Quantum kinetic models of open quantum systems
in semiconductor theory
Chiara Manzini
Scuola Normale Superiore di Pisa
Westf¨alische Wilhelms-Universit¨at M¨unster
29 Aprile 2005
Outlook
1- Motivation
2- Quantum kinetic formulation 3- Analytical difficulties
4- Open quantum systems: two examples 5- The three well-posedness results
6- New tools and perspectives
7- Final considerations
1- Motivation
Trend in semiconductor technology: miniaturization
=⇒ Challenge: simulation tools
Features of novel devices =⇒ Quantum effects
Prototype: Resonant Tunneling Diode
Doping
Density
Density
Quantum description
QS Quantum System with d degrees of freedom:
• a pure state of QS ↔ ψ ∈ L
2(R
d; C),
• a physical state of QS ↔ ρ ˆ “density matrix” on L
2(R
d; C) , i.e. ρ(x, y) ∈ L
2(R
2d; C) kernel
ρ(x, y) = X
j∈ N
λ
jψ
j(x)ψ
j(y)
| {z }
pure state
,
{ψ
j}
jcomplete orthonormal set, λ
j≥ 0, P
j
λ
j= 1 . The position density:
n(x) = X
λ
j|ψ
j|
2(x) ∈ L
1(R
d; R
+) ⇔ finite mass.
Quantum Evolution: reversible dynamics
( i
dtdψ
j= −
12∆
xψ
j+ V ψ
j, j ∈ N, x ∈ R
d−∆V (x, t) = n(x, t) = P
j
λ
j|ψ
j|
2(x, t) ρ(x, y, t) = X
j∈ N
λ
jψ
j(x, t)ψ
j(y, t) , {ψ
j(t)}
j⊂ L
2(R
d; C) E.g. evolution of an electron ensemble with d d.o.f.,
ballistic regime, mean-field approximation
Quantum Evolution: reversible dynamics
( i
dtdψ
j= −
12∆
xψ
j+ V ψ
j, j ∈ N, x ∈ R
d−∆V (x, t) = n(x, t) = P
j
λ
j|ψ
j|
2(x, t) w(x, v) = F
η→vX
j∈N
λ
jψ
j(x + η
2 )ψ
j(x − η 2 )
Quantum Liouville equation: Wigner-Poisson
w
t+ v · ∇
xw − Θ[V ]w = 0
−∆
xV (x, t) = n[w](x, t) = R
w(x, v, t) dv
F
η→v{Θ[V ]w}(x, v) := i [V (x + η/2)−V (x − η/2)]F
v→ηw(x, η)
2- Quantum Kinetic description
A physical state of QS ←→ a quasiprobability on the phase space, w : (x, v) ∈ R
2d7→ R =⇒
w(x, v) := F
η→vρ
x + η
2 , x − η 2
∈ L
2(R
2d; R )
w is the Wigner function for the physical state of QS The position density
n[w](x, t) :=
Z
w(x, v, t) dv
w ∈ L
2(R
2d; R ) 6⇒ n[w] well-defined, non-negative
3- Analytical difficulties
• w ∈ L
2(R
2d; R ) NOT ⇒ n[w] well-defined w(x, v, t)
><0 ⇒ n[w](x, t) :=
Z
w(x, v, t) dv
><0 Quantum counterpart :
{ψ
j(t)}
j⊂ L
2(R
d; C), λ
j≥ 0, P
j
λ
j= 1
⇒ n(t) = P
j
λ
j|ψ
j|
2(t), kn(t)k
L1< ∞ , ∀ t
⇒ (mass cons.) kψ
j(t)k
2L2= const.
First a-priori estimate for Schr.-Poisson systems
3- Analytical difficulties
• w ∈ L
2(R
2d; R ) NOT ⇒ n[w] well-defined
• e
kin(x, t) :=
12R
|v|
2w(x, v, t) dv:
w(x, v, t)
><0 ⇒ e
kin(x, t)
><0 Quantum cp. : E
kin(t) :=
12P
j
λ
jk∇ψ
j(t)k
L21 2
P
j
λ
jk∇ψ
j(t)k
2L2+ 1
2 k∇V (t)k
2L2| {z }
Epot2 (t)
=const.
(energy cons.) ⇒ k∇ψ
j(t)k
2L2= const.
Advantages
• w ∈ L
2(R
2d; R ) is necessary for a physical state of QS
• L
2-setting is suitable for numerical approximation (cf.[Arnold. . . 96])
• pseudo-differential operator
kΘ[V ]wk
L2x,v= k [V (x + η/2)−V (x − η/2)]F
v→ηwk
L2x,η< Θ[V ]w, w >
L2v= 0 (skew-simmetry) w
t+ v · ∇
xw − Θ[V ]w = 0
=⇒ kw(t)k
L2x,v= const.
4- Open QS: irreversible dynamics
Motivation for quantum kinetic approach Two examples:
• electron ensemble with d d.o.f. in RTD
[Frensley90]: Time-dependent boundary conditions (b.c.) for w model ideal reservoir
⇒ Boundary-value problem for Wigner-Poisson:
• inflow time-dependent b.c. for w
• mixed Dirichlet-Neumann b.c. for V
Remark: Reformulation with {ψ
j}
jimpossible:
λ NOT const.
4- Open QS: irreversible dynamics
Motivation for quantum kinetic approach Two examples:
• electron ensemble with d d.o.f. in RTD
• electron ensemble with d d.o.f. in GaAs:
electron-phonon interaction
Semiclassical cp. : Boltzmann for semiconductors
Density-matrix form. : NOT suitable for boundary-
value problems
Open QS: irreversible dynamics
w
t+ v · ∇
xw − Θ[V ]w = Qw
•
Qw =
1τ(w − M
ρ) ⇒ [Degond,. . .03] QDD,QET
Open QS: irreversible dynamics
w
t+ v · ∇
xw − Θ[V ]w = Qw
•
Qw =
1τ(w − M
ρ) ⇒ [Degond,. . .03] QDD,QET
•
Boltzmann-like collision operator
[Demeio, . . .04] ⇒ simulation
Open QS: irreversible dynamics
w
t+ v · ∇
xw − Θ[V ]w = Qw
•
Qw =
1τ(w − M
ρ) ⇒ [Degond,. . .03] QDD,QET
•
Boltzmann-like collision operator [Demeio, . . .04] ⇒ simulation
•
scattering term for electron-phonon interaction
[Fromlet, . . .99]
Open QS: irreversible dynamics
w
t+ v · ∇
xw − Θ[V ]w = Qw
•
Qw =
1τ(w − M
ρ) ⇒ [Degond,. . .03] QDD,QET
•
Boltzmann-like collision operator [Demeio, . . .04] ⇒ simulation
•
scattering term for electron-phonon interaction [Fromlet, . . .99]
•
quantum Fokker-Planck term [Castella, . . .00]
⇒ [Jüngel,. . .05] QHD
Classical cp. :Vlasov-Fokker-Planck
Quantum Fokker-Planck term
Qw = βdiv
v(vw)+σ∆
vw+2γdiv
v(∇
xw) + α∆
xw (QFP) with α, β, γ ≥ 0, σ > 0.
[Caldeira . . .83]: QS= {electrons+reservoir}
β ∼ ξ coupling, σ ∼ T temp., γ, α ∼
T1, ǫ ∼
TξLindblad condition: α σ ≥ γ
2+ β
2/16
Model Term Accuracy Lindblad
Classical β div
v(vw) + σ∆
vw O(ǫ
2) NOT ok
Frictionless σ∆
vw O(ǫ) ok
Quantum (QFP) O(ǫ
3) ok
5- Three well-posedness results
•
WP system with inflow, time-dependent b.c. :
•
d = 1, global-in-time well-p. [M,Barletti04]
•
d = 3, local-in-time well-p. [M05]
•
WPFP system, d = 3 [Arnold,Dhamo,M05]:
•
global-in-time well-p., NEW a-priori estimates
⇒ NEW strategy for well-p. of WP(FP)
Common feature: L
2-setting, ONLY kinetic tools
WP system with inflow b.c. d = 1, 3
( {∂
t+ v · ∇
x− Θ[V (t) + V
e(t)]} w(t) = 0, t ≥ 0,
−∆
xV (x, t) = n[w](x, t) = R
w(x, v, t) dv
(x, v) ∈ Ω × R
d, Ω ⊂ R
dopen, bounded, convex, V
e≡ V
e(x, t), (x, t) ∈ R
d× R
+
w(s, v, t) = γ(s, v, t), (s, v) ∈ Φ
in, t ≥ 0, V (x, t) = 0, x ∈ ∂Ω, t ≥ 0,
w(x, v, 0) = w
0(x, v) Φ
in:=
(s, v) ∈ ∂Ω × R
d| v· n(s) > 0
• Weighted space:
X
k:= L
2(Ω × R
d; (1 + |v|
2k)dxdv), d = 1, 3 ⇒ k = 1, 2 cf. [Markowich,. . . 89,Arnold,. . . 96]
Prop. : d = 3, w ∈ X
2⇒ kn[w]k
L2≤ Ckwk
X2• Weighted space:
X
k:= L
2(Ω × R
d; (1 + |v|
2k)dxdv), d = 1, 3 ⇒ k = 1, 2 cf. [Markowich,. . . 89,Arnold,. . . 96]
Prop. : d = 3, w ∈ X
2⇒ kn[w]k
L2≤ Ckwk
X2=⇒ V solution of Poisson pb. with n[w] ∈ L
2(Ω) : V ∈ W
01,2(Ω) ∩ W
2,2(Ω), kV k
W2,2≤ Ckn[w]k
L2Define P : X
2−→ W
2,2(R
3)
P : w ∈ X
27−→ V 7−→ P w ≡ V a.e. Ω, P w = 0 outside Σ
WP system with inflow b.c. in X 2
w
t+ v · ∇
xw − Θ[P w]w = 0, t ≥ 0,
w(s, v, t) = γ(s, v, t), (s, v) ∈ Φ
in, t ≥ 0, w(x, v, 0) = w
0(x, v) ∈ X
2• for all u ∈ X
2, kΘ[U ]uk
X2: contains kv
i2Θ[U ]uk
L2F
η→vv
i2Θ[U ]u
= ∂
η2i{ [U (x + η/2)−U (x − η/2)]F
v→ηu }
contains ∂
ηi, ∂
x2iU and v
i, v
i2u
WP system with inflow b.c. in X 2
w
t+ v · ∇
xw − Θ[P w]w = 0, t ≥ 0,
w(s, v, t) = γ(s, v, t), (s, v) ∈ Φ
in, t ≥ 0, w(x, v, 0) = w
0(x, v) ∈ X
2• for all u ∈ X
2, kΘ[U ]uk
X2: contains kv
i2Θ[U ]uk
L2F
η→vv
i2Θ[U ]u
= ∂
η2i{ [U (x + η/2)−U (x − η/2)]F
v→ηu } Prop. : kΘ[U ]uk
X2≤ CkU k
W2,2(R3)kuk
X2⇒ kΘ[P w]wk
X2≤ CkP wk
W2,2(R3)kwk
X2≤ Ckwk
2XWP system with inflow b.c. in X 2
( {∂
t− T
γ(t)− Θ[P w](t)} w(t) = 0, t ≥ 0, w(x, v, 0) = w
0(x, v) ∈ X
2w(t) ∈ D(T
γ(t)) := {u ∈ X
2| v · ∇
xu ∈ X
2, u|
Φin= γ(t)}
Remark ∀u
1, u
2∈ D(T
γ(t)) ⇒ u
1− u
2∈ D(T
0)
∀ p : [0, +∞) → X
2, s.t. p(t) ∈ D(T
γ(t)), ∀t ≥ 0
⇒ D(T
γ(t)) = p(t) + D(T
0)
(cf.[Barletti00])
WP system with inflow b.c. in X 2
( {∂
t− T
γ(t)− Θ[P w](t)} w(t) = 0, t ≥ 0, w(x, v, 0) = w
0(x, v) ∈ X
2∀ w : [0, +∞) → X
2, s.t. w(t) ∈ D(T
γ(t)), ∀t ≥ 0 w(t) = p(t) + u(t), with u : [0, +∞) → D(T
0)
( {∂
t− T
0− L
p(t) − Θ[P u](t)} u(t) = Q
p(t) u(x, v, 0) = w
0(x, v) − p(x, v, 0) ∈ D(T
0) L
p(t)u(t) := Θ[P p](t)u(t) + Θ[P u](t)p(t)
Q
p(t) := −{∂
t− T
γ(t)− Θ[P p](t)} p(t)
WP system with inflow b.c. in X 2 : local-in- t solution
( {∂
t− T
γ(t)− Θ[P w](t)} w(t) = 0, t ≥ 0, w(x, v, 0) = w
0(x, v) ∈ X
2Theorem
∀ γ s.t. ∃ p as in (1) , ∀ w
0∈ D(T
γ(0)), ∃t
max≤ ∞ s.t.
∃! classical solution w(t), ∀ t ∈ [0, t
max).
If t
max< ∞, then lim
tրtmaxkw(t)k
X2= ∞.
• Ex. :
γ ∈ C
1([0, ∞); L
2(∂Ω × R
d; (v· n(s))(1 + |v|
4)dsdv))
• A priori estimates are needed for global-in-t result
WPFP system, d = 3
{∂
t+ v · ∇
x− Θ[V (t)]} w(t) = Qw(t)
−∆
xV (x, t) = n[w](x, t) w(x, v, t = 0) = w
0(x, v)
w ≡ w(x, v, t), (x, v, t) ∈ R
2d× [0, ∞)
Qw = βdiv
v(vw) + σ∆
vw + 2γdiv(∇
xw ) + α∆
xw Assume α σ > γ
2⇒ Q uniformly elliptic in x and v.
•
Classical cp. : Vlasov-PFP with
•
non-linear term: ∇
xV · ∇
vw ,
•
FP term: β div
v(vw) + σ∆
vw.
Existing results
•
“density matrix” ρ ˆ : [Arnold, . . .03]
global-in-t solution, by conservation laws.
•
Quantum Kinetic: L
1-analysis [Cañizo, . . .04]
global-in-t solution, by ρ(t) ≥ 0 ˆ , cons. laws:
w(t) ∈ L
1(R
2d; R) ⇒ n[w](t) ∈ L
1(R
d) (mass cons.) ⇒ kn[w](t)k
L1= const.
e
kin[w](x, t)> −∞ ⇒ e
kin[w](x, t)< ∞
INSTEAD keep to L
2-setting, purely kinetic analysis
• Weighted space: X
2:= L
2(R
6; (1 + |v|
4)dxdv) Prop. : w ∈ X
2⇒ kn[w]k
L2≤ Ckwk
X2=⇒ V (x) := 1
4π|x| ∗ n[w](x), x ∈ R
3:
• V ∈ L /
p, ∀p, ∇V = − x
4π|x|
3∗ n[w] ∈ L
6. F
vΘ[V ]z(x, η) = i (V (x + η/2) − V (x − η/2))
| {z }
=:δV (x,η)
F
vz(x, η)
For η fixed, kδV (. , η)k
L∞≤ C |η|
12kn[w]k
L21
• Weighted space: X
2:= L
2(R
6; (1 + |v|
4)dxdv)
• V ∈ L /
p, ∀p, ∇V = − x
4π|x|
3∗ n[w] ∈ L
6.
• kΘ[V ]zk
L2≤ Ckn[w]k
L2k|η|
12F
vzk
L2Prop. : w, z, ∇
vz ∈ X
2⇒
kΘ[V ]zk
X2≤ C{ kzk
X2+ k∇
vzk
X2} kwk
X2⇒ parabolic reguralization is needed
The linear equation
w
t= Aw(t), t > 0, Aw := −v · ∇
xw + Qw w(t = 0) = w
0∈ X
2,
• {e
tA, t ≥ 0} C
0–semigroup on X
2(cf. [Arnold, . . .02])
⇒ kw(t)k
L2≤ e
23βtkw
0k
L2, t ≥ 0
• w(x, v, t) = G(x, v, t)∗ w
0(cf. [Sparber, . . .03]) ⇒
k∇
vw(t)k
X2≤ B t
−1/2e
κtkw
0k
X2, 0 < t ≤ T
0(cf. classical cp. VPFP [Carpio98])
The non-linear equation
∀ T ∈ (0, ∞) fixed
w
t= Aw(t)+(Θ[V ]w)(t), t ∈ [0, T ], w(t = 0) = w
0, Y
T:= {z ∈ C([0, T ]; X
2) | ∇
vz ∈ C((0, T ]; X
2),
t
1/2k∇
vz(t)k
X2≤ C
T∀t ∈ (0, T )}.
∀ T < T
0, the map w ∈ Y
T7−→ w s. t. e e
w(t) = e
tAw
0+
Z
t0
e
(t−s)A(Θ[ e V ]w)(s) ds, t ∈ [0, T ] is a strict contraction on some ball of Y
τwith
τ (kw
0k
X2).
The non-linear equation
∀ T ∈ (0, ∞) fixed
w
t= Aw(t)+(Θ[V ]w)(t), t ∈ [0, T ], w(t = 0) = w
0, Y
T:= {z ∈ C([0, T ]; X
2) | ∇
vz ∈ C((0, T ]; X
2),
t
1/2k∇
vz(t)k
X2≤ C
T∀t ∈ (0, T )}.
Existence and Uniqueness Theorem
∀ w
0∈ X
2, ∃ t
max≤ ∞ s.t. ∃ ! mild solution w ∈ Y
T, ∀ T < t
max. If t
max< ∞, then
lim
tրtmaxkw(t)k
X2= ∞.
Global-in- t well-posedness
A priori estimates
•
kw(t)k
2≤ e
32βtkw
0k
2, t ≥ 0,
•
for k|v|
2w(t)k
2< ∞ 0 ≤ t ≤ T,
some L
p-bounds for ∇
xV (t) are needed
⇒Idea: get bounds for ∇
xV (t) depending on kw(t)k
2Strategy: exploit dispersive effects of the free-str.
(cf. [Perthame96]) =⇒
estimate(define) ∇
xV via integral equation
⇒ definition of n[w] is by-passed
Anticipation: global-in- t well-posedness WITHOUT
A priori estimate for electric field: WP case
E (x, t) = −∇V (x, t) = −
4π1 |x|x3∗ n(x, t) Z
R3
v
w(x, v, t) dv = Z
R3
v
w
0(x − vt, v) dv + Z
R3
v
Z
t0
(Θ[V ]w)(x − vs, v, t − s) ds dv Correspondingly E into
E
0(x, t) = − 1 4π
x
|x|
3∗
xZ
R3
v
w
0(x − vt, v) dv E
1(x, t) = − 1
4π
x
|x|
3∗
xZ Z
t(Θ[V ]w)(x − vs, v, t − s) ds
A priori estimate for electric field: WP, E 1
n
1(x, t) = R R
t0
(Θ[V ]w)(x − vs, v, t − s) ds dv Classical cp. : Vlasov-Poisson [Perthame96]
n
1(x, t) = Z
R3
v
Z
t0
E · ∇
vw (x − vs, v, t − s) ds dv
= div
xZ
R3
v
Z
t0
s(E w)(x − vs, v, t − s) ds dv
• Θ[V ]w = F
η−1(δV w) = F b
η−1( W [E]· d ∇
vw ) since δV (x, η) = V (x +
η2) − V (x −
η2)
= η · R
12−12
E(x − rη)dr =: W [E](x, η)· η
A priori estimate for electric field: WP, E 1
n
1(x, t) = R R
t0
(Θ[V ]w)(x − vs, v, t − s) ds dv
• Θ[V ]w = F
η−1(δV w) = F b
η−1( W [E]· d ∇
vw ) since δV (x, η) = V (x +
η2) − V (x −
η2)
= η · R
12−12
E(x − rη)dr =: W [E](x, η)· η
n
1(x, t) = Z
R3
v
Z
t0
F
η−1( W [E]· d ∇
vw )(x − vs, v, t − s) ds
= div
xZ
R3
Z
t0
s F
η−1(W [E] w)(x − vs, v, t − s)ds b
A priori estimate for electric field: WP, E 1
E
1(x, t) = −
4π1 |x|x3∗ n
1(x, t) =
− div
x 1 4πx
|x|3
∗ R
t0
s R
F
η−1(W [E] w)(x − vs, v, t−s)dvds b Lemma :
RF
η−1(W [E] w)(x−vs, v, t−s)dv b
L2x≤ s
−3/2kE(t − s)k
L2x
kw(t − s)k
L2x,vQuantum case: just in L
2.
Classical case: L
p-version.
A priori estimate for electric field: WP, E 1
E
1(x, t) = −
4π1 |x|x3∗ n
1(x, t) =
− div
x 1 4πx
|x|3
∗ R
t0
s R
F
η−1(W [E] w)(x − vs, v, t−s)dvds b Lemma :
RF
η−1(W [E] w)(x−vs, v, t−s)dv b
L2x≤ s
−3/2(kE
0(t − s)k
L2+ kE
1(t − s)k
L2) kw(t − s)k
L2Prop. : kE
0(t)k
L2≤ C
Tt
−ωwith ω ∈ [0, 1)
⇒ kE
1(t)k
L2≤ C
T, sup
s∈[0,T ]kw(s)k
L2t
1/2−ω.
NEW : w ∈ C([0, T ]; L
2) 7→ E [w] ∈ L
1((0, T ]; L
2)
A priori estimate for electric field: WP, E 0
E
0(x, t) = − 1 4π
x
|x|
3∗
xZ
w
0(x − vt, v) dv Ex. Strichartz for free-str. [Castella-Perthame ’96],
Z
w
0(x − vt, v)dv
L6x/5
≤ t
−12kw
0k
L1x(L6v/5)
, t > 0.
Let ∀ t ∈ (0, T ]
Z
w
0(x − vt, v)dv
L6/5x
≤ C
Tt
−ω, ω ∈ [0, 1)
(HP1)
⇒ kE
0(t)k
2≤ C
Z
w
0(x − vt, v)dv
L6/5
≤ CC
Tt
−ωWPFP: global-in- t, smooth solution
w(x, v, t) = G(t, x, v)∗w
0+
Z
t0
G(s, x, v)∗(Θ[V ]w)(t−s)ds
E
1∈ L
1t((0, T ]; L
[2,6]x) by parabolic reguralization.
Classical cp. : VPFP [Castella98]
Remark: E(t) ∈ L
3is needed for bootstraping kvw(t)k
2, k|v|
2w(t)k
2≤ C, 0 ≤ t ≤ T, ∀ T.
Theorem[Arnold,Dhamo,M05]
∀ w
0∈ X
2s.t. (HP1) holds, ∃!w global mild solution, w(t) ∈ C
∞(R
6), ∀t > 0,
n(t), E (t) ∈ C
∞(R
3), ∀t > 0.
WPFP: global-in- t, smooth solution
w(x, v, t) = G(t, x, v)∗w
0+
Z
t0
G(s, x, v)∗(Θ[V ]w)(t−s)ds
E
1∈ L
1t((0, T ]; L
[2,6]x) by parabolic reguralization.
Remarks
• NOT used pseudo-conformal law (cf. classical cp.)
• w ∈ C([0, T ]; L
2x,v) 7→ E
1[w] ∈ L
1t((0, T ]; L
[2,6]x) 7→
V
1[w](x, t) := −
4π1 |x|x3∗ E
1[w](x, t)
⇒ V
1[w] ∈ L
1t((0, T ]; L
[6,∞]x) ⇒ fixed-point-map for w
WPFP: global-in- t, smooth solution, without weights
7- Final considerations
• kinetic analysis, L
2-framework:
•
physically consistent,
•
suitable for real device simulation,
•