P 11 Milisic
P 11 Milisic
P 11 Milisic
xx
c American Institute of Mathematical Sciences
Volume x, Number 0x, xxx 2011 pp. 1–xx
Ansgar Jüngel
Institute for Analysis and Scientific Computing,
Vienna University of Technology, Wiedner Hauptstr. 8-10, 1040 Wien, Austria
Josipa-Pina Milišić
Department of Applied Mathematics, Faculty of Electrical Engineering and Computing,
University of Zagreb, Unska 3, 10000 Zagreb, Croatia
2000 Mathematics Subject Classification. Primary: 35Q30, 76Y05; Secondary: 35Q40, 82D37.
Key words and phrases. Quantum Navier-Stokes equations, density-dependent viscosity,
Chapman-Enskog expansion, finite differences, tunneling diode, semiconductors.
The first author acknowledges partial support from the Austrian Science Fund (FWF), grants
P20214 and I395, and from the Austrian-French and Austrian-Spanish Projects of the Austrian
Exchange Service (ÖAD). Both authors have been supported by the Austrian-Croatian Project
of the ÖAD (HR 01/2010) and the Ministry of Science, Education, and Sports of the Republic of
Croatia (MZOS).
1
2 ANSGAR JÜNGEL AND JOSIPA-PINA MILIŠIĆ
[21, 22] (also see [7]). Similar model equations were derived by Gardner [15] who
used another equilibrium distribution motivated from classical gas dynamics.
Quantum diffusion models can be derived from collisional Wigner equations [20].
In some physical applications, such as semiconductor theory, momentum is trans-
ferred to the crystal lattice such that only mass and energy are conserved. Then the
mean momentum of the thermal equilibria vanishes and the hydrodynamic equa-
tions are not well adapted to the macroscopic fluid transport [9]. Instead, one
may employ a diffusive scale in the Wigner equation with suitable scattering terms,
which leads to quantum drift-diffusion and quantum energy-transport models, see
[6, 10].
All these models were obtained by performing the zero-mean-free-path limit in
the collisional Wigner equation. In the first-order approximation of the mean-free
path, it is well known in classical gas dynamics that this leads to viscous correc-
tions to the Euler equations. This idea was applied by Brull and Méhats in [4] to
derive the isothermal quantum Navier-Stokes equations. In this paper, we comple-
ment the result of [4] by computing the full quantum Navier-Stokes equations, i.e.
including the energy equation, and its local approximation. Moreover, we discuss
the energy and entropy dissipation of the resulting model and present, for the first
time, numerical results for these models.
To this end, we start from a collisional Wigner equation in the hydrodynamic
scaling,
1 α
wt + p · ∇x w + θ[V ] = (M [w] − w) + ∆p w + divp (pw) , (1)
α τ
where w(x, p, t) is the Wigner function in the phase-space variables (x, p) ∈ R3 × R3
and time t > 0, α is the scaled mean-free path (which is assumed to be small com-
pared to one), τ is a relaxation time, and θ[V ] is a nonlocal potential operator,
involving the scaled Planck constant ε (see Section 2.1 for a definition). The elec-
tric potential V may be a given function or self-consistently coupled to the particle
density through the Poisson equation. The collision operator on the right-hand
side of (1) consists of two terms: The dominant term is of relaxation type, i.e., the
Wigner function has the tendency to tend to the quantum equilibrium or quantum
Maxwellian M [w] (see Section 2.1). The second term is a Caldeira-Leggett-type
collision operator [5]. Its moments lead to momentum and energy relaxation ex-
pressions which are often used in quantum hydrodynamic equations [17].
Performing a Chapman-Enskog expansion w = M [w] + αg around the quantum
Maxwellian gives nonlocal equations for the first moments of w, the particle density
n, the momentum nu, and the energy density ne, including viscous corrections of
order α. Local expressions in the particle density n, velocity u, and temperature
T are obtained by expanding the higher-order moments in powers of ε2 . Then,
assuming further that the temperature variations and the vorticity A(u) = 21 (∇u −
∇u⊤ ) are of order ε2 , i.e. ∇ log T = O(ε2 ) and A(u) = O(ε2 ), the expansion up to
order O(α2 + αε2 + ε4 ) gives the following quantum Navier-Stokes system:
2 3
= div(Su) − ne − n , (4)
τ0 2
where u ⊗ u is a matrix with components uj uk , ∇2 is the Hessian matrix, τ0 = τ /α,
the energy density (up to terms of order O(ε4 )) is given by
3 1 ε2
ne = nT + n|u|2 − n∆ log n, (5)
2 2 24
the viscous stress tensor S and the total heat flux q0 are defined by
2 5 ε2
S = 2αnT D(u) − αnT div u I, q0 = − αnT ∇T − n(∆u + 2∇div u), (6)
3 2 24
D(u) = 21 (∇u + ∇u⊤ ) is the symmetric part of the velocity gradient, and I is the
unit matrix in R3×3 .
The mass and momentum equations (2)-(3) with constant temperature have been
derived by Brull and Méhats in [4]. The novelty of this paper is the derivation of
the energy equation and the numerical solution of the full Navier-Stokes model.
In the following, we comment the form of the pressure, total heat flux, and the
viscous corrections. The pressure p(n, T ) = nT can be interpreted as the Boyle law
for a perfect gas [14]. The total
R heat flux q0 consists of two contributions: one from
the third-order moment q = R3 12 (p−u)|p−u|2 M [w]dp/(2πε)3 (see (15)), which is of
quantum mechanical origin, and another contribution from the heat flux 52 αnT ∇T
(computed in Section 2.4), which arises from the Chapman-Enskog expansion. The
heat conductivity equals κ(n, T ) = 52 αnT , expressing the Wiedemann-Franz law.
This term is often used in numerical simulations as a stabilizing term in the quantum
hydrodynamic equations, see e.g. [15, 22].
The viscous stress tensor for Newtonian fluids with vanishing bulk viscosity coef-
ficient reads as S = 2µ(D(u) − 13 div u I) [14]. In our model, the viscosity coefficient
is equal to µ = αnT . Thus, in the semi-classical limit ε → 0, we recover the clas-
sical Navier-Stokes equations with the above viscosity coefficient which can be also
obtained from a Chapman-Enskog expansion of the classical BGK model.
The global-in-time existence of weak solutions to the barotropic quantum fluid
model (2)-(3) with T = 1 was proved, in a one-dimensional setting, in [19] for ε = α
and, in several space dimensions, in [18] for ε > α, in [12] for ε = α, and in [16]
for ε < α. Up to our knowledge, no analytical results are available for the system
(2)-(4).
The paper is organized as follows. In Section 2, we perform the Chapman-
Enskog expansion of the Wigner function and derive the local quantum Navier-
Stokes equations (2)-(4). We show in Section 3 that the total energy, consisting
of the thermal, kinetic, and quantum energy, is conserved for all time if 1/τ =
0. Furthermore, the fluid entropy is dissipated for incompressible fluids. Finally,
numerical simulations of a one-dimensional tunneling diode and stationary current-
voltage characteristics are presented in Section 4.
2. Derivation of the model. In this section, we derive the system (2)-(4) from
the Wigner equation (1).
2.1. The Wigner-BGK equation. We recall the kinetic quantum model [3] in-
troduced in the introduction:
1 α
wt + p · ∇x w + θ[V ]w = (M [w] − w) + ∆p w + divp (pw) . (7)
α τ
4 ANSGAR JÜNGEL AND JOSIPA-PINA MILIŠIĆ
Then, using the relations hM [w]i = hwi = n, hpM [w]i = nu, and h 21 |p|2 M [w]i = ne,
which follow from (8), we compute
Lemma 2.2. Up to terms of order O(α2 ), the moment equations of the Wigner
equation read as follows:
ε2
P = nT I − n∇2x log n + O(ε4 ), (19)
12
ε2
q=− n(∆x u + 2∇x divx u) + O(ε4 ),
24
3 1 ε2
ne = nT + n|u|2 − n∆x log n + O(ε4 ). (20)
2 2 24
(Notice that the formula for q in [22] contains a misprint.) Moreover, the Lagrange
multiplier v is related to the macroscopic velocity u by [22]
nu = nv + O(ε2 ). (21)
In the following subsections, we derive local expressions for the viscous terms S1
and S2 up to order O(αε2 + α2 ).
FULL QUANTUM NAVIER-STOKES EQUATIONS 7
Hence,
S12 = divx (nu ⊗ u ⊗ u) + divx (nT u)I + 2∇x (nT ) ⊗s u + 2nT D(u) + O(ε2 ), (24)
hp ⊗ pM [w]i = P + nu ⊗ u = nT I + nu ⊗ u + O(ε2 ),
Then, employing (18) and the expansions for P , ne, and q, i.e.
5 1
(ne)t = −divx nT u + nu|u|2 + nu · ∇x V + O(ε2 + α),
2 2
8 ANSGAR JÜNGEL AND JOSIPA-PINA MILIŠIĆ
we find that
2 2 1
S11 = (ne)t I − (nu)t · u I + 2(nu)t ⊗s u + divx (nu) u ⊗ u − |u|2 I + O(ε2 )
3 3 3
2 5 1 2 2 2
= − divx nu T + |u| I + nu · ∇x V I + divx (nu ⊗ u) · u I
3 2 2 3 3
2 2
+ ∇x (nT ) · u I − nu · ∇x V I − 2divx (nu ⊗ u) ⊗s u − 2∇x (nT ) ⊗s u
3 3
1 2
+ 2n∇x V ⊗s u + divx (nu) u ⊗ u − |u| I + O(ε2 + α).
3
The terms involving T simplify to
2
−divx (nT u)I − nT divx u I − 2∇x (nT ) ⊗s u.
3
The terms involving third powers of u sum up to −divx (nu ⊗ u ⊗ u). Hence,
2
S11 = −divx (nT u)I − nT divx u I − 2∇x (nT ) ⊗s u − divx (nu ⊗ u ⊗ u)
3
+ 2n∇x V ⊗s u + O(ε2 + α).
In view of (22) and (24), we conclude that
2
S1 = 2nT D(u) − nT divx u I + O(ε2 + α).
3
2.4. Calculation of S2 . Similarly as in the previous subsection, we set S2 = S21 +
S22 + S23 , where
S21 = ∂t h 21 p|p|2 M [w]i, S22 = divx h 12 p ⊗ p|p|2 M [w]i, S23 = h 12 p|p|2 θ[V ]M [w]i.
Formula (11) and the expansions (14), (19), and (20) show that
S23 = − P + nu ⊗ u + ne I ∇x V + O(ε2 )
5 1
=− nT I + n|u|2 I + nu ⊗ u ∇x V + O(ε2 ).
2 2
√
Next, we calculate the fourth moment h 12 p ⊗ p|p|2 M [w]i. Setting s = (p − v)/ T
and employing Lemma 2.3 and (21), we obtain
√ √ √
h 21 p ⊗ p|p|2 M [w]i = 12 (v + T s) ⊗ (v + T s)|v + T s|2 M [w]
√
= h 12 v ⊗ v|v|2 M [w]i + T v ⊗s hsM [w]i|v|2
+ (v ⊗ v)v · hsM [s]i + T 21 hs ⊗ sM [w]i|v|2
+ v ⊗ vh 21 |s|2 M [w]i + 2h(v ⊗s s)v · sM [w]i
+ T 3/2 v ⊗s hs|s|2 M [w]i + hs ⊗ s ⊗ sM [w]i · v
+ T 2 h 12 s ⊗ s|s|2 M [w]i
1 1 1
= nu ⊗ u|u|2 + T |u|2 hs ⊗ sM [w]i + T u ⊗ uh|s|2 M [w]i
2 2 2
1 2
+ 2T u ⊗s hs(u · s)M [w]i + T hs ⊗ s|s|2 M [w]i + O(ε2 )
2
1 1 7 5
= nu ⊗ u|u| + nT |u| I + nT (u ⊗ u) + nT 2 I + O(ε2 ).
2 2
2 2 2 2
FULL QUANTUM NAVIER-STOKES EQUATIONS 9
Therefore,
1 1 7 5
S22 = divx (nu ⊗ u|u|2 ) + ∇x (nT |u|2 ) + divx (nT u ⊗ u) + ∇x (nT 2 ) + O(ε2 ).
2 2 2 2
It remains to calculate the time derivative of h 21 p|p|2 M [w]i. Setting k = ℓ in (23)
and summing over j, we find that
5 1
h 21 p|p|2 M [w]i = nT u + nu|u|2 + O(ε2 ).
2 2
Hence,
5 1 5 1
S21 = ∂t nT u + nu|u|2 + O(ε2 ) = ∂t ne u − nu|u|2 + O(ε2 )
2 2 3 3
5 5 3 1 1 1
= (ne)t u + T + |u|2 nut − (n|u|2 )t u − |u|2 nut + O(ε2 )
3 3 2 2 3 3
5 5 1 1
= (ne)t u + T + |u|2 nut − (n|u|2 )t u + O(ε2 ).
3 2 2 3
By (25) and nut = (nu)t + divx (nu)u (which is a consequence of (16)), it follows
that
5 5 1 5 1
S21 = (ne)t u + T + |u|2 (nu)t + T + |u|2 divx (nu)u
3 2 2 2 2
1 2
− divx (nu)|u|2 u − uu · (nu)t + O(ε2 )
3 3
5 5 1 2 5 1
= (ne)t u + T + |u| (nu)t + T + |u|2 divx (nu)u
3 2 2 2 6
2
− uu · (nu)t + O(ε2 ).
3
Inserting the momentum equation (17) and the energy equation (18),
5 1
(ne)t = −divx nT u + nu|u| + nu · ∇x V + O(ε2 + α),
2
2 2
(nu)t = −divx (nT I + nu ⊗ u) + n∇x V + O(ε2 + α),
in the expression for S21 gives
25 5 5
S21 = − divx (nT u)u − divx (nu|u|2 )u + nu · ∇x V u
6 6 3
5 1 2 5 1
− T + |u| divx (nT I + nu ⊗ u) + T + |u|2 n∇x V
2 2 2 2
5 1 2
+ T + |u|2 divx (nu)u + divx (nT I + nu ⊗ u) · uu
2 6 3
2
− nuu · ∇x V + O(ε2 + α).
3
The terms involving the potential V sum up to
5 1
nuu · ∇x V + nT ∇x V + n|u|2 ∇x V,
2 2
2
which equals −S23 up to terms of order O(ε ). We simplify the terms involving T :
7 2 5 1
− divx (nT u ⊗ u) − nT udivx u + nT (∇u + ∇u⊤ )u − T ∇x (nT ) − ∇x (nT |u|2 ).
2 3 2 2
10 ANSGAR JÜNGEL AND JOSIPA-PINA MILIŠIĆ
Summing S21 , S22 , and S23 , some terms cancel, and we end up with
5 2
S2 = nT ∇x T + 2nT D(u)u − nT udivx u + O(ε2 + α).
2 3
Notice that, up to the Fourier heat term (5/2)nT ∇x T , S2 equals S1 times u.
We have proved (formally) the following result.
Theorem 2.4. Assume that A(u) = (∇u − ∇u⊤ )/2 = O(ε2 ) and ∇ log T = O(ε2 ).
Then, up to terms of order O(α2 + αε2 + ε4 ), the moment equations of the Wigner
equation (1) are given by (2)-(4), where the energy density ne, viscous stress tensor
S, and the total heat flux q0 are given by (5) and (6).
3. Energy and entropy estimates. In this section, we prove energy and entropy
estimates for the quantum Navier-Stokes system (2)-(4). Let the electric potential
V solve the Poisson equation
λ2 ∆V = n − C(x), (26)
where λ > 0 is the scaled Debye length and C(x) models fixed background ions
(doping profile).
Proposition 1. Let (n, nu, ne) be a smooth solution to (2)-(4) and let V be the
solution to (26). We assume that the variables decay R sufficiently fast to zero as
|x| → ∞ uniformly in t. Then the total mass N (t) = R3 n(x, t)dx is conserved, i.e.
dN (t)/dt = 0, and the total energy
Z
λ2 2
E(t) = ne + |∇V | dx
R3 2
is dissipated according to
Z
dE 2 3 1 2 ε2 √ 2
+ n(T − 1) + n|u| + |∇ n| dx = 0.
dt τ0 R 3 2 2 6
In particular, without relaxation (1/τ0 = 0), the energy is conserved. Furthermore,
the energy can be written as
Z
3 1 ε2 √ λ2
E(t) = nT + n|u|2 + |∇ n|2 + |∇V |2 dx ≥ 0.
R3 2 2 6 2
The total energy is the sum of the thermal, kinetic, quantum, and electric energy.
FULL QUANTUM NAVIER-STOKES EQUATIONS 11
Proof. The conservation of the mass N is clear. The dissipation of E follows from
Z
dE
= (ne)t + λ2 ∇V · ∇Vt dx
dt 3
ZR Z
2 3
= nu · ∇V − λ2 V (∆V )t dx − ne − n dx
R3 τ0 R 3 2
Z Z
2 3 1 ε2 |∇n|2
= − div(nu)V − V nt dx − n(T − 1) + n|u|2 + dx
R3 τ0 R 3 2 2 24 n
Z
2 3 1 ε2 √
=− n(T − 1) + n|u|2 + |∇ n|2 dx
τ0 R 3 2 2 6
Next, we calculate the evolution of the thermal energy 23 nT . By (5), we find that
3 1 ε2
(nT )t = (ne)t − (n|u|2 )t + (n∆ log n)t . (27)
2 2 24
We employ (4) and (25) to reformulate the first two terms on the right-hand side:
5 1
(ne)t =div(−nT u + αnT ∇T ) − div(nu|u|2 ) + div(Su) + nu · ∇V
2 2
ε2
+ div nu∆ log n + 2n(∇2 log n)u + n∆u + 2n∇div u
24
2 3
− ne − n ,
τ0 2
1 1
− (n|u| )t = −(nu)t · u − div(nu)|u|2
2
2 2
1 ε2
= div(nu ⊗ u) · u − div(nu)|u|2 + ∇(nT ) · u − div(n∇2 log n) · u
2 12
n 2
− nu · ∇V − (div S)u + |u| .
τ0
1 1
− div(nu|u|2 ) + div(nu ⊗ u) · u − div(nu)|u|2 = 0,
2 2
where the double point “:” means summation over both matrix indices. We notice
that
⊤ 2 2
S : ∇u = αnT (∇u + ∇u ) : ∇u − (div u)
3
1 ⊤ ⊤ 2 2
= αnT (∇u + ∇u ) : (∇u + ∇u ) − (div u)
2 3
2
= αnT 2|D(u)|2 − (div u)2 .
3
The remaining time derivative becomes, by (2),
Since
it follows that
ε2 ε2 ε2
(n∆ log n)t = − div nu∆ log n + 2n(∇2 log n) · u + n∆u + div(n∇2 log n) · u.
24 24 12
Inserting this identity and (28) into (27), we conclude that
3 3 5 ε2
(nT )t + div nT u − αnT ∇T − n∇div u
2 2 2 12
2
= −nT div u + αnT 2|D(u)|2 − (div u)2
3
2 3 ε 2
− n(T − 1) − n∆ log n . (29)
τ0 2 24
This formulation allows us to formulate the evolutions of the thermal energy and
entropy.
Proposition 2. Let (n, nu, ne) be a smooth solution to (2)-(4). We assume that
the variables decay sufficiently fast to zero as |x| → ∞ uniformly in t. Then
Z Z Z
3 d 2
nT dx = ∇(nT ) · udx + α nT 2|D(u)|2 − (div u)2
2 dt R3 R3 R3 3
Z 2
2 3 ε √
− n(T − 1) + |∇ n|2 dx, (30)
τ0 R 3 2 6
Z
d n 3
n log 3/2 + nT dx
dt R3 T 2
Z
5 n 2 3 n
=− α |∇T |2 + αn 2|D(u)|2 − (div u)2 + (T − 1)2 dx
R3 2 T 3 τ0 T
2 Z
ε n 1 n
− ∇T · ∇div u − (T − 1)∆ log n dx. (31)
12 R3 T 2 τ0 T
FULL QUANTUM NAVIER-STOKES EQUATIONS 13
The first integral on the right-hand side of (30) expresses the energy change due
to the work of compression. In incompressible fluids, this term vanishes. The second
integral is nonnegative since
2
2 1 X ∂uj ∂uk 4
2|D(u)|2 − (div u)2 = + + (div u)2 ≥ 0.
3 2 ∂xk ∂xj 3
j6=k
The third integral represents the dissipation due to relaxation. The first term on
the right-hand side of (31) expresses the dissipation due to temperature variations,
viscosity, and relaxation, the second term vanishes for incompressible fluids without
relaxation. In this situation, the entropy is nondecreasing.
Proof. The first identity follows immediately from (29) after integration over x ∈ R3
and integrating by parts in the terms nT div u and n∆ log n. For the proof of the
second identity, we employ (2) and (29) to obtain
Z Z
d n 3 d 5 3 3
n log 3/2 + nT dx = n log n − n log(nT ) + nT dx
dt R3 T 2 dt 3 2 2 2
Z R
3 5 3 3
= nt log n − log T − nt − (nT )t + (nT )t dx
3 2 2 2T 2
ZR h
3
= − div(nu) log n − log T
R3 2
1 3 5 ε2
+ div nT u − αnT ∇T − n∇divu + ndiv u
T 2 2 12
2
− αn 2|D(u)|2 − (div u)2
3
2 1 3 ε2 i
− 1− n(T − 1) − n∆ log n dx.
τ0 T 2 24
After integration by parts, some terms cancel and we infer that
Z
d n 3
n log 3/2 + nT dx
dt R3 T 2
Z 3 n
5 n 2 2 2 2 2
=− α |∇T | + αn 2|D(u)| − (div u) + (T − 1) dx
R3 2 T 3 τ T
Z
ε2 n 1 n
− 2
∇T · ∇div u − (T − 1)∆ log n dx.
12 R3 T τ0 T
This concludes the proof.
16 −3
2 10 cm
AlxGa As
AlxGa As
1−x
1−x
GaAs
GaAs
GaAs
−3
−3
Energy [eV]
GaAs 10 cm
GaAs 10 cm
18
18
0.4
0 25 30 35 40 45 50 75
Position [nm]
where C ∗ is the maximal doping value, V ∗ = kB T0 /q, and (u∗ )2 = kB T0 /m. The
scaled Planck constant and the scaled Debye length are defined by
~2 εs kB T0
ε2 = , λ2 = ,
mL2 kB T0 q 2 C ∗ L2
(nu)x = 0, (36)
2
ε 4 nu
(nu2 )x + (nT )x − (n(log n)xx )x − nVx = α(nT ux )x − , (37)
12 3 τ1
5 1 ε2
(nT u)x + (nu3 )x − (nu(log n)xx + nuxx )x − nuVx
2 2 8
5 4 1 3
= ακτ (nT Tx )x + ατ (nT uux )x − ne − n , (38)
2 3 τ2 2
λ2 Vxx = n − C, x ∈ (0, 1), (39)
where
τp τw τ0
τ1 = , τ2 = , τ= .
τr τr τr
These equations are solved in the interval (0, 1) with the following boundary con-
ditions taken from [15]:
(nu)x = 0, (40)
2
ε nu
(nu2 )x + (nT )x − (n(log n)xx )x − nVx = 2α(nux )x − , (41)
12 τ1
λ2 Vxx = n − C, x ∈ (0, 1). (42)
We are interested in the influence of the relaxation term −nu/τ1 on the static
current-voltage characteristics of the tunneling diode. In our tests, we have chosen
the (augmented) effective mass m = 0.126 · m0 , m0 = 10−31 kg, the barrier height
is B = 0.4 eV, and the momentum relaxation time τp = τ0 , which yields τ1 = τ0 /τr .
With the above values for T0 = 77 K and L = 75 nm, the scaled Planck constant
becomes ε ≈ 0.0162.
α=ε⋅r
4 α=ε⋅r 26
10
x 10
12
r = 0.05
r = 0.1
24
10 r = 0.5 10
r=1
Electron density n [m ]
Current Density J [Am ]
−3
−2
8 22
10
6
20
10
4
18
10 r = 0.05
2 r = 0.1
r = 0.5
16
r=1
0 10
0 0.1 0.2 0.3 0.4 0.5 0 20 40 60
Voltage U [V] Position x [nm]
Figure 2 (left) shows the dependence of the viscosity parameter α on the current-
voltage characteristics and the electron density when the relaxation term vanishes,
i.e. 1/τ1 = 0. As physically expected, the current density decreases for certain
applied voltages, which is called negative resistance effect. An important parame-
ter in applications is the peak-to-valley ratio, i.e. the ratio between the first local
maximum and the next local minimum of the current density. We observe that the
ratio increases for smaller values of the viscosity parameter α. Moreover, a small
value for α produces multiple negative resistance effects, related to the appearance
of “wiggles” in the electron density behind the right barrier (see Figure 2, right).
Next, we include the relaxation term −nu/τ1 in the momentum equation. Figure
3 shows the influence of this term on the current-voltage characteristics and the
electron density. We observe that relaxation term eliminates the multiple negative
resistance regions. The current density decreases with decreasing relaxation times
τ1 , and the electron density varies only near the right contact.
FULL QUANTUM NAVIER-STOKES EQUATIONS 17
τ =τ⋅r τ =τ⋅r
1 1
26
8000 10
−5
r = 10
7000 −4
r = 10
24
−3 10
r = 10
Current Density J [Am−2]
6000
Electron density n [m ]
−3
−2
r = 10
−1
5000 r = 10 22
10
4000
20
3000 10
2000 r = 10−5
18
10 −3
r = 10
1000 −1
r = 10
16
r=1
0 10
0 0.1 0.2 0.3 0.4 0 20 40 60
Voltage U [V] Position x [nm]
4 m = 0.092 ⋅ m m = 0.126 ⋅ m
0 0
x 10
16 18000
14 16000
14000
Current Density J [Am ]
−2
12000
10
10000
8
8000
6
6000
4
4000
2 2000
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Voltage U [V] Voltage U [V]
m = 0.092 ⋅ m m = 0.126 ⋅ m
0 0
25 25
10 10
23 23
10 10
Electron density n [m−3]
Electron density n [m ]
−3
21 21
10 10
19 19
10 10
U = 0.179 V U = 0.155 V
17
U = 0.286 V 17
U = 0.296 V
10 10
0 20 40 60 0 20 40 60
Position x [nm] Position x [nm]
m = 0.092 ⋅ m0 m = 0.126 ⋅ m0
110 110
105 105
100 100
Electron temperature T [K]
95 95
90 90
85 85
80 80
75 75
70 70
65 U = 0.179 V 65 U = 0.155 V
U = 0.286 V U = 0.296 V
60 60
0 20 40 60 0 20 40 60
Position x [nm] Position x [nm]
the electron density is relatively large in the quantum well but very small in the
right barrier. On the other hand, the temperature changes drastically. Indeed,
the particle temperature is almost constant except in the double-barrier structure.
FULL QUANTUM NAVIER-STOKES EQUATIONS 19
m = 0.092 ⋅ m m = 0.126 ⋅ m
0 0
1200 120
1000 100
Current Density J [Am−2]
600 60
400 40
200 20
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Voltage U [V] Voltage U [V]
m = 0.092 ⋅ m0 m = 0.126 ⋅ m0
25 25
10 10
23 23
10 10
Electron density n [m−3]
Electron density n [m ]
−3
21 21
10 10
19 19
10 10
U = 0.179 V U = 0.152 V
17
U = 0.289 V 17
U = 0.298 V
10 10
0 20 40 60 0 20 40 60
Position x [nm] Position x [nm]
m = 0.092 ⋅ m0 m = 0.126 ⋅ m0
77 77
76.9
Electron temperature T [K]
76
76.8
75
76.7
74
76.6
73
76.5
U = 0.179 V U = 0.152 V
U = 0.289 V U = 0.298 V
76.4 72
0 20 40 60 0 20 40 60
Position x [nm] Position x [nm]
Compared to the case without relaxation terms, the temperature variations are
small and the relaxation eliminates the temperature oscillations.
The dependence of the viscosity α on the current-voltage characteristics and the
electron temperature is presented in Figure 6. As in the isothermal model, we
20 ANSGAR JÜNGEL AND JOSIPA-PINA MILIŠIĆ
find that the peak-to-valley ratio increases for smaller values of α. However, in
contrast to the isothermal model, no multiple negative resistance regions occur and
the temperature has no sharp peaks, at least for α larger than ε (compare to Figure
2).
α=ε⋅r α=ε⋅r
900 80
r = 10
800 r = 50
r = 150 70
700 r = 450
60
600
500 50
400 40
300
30
200 r = 10
20 r = 50
100 r = 150
r = 450
0 10
0 0.1 0.2 0.3 0.4 0 20 40 60
Voltage U [V] Position x [nm]
Up to now, we have used the same value of α in front of the thermal conductivity
term as well as the viscosity terms in equations (36)-(38). We are interested in the
influence of the thermal conductivity term
5
α1 κτ (nT Tx )x
2
and the viscosity terms
4 4
α2 τ (nT ux )x , α2 τ (nT uux )x
3 3
on the current-voltage characteristics and the electron temperature. Figure 7 shows
the results obtained by taking the fixed value of the thermal conductivity coefficient
α1 = 100ε and by varying the viscosity constant α2 . Our numerial results are
similar to those for the isothermal model; see Figure 2. For α much smaller than ε,
there are multiple negative resistance regions in the current-voltage characteristics.
Moreover, we observe oscillations in the electron temperature, which dissappear for
larger values of α2 , see Figure 7 (bottom).
Finally, Figure 8 shows the influence of the thermal conductivity term on the
current-voltage characteristics and the electron temperature. We have fixed the
viscosity parameter α2 = 100ε, whereas the value of the thermal conductivity con-
stant α1 varies. It turns out that the thermal conductivity stabilizes the numerical
scheme. Indeed, if the constant α1 has been chosen too small, we observed strong
oscillations in the electron density and the electron temperature became negative.
Figure 8 (bottom) shows that the temperature becomes very small (the minimal
value is about 0.05 K only) in the quantum well when the parameter α1 is of the
order of or smaller than ε.
FULL QUANTUM NAVIER-STOKES EQUATIONS 21
4 α2 = ε ⋅ r α2 = ε ⋅ r
x 10
18 100
r = 0.05 r = 0.05
16 r = 0.075 95 r = 0.075
r = 0.1 r = 0.1
14
90
−2
12
85
10
80
8
75
6
70
4
2 65
0 60
0 0.1 0.2 0.3 0.4 0.5 0.6 0 20 40 60
Voltage U [V] Position x [nm]
α =ε⋅r
2
78
76
Electron temperature T [K]
74
72
70
r=1
r = 100
68
0 20 40 60
Position x [nm]
REFERENCES
α1 = ε ⋅ r α1 = ε ⋅ r
120 80
r = 0.5
r = 0.75 70
100 r=1
60
80 50
40
60
30
40 20
10
20 r = 0.5
0 r = 0.75
r=1
0 −10
0 0.1 0.2 0.3 0.4 0 20 40 60
Voltage U [V] Position x [nm]
α1 = ε ⋅ r
80
70
Electron temperature T [K]
60
50
40
30
20
r=1
10 r = 10
r = 100
0
0 20 40 60
Position x [nm]
[9] P. Degond, S. Gallego, F. Méhats, and C. Ringhofer, Quantum diffusion models derived from
the entropy principle, in: L. Bonilla et al. (eds.), Progress in Industrial Mathematics at ECMI
2006, pp. 106–122, Mathematics in Industry 12. Springer, Berlin, 2008.
[10] P. Degond, F. Méhats, and C. Ringhofer, Quantum energy-transport and drift-diffusion mod-
els, J. Stat. Phys., 118 (2005), 625–665.
[11] P. Degond and C. Ringhofer, Quantum moment hydrodynamics and the entropy principle, J.
Stat. Phys., 112 (2003), 587–628.
[12] J. Dong, A note on barotropic compressible quantum Navier-Stokes equations, Nonlin. Anal.,
73 (2010), 854–856.
[13] W. Dreyer, Maximisation of the entropy in non-equilibrium, J. Phys. A, 20 (1987), 6505–6517.
[14] E. Feireisl, Dynamics of Viscous Compressible Fluids, Oxford University Press, Oxford, 2004.
[15] C. Gardner, The quantum hydrodynamic model for semiconductor devices, SIAM J. Appl.
Math., 54 (1994), 409–427.
[16] F Jiang, A remark on weak solutions to the barotropic compressible quantum Navier-Stokes
equations, to appear in Nonlin. Anal., 2011. doi:10.1016/j.nonrwa.2010.11.005.
[17] A. Jüngel, Transport Equations for Semiconductors, Lecture Notes in Physics 773, Springer,
Berlin, 2009.
[18] A. Jüngel, Global weak solutions to compressible Navier-Stokes equations for quantum fluids,
SIAM J. Math. Anal., 42 (2010), 1025–1045.
FULL QUANTUM NAVIER-STOKES EQUATIONS 23
[19] A. Jüngel, Effective velocity in compressible Navier-Stokes equations with third-order deriva-
tives, to appear in Nonlin. Anal., 2011. doi:10.1016/j.na.2011.01.002.
[20] A. Jüngel, Dissipative quantum fluid models, to appear in Revista Mat. Univ. Parma, 2011.
[21] A. Jüngel and D. Matthes, Derivation of the isothermal quantum hydrodynamic equations
using entropy minimization, Z. Angew. Math. Mech., 85 (2005), 806–814.
[22] A. Jüngel, D. Matthes, and J.-P. Milišić, Derivation of new quantum hydrodynamic equations
using entropy minimization, SIAM J. Appl. Math., 67 (2006), 46–68.
[23] C. Levermore, Moment closure hierarchies for kinetic theories, J. Stat. Phys., 83 (1996),
1021–1065.
[24] T. Li, Z. Yu, Y. Wang, L. Huang, and C. Xiang, Numerical simulation of negative differential
resistance characteristics in Si/Si1−x Gex RTD at room temperature, Proceedings of the 2005
IEEE Conference on Electron Devices and Solid-State Circuits, pp. 409–412, ISBN 0-7803-
9339-2, 2005.
[25] F. Méhats and O. Pinaud, An inverse problem in quantum statistical physics, J. Stat. Phys.,
140 (2010), 565-602.
E-mail address: juengel@tuwien.ac.at
E-mail address: pina.milisic@fer.hr