L-S terms? [WHY? 1/rij is scalar with respect to L,S, and J, thus eigenenergies are independent of ML, MS and MJ.] EASY because can read Lz = liz and Sz = siz directly from the spin-orbital labels.

32 - 3

L z 21 =
i =1

l iz

21 = h[2 + 1] 21

ML = 3
ML is sum of ml s MS is sum of mSs NONLECTURE
What about L2 ? Can do this in either of two ways: * as below (very cumbersome) * L2 = L2z + (1 / 2)(L +L + L L + ) [separately apply each 1e operator rather than treat entire operator as a 2e operator.] very laborious because L2 =
i, j

li lj

=
i

li

+ 2
i> j

l il j

L2 21
i

one e
h li li
2

twoe
li

+ 1) 21

= 3 for f

WORK OUT L2 matrix for M L = 3, MS = 0 block of f 2 for future reference

L2 =
i, j

li

l j =
i

[l ] + 2
2 i i> j

l l + 1 l l + l l i j + ) iz jz 2 ( i + j
l = 0, Ml = 0 ml1 = ml2 = 1
nonstandard order

l = 0, Ml = 0
all are M S = mS1 = mS2 = 0

l l

( + 1)

ml1ml2

(12 + 12) 21 + 2(2 1) 21 + 1/ 2 1/ 2 L2 21 = h2 [3 4 2 1] [3 4 1 2] 1 2 + 1/ 2 1/ 2 [3 4 2 3] [3 4 1 0] 3 0

l 1 l 2 + l 1+ l 2

= h2 [28 21 10 21 + 12 2 1/ 2 3 0 ]

All of the 12, 21 type matrix elements are 0 because of ms mismatch.

e.g. 21 spatial part 12 = 0

Recall (12|G|12 12|G|21).

updated September 19,

5.73 Lecture #32

L2 21 = L2 3 0 = L2 3 0 =
Many steps skipped
h
2

32 - 4

h h

2 2

[28 21 10 21 + 12 2 3 0 ] [(24 + 3 0) 3 0 + (12 2 ) 21 ] [24 3 0 + (12 2 ) 21 ]

1/2 1/2 1/2

3 0 24 0 12 21/2 0 21 0 24 0 12 21/2 2 2 L =h 1 2 12 21/2 0 28 10 12 21/2 10 28 0 3 0

[the bottom two Slater determinants are intentionally out of standard order to display decreasing values of ml(1) and increasing values of ml(2).] find eigenvalues and eigenvectors of this block ML = 3, MS = 0 of f2

L2
h
2

[3 [6

1 / 2

3 0 + 3

1 / 2

3 0 + 6

1 / 2

21 + 6

1 / 2

2 1 ] = 30[

L=5

] ]
L=6

L2
h
2

1 / 2

3 0 + 6 1/ 2 3 0 31/ 2 21 31/ 2 2 1 ] = 12[

L=3

L2
h
2

[11

1 / 2

3 0 111/ 2 3 0 + 3 22 1/ 2 21 3 22 1/ 2 2 1 ] = 42[ 3 0 3 22
1 / 2

] ]

L2
h
2

[3 22

1 / 2

3 0 11

1 / 2

21 + 11

1 / 2

2 1 ] = 20[

L=4

(Note how easy it is to see that normalization is correct.) a lot of algebra is not presented here! * each Slater basis state gets used up * first 2 eigenfunctions are in form + S = 1 second 2 eigenfunctions are in form S = 0 prove this by applying S2 to above eigenfunctions of L2

END OF NON-LECTURE
updated September 19,

5.73 Lecture #32

32 - 5

Nonlecture pages were intended to show that applying L2 and S2 to Slater determinants is laborious much moreso than Lz and Sz.

This is one reason why we use the crossing out ML, MS microstates method to figure out which L,S states must be considered. Often this is sufficient or can be the basis for some shortcut tricks! ML, MS method works because: * each configuration generates the full (2L + 1) (2S + 1) manifold of ML, MS states associated with a given L,S term. Why? If you have one |MLMS you can generate all of the others using L and S operators. This must be true because, starting with ML = L, MS = S, L and S can be used to generate the full L,S term without the need to go outside the specific configuration. ML, MS method
ML= LMAX MS = SMAX S1 list all Slater determinants L1 L2 0

SMAX = (# of e)/2.

No need to include negative values of MS or ML.

updated September 19,

5.73 Lecture #32

ML MS

f2
4 (G) 31 2 2 3(F) 3 0 21 2(D) 2 0 3 1 11 1(P) 3 2 2 1 1 0 0(S)

32 - 6

6(I) 1 3 3

5(H) 3 2

3 3 2 2 1 1 0 0 3 3 3 3 2 2 2 2 1 1 1 1 0 0

3 3

3 2 32

31 31 2 2

3 0 30 21 21

2 0 20 3 1 3 1 11

3 2 3 2 2 1 2 1 1 0 10

Slaters for f2

need not include MS < 0 or ML < 0 because these are identical to the ML > 0 and MS > 0 quadrant. Notice that as you go down in ML, the number of Slater determinants in each ML, MS box increases only by one. This is a prerequisite for using the L plus orthogonality method! This useful simplicity does not occur as you go down a column in MS. This convenient situation does not occur for d3 or f3. Why? Because there are more than one L-S term of a given symmetry.

updated September 19,

5.73 Lecture #32

No J

32 - 7

S P D F G H I K L= 0, 1, 2, 3, 4, 5, 6, 7
Start in extreme ML, MS corner This generally contains only one Slater determinant

L = MLMAX

S = MS MAX

so we have one of the L - S terms

This L-S term includes one of each ML, MS in the range

L ML L S MS S

This means this L-S term will use up the equivalent of one Slater determinant in each ML,MS box

bookkeeping cross out one Slater determinant, any one, from each relevant ML,MS box now repeat, again starting at the extreme ML,MS corner

etc. * 3H * 1I * 3F * 1G * 3P * 1D * 1S

3 11 = 33 1 13 = 13 3 7 =21 19 =9 33 =9 15 =5 11 =1
91 as required!

Since there is only one Slater determinant in the ML = 5, MS = 1 box, generate all triplets by repeated application of L to ||32|| (plus orthogonality) and generate all singlets by L on ||33||. Many orthogonalization steps needed! Especially for singlets. Need S also.

updated September 19,

5.73 Lecture #32

32 - 8

Before illustrating the ladders plus orthogonality method, it is useful to show some patterns and list some tricks.

Most difficult cases are (nl)m where m = 2, 3, 2l. Easy to combine nl with nl because no need for special bookkeeping.
l

s p d f g

(n l )2 1S
1D, 3P, 1S 1G, 3F, 1D, 3P, 1S 1I, 3H, 1G, 3F, 1D, 3P, 1S

(n l )3
4S, 2D, 2P
2 2H, 2G, 2F, 4F, 2D(2), 4P, 2P H,

a simple, memorable pattern

rather complicated

same L-S states for 2 and 3 holes instead of electrons.

(nl) 2 n l[nl 2 2S +1 L ] ( 2 l )= 2S + 2, and 2S (L + l, L + l 1,L L l )

( )

simple vector coupling

updated September 19,

5.73 Lecture #32

Ladder and Orthogonality Method f2 example Start with 2 extreme UNIQUE states 3 H M L = 5, M S = 1 = 3 2

32 - 9

Use this to generate all triplets by applying L repeatedly and using orthogonality when necessary. Note that # of determinants in each ML,MS=1 box increases no faster than in steps of 1. To get to 3P, must not only apply orthogonality several times, but must follow each L state down to the ML = 1 box!
1 To get singlets, start with I M L = 6, M S = 0 .

Again, as L takes us to successively lower ML boxes, # of determinants increases in steps of 1. But some of these steps are due to triplets with MS = 0. Need to step triplets down into MS = 0 territory using S once. Lots more orthogonality steps, lots more trails being followed. AWFUL, but do-able. Nonlecture
3

H ML MS = L 3 H 51 = 2l i 3 2

0
big surprise!

[ 5 6 5 4]1/2 3 H 41 = h[ 3 4 3 2]1/2 2 2 + ( 3 4 2 1)1/2 31

3

H 41 = 3 1 H 31 = (1 / 3)
1/2

L 3 H 41 = l i 3 1
3

2 1 + ( 2 / 3)

1/2

3 0

updated September 19,

5.73 Lecture #32

orthogonality:
3

2 F 31 = 3

1 /2

1 21 3

1 /2

32 - 10
3 0

ML
and so on, to get all
3

MS

L L 1 many electron functions.

MS = 0 Try a detour into singlet territory, and then check for self-consistency. 1 /2 1/2 3 2 21 1 3 0 S F 31 = s i 3 3 i 1 /2 1 /2 1 /2 2 1 3 1 1 3 F 30 = h 21 + 21 h[1 2 1 0] 3 2 2 2 2

1 3
3

1 /2

[1]1/2 ( 3 0
1 /2

+ 3 0

1 F 30 = 3
3

1 /2

1 21 + 21 6

[ 3 0

+ 3 0

S
3

1 /2 1/2 2 1 21 + 3 0 H 31 = s i 3 3 i

1 H 30 = 6

1 /2

1 21 + 21 + 3

1 /2

[ 3 0

+ 3 0

There are 4 Slater determinants in ML = 3, MS = 0 box. We cant find the other two singlet linear combinations uniquely without using L on the extreme singlets.
L 1I 60 =
h

l i 3 3

[6 7 6 5]1/2

I 50 =

[3 4 3 2]1/2 ( 2 3

+ 3 2

)
1 /2

wrong order
1 I 50 = 2
1 /2

[ 3 2

3 2

] orthogonality
1 /2

1 H 50 = 2

[ 3 2

+ 3 2

1 L 1I 50 = l i 2

[ 3 2

3 2

]
updated September 19,

5.73 Lecture #32

wrong order
1

32 - 11
1/ 2

1 I 40 = 44
5 I 40 = 22

1/ 2

[(10) ( 31 31 ) + 6 ( 22 22 )]
1/ 2
1 /2 31 31 + 6 2 2 11

1 /2

1 H 40 = 20
1 /2

1 /2

[(6)

1 /2

( 2 2
+ 31

+ 2 2 + (10)

1 /2

( 31

+ 31

wrong order

)]

1 H 40 = 2

( 31 (

) )

orthogonality 1 /2 1 /2 3 5 1 G 40 = 31 31 2 2 11 11 At last we are ready to enter the ML = 3, MS = 0 block! It is clear that if we apply L- to 3 H 40 we will get the same form we
already derived starting from 3 H 51 . Let' s lower 1I 40
1 /2 1/2 5 6 L I 40 = l i 31 31 + 2 2 11 22 i 1 /2 1 /2 1 (30)1/2 5 (6)1/2 21 21 + 5 (12)1/2 3 0 3 0 I 30 = 22 22 1

6 + 11
1

1 /2

(10)1/2 ( 21

1 /2 1/2 1/2 1 + 4 21 21 + 2 I 30 = 3 0 3 0 22 22 22

21

9 I 30 = 22

1 /2

2 21 21 + 22

1 /2

( 3 0

3 0

Finally, by orthogonality:
IMPORTANT

1 1 G 30 = 11

1 /2

9 21 21 + 22

1 /2

( 3 0

3 0

Does this match what one would get from L 1G 40 ?

updated September 19,

5.73 Lecture #32

1

32 - 12

1 /2 1/2 3 5 L G 40 = l i 31 31 2 2 11 11 1 /2 1 /2 1/2 3 1 /2 1 /2 3 1 G 30 = (8) (6) 21 21 + (12) 3 0 3 0 11 11 1 /2 5 (10)1/2 21 21 11

1 IMPORTANT 1G 30 = 11
checks!

1 /2

9 21 21 + 22

1 /2

( 3 0

3 0

End of Non-Lecture
As you see, this is extremely laborious. There is a better way!
There are several patterns: singlets for MS = 0 always have the ** form ( ) and triplets always ( + ).

This can be generalized for any value of S (page 61 of Hlne LefebvreBrion-Robert Field Perturbations book) [M. Yamazaki, Sci. Rep. Kanezawa Univ. 8, 371 (1963).] 2. Failure and Inconvenience of ladder method The ladder method is OK when you have a single target LML SMS state, especially when it is near an edge of the ML,MS box diagram. Essential that # of Slater determinants in each MLMS box increases in steps of 1 as you step down in ML or MS. Fails when there are 2 L-S terms of same L and S in a given configuration must set up 2 2 secular equation anyway.

eg.

(nd )3

H ,2 G,2 F ,4 F ,

D( 2), 4 P ,2 P

3. L2 and S2 Matrix Method Another method is based on constructing L2 and S2 matrices in the Slater determinantal basis set. This is no cakewalk either!

Since usually SMAX << LMAX for a configuration, it is best to start with S2 because it is simpler.

updated September 19,

5.73 Lecture #33

33 - 1

L-S Terms via L2, S2 and Projection

LAST TIME: method of ML, MS boxes. [For 3L states, cross out boxes starting from both (ML=L,MS=1) and (ML=L,MS=0).] complete (2L + 1)(2S + 1) dimension for each L-S term [# of boxes] * nl2 pattern * (nl)2 nl * method of ladders plus orthogonality *

TODAY: L2, S2 method to obtain |LMLSMS, especially for ML,MS boxes in which the where method of ladders plus orthogonality is most inconvenient: ML = 0, MS = 0 * L2 L+L only for ML = 0 block. Every LS term is represented in this most evil block. * set up and diagonalize S2 easy by forming linear combinations (singlet and triplet) + 2 * transform L to singlet, triplet basis (block diagonalization), then diagonalize L2 by knowing (from crossing out boxes method) eigenvalues: L(L + 1)

other, strong spin-orbit basis sets Modern calculations use projection operators: designed to project away all unwanted parts of yet preserve normalization.

updated September 19,

5.73 Lecture #33

33 - 2

Look at the ML = 0, MS = 0 block of f2 and construct all L S basis states. All extant L-S terms of f2 are present once in the ML = MS = 0 block. Never try to get to this block by ladders and orthogonality!

1 = 3 3 2 = 3 3 3 = 2 2 4 = 2 2 5 = 1 1 6 = 1 1 7 = 0 0
Cute trick that works especially well in ML = 0 and MS = 0 blocks because many otherwise awful terms vanish.

Do d2 in lecture

1 1 (L +L + L L + ) = L2z + 2 (L +L + L +L [L + , L ]) 2 [L+ ,L ] = 2hLz L2 = L2z + L2 = L2z hL z + L +L

diagonal but vanishes in ML = 0 nondiagonal

(same as L2 = L2z + hL z + L L + )

So for ML = 0 block only, can replace L2 by L+L (or LL+) and, for MS = 0 only, replace S2 by S+S.

updated September 19,

5.73 Lecture #33

For f2:

33 - 3

ignore factors of h2

S2 3 3 = S+ S 3 3 = S+ 3 3 = 3 3 + 3 3 L2 3 3 = L + L 3 3 = L + 61/ 2 2 3 =
1/ 2 1/ 2 61/ 2 [12 6] 3 3 + [12 6] 2 2

= 6[ 3 3 + 2 2 ] etc.

1 = 3 3 2 = 3 3 3 = 2 2 4 = 2 2 5 = 1 1 6 = 1 1 7 = 0 0

S2 1 = 1 + 2 S2 2 = 1 + 2 S2 3 = 3 + 4 S2 4 = 3 + 4 S2 5 = 5 + 6 S2 6 = 5 + 6 S2 7 = 0
all easy

L2 1 = 6 1 + 6 3 L2 2 = 6 2 + 6 4 L2 3 = 6 1 + 16 3 + 10 5 L2 4 = 6 2 + 16 4 + 10 6 L2 5 = 10 3 + 22 5 + 12 7 L2 6 = 10 4 + 22 6 12 7 L2 7 = 12 5 12 6 + 24 7
require a bit more work

now we know, for 2e, S2 can only have 2h2 and 0h2 eigenvalues (triplet and singlet) diagonalize S2 by inspection
t : + s :

1t = 2 1/ 2 ( 1 + 2 ) 2 t = 2 1/ 2 ( 3 + 4 ) 3t = 2 1/ 2 ( 5 + 6 )

1s = 2 1/ 2 ( 1 2 ) 2 s = 2 1/ 2 ( 3 4 ) 3s = 2 1/ 2 ( 5 6 ) 4s = 7
This also has form
updated September 19,

5.73 Lecture #33

33 - 4

Confirm that these functions diagonalize S2 and give correct diagonal elements. 1 a diagonal element: 1t S2 1t = ( 1 + 2 ) S2 ( 1 + 2 ) 2 1 1 = h2 ( 1 + 2 )(2 1 + 2 2 ) = h2 (2 + 2) = 2 h2 2 2 1 an off - diagonal element: 1t S2 1s = ( 1 + 2 ) S2 ( 1 2 ) 2 1 = h2 ( 1 + 2 )( 1 + 2 1 2 ) = 0 2
also 1s S 2 1s = 0

S2 = h 2 2

2 2 0

0
0 0

0
1 2 2 h [6 + 6 ] = 6 h 2

1t 2t 3t 1s 2s 3s 0 4 s

What does L2 look like in basis set that diagonalizes S2?

L2 1t = 2 1/ 2 h2 [6 1 + 6 3 + 6 2 + 6 4 ]

1t L2 1t =

NONLECTURE
2 t L2 1t = 2 t L2 2 t 1 2 2 h [ 6 + 6] = 6 h 2 1 2 = h2 3 + 4 L 3 + 4 2 1 = h2 3 + 4 6 1 + 16 3 + 10 5 + 6 2 + 16 4 + 10 6 2 1 = h2 (16 + 16) = 16 h2 2

updated September 19,

5.73 Lecture #33

33 - 5
0 10 22 6 6 0 0 6 16 10 0

L2 = h2 6 6 0

6 16 10

0
0 10 22 24 2 1/ 2

0
6 6 0 6 16 10

1t 2t 3t 0 1s 0 2s 24 2 1/ 2 3s 24 4 s

These 2 matrices are easier to diagonalize than the full 7 7 matrix, especially because we know the eigenvalues in advance! Our goal is actually the eigenvectors not the eigenvalues

TRIPLETS

L2 3 H M L = 0, MS = 0 = h2 30 3 H 00
0 a a 10 b = 30 b 22 c c eigenvector equation

(L2 )
6 a + 6b + 0c = 30 a 6 a + 16b + 10c = 30b 0 a + 10b + 22c = 30c b =
1 = [a 2 + b 2 + c 2 ]
1/ 2

b=

24 a = 4a 6 8 c 10

a=b 4

c=

5 b 21

a = 42 1/ 2 b = (8 / 21)
1/ 2 1/ 2

c = (25 / 42)

updated September 19,

5.73 Lecture #33 1 /2 1 /2 3 H 00 = 421/2 1t + (8 / 21) 2 t + (25 / 42) 3 t

Similarly,
3

33 - 6

F 00 = 31/2 1t + 2 t 3 t 9 P 00 = 14
1/2

2 1t + 2 t 14 1/2 3 t 7

1 /2

Note that each nt basis state gets completely used up and all eigenvectors are normalized and mutually orthogonal. You check both used up and orthogonality. Nonlecture: Singlets

L2 1I 00 =

42 1I 00 a 0 a b 0 b = 42 1/2 c 24 2 c d 24 d
6b = 36a a = b/6 5 b 2 d=
1 /2

6 6 0 6 16 10 0 10 22 0 0 24 21/2
6a + 6b = 42a 6a + 16b + 10c = 42b 24 21/2 c + 24d = 42d normalization:
1/ 2 1

6a + 10c = 26b 10c = 25b c = 24 21/2 c = 18d 1 25 50 + 1 = b + 1 + 4 9 36

1/ 2 1/ 2

10 3 21/2

b = (6 77)
1/ 2

1 /2

1 6 6 5 6 I 00 = 1s + 2 s + 77 6 77 2 77 A lot of algebra skipped here:

1

10 3 3s + 3 77
1/ 2

4s

9 49 1 G 00 = 1s + 2 s + 77 77 77 25 D 00 = 42 2 S 00 = 7
1/ 2

1/ 2

1/ 2

1/ 2

18 3s 77
1/ 2

4s

9 1s + 0 2 s + 42
1/ 2

1/ 2

8 3s 42

4s
1/ 2

1/ 2

2 1s + 7

2 2s 7

1/ 2

1 3s + 7

4s
updated September 19,

5.73 Lecture #33

33 - 7

Again note that each ns is used up. Check for orthogonality! Two opposite strategies:
1. ladder down from extreme ML, MS 2. L2 + S2 matrices are large but easy to write out for ML = 0 and MS = 0 ONLY could then ladder up from any L2, S2 eigenfunction (no need to use orthogonality).

Before going to Projection Operators, look at the problems associated with getting 2 other kinds of basis states.
mj coupled orbitals important for strong spin-orbit limit with HEAVY ATOMS. (HSO is diagonal in j and in JMLS) nl >> energy separations between L-S terms (all nl are 0)

jls

JM J LS

coupled many-electron L-S-J states. Again useful in strong spin-orbit limit

JM J LS
J2 (ladders are almost useless)

3- j or ladders

LM L SMS
ladders and L2 and S2 and 3-j in multiple steps

(n j
1 1

1l 1 1

s 3- j or ladders

(n l m
1 1

l1

s1ms1 )( )

to get here, must go long way around or use projection operators.

updated September 19,

5.73 Lecture #33

NONLECTURE
BOX METHOD FOR jls orbital basis: No need to specify l and s.

33 - 8

(nf )

Example

Standard Order: (7/2 7/2), (7/2 5/2), (7/2 3/2), (7/2 1/2), (7/2, -1/2), (7/2, -3/2), (7/2 -5/2), (7/2 -7/2), (5/2 5/2), (5/2 3/2), (5/2 1/2), (5/2 -1/2), (5/2 -3/2), (5/2 -5/2) 14 functions. List only Slater determinants with MJ 0. Suppress the /2s
# ( 0) ( 2) ( 3) ( 6) (7) (10) (11) (13) MJ 7 6 5 4 3 2 1 0 7777 7775 7773 7771 777 1 777 3 555 1 777 5 7 355 777 7 7 151 7755 7753 7751 775 1 775 3 5351 775 5 555 3 757 5 7 373

7555 7573 7571 777 1 757 3 535 1 755 5 7 353

7553 7551 775 1 755 3 737 3 7 555

7355 7353 7371 737 1 735 3 555 5

5553 5551 7351 735 1 717 1 535 3

7155 7153 7151 715 1 515 1 7 155 7 153 7 171

AWFUL! The number of Slater determinants increases in steps larger than 1

as you move down from MJ = J.

Work in the 13 member MJ = 0 block

J 2 = J 2 hJ z + J + J z
diagonal

Worst possible one for ladders plus orthogonality.

2

[Hopeless to attempt to set up L

and S2 matrices!]

Dimension of Various J blocks: J = 6 5 4 3 2 1 0

Dimension = 2 1 3 1 3 1 2

All blocks are manageable! Ladder up from MJ = 0.

updated September 19,

5.73 Lecture #33

coupled basis sets are convenient for LS and l i si uncoupled basis sets are convenient for (Lz + 2Sz)

33 - 9

Either of the two many-electron basis sets is OK for The big problem for e2/rij is that it has many off-diagonal matrix elements in the Slater determinantal basis set. These are extremely tedious to evaluate. The solution to this is the Slater Sum Rule method. It is based on the fact that the trace of a matrix is equal to the sum of the eigenvalues. This is true regardless of what representation is used to express the matrix.
SUM RULE METHOD: diagonal matrix elements of e 2 rij in the Slater determinantal basis set

e2 rij .

NEXT TIME

updated September 19,

5.73 Lecture #33

NONLECTURE: Projection Operators

33 - 10

Alternative method to set up LMLSMS or JLSMJ basis sets in terms of either nlmlsms or njls orbital Slater basis sets. 1. Work out L2 and S2 matrices for nlmlsms (or J2 for njls). These matrices are block diagonal in ML,MS (or MJ). 2. Construct an operator which, when applied to an arbitrary function, annihilates the undesired part of that function. e.g. annihilate L by
[L2 h2L(L + 1)]

3. Modify the above operator so that it preserves the amplitude of the L component of . e.g. annihilate L , preserve amplitude of L L2 h 2 L ( L + 1) 2 ( h [ L L + 1) L ( L + 1)]
show how this works by applying it to = a L + b L
P( a L + b L ) = a L ( L + 1) L ( L + 1) L ( L + 1) L ( L + 1) L + b L ( L + 1) L ( L + 1) L ( L + 1) L ( L + 1) L = a L + 0 L

4. Now recognize that one can build a projection operator that annihilates all undesired L components by taking a product of operators like that in #3, one for each L .

PL =

all L L

L2 h2 L ( L + 1) 2 2 h L ( L + 1) h L ( L + 1)

updated September 19,

5.73 Lecture #33

33 - 11

5. Recognize that PL = aLL, which is not normalized, because aL is the amplitude of L in . Get a normalized L by recognizing that L = aL

L = L

This method is useful for dealing with JMJLS in the jls orbital basis because there is no simple way of block diagonalizing J2 in terms of L2 and S2, can only block diagonalize J2 in terms of MJ. Modern calculations will simply set up the J2,Jz matrix, diagonalize J2, and then discover to which eigenvalues of L2 and S2 each J2, Jz basis function belongs. In many cases two or more LS terms will contain LSJ components which belong to the same eigenvalue of J2.

updated September 19,

5.73 Lecture #34

e2/rij and Slater Sum Rule Method
LAST TIME:

34 - 1

1. L2,S2 method for setting up NLMLSMS many-electron basis states in terms of linear combination of Slater determinants 2 * ML = 0, MS = 0 block: L L + L S2 S + S * diagonalize S2 (singlets and triplets) * diagonalize L2 in same basis that diagonalizes S2 [Recall: to get matrix elements of L2, first evaluate L2 i and then left multiply by j 2. coupled representations njls and NJLSMJ 3. Projection operators: automatic projection of L2 eigenfunctions* remove unwanted L part * preserve normalization of wanted L part * remove overlap factor

TODAY: 1. Slater Sum Rule Trick (trace invariance): MAIN IDEA OF LECTURE. 2 2. evaluate e rij matrix elements (tedious, but good for you)
i> j

[2-e operator, spatial coordinates only, scalar wrt J,L,S] * multipole expansion of charge distribution due to other electrons * matrix element selection rules for e2/rij in both Slater determinantal and many-e basis sets * Gaunt Coefficients (ck) (tabulated) and Slater-Condon (Fk,Gk) Coulomb and Exchange parameters. Because of sum rule, can evaluate mostly ab r1 ab and ab r1 ba type matrix elements and never ab r1 cd ij ij ij type matrix elements. 3. Apply Sum Rule Method 4. Hunds 1st and 2nd Rules

updated September 19,

5.73 Lecture #34

1. Slaters Sum Rule Method

34 - 2

It is almost always possible to evaluate e2/rij matrix elements without solving for all |LMLSMS basis states. * trace of any Hermitian matrix, expressed in ANY representation, is the sum of the eigenvalues of that matrix (thus invariant to unitary transformation) *

e
i> j

/ rij and every scalar operator with respect to J (or L,S) has

nonzero matrix elements diagonal in J and MJ (or L and ML) and independent of MJ or (ML,MS) [W-E Theorem: J is a GENERIC ANGULAR MOMENTUM with respect to which e2/rij is classified] Recall from definition of r12, that e2/rij is a scalar operator with respect to J, L, S but not with respect to ji or li.
Interelectronic Repulsion:

e
i> j

rij

* destroys orbital approximation $$ for electronic structure calculations * correlation energy, shielding
e1
r r1
e2

r r12

e1 at (r1 ,1 , 1 )
e2 at (r2 , 2 , 2 )

r r2

r r r r12 = r2 r1
scalar with respect to J, L, S, si but not ji, l i

2 r12 = r12 2r1 r2 + r22

r r 1/ 2 r12 = r + r 2 r1 r2 cos(r1 , r2 )

2 1

2 2

updated September 19,

5.73 Lecture #34

r expand r121 as power series in < r> where r< is smaller of r1 , r2

34 - 3

(integrals evaluated in 2 regions :

r1 < r2 , r2 < r1 )

lengthy algebra see Eyring, Walter, and Kimball Quantum Chemistry pages 369 - 371 and, for relationship between Legendre polynomials and Y m , , pages 52 - 59.
l

( )

2n-pole moment (n=0 monopole, n=1 dipole,)

multipole expansion

1 = rij n = 0
not principal q.n.!

m=n

* 4 r<n m m Y , i ) Yn ( j , j ) n +1 n ( i 2 n + 1 r>

convergent series

angular momenta magnitude n, projection m

scalar product of 2 angular momenta, one for i-th particle, one for j-th * converts m to m

n-pole charge distribution n-th rank tensor 2n+1 components

[Y

No dependence on s, so 1 / rij is scalar with respect to S, si , s j .

m n

(i , i ) =

i , i

li

= n, m i = m
l

updated September 19,

5.73 Lecture #34

34 - 4

The reason for this rather complicated looking expansion is that it is well suited for integrals over atomic orbitals which are expressed in terms of ri, i, i, which are coordinates of the i-th e with respect to the center of symmetry (nucleus) rather than the other e. It enables use of AO basis states. Otherwise 1/rij integrals would be nightmares.

Selection rules for matrix elements:

not principal q. n.

l i n , m = m , ms = 0 i i orbitals l j n , m j = m , ms j = 0
l l

term in multipole expansion

triangle rule,

li

li n l i + li

(steps of 2 because of parity)

overall: L = 0, S = 0, ML = 0, MS = 0, and indep. of ML, MS Can use any ML, MS from box diagram.
It is also clear how to evaluate the angular factors of the atomic orbital matrix elements using 3-j coefficients. Special tables of Gaunt Coefficients (also C&S pages 178-179, Golding, page 41, see handout).

updated September 19,

5.73 Lecture #34

general 1/r12 matrix element (so = 0, 1, and 2 are possible)

34 - 5

e2

ab 1 r12 cd =
e1

ab

1 cd r12

ab

1 dc r12

ab

1 cd = ( ms a , msc ) ( msb , ms d ) ( m a + m b , m c + m r12

l l l

ld

**
k =0 {
tensor rank for product of AOs occupied by e # 1 must be same as for # 2 for scalar product of n -th rank tensors

1/r12 does not operate on spin coodinates

**
k

1/r12 scalar with respect to L = + l l

12 1 2

ck

la

m a , l cm
l

lc

e1

)c (

lb

m b , l dm
l

ld

e2

GAUNT COEFFICIENTS ANGULAR FACTOR OF INTEGRAL

R k (nal a nbl b ncl c nd l d )

radial factor

(m,
l
l

m
l

)
l l

tabulated

2l + 1 2l + 1

1 /2 k k A000 Am m ,m , m
ll l l
l l l

Clebsch-Gordan coefficients that result from integral over product of 3 spherical harmonics one from operator, two from orbitals
kll 000

triangle rule:

- l k l + l + l + k = even

(from properties of A )

( parity)
updated September 19,

5.73 Lecture #34

restrictions on k and m:
e1 integral

34 - 6
m 1 + m = m
l l1

n1l 1m 1 Ykm n1l 1m 1

l l

triangle rule for intraconfiguration matrix elements, Rk(abcd) takes on especially simple form (because the same one or two orbitals appear in the bra and in the ket).

Rk ( ab, ab) F k ( a, b) Slater Condon parameters k k elements dependent only on R ( ab, ba) G ( a, b) (these are reduced matrix the m quantum numbers.) All, , , and not on any of
la lb l c ld

L-S states from one configuration are expressed in terms of the same set of Fk, Gk parameters.

e2 ab ab r12
e2 J(a , b) ab ab = r12

spins must match or K term vanishes

= J( a, b) ( ms a , msb )K( a, b)
DIRECT

EXCHANGE

k =0

c k (l am , l am )c k (l bm , l bm

la la lb

lb

) F (n
k

al a

, nb , l b
1

ak

la

m a , l bm
l

lb

) [ a * (1)a(1)Opb * (2)b(2)d d ]
2 charge distributions
lb

e2 K (a , b) ab ba = msa ,msb r12

) [ (

ck

la

k =0

m a , l bm
l
l

)] G (n
2 k

al a

, nbl b

)
1 2

b k (l a m a , l b m

lb

)
something else!

[ a * (1)b(1)Opa(2)b * (2)d d ]

for special cases, such as nd 2 , nal a = nbl b and F k = G k

Now we are ready to use tables of c k (or, more conveniently, a k and b k ) to set up e 2 rij matrix
updated September 19,

5.73 Lecture #34

Easy example: nf 2

34 - 7
( recall 1I, 3 H, 1G, 3 F, 1D, 3 P, 1S)

I 60 = 3 3 3 H 51 = 3 2
1

these are the only L-S states represented by a single Slater determinant extremes of ML,MS box diagram

since e2/rij is a scalar operator with respect to L, S, J , matrix elements are ML, MS, MJ independent so we can use any ML,MS component to evaluate the matrix element whichever is most convenient! e1 e2 2
1

2 e 1 I = c k (33, 33)c k (33, 33) F k (nf , nf ) ( , ) [c k (33, 33)] G k (nf , nf ) m r12 k = 0, 2, 4, 6 k

l
l

=
3

k = 0, 2, 4, 6

[c (33, 33)] F (nf , nf )

k 2 k
k k k

2l

one spin other spin

=0

Fk(nf2)

e2 3 H = r12

k = 0 ,2 ,4 ,6

{[c (33,33)c (32,32)]F (nf ,nf ) [c (33,32)] G (nf ,nf )}

k 2 k

e1

e2

both spins

F k (nf 2 )

(a,a)

(b,b)

(a,b)

Here is where everyone makes mistakes! Use table of ck in Golding/C&S handout (C&S page 179). Note that [1/736164]1/2 is implicit after the first entry for f2, k = 6. k=0 ck(33,33) ck(32,32) ck(33,32) Dk convenient factor 1 1 0 1 2 1/3 0 +1/3 225 4 1/11 7/33 301/2/33 1089 = 332 6 [1/736164]1/2 [6/736164]1/2 [7/736164]1/2 736164

updated September 19,

5.73 Lecture #34

34 - 8

Dk is a factor that simplifies the expressions. Each term has the form Fk/Dk. Call this ratio Fk. Get simpler looking expressions when you replace Fk by DkFk (Dk appears in denominators of ck as [/Dk]1/2 )
1

1 1 4 6 e2 1 1 I I = F0 + F2 + F + F 9 121 7361 64 r12 = F0 + 25 F2 + 9 F4 + F6

Always have two factors of ck. Thus Fk gets divided by Dk to yield Fk.

1 1 7 2 30 4 6 7 6 e2 3 H H = F 0 + (0) (1 3) F 2 + F + F r12 7361 64 3 11 33 33 33 = F0 13 1 2 51 F F4 F6 2 9 7361 64 (33)

= F0 25 F2 51F4 13 F6

A lot of book keeping, but easy to learn how to use tables of ck , a k , b k , D k . But it is much more work for f 3 than for f 2 .

SUM RULE METHOD:

Basic idea is that the sum of diagonal elements in the single Slater determinant basis set within an ML, MS box is equal to the sum of the eigenvalues!
Look at M L = 3, M S = 1 box: but trace is E ( 3 H ) + E ( 3 F ) and we already know E ( 3 H )! 3 0 and 2 1 . This box generates
3

H 31 and

F 31 ,

So

E( 1 I ) = 33

E( 3 H ) = 3 2 E( 3 F ) = 3 0 + 21 E( 3 H ) E( 1G) = 31 + 31 + 2 2 E( 1 I ) E( 3 H ) E( 1 D) = 3 1 + 3 1 + 2 0 + 20 + 11 E( 1 I ) E( 1G) E( 3 H ) E( 3 F ) E( 3 P) = 3 2 + 2 1 + 1 0 E( 3 H ) E( 3 F ) E( 1S ) = sum of seven sum of six E( 2 S +1 L)

updated September 19,

5.73 Lecture #34

*

34 - 9

This seems rather laborious, but it is much easier than: generating each LML = L SMS = S as an explicit linear combination of Slater determinants * then calculating matrix elements of e2/rij, because there are many nonzero offdiagonal matrix elements between Slater determinants in the same ML,MS box.
2 2S+1

Here is the final result for the energies of all (nf)

L terms: Bare nucleus hydrogenic orbital energy or partly shielded by filled shells.

E = E0 +E1 +E2 E
(0 )

( )

( )

( ) (0 )

= sum of orbital energies from h

Z 2 n2

= n

2 ( ) H3 E1 = e + 1 SO 2 rij 123 next lecture ready now

(2 )

= (intraconfigurational spin - orbit) + interconfigurational e 2 rij

CI shielded by all filled subshells
1

For nf2

shielded by same subshell

+ 25 F2 nf 25 F2 30 F2 10 F2 + 19 F2 + 45 F2 + 60 F2

2 nf + F0 (nf 2 )

+ 9 F4 nf 51 F4 + 97 F4 33 F4 99 F4 + 33 F4 + 198 F4

+ F6 nf

H
G

2 nf
2 nf

F0

13 F6 + 78 F6 286 F6 + 715 F6 1287 F6 + 1716 F6

+ F0 + + + +

F
D

2 nf 2 nf 2 nf 2 nf

F0 F0 F0 F0

3 1

P S

shielded-core configurational energy

intraconfiguration L-S term splittings

(there is NO center of Gravity Rule for degeneracy weighted L-S terms)

updated September 19,

5.73 Lecture #34

34 - 10

Now it is easy to show that all Fks are > 0 and Fk >> Fk+2 etc. (roughly factor of 10 per step in k) From this we get an empirical rule Lowest E of all LS terms is the one with * * MAXIMUM S of those with Maximum S, lowest is the one with MAXIMUM L

These are Hunds first and second (of three) rules.

Note also that Hunds rules do nothing about predicting the energy order of L-S terms except for the identity of the single, lowest energy L-S term.

updated September 19,

5.73 Lecture #34

Nonlecture

34 - 11

There are several interesting problems also solved by this e2/rij formalism. 1. Energy splittings between and Slater determinantal characters of two or more L,S terms of the same L and S that belong to the same L,S configuration
e.g. d 3 two 2D terms see pages 47 - 50 of Golding for 2 2 secular determinant for 2D of d 3

2. matrix elements of e2/rij between sameL,S terms that belong to two different configurations
e.g. nd 2
1

S,3P,1D,3F,1G

1S,3P,1D,3F,1G ndn d 3 1 3 1 3 no Pauli restrictions S, P, D, F, G

so there will be

S ~1S P ~ 3P F ~ 3F D~1D

3 1

3 1

G ~1G

interconfigurational CIs, and each of these 5 interaction matrix elements will NOT be of the same magnitude.

updated September 19,

5.73 Lecture #35

35 - 1

SPIN - ORBIT: Many-Electron (N, L, S) Single-Orbital nl Coupling Constants

LAST TIME:

e2 rij
i j

death of orbital picture expansion of 1/rij: multipoles, integrals over AOs in nucleus-centered coordinates SELECTION RULES: orbital and many-e basis sets Gaunt Coefficients: a k , b k , c k [products of 3-j] k k Slater-Condon ( F , G ) ( Fk , Gk ) Sum Rule Method - avoid necessity to derive: *eigenvectors * off diagonal elements in Slater basis Hunds 1st and 2nd Rules

TODAY:
A. General Importance of spin - orbit term HSO = a ( ri )l i s i
i

1 e operator

B. C. D.

Trick: replace 1 - e operator by more convenient (N , L , S)L S for S = 0, L = 0 special case Pattern: Land Interval Rule (Patterns are for breaking! Information about "dark" states) HSO matrix elements in Slater Determinantal Basis Set * another operator replacement * Fundamental control parameter: n , l - scaling of nl

* nl (N , L , S) * off - diagonal spin - orbit matrix elements: L 0, S 0, J = 0.

next time Hunds 3rd Rule and Lande gJfactors

updated September 19,

5.73 Lecture #35

A. Importance of spin-orbit

35 - 2

1. HSO produces diagnostically significant fine structure CONFIGURATIONAL ASSIGNMENTS L,S term assignments PATTERNS:* # components * sign of pattern (largest splitting on top or bottom) * statistical weight (2J + 1) of lowest vs. highest energy component * overall magnitude of splitting 2. for heavy atoms, HSO responsible for such large splittings and offdiagonal interactions that L-S terms vanish, S selection rules are violated. Inter-System Crossing (ISC). Need to deperturb to recover Fk, Gk parameters which should vary smoothly from atom to atom (isoelectronic series) (shielding rules) 3. spin-forbidden transitions provide energy linkages between manifolds of states with different values of S. InterSystem Crossing (ISC) 4. non-textbook Zeeman tuning coefficients (clues about unobserved dark states) Atoms, Molecules, Quantum Dots, solids: in an electronic transition, light acts on single e and operates exclusively on spatial part of spin-flips are forbidden except when HSO mixes states of different S forbidden transitions borrow intensity from allowed transitions. In time-domain: a short pulse prepares, at t = 0, a non-eigenstate that is a pure S = 0 excitation (and l = 1) basis state. The pluck! Language: the name of each eigenstate is based on dominant (i.e., nominal) character. It is the name of the dominant basis states. Use of same name for both eigenstate and basis state is a source of confusion.

updated September 19,

5.73 Lecture #35

B. Operator Replacement for HSO
H SO = a(ri )l i si
i

35 - 3
a one - e operator

Wigner-Eckart Theorem for vector operator operator replacement for special case of J = 0 matrix elements.

JM J A JM J = J A J JM J J JM J
J = 0
proportionality constant

matrix element of corresponding component of J

CTDL p. 1054 use projection theorem: V = J V J. J2 Especially useful when V is an angular momentum that is included in J. useful trick for HSO
a(ri )l i si vector with respect to L vector with respect to S

Special case of L = 0, S = 0 matrix elements in NLM LSM S basis set

configuration label

H SO = a(ri )l i si ( N , L, S )L S
i

operator replacement!

* *

( N , L, S ) L a(ri )l i L S si S
i

a different spin-orbit coupling constant for EACH L-S term of the N configuration

* convenient because it is easy to evaluate matrix elements of LS without having to resort to Slater determinatal basis set
ASIDE: a ri l i and s i are both vectors with respect to J, thus HSO is scalar with respect to J, hence matrix elements in NJLSM J basis set are J = 0, M J = 0, and independent of MJ .

( )

CAUTION: LS seems to imply S = 0 selection rule, but we assumed S = 0 in deriving the simplified form of HSO: LS
updated September 19,

5.73 Lecture #35

35 - 4

C. Land Interval Rule (useful for recognizing and assigning an isolated L-S term) J=L+S
J 2 = L2 + S2 + 2 L S J 2 L2 S2 LS = 2

NJLSM J H

SO

NJLSM J =

( N , L, S )[ J ( J + 1) L( L + 1) S( S + 1)]
can be positive, zero, or negative

So, within an L-S term, HSO causes splitting into 2S+1 (or 2L+1 if S > L) components. J

( h2 2) ( N , L, S)[ J ( J + 1) ( J 1) J ] = h ( N , L, S) J
Land interval rule J1

P2

2
P 1

1
3

spacing is to larger of 2 Js largest spacing locates largest J large spacing on top: regular L-S term bottom: inverted L-S term [Convince yourself that you should be able to tell 3D from 3P] * * *

P0

regular 3P [2 intervals: 2:1]

Easy to show that degeneracy weighted average spin-orbit energy of a multiplet = 0 (easiest to show from SO trace of H in LMLSMS basis, followed by trace invariance). The interval rule plus the number of Jcomponents of a multiplet determine the values of L and S. [4P 5:3, 2 intervals; 2P 1 interval, 4 D 7:5:3, 3 intervals]

L+S

J = LS

(2 J + 1)EJ = 0

updated September 19,

5.73 Lecture #35

D. Matrix Elements of HSO in Slater Determinantal Basis Set GOALS: * * S 0 matrix elements, L 0 matrix elements relationships between (N,L,S) and nl
L-S term orbital

35 - 5

no interconfigurational HSO matrix elements

[except n

~ nl (n n)

3 /2

(Rydberg scaling rule.)

NONLECTURE: alternative operator replacement for HSO that is appropriate for orbital matrix elements

HSO =

a ri

( )

li

si

l l replace a(ri ) i by nl i using completeness:

nl msms HSO nlm sms =

l l

nl msms a(ri ) nl msms nl msmsl i s i nlm sms

l l l l

a(ri ) l
i

a(ri )

is scalar with respect to si ms = ms and ms independent is scalar with respect to l i l = l , ml = ml , ml independent can' t change l in lml l = l

li

si does not operate on radial part of n = n

thus n l msms HSO nlm sms =

l l

l l

n l a ri nl

( )

msm l s lm sms
l l l

n l a ri nl = (nn)

( )

3 /2

Rydberg scaling for inner part of orbital

n = n
l

3 /2

n 3

so, for n = n ,

nl a ( ri ) nl = n = n 3
l

updated September 19,

5.73 Lecture #35

35 - 6

This reduction of HSO shows that, for atoms, HSO acts exclusively within a configuration except for interconfigurational matrix elements where the two configurations differ by a single spin-orbital of the same value of l: nl nl same l Examples A is a single Slater determinant

A HSO A =

k spin-orbitals

u k a ( rk )l k ss k u k
lk

= n k lk
k

m k s k ms k l s l k m k s k ms k
l l

= h2 n k k m k ms k
l l

diagonal element picks out lzsz

k spin-orbitals

if A is also an eigenfunction of L2 , L z , S2 , and S z then NLML SMS HSO NLML SMS = (N , L , S) h2 ML MS (N , L , S) =

n k k m k ms k
l l

ML MS

The matrix element is evaluated 2 ways in order to reduce a many-e spin-orbit coupling constant ((N,L,S)) to a sum of one-e orbital coupling constants (nl)!

updated September 19,

5.73 Lecture #35

Example 1. nf2 uncoupled

35 - 7

nf 2

(nf

2 3

( ) ( ) ) = [3 1 / 2 + 2 1 / 2 ] = ,H
nf

M L = 5 M S = 1 = 3 2 5 1
nf

Single Slater determinant!

(fill in the steps!) Example 2. nf2 coupled

nf 2

H6

MJ = 6
Land :

= 3 2 nf 2 3H6 M J = 6 HSO nf 2 3H6 6 =

h
2

= h2 5(nf 2 ,3H )

[ J ( J + 1) L( L + 1) S(S + 1)](nf 2 ,3H )

67 56 12 from many-e form from orbital form

= h2 nf [3(1 / 2) + 2(1 / 2)] (nf 2 ,3H ) = nf 2

Example 3. (nf2,3F) Evaluate 2 ways:
3

F is never a single Slater determinant for any value of (ML,MS)

a. Obtain explicit linear combination of Slater determinants using ladders and orthogonality or using L2 , S2 to get nf 2 3 F M L = 3 MS = 1 [laborious]. b. Slater sum rule method [simple].

M L = 3, MS = 1 box: 3 0 , 21 3 0 + 21 = E( 3 H M L = 3, MS = 1) + E( 3 F M L = 3, MS = 1) 3 3 0 = 3 0 HSO 3 0 = h2 nf + 0 2

updated September 19,

5.73 Lecture #35

21 = 21 HSO 21 =
h
2

35 - 8
1 nf 1 + 2

trace of M L = 3, M S = 1 box is 3h2 nf

E( 3 H M L = 3 MS = 1) =
3

H M L = 3 MS = 1 HSO 3 H 3 1 = (nf 2 ,3H )h2 3 1

but already showed nf 2 ,3 H = nf 2

E( 3 H M L = 3, MS = 1) = h2 nf (3 2)
E( 3 F ML = 3 , MS = 1) = 3 h2 nf ( 3 2) h2 nf = ( 3 2) h2 nf =
3

F 3 1 HSO 3 F 3 1 = ( nf 2 ,3 F)( 3 1) h2

1 ( nf 2 ,3 F) = nf 2
1 2 2 3 actually would find, for nf , nf , L = nf for all L 2 [not true for all configurations]

We are not done. There are some off-diagonal matrix elements between the L-S-J terms of the same configuration. nf 2 ,1 I , 3 H , 1G, 3 F , 1 D, 3 P, 1S,

6 5 4 4 4 3 2 2 2 1 0 0

HSO is diagonal in J must diagonalize 2 3 3 and 2 2 2 matrices.

updated September 19,

5.73 Lecture #35

set up J = 6 matrix because it is simplest
1 3 1

35 - 9

I6 6 = 3 3

H 6 6 = 3 2

I6 6 HSO 3 H 6 6 = 3 3 HSO 3 2

Mismatch is in 2nd spin-orbital. Needs 1/2 l+s operator to give nonzero spin-orbital matrix element.

= 3 =
h
2

1 l + s 2 nf 2
1 /2 1[ 3 4 2 3] = 2

nf

nf 3 2

1 /2

zero for all singlet states

HSO 6 J=

I6 0 =3 1 /2 H 6 (3 / 2)
1
SO 3

(3 / 2)1/2 h2
5/2
h
2

nf

H6 6 H

H6 6 = =

[J (J + 1) L(L + 1) S(S + 1)] (nf 2 5 (nf H ) = 5/2

2 2 3

, 3H

nf

for more complex configurations such as (nl)a(nl)b nl and nl: two parameters needed for the two open subshell orbitals. But can use the value of nl determined from some other configuration: e.g. 3d from 3d64s2 can be used to predict the 3d part of HSO in 3d6 4s4p. Unexpected inter-relationships between superficially unrelated observables. Small number of control parameters. Hunds 3rd Rule: lowest energy J of lowest energy L- S term is J = L S if subshell is less than 1 / 2 full, is J = L + S if subshell is more than 1 / 2 full, and J = S because L = 0 for half filled
subshell. Sign of ( N , L , S ) as diagnostic!

NEXT TIME!
updated September 19,

5.73 Lecture #36

Read CTDL, pp. 1156-1178 LAST TIME:
HSO = a(ri )l i si
i

36 - 1

( N , L, S )L S n l i si
l

(one for each L - S term) (one for entire configuration)

Land interval rule (assignment!)

(N , L, S ) n
l

examples

evaluate matrix elements in Slater determinantal basis and in many-e |NJLSMJ or |NLMLSMS basis

off - diagonal ( J = 0) intraconfigurational HSO matrix elements: 1 SO 3 e.g. I6 H H6 = ? See notes [page 35 - 9]!

TODAY:

1. electrons vs. holesa shortcut: e 2 rij vs. HSO (holes are a convenience in spectra of isolated atoms and molecules, but they are an essential part of the interpretive picture for solids) 2. Hunds 3rd rule 3. Zeeman effect: Land g-factor formula via W-E Theorem (done previously by projection theorem) 4. Matrix elements of HZeeman in Slater determinantal basis set. No difference between electron and hole as far as Zeeman effect is concerned.

NEXT TIME:

e in solids (CTDL, pages 1156-1168)

updated September 19,

5.73 Lecture #36

1. relationship between configurations with N e vs. N holes
subshell 1 / 2 full subshell # e

36 - 2

(nl)N
s 1 p d f 3 5 7

for p5 is it necessary to consider all 5 e?

e.g. 1 10 0 1 = np5 2 P ML = 1, MS = 1 / 2 (1 is the unoccupied spin - orbital. It is the "hole") HSO np5 2 P ML = 1, MS = 1 / 2 = np
i

l iz s iz

1 10 0 1 1 5e 2

1 1 = h2 np + (0 0) 2 2

so expectation value of HSO :

but for single e (with the same ML, MS as the five e)

5e

1 = np h2 2

HSO np1 2 P ML = 1, MS = 1 / 2 = npl z s z 1 1e 1 = + np h2 2

is the sign flip just a coincidence? NO! TRICK: Hole is exactly equivalent to e (for identical LMLSMS or JLSMJ) except that the sign of its charge is reversed. * no effect on e2/rij because 2 interacting particles have charge of the same sign (either both e or both hole), so e2/rij is always a repulsive interaction. [What happens for f13p? Certainly different from fp!] reverse sign for HSO because HSO is a relativistic electrostatic interaction between e and nucleus (+ charge). Replacing e by h+ and leaving the sign on the nucleus the same reverses the sign of HSO!

updated September 19,

5.73 Lecture #36

36 - 3
d 9 etc. d8 d7 d6

p1 p5 p2 p4

d1 d2 d3 d4

pretend that holes are e, Slater determinants describe spin-orbitals occupied by holes. * all Fk, Gk, nl remain positive (repulsions)

* all e2/rij energy level patterns are unaffected * all (N,L,S) reverse sign

Look at Tinkham 6-2, page 187 figure.

nd vs. (N,L,S) for lowest L-S term of (3d)N configuration

sign change, too rapid evolution with Z
3 Z eff periodicity, isoelectronic series, aufbau too

INSIGHT regularization of trends EXTRAPOLATION ASSIGNMENT LABOR SAVING! Z Z+1 Zeff Zeff + 1 0.5 : shielding Burns Rules. G. Burns, J. C. P. 41, 1561 (1964).

Shielding systematics:

updated September 19,

5.73 Lecture #36

36 - 4

updated September 19,

5.73 Lecture #36

2. Hunds Third Rule

36 - 5

Consider only MAX-S, MAX-L L-S term, which Hunds 1st and 2nd rules identify as the lowest lying within the (nl)N configuration This L-S term will always be a single Slater determinant for the ML = LMAX, MS = SMAX component

L MAX , M L = L MAX , SMAX , MS = SMAX =

(as many spins as possible) diagonal element of HSO

(l - 1)

( (( n )
l

( nl) , L MAX , S MAX L MAX S MAX = nl ml i msi

N i N

, L MAX , S MAX

) )=

ML

MS

ml i msi
nl

L MAX S MAX

S MAX ?

shell less than 1 / 2 full, N < 2l + 1, all spins are S = N / 2

LMAX?

if all spins are , maximize ML by putting 1e into each ml starting at ml= l and working downward.

M L MAX = l + (l 1)4 (l N +3 = N[l (N 1) 2] 1 1444 +24444 )

N terms in sum get l from each term in sum

all spins

nl N L MAX , S MAX = nl

1 1 ml i ML 2 2 = nl L MAX ( N / 2) L MAX N 2

LMAX

SMAX

= nl / N

WHICH OR IMPLIES

nl 2S MAX

updated September 19,

5.73 Lecture #36

Shell 1/2 full
N = 2l + 1, all spins ,

36 - 6

mi =0
l

S = N 2, L = 0 lowest L - S term is
2S+1

LJ = N + 1SN

(single J for all L = 0 terms) - no fine structure

Shell more than 1/2 full SMAX?

2l + 1 spins MS = SMAX N (2l + 1) spins 1 ( 2l + 1) N (2l + 1) = 2l + 1 N 2 2 = 2l + 1 N 2

]]

LMAX?

for the 2l + 1 spins

mi =0
l

for the N (2l + 1) spins,

li

= l + (l 1) + = M L = L MAX

spins

spins LMAX

nl N L MAX , S MAX

1 1 nl ml i ml i 2 2 ( ) ( ) = L MAX S MAX = L MAX (2l + 1) N 2

nl ( 1 2)L MAX

nl 2 S MAX

2(2l + 1) N
# of holes

nl

Summary for lowest energy L - S term: ** ** (nl N , L MAX ,SMAX ) > 0 for less than 1 / 2 full, = 0 for 1 / 2 full, < 0 for more than 1 / 2 full (nl N , L MAX ,SMAX ) = # # n of e + of h
l

updated September 19,

5.73 Lecture #36

36 - 7

Hunds third rule: ONLY FOR LOWEST ENERGY L-S term, lowest J component is
J = LS J=S J = L+S for N < 2l + 1 regular N = 2l + 1 no fine structure N > 2l + 1 inverted

Assignments:

sign of (NLS) # of J components extreme J values (recognize via interval rule) magnitude of nl # of MJ components Zeeman tuning rates

3. Zeeman effect in many-e atoms

HZeeman = 0

)(L

+ 2S z Bz

1.399613 MHz/Gauss Bohr magneton (Used previously) remember that HZeeman is awkward in JMJLS basis set WE Theorem trick to simplify HZeeman: consider only matrix elements diagonal in J [There are also nonzero matrix elements of HZeeman off-diagonal in J.]
HSO and e2 r are strictly diagonal in J . Since HZeeman has sum of 2 vectors with respect to ij J, W - E Theorem says it can have J = 0, 1 matrix elements. When we evaluated matrix elements of Lz and S z in JM J LS the hard way, we saw that there were nonzero J = 1 matrix elements.

Our special case J = 0 is useful as long as J HZeeman J << EJ0 EJ0

Bz (This fails at high Bz when (nLS) is small.)
updated September 19,

( )

( )

5.73 Lecture #36

36 - 8

for J = 0 matrix elements, replace both L z and Sz by J z JM LS L JMLS = JLS L JLS JM LS J JMLS JM LS S JMLS = JLS S JLS JM LS J JMLS but J = L + S. Add the 2 equations J =( L + S ) J 1442443
(1 )
=1

()

[This trick is equivalent to, but not as elegant as, the projection Theorem.]

H Zeeman =

0
h

)J (1 24z + 2J z Bz = 1 3 { 4 Lz 2 Sz part part

0
h

Bz (1 + )J z !

Trick to evaluate : L2 = (J S) = J2 + S2 2J S
2

diagonal JM J LS matrix element of both sides

h
2

L (L + 1) = h 2 J( J + 1) + h 2S(S + 1) 2 JMLS J S JMLS **

completeness: J has J = 0 selection rule, L has L = 0, S has S = 0, J S is scalar with respect to J, M = 0

JMLS J S JMLS =

J M L S

JMLS J J M L S J M L S S JMLS
J

= JMLS J JMLS JMLS S JMLS = JMLS J JMLS = J J + 1 h2

2

Plug this into the ** equation above and rearrange: = J( J + 1) + S(S + 1) L (L + 1) 2 J( J + 1)

updated September 19,

5.73 Lecture #36

from from

36 - 9

H Zeeman

2S z 7 6L z8 } = 0 Bz M J (1 ) + 2

= 0 Bz M J [1 2 ] 1+ 3
gJ

Land g-value gJ 1 + = 1 + J ( J + 1) + S( S + 1) L( L + 1) 2 J ( J + 1)

1 dE 1 = gJ * gJ is Zeeman tuning coefficient = 0 dB z M J * equally spaced MJ components * excellent diagnostic for different L,S of same J r r gJ is large when L and S are parallel (i.e. since J = L + S , r r parallel L,S at constant J means smallest possible L in order to

have largest possible S)

L

r r gJ small when L,S are antiparallel e.g. J =3 : L = 0, S = 3

gJ L = 3, S = 1 :

2.000 J =4 (parallel) 1.250

L = 1, S = 2 1.667 J=3 1.1667

L = 2, S = 1 L = 3, S = 0 1.333 1.000 J=2 (antiparallel) 0.667

* gJ decreases at constant J when S is replaced by L. * gJ decreases at constant L and S as J decreases from L+S to |LS|. How to determine J: * apply B-field and count MJ components (constant splittings in upper and in lower L-S term) measure gJ (Quantum Beats) polarization dependent Zeeman splitting pattern: MJ = 0 for z polarized, MJ = 1 for x or y polarized, MJ = +1 or 1 for circularly polarized

* *

updated September 19,

5.73 Lecture #36

36 - 10

Compare direct evaluation of Zeeman matrix element to gJ determined independently. Matrix Elements of HZeeman in Slater determinantal basis set?

e.g.

f 2 3H 6 M J = 6 = 3 2

HZeeman = 0

)B (
z i

l iz

+ 2s iz

3 2 H Zeeman 3 2 = ( 0 B z )[( 3 + 1) + ( 2 + 1)] = 7 0 B z Now compare with gJ equation:

3

H6 6 H Zeeman 3 H6 6 = ( 0 B z ) gJM J
6 7 + 1 2 5 6 1 7 = 1+ = 2 6 7 6 6
0 0

gJ = 1 + = 0 Bz

) 7 6 = 7 B
6

agrees!

updated September 19,

5.73 Lecture #36

Hole vs. e for Zeeman effect.

36 - 11

What about a single hole state? Does Zeeman effect reverse sign?

E Zeeman f 13 2 F7/2 7 / 2 = ( 0 B z )[(0 + 7 ) + ( 3 6)] = 4 0 B z

-spins 7e -spins 6e

f 13 2 F7/2 7 / 2 = 3 3 3 2 same ML , MS 1 2 f F7/2 7 / 2 = 3

E Zeeman f 1 2 F7/2 7 / 2 = ( 0 B z )[ 3 + 1] = 4 0 B z same as f 13

no sign change for Zeeman for e vs. h+. WHY?

updated September 19,

5.73 Lecture #37 Infinite 1-D Lattice

CTDL, pages 1156-1168 LAST TIME:
hole ( h + ) vs. e configurations:
l
N

37 - 1

l2 2

+ 1) N

for N > 2l + 1

( NLS ) ( NLS )
N

e rij unchanged

[
S
2 l +1 2

nl

unchanged

Hunds 3rd Rule (Lowest L - S term of l only) N < 2l + 1 N = 2l + 1 N > 2l + 1 E MIN for J = L S
( 2 l +1) +1
J=

regular S state: no fine structure inverted

E MIN for J = L + S

Zeeman Effect Wigner-Eckart Theorem used to define gJ EZeeman = 0 M J g J Bz

gJ = 1 + J (J + 1) + S(S + 1) L( L + 1)

equal spacings gJ as L,S,J diagnostic

Confirm by H TODAY:

Zeeman

2J (J + 1) in Slater determinantal basis

+ 1. H2 as example of localization, delocalization, tunneling

2. secular equation for simplified 1- D lattice 3. eigenvectors by equal probability trick 4. restrict k to k < 6. 7. 8. 9. 5. E ( k) = E0 2 A cos kl
l

: 1st Brillouin Zone

(all of the allowed states?)

Bloch functions k ( x ) = e ikxuk ( x ) wavepackets,motion, group velocity transitions energy bands and intensity profiles conductivity

next lecture

updated September 19, 2003 @ 2:22

5.73 Lecture #37

+ Start with H2 , a lattice with only 2 equivalent sites.

37 - 2

qualitative picture:

atomic energy levels tunneling between identical localized states slow behind big barrier (small splitting) fast behind small barrier (large splitting) levels bands, of width related to tunneling rate
R 2 +R 2

+ H2

) E(n0+1

E(n0 )

2 atoms 2 states

R >> a0

Elowest =

12

and doubly degenerate

for exact degeneracy, can choose any linear combination Localized basis set Delocalized basis set
(0) (0) localized = left or right (0) (0) delocalized = 21/2 left right

updated September 19, 2003 @ 2:22

5.73 Lecture #37

If initially in localized state, tunneling rate depends on * height relative to E(0) of barrier n * width of barrier
(0) (0) * size of overlap between exponential tails of left and right

37 - 3

clear that tunneling rate (i.e. splitting) increases * as n at constant R (internuclear separation) * as R at constant n E 0 E1 double degeneracy at R R is tunneling splittinggets larger as R

N ATOMS ALONG A STRAIGHT LINE

wide narrow

N atoms, N states

each electronic state of isolated atom becomes band of states for lattice. Energy width of each band increases as the principal q.n. increases because atomic states require more room: rn a0n2. Tunneling gets faster. Greater sensitivity to world outside one atom.

updated September 19, 2003 @ 2:22

5.73 Lecture #37

37 - 4

Simplified Model for 1Dimensional Lattice: basis for qualitative insights and early time predictions. 1. Each ion, called q, has one bound state, q at E0 = qHq [diagonal element of H] (actually 2 spin-orbitals)

2. permit orbitals only on adjacent ions to interact [simplifying assumption] like Hckel theory. 3. symmetry: all ions are equally spaced, xq+1 xq = l, and all adjacent-orbital interaction matrix elements are identical [off-diagonal elements of H] qHq+1 (A would increase as l 0) [reasons for A sign choice later.]
E0 A so H = A O O O E0 A A O O O

tridiagonal infinite matrix

since this is infinite, need a trick to diagonalize it. general variational function

q =
get requirements on cq by plugging this into Schrdinger equation

cq q

superposition of AOs at each site

H = E left multiply by q

updated September 19, 2003 @ 2:22

5.73 Lecture #37

qth position LHS (010) picks out q-th row of H
q H = E q
(0 AE 0 A0 )

37 - 5

RHS

E q

] = E [c ]
q

c M = Acq 1 + E0cq Acq + 1 M c+

0 = cq E0 E cq 1 A cq + 1 A

comes from the assumed simple form of model

TRICK: probability of finding e on each lattice site should be the same for all sites (complex amplitudes might differ but probabilities will be constant)

let cq = e

ikql

cq = 1

for all q

This choice of cq is a good guess that is consistent with expectation of equal probabilities on each lattice site. is distance between adjacent atoms q is integer ql is the coordinate of the q-th site: looks like eikx plane wave k is of dimension l1 problem reduces to finding allowed values of k.
l

periodicity of lattice provides the important result that if k is replaced by k , where k = k + 2

l

, the wavefunction does not change (translational symmetry)

updated September 19, 2003 @ 2:22

5.73 Lecture #37

2 ql ikql + i l

37 - 6

cq = e ik q = e
l

= e ikq e i 2 q = cq {
l

=1

Since all distinguishable may be generated by choosing k in the interval

l

k<

, restrict k to this range: called First Brillouin zone.

Return to question about what happens when k is not in 1st Brillouin Zone next time [get another part of the band structure using qualitative perturbation theory rather than a matrix diagonalization calculation].

Plug

cq = e ikql into Schrdinger Equation 0 = cq ( E0 E ) A(cq +1 + cq 1 )

divide by e ikq and rearrange
l

ik q +1 l ik q 1 l 0 = e ikql ( E0 E ) A(e ( ) + e ( ) )

E = E0 A[e ik 243 + e ik ] 14
l l

2 cos kl

This is the condition on E,k that must be satisfied for all eigenfunctions of the Schrdinger equation

E = E0 2 A cos kl

updated September 19, 2003 @ 2:22

5.73 Lecture #37

E varies continuously over finite interval E0 2 A

37 - 7

E(k) E0
E0 2 A

E0 + 2 A

The choice q H q +1 = A leads to minimum E at k = 0.

Are these all of the allowed energy levels that arise from a single orbital at each lattice site? Apparently not see next time. Only half of the states. [One orbital per atom two spin-orbitals per atom. Antisymmetrization gives another separate band.] Could repeat calculation for a higher energy state at each site. Would get a broader band centered at higher energy.

updated September 19, 2003 @ 2:22

5.73 Lecture #37

closer look at spatial form of k ( x ) x k

37 - 8

k ( x) = x k =

q =

e ikq x q 1 3 2
l

q ( x)

goal is to replace infinite sum by single term:

show that:

6 74 4 8 k ( x ) ~ e ikxuk ( x )
periodicity of lattice

This is called a Bloch function

plane wave (Free particle)

begin by requiring that k ( x) =

q =

e ikql q ( x)

Translational symmetry imposes a relationship between q(x) and 0 each q(x) is localized at site q.
q ( x ) = 0 ( x ql ) k( x ) = k( x + l ) =
q =

e 0 ( x ql)
ikql

q =

ql x e ikq 0 (4+ l44) 1 4 2 3

l

shift x by ql to get from site q to site 0

= 0 x q 1
l

( (

))
l

= e ik

ik( q 1) 0 ( x (q 1)l ) e

updated September 19, 2003 @ 2:22

5.73 Lecture #37

re-index sum (replace q1 by q)

37 - 9

k( x + l ) = e k( x )
ikl
translation by l!

This form of k has all of the symmetry properties we will need. This form is sufficient to satisfy the symmetry requirements (boundary conditions). This means, instead of writing k ( x ) as sum over atom - localized q ( x )s, it is possible to write k ( x ) as product of 2 factors

k ( x ) = e ikxuk ( x )
1st factor conveys translational symmetry of a plane wave with wavevector k, 2nd factor builds in translational symmetry of lattice with spacing l. This is a more general expression that incorporates all of the properties of the original definition of k(x) as a sum over localized orbitals.

note that

k ( x + l ) = e ikx e ik uk ( x + l ) = e ik e ikxuk ( x )
l l

u k ( x + l ) = uk ( x )

= e ik k ( x )
l

as required.

Note also that k(x + nl)2 = k(x)2 implies that, as required, e has equal probability of being found on each site.

updated September 19, 2003 @ 2:22

5.73 Lecture #38 Infinite 1-D Lattice II

LAST TIME:
+ H2

38 - 1

localization tunneling: overlap bonding and antibonding orbitals

R vs. a0n2 distance below top of barrier

TIGHT-BINDING (Kronig-Penney) Model (see Baym pp. 116-122)

1-D lattice: 1 state per ion tunneling only between nearest neighbors H matrix E 0 A A E 0 H= 0 O 0 0 q H = E q =
q =

O 0 O 0 O 0 O O

c q q

Variational wavefunction. Minimize E. # of coupled equations

0 = cq (E0 E) A(cq1 + cq+1)

Usually solve for {cq} by setting determinant of coefficients = 0 and solving for E. Cant do this because determinant is . TRICK: expect equal probability of finding e on each lattice site by analogy to plane wave eikx, where probability density is uniform at all sites along x, try

cq = eikql
integer distance between lattice sites

xq = ql

Notice that this is similar to free particle eikx, which seems rather strange because particle is never really free in tight-binding model.
updated September 19, 20032:22 PM

5.73 Lecture #38

cq = 1
2

38 - 2

plug trial form for c q into 0 = c q (E 0 E ) A(c q 1 + c q +1 )

l l l

0 = e ikq (E 0 E ) Ae ikq (e ik + e + ik

divide through by e ikq

0 = E0 E A2 cos kl

E ( k) = E0 2 A cos kl
E(k)
E 0 + 2A

E0

E 0 2A
+
l

E varies continuously over an interval 4A, where A is the adjacent site interaction strength or the tunneling integral What happens when we look at k outside -/l k < /l 1st Brillouin Zone

ck = e ikq k = k + ck = e

2
l

2 i k + ql l

(one additional wavelength per lattice spacing l)

= e ikq e i 2 q = e ikq
l

wavefunction is unchanged! So if k goes outside 1st Brillouin Zone, get same , so get same E nothing new! No point in allowing k to vary more widely than /l k /l.

updated September 19, 20032:22 PM

5.73 Lecture #38

Unanswered Questions:

38 - 3

1.

How many distinct orbitals are there in a band? N-atom periodic array. Periodic Boundary conditions:
longest = lN N possible value of = 2 / k shortest = l 2 2 k in N steps lN l infinite lattice: < k < contains all the states generated
l l

from one state per atom.

2.

What happens at E > E0 + 2A? gap no states allowed next higher state of each atom? free particle if E > work function

3.

Orbitals not states! Two spin-orbitals per orbital. Antisymmetrization. Lowest band: all spins paired. No G term. e e repulsion raises overall E above that of single state of each atom Work function is smaller than single atom IP

4.

How many e does each atom contribute to ? alkali: 1e half full band alkaline earth: 2e full band

updated September 19, 20032:22 PM

5.73 Lecture #38

Now take a closer look at k ( x )
+

38 - 4

( x ) = x k =

x q e ikq 123 q =
l

q ( x ) = 0 ( x ql )
l

q( x )

shift x by ql to get from site q to site 0

k ( x ) = e ikq 0 ( x ql )
q

translate entire by l

k ( x + l ) = e ikq 0 ( x ql + l ) 14 244 4 3 q 0 ( x ( q 1) )
l l

= e ik

e ik q 0 x (q 1)l
(
1

k ( x + l ) = e ik

re-index summation
l

e ikq 0 ( x ql ) = e ik k ( x )
l l

translation of plane wave by l

implies that it is possible to write k ( x ) in more general form

k ( x ) = e ikxuk ( x )
where uk ( x + l ) = uk ( x )

Bloch wave function

perodicity of l

e ikx conveys translational symmetry of plane wave with wavevector k uk ( x ) conveys translational symmetry of lattice with spacing l
updated September 19, 20032:22 PM

5.73 Lecture #38

Localized time dependent state : wavepacket We are gong to build intuitive insight by comparison to free particle. Recall free particle:

38 - 5

( x, t ) = (2 )

1/ 2

) dk g(k ) e i[ kx E ( ktt / h ] { envelope of k centered at k 0

Group velocity: motion of stationary phase point (stationary with respect to k near k0)

d [kx Et / h] k = k 0 dk dE t h x center (t ) = dk k 0 0= v center = E= dE dk dE dk

h

d take dt

1
k0 h

k0

k 2m 2 h k0 = m 1 dE = vG = h dk hk 0 = m k0
use this to understand motion in a periodic lattice
updated September 19, 20032:22 PM

2 2

v center

free particle relationship between vG and hk,m

5.73 Lecture #38

Up to here we have been analyzing the free particle. for 1D lattice

38 - 6

(t ) = (2 ) 1/ 2 dkg(k )e iE( k )t / k {
h

peak at k 0

instead of asking for location of stationary phase point, ask for time dependent overlap of (t) with specific lattice site q.

q (t ) = (2 ) 1/ 2 dk g(k )e i[ kq because k =
q =

E ( k )t / h ]

ikql

q
l

same thing as k ( x) = e ikq x | q q =

We can use either state vector or wavefunction picture.

and q picks out only the e ikq term

l

because p q = pq

recall that x = ql, so we can think of q (t ) as function of x, t Overlap of (t) with particluar lattice site q . (t) moves and sequentially overlaps succesive lattice sites.

meaningful only for regions of x near ql

q (t) = ( x ,t) = (2)

1/2

dkg ( k)e i [ kx E( k )t / ]
h

real part

phase factor

ask for stationary phase factor (near x = 0, l, 2l, ) with respect to k

updated September 19, 20032:22 PM

5.73 Lecture #38

38 - 7

d 0 = [kx E (k )t / h] dk dE t /h x c (t ) = dk k 0 dx c dE vG = = dt dk dE dk 1
k0 h

wavepacket is created centered at k=k0

Up to here, everything is identical for free particle and motion in a periodic lattice.

E (k ) = E 0 2 A cos kl = 2 Al sin k 0l
k0

Now use Ek relationship derived for periodic (tight binding) lattice.

vG =

2 Al
h

sin k 0l

quite different from plane wave result hk VG = 0 m

Note that vG = 0 when k0 is at bottom (k0 = 0) or top (k0= /l) of band.

updated September 19, 20032:22 PM

5.73 Lecture #38 Building of intuition:

38 - 8

* vG A

[as | A | increases it becomes easier to take a step]

(because tunneling rate decreases as l increases) but if A is kept constant as l increases, each step is longer so velocity will be higher

* v G l (but A as l )

* v G = 0 when k 0 = 0 and when k 0 = / l

bottom of band Not a surprise because expect k=0v=0 top of band Big surprise. Use concept of effective mass to rationalize.

e cannot move if it is too close to edges of band

Effective Mass: free

vs.

lattice

k vG = m

h 0

vG =

2 Al sin k 0l
h

2 Al 2 hk 0 2 h
near bottom of band

at small k0l sin k0l k0l

compare the terms and identify reciprocal of the coefficient of hk0:

meff =

2 Al 2

at small k 0l

* large interaction strength makes meff small * large l makes meff small (large jumps)

Next: How do we show that meff increases to at band edges (k = / l)?

updated September 19, 20032:22 PM

5.73 Lecture #38

38 - 9
l

When k 0 is near k0 = l

sin k 0l = sin

l l l

2 Al 2 Al v G = hk 0 2 sin k 0l hk 0 2 l h k0 h k0 2 h k0 meff = as 0 2 Al 2

Full band: no e transport. 1 / 2 Full band: meff 2 h2 = 2 Al 2

(slightly heavier than at bottom of band)

Alternative approach to meff:

E = p 2 / 2m d 2E 2 dp
1

for free particle Use this to define meff

=m

E(k) = E 0 2 A cos kl E (p) = E 0 2 A cos(pl / h) d 2E = (2 Al 2 / h 2 ) cos kl dp 2

h
2

cos kl = 1 1 (kl) 2 + 2

at small kl

meff =

2 Al 2

updated September 19, 20032:22 PM

5.73 Lecture 39

39 - 1

One Dimensional Lattice: Weak Coupling Limit

See Baym Lectures on Quantum Mechanics pages 237-241. Each atom in lattice represented as a 1-D V(x) that could bind an unspecified number of electronic states. Lattice could consist of two or more different types of atoms. Periodic structure: repeated for each unit cell, of length l. Consider a finite lattice (N atoms) but impose periodic (head-to-tail) boundary condition.

L = Nl Each unit cell, eq: A+ B A

Vi(x) i-th unit cell

This is an infinitely repeated finite interval: Fourier Series

V( x) =

n =

e iKnx Vn

K=

2
l

reciprocal lattice vector

updated September 19, 2003

5.73 Lecture 39

39 - 2

Vn is the (possibly complex) Fourier coefficient of the part of V(x) that looks like a free particle state with wave-vector Kn (momentum hKn). Note that Kn is larger than the largest k (shortest ) free particle state that can be supported by a lattice of spacing l.
Kn = n 2
l

first Brillouin Zone for k

l

We will see that the lattice is able to exchange momentum in quanta of hnK with the r free particle. In 3-D, K is a vector. To solve for the effect of V(x) on a free particle, we use perturbation theory.

1.

Define basis set.

2 p2 h d2 H = = 2m 2m dx 2 V( 0 ) = cons tan t ( 0)

(k0 ) = L1/ 2e ikx E

( 0) k

k2 = 2m
h

2.

H =
(1)

n =

e iKn Vn

Matrix elements: H
H
(1) k k

1 L = dx L 0 n

= dx[L1/ 2e ik x ] e iKnx Vn [L1/ 2e ikx ] 0 n ( ) e ix k + Kn k Vn

(1) k k L

integral = 0 if k + Kn k 0 k = k + Kn

H(k1)k =

1 L Vn k ,k + Kn = Vn k ,k + Kn L n n
updated September 19, 2003

5.73 Lecture 39
1 Must be careful about H(kk) (relative to H(k1)k )

39 - 3

1 L ( ) H = dx e ix k + Kn + k Vn = Vn k ',k Kn L 0 n n 1 but Hermitian H requires H(kk) = H(k1)* k

(1) kk

* Vn k ',k Kn = Vn k ',k + Kn n n * true if Vn = V n

So now that we have the matrix elements of H(0) and H(1), the problem is essentially solved. All that remains is to plug into perturbation theory and arrange the results.

3.

( 0) (1) Solve for k = k + k

(k0 ) = L1/ 2e ikx =L

(1) k (1) k 1/ 2 ik x 1 H(kk) e ik x 1/ 2 Vn k , k Kn e ( means k k ) ( 0) ( 0) = L ( 0) ( 0) n E n E k E k Kn k Ek

=L

1/ 2

Vn e i( k Kn )x ( 0) ( 0) n E k E k Kn
(1)* k

=L

1/ 2

* i( k Kn )x Vn e ( 0) ( 0) n E k E k Kn

* Vn = V n

(1)* k

=L

1/ 2

Vn e i( k + Kn )x V n e i( k Kn )x 1/ 2 = L ( 0) ( 0) ( 0) ( 0) n E n E k E k Kn k E k + Kn

But n is just a dummy index, so replace n by n.

updated September 19, 2003

5.73 Lecture 39
4. Use k and * to compute E k = E (k0 ) + E (k1) + E (k2 ) . k

39 - 4

Rather than use the usual formula for E(2), go back to the n formulation of perturbation theory.

E k = 0E (k0 ) + 1E (k1) + 2E (k2 ) = k 0H( 0 ) + 1H(1) k

Retain terms only through 2

Vn e i( k + Kn )x h 2 d 2 1 L ikx iKmx + Vm e E k = dx e + ( 0) n E m E (k0+)Kn 2m dx 2 L 0 k ikx Vn e i( k Kn )x e + ( 0 ) n E E (k0)Kn k

1 h2 ( 2 ) 0 h2k 2 = k L = 2m L 2m
0 1

E (k0 ) E (k0 )

d 2 ikx recall 2 e = k 2e ikx dx

h2 d 2 1 iKmx ikx ikx = dx e e Vme + 2 terms involving 2m dx2 L m

1st term, only m = 0 term in sum gives nonzero integral. 2nd terms, need n or n = 0 term from sum, but these are excluded by .

1 E = LV0 = 1V0 L
(1) k 1

updated September 19, 2003

5.73 Lecture 39

39 - 5

Vn e i( k Kn )x 1 2 E (k2 ) = dx e ikx Vme iKmx ( 0 ) L E k E (k0)Kn n = m =

ikx Vne i( k + Kn )x iKm + dx ( 0 ) Vme e ( 0) n 0 E k E k + Kn m

1st term 2nd term 0 = k + Km + k Kn, requires m = n 0 = k Kn + Km + k, requires m = n

E E

( 2) k

2 2 Vm Vm 1 2 + dx ( 0 ) = dx ( 0 ) m E m E E (k0+)Km E (k0)Km L k k

( 2) k

Vn2 = 2 ( 0 ) ( 0) m = E k E k + Kn
2

Combine terms for n and n and sum

n =1

E (k0 ) E (k0) Kn =

[k 2m
h

(k Kn) 2 ] =

1 1 + ( 0) E (k0 ) E (k0+)Kn E k E k Kn E
( 2) k

Kn [Kn 2k ] 2m 4m 1 = 2 2 2 2 h K n 4k
h

8m
h
2

n =1

Vn2 K 2n 2 4 k 2
updated September 19, 2003

5.73 Lecture 39
But there are many zeroes in this denominator as n goes 0 . Must use degenerate perturbation theory for each small denominator.

39 - 6

E k Recall V

2 V E k + E k E k E k 2 +V E = E k 2 2

1/ 2

Vn2 k 8m Ek = + V0 + 2 2 2 2 h n =1 K n 4 k 2m
h
2 2

Kn 2 n = n= except n = 0 2 2l l at k = 0, there are no nearby zeroes zeroes at k =

dE k dk

k =0

k m

(minimum at k = 0)

d 2E k h (positive curvature) = 2 dk k = 0 m just like free particle!

K , there are zeroes in denominator, so there is a gap in energy of 2 K 2 V1 at k = 2 2 V2 at k = K At k = M 2 Vn at k = nK 2

What does this look like?

updated September 19, 2003

5.73 Lecture 39

39 - 7

h2 2 E = V0 + k 2m
look at text Baym page 240.

updated September 19, 2003

5.73 Lecture 39
But we want to shift each of the segments by integer times K to left or right so that they all fit within the K K k first Brillouin Zone. 2 2 E

39 - 8

2V3

2V2

2V1

V0

-K/2

K/2

k diagram. Curvature gives meff 3 D k - diagram much more information. Tells where to find allowed r r transitions as function of 3- D k vector in reciprocal lattice of lattice vector K. Scattering of free particle off lattice. Conservation of momentum in the sense r r r k final k initial = K.

updated September 19, 2003