University of Kurdistan logoNanoelectronics · Week 13 Study ToolkitProf. Mohammad Razaghi
University of Kurdistan logo
WEEK 13 · THE NEGF METHOD

The NEGF method: contacts for the Schrödinger equation

Study toolkit for students · based on S. Datta, Lessons from Nanoelectronics, Part B, Chapter 18
Prof. Mohammad Razaghi · University of Kurdistan

1This week's plan

Read
Datta Part B, Chapter 18: NEGF Method (Secs. 18.1–18.4).
Run
Colab notebook week13_negf.ipynb: the rate equation, the Lorentzian, incoherent sources, the NEGF function negf(), current and conductance, self-consistent dephasing.
Practise
Quiz (aim for 8/10), the practice problems and one mini-investigation. Next week these equations are applied to wires (C7 follows).

By the end of the week you should be able to

  1. Solve the one-level resistor with the rate equation and with E ψ = (ε − iγ/2)ψ + s, and show the two currents agree.
  2. Explain broadening: D(E) is a Lorentzian of width γ = γ₁ + γ₂ and area one; τ = ħ/γ.
  3. Explain why sources add incoherently and why Gn ~ ψψ† is the right object.
  4. Write GR, Gn, A, the current Ĩm, and G(E) = (q²/h) Trace[Γ₁GRΓ₂GA].
  5. State Σ₀ = D × GR and what Models A and B destroy.

Optional video: Datta's nanoHUB-U course, Unit 2: L2.2–L2.5, L2.9.

2Key ideas in five sentences

  1. Contacts in the Schrödinger equation. E ψ = [H + Σ₁ + Σ₂] ψ + s₁: outflow (non-Hermitian Σ) and inflow (s₁) from a contact; Γ = i[Σ − Σ†] plays the role of γ in the one-level case.
  2. Broadening. Coupling spreads a sharp level into D(E) = (γ/2π)/[(E − ε)² + (γ/2)²]; the current agrees with the rate equation, I = (q/ħ)γ₁γ₂/γ, but now we know the energy resolution.
  3. Incoherent sources. Different contacts do not interfere; we insert one source at a time and add the bilinear quantities Gn = GRΣinGA.
  4. Current and conductance. Ĩm = (q/h) Trace[ΣinmA − ΓmGn], and for coherent transport G(E) = (q²/h) Trace[Γ₁GRΓ₂GA].
  5. Dephasing. Fluctuating potentials add Σ₀ = D × GR, Σ₀in = D × Gn; Model A (Dij = D₀) destroys phase only, Model B (Dij = D₀δij) phase and momentum.

3Equation sheet

dN/dt = −(ν₁ + ν₂)N + S₁ + S₂,   Sm = νmfmEqs. 18.5–18.7
Semiclassical one-level rate equation.
N = (ν₁f₁ + ν₂f₂)/(ν₁ + ν₂),   I = q ν₁ν₂/(ν₁ + ν₂) (f₁ − f₂)Eq. 18.10
Steady state.
E ψ = (ε − iγ/2)ψ + s₁,   ψ = s₁/(E − ε + iγ/2),   γ = ħ(ν₁ + ν₂)Eqs. 18.13–18.16
Contact-ed Schrödinger equation.
2π s₁s₁* = γ₁,   D(E) = (γ/2π)/[(E − ε)² + (γ/2)²]Eqs. 18.20, 18.23
Source strength; broadened level.
I = (q/h) ∫ dE γ₁γ₂/[(E − ε)² + (γ/2)²] (f₁ − f₂)Eq. 18.22
One-level current; peak G = q²/h for γ₁ = γ₂.
E{ψ} = [H + Σ₁ + Σ₂]{ψ} + {s₁}Eq. 18.24
N-level channel.
GR = [E I − H − Σ]−1,   Gn = GRΣinGAEqs. 18.1–18.2
Σin = Γ₁f₁ + Γ₂f₂ (18.26).
A = GRΓGA = i[GR − GA],   Γ = i[Σ − Σ†]Eq. 18.3
Spectral function (2π × DOS).
Ĩm = (q/h) Trace[ΣinmA − ΓmGn]Eq. 18.4
Current per unit energy at contact m.
G(E) = (q²/h) Trace[Γ₁GRΓ₂GA]Eq. 18.30
Coherent conductance function.
Ĩm = (q/h) Σn T̄mn(fm − fn)Eqs. 18.31–18.32
Multi-terminal (Büttiker).
Σ₀ = D × GR,   Σ₀in = D × Gn,   Γ₀ = D × AEq. 18.35
Elastic dephasing; Dij = ⟨UR(i)UR(j)⟩.

4Physics refresher: rates, Lorentzians, matrix inverses

Rate equation

dN/dt = −νN + S gives N(t) = (S/ν)(1 − e−νt): the leaky bucket. The steady state S/ν is where inflow equals outflow.

Lorentzian

L(E) = (γ/2π)/[(E − ε)² + (γ/2)²] has half-maximum width γ and area 1. Its Fourier transform in time is e−iεt/ħe−γ|t|/2ħ: exponential decay of the level.

Why ∫dE γ/[(E−ε)² + (γ/2)²] = 2πSubstitute E − ε = (γ/2) tan θ; dE = (γ/2) sec²θ dθ and the denominator is (γ/2)² sec²θ, so the integrand becomes (γ/2)·(2/γ)·... = 2 dθ over (−π/2, π/2): total 2π.
Why the current is a traceTrace[ψψ†] = ψ†ψ is the number of electrons. Its time derivative is a sum of terms, one for each contact; each is a trace of a matrix product (the trace of AB equals the trace of BA).
Matrix inverse numericallyNever invert explicitly if you only need GRs: np.linalg.solve(E*I - H - Sigma, s). For a tridiagonal channel the cost is linear in N.

5Glossary

TermMeaning
Self-energy ΣMatrix added to H describing the coupling to a contact; its anti-Hermitian part is Γ.
Broadening ΓΓ = i[Σ − Σ†]: how easily electrons enter and leave the channel (γ for one level).
Green's function GR[E I − H − Σ]−1: the response of the channel to a source.
Spectral function A2π × the (matrix) density of states.
Electron density Gn2π × (matrix) number of electrons per unit energy.
In-scattering ΣinRate of injection of electrons: Γ1f1 + Γ2f2 (+ Σ₀in).
Transmission T̄(E)Trace[Γ₁GRΓ₂GA]: conductance function in units of q²/h.
Dephasing Σ₀Self-energy of a conceptual contact that randomises the phase (Model A) or phase and momentum (Model B).

6Calculators

One-level resistor

γ = γ₁ + γ₂
Dwell time τ = ħ/γ
Peak G = 4γ₁γ₂/γ² (q²/h)
G at µ₀ for kT ≪ γ (q²/h)
Rate-equation I, f₁ = 1, f₂ = 0
N, f₁ = 1, f₂ = 0

One-level current at T = 0

I = (q/h)∫windowT dE
Rate-equation limit (q/ħ)γ₁γ₂/γ
Ratio

Two-site molecule

T(E) = γ₁γ₂t²/|det|²
Levels of isolated H (eV)

7Self-check quiz

Click an answer: you get instant feedback. Score: 0 / 0

8Practice problems

ħ = 0.6582 meV·ps; q²/h = 38.74 µS; h/q² = 25.81 kΩ; kT = 25.85 meV at 300 K.

P1. Dwell timeEASY

A level is coupled with γ₁ = γ₂ = 20 meV. Find its dwell time and the width of its D(E).

Solutionγ = 40 meV ⇒ D(E) has full width at half maximum 40 meV; τ = ħ/γ = 0.6582/40 ps = 16.5 fs.
P2. Rate-equation currentEASY

Same device, f₁ = 1, f₂ = 0. Find N and I.

SolutionN = ν₁/(ν₁ + ν₂) = 0.5; ν = γ/ħ = 3.04 × 10¹³ s⁻¹; I = qν₁ν₂/(ν₁ + ν₂) = 2.43 µA.
P3. Unequal contactsEASY

γ₁ = 10 meV, γ₂ = 40 meV at very low temperature. Peak conductance and resistance?

Solution4γ₁γ₂/γ² = 4 × 400/2500 = 0.64 q²/h = 24.8 µS ⇒ R = 25.81/0.64 = 40.3 kΩ.
P4. Incoherent sourcesMEDIUM

Two sources s₁, s₂ with |s₁|² = |s₂|² have a random relative phase. What is ⟨ψψ*⟩ at E = ε compared with the coherent in-phase and opposite-phase cases?

HintAverage the cross term.
SolutionCoherent in phase: 4|s₁|²/(γ/2)²; opposite phase: 0; random phase: the cross terms vanish on average, leaving 2|s₁|²/(γ/2)², half the in-phase value.
P5. Window much larger than γMEDIUM

ε = 0, γ₁ = γ₂ = 20 meV, qV = 200 meV, T = 0. Find I from Eq. 18.22 and compare with the rate-equation value 2.43 µA.

SolutionI = (q/h)(2γ₁γ₂/γ)[atan(2(qV/2 − ε)/γ) − atan(2(−qV/2 − ε)/γ)] = (q/h)(0.02 eV)(2 atan 5) = 38.74 µS × 0.0549 V = 2.13 µA, 87 % of 2.43 µA: the Lorentzian tails outside the window are lost.
P6. N = 1 from the general formulaMEDIUM

Show that Trace[Γ₁GRΓ₂GA] = γ₁γ₂/[(E − ε)² + (γ/2)²] when all matrices are numbers.

SolutionGR = 1/(E − ε + iγ/2), so GRGA = 1/[(E − ε)² + γ²/4], and Γ₁GRΓ₂GA = γ₁γ₂ × that.
P7. Two-site moleculeMEDIUM

ε₁ = ε₂ = 0, t = −1 eV, γ₁ = γ₂ = 0.1 eV on sites 1 and 2. T at E = ±1 eV and at E = 0?

Solutiondet = (E + 0.05i)² − 1. At E = 1: det = −0.0025 + 0.1i, |det|² = 0.0100 ⇒ T = γ²t²/|det|² = 0.01/0.0100 = 0.999. At E = 0: det = −1.0025 ⇒ T = 0.01/1.005 = 0.0100.
P8. Spectral sum ruleHARDER

Why does ∫ Trace[A] dE/2π equal N? Estimate the value if the integral is cut at ±4 eV for a five-site chain with γ = 0.1 eV.

SolutionEach level contributes a Lorentzian of area 1. The tails missing beyond ±4 eV amount to about γ/(π × distance) ≈ 0.1/(π × 2.3) per level: total ≈ 4.98, as in the notebook.
P9. Zero current at the dephasing contactHARDER

Show that Trace[Σ₀inA − Γ₀Gn] = 0 for Model A with Dij = D₀.

HintWrite the trace as a double sum.
SolutionΣij Dij[GnijAji − AijGnji]. Dij is symmetric, and swapping i ↔ j changes the sign of the bracket ⇒ the sum is 0.

9Common mistakes

MISTAKE"The one-level device has conductance q²/h whatever the contacts."
CORRECTOnly for γ₁ = γ₂ and kT ≪ γ; otherwise the peak is 4γ₁γ₂/γ² of q²/h (and thermal smearing lowers it when kT is not small).
MISTAKE"Σ is Hermitian like H."
CORRECTΣ is non-Hermitian: its anti-Hermitian part Γ is the broadening, which makes the equation describe leakage.
MISTAKE"Gn = ψψ† for the whole device."
CORRECTGn adds the contributions ψψ† from each contact, weighted by its Fermi function; ψ alone belongs to one source.
MISTAKE"Dephasing is just averaging over random potentials in H."
CORRECTIt is not exactly equivalent: Σ₀ also averages over different sections of the same sample (Sec. 18.4).
MISTAKE"A and the density of states are the same thing."
CORRECTA = 2π × the matrix density of states: ∫ Trace[A] dE/2π = N.
MISTAKE"Model A and Model B give the same result."
CORRECTA keeps momentum, B randomises it; B also needs only the diagonal of G, A the full matrix.

10Explore

Files are in the course folder (course_materials).

Mini-investigations (15 min each)

  1. In the one-level widget, set γ₁ = γ₂ = 20 meV and find the temperature above which the peak of G(µ₀) falls below 0.8 q²/h. How does it compare with γ/kB?
  2. In the NEGF widget choose the benzene ring and compare T(E) for para and meta contacts at E = 0 and E = 1 eV. Explain the zero.
  3. In the notebook, change the dephasing strength D₀ from 0 to 0.5 eV² for Models A and B and plot the peak transmission of the middle level against D₀.

11Can I…?

Tick what you can do without notes. Your ticks are saved in this browser.

12Looking ahead: Week 14 and circular session C7

Using the equations (Datta Part B, Ch. 19–20), then C7: an NEGF lab.

  1. How does a wire get its contact self-energy t eika?
  2. Can two scatterers in series offer less resistance than one?
  3. Why does the conductance of a narrow 2D wire go in steps of q²/h?