Chapter 6

Time-Dependent Potentials and Light–Matter Interaction

6.1Introduction

Up to now the Hamiltonian was time independent, so the main questions were energy levels and stationary states. Here the Hamiltonian depends on time because an external field drives the atom, and the central questions become: which transitions occur, how fast, and with what probability.

This chapter is the bridge from atomic structure (Chapters 4–5) to the time-domain spectroscopy model used in the final chapters. The logic is deliberately linear:

  1. derive the exact coupled-amplitude equations from the TDSE;

  2. insert the semiclassical dipole interaction and pulse model;

  3. solve a near-resonant two-level system (Rabi oscillation);

  4. derive first-order perturbation theory and Fermi's golden rule;

  5. state clearly where each approximation fails.

If these five steps are clear, the light–matter formalism in the rest of the book is transparent rather than memorized.

Definition 6.1.1Core route and research extension

The first-pass core is Sections 6.2– 6.4 together with the three self-checks before the summary. It asks one question: how does the two-state Hamiltonian of Chapter 1 change when its coupling depends on time? Dressed states, dark states, continuum transitions, and the project are a research extension. They can be postponed without blocking Chapters 7–9, Quantum Chemistry, or Condensed Matter.

6.2The Coupled Amplitude Equations

Let the Hamiltonian split into a solved part and a time-dependent part,

H ˆ ( t ) = H ˆ 0 + V ˆ ( t ) , H ˆ 0 | n ⟩ = E n | n ⟩ , ⟨ m | n ⟩ = δ m n . (6.1)

The eigenstates | n ⟩ form a complete basis, so any state expands in them. We pull out the free-evolution phase explicitly:

| Ψ ( t ) ⟩ = ∑ n c n ( t ) e − i E n t / ℏ | n ⟩ . (6.2)
Theorem 6.2.1Coupled amplitude equations

The coefficients in Eq. (6.2) obey, exactly,

i ℏ c ˙ k ( t ) = ∑ n c n ( t ) V k n ( t ) e i ω k n t (6.3)

where

V k n ( t ) = ⟨ k | V ˆ ( t ) | n ⟩ , ω k n = E k − E n ℏ . (6.4)

Substitute Eq. (6.2) into i ℏ ∂ t | Ψ ⟩ = ( H ˆ 0 + V ˆ ) | Ψ ⟩ . The left-hand side, by the product rule, is

i ℏ ∑ n [ c ˙ n e − i E n t / ℏ + c n ( − i E n ℏ ) e − i E n t / ℏ ] | n ⟩ = ∑ n [ i ℏ c ˙ n + E n c n ] e − i E n t / ℏ | n ⟩ .

The right-hand side is

∑ n c n e − i E n t / ℏ [ E n | n ⟩ + V ˆ | n ⟩ ] .

The terms E n c n appear on both sides and cancel — which is the whole point of extracting the phase in Eq. (6.2). What remains is

∑ n i ℏ c ˙ n e − i E n t / ℏ | n ⟩ = ∑ n c n e − i E n t / ℏ V ˆ | n ⟩ .

Project onto ⟨ k | and use orthonormality on the left:

i ℏ c ˙ k e − i E k t / ℏ = ∑ n c n e − i E n t / ℏ V k n .

Multiplying by e + i E k t / ℏ gives Eq. (6.3).

Three remarks on this result, since it is the skeleton of everything below.

Remark 6.2.1It is exact

No approximation has been made: one partial differential equation has been rewritten as an infinite set of coupled ordinary differential equations. Approximations enter only later, when we truncate the basis or simplify V ˆ ( t ) . That separation is useful: convergence can be tested by adding states and checking how much the observables change.

Definition 6.2.1Time-ordered propagator

For an arbitrary time-dependent Hamiltonian the exact evolution operator is

U ˆ ( t , t 0 ) = 𝒯 exp [ − i ℏ ∫ t 0 t H ˆ ( t ′ ) d t ′ ] . (6.5)

The symbol 𝒯 places later-time Hamiltonians to the left. It may be omitted only when [ H ˆ ( t ) , H ˆ ( t ′ ) ] = 0 for all relevant pairs of times. The RWA introduced below is an approximation to a driven problem; it is not the general exact solution.

For one constant two-state Hamiltonian, H ˆ = a I ˆ + b → ⋅ σ → , the exponential is exact because ( b → ⋅ σ → ) 2 = | b → | 2 I ˆ :

e − i H ˆ t / ℏ = e − i a t / ℏ [ I ˆ cos ( | b → | t ℏ ) − i b → ⋅ σ → | b → | sin ( | b → | t ℏ ) ] . (6.6)

For Chapter 1's matrix ( 0 g g D ) , a = D / 2 and b → = ( g , 0 , − D / 2 ) . This is the exact static benchmark against which any time integrator should be tested.

Remark 6.2.2The phase factor is where resonance lives

The factor e i ω k n t carries the energy difference between the two states. If V ˆ ( t ) oscillates at a frequency near ω k n , the product V k n e i ω k n t contains a slowly varying piece whose time integral accumulates; otherwise the product oscillates rapidly and averages to nearly nothing. That observation is resonance, and it is the mechanism behind every selection rule and every bandwidth argument in the rest of the book.

Remark 6.2.3Probability conservation as a numerical test

P n ( t ) = | c n ( t ) | 2 is the probability of finding the system in | n ⟩ , and unitarity guarantees ∑ n | c n | 2 = 1 for all time. When Eq. (6.3) is integrated numerically, monitoring that sum is the single most useful check available.

6.3How Light Couples to an Atom

We treat the field classically — a given E → ( t ) , not a quantized photon field. This is the semiclassical approximation, excellent for lasers, where the photon number is enormous.

A charge distribution in a uniform field has interaction energy − 𝒅 ˆ ⋅ E → , where for an electron at 𝒓 ˆ the dipole moment operator is 𝒅 ˆ = − e 𝒓 ˆ . Hence

V ˆ ( t ) = − 𝒅 ˆ ⋅ E → ( t ) = e 𝒓 ˆ ⋅ E → ( t ) (6.7)
Definition 6.3.1Dipole approximation

The dipole approximation replaces the field at the electron's position by its value at the nucleus, E → ( r → , t ) → E → ( t ) . It is valid when the wavelength of the radiation greatly exceeds the size of the atom.

Example 6.3.1Checking the dipole approximation

Is Def. 6.3.1 safe for 800 nm infrared light and for 17.6 eV extreme ultraviolet light?

Solution.

The relevant atomic size is ⟨ r ⟩ = 1.5 a 0 = 0.079 nm for a ground state (Chapter 4, Example 4.7.1).

For 800 nm the ratio is 800 / 0.079 ≈ 10 4 : the field varies by one part in ten thousand across the atom. Entirely safe.

For 17.6 eV, λ = h c / E = 1239.8 / 17.6 = 70.4 nm, and the ratio is 70.4 / 0.079 ≈ 900 . Still safe, though by three orders of magnitude rather than four.

The approximation fails only for hard X-rays, where λ approaches a 0 . Note however that a Rydberg state with n = 20 has ⟨ r ⟩ ≈ 600 a 0 = 32 nm, comparable to the XUV wavelength — so “the atom is small compared with the wavelength” is a statement about the state, not just the atom.

Combining Eqs. (6.7) and (6.4), every transition is governed by the dipole matrix element

d → k n = ⟨ k | 𝒅 ˆ | n ⟩ = − e ⟨ k | 𝒓 ˆ | n ⟩ , (6.8)

and for fixed polarization this reduces to the scalar form V k n ( t ) = − d k n E ( t ) .

This is exactly the object analyzed in Chapter 3: 𝒓 ˆ is odd, so d → k n vanishes unless the states differ in parity (Thm. 3.6.1), giving Δ l = ± 1 and Δ m = 0 , ± 1 .

6.3.1What a Pulse Is

The envelope is not a detail. Confining the field in time spreads it in frequency by the same Fourier relation that governed Chapter 1's uncertainty relation, which is why a few-femtosecond pulse drives transitions it was never tuned to — the fact Chapter 11 depends on.
Figure 6.1. The envelope is not a detail. Confining the field in time spreads it in frequency by the same Fourier relation that governed Chapter 1's uncertainty relation, which is why a few-femtosecond pulse drives transitions it was never tuned to — the fact Chapter 11 depends on.
Definition 6.3.2Laser pulse

A pulse is a carrier wave inside an envelope,

E → ( t ) = ϵ ˆ E 0 h ( t ) cos ⁡ ( ω t + ϕ ) , (6.9)

with ϵ ˆ the polarization direction, E 0 the peak field amplitude, h ( t ) a dimensionless envelope of peak value 1 , ω the carrier frequency and ϕ the carrier–envelope phase. A common choice is the Gaussian

h ( t ) = e − t 2 / 2 τ 2 . (6.10)

Experiments report intensity, theory needs field; they are related by

I = 1 2 ϵ 0 c E 0 2 , (6.11)

and one atomic unit of field ( 5.14 × 10 11 V/m) corresponds to I = 3.51 × 10 16 W/cm 2 .

Duration and bandwidth are conjugate. For Gaussian pulses the full-width-at-half-maximum values satisfy

Δ E Δ t ≈ 1.8   eV fs , (6.12)

a relation derived in Example 6.6.1 rather than assumed.

6.3.2A Practical Workflow

For most light–matter calculations in this course, the steps are:

  1. Choose the active basis states { | n ⟩ } and energies E n .

  2. Compute the relevant dipole elements d k n and apply selection rules early to discard forbidden channels.

  3. Specify the field E ( t ) = E 0 h ( t ) cos ⁡ ( ω t + ϕ ) and convert given intensity to E 0 with Eq. (6.11).

  4. Build Eq. (6.3) and choose the model by regime: two-level near resonance → Rabi; weak excitation to many states → first-order perturbation; otherwise integrate the coupled equations numerically.

  5. Extract populations P n ( t ) = | c n ( t ) | 2 and phases arg ⁡ c n ( t ) , then compare directly with measured line strengths, widths, and beats.

6.4Two Levels, Solved Exactly

Four frequencies are about to appear, and only two of them belong to the atom. Figure 6.2 puts all four in one picture before any of them is used, because the pair that survives into the answer — the detuning and the Rabi frequency — are precisely the two that belong to neither the atom nor the laser alone.

The two-level problem, stated. The atom contributes 0; the laser contributes and, through its amplitude E0, the coupling =dE0. Neither of the first two appears in the solution on its own: what does appear is their difference =0-, together with . The curved arrow is drawn curved on purpose — is a coupling between the two levels, not a third statement about the gap between them.
Figure 6.2. The two-level problem, stated. The atom contributes ω 0 ; the laser contributes ω and, through its amplitude E 0 , the coupling ℏ Ω = d E 0 . Neither of the first two appears in the solution on its own: what does appear is their difference Δ = ω 0 − ω , together with Ω . The curved arrow is drawn curved on purpose — ℏ Ω is a coupling between the two levels, not a third statement about the gap between them.

Truncate Eq. (6.3) to two states, | 1 ⟩ and | 2 ⟩ , separated by ℏ ω 0 = E 2 − E 1 . Since the dipole operator has no diagonal elements between parity eigenstates, V 11 = V 22 = 0 , and writing d ≡ − ⟨ 2 | 𝒅 ˆ ⋅ ϵ ˆ | 1 ⟩ (real, by a choice of basis phase) with a constant envelope h = 1 and ϕ = 0 , Eq. (6.3) reads

i ℏ c ˙ 1 = d E 0 cos ⁡ ( ω t ) e − i ω 0 t c 2 , i ℏ c ˙ 2 = d E 0 cos ⁡ ( ω t ) e + i ω 0 t c 1 . (6.13)

The ordered basis is ( | 1 ⟩ , | 2 ⟩ ) . It is the same ordered two-state basis called ( | 0 ⟩ , | 1 ⟩ ) in Chapter 1: here the labels are shifted so that subscripts agree with conventional lower- and upper-level notation. With the lower energy chosen as zero, the full laboratory-frame matrix is

H l a b ( t ) = ( 0 ℏ Ω cos ⁡ ( ω t ) ℏ Ω cos ⁡ ( ω t ) ℏ ω 0 ) , Ω = d E 0 ℏ . (6.14)

Changing the phase of either basis ket changes the sign of d and Ω , but cannot change a probability. Direct numerical integration of i ℏ 𝒂 ˙ = H l a b ( t ) 𝒂 retains both rotating terms and is the reference calculation when the RWA is doubtful.

6.4.1The Rotating-Wave Approximation

Expand the cosine as cos ⁡ ω t = 1 2 ( e i ω t + e − i ω t ) and look at the second equation:

i ℏ c ˙ 2 = d E 0 2 [ e i ( ω 0 + ω ) t ⏟ fast + e i ( ω 0 − ω ) t ⏟ slow near resonance ] c 1 . (6.15)

Define the detuning

Δ = ω 0 − ω . (6.16)

Near resonance | Δ | ≪ ω 0 , so the two exponents are wildly different: the first oscillates at about 2 ω 0 and the second at the small frequency Δ .

Definition 6.4.1Rotating-wave approximation

The rotating-wave approximation (RWA) discards the counter-rotating term e i ( ω 0 + ω ) t , retaining only the term oscillating at the detuning.

The discarded term has the large denominator ω 0 + ω . Its effect is small when | Ω | / ( ω 0 + ω ) , | Δ | / ( ω 0 + ω ) , and the fractional envelope change during one optical cycle are all small. There is no universal RWA error proportional only to Δ / ω 0 : even on exact resonance a sufficiently strong field produces counter-rotating corrections, including the Bloch–Siegert shift.

Definition 6.4.2Rabi frequency
Ω = d E 0 ℏ . (6.17)

In the rotating frame the retained constant coupling is therefore ℏ Ω / 2 off diagonal. The Rabi formula below is exact for that RWA Hamiltonian, but approximate for Eq. (6.14).

Within the RWA, Eq. (6.13) becomes

i c ˙ 1 = Ω 2 e − i Δ t c 2 , i c ˙ 2 = Ω 2 e + i Δ t c 1 . (6.18)

6.4.2Solving the Two-Level Problem

Theorem 6.4.1Rabi oscillation

With c 1 ( 0 ) = 1 and c 2 ( 0 ) = 0 , the solution of Eq. (6.18) gives

P 2 ( t ) = | c 2 ( t ) | 2 = Ω 2 Ω 2 + Δ 2 sin 2 ( Ω R t 2 ) , Ω R = Ω 2 + Δ 2 (6.19)

Ω R is the generalized Rabi frequency.

Differentiate the second equation of Eq. (6.18):

i c ¨ 2 = Ω 2 [ i Δ e i Δ t c 1 + e i Δ t c ˙ 1 ] .

From the second equation, e i Δ t c 1 = 2 i c ˙ 2 / Ω ; from the first, c ˙ 1 = − i Ω 2 e − i Δ t c 2 . Substituting both,

i c ¨ 2 = Ω 2 [ i Δ ⋅ 2 i c ˙ 2 Ω + e i Δ t ( − i Ω 2 e − i Δ t c 2 ) ] = − i Δ c ˙ 2 − i Ω 2 4 c 2 .

Dividing by i gives a linear equation with constant coefficients,

c ¨ 2 + i Δ c ˙ 2 + Ω 2 4 c 2 = 0 .

Trying c 2 ∝ e i λ t gives − λ 2 − Δ λ + Ω 2 / 4 = 0 , hence

λ = − Δ ± Δ 2 + Ω 2 2 = − Δ ± Ω R 2 .

The general solution is therefore c 2 = e − i Δ t / 2 [ A sin ⁡ ( Ω R t / 2 ) + B cos ⁡ ( Ω R t / 2 ) ] . The initial condition c 2 ( 0 ) = 0 forces B = 0 , and c ˙ 2 ( 0 ) = − i Ω / 2 (from the second equation with c 1 ( 0 ) = 1 ) fixes A = − i Ω / Ω R . Hence

c 2 ( t ) = − i Ω Ω R e − i Δ t / 2 sin ( Ω R t 2 ) ,

whose squared modulus is Eq. (6.19).

Equation (6.19) plotted. Detuning does two things at once and they pull against each other: it lowers the ceiling the population can reach, by ^2/R^2, while raising the frequency at which it gets there, to R. Only at =0 does the transfer complete.
Figure 6.3. Equation (6.19) plotted. Detuning does two things at once and they pull against each other: it lowers the ceiling the population can reach, by Ω 2 / Ω R 2 , while raising the frequency at which it gets there, to Ω R . Only at Δ = 0 does the transfer complete.

Three consequences follow, and each is stated separately below.

On resonance the transfer is complete. With Δ = 0 , Ω R = Ω and P 2 = sin 2 ⁡ ( Ω t / 2 ) , which reaches 1 at t = π / Ω . The atom is then certainly excited. Continue driving and it returns. This is coherent, reversible oscillation, not decay: a pulse delivering Ω t = π — a “ π pulse” — inverts the population exactly.

Off resonance the transfer is incomplete. The prefactor Ω 2 / ( Ω 2 + Δ 2 ) is a Lorentzian in the detuning of half-width Ω . For | Δ | ≫ Ω the maximum transfer is only Ω 2 / Δ 2 , however long one waits.

Strong driving broadens the resonance. The width of that Lorentzian is Ω ∝ E 0 : a strong field drives a transition it is badly detuned from. “Resonant” is a statement about field strength as well as frequency.

A common misreading 6.4.1Pulse area is not sufficient for every pulse

On resonance with fixed optical phase, the rotating-frame Hamiltonian is proportional to one fixed Pauli matrix, so Hamiltonians at different times commute and the result depends only on the pulse area A = ∫ Ω ( t ) d t . If the detuning or phase changes, the rotation axis changes and the ordered history matters. For example, an x pulse followed by a z pulse gives U z U x = e − i β σ z / 2 e − i α σ x / 2 , whereas reversing the pulses gives U x U z . Their difference begins with [ σ z , σ x ] = 2 i σ y , so equal scalar pulse areas do not make the two experiments equivalent.

Example 6.4.1Field required for a π pulse

A transition has dipole matrix element d = 1 a.u. What peak field and intensity invert the population in 10 fs?

Solution.

A π pulse requires Ω t = π , so with t = 10 fs = 10 / 0.02419 = 413.4 a.u.,

Ω = π 413.4 = 7.60 × 10 − 3   a.u.

With d = 1 , Eq. (6.17) in atomic units ( ℏ = 1 ) gives E 0 = Ω = 7.60 × 10 − 3 a.u., that is

E 0 = 7.60 × 10 − 3 × 5.14 × 10 11 = 3.91 × 10 9   V / m .

The intensity follows from Eq. (6.11):

I = 1 2 ( 8.854 × 10 − 12 ) ( 3 × 10 8 ) ( 3.91 × 10 9 ) 2 = 2.0 × 10 16   W / m 2 = 2.0 × 10 12   W / cm 2 .

This is a readily available laboratory intensity, and it is well below the 3.5 × 10 16 W/cm 2 at which the field reaches one atomic unit — so the RWA and the two-level truncation are both defensible here.

6.5Dressed States, and the Names They Go By

Everything so far has followed the amplitudes c 1 ( t ) and c 2 ( t ) in the basis of the bare atomic states. There is a second way to read the same equations, and it is the one that the strong-field literature is written in.

Written as a matrix in the rotating frame, Eq. (6.18) is

H ˆ RWA = ℏ 2 ( 0 Ω Ω 2 Δ ) , (6.20)

and nothing stops us from diagonalizing it. The eigenvalues are

E ± = ℏ 2 ( Δ ± Δ 2 + Ω 2 ) (6.21)

with eigenvectors that are superpositions of | 1 ⟩ and | 2 ⟩ . These are the dressed states of the atom in the field. They are not states of the atom, and they are not states of the field; they are states of the two together, and they exist only while the field is on.

Dressed energies against detuning. The dashed lines are the bare levels in the rotating frame, and they cross at resonance. The solid ones cannot: the coupling opens a gap at =0, and the two branches turn away from each other instead. Far from resonance each dressed state relaxes back onto a bare one, which is why the language of bare states works at all when the field is weak or badly detuned.
Figure 6.4. Dressed energies against detuning. The dashed lines are the bare levels in the rotating frame, and they cross at resonance. The solid ones cannot: the coupling opens a gap ℏ Ω at Δ = 0 , and the two branches turn away from each other instead. Far from resonance each dressed state relaxes back onto a bare one, which is why the language of bare states works at all when the field is weak or badly detuned.

Two things follow immediately. The gap at resonance is ℏ Ω , which is directly measurable: probe the system with a second, weak field and the line splits into a doublet separated by Ω . That is the Autler–Townes splitting, and Eq. (6.19) is the same physics seen in the time domain — Ω R = Ω 2 + Δ 2 is exactly the separation ( E + − E − ) / ℏ of the dressed levels. A splitting in a spectrum and an oscillation in time are one fact.

6.5.1One Object, Four Names

A reader moving into the strong-field literature meets several terms for states produced by a laser — virtual, transient, light-induced, dressed — usually without being told how they are related. They are related very simply: they are the same object, and the names record how long it lives.

The connection is the uncertainty relation of Sec. 6.3.1, Δ E ≃ ℏ / τ . A field-dressed state that survives a long time is spectrally sharp and can be pointed at in a measured spectrum. One that survives briefly is broad, and cannot. Nothing else distinguishes them.

The four names, plotted on the curve that relates them. A state is called virtual when it is too short-lived and too broad to appear as a feature, and dressed when it lasts as long as the pulse and appears as a resolved splitting; transient and light-induced are the names for the stretch in between. There is no boundary in the physics at any of the three places the name changes.
Figure 6.5. The four names, plotted on the curve that relates them. A state is called virtual when it is too short-lived and too broad to appear as a feature, and dressed when it lasts as long as the pulse and appears as a resolved splitting; transient and light-induced are the names for the stretch in between. There is no boundary in the physics at any of the three places the name changes.

The same statement can be made algebraically. Second-order perturbation theory for a two-photon transition through an intermediate state | k ⟩ gives an amplitude

U i j ( 2 ) = ∑ k V i k V k j E k − E j − ℏ ω + i 0 + , (6.22)

and stopping here is what makes | k ⟩ a virtual state: it is a term in a series, entered and left without energy being conserved, for a time τ ∼ ℏ / | E k − E j − ℏ ω | . Sum the series to all orders instead — which is what diagonalizing Eq. (6.20) does — and the denominator acquires a correction,

1 E k − E j − ℏ ω − Σ ( ω ) , (6.23)

whose pole sits at a shifted energy with a finite width. The state is now called dressed. The physics did not change between Eqs. (6.22) and (6.23); the approximation did.

Remark 6.5.1Dressed states are dipole allowed

It is sometimes said that a virtual state is “not a real state” and therefore cannot be reached by an allowed transition. The second half does not follow from the first. A dressed state is a superposition | Ψ ⟩ = α | 1 ⟩ + β | 2 ⟩ + ⋯ of bare states, so

⟨ k | 𝒅 ˆ | Ψ ⟩ = α ⟨ k | 𝒅 ˆ | 1 ⟩ + β ⟨ k | 𝒅 ˆ | 2 ⟩ + ⋯ ,

which is a sum of allowed matrix elements and is generically nonzero. Indeed it must be: the field could not have mixed | 1 ⟩ and | 2 ⟩ in the first place unless ⟨ 1 | 𝒅 ˆ | 2 ⟩ ≠ 0 . Dressed states are built out of dipole activity, and the Autler–Townes doublet is the experimental proof — a state nothing couples to leaves no line to split.

6.5.2Dark States: A Different Thing Entirely

There is one more term in the same literature, and it does not belong on the curve of Fig. 6.5.

Take three levels in a Λ arrangement: two lower states | 1 ⟩ and | 2 ⟩ , one upper state | 3 ⟩ , and two fields driving | 1 ⟩ → | 3 ⟩ and | 2 ⟩ → | 3 ⟩ with Rabi frequencies Ω 1 and Ω 2 . Consider the superposition

| Ψ dark ⟩ = Ω 2 | 1 ⟩ − Ω 1 | 2 ⟩ | Ω 1 | 2 + | Ω 2 | 2 . (6.24)

The amplitude for the field to lift it to | 3 ⟩ is

⟨ 3 | V ˆ | Ψ dark ⟩ ∝ Ω 2 Ω 1 − Ω 1 Ω 2 = 0 , (6.25)

identically, for any field strengths. An atom placed in this state does not absorb. It is transparent, it does not fluoresce, and in the ideal case it stays there forever. This is the mechanism behind electromagnetically induced transparency and coherent population trapping.

A common misreading 6.5.1a dark state is not a forbidden state

The two look identical in a spectrum — no absorption — and they are opposite in origin.

A dipole-forbidden transition has ⟨ i | 𝒅 ˆ | f ⟩ ≡ 0 because the two states have incompatible symmetry, by Theorem 3.6.1. Each matrix element is separately zero. No field, however intense, and no superposition, however clever, can make the transition happen.

A dark state has every matrix element nonzero. Both routes to | 3 ⟩ are fully allowed, and Eq. (6.25) says the two amplitudes cancel. The transparency is interference, not symmetry, and it is fragile in a way that a selection rule is not: change the ratio Ω 1 / Ω 2 and the same superposition becomes bright.

One test tells them apart. A forbidden transition stays dark when you change the field. A dark state does not.

So there are three categories here, not one:

originsignature
dressed (virtual, transient, LIS)field-induced mixingpeaks, shifts, splittings
darkdestructive interferencea transparency window
dipole-forbiddensymmetryno line, at any intensity

The first row is one object under four names. The second and third look alike in the data and have nothing else in common. Chapter 10 returns to the first two, where a bound state embedded in a continuum interferes with it and the resulting line shape carries the whole story in a single parameter.

6.6Perturbation Theory and the Golden Rule

When many states participate, the two-level truncation fails. The alternative is to assume the transfer out of the initial state is small and iterate Eq. (6.3).

Set c 1 ≈ 1 and all other c n ≈ 0 on the right-hand side. Then for any k ≠ 1 ,

i ℏ c ˙ k ≃ V k 1 ( t ) e i ω k 1 t (6.26)

which integrates immediately:

c k ( t ) ≃ 1 i ℏ ∫ − ∞ t V k 1 ( t ′ ) e i ω k 1 t ′ d t ′ (6.27)

The formula is physically direct: the amplitude to reach state k is the Fourier component of the perturbation at the transition frequency. This is the basic selection mechanism of ultrafast spectroscopy.

Example 6.6.1The pulse spectrum decides what is excited

Evaluate Eq. (6.27) for the Gaussian pulse of Def. 6.3.2, and deduce Eq. (6.12).

Solution.

With V k 1 = d k 1 E 0 h ( t ) cos ⁡ ω t and the RWA retaining only the e − i ω t half of the cosine,

c k ( ∞ ) ≃ d k 1 E 0 2 i ℏ ∫ − ∞ ∞ e − t 2 / 2 τ 2 e i ( ω k 1 − ω ) t d t .

The integral is the Fourier transform of a Gaussian, a standard result:

∫ − ∞ ∞ e − t 2 / 2 τ 2 e i Δ k t d t = τ 2 π e − Δ k 2 τ 2 / 2 , Δ k = ω k 1 − ω .

Hence

P k = | c k ( ∞ ) | 2 = | d k 1 | 2 E 0 2 τ 2 π 2 ℏ 2 e − Δ k 2 τ 2 .

The probability falls as a Gaussian in the detuning, with 1 / e half-width Δ k = 1 / τ in angular frequency. Converting to an energy FWHM and pairing it with the intensity FWHM of the pulse (itself 2 τ ln ⁡ 2 ) gives the product quoted in Eq. (6.12), Δ E Δ t ≈ 1.8 eV fs.

Two lessons. A short pulse populates a band of states, each weighted by its own dipole matrix element times the pulse spectrum at its own frequency — so two states with equal dipoles but different detunings receive different amplitudes. And this is why a research calculation must use the measured pump bandwidth: it weights every state that gets excited.

Example 6.6.2Reading a pulse specification

An attosecond XUV pulse has bandwidth 0.4 eV centered at 17.6 eV. How long is it, and how many states can it excite at once if the levels are spaced by 0.1 eV?

Solution.

From Eq. (6.12),

Δ t ≈ 1.8   eV fs 0.4   eV = 4.5   fs .

A bandwidth of 0.4 eV covers states within roughly ± 0.2 eV of the carrier, so with 0.1 eV spacing it reaches four or five levels simultaneously and coherently.

This inverts the usual spectroscopic picture. A narrowband laser selects one level and measures it; a broadband pulse creates a superposition of several, whose relative phases then advance at different rates so that the system beats. Chapter 11 is about reading those beats.

6.6.1Transitions into a Continuum

Why a rate needs a continuum. Against a single final state the population oscillates back, as the Rabi solution above shows; against a dense band the contributions never re-phase, the return never happens, and what is left is a constant probability per unit time.
Figure 6.6. Why a rate needs a continuum. Against a single final state the population oscillates back, as the Rabi solution above shows; against a dense band the contributions never re-phase, the return never happens, and what is left is a constant probability per unit time.

Now let the final states form a continuum — the situation whenever the atom ionizes, since the freed electron may carry away any energy. Chapter 3 flagged the normalization this requires, ⟨ ε | ε ′ ⟩ = δ ( ε − ε ′ ) .

Theorem 6.6.1Fermi's golden rule

For a perturbation of constant strength V ˆ switched on at t = 0 , the transition rate from a discrete state | i ⟩ into a continuum of final states is

Γ = 2 π ℏ | ⟨ f | V ˆ | i ⟩ | 2 ρ ( E f ) , (6.28)

evaluated at the energy-conserving final energy, with ρ the density of final states per unit energy.

From Eq. (6.27) with V f i constant for 0 < t ′ < t ,

c f ( t ) = V f i i ℏ ∫ 0 t e i ω f i t ′ d t ′ = V f i i ℏ e i ω f i t − 1 i ω f i ,

so

P f ( t ) = | c f | 2 = | V f i | 2 ℏ 2 ⋅ 4 sin 2 ⁡ ( ω f i t / 2 ) ω f i 2 = | V f i | 2 t 2 ℏ 2 sinc 2 ( ω f i t 2 ) .

This function of ω f i has height ∝ t 2 and width ∝ 1 / t , so its integral grows as t . Summing over the continuum, ∑ f → ∫ ρ ( E f ) d E f = ℏ ∫ ρ d ω f i , and using

∫ − ∞ ∞ 4 sin 2 ⁡ ( ω t / 2 ) ω 2 d ω = 2 π t ,

the total probability is

P ( t ) = | V f i | 2 ℏ 2 ℏ ρ ( E f ) 2 π t = 2 π ℏ | V f i | 2 ρ ( E f ) t ,

assuming ρ and V f i vary slowly over the narrow width of the sinc. Since P grows linearly, the rate Γ = d P / d t is constant and equal to Eq. (6.28).

Two features of this result are used directly at the end of the book.

Remark 6.6.1Energy normalization absorbs ρ

If the continuum states are normalized as ⟨ ε | ε ′ ⟩ = δ ( ε − ε ′ ) then the density of states is already carried by the states themselves, ρ = 1 , and in atomic units Eq. (6.28) reduces to

Γ = 2 π | ⟨ ε | V ˆ | i ⟩ | 2 . (6.29)

This is why the research literature uses that convention: it removes a bookkeeping factor that is otherwise easy to get wrong.

Remark 6.6.2The perturbation need not be light

Nothing in the proof assumed V ˆ was a laser field. Any coupling between a discrete state and a degenerate continuum gives a rate of this form — including the electron–electron repulsion that autoionized helium's 2 s 2 p state in Chapter 5. Reading Eq. (6.29) backwards, | ⟨ ε | V ˆ | i ⟩ | = Γ / 2 π , converts a measured width into a coupling strength. That is exactly how autoionization enters the model in Chapter 10.

6.6.2Where These Methods Fail

Perturbation theory assumed c 1 ≈ 1 ; the golden rule assumed further that the initial state persists while the rate acts. Both fail, and the two failures define the research frontier this course points at.

Strong fields. When E 0 approaches one atomic unit ( 5.14 × 10 11 V/m, I ∼ 10 16 W/cm 2 ) the field rivals the atom's own binding and no expansion in it converges. The atom no longer absorbs one photon at a time; the barrier confining the electron is bent down until the electron tunnels through. Describing that requires the Keldysh parameter, tunnelling rates, and classical propagation of the freed electron — a subject of its own, not treated in these notes.

Decaying states. The golden rule gives a constant rate, hence exponential decay and a Lorentzian line. But if the state being driven is itself decaying into the same continuum the laser ionizes into, the two routes to that continuum interfere. The line is then not Lorentzian and no rate equation describes it: the amplitudes must be kept and their phases tracked. That is Eq. (6.3) with no perturbative expansion at all, and it is the subject of Chapters 10 and 11.

First-pass self-checks

When the rotating-wave approximation is safe, and when it is not. Both panels integrate the full laboratory Hamiltonian Eq. (6.14) numerically, with no approximation, and compare it with ^2( t/2). On resonance and at =0.1\,0 the two are indistinguishable at this scale. At =0.5\,0 the counter-rotating term the RWA discarded shows up as a visible ripple and a shifted period — the Bloch–Siegert effect. Note what changed: not the detuning, which is zero in both panels, but the field strength. The RWA is a statement about /0, not only about /0.
Figure 6.7. When the rotating-wave approximation is safe, and when it is not. Both panels integrate the full laboratory Hamiltonian Eq. (6.14) numerically, with no approximation, and compare it with sin 2 ⁡ ( Ω t / 2 ) . On resonance and at Ω = 0.1 ω 0 the two are indistinguishable at this scale. At Ω = 0.5 ω 0 the counter-rotating term the RWA discarded shows up as a visible ripple and a shifted period — the Bloch–Siegert effect. Note what changed: not the detuning, which is zero in both panels, but the field strength. The RWA is a statement about Ω / ω 0 , not only about Δ / ω 0 .
Two pulse sequences with identical areas and opposite outcomes. Each row applies a /2 rotation about x and one about z to the same initial state, differing only in which comes first. The bars are the computed probability of finding + afterwards: 1 and 12. Pulse area alone therefore cannot predict a result once the rotation axis moves, which is Misreading 6.4.1 made numerical and Self-check 3 made visible.
Figure 6.8. Two pulse sequences with identical areas and opposite outcomes. Each row applies a π / 2 rotation about x and one about z to the same initial state, differing only in which comes first. The bars are the computed probability of finding | + ⟩ afterwards: 1 and 1 2 . Pulse area alone therefore cannot predict a result once the rotation axis moves, which is Misreading 6.4.1 made numerical and Self-check 3 made visible.
Example 6.6.3Self-check 1: exact static propagation

In units with ℏ = 1 , take Chapter 1's matrix H = ( 0 0.2 0.2 1 ) . Starting in the first basis state, find the largest probability of the second state.

Solution.

Here D = 1 , g = 0.2 , and D 2 + 4 g 2 = 1.16 = 1.07703296 . Equation (6.6) gives

P 2 ( t ) = 4 g 2 D 2 + 4 g 2 sin 2 ( D 2 + 4 g 2 t 2 ) .

The sine can reach one, so P 2 , max = 0.16 / 1.16 = 0.137931 . The lower eigenvalue is E − = ( 1 − 1.16 ) / 2 = − 0.0385164807 , close to the weak-coupling estimate − g 2 / D = − 0.04 . This checks both the matrix exponential and the perturbative scale.

Example 6.6.4Self-check 2: weak response versus Rabi transfer

Let ℏ = 1 , ω 0 = ω = 1 , Ω = 0.1 , and E ( t ) = E 0 cos ⁡ t . Compare first-order and RWA probabilities at t = 2 .

Solution.

The resonant slowly varying part of the first-order integral is c 2 ( 1 ) ≃ − i Ω t / 2 , so P 2 ( 1 ) ≃ ( 0.1 × 2 / 2 ) 2 = 0.010000 . The RWA result is P 2 = sin 2 ⁡ ( 0.1 × 2 / 2 ) = sin 2 ⁡ ( 0.1 ) = 0.009967 . Their agreement is the quadratic weak-excitation limit of the sine. At t = π / Ω = 31.416 , the RWA predicts complete transfer while the first-order expression predicts a probability larger than one and has therefore left its domain. Direct integration of Eq. (6.14) is the independent full-drive check.

Example 6.6.5Self-check 3: order two noncommuting pulses

A state starts at the north pole of the Bloch sphere. Compare a π / 2 rotation about x followed by one about z with the reverse order.

Solution.

The ordered propagators are U z U x = e − i π σ z / 4 e − i π σ x / 4 and U x U z = e − i π σ x / 4 e − i π σ z / 4 . In Bloch-vector form, the first sequence maps + z → − y → + x ; the reverse sequence maps + z → + z → − y . A measurement of σ x therefore has expectation 1 after the first order and 0 after the second. Equal rotation angles do not erase time ordering.

6.7Summary

The chapter reduces light–matter interaction to one working pipeline:

6.8Exercises

  1. The coupled equations. Reproduce the proof of Theorem 6.2.1, showing explicitly where the E n c n terms cancel. Then explain in one sentence what would change if the phase e − i E n t / ℏ had not been extracted in Eq. (6.2).

  2. The RWA, term by term. Starting from Eq. (6.13), expand both cosines and write out all four resulting terms with their exponents. Identify which two are discarded by Def. 6.4.1 and estimate the size of the error for ω 0 = 0.4 a.u. and Ω = 0.01 a.u.

  3. Rabi, checked.

    1. Verify that Eq. (6.19) satisfies P 2 ( 0 ) = 0 , reaches 1 on resonance, and has maximum Ω 2 / Δ 2 for Δ ≫ Ω .

    2. Show that P 1 + P 2 = 1 at all times, using the explicit c 2 ( t ) from the proof and the first equation of Eq. (6.18).

    3. Repeat Example 6.4.1 for a 100 fs pulse. By what factor does the required intensity change, and why?

  4. The golden rule. Fill in the two steps of Theorem 6.6.1's proof that are quoted: the elementary integral giving the sinc form, and ∫ 4 sin 2 ⁡ ( ω t / 2 ) / ω 2 d ω = 2 π t . Then explain why the rate is constant even though P f ( t ) for a single final state oscillates.

  5. Bandwidth in practice. An XUV pulse has a Gaussian envelope with τ = 2 fs.

    1. Using Example 6.6.1, find the spectral width in eV over which it can excite states.

    2. Two states lie 0.15 eV apart, both within the bandwidth and with equal dipole matrix elements. Compute the ratio of their final amplitudes if the carrier is tuned exactly to the lower one.

    3. How long after the pulse do the two amplitudes return to their initial relative phase?

6.9Project: Driving a Two-Level Atom

The problem. Integrate the coupled amplitude equations numerically for a two-level atom driven by a Gaussian pulse, and map out where perturbation theory works. This project builds the tool the last chapters need: the research code is this same integration with more amplitudes.

Work in atomic units. The exact equations, with the envelope restored, are

i c ˙ 1 = d E 0 h ( t ) cos ⁡ ( ω t ) e − i ω 0 t c 2 , i c ˙ 2 = d E 0 h ( t ) cos ⁡ ( ω t ) e + i ω 0 t c 1 ,

with h ( t ) = e − ( t − t 0 ) 2 / 2 τ 2 . Take ω 0 = 0.4 , d = 1 , τ = 200 , and t 0 large enough that the pulse begins from nothing.

  1. On paper. Apply the RWA by hand as in Sec. 6.4.1 and write down the resulting pair of equations.

  2. On paper. Predict the peak field for a π pulse from ∫ Ω ( t ) d t = π with Ω ( t ) = d E 0 h ( t ) and ∫ h d t = τ 2 π . This is the number the simulation must confirm.

  3. On the computer. Integrate the exact equations from c 1 = 1 , c 2 = 0 . Plot P 1 ( t ) and P 2 ( t ) through the pulse and confirm P 1 + P 2 = 1 ; report the actual deviation, since it is your error bar for everything else.

  4. On the computer. Scan E 0 from far below your π -pulse value to several times it, and plot P 2 ( ∞ ) against E 0 . Overlay the first-order prediction from Example 6.6.1. Identify the field at which the two part company by more than ten per cent.

  5. On the computer. Fix E 0 weak and scan the detuning. Plot P 2 ( ∞ ) against Δ and compare the width with both 1 / τ and Ω . Which sets it at weak field, and which takes over as E 0 grows?

  6. On the computer. Integrate the RWA equations alongside the exact ones and plot the difference. Where does the RWA visibly fail — in detuning, in field strength, or in the fast wiggles?

  7. What has to move. One animation: the two populations against time with the pulse envelope drawn behind, played for increasing E 0 so the flopping speeds up. Caption it with what to watch.

The check. Part (c)'s norm conservation is the test that matters. Drift in P 1 + P 2 almost always means the time grid does not resolve the optical period 2 π / ω — you must resolve the carrier, not merely the envelope.

Be ready to answer. Perturbation theory predicts P 2 growing as E 0 2 without limit, so it must eventually exceed 1 . Point at the specific step in the derivation of Eq. (6.27) that allows a probability greater than one, and say what the exact calculation does that the perturbative one cannot.