Why Can Green’s Functions Be Calculated with Diagrams?
— Adding an Electron to the Electron Sea and Removing It

This is the theory companion to the simulator (Green’s Functions and Feynman Diagrams). Starting from the definition of the one-particle Green’s function, we read through all six first-order diagrams (Fig. 9.1) and all ten second-order diagrams (Fig. 9.8), and tell it as one continuous story that reaches the point where the polarization bubble leads to the GW approximation and crossed exchange leads to the vertex correction. In §11–13 the lead role passes to the phonon: we reread the harmonic oscillator as G0 and the cubic and quartic anharmonic terms as vertices in the same language, and check the Dyson equation against the Morse-type potential, which has an exact solution. The notation follows Fetter–Walecka, Quantum Theory of Many-Particle Systems (Japanese translation) (iG = ⟨T[ψψ†]⟩, Gaussian units, a factor of −1 for each closed loop). The equation numbers and the names of the diagrams correspond to Appendix C of the Mochizuki Research Group’s Introduction to Linear Response Theory (in Japanese).

0. The Big Picture — From Free Particles to the Vertex Correction

The lead role of this explanation is a single question. “If we add one electron to a sea of non-interacting electrons (a Fermi sea) and remove it after a while, how much of it survives as ‘the same single particle’?” This survival amplitude is the one-particle Green’s function, and as the order of perturbation is raised, the answer becomes richer in the following five stages.

0th order: free propagationOnly the phase rotates. The amplitude does not decrease. Lifetime ∞.
1st order: Hartree–FockThe mean field and exchange shift the levels. The lifetime is still ∞.
2nd order: polarization and crossed exchangeParticle–hole pairs are created, and the lifetime becomes finite for the first time.
Summation: RPA / WSum the ring diagrams to infinite order to get the screened interaction W.
Beyond: GW and ΓΣ = iGWΓ. Crossed exchange is the seed of the vertex correction Γ.

And there is one caution that runs through the whole page. A Feynman diagram is not a classical trajectory of an electron; it is a ledger that shows how the lines of an integral expression are connected. We picture the vertical direction of a diagram as time (up is the future), but the times of the internal vertices are all integrated over, so a single diagram represents all possible time orderings together. This caution will come back again and again later (§4, §8, §10).

About the figure numbers On this page and in the simulator, we call the diagrams by the numbers of the original source, Fetter–Walecka (Japanese translation), Fig. 9.1, Fig. 9.7, Fig. 9.8, and Fig. 9.16. In Appendix C of the Mochizuki Research Group’s Introduction to Linear Response Theory (in Japanese), the same figures are called Fig. C.2, Fig. C.7, Fig. C.11, and Fig. C.12, respectively. The original figures themselves are reproduced in the text below.

1. What Does the Green’s Function Keep Track Of?

Let the exact ground state of an interacting N-electron system be |Ψ0⟩ and let the field operators in the Heisenberg picture be ψ^(x), ψ^†(y) (where x=(r,t) is a spacetime point and α, β are spin indices). The one-particle Green’s function is defined by

iGαβ (x,y) = ⟨Ψ0| T[ ψ^α(x) ψ^β†(y) ] |Ψ0⟩ ⟨Ψ0|Ψ0⟩ (1) Definition in the F–W style

Unwinding the time-ordered product T shows that this single formula contains two stories. For t>t′ we have

iGαβ (x,y) = ⟨Ψ0| ψ^α(x) ψ^β†(y) |Ψ0⟩ (2) Adding a particle

Reading from the right, it says “① create one electron at y, ② evolve in time as an (N+1)-particle system, and ③ remove one electron at x.” In other words, it is “the amplitude for the electron added at y to be detected at x.” Conversely, for t<t′ (with a sign − from interchanging fermions) the story is that an electron is pulled out first, that is, the propagation amplitude of a hole. The T product is a device that combines the two stories, adding a particle and removing one, into a single analytic function — which is exactly why a single solid line in a diagram can represent both a particle and a hole.

Supplement: the Green’s function is not a product of perturbation theory Eq. (1) is a quantity defined exactly for the interacting Hamiltonian; the perturbation expansion and Feynman diagrams are merely one method for calculating it. The factor i on the left-hand side is the Fetter–Walecka convention; it is there because a single Wick contraction then corresponds exactly to iG0, which keeps the dictionary of diagrams clean. The physics is the same as in the G = −i⟨Tψψ†⟩ of many textbooks; this i carries no new physics.

Without interactions the answer is easy. The free Green’s function for momentum k and frequency ωk=εk/ℏ is

G0 (k,ω) = θ(k−kF) ω−ωk+iη + θ(kF−k) ω−ωk−iη (3) Free Green’s function

The first term is above the Fermi surface (particles), and the second is below it (holes). In the time representation only the phase e−iωkt rotates and the amplitude never decreases — the lifetime is ∞. When interactions are switched on, the excitation ① shifts in energy (renormalization), ② loses weight as a single particle (Z < 1), ③ acquires a finite lifetime if it can decay into real final states, and ④ the lost weight moves to an incoherent background and to satellites. The yardstick that measures this is the spectral function.

A(k,ω) = −1π Im⁡GR (k,ω) , D(ω) = ∑k Tr⁡ A(k,ω) (4) Spectral function and density of states
−Im G is not the density of states as it stands What produces the spectral function A(k,ω) is the retarded Green’s function GR. What we calculate with diagrams is the time-ordered GT, so the two must not be confused. The density of states D(ω) is obtained only after taking the spin trace and the momentum sum of Eq. (4). If you plot A(k,ω) as an intensity map on the ω–k plane, you can see the renormalized band dispersion, the linewidth, and the satellites (this is what ARPES measures).

2. Why Does It Become “Diagrams”? — Wick’s Theorem and the Dictionary

Expanding Eq. (1) in the interaction picture, the nth-order term contains n interaction Hamiltonians in a row, and a total of 4n+2 field operators appear. The bare Coulomb interaction is

U(x1,x1′) = V(r1−r1′) ⁢ δ(t1−t1′) (5) Instantaneous interaction

and both ends of a wavy line are always at the same time. Wick’s theorem breaks the ground-state expectation value of this long string of operators down into a sum of “products of pairwise contractions (= iG0).” Each way of choosing the contractions is one diagram. At first order there are three ψ and three ψ†, so there are 3! = 6 ways — these are the six diagrams of Fig. 9.1. At second order the 5! = 120 ways are reduced to 10 by grouping equivalent ones (§6). The dictionary for turning diagrams back into formulas is as follows.

Table 1. Dictionary of Feynman diagrams (coordinate space, Fetter–Walecka style)
Diagram elementFactor in the formulaHow to read it
Solid line with arrow y → xiG0αβ(x, y)Free propagation. For tx > ty it represents a particle, and for tx < ty a hole
Wavy lineU(x₁, x₁′)Instantaneous Coulomb interaction. Both ends are at the same time (Eq. (5))
Vertex (black dot)∫d⁴x₁ and a spin sumAll internal spacetime points are integrated over
Closed Fermi loopFactor (−1) plus a spin traceA sum over occupied states. A sign − for each loop
Equal-time line with the same start and end pointiG0(rt, rt⁺) = −n/(2s+1)·δαβThe density n of the electron sea. This is what is inside a tadpole
Overall factor at order n(−i/ℏ)ⁿ (−1)FF is the number of loops (it counts the (−1) of the loop row above; it must not be multiplied in twice)

3. Reading All Six First-Order Diagrams (Fig. 9.1)

First-order contributions (A)–(F) to the one-particle Green’s function
Fig. 9.1. First-order contributions (A)–(F) to the one-particle Green’s function. Time runs vertically (up is the future) and space horizontally. Solid lines are G0, and wavy lines are the interaction U. (A) and (B) are vacuum bubbles that are not connected to the external line, (C) = (E) is the exchange term, and (D) = (F) is the direct term. Source: A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems (Japanese translation), Fig. 9.1 (Fig. C.2 in Appendix C)

All six have “an external line from y to x” and “one wavy line x₁–x₁′”; the only difference is how the remaining two solid lines are connected. Please read them while checking them one by one in tab ① of the simulator.

Table 2. How the six first-order diagrams are read and what becomes of them ((A)–(F) of Fig. 9.1). The “representative story” is only one time ordering chosen for intuition; in reality all internal times and internal coordinates are integrated over.
DiagramShapeRepresentative storyFate
(A)Dumbbell (two densities + wavy line)The two equal-time densities of the sea interact directly through the Coulomb interaction — the Hartree energy of the ground stateDisconnected from the external line → cancels against the denominator ⟨Φ₀|Ŝ|Φ₀⟩ and vanishes
(B)Oyster (exchange loop + wavy line)The exchange amplitude of two occupied electrons of the sea — the Fock energy of the ground stateDisconnected → cancels against the denominator and vanishes
(C)Exchange: the external line passes through both ends of the wavy line in turnThe added electron exchanges with an occupied electronConnected. Together with (E), it becomes the Fock term of Fig. 9.7(b)
(D)Tadpole: a density loop is attached to the external lineThe added electron feels the Coulomb field of the average densityConnected. Together with (F), it becomes the Hartree term of Fig. 9.7(a)
(E)Endpoints x₁ ↔ x₁′ of (C) interchangedMerely renaming the integration variablesSame value as (C). The factor 2 cancels the 1/2 of H₁
(F)Endpoints of (D) interchangedSame as aboveSame value as (D)
What it means that (A) and (B) “vanish” It is not that there is no physics there. A vacuum bubble is the energy and phase of the ground state, which exist even if not a single electron is added, and since G is a quantity that records “the difference caused by the addition,” it is divided out as a common factor (the linked-cluster theorem).

In the end, the first-order Green’s function is reduced to two topologically distinct connected diagrams.

G(1) = G0ΣHG0 + G0ΣFG0 (6) First order = Hartree + Fock
ΣH(r) = 1ℏ ∫d3r′ V(r−r′) ⁢n(r′) , ΣF,αβ(r,r′) = −1ℏ V(r−r′) ⁢ ραβ(r,r′) (7) Self-energies of Hartree and Fock

Here ραβ(r,r′) is the one-body density matrix. In the uniform electron gas (jellium), the Hartree term nV(0) exactly cancels against the potential of the uniform positive background charge. The Hartree–Fock approximation is nothing other than stopping the Feynman diagrams at first order.

4. Hartree and Fock Do Not Differ by “Before and After in Time”

A common misreading “The tadpole has a horizontal wavy line, so it can only happen at one particular time. The Fock wavy line is slanted, so an electron in the past and an electron in the future are interacting” — neither is correct. ① The times of the internal vertices are all integrated over, and “an event that happens at one particular time” is written nowhere in the diagram. ② Both ends of a wavy line are always at the same time (Eq. (5)); the slant is just a matter of how the diagram is drawn.
First-order Feynman diagrams: (a) Hartree and (b) Fock
Fig. 9.7. First-order Feynman diagrams after grouping. (a) Direct term = Hartree (tadpole), (b) exchange term = Fock. (D) and (F) of Fig. 9.1 are grouped into (a), (C) and (E) into (b), and (A) and (B) have disappeared. Source: Fetter–Walecka (Japanese translation), Fig. 9.7 (Fig. C.7 in Appendix C)
Table 3. So what is different? — Only whether it is direct or exchange
(a) Direct term = Hartree(b) Exchange term = Fock
Fact about the diagramThe external line passes through only the vertex x₁, and a closed loop is attached to x₁′The external line passes in turn through both ends x₁, x₁′ of the wavy line
PhysicsThe added electron feels the Coulomb field created by the average density of the seaExchange due to the antisymmetry of identical fermions. Since the particles cannot be given name tags, the interchanged amplitude is added with a sign −
Fermi loopOne (sign −)None (sign +)
NonlocalityLocal in r (a multiplication operator)A nonlocal operator connecting the two points r and r′

In (C) of tab ① of the simulator, the “amplitude that goes straight through” and the “amplitude that is interchanged” overlap semi-transparently, and a − tag is attached. Exchange is not an event in which two classical particles swap positions; it is an interference term of the antisymmetrized many-body amplitude.

5. Why Is There No Lifetime at First Order? — How to Read δ(t−t′) Correctly

Writing the Hartree–Fock self-energy in the time representation gives

ΣHF (t,t′) = δ(t−t′) ⁢ ΣHF (8) Time-local kernel
δ(t−t′) is not “an event that occurs for only an instant” In the Dyson equation the self-energy acts as a convolution ∫dt′ Σ(t,t′)G(t′,t₀). Inserting Eq. (8) gives ΣHF·G(t,t₀) — that is, at any time t the self-energy acts only on the state at that very instant and does not refer to the history of the past (it has no memory). It does not act at only one absolute time t = t′; t itself runs over all times.

From here a single chain of logic extends.

Time-local (memoryless) ⇒ static ⇒ real ⇒ lifetime ∞ Fourier transforming with respect to the time difference τ = t−t′ gives δ(τ) ↔ 1; that is, ΣHF(ω) is independent of frequency (static). Since the static Hartree–Fock self-energy is real, Im ΣHF = 0. The linewidth is 0 and the lifetime is infinite. The levels are renormalized, yet the electron lives forever — this is the world of first order. Conversely, a finite lifetime requires “a self-energy that has a memory of time (that depends on frequency),” and that first appears in (i) and (j) at second order.
Supplement: the time kernel is not a probability distribution Even at second order and beyond, where Σ(t,t′) is no longer a δ function, it does not in general become a positive Gaussian distribution. The time kernel is a complex-valued memory kernel, which can oscillate, change sign, and decay as a power law. It is also inappropriate to call its “width” an uncertainty. Moreover, a kernel having a finite time width does not by itself produce a lifetime — only when there are real final states that satisfy energy conservation (such as the particle–hole continuum) does Im ΣR ≠ 0 (§9).

6. The Ten Second-Order Diagrams (Fig. 9.8) — Four Reducible and Six Proper

At second order there are two wavy lines and five solid lines, and there are ten topologically distinct connected diagrams. They fall into two groups.

G(2) = G0 [ Σ(1) G0 Σ(1) + Σ(2) ] G0 (9) Structure of second order
Second-order Feynman diagrams (a)–(j)
Fig. 9.8. Second-order contributions (a)–(j) to the one-particle Green’s function. (a)–(d) are reducible (cutting one intermediate solid line splits them into first order × first order), and (e)–(j) are proper. Source: Fetter–Walecka (Japanese translation), Fig. 9.8 (Fig. C.11 in Appendix C)
Table 4. The ten second-order diagrams ((a)–(j) of Fig. 9.8). The “representative story” is just one way of reading chosen for intuition; in reality all internal times and internal momenta are integrated over.
Fig. 9.8NameRepresentative storyClassificationFig. 9.16
(a)Hartree–HartreeFeels the mean field, propagates, and feels it once more: G⁰ΣHG⁰ΣHG⁰Reducible (cutting one solid line splits it into two)—
(b)Fock–FockExchange → propagation → exchange once moreReducible—
(c)H–F mixed 1One Hartree and one Fock (connection order 1)Reducible—
(d)H–F mixed 2Opposite connection order to (c)Reducible—
(e)Tadpole with exchangeAn electron in the density loop exchanges with another occupied electron — the Hartree field created by an exchange-corrected densityProper, static(a)
(f)Tadpole of a tadpoleA nesting in which a density loop receives the Hartree field of another density loopProper, static(b)
(g)RainbowThe internal line responsible for the Fock exchange undergoes a further exchange (unlike in (b), it is not the external line)Proper, static(c)
(h)Exchange with a tadpoleThe internal line of the exchange receives the Hartree field of the seaProper, static(d)
(i)Polarization-bubble insertionThe electron passes (q,ω) to the sea → a particle–hole pair (polarization cloud) is created, propagates, and recombines, and the back-action is returnedProper, dynamic — the seed of GW(e)
(j)Crossed exchangeEmit q, emit q′, absorb q first, and absorb q′ last. Several exchange histories interfere, with signsProper, dynamic — the seed of the vertex correction(f)
Do not mix up the label correspondence Fig. 9.8 and Fig. 9.16 carry the same symbols (a)–(f), which makes them easy to confuse, but what is called “the core of the correlation” is (e) and (f) of Fig. 9.16, that is, (i) and (j) of Fig. 9.8. (e) and (f) of Fig. 9.8 are merely a static rearrangement of structure, in which the internal lines of the Hartree/Fock diagrams are dressed.

(a)–(d) are merely reducible diagrams in which “a first-order block is used twice on the external line,” and they are generated automatically by iterating the Dyson equation (§9). As long as ΣH and ΣF are real and static, no matter how many of these are stacked in series, Im Σ = 0 remains — the levels and orbitals change, but the linewidth is 0. (e)–(h) are proper (they do not split when one solid line is cut), but they are again static dressings. The new physics is in (i) and (j).

7. The Polarization Bubble (i) — Screening and the Seed of the GW Approximation

Second-order proper self-energy (a)–(f)
Fig. 9.16. The six terms of the second-order proper self-energy Σ★(2). These are the forms obtained from (e)–(j) of Fig. 9.8 by cutting off the external lines. (e) is the seed of the GW approximation and (f) is the seed of the vertex correction — these two are the core of dynamic correlation. Source: Fetter–Walecka (Japanese translation), Fig. 9.16 (Fig. C.12 in Appendix C)

Writing down Fig. 9.16(e) (= the proper part of Fig. 9.8(i)) with the Feynman rules in momentum space gives (g = 2s+1 is the spin degeneracy)

Σ(2e) (k) = (i/ℏ)2 (−g) ∫d4q(2π)4 ∫d4p(2π)4 [V(q)]2 G0(k−q) G0(p) G0(p+q) (10) Self-energy of the polarization bubble

The pair G⁰(p)G⁰(p+q) is the particle–hole pair (polarization). Writing the free polarization as P0(q)=−i∫dpG0(p)G0(p+q), the interaction part of (10) is V(q)P⁰(q)V(q) — the first term of “the bare interaction polarizes the sea, and the polarization weakens the interaction.” Now putting in the bare Coulomb V(q) = 4πe²/q², since the polarization loop is finite at q → 0 while [V(q)]² ~ q⁻⁴, this term diverges at small q. It is not that perturbation theory breaks down at second order; it is a signal to “add up diagrams of the same form (rings) to infinite order.” Summing the geometric series gives

W = v+vP0v +vP0vP0v +⋯ = v1−vP0 (11) Screened interaction in the RPA

— the screened interaction W (RPA), which is finite even at q → 0. Replacing the wavy line v of the exchange term by W, Σ = iGW is the GW approximation. It can also be described as the approximation that sums (e) of the six terms of Fig. 9.16 to infinite order and discards (f) (crossed exchange). Please check in Fig. 1 below how “it diverges when stopped at finite order and converges when summed completely.”

Fig. 1. Summing the rings: v + vP⁰v + … → W. Even if the number n of bubbles is increased, the partial sums oscillate and diverge at small q, and only the infinite sum (RPA) converges to the finite screened W. The curves are a model for teaching (v = 0.8/q², −P⁰ = 1/(1+q²/3)).

8. Crossed Exchange (j) — The Seed of the Vertex Correction

The other dynamic diagram, Fig. 9.16(f) (= Fig. 9.8(j)), is

Σ(2f) (k) = (i/ℏ)2 ∫d4q(2π)4 ∫d4q′(2π)4 V(q) V(q′) G0(k−q) G0(k−q−q′) G0(k−q′) (12) Self-energy of crossed exchange

The external line emits the wavy line q, then emits q′, and absorbs first the q it emitted first, and absorbs q′ last — this “order of connection of the lines” produces the crossing in the diagram. The crossing in the diagram does not mean that the past and the future interacted. Physically, it expresses that, because identical Fermi particles cannot be given name tags, several exchange histories cannot be distinguished and interfere as quantum amplitudes that include signs.

Why is it “the seed of the vertex correction”? No matter how many times one repeats the operation of inserting a polarization bubble into a wavy line (= the operation that builds GW), this crossed diagram can never be built. In Hedin’s exact form Σ = iGWΓ, the GW approximation assumes the bare vertex Γ = 1, while crossed exchange corresponds to the first nontrivial correction Γ⁽¹⁾ in Γ = 1 + Γ⁽¹⁾ + ⋯. It is the entrance to “the physics that GW misses,” such as consistency with the Ward identity (conservation laws) and excitonic effects. (12) is finite even for the bare Coulomb interaction, and gives the remainder of the second-order exchange–correlation.

9. The Dyson Equation and the Lifetime — Until Im Σ Appears in the Spectrum

Once the proper self-energy Σ★ (the sum of diagrams that do not split into two when an external line is cut) has been built, the Dyson equation automatically sums the reducible diagrams to infinite order. In momentum space,

G(k,ω) = 1 ω−ωk −Σ★(k,ω) (13) Dyson equation

Expanding near the pole, we can read off the quasiparticle triplet.

ε~k = εk+ℏRe⁡Σ★ , γk = −ZkIm⁡Σ★ , Zk = [1−∂Re⁡Σ★/∂ω]−1 (14) The quasiparticle triplet

Re Σ★ is the shift of the level, −Im Σ★ is the linewidth (the inverse of the lifetime), and Z is the remaining share of “single-particle character.” At first order (HF), Σ★ is real and static, so γ = 0 and Z = 1. When Σ★ becomes frequency dependent through (i) and (j) at second order, Im Σ★ < 0 arises in the region where there are real final states that satisfy energy conservation (the particle–hole continuum), the peak of the spectrum broadens into a Lorentzian of width 2γ, and in the time domain the survival amplitude decays as e−γt. Conversely, in a region with no final states (for example, inside a gap), the lifetime remains strictly infinite even if Σ★ depends on ω. In a Fermi liquid, the Pauli exclusion principle blocks the final states, so near the Fermi surface

−Im⁡Σ★R (ξ,T) ∝ ξ2 + (πkBT)2 (15) Scattering rate of a Fermi liquid

— the lower the temperature and the closer to the Fermi surface, the longer-lived the quasiparticle (ξ = εk − εF). The intuition that “lowering the temperature strengthens the quantum effects of the electron–electron interaction and makes the scattering fiercer” is the wrong way around. Please check this in Fig. 2.

Fig. 2. Spectral function A(k,ω) and lifetime. Switch the level of approximation and move ξ and T. Up to first order the peak only moves, and its width is 0. A width appears for the first time at second order; please check that it narrows as ξ → 0 and T → 0 (Eq. (15)). The model is a teaching model that has the form of Eq. (15), and λ is the interaction strength (γ is given directly as the half width at half maximum, and the Z factor of Eq. (14) has already been absorbed into γ).

10. Common Misconceptions and How to Read Things Correctly

Table 5. A list of misconceptions about Feynman diagrams and Green’s functions (if you stumble, go to the corresponding section)
A reading that is easy to stumble onThe correct reading§
A diagram is a photograph of the trajectory the electron followedA ledger showing “how the lines are connected” in an integral expression. All internal times and coordinates are integrated over0, 2
The wavy line is a “Coulomb ray” flying at finite speed, and a slanted wavy line connects the past and the futureBoth ends of a wavy line are always at the same time (U ∝ δ(t₁−t₁′)). The slant is just a matter of how the diagram is drawn2, 4
Hartree “happens at one particular time,” while Fock “spans times”The only difference is direct or exchange (how the lines are connected)4
δ(t−t′) is “an event that happens only once”It means “has no memory (time-local).” t runs over all times5
If HF is stacked up to high order, a lifetime will eventually appearHowever many times a real, static Σ is stacked, Im Σ = 0. The lifetime comes from the dynamic correlation of (i) and (j)5, 6, 9
The time kernel is a Gaussian distribution, and its width is “the uncertainty of the self-energy”A complex-valued memory kernel. Width alone gives no lifetime; real final states are required5, 9
−Im G is the density of states as it stands. Resolving it in k gives a band diagramA(k,ω) = −(1/π)Im GR (retarded type) → D(ω) from the trace and Σk1
The lower the temperature, the fiercer the scattering between electrons and the “stickier” things getBy the Pauli exclusion principle, −Im Σ ∝ ξ²+(πkBT)². At low temperature and near the Fermi surface the lifetime is long9
A short one-particle lifetime = a large electrical resistanceResistance is the relaxation of the current (momentum). A separate transport calculation that includes Umklapp processes, impurities, and vertex corrections is needed9
(a)–(d) at second order are also new second-order self-energiesReducible diagrams = iteration of first-order blocks. The Dyson equation generates them automatically6, 9
The “core of the correlation” is (e) and (f) of Fig. 9.8(e) and (f) of Fig. 9.16 = (i) and (j) of Fig. 9.8. Beware of mixing up the labels6

11. The Phonon Green’s Function — The Harmonic Oscillator Is G0, and the Anharmonic Terms Are the Vertices

From here on we change the lead role from the electron to the phonon (corresponding to tab ④ of the simulator). What corresponds to the electron’s “sea of non-interacting particles” is the harmonic oscillator, and what corresponds to the electron’s Coulomb interaction is the cubic and quartic terms of the atomic displacement. The notation follows Tadano and Tsuneyuki, PRB 92, 054301 (2015) [the SCP paper of ALAMODE]. Expanding the interatomic potential in the displacement u and writing the displacement in terms of the normal coordinate Aq = bq + b†−q, we have

U=U0+U2+U3+U4+⋯ Un=1n! (ℏ2)n/2 ∑{q} Δ(q1+⋯+qn) Φ(q1,…,qn) ωq1⋯ωqn Aq1⋯Aqn (16) Anharmonic terms (Tadano Eq. (13))

Φ(q1,…,qn) is the nth-order force constant transformed to normal coordinates (divided by the masses), and Δ expresses momentum conservation. The one-phonon Green’s function is the imaginary-time correlation function obtained by replacing ψ, ψ† of the electron case with Aq:

Gq(τ)= ⟨TτAq(τ) Aq†(0)⟩ , Gq0(iωm)= −2ωq (iωm)2−ωq2 (17) Phonon G0 (Tadano Eqs. (5), (7))

Compared with the electron’s G0 = 1/(ω − ωk), the phonon has poles at both positive and negative frequencies (because A = b + b† contains both creation and annihilation). Written in one line, the calculation in the harmonic-oscillator notes (“the response function in imaginary time” in lecture notes 2/14) is as follows: take ⟨A(τ)A(0)⟩ = (n+1)e−ωqτ + n eωqτ and apply the transform ∫0βħdτ eiωmτ; the temperature then disappears through the two identities (n+1)(1−e−βħω) = 1 and n(eβħω−1) = 1, and (17) results. Why imaginary time (Q1 of the lecture notes)? Because it lets us combine the finite-temperature expectation value Tr[e−βH…] and the time evolution e−iHt/ħ into a single exponential, so that Wick’s theorem and Feynman diagrams can be used as they are; for the retarded function at real frequency, we return at the end by the analytic continuation iωm → ω + iε. The Dyson equation has the same form as for electrons, and for matrices (including the mixing of bands j, j′) it reads

[Gq(ω)]−1 = [Gq0(ω)]−1 −Σq(ω) ⟹ det[ω2−Vq(ω)]=0 Vqjj′= ωqj2δjj′ −(2ωqj)12 (2ωqj′)12 Σqjj′ (18) Dyson equation and renormalized frequencies (Tadano Eqs. (8)–(11))

For a single mode, Ω2 = ωq2 − 2ωqΣq(Ω), that is, Ω ≈ ωq − Σq (in this sign convention, Σ < 0 means hardening). In what follows, for ease of reading, we speak in terms of the “frequency shift” Δ ≡ Ω − ωq and the “linewidth” Γ ≡ the Im part (Γ is the half width at half maximum, and the lifetime is τ = 1/(2Γ)).

12. Feynman Diagrams of the Cubic and Quartic Terms — Loop, Tadpole, Bubble

Just as the six first-order diagrams for electrons were gathered into Hartree and Fock, for phonons three proper diagrams remain up to second order in the coupling constants (the column of diagrams in tab ④ of the simulator. The wavy lines are the phonon lines G0, and ● is the vertex of Φ3 (three legs) or Φ4 (four legs)).

Table 6. Proper self-energies (phonon) up to second order in the coupling constants, and their correspondence with the electron diagrams. The last row is for reference
DiagramOrderContentω dependenceLinewidthElectron counterpart
LoopFirst order in Φ4A mean field obtained by closing two of the four legs, giving ⟨A²⟩ = 2n+1. SCP adds this up self-consistentlyNone (real, static)0Hartree–Fock (Fig. 9.1)
TadpoleSecond order in Φ3One of the three legs goes through a line with q = 0 and ω = 0 to the loop of another three-leg vertex. A change of curvature due to a static displacement ⟨u⟩ (thermal expansion, relaxation of internal coordinates)None0The “density shift” of Hartree
BubbleSecond order in Φ3The external phonon breaks into two phonons and returns. If there is a final state with ω = ω1+ω2, there is an imaginary partYesYesPolarization bubble (i) (Fig. 9.8)
Second order in Φ4 (three lines)Second order in Φ4Three-phonon intermediate state. Omitted in Tadano 2015YesYes—
Σq(L)=− 12N ∑q1 ℏΦ(q,−q,q1,−q1) 4ωqωq1 [1+2n(ωq1)] Ωq2=ωq2+2ΩqIq (SCP, Iq≡−Σq(L)) (19) Loop (Tadano Eqs. (15)–(17))
Σq(B)(iωm)= 12N ∑q1,q2 ℏ|Φ(−q,q1,q2)|2 8ωqω1ω2 Δ(−q+q1+q2) F(iωm) F=∑σ=±1 [ 1+n1+n2 iωm+σ(ω1+ω2) + n2−n1 iωm+σ(ω1−ω2) ] (20) Bubble (Tadano Eqs. (18), (19))

Letting iωm → ω + iε, the imaginary part separates out of 1/(ω − ω1 − ω2 + iε) = P[…] − iπδ(ω − ω1 − ω2), and this is Tadano’s Eq. (30) — the linewidth copied into lecture notes 2/16:

Γq(ω)= πℏ16N ∑q′,q″ |Φ(−q,q′,q″)|2 ΩqΩq′Ωq″ × [ (nq′+nq″+1) δ(ω−Ωq′−Ωq″) −2(nq′−nq″) δ(ω−Ωq′+Ωq″) ] (21) Linewidth (Tadano Eq. (30))

The first term is the decay “one → two” (the Klemens process), and the second is merging with a thermally excited phonon; both are nonzero only when there is a final state that satisfies conservation of energy and momentum. At high temperature n ∝ kBT/ħω, so Γ ∝ T and the lifetime τ = 1/(2Γ) ∝ 1/T. The real part (frequency shift) is obtained by replacing the δ of the same expression with a principal value such as P[1/(ω − Ω′ − Ω″)], and if the frequency of the external line lies below the two-phonon continuum, it softens by level repulsion. The tadpole is not contained in Tadano’s paper (the end of Section IV B of the paper states explicitly that “thermal expansion, relaxation of internal coordinates, and the intrinsic shift due to the bubble are not considered”), but it is exactly the effect of the cubic term displacing the equilibrium position, so it cannot be left out when considering negative thermal expansion. Written for a single mode,

⟨x⟩=−3α⟨x2⟩Mω02 Δ(T)=3α⟨x⟩Mω0 =−9α2⟨x2⟩M2ω03 , ⟨x2⟩=ℏ2Mω0(2n+1) (22) Tadpole = change of curvature due to a static displacement (U = ½Mω0²x² + αx³ + βx⁴)

12.1 Reducing to one mode: the values of the three diagrams

Reduce the crystal formulas to a single oscillator (N = 1, one mode) and write H = ħω0(b†b+½) + gA³ + hA⁴, g = αλ3/2, h = βλ², λ = ħ/2Mω0. The on-shell (ω = ω0) values of the three diagrams are all proportional to ⟨A²⟩ ∝ 2n+1:

ℏΔL=12h(2n+1) , ℏΔT=−36g2ℏω0(2n+1) ℏΔB(ω0)=−24g2ℏω0(2n+1) (23) The three diagrams for one mode

For a single mode the imaginary part of the bubble vanishes because δ(ω0 − 2ω0) = 0 — a single anharmonic oscillator has no lifetime (its eigenstates are strictly stationary). Whether these are correct can be checked with Rayleigh–Schrödinger perturbation theory. The first order in the quartic term and the second order in the cubic term (the problem in Landau–Lifshitz, Quantum Mechanics, §38) give

En=ℏω0(n+12) +3h(2n2+2n+1) −g2ℏω0(30n2+30n+11) ℏωn+1,n=En+1−En =ℏω0+(n+1) [12h−60g2ℏω0] (24) Levels and transition frequencies from perturbation theory

At T = 0, 12h − 60g²/ħω0 = ħ(ΔL + ΔT + ΔB)|n=0 (12 − 36 − 24 = −60), in agreement with Eq. (23). At finite temperature, the absorption intensity for n → n+1 is (n+1)pn, so the weighted average of the transition frequency is, using ⟨(n+1)²⟩/⟨n+1⟩ = 2nB+1 (⟨n²⟩ = 2nB²+nB in Bose statistics), ħω0 + (2nB+1)[12h − 60g²/ħω0]. This is the same as the quasiparticle frequency ω0 + ΔL + ΔT + ΔB of the Dyson equation. In other words, the peak of the lowest-order Dyson equation gives the centroid of the exact spectrum (a comb of transitions lined up for each n).

13. The Morse-Type Potential — Checking the Diagrams Against the Exact Solution

The Morse-type potential (lecture notes 11/7) has expansion coefficients that are fixed by a single parameter, and its Schrödinger equation can be solved exactly:

UM(x)= D(1−e−ax)2 =12Mω02x2 −ξx3 +76ξ2Mω02x4+⋯ D=M3ω068ξ2 , a=2ξMω02 EnM= (n+12)ℏω0 −(ℏω0)24D (n+12)2 , (ℏω0)24D =2ℏ2ξ2M3ω04 (25) Morse type (Morse 1929) and the exact levels

Putting α = −ξ and β = (7/6)ξ²/Mω0² into Eq. (23) and rearranging with λ = ħ/2Mω0 and D gives 12h = (7/16)(ħω0)²/D, 36g²/ħω0 = (9/16)(ħω0)²/D, and 24g²/ħω0 = (6/16)(ħω0)²/D, so that

ℏ(ΔL+ΔT+ΔB) =7−9−616 (ℏω0)2D (2nB+1) =−(ℏω0)22D (2nB+1) Exact solution: ℏωn+1,n= ℏω0−(ℏω0)22D (n+1) Mean weighted by absorption intensity: ℏω0−(ℏω0)22D (2nB+1) (26) Check: sum of the three diagrams = mean of the exact solution
What is happening in the Morse potential The loop of the quartic term hardens the mode (+7), and the tadpole and bubble of the cubic term soften it (−9, −6), for a net of −8/16 = −½. Because the exact En closes as a quadratic in (n+½), perturbation theory up to second order in the coupling constants (= these three diagrams) becomes exact for every n, and the contributions of higher-order diagrams cancel one another. In other words, the Morse potential is the ideal model for seeing “what is right and what is missed at the lowest order of the Dyson equation”: the position of the peak (the centroid) comes out correctly, but the spread of the comb, in which the frequency differs for each n — the standard deviation (ħω0²/2D)√(2nB(nB+1)), ∝ T at high temperature — does not appear in the lowest-order Σ. This is not a lifetime but an “inhomogeneous broadening,” and in a crystal it is observed overlapping with the imaginary part of the bubble (the true lifetime).

Fig. 3. Spectrum A(ω) of the Morse type: the comb of the exact solution and the peak of the Dyson equation. Each line of the comb is an n → n+1 transition (intensity (n+1)pn), and the band is the mean ± standard deviation. The vertical lines are loop only (lowest order of SCP) and the sum of the three diagrams (Dyson). Units are ħ = ω0 = 1.

Fig. 4. Renormalized frequency Ω(T) and linewidth Γ(T). Left: loop only (hardening), the sum of three diagrams (Dyson, red dashed line), the mean ± σ of the exact-solution comb (green), and the lecture-notes substitution n → nB (gray). Right: Γ ∝ (2n+1) when a two-phonon continuum exists in the crystal, and the difference between taking the argument of the Bose factor as ω0/2 (energy conservation) or ω0 (the formula in the lecture notes).

13.1 On the substitution n → nB in the lecture notes

At the end of lecture notes 11/7, an internal energy is constructed by substituting the Bose–Einstein nB into the Morse levels: UM = (nB+½)ħω0 − (nB+½)²(ħω0)²/4D. The exact thermal average is ⟨(n+½)²⟩ = (nB+½)² + nB(nB+1) (the fluctuation of Bose statistics), so the substitution drops the fluctuation term nB(nB+1)(ħω0)²/4D (∝ T² at high temperature). For the same reason, the transition frequency with n → nB substituted, ω0 − (ħω0²/2D)(nB+1), softens less than the Dyson result with (2nB+1) (at high temperature it is half as large, and at low temperature the two agree), and this is the gray dashed line in Fig. 4. What the mean field (loop, SCP) picks up is the “second moment including fluctuations” ⟨A²⟩ = 2n+1, and this is the reason it becomes (2nB+1) rather than (nB+1). The Γ ∝ (2n+1) and τ ∝ tanh(ħΩ/2kBT) of lecture notes 2/16 become τ ∝ tanh(ħΩ/4kBT) when written in terms of the occupation numbers of the two daughter phonons (Ω/2 each) that satisfy the δ function (the Balkanski–Wallis–Haro formula Γ = Γ0[1 + 2/(eħω/2kBT − 1)] for the Raman linewidth of Si has this form).

One answer to Q2 of the lecture notes (is first-order perturbation theory enough for ALAMODE?) SCP (the self-consistent summation of loops) is the best one can do within a “static, real” Σ, and in the Morse potential it sees only the hardening of +7/16. The cubic terms responsible for the softening (−9/16 − 6/16) lie outside SCP, and Tadano’s paper also says that “bubbles, thermal expansion, and relaxation of internal coordinates are left for future work” (later versions of ALAMODE implement “SCP+B,” which adds the bubble on top of SCP perturbatively, and “SCP+T,” which includes the tadpole). First-order perturbation theory works well for the 1s state of the He atom because it is a system in which the perturbation is exhausted by a “static mean field”; quantities for which ω-dependent, second-order processes are essential, such as lifetimes and softening, do not come out at first order — the same conclusion as §5 and §9 for electrons.

14. Summary — As One Story

  1. 0th order: When one electron is added to a non-interacting Fermi sea, it propagates without decay, only its phase rotating (Eq. (3)).
  2. 1st order (Fig. 9.1 → Fig. 9.7): Of the six diagrams, the two vacuum bubbles cancel against the denominator, and the remaining four are equivalent in pairs and are gathered into Hartree (the field of the average density) and Fock (the exchange from antisymmetry) (Eqs. (6), (7)). Σ is real and static — the levels shift, but the lifetime remains ∞ (§5).
  3. 2nd order (Fig. 9.8 → Fig. 9.16): Ten diagrams = four reducible + six proper. Of the proper ones, four are dressings of internal lines, and the remaining (i) polarization bubble and (j) crossed exchange are the core of dynamic correlation (§6).
  4. (i) → RPA and GW: With the bare Coulomb interaction, (i) diverges at small q, and summing the rings to infinite order gives the screened W (Eq. (11)). Σ = iGW is the GW approximation.
  5. (j) → vertex correction: Crossed exchange cannot be built by bubble insertions and is the first correction Γ⁽¹⁾ beyond Γ = 1 (Eq. (12)).
  6. Phonons (§11–13): The harmonic oscillator is G0, and the cubic and quartic terms are the vertices. Loop (Φ4, static, hardening), tadpole (Φ3², static displacement, softening), and bubble (Φ3², ω-dependent, softening, linewidth). For the Morse type, the sum of the three diagrams, −(ħω0)²(2nB+1)/2D, coincides with the centroid of the exact transition frequencies, and the spread of the comb does not appear in the lowest-order Σ (Eqs. (23)–(26)).
  7. Lifetime: When real final states in the form of particle–hole pairs open up, Im Σ★ ≠ 0. The peak broadens and |G(t)| decays. In a Fermi liquid −Im Σ ∝ ξ²+(πkBT)², so the lifetime is longer at low temperature and near the Fermi surface (Eq. (15)).

Companion simulator: Green’s Functions and Feynman Diagrams — Adding an Electron to the Electron Sea and Removing It Again (④ Phonons: anharmonic-term diagrams and the Morse type). Related resource: the harmonic oscillator (an entrance to second quantization). The diagrams are quoted from Figs. 9.1, 9.7, 9.8, and 9.16 of Fetter–Walecka (Japanese translation); in Appendix C of the Mochizuki Research Group’s Introduction to Linear Response Theory (in Japanese), the same figures are called Fig. C.2, Fig. C.7, Fig. C.11, and Fig. C.12, respectively. Figs. 1 and 2 are teaching models for seeing the shapes and are not calculations for any particular material.
Main sources

  1. A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems (McGraw-Hill, 1971) [Japanese translation: the theory volume], Chapter 3 (Wick’s theorem, Figs. 9.1–9.16, Dyson equation).
  2. R. D. Mattuck, A Guide to Feynman Diagrams in the Many-Body Problem, 2nd ed. (Dover, 1992). The best introduction to diagrams.
  3. G. D. Mahan, Many-Particle Physics, 3rd ed. (Kluwer, 2000) Chapter 5 (second-order Born approximation).
  4. L. Hedin, “New Method for Calculating the One-Particle Green's Function with Application to the Electron-Gas Problem,” Phys. Rev. 139, A796 (1965). doi:10.1103/PhysRev.139.A796
  5. G. Onida, L. Reining, and A. Rubio, “Electronic excitations: density-functional versus many-body Green's-function approaches,” Rev. Mod. Phys. 74, 601 (2002). doi:10.1103/RevModPhys.74.601
  6. N. E. Dahlen and R. van Leeuwen, “Self-consistent solution of the Dyson equation for atoms and molecules within a conserving approximation,” J. Chem. Phys. 122, 164102 (2005). doi:10.1063/1.1884965 (explicit formula for the second-order self-energy = GF2)
  7. T. Tadano and S. Tsuneyuki, “Self-consistent phonon calculations of lattice dynamical properties in cubic SrTiO3 with first-principles anharmonic force constants,” Phys. Rev. B 92, 054301 (2015). doi. The source of Eqs. (16)–(21) in §11–12 (the “Tadano Eq.” numbers refer to the original paper). T. Tadano, Y. Gohda, and S. Tsuneyuki, J. Phys.: Condens. Matter 26, 225402 (2014): ALAMODE.
  8. A. A. Maradudin and A. E. Fein, “Scattering of Neutrons by an Anharmonic Crystal,” Phys. Rev. 128, 2589 (1962); R. A. Cowley, “The lattice dynamics of an anharmonic crystal,” Adv. Phys. 12, 421 (1963). The original sources for the self-energies of the cubic and quartic terms (loop, tadpole, bubble).
  9. N. R. Werthamer, “Self-Consistent Phonon Formulation of Anharmonic Lattice Dynamics,” Phys. Rev. B 1, 572 (1970). The SCP theory.
  10. A. Togo, L. Chaput, and I. Tanaka, “Distributions of phonon lifetimes in Brillouin zones,” Phys. Rev. B 91, 094306 (2015). First-principles calculation of the imaginary part of the bubble (the linewidth).
  11. P. M. Morse, “Diatomic Molecules According to the Wave Mechanics. II. Vibrational Levels,” Phys. Rev. 34, 57 (1929). The exact solution of the Morse type.
  12. P. G. Klemens, “Anharmonic Decay of Optical Phonons,” Phys. Rev. 148, 845 (1966); M. Balkanski, R. F. Wallis, and E. Haro, “Anharmonic effects in light scattering due to optical phonons in silicon,” Phys. Rev. B 28, 1928 (1983). Two-phonon decay and the linewidth Γ0[1 + 2n(ω/2)].
  13. L. D. Landau and E. M. Lifshitz, Quantum Mechanics, 3rd ed. (Pergamon, 1977) §38, Problem 3: perturbation theory for the anharmonic oscillator (Eq. (24)).
  14. J. J. Phillips and D. Zgid, “Communication: The description of strong correlation within self-consistent Green's function second-order perturbation theory,” J. Chem. Phys. 140, 241101 (2014). doi:10.1063/1.4884951