Spohn H. Dynamics of Charged Particles and their Radiation Field

Notes and references 243

⊗ 

⊗F

. The net result is



= 1 −



k|ϕ(k/λ



|k|



1 +

|k|





−1

−



k|ϕ(k/λ



|k|



1 +

|k|





−1

+ O(e

). (16.110)

For



only one of the two ground states is expanded to order e. Hence

one has a contribution proportional to χ

⊗ , ( A

ϕ2

−1

σ · B

− σ ·

−1

ϕ1

)χ

⊗ 

⊗F

. The net result is



=−



k|ϕ(k/λ



|k|



1 +

|k|





−1

. (16.111)

Adding up (16.109), (16.110), and (16.111), the g-factor to order e

is given by

g = 2



1 +



k|ϕ(k/λ





1 +

|k|





−1



O(e

). (16.112)

In Heaviside–Lorentz units e

= 4πα.Wealso set the sharp cutoff ϕ(k) =

(2π)

−3/2

for |k|≤, ϕ(k) = 0 for |k| >. Then

g = 2



1 +



2π



(1 − (1 + (/2λ

))

−2

) + O(α

)



. (16.113)

Clearly g > 2, as observed experimentally. It is remarkable that g stays bounded

in the limit  →∞and

∞

= 2



1 +



2π



O(α

), (16.114)

which is to be compared with 2



1 + (α/2π)



(α

) from fully relativistic QED.

Evidently the nonrelativistic Pauli–Fierz model overestimates the contribution

from large wave numbers by a factor 8/3. The result (16.114) is satisfactory, since

it nourishes the hope that the Pauli–Fierz model makes reasonable predictions even

when the ultraviolet cutoff  is removed.

Notes and references

Section 16.1

In the old quantum theory classical adiabatic invariants were associated with

good quantum numbers (Ehrenfest 1916). Thus the time-adiabatic theorem was an

244 States of lowest energy: dynamics

important consistency check of the Heisenberg–Schr

odinger quantum mechanics

(Born 1926; Born and Fock 1928). Kato (1958) proves the adiabatic theorem under

the condition that the relevant subspace has ﬁnite dimension and is separated by a

spectral gap. In fact, the theorem holds in much greater generality than explained

in the text. Only a corridor separating the relevant energy band from the rest is

needed. The spectrum inside the band can be arbitrary. The error in (16.8) may be

improved to any order at the expense of a slight tilt of the subspace P(εt)

H,asﬁrst

recognized by Lenard (1959) and further reﬁned by Garrido (1965), Berry (1990),

Joye et al. (1991), Nenciu (1993), and Joye and Pﬁster (1994). We refer also to the

interesting collection of articles by Shapere and Wilczek (1989). Sj

ostrand (1993)

discusses the higher-order corrections from the point of view of pseudodifferential

operators; compare with section 16.4 and Panati et al. (2003a). If H (t) depends

analytically on t, the error becomes e

−1/ε

, which complements the Landau–Zener

formula for almost crossing of eigenvalues (Joye and Pﬁster 1993). If there is no

gap, but a smooth t-dependence as before, the adiabatic theorem still holds (Avron

and Elgart 1999; Bornemann 1998; Teufel 2001). The error depends on the con-

text. It can be as small as in (16.8), but in general it will be larger.

Section 16.2

Our discussion of the space-adiabatic limit ignores technical details on purpose.

They are supplied in Teufel and Spohn (2002), Spohn and Teufel (2001), and

Teufel (2003). Most importantly, since p

< ∞, one needs a local version of

the result explained in the text in the following sense. In the limit ε → 0 the

initial state deﬁnes a classical probability measure ρ

p) on phase space

; compare with section 16.5. ρ

is transported by the classical ﬂow 

with

Hamiltonian (16.22) as ρ

◦ 

−t

.Ifρ

is supported in R

×{p||p| < p

}, then

there is a ﬁrst time t

hit

at which the support of ρ

◦ 

−t

hits the boundary

×{p||p|= p

}. The approximation through an effective Hamiltonian is valid

for times 0 ≤ t <ε

−1

hit

Section 16.3

Weyl quantization, the Moyal product, and matrix-valued symbols are discussed in

Robert (1987, 1998), Dimassi and Sj

ostrand (1999), Martinez (2002), and Panati

et al. (2003a). The Moyal product is introduced in Moyal (1949).

Section 16.4

The methods explained in this section have a rich history with motivations ranging

from singular partial differential equations and Fourier integral operators to the

motion of electrons in solids subject to a small magnetic ﬁeld. Blount (1962a, b, c)

Notes and references 245

develops a similar scheme for computing effective Hamiltonians and applies it to

Bloch electrons and to the Dirac equation. In particular, he computes the second-

order symbol h

.Inthe solid state physics literature his work is a standard ref-

erence, but his method is hardly applied to concrete problems. We refer to the

discussion in Panati et al. (2003b) for an example in the case of magnetic Bloch

bands. Starting from coupled wave equations Littlejohn and Flynn (1991) and Lit-

tlejohn and Weigert (1993) develop the technique of unitary operators close to the

identity on the level of symbols in the case where the principal symbol is a nonde-

generate matrix. They apply their scheme to Born–Oppenheimer-type problems,

where H

(q, p) = p

1l + V (q) with V (q) an n × n matrix. On an abstract level

the Born–Oppenheimer approximation is similar to the Pauli–Fierz model with a

slowly varying external electrostatic potential only. The role of the invariant sub-

space is emphasized by Nenciu (1993). The formal power series for the projector

π(q, p) is constructed by Brummelhuis and Nourrigat (1999) for the Dirac equa-

tion, by Martinez and Sordoni (2002) for Born–Oppenheimer-type Hamiltonians

and in the general matrix-valued case by Nenciu and Sordoni (2001). Our discus-

sion is based on Panati et al. (2003a). The lecture notes by Teufel (2003) give de-

tailed coverage with many examples, including the case of Bloch electrons (Panati

et al. 2003b). There also a more complete listing of the literature can be found.

Section 16.5

There is a vast literature on semiclassical methods, both on the theoretical physics

and on the mathematical side; to mention only a few representatives: Maslov

and Fedoriuk (1981), Gutzwiller (1990), and Robert (1987, 1998). These works

are mostly concerned with the scalar case. An alternative technique is to employ

matrix-valued Wigner functions (G

erard et al. 1997; Spohn 2000b). In this

approach the adiabatic and semiclassical limits are fused, which is conceptually

misleading. Also higher-order corrections are not accessible. An important

example is the Dirac equation which has matrix dimension n = 4 and degeneracy

 = 2 of, for example, the electron subspace. The adiabatic limit yields the BMT

equation of chapter 10, as discussed in Panati et al. (2003a). Blount (1962c)

computes the next-order correction. It seems to be of interest in accelerator

physics (Heinemann and Barber 1999), despite its fairly complicated structure.

Yajima (1992) studies the derivation of the BMT equation using WKB methods,

which are rather difﬁcult to handle because of the necessity to switch coordinate

systems on the Lagrangian manifold.

The classical limit of the free Maxwell ﬁeld with classical sources is regarded as

sort of obvious. An instructive discussion is Thirring (1958) and Sakurai (1986).

Photon counting statistics is covered by Carmichael (1999).

246 States of lowest energy: dynamics

Section 16.6

The gyromagnetic ratio of the electron is the most famous and precise predic-

tion of QED with the current value g

theor

/2 = 1.001 159 652 459 (135) as based

on an eight-loop computation, see Kinoshita and Sapirstein (1984) for a review.

This result compares extraordinarily well with the experimental value g

exp

/2 =

1.001 159 652 193 (4) of van Dyck, Schwinberg and Dehmelt (1986) based on

measurements on a single electron in a Penning trap, see also Brown and Gabrielse

(1986), and Dehmelt (1990). The nonrelativistic theory yields g

non

/2 = 1.0031,

with no cutoffs. The nonperturbative formula (16.108) seems to be novel and is

described in Panati et al. (2002b). A rough approximation is provided by Welton

(1948). Grotch and Kazes (1977) discuss the g-factor for the Pauli–Fierz model

and obtain the second-order result (16.113) through computing energy shifts; com-

pare with section 19.3.5. Surprisingly, they do not stress the obvious point: the

g-factor is not too far off the truth even in the limit  →∞. After all, the mis-

trust in QED up to the early 1940s was based mainly on the results being cutoff-

dependent and diverging as  →∞; see Schweber (1994).

Radiation

The theoretical understanding of the emission of light from atoms is inseparably

linked with the development of quantum mechanics – the ﬁrst glimpse of the full

answer unraveled by P. A. M. Dirac in February 1927. A minimal model for radi-

ation has to consist of at least one atom and the photons. Thus we ﬁx an inﬁnitely

heavy nucleus at the origin, say, and describe the motion of a single electron by

the spinless Pauli–Fierz Hamiltonian

H =

( p − eA

(x))

+ V

ϕcoul

(x) + H

(17.1)

with V

ϕcoul

(x) =−e



ϕ(x

)ϕ(x

)(4π|x + x

− x

−1

, the smeared

Coulomb potential. Besides radiation, (17.1) describes a multitude of physical pro-

cesses of interest. If the electron is free, i.e. far away from the nucleus, photons

scatter off the electron (Compton effect). As the electron approaches the nucleus it

will be scattered under the emission of bremsstrahlung (Rutherford scattering). In

contrast, in this chapter we are interested in processes where the electron remains

tightly bound to the nucleus. Of course, these two worlds are not strictly separated.

The electron might be captured by the nucleus at the expense of radiated energy.

Conversely, the atom may become ionized by hitting it with sufﬁciently energetic

radiation (photoelectric effect). Even in the realm of a bound electron, several

processes should be distinguished. The most basic one is spontaneous emission,

through which the atom in an excited state loses energy and ends up in the radi-

ationless ground state. A photon may be scattered by the atom leaving the atom

behind in either its ground state (elastic Rayleigh scattering) or in an excited state

(inelastic Rayleigh scattering) which is then followed by spontaneous emission.

Both processes will be discussed in separate sections.

Under usual circumstances the wavelength of emitted light is much larger than

the size of an atom. In this case one can ignore the variation of the vector poten-

tial in (17.1) and replace A

(x) by A

(0), the so-called dipole approximation. In

247

248 Radiation

addition we want to restrict the electron Hilbert space to bound states only. Tak-

ing into account the ﬁrst N of them results in an N -level system coupled to the

radiation ﬁeld. We point out that an enormous effort has been invested precisely

to avoid such a mutilation of the Pauli–Fierz Hamiltonian (17.1). Still, in the ﬁrst

round a simpliﬁed version will sufﬁce.

Radiation as discussed here has no classical counterpart. Of course, as ex-

plained, in the context of the Abraham model a charge loses energy through radia-

tion. Its analog would be an extension of the results given in the previous chapter.

There one has to give up m

> 0. Then the spectral gap closes and the strict adia-

batic protection is lost. For example, (16.105) would have a dissipative correction

at the next order associated with a gradual emission of photons. In contrast, for

the radiation processes studied here the emission of photons occurs on the atomic

scale.

17.1 N-level system in the dipole approximation

The dipole approximation reads

H =

( p − eA

(0))

+ V (x) + H

. (17.2)

If in addition we were to choose V to be harmonic, V (x) =

mω

, then

(17.2) is a quadratic Hamiltonian, as can be seen, if on top of the Bose ﬁelds

a(k, λ), a

∗

(k,λ) one introduces the annihilation and creation operators b, b

∗

for

the particle; compare with section 13.7(i). The analysis of this model can be

reduced to a Hamiltonian on the one-particle space

⊕ (L

) ⊗ C

), where

corresponds to the b, b

∗

degrees of freedom. While such an analysis is very

instructive, we stick here to the more realistic Coulomb-type potential. We rewrite

(17.2) as

H =

+ V (x) + H

−

p · A

(0) +

(0)

, (17.3)

drop the A

(0)

term, and expand in the eigenbasis of

+ V (x) up to the

N -th eigenvalue, including multiplicity. This results in

= H

⊗ 1 + 1 ⊗ H

+ λ

Q · A

(0). (17.4)

Here H

and

Q = (

) are symmetric N × N matrices. In our repre-

sentation H

is diagonal with nondegenerate smallest eigenvalue ε

and

Q is

proportional to the dipole moments

=ψ

, pψ

=im(ε

− ε

)ψ

, xψ

, (17.5)

i, j = 1,... ,N , where we used the facts that i[

+ V (x), x] =

p and

17.1 N -level system in the dipole approximation 249

(

+ V (x))ψ

= ε

counting eigenvalues and eigenfunctions including

their multiplicity. We also introduced explicitly the dimensionless small coupling

parameter λ.Ifone follows the conventions of section 13.4, then λ = α

3/2

Note that in the functional integral representation of e

−tH

, H

of (17.4), the

effective action is quadratic with the interaction potential

dip

(t) = λ



k|ϕ|

2ω

−ω|t|

, (17.6)

which decays as t

−2

for large t. Thus (17.4) is marginally infrared divergent.

Generically H

will lose its ground state at strong enough coupling, in contrast

to the full Pauli–Fierz model, and (17.4) can be trusted only at small coupling.

An alternative route to the N -level approximation is ﬁrst to transform to the

x · E

(0) coupling through the unitary transformation

U = e

iex·A

(0)

. (17.7)

Then

∗

pU = p + eA

(0), U

∗

xU = x ,

∗

a(k,λ)U = a(k,λ)+ i(e

(k) · x)eϕ(k)/



2ω(k) (17.8)

and therefore

∗

HU =

+ V (x) +





k|ϕ|



+ H

− ex · E

(0). (17.9)

As before, we expand in the eigenbasis of

+ V (x) up to the N -th eigenvalue.

This results in the Hamiltonian

= H

+ H

+ λQ · E

(17.10)

with the matrix of dipole moments Q

=ψ

, xψ

, E

= E

(0), and λ =−e.

Since now the coupling is to E

(0), the effective action (17.6) gains an extra factor

of ω

and therefore has a decay as t

−4

in accordance with the full model.

For the remainder of the chapter, we take (17.10) as the starting point. The par-

ticular origin of H

and Q is of no importance. We only record that they satisfy

∗

= H

, Q

∗

= Q. H

has the spectrum σ(H

) ⊂ R.Itconsists of the eigen-

values labeled without multiplicity as ε

<ε

< ···<ε

N ≤ N . The corre-

sponding spectral projections are denoted by P

,... ,P

. Their degeneracies are

tr[P

] = m

with m

= 1 and



j=1

= N .Inparticular one has the spectral

representation



j=1

. (17.11)

250 Radiation

17.2 The weak coupling theory

We plan to study the emission of light from atoms. The atom is assumed to have

already been prepared in an excited state and thus the initial state of the coupled

system is of the form ψ ⊗ with the atomic wave function ψ ∈

.Todeter-

mine the radiated ﬁeld one has to understand the long-time asymptotics of the

solution e

−iHt

ψ ⊗ of the time-dependent Schr

odinger equation. For small cou-

pling, which is well satisﬁed physically, the dynamics approximately decouples:

the atom is governed by an autonomous reduced dynamics and the ﬁeld evolves

with the decaying atom as a source term. In this section we will ﬁrst study the

reduced dynamics of the atom in the weak coupling regime with our results to be

supported through a nonperturbative resonance theory in section 17.3. In a follow-

up we discuss the spectral characteristics of the emitted light.

By deﬁnition, the reduced dynamics refers to the reduced state of the atom,

which allows one to determine atomic observables such as the probability of being

in the n-th level at time t.Although by assumption the initial state of the atom is

pure, it will not remain so because of the interaction with the radiation ﬁeld. Thus

it will be more natural to work directly in the set of all density matrices. The initial

state is then of the form ρ ⊗ P



with ρ the atomic density matrix, and P



the

projection onto Fock vacuum. The time evolution is given through

−iH

ρ ⊗ P



= e

−i

ρ ⊗ P



. (17.12)

Here L

W = [H

, W ]isthe Liouvillean as acting on T

⊗ F), the trace class

over

⊗ F

.Todistinguish typographically, L

is written as a slanted symbol,

like other operators, sometimes called superoperators, which act either on

or on

⊗ F), the space of bounded operators on C

⊗ F. Clearly, states evolve

into states, i.e. if W ∈

is positive and normalized, so is e

−iL

W . Sometimes, it

is convenient to think of (17.12) as a Schr

odinger evolution in a Hilbert space. This

can be done by adopting the space

⊗ F) of Hilbert–Schmidt operators with

inner product A|B=tr[ A

∗

B]. In this space the Liouvillean L

is a self-adjoint

operator, which explains our sign convention in front of the commutator. A fur-

ther choice comes from regarding B(

⊗ F) as the space dual to T

⊗ F)

through the duality relation W → tr[AW], W ∈

⊗ F), A ∈ B(C

⊗ F).

Then the dual of L

is −[H

, ·], which generates the Heisenberg evolution of op-

erators.

The reduced dynamics is deﬁned through

ρ = ρ

(t) = tr

−iL

ρ ⊗ P



], (17.13)

where tr

[·] denotes the partial trace over Fock space.

acts on B(C

).Itis

linear, preserves positivity and normalization. In fact, since it originates from a

17.2 The weak coupling theory 251

Hamiltonian dynamics, the even stronger property of complete positivity is satis-

ﬁed. In such generality, T

is intractable. But scales become separated for small

λ into atomic oscillations of the uncoupled dynamics e

−iH

and the weak radia-

tive damping of order λ

(= α

= 1/137

). When viewed on the dissipative scale

the atomic oscillations are very rapid and effectively time-averaged. For small λ

memory effects are negligible and T

becomes a dissipative semigroup, which is

the autonomous dynamics we are looking for.

To write a formal evolution equation for ρ

(t) one employs the Nakajima–

Zwanzig projection operator method. We deﬁne the Liouvilleans L

= [H

, ·]

as acting on B(

) = T

), L

= [H

, ·]asacting on T

(F), and L

int

= [Q ·

, ·]asacting on T

⊗ F).For an arbitrary density matrix W on C

⊗ F

the Nakajima–Zwanzig projection is

PW = (tr

W ) ⊗ P



. (17.14)

Clearly P

= P and

−i

ρ ⊗ P



= ρ

(t) ⊗ P



. (17.15)

Let W (t) = e

−iL

ρ ⊗ P



. Then

PW (t) = PL

W (t) = PL

PW(t) + PL

(1 − P )W (t), (17.16)

(1 − P )W (t) = (1 −P )L

W (t) = (1 − P )L

PW(t)

+ (1 − P )L

(1 − P )W (t). (17.17)

Substituting (17.17) back in (17.16) and using PL

int

P = 0, we obtain

(t) =−iL

(t)

−λ



ds tr

int

(1 − P )e

−i(1−P)L

(1−P)(t−s)

(1 − P )L

int



]ρ

(s),

(17.18)

which is an exact memory-type equation.

As argued traditionally, the memory decays rapidly on the time scale of the

variation of ρ

(t).For small λ one may ignore the interaction and replace L

+ L

in the exponential. In this approximation for small λ

ρ(t) = (−iL

+ λ

)ρ(t) (17.19)

252 Radiation

is obtained as reduced dynamics with

ρ =−



∞

dt tr

int

−i(L

int



]ρ. (17.20)

This argument misses the point that both ρ

(t) and the memory kernel have

oscillatory contributions from e

−iH

.Ingeneral, their products cannot be approx-

imated as in (17.19), (17.20). To subtract the oscillations from the memory kernel

we rewrite (17.18) as an integral equation,

(t) = e

−iL

ρ − λ



dse

−iL

(t−s)



du tr

int

(1 − P )e

−i(1−P )L

(1−P )(s−u)

(1 − P )L

int



]ρ

(u).

(17.21)

After the change of variables v = s − u, one has

(t) = e

−iL

ρ − λ



du e

−iL

(t−u)





t−u

dv e

int

(1−P ])e

−i(1−P )L

(1−P )v

(1−P )L

int



]



(u).

(17.22)

Now in the memory kernel the fast oscillations are properly counterbalanced and

to a good approximation ρ

(t) is governed by

ρ(t) = (−iL

+ λ

K )ρ(t), (17.23)

where

Kρ =−



∞

dt e

int

−i(L

int



]ρ. (17.24)

We state our result as

Theorem 17.1 (Weak coupling quantum master equation). Let

, E

ϕα

−iH

ϕβ



= h

αβ

(t) = δ

αβ

h(t), (17.25)

h(t) =



k|ϕ|

ω(k)e

−iω(k)t

, (17.26)

α, β = 1, 2, 3. If



∞

dt|h(t)|(1 + t)

< ∞ (17.27)

Spohn H. Dynamics of Charged Particles and their Radiation Field

Подождите немного. Документ загружается.