Lectures II, III: Quantum Noise

Leonid Levitov (MIT)

Boulder 2005 Summer School
—Background on electron shot noise exp/th
— Scattering matrix approach (intro)
— Noise in mesoscopic systems: current partition, binomial statistics
• Counting statistics:
— Generating function for counting statistics
— Passive current detector; Keldysh partition function representation
• Tunneling
— Odd vs even moments, nonequilibrium FDT theorem
— Third moment S3
• Driven many-body systems
— Many particle problem → one particle problem (the determinant formula)
— Coherent electron pumping
— Phase-sensitive noise, ‘Mach-Zender effect’, orthogonality catastrophe
— Coherent many-body states — noise-minimizing current pulses
 Noise Introduction

Fluctuating current I(t)

Correlation function G2(τ ) = I(t)I(t + τ ) (time average, stationary flow)

Temporal correlations due to quantum statistics and/or source
Electrons counted in a flow, integrity preserved, without being pulled out of
a many-body system, no single-electron resolution yet — ensemble average
picture (but, turnstiles)
R ∞ −iωτ
Noise spectrum S(ω) = −∞ e S(τ )dτ ,
2
where S(τ ) = hhI(t)I(t + τ )ii = I(t)I(t + τ ) − I

L Levitov, Boulder 2005 Summer School Quantum Noise 1

 Electron transport
Coherent elastic scattering (mesoscopic systems, point contacts, etc.)
— scattering matrix approach
Interactions (nanotubes, quantum wires, QHE edge states) — Luttinger
liquid theory, QHE fractional charge theories
Quantum systems driven out of equilibrium (quantum dots, pumps,
turnstiles, qubits)
Current autocorrelation function:
1
G2(τ ) = h (I(t)I(t + τ ) + I(t + τ )I(t))i
2
(Note: no normal ordering, electrons counted without being destroyed)
If you’re an electron,

L Levitov, Boulder 2005 Summer School Quantum Noise 2

 Mesoscopic transport (crash course)
A quintessential example (point contact): 1d single channel QM scattering
~2 00
on a barrier, − 2m ψ + U (x)ψ = ψ. Scattering states in asymptotic form:

Functions u (x) Functions v (x)

ikx k k
e e
−ikx
1/2 ikx
t e 1/2
t
−ikx
e
1/2 −ikx
ir e ir1/2 eikx

P  
Express electric current through ψ(x) = k âk uk (x) + b̂k vk (x) :
+ 0 0
e
ei(k−k )x k+k +
P
ĵ(x) = 2m (−ψ (x)∂x ψ(x) + h.c.) = e k,k0 2m ψk0 (x)ψk (x)

X k + k0  +  √  
0 ak 0 t i rt ak
ĵ(x) = e ei(k−k )x + √ (x  0)
0
2m bk 0 −i rt r − 1 bk
k,k
Time-averagedP
current (at eV  EF only energies near EF contribute):
+ + dk
R
hj(x)i = evF k thak ak i + (r − 1)hbk bk i = evF t 2π ~ [nL() − nR ()] =
2
(et/h) [f ( − eV ) − f ()] d = eh tV
R

e2
Ohm’s law: I = gV (IR = V ) with conductance g = 1/R = ht (Landauer)

L Levitov, Boulder 2005 Summer School Quantum Noise 3

 Multiterminal system, reservoirs, scattering states
√ √ 
t i√ r
Single channel S-matrix: S = √ (optical beam splitter)
i r t
Scattering manifest in transport — quantization of g in point contacts
of adjustable width (many parallel channels which open one by one as a
function of Vgate )
2e2
P
g= h n tn

Conductance quantum:
2e2/h = 1/13 kΩ−1
adapted from van Wees et al. (1991)

L Levitov, Boulder 2005 Summer School Quantum Noise 4

 Electron beam partitioning

e e e e e

Incident electrons transmitted/reflected with probabilities t, r = 1 − t;

At T = 0, bias voltage eV = µL − µL, transport only at µR <  < µL:

(i) Fully filled Fermi at  < µL, µR; (ii) All empty states at  > µL, µR.
e2
R µL d
Mean current I = 2eNL→R = e µ t 2π~ = ~ tV (two spin channels)
R

Noiseless source (zero temperature) — binomial statistics with the number

of attempts during time τ : Nτ = (µL − µR)τ /2π~ = eV τ /h
Transmitted charge hQi = 2eNτ τ = (2e2/h)V τ — agrees with microscopic
calculation!

L Levitov, Boulder 2005 Summer School Quantum Noise 5

 Mesoscopic noise
Find noise power spectrum for a single channel conductor (no spin):
Sω = hhĵ(t)ĵ(0) + ĵ(0)ĵ(t)iie−iωtdt „ « „ √ «„ «
+
−i(k −k0 )t a k0
t i rt ak
√
P
ĵ(t) = 0 ev F e
k,k b+
k0
−i rt r − 1 bk
√ 2 EE
a+
»„ « „ « „ «–
P 2
DD
k 0 t
√ i rt a k
Sω=0 = k,k0 (evF ) δ(k − k0 ) +
bk 0 −i rt r − 1 bk
P 2 ˆ + +
√ + +
˜
= k e vF hh t(ak ak − bk bk ) + i rt(ak bk − bk ak ) × [h.c.]ii

Averaging with the help of Wick’s theorem, obtain

e2
d t2 (nL(1 − nL) + nR (1 − nR )) + rt(nL(1 − nR ) + nR (1 − nL))
R ˆ ˜
S0 = h

For reservoirs at equilibrium, with nL,R () = f ( ∓ 12 eV ), have

(
e2 2
 eV
 gkT eV  kT, thermal noise;
S0 = h t kT + rteV coth 2kT =
re2geV eV  kT, shot noise

Note: S0(eV  kT ) = re2 I — Shottky noise, suppressed by r = 1 − t (Lesovik ’89)

L Levitov, Boulder 2005 Summer School Quantum Noise 6

 Noise due to electron beam partitioning

e e e e e

Incident electrons with µR <  < µL transmitted/reflected with probabilities

t, r = 1 − t (No transport at  < µL, µR and  > µL, µR)
Noiseless source (zero temperature) — binomial statistics with the number
of attempts during time τ : Nτ = (µL − µR)τ /2π~ = eV τ /h
Probability of m out of Nτ electrons to be transmitted:
m m N −m m
Pm = CN t r (CN = N !/m!(N − m)! — binomial coefficients)
PN
Mean value: m = 0 mPm = t∂t(t + r)N = tN
Variance: δm2 = m2 − m2 = (t∂t)2(t + r)N − m2 = rtN
Variance = (1 − t) × Mean
— agrees with microscopic calculation!

L Levitov, Boulder 2005 Summer School Quantum Noise 7

 Noise in a point contact, experiment

2e2
P
Shot noise summed over channels, S0 = n h tn (1
− tn) — minima on
QPC conductance plateaus (adapted from Reznikov et al. ’95)

L Levitov, Boulder 2005 Summer School Quantum Noise 8

Example from optics: photon beam splitter with noiseless source
Consider n identical photons, in a number state |ni, incident on a beam
splitter.
  √ √  
ain t i√ r aout
= √
bin i r t bout

|ni = √1 (a+ )n |0i

n! in
√ √ + n
= √1 (
n!
ta+
out + i rbout ) |0i
Pm=n
√1 n−m
Cnmtm/2r (n−m)/2(a+ m + n−m


= n! m=0 i out ) (bout) |0i
Pm=n n−m m 1/2 m/2 (n−m)/2
= m=0 i (C n ) t r |n, n − mi

Probability to transmit m out of n photons is Pm = Cnmtmr n−m — binomial

statistics
Coherent, but noiseless, source → classical beam partitioning

L Levitov, Boulder 2005 Summer School Quantum Noise 9

 Shot noise highlights (for experts)
• Noise suppression relative to Scottky noise S2 = eI (tunneling current). In a point
contact S2 = (1 − t)eI (Lesovik ’89, Khlus ’87)

• Multiterminal, multichannel generalization; Relation to the random matrix theory;

Universal 1/3 reduction in mesoscopic conductors (Büttiker ’90, Beenakker ’92)

• Measured in a point contact (Reznikov ’95, Glattli ’96)

• Measured in a mesoscopic wire (Steinbach, Martinis, Devoret ’96, Schoelkopf ’97)

• Fractional charge noise in QHE (Kane, Fisher ’94, de Picciotto, Reznikov ’97, Glattli
’97 (ν = 1/3), Reznikov ’99(ν = 2/5))

• Phase-sensitive (photon-assisted) noise (Schoelkopf ’98, Glattli ’02)

• Noise in NS structures, charge doubling (Kozhevnikov, Schoelkopf, Prober ’00)

• Luttinger liquid, nanotubes (Yamamoto, ’03)

• QHE system; Kondo quantum dots; Noise near 0.7e2/h structure in QPC (Glattli
’04)
• Third moment S3 measurement (Reulet, Prober ’03, Reznikov ’04)

L Levitov, Boulder 2005 Summer School Quantum Noise 10

 Experimental issues, briefly
• Actually measured is not electric current but EM field. Photons
detached from matter, transmitted by ∼ 1 m, amplified, and detected;
• Matter-to-field conversion harmless if there is no backaction;
• Device + leads + environment. Engineer the circuit so that the
interesting noise dominates (e.g. a tunnel junction or point contact of high
impedance)
• Detune from 1/f noise
• Limitations due to heating and detection sensitivity

L Levitov, Boulder 2005 Summer School Quantum Noise 11

 Full counting statistics

L Levitov, Boulder 2005 Summer School Quantum Noise 12

 Counting statistics generating function
Probability distribution Pn → cumulants mk = hhnk ii
m1 = n, the mean value;
m2 = δn2 = n2 − n2, the variance;
m3 = δn3 = (n − n)3, the skewness;
...
P iλk
Generating function χ(λ) = n e Pk (defined by Fourier transform),
X mk
ln [χ(λ)] = (iλ)k
k!
k>0

While Pn is more easy to measure, χ(λ) is more easy to calculate!

The advantage of χ(λ) over Pn similar to partition function in a ‘grand
canonical ensemble’ approach
n n
Ex I: Binomial distribution, Pn = CN p (1 − p)N −n with N the number
of attempts, p the success probability → χ(λ) = (peiλ + 1 − p)N
n̄n −n̄
Ex II: Poisson distribution, Pn = n! e , χ(λ) = exp(n̄(eiλ − 1))

L Levitov, Boulder 2005 Summer School Quantum Noise 13

 Counting statistics, microscopic formula I
WANTED:
P A microscopic expression for the generating function
iλq
χ(λ) = q P (q)e for a generic many-body system
Spin 1/2 coupled to current: Hel,spin = Hel (p − aσ3, q)
Counting field a = − λ2 x̂δ(x − x0) measures current through cross-section
x = x0
Ex. Coupling to classical current H = λ2 σ3I(t); time evolution of spin:
| ↑i = e−iθ(t)/2| ↑i, | ↓i = eiθ(t)/2| ↓i
Rt 0 0
Spin precesses in the XY plane, precession angle θ(t) = λ 0 I(t )dt
measures time-dependent transmitted charge
x=x0

 



 
 

  


  

 

  

  


  
  

  




 

θ (t)
Disclaimer: Our goal is to clarify microscopic picture of current
fluctuations, not to describe realistic measurement

L Levitov, Boulder 2005 Summer School Quantum Noise 14

 Counting statistics, microscopic formula II
In QM current is an operator and [I(t), I(t0)] 6= 0 → Need a more careful
analysis!
Spin density matrix evolution (ensemble-averaged):
(0) (0)
!
ρ↑↑ heiH−λt e−iHλtiel ρ↑↓
ρ(t) = he−iHtρ0eiHt iel = (0) (0)
heiHλte−iH−λt iel ρ↓↑ ρ↓↓
h...iel = Trel (...ρel )

Examine the classical current case: spin precesses by θn = λn for n

transmitted particles.
(0) (0) (0) (0)
! !
P ρ↑↑ e−iθn ρ↑↓ ρ↑↑ he iλn
iρ↑↓
ρ(t) = n Pn (0) (0) = (0) (0)
eiθn ρ↓↑ ρ↓↓ he−iλniρ↓↑ ρ↓↓

Identify heiθ(t)i = heiλq i with χ(λ), the generating function

L Levitov, Boulder 2005 Summer School Quantum Noise 15

 Main result:
χ(λ) is given by Keldysh partition function
I !
D E
χ(λ) = TK exp −i Ĥλ(t0)dt0
C0,t

with the counting field λ(t) = ±λ antisymmetric on the forward and

backward parts of the Keldysh contour C0,t ≡ [0 → t → 0]
Properties:
P
1. Normalization: q Pq = 1, since χ(λ = 0) = 1
χ(λ) dλ
R −iqλ
2. Pq = e 2π ≥ 0
3. Charge quantization: χ(λ) is 2π-periodic in λ (for noninteracting
particles)
Features:
— Describes not just spin 1/2 but a wide class of passive charge detectors,
such as heavy particle H = p2/2M − λf (t)q at large M (no recoil);
— Minimal backaction, measurement affects only forward scattering:
[H, σz ] = 0;
— Good for generic many-body system

L Levitov, Boulder 2005 Summer School Quantum Noise 16

 Compare: Larmor clock for QM collision
Nuclear reaction, tunneling, resonance scattering, etc. — How long does it take?

Add a fictitious spin 1/2 to particle: U (x) → Uef f = U (x) + 12 ω(x)σz

with fictitious field ω(x) nonzero in the spatial region of interest.

Potential barrier Resonance scattering

Spin precesses about the Z axis during collision: precession angle measures time.
Analysis similar to passive detector (one particle!) yields
Z
−1 −iωτ
χ(ω) = Tr(S−ω Sω ρ) = e P (τ )dτ

with P (τ ) interpreted as probability to spend time τ in the region of interest

− +iγ/2
Resonance scattering: S() = −0−iγ/2 gives χ(ω) = ω−∆−iγ ω−∆+iγ × ω+∆−iγ
ω+∆+iγ with
0
detuning ∆ = 2( − 0). Obtain positive or negative probabilities! (Caution)

L Levitov, Boulder 2005 Summer School Quantum Noise 17

 Variety of topics
• Tunneling problem. S2, S3 Nonequilibrium FDT theorem. Relation with
Glauber theory of photocounting.
• Driven many-body systems. For noninteracting particles (fermions or
bosons) χ(λ) can be expressed through time-dependent one-particle S-
matrix. Pumps, coherent current pulses, photon-assisted noise → next
lecture
• Mesoscopic noise in normal and superconducting systems (Nazarov,
Nagaev)
• Mesoscopic photon sources (Beenakker)
• Entangled EPR states, counting statistics (Fazio)
• Spin current noise (Lamacraft)
• Backaction of spin 1/2 counter (Muzykantsky)
• Role of environment (Kindermann)
• Quantum information/entropy (Callan & Wilczek, Vidal, Kitaev)
• Orthogonality catastrophe, Fermi-edge singularity

L Levitov, Boulder 2005 Summer School Quantum Noise 18

 Counting statistics of tunneling current

Focus on the tunneling problem (generic interacting system). Tunneling

Hamiltonian
Ĥ = Ĥ1 + Ĥ2 + V̂
where Ĥ1,2 and V̂ = Jˆ12 + Jˆ21) describe leads and tunneling coupling). The
counting field λ(t) is added to the phase of the tunneling operators Jˆ12, Jˆ21
as i i
V̂λ = e 2 λ(t)Jˆ12(t) + e− 2 λ(t)Jˆ21(t)
with λ0<t<τ = λ. (Justified using one-particle tunneling problem.)

Transform the bias voltage into a phase factor, Jˆ12 → Jˆ12e−ieV t,

Jˆ21 → Jˆ21eieV t. In the interaction representation, write

I !
D E
0 0
χ(λ) = TK exp −i V̂λ(t0)(t )dt
C0,t

using cumulant expansion, as a sum of linked cluster diagrams.

L Levitov, Boulder 2005 Summer School Quantum Noise 19

The lowest order in the tunneling coupling Jˆ12, Jˆ21 is given by linked clusters
of order two. Obtain χ(λ) = eW (λ), where

1
II D E
W (λ) = − TKV̂λ(t0)(t0)V̂λ(t00)(t00) dt0dt00
2 C0,t

More explicitly,

W (λ) = (eiλ −1)N1→2(t) + (e−iλ −1)N2→1(t)

Z tZ t Z tZ t
N1→2 = hJˆ21(t )Jˆ12(t )i dt dt
0 00 0 00
N2→1 = hJˆ12(t0)Jˆ21(t00)i dt0dt00
0 0 0 0
with Nj→k (t) = njk t the mean particle number transmitted between the
contacts in a time t (cf. Kubo formula).
Resulting counting statistics is bi-directional Poissonian:
iλ −iλ
 
χ(λ) = exp (e −1)N1→2(t) + (e −1)N2→1(t)

True in any interacting system, in the tunneling regime.

L Levitov, Boulder 2005 Summer School Quantum Noise 20

 Nonequilibrium Fluctuation-Dissipation theorem
P∞ (iλ)k hhδq k ii
The cumulants are generated as ln χ(λ) = k=1 k! qk
with q0 the
0
tunneling charge. Obtain
(
(n12 − n21)t, k odd
hhδq k ii = q0k
(n12 + n21)t, k even

Setting k = 1, 2, relate n12 ± n21 with the time-averaged current and the
low frequency noise power:

n12 − n21 = I/q0, n12 + n21 = S2/q02.

Relate the second and the first correlator:

S2 = hhδq 2ii/t = (N1→2 +N2→1)/(N1→2 −N2→1)q0I = coth(eV /2kBT )q0I

Nyquist for eV < kBT , Schottky for eV > kBT .

A universal relation — holds for any I − V characteristic (linear responce
not required, cf. FDT in equilibrium)

L Levitov, Boulder 2005 Summer School Quantum Noise 21

 Application in metrology

L Levitov, Boulder 2005 Summer School Quantum Noise 22

 Generalized Schottky formula for S3
3

3
Relate the third cumulant hhδq ii ≡ δq − δq = S3t with hδqi = It.
Obtain a Schottky-like relation for the third correlator spectral power S3:

S3 ≡ hhδq 3ii/t = q02I

— independent of the mean/variance ratio (n12 − n21)/(n12 + n21).

Since N1→2/N2→1 ≡ n12/n21 = exp(eV /kBT ) (detailed balance), the
relation S3 = q02I holds at any voltage/temperature ratio.

Good for using shot noise to determine particle charge in Luttinger liquids
and fractional QHE (heating limitation: S2 = q0I requires eV > kBT ).
A possibility to measure tunneling quasiparticle charge at temperatures
kBT ≥ eV

L Levitov, Boulder 2005 Summer School Quantum Noise 23

The 3-rd correlator in terms of the counting distribution profile:
hqi = It = mean; hhδq 2ii = S2t = variance; hhδq 3ii = S3t = skewness
−1
10

−2
10

Counting probability
−3
10

−4
10 N =20, N =0
12 21
Gaussian

−5
10
0 5 10 15 20 25 30 35 40
*
Transmitted charge q/e

The third moment determines skewness of the distribution P (q) profile. This is illustrated by a bi-directional Poissonian distribution
and a Gaussian with the same mean and variance. For S 3 > 0 the peak tails are stretched more to the right than to the left.

L Levitov, Boulder 2005 Summer School Quantum Noise 24

 Measurement of S3 in a Tunnel Junction:

First experiment, low impedance (50 Ohm) tunnel junction (B. Reulet, J.
Senzier, and D.E. Prober, Phys. Rev. Lett. 91, 196601 (2003));

High impedance junction (M. Reznikov et. al. ’04):

6
10
3
counts

4 0

S(3) ( 10−46 A3/Hz2)

S(2) (10−27 A2/Hz)

2 −10

2
Voltage 0 −20 −10 0 10
counts

counts
−2 1
1
0.8
−4
Voltage bin
bin number
number
0 −6
0 5 10 15 20 −20 −10 0 10 20
−9 −9
I (10 A) I (10 A)

Transmitted charge distribution and its moments S2 and S3. Poisson

behavior S3 = e2I confirmed.

L Levitov, Boulder 2005 Summer School Quantum Noise 25

 Tunneling summary:

1) The counting statistics of tunneling current is bi-directional Poissonian,

universally and independently of the character of interactions and thus of
the form of I − V dependence.
2) Nonequilibrium FDT relation S2 = coth(eV /2kBT )q0I
3) Shottky-like relation for the third correlator S3 = q02I at both large and
small eV /kBT .

L Levitov, Boulder 2005 Summer School Quantum Noise 26

 Counting statistics of a driven many-body system
Time-dependent external field and scattering (noninteracting fermions)
We obtain the generating function in the form of a functional determinant
in the single-particle Hilbert space:
  
χ(λ) = det 1̂ + n(t, t0) T̂λ(t) − 1̂

λ λ
† i 4i −i 4j
T̂λ(t) = S−λ (t)Sλ(t), Sλ(t)ij = e
S(t)ij e
i 0
with reservoirs density matrix n(t, t )T =0 = 2π(t−t0+iδ) = <0 e−i(t−t ), a
0
P

time-dependent S-matrix S(t) and separate λi for each channel

2 3

1
4
... 5

Time-dependent field (voltage V (t), etc.) included in Sλ(t). No time delay:

S(t, t0) ' S(t)δ(t − t0) (instant scattering apprx. — nonessential)

L Levitov, Boulder 2005 Summer School Quantum Noise 27

Simple and not-so-simple facts from matrix alebra
Useful relations between 2-nd quantized and single-particle operators:

N
X
A −→ Γ(A) = Aij a+
i aj
i,j=1

(mapping of matrices N × N −→ 2N × 2N ).

Γ(A) A


Tr e = det 1 + e
— fermion partition function Z = Tr e−βH with −βH = Γ(A))
Note: For A = 0 obtain 2N = 2N
 
Γ(A) Γ(B) A B


Tr e e = det 1 + e e
 
Γ(A) Γ(B) Γ(C) A B C


Tr e = det 1 + e e e
e e
...
Proven using Baker-Hausdorff series for ln(eX eY ) (commutator algebra for
X, Y the same as for Γ(X), Γ(Y ))

L Levitov, Boulder 2005 Summer School Quantum Noise 28

 Deriving the determinant formula I
Write c.s. generating function as

eiH−λte−iHλt


χ(λ) = Tr ρel

with H±λ = Γ(h±λ), ρ = Z1 e−βH0 . Obtain

“ ” »“ ”−1 “ ”–
−1
χ=Z det 1+ e−βh0 eih−λt e−ihλt = det 1+e −βh0
1+ e−βh0 eih−λte−ihλt

βh0 −1


Finally, with n̂ = 1 + e , have

n̂eih−λte−ihλt


χ(λ) = det 1 − n̂ +

L Levitov, Boulder 2005 Summer School Quantum Noise 29

 Deriving the determinant formula II
Relate forward-and-backward evolution in time with scattering operator:
t
-1
S1 S0 ψ
^
-ih τ
e 1

-1
S0 ψ
^
ih τ
e 0
ψ x

−1
Thus ht0|eih−λte−ihλt|ti = S−λ (t)Sλ(t)δ(t − t0)
with |ti a wavepacket arriving at scatterer at time t.
−1


χ(λ) = det 1 − n̂ + n̂S−λ Sλ

A single-particle quantity — Fermi-statistics accounted for by det!

(Generalize to bosons: det(1 + ...) −→ det(1 − ...)−1)

L Levitov, Boulder 2005 Summer School Quantum Noise 30

 A more intuitive approach: DC transport
For elastic scattering different energies contribute independently:
 Z 
Y d
χ(λ) = χ(λ) , i.e. χ(λ) = exp t ln χ(λ) ,

2π~

(quasiclassical dV = ddt/2π~).

2 3

1
4
... 5

Sum over all multiparticle processes:

X i
χ(λ) = e 2 (λi1 +...+λik −λj1 −...−λjk ) Pi1,...,ik | j1,...,jk ,
i1 ,...,ik ,j1 ,...,jk

L Levitov, Boulder 2005 Summer School Quantum Noise 31

with the rate of k particles transit i1, ..., ik −→ j1, ..., jk is given by

2
j ,...,j
Y Y
Pi1,...,ik | j1,...,jk = Si11,...,ikk (1 − ni()) ni() .
i6=iα i=iα

j ,...,j
with Si11,...,ikk antisymmetrized product of k single particle amplitudes.
This is equal to our determinant (Proven by reverse engineering)
Positive probabilities!

L Levitov, Boulder 2005 Summer School Quantum Noise 32

 Point contact (beam splitter): 2 × 2 matrices

i
χ(λ) = (1 − n1)(1 − n2) + (|S11|2 + e 2 (λ2−λ1)|S21|2)n1(1 − n2)
2 i
+ (|S22| + e 2 (λ1 −λ2 ) |S12|2)n2(1 − n1) + |det S|2n1n2 , (1)

with n1,2() = f (∓ 21 eV ) and Sij is unitary: |S1i|2 +|S2i|2 = 1, |det S| = 1.

Simplify: χ(λ) = 1 + t(eiλ − 1)n1(1 − n2) + t(e−iλ − 1)n2(1 − n1) Here

t = |S21|2 = |S12|2 is the transmission coefficient and λ = λ2 − λ1.
At T = 0, since nF () = 0 or 1, for V > 0 have
(
eiλt + 1 − t , || < 12 eV ;
χ(λ) =
1, || > 12 eV

Full counting statistics: χτ (λ) = (eiλp + 1 − p)N (τ ) — binomial, with the

number of attempts N (τ ) = (eV /h)τ .
Similar at V < 0, with eiλ −→ e−iλ (DC current sign reversal).

L Levitov, Boulder 2005 Summer School Quantum Noise 33

 Note: the noninteger number of attempts is an artifact of a quasiclassical
calculation. More careful analysis gives a narrow distribution PN of the
number of attempts
P peaked at N = N (τ ), and the generating function as a
weighted sum N PN χN (λ). The peak width is a sublinear function of the
measurement time τ (in fact, δN 2T =0 ∝ ln τ ), the statistics still binomial,
to leading order in t.

L Levitov, Boulder 2005 Summer School Quantum Noise 34

 Strategies for handling the determinant
Two strategies:
1) For periodic S(t), in the frequency representation n̂ is diagonal,
n(ω) = f (ω), while S(t) has matrix elements Sω0,ω with discrete frequency
change ω 0 − ω = nΩ, with Ω the pumping frequency. In this method the
energy axis is divided into intervals nΩ < ω < (n + 1)Ω, and each interval
is treated as a separate conduction channel with time-independent S-matrix
Sω0,ω .
2) The determinant can also be analyzed directly in the time domain:
h i
−1
∂λ ln χ(λ) = Tr (1 + n(Tλ − 1)) ∂λ Tλ

−1 i
with Tλ(t, t0) = S−λ (t)Sλ(t)δ(t − t0), n(t, t0) = 2π (t − t0 + iδ)−1.

The problem of inverting the integral operator R = 1 + n(Tλ − 1) is the

so-called Riemann-Hilbert problem (matrix generalization of Wiener-Hopf,
well-studied, exact and approximate solutions can be constructed)

L Levitov, Boulder 2005 Summer School Quantum Noise 35

 Phase-sensitive noise
Charge flow induced by voltage pulse V (t) in a point contact
A pulse V (t)corresponds to a step ∆θ in the forward scattering phase:
 −iθ(t)
√ √  t
e
Z
e t i r√
S(t) = √ iθ(t) θ(t) = V (t0)dt0
i r e t ~ −∞

The mean transmitted charge q = te∆θ/2π — independent of pulse shape

In contrast, the variance δq 2 exhibits complex dependence:
t2
1 − eiθ12 e
Z Z Z
δq 2 = 2t(1 − t)e2 dt1dt2 θ12 = V (t0)dt0
(t1 − t2)2 ~ t1

Gives δq 2 ∝ (1 − cos ∆θ) ln(tmax/t0) + const — periodic in the pulse area

and log-divergent (tmax ∼ ~/kBT )
Interpret the log as orthogonality catastrophe: long-lasting change of
scatterer at ∆θ 6= 2πn causes infinite number of soft particle-hole excitations
Shot noise is phase-sensitive — Mach-Zender effect in electron noise

L Levitov, Boulder 2005 Summer School Quantum Noise 36

 Minimizing unhappiness
t2 e
“ R ”
0 0
RR 1−exp i t ~ V (t )dt
1
Find noise-minimizing pulse shapes: (t1 −t2 )2
dt1dt2 → min
— an interesting variational problem, solved by pulses of integer area 2πn:

~ X 2τi
V (t) = (τi > 0)
e i=1...n (t − ti)2 + τi2

Lorentzian pulses (overlapping or nonoverlapping)

Degeneracy: δq 2 = e2t(1 − t)n, the same for all ti, τi

L Levitov, Boulder 2005 Summer School Quantum Noise 37

 Coherent time-dependent many-body states
Variance of transmitted charge δq 2 = e2t(1−t)n — independent of pulse
parameters ti, τi, same value as for binomial distribution with the number
of attempts n
Binomial counting statistics from the functional determinant (exactly
solvable Riemann-Hilbert problem):

iλ

n
χ(λ) = te +1−t

Interpretation: pulses ' independent attempts to transmit charge

Coherent current pulses:
Noise reduced as much as the beam splitter partition noise permits
Similarity to coherent states (QM uncertainty minimized)
Many-body objects which behave like songle particles. Fully entangled?

L Levitov, Boulder 2005 Summer School Quantum Noise 38

 Measurement of phase sensitive noise
Instead of a train of pulses (which is difficult to realize) used a
combination of DC and AC voltage, V = VDC + VAC cos Ωt — oscillations
in noise power (Bessel functions), while DC current is ohmic:
!
∂S0 2 3 X X 2
= e tm(1 − tm ) Jn(VAC /~Ω)θ(eVDC − n~Ω)
∂VDC π m n

Observed in mesoscopic wires (Schoelkopf ’98), point contacts (Glattli ’02)

L Levitov, Boulder 2005 Summer School Quantum Noise 39

 Case studies
(i) Voltage pulses of different signs (D Ivanov, H-W Lee, and LL, PRB 56, 6839 (1997)):
 
h 2τ1 2τ2
V (t) = −
e (t − t1)2 + τ12 (t − t2)2 + τ22

give rise to the counting distribution

iλ −iλ z1∗ − z2 2
χ(λ) = 1 − 2F + F (e +e ), F = t(1 − t)
z1 − z 2

with z1,2 = t1,2 + iτ1,2. The quantity A = |...|2 is a measure of pulses’

overlap in time: A = 0 (full overlap), A = 1 (no overlap).
Note: χ(λ) factorizes for nonoverlapping pulses
(ii) Two-channel model of electron pump ((D Ivanov and LL, JETP Lett 58, 461 (1993)):

0 −iΩτ iΩτ
   
r t B + be Ā + āe
S(τ ) ≡ = ,
t r0 A + ae−iΩτ −B̄ − b̄eiΩτ

L Levitov, Boulder 2005 Summer School Quantum Noise 40

which is unitary provided |A|2 + |a|2 + |B|2 + |b|2 = 1, Aā + B b̄ = 0
(time-independent parameters).
For T = 0 and µL = µR the charge distribution for m pumping cycles is
described by

iλ −iλ

m
χ(λ) = 1 + p1(e − 1) + p2(e − 1)

with p1 = |a|4/(|a|2 + |b|2) and p2 = |b|4/(|a|2 + |b|2): at each pumping

cycle an electron is pumped in one direction with probability p1, or in
the opposite direction with probability p2, or no charge is pumped with
probability 1 − p1 − p2.
Also can be solved at µL 6= µR: more complicated statistics.

L Levitov, Boulder 2005 Summer School Quantum Noise 41

 Counting statistics of a charge pump
DC current from an AC-driven open quantum dot. (Exp: Marcus group ’99)

The time-averaged pumped current is a purely geometric property of the

path in the S-matrix parameter space, insensitive to path parameterization
(Brouwer ’98, Buttiker’94).

Noise dependence on the pumping cycle? Bounds on the ratio noise/current?

Here, consider generic but small path V1(t), V2(t):
V1

V2

L Levitov, Boulder 2005 Summer School Quantum Noise 42

 Counting statistics for a generic pumping cycle
Focus on the weak pumping regime, a small loop in the matrix parameter
space: S(t) = eA(t)S (0) with perturbation A(t) (antihermitian, trA†A  1)
and S (0) is the S-matrix in the absence of pumping.
  
†
Expand ln χ(λ) = det 1̂ + n(t, t0) S−λ(t)Sλ(t) − 1̂ in A(t):

1

1
ln χ = tr n̂ A−λ + Aλ −2A−λAλ − tr(n̂Bλ)2
2 2
2 2
λ λ
with Aλ(t) = ei 4 σ3 A(t)e−i 4 σ3 , Bλ(t) = Aλ(t) − A−λ(t)
Using n̂2T =0 = n̂T =0, separate a commutator:

1 1 2 2

2


ln χ(λ) = tr (n̂[Aλ, A−λ]) + tr n̂ Bλ − tr(n̂Bλ)
2 2
The commutator is regularized as the Schwinger anomaly (splitting points,
t0, t00 = t ± /2), which gives

1
I
n(t0, t00)tr (A−λ(t00)Aλ(t0) − Aλ(t00)A−λ(t0)) dt
2

L Levitov, Boulder 2005 Summer School Quantum Noise 43

Average over small  (insert additional integrals over t0, t00, or just replace
Aλ(t) → 12 (Aλ(t) + Aλ(t0)), etc.) In the limit  → 0, obtain

i
I
(ln χ)1 = tr (A−λ∂tAλ − Aλ∂tA−λ) dt
8π

The second term of ln χ is rewritten as

2
1 tr (Bλ(t) − Bλ(t0))
II
0
(ln χ)2 = dtdt
4(2π)2 (t − t0)2

Now, combine (ln χ)1 with (ln χ)2, and simplify

L Levitov, Boulder 2005 Summer School Quantum Noise 44

 Bi-directional Poissonian statistics
†
Convenient decomposition A = a 0 + z + z , such that [σ3, a0] = 0,
[σ3, z] = −2z, σ3, z † = 2z †, gives


−i λ
4 σ3 iλ
4 σ3 iλ
2 † −i λ
Aλ ≡ e Ae = a0 + e z + e z 2
 λ

−i λ
B λ = ei 2 − e 2 W, W ≡ z † − z

In this representation,
2
sin λ (1 − cos λ) tr (W (t) − W (t0))
I II
0
ln χ = tr ([σ3, W ] ∂tW ) dt + dtdt
8π 2(2π)2 (t − t0)2

The first term is identical to the Brouwer result (invariant under

reparameterization and has a purely geometric character), the second term
describes noise.
Note the λ-dependence: u(eiλ − 1) + v(e−iλ − 1) → 2P statistics
iλ −iλ


χ(λ) = exp u(e − 1) + v(e − 1)

L Levitov, Boulder 2005 Summer School Quantum Noise 45

Prove that
1) u, v ≥ 0 for generic pumping cycle W (t);
2) u = 0 or v = 0 for special paths W (t) holomorphic in the upper/lower
half-plane of complex time t;
Then the ratio current/noise = (u − v)(u + v) is maximal or minimal
(equal ±q0−1 per cycle), when u = 0 or v = 0. The counting statistics in
this case is pure poissonian.

L Levitov, Boulder 2005 Summer School Quantum Noise 46

 Example of a single channel system (two leads) driven by V1(t) =
a1 cos(Ωt + θ), V2(t) = a2 cos(Ωt). The pumping cycle:
 
0 z(t)
W (t) = , z(t) = z1V1(t)+z2V2(t), (z1,2 system parameters)
z ∗(t) 0

−1
Current to noise ratio, I/J, in the units of e
1

0.5

−0.5

1.5
−1 1
−1.5 0.5
−1 −0.5 0
0 −0.5 Re w
Im w 0.5 1 −1
1.5 −1.5

Current to noise ratio, I/J = q −1 (u − v)/(u + v), as a function of the driving signal parameters for a single channel pump.
0
The two harmonic signals driving the system are characterized by relative amplitude and phase, w = (V 1 /V2 )eiθ . Maximum
and minimum, as a function of w , are I/J = ±q0 −1 .

L Levitov, Boulder 2005 Summer School Quantum Noise 47

 Pump noise summary:

1) Pumping noise is super-poissonian; counting statistics is double-

poissonian;
2) The current/noise ratio can be maximized by varying the pumping
cycle (relative amplitude or phase of the driving signals);
3) Extremal cycles correspond to poissonian counting statistics with a
universal ratio
current/noise = ±q0−1
(another generalization of the Schottky formula).

L Levitov, Boulder 2005 Summer School Quantum Noise 48