Sensitivity analysis for a class ofsemi-coercive variational inequalities
using recession tools.
Samir ADLYUniversite de Limoges
France
Workshop Internationalsur les mathematiques et l’environnement
Essaouira, 23-24 novembre 2012.
References
The talk is based on the following papers :
I K. ADDI, S. ADLY, D. GOELEVEN, H. SAOUD,A sensitivity analysis of a class of semi-coercivevariational inequalities using recession tools, JOGO 2007.
I K. ADDI, S. ADLY, B. BROGLIATO, D. GOELEVEN,A method using the approach of Moreau andPanagiotopoulos for the mathematical formulation ofnon-regular circuits in electronics, Nonlinear Analysis,Hybrid Systems 2007.
I S. ADLY, E. ERNST, M. THERA, Stability of Non-coerciveVariational Inequalities, Communications in ContemporaryMathematics Vol. 4, 1, 145-160 (2002).
Outline
1. Non-coercive Variational inequalities : a state of art2. Stability of semi-coercive variational inequalities3. Ideal diode model : a complementarity formulation4. Set-Valued Ampere-Volt characteristics in
electronics : the convex case5. Some applications in electronics and mechanics
Outline
Non-coercive Variational inequalities : a state of art
Stability of semi-coercive variational inequalities
Ideal diode Model
Set-Valued Ampere-Volt characteristic in electronics
Some applications in electronics and mechanics
Position of the problem
Let us consider the problem :
V .I.(A, f ,Φ,K )
Find u ∈ K such that
〈Au − f , v − u〉+ Φ(v)− Φ(u) ≥ 0,
∀v ∈ KI X is a reflexive Banach space ,I K ⊂ X is a closed convex subset (nonemty),I A : X → X ∗ is an operator,I f ∈ X ∗,I Φ ∈ Γ0(X ) is a convex l.s.c. and proper function.
The coercive case
Several existence results for V .I.(A, f ,Φ,K ) are knownwhen the operator A
(i) is linear and coercive , i.e. ∃α > 0 such that :
〈Au,u〉 ≥ α‖u‖2, ∀u ∈ X,
(ii) is non-linear and coercive i.e. :
lim‖u‖→+∞
〈Au,u〉‖u‖ = +∞.
See the contributions of Brezis, Browder, J.L. Lions,Mosco, Stampacchia, Fichera etc . . .
The non-coercive case
For the Non-coercive case, we refer to the works of :I Fichera (1964)I J.L. Lions & G. Stampacchia (1967)I C. Baiocchi, F. Gastaldi, F. Tomarelli (1985)I C. Baiocchi, G. Buttazzo, F. Gastaldi, F. Tomarelli
(1988)I F. Tomarelli (1993)I D. Goeleven (1994)I S. Adly, D. Goeleven, M. Thera (1996) etc ...I A. Auslender (1996).
MotivationLet us consider the following classical Neumann problem :
N (f )
find u ∈ H1(Ω) such that−∆u = f , in Ω∂u∂n
= 0, on ∂Ω
It is well-known that N (f ) has a solution if and only if∫Ω
f (x)dx = 0.
If we replace f by fε = f + ε (with ε > 0), then N (fε) hasno solutions.
The problem N (f ) is unstable.
Obstacle problem without friction.Let Ω be an open bounded subset of R2 (representing athin elastic membrane).For f ∈ L2(Ω) and Ψ a given obstacle, we consider thefollowing problem :
(P)
−∆u ≥ f , in Ω(−∆u − f )(u −Ψ) = 0, in Ωu = 0, on ∂Ω
fdx
x1
x2
xu(x)
The variational formulation of the problem (P) is :
(P)
Find u ∈ K such that〈Au − f , v − u〉 ≥ 0, ∀v ∈ K
I X = H10 (Ω)
I K = v ∈ X | v ≥ ΨI 〈Au, v〉 =
∫Ω
∇u · ∇v dx
I 〈f , v〉 =
∫Ω
f vdx
If the condition u = 0 on ∂Ω is replaced by∂u∂n
= 0 on ∂Ω
then the operator A : X := H1(Ω)→ X ′ is no morecoercive.
If the problem (P) has a solution, then∫
Ω
fdx ≤ 0.
If∫
Ω
fdx < 0, then the problem has a solution.
Equivalently :
Necessary Condition : f ∈[
ker A ∩ K∞]
.
Suffisante Condition : f ∈ Int(
[ker A ∩ K∞])
.
The problem is unstable on the boundary.
(ker(A) ∩ K∞
)
Stable
instableinstable
The puffed-up membrane
Membrane
Obstacle
The puffed-up membrane
Find u ∈ K = v ∈W 1,2(Ω) : v ≥ Ψ on Ω∫
Ω
∇u · ∇(v − u)dx +
∫∂Ω
g(|v | − |u|)dσ ≥∫
Ω
f (v − u)dx ,
∀v ∈ K .
Necessary condition :∫∂Ω
gdσ ≥∫
Ω
fdx .
Suffisante condition :∫∂Ω
gdσ >∫
Ω
fdx .
Outline
Non-coercive Variational inequalities : a state of art
Stability of semi-coercive variational inequalities
Ideal diode Model
Set-Valued Ampere-Volt characteristic in electronics
Some applications in electronics and mechanics
We consider the following finite dimensional variationalinequality
VI(A, f , ϕ,K )
Find u ∈ K such that〈Mu − q, v − u〉+ Φ(v)− Φ(u) ≥ 0, ∀v ∈ K
I M ∈ Rn×n is a symmetric and positive semidefinitematrix ;
I Φ : Rn → R ∪ +∞ is a proper, convex, lowersemicontinuous and bounded from below ;
I K ⊂ Rn is a closed and convex set such that0 ∈ Dom Φ ∩ K .
I The recession function :
Φ∞(x) = limt→+∞
Φ(x0 + tx)− Φ(x0)
t, x0 ∈ dom(Φ).
(epi (Φ)
)∞
= epi (Φ∞).
I The polar cone :
C = f ∈ X ∗ : 〈f , x〉 ≤ 0, ∀x ∈ C.
C
C
I The support function :
σC(f ) = supx∈C〈f , x〉.
σC (f
2 )
.
f2
f1
C
σC (f1 )
I The Barrier cone : B(C) = dom σC.
I The Fenchel-Legendre conjugate :
Φ∗(f ) = supx∈X〈f , x〉 − Φ(x).
Φ(x)
−Φ∗(f)
f
〈f,x〉
=0
Let us finally recall the following proposition
PropositionLet Ψ ∈ Γ0(Rn) and p ∈ Rn be given. We have :
(i) p ∈ Dom Ψ∗ ⇐⇒ Ψ∞(w) ≥ 〈p,w〉, ∀w ∈ Rn ;
(ii) p ∈ Int (Dom Ψ∗)⇐⇒ Ψ∞(w) > 〈p,w〉, ∀w ∈ Rn\0.
The solutions set of VI(A, f , ϕ,K ) will be denoted bySol(M,q,Φ,K ).The following resolvent set will also play an important role
R(A,Φ,K ) = q ∈ Rn : Sol(M,q,Φ,K ) 6= ∅.
Let us introduce the following functionΨ : Rn → R ∪ +∞ defined by
Ψ(u) =12‖Qu‖2 + Φ(u) + IK (u), (1)
where Q = I − Pker(M) and Pker(M) denotes the orthogonalprojector from Rn to ker(M).
LetΨ(u) =
12‖u − Pker(M)(u)‖2 + Φ(u) + IK (u),
We have the following lemma :
LemmaSuppose that the assumptions (H) hold. We have
I Dom Ψ∗ = R(M) + Dom(
Φ + IK)∗.
I Ψ∞(w) = Iker(M)(w) + Φ∞(w) + IK∞(w),
PropositionA necessary condition for the existence of a solution ofVI(A, f , ϕ,K ) is that
Φ∞(w) ≥ 〈q,w〉, ∀w ∈ ker(M) ∩ K∞. (2)
PropositionSuppose that assumptions (H) hold. We have
IntR(M,Φ,K ) =
q ∈ Rn : Φ∞(w) > 〈q,w〉,∀w ∈ ker(M) ∩ K∞ \ 0
DefinitionThe variational inequality VI(A, f , ϕ,K ) is stable if thereexists ε > 0 such that for any Mε ∈ S+
n (R), any vectorqε ∈ q + εBn, any Φε ∈ Γ0(Rn) bounded from below, andany nonempty closed convex set Kε satisfying thefollowing conditions
0 ∈ Dom Φε ∩ Kε
ker(M) ∩ ker(Φ∞) ∩ K∞ = ker(Mε) ∩ ker((Φε)∞) ∩ (Kε)∞
the perturbed problem VI(Mε,qε,Φε,Kε) has at least onesolution.
TheoremAssume that assumptions (H) are satisfied. Then thevariational inequality VI(A, f , ϕ,K ) is stable in the senseof Definition 2 if and only if
Φ∞(w) > 〈q,w〉, ∀w ∈ ker(M) ∩ K∞, w 6= 0.
Ideal diode Model
THE APPROACH OF MOREAU AND PANAGIOTOPOULOS:
USE IT IN ELECTRONICS
K. ADDI and D. GOELEVEN
IREMIA, University of La Reunion
97400 Saint-Denis, France
[email protected], [email protected]
S. ADLY
DMI-XLIM, University of Limoges
87060 Limoges, France
B. BROGLIATO
INRIA Rhones-Alpes
38334 Saint-Ismier, France
Ideal Diode
i (mA)
V (Volts)
+ !
i
V
• V < 0, i = 0 =⇒ diode is blocking ;• i > 0,V = 0 =⇒ diode is conducting ;
Complementarity formulation
V ≤ 0, i ≥ 0, Vi = 0 ⇐⇒ min−V , i = 0
V ∈ ∂ΨR+(i) ⇐⇒ i ∈ ∂Ψ∗R+(V ) = ∂ΨR−(V )
ΨR+(x) :=
0 if x ≥ 0+∞ if x < 0 ∂ΨR+(x) :=
R− if x = 00 if x > 0∅ if x < 0
Complementarity formulation
u =
Ri︷︸︸︷UR +
∈ ∂ΨR+ (i)︷︸︸︷V +E ⇐⇒ E + Ri − u ∈ −∂ΨR+(i)
Complementarity formulation
ER
+ i − uR∈ −∂ΨR+(i)⇐⇒ −E
R+
uR∈ i + ∂ΨR+(i)
i = (id + ∂ΨR+)−1(u − E
R) =
1R
max0,u − E
u < E =⇒ diode is blocking
u ≥ E =⇒ diode is conducting
Numerical simulation : Ideal diode
INPUT SIGNAL t 7→ u(t)
D. Goeleven - Variational inequalities in electronics - 16
Numerical simulation: Ideal diode
INPUT SIGNAL t !" u(t)
0 1 2 3 4 5 6 7 8 9 10!1.5
!1
!0.5
0
0.5
1
1.5
time
input vo
ltage
OUTPUT SIGNAL t !" Vo(t) := V (t) + E = u(t) # Ri(t)
= u(t) # R max0,u(t) # E
R = u(t) + min0, E # u(t)
= minu(t), E
0 1 2 3 4 5 6 7 8 9 10!1.5
!1
!0.5
0
0.5
1
1.5
time
outp
ut vo
ltage
Numerical simulation : Ideal diode
OUTPUT SIGNAL t 7→ Vo(t) := V (t) + E = u(t)− Ri(t)
= u(t)− R max0, u(t)− ER
= u(t) + min0,E − u(t)
= minu(t),E
D. Goeleven - Variational inequalities in electronics - 16
Numerical simulation: Ideal diode
INPUT SIGNAL t !" u(t)
0 1 2 3 4 5 6 7 8 9 10!1.5
!1
!0.5
0
0.5
1
1.5
time
input vo
ltage
OUTPUT SIGNAL t !" Vo(t) := V (t) + E = u(t) # Ri(t)
= u(t) # R max0,u(t) # E
R = u(t) + min0, E # u(t)
= minu(t), E
0 1 2 3 4 5 6 7 8 9 10!1.5
!1
!0.5
0
0.5
1
1.5
time
outp
ut vo
ltage
Outline
Non-coercive Variational inequalities : a state of art
Stability of semi-coercive variational inequalities
Ideal diode Model
Set-Valued Ampere-Volt characteristic in electronics
Some applications in electronics and mechanics
Set-Valued Ampere-Volt characteristic inelectronics
• Diode• Zener Diode• Varactor• Triode• Tetrode• Transistor• Diac• Triac• SiliconControlledRectifier
Set-Valued Ampere-Volt characteristic : theDiode model
V
i
+ -
V1
V2
1
-100
V (Volts)
i (mA)
There is a voltage point, called the knee voltage V1, at whichthe diode begins to conduct and a maximum reverse voltage,called the peak reverse voltage V2,that will not force the diodeto conduct.
Set-Valued Ampere-Volt characteristic : thecomplete diode model
V1
V2
IR
i (mA)
-100
100
1
-0.001
V (Volts)
Illustrates a complete diode model which includes theeffects of the natural resistance of the diode
Set-Valued Ampere-Volt characteristic : theZener diode model
V
i
+ -
V1
V2
I1I2
-10
-10
1
V (Volts)
i (mA)
Zener diode is a good voltage regulator to maintain aconstant voltage regardless of minor variations in loadcurrent or input voltage.
Set-Valued Ampere-Volt characteristic inelectronics
V
V
i
+ -
V1
V2
i
VV1
V2
i
(a) (b)
I1
I2
Illustrates a typical voltage current characteristics of adiac.
Set-Valued Ampere-Volt characteristic inelectronics
V
I
IG
i
V
i
(a) (b)
V
i
(c)
V
IG3
IG2
IG1
<IG3<IG2=IG10
+
_
Illustrates the AV-characteristic of a three-terminal siliconcontrolled rectifier which is used for start/stop controlcircuit for a direct current motor, lamp...
Outline
Non-coercive Variational inequalities : a state of art
Stability of semi-coercive variational inequalities
Ideal diode Model
Set-Valued Ampere-Volt characteristic in electronics
Some applications in electronics and mechanics
Clipping circuit
u(t)
R
V+
-
+-
E
i
V
i
+ -
V1
V2
1
-100
V (Volts)
i (mA)
The electrical superpotential of the practical diode is
ϕPD(x) =
V1x if x ≥ 0
V2x if x < 0, (x ∈ R).
Thenϕ∗PD(z) = I[V2,V1](z), (z ∈ R)
We see that
∂ϕPD(x) =
V2 if x < 0
[V2,V1] if x = 0
V1 if x > 0
, (x ∈ R)
∂ϕ∗PD(z) =
R− if z = V2
0 if z ∈]V2,V1[
R+ if z = V1
∅ if z ∈ R\ [V2,V1]
, (z ∈ R).
recovers the volt-ampere characteristic (V , i). Theampere-volt characteristic of the practical diode can thusbe written as
V ∈ ∂ϕPD(i)⇐⇒ i ∈ ∂ϕ∗PD(V )⇐⇒ ϕPD(i) + ϕ∗PD(V ) = iV .
Using Kirchhoff’s law the problem is equivalent toVI(R,E− u, ϕPD,R), i.e.
i ∈ K := R : (Ri+E−u)(v−i)+ϕPD(v)−ϕPD(i) ≥ 0,∀v ∈ R.
Here R > 0 and for each E ,u ∈ R.
Moreover :
i(t) = (idR + ∂ϕPD)−1(u(t)− E
R)
= argminx∈R12|x − (
u(t)− ER
)|2 + ϕPD(x).
andVo(t) = u(t)− Ri(t). (3)
A static model
E1 < E2
D2D1
u(t)
R
V1 V2
i2
i1
i
KIRCHOFF’S LAWS
E1 + R(i1 + i2)− u = +V1 ∈ −∂ΨR−(i1)
E2 + R(i1 + i2)− u = −V2 ∈ −∂ΨR+(i2)
Non-coercive complementarity formalism
A︷ ︸︸ ︷(R RR R
) I︷ ︸︸ ︷(i1i2
)+
U︷ ︸︸ ︷(E1 − uE2 − u
)∈ −∂ΨR−×R+(I).
Here the matrix A is positive semidefinite and symmetric.A sufficient condition for the existence of at least onesolution is :
〈q, v〉 > 0, ∀v ∈ kerA ∩ (R− × R+) \ 0.
Non-coercive complementarity formalism
〈q, v〉 > 0, ∀v ∈ kerA ∩ (R− × R+) \ 0.We have
kerA∩(R−×R+) = v = (v1, v2) ∈ R2 : v1 ≤ 0, v2 = −v1.
Then, for all v ∈ kerA ∩ R− × R+, v 6= 0, we get
qT v = (E1 − u)v1 + (E2 − u)v2 = v2(E2 − E1) > 0.
Then I is determined as the unique solution of thecomplementarity system :
I ∈ K , AI + U ∈ K ∗, IT (AI + U) = 0
m
VARIATIONAL INCLUSION SYSTEM
AI + U ∈ −∂ΨR+×R+(I)
Moreover
i1 + i2 =
u − E1
Rif u < E1
0 if E1 ≤ u ≤ E1u − E2
Rif u > E2
.
0 1 2 3 4 5 6 7 8 9 10!1.5
!1
!0.5
0
0.5
1
1.5
time
inpu
t vol
tage
D. Goeleven - Variational inequalities in electronics - 4
INPUT t !" u(t)
0 1 2 3 4 5 6 7 8 9 10!1.5
!1
!0.5
0
0.5
1
1.5
time
input voltage
OUTPUT t !" Vo(t) = Ri(t)
0 1 2 3 4 5 6 7 8 9 10!1.5
!1
!0.5
0
0.5
1
1.5
time
outp
ut voltage
Non-regular electronic circuits
• A ∈ Rn×n, B ∈ Rn×m, C ∈ Rm×n, D ∈ Rn×p,• j : R×Rm → R, x 7→ j(t , x) locally Lipschitz for all t ≥ 0,• u ∈ L1
loc(0,+∞;Rp), x0 ∈ Rn.Problem P(x0) : Find x : [0,+∞[→ Rn; andyL : [0,+∞[→ Rm such that :
x ∈ C0([0,+∞[;Rn), ByL ∈ L1loc(0,+∞;Rn),
dxdt∈ L1
loc(0,+∞;Rn), x(0) = x0,
dxdt
(t) = Ax(t)− ByL(t) + Du(t), a.e. t ≥ 0,
y(t) = Cx(t), ∀ t ≥ 0, and yL(t) ∈ ∂2j(t , y(t)), a.e. t ≥ 0.
Non-regular circuit with VAP-admissibledevice
(H) There exists a symmetric and invertible matrixR ∈ Rn×n such that
R−2CT = B.
Problem Q(x0) : Find z : [0,+∞[→ Rn; t 7→ z(t) suchthat :
z ∈ C0([0,+∞[;Rn),dzdt∈ L1
loc(0,+∞;Rn), z(0) = Rx0,
dzdt
(t) ∈ RAR−1z(t) + RDu(t)− R−1CT∂2j(t ,CR−1z(t)).
Non-regular circuit with VAP-admissibledevice
Suppose that assumption (H) is satisfied.
If (x , yL) is solution of Problem P(x0) then z = Rx issolution of Problem Q(x0).
Reciprocally, if z is solution of Problem Q(x0) then thereexists a function yL such that (R−1z, yL) is solution ofProblem P(x0).
Non-regular circuit with VAP-admissibledevice
SetL = CR−1.
andJ(t ,X ) = j(t ,LX ), (X ∈ Rn, t ≥ 0)
then
∂2J(t ,X ) ⊂ R−1CT∂2j(t ,CR−1X ), (X ∈ Rn, t ≥ 0)
⇓
DIFFERENTIAL INCLUSION
dzdt
(t) ∈ RAR−1z(t) + RDu(t)− ∂2J(t , z(t)), a.e. t ≥ 0
Non-regular circuit with VAP-admissibledevice
DIFFERENTIAL INCLUSION
dzdt
(t) ∈ RAR−1z(t) + RDu(t)− ∂2J(t , z(t)), a.e. t ≥ 0
⇓
VARIATIONAL INEQUALITY
〈dzdt
(t)− RAR−1z(t)− RDu(t), v − z(t)〉+
+J(t , v)− J(t , z(t)) ≥ 0,∀v ∈ Rn, a.e. t ≥ 0.
Example
IG
u(t)
R
L
C0
Example
dx1
dtdx2
dt
=
A︷ ︸︸ ︷ 0 1
− 1LC0
−RL
( x1
x2
)−
B︷ ︸︸ ︷(0
−1L
)yL+
D︷ ︸︸ ︷(01L
)u,
y =
C︷ ︸︸ ︷(0 −1
)( x1
x2
)and
yL ∈ ∂C,2jSCR(., y).
(H) There exists a symmetric and invertible matrixR ∈ Rn×n such that
R−2CT = B.
R =
1√C0
0
0√
L
Hence assumption (H) is satisfied.
Example 2
u
L2
L3
R1
R2
R3
C4
dx1
dtdx2
dtdx3
dt
=
A︷ ︸︸ ︷0 1 0
− 1L3C4
−(R1 + R3)
L3
R1
L3
0R1
L2−(R1 + R2)
L2
x1
x2
x3
−
B︷ ︸︸ ︷0 0
1L3
1L3
− 1L2
0
(
yL,1
yL,2
)+
D︷ ︸︸ ︷00
1L2
u,
yL,1 ∈ ∂jD(−x3 + x2)yL,2 ∈ ∂jZ (x2)
(4)
Setting
y =
C︷ ︸︸ ︷(0 1 −10 1 0
) x1
x2
x3
jZD(X ) = jZ (X1) + jD(X2), (X ∈ R2)
yL,1 ∈ ∂jD(−x3 + x2)yL,2 ∈ ∂jZ (x2)
⇐⇒ yL ∈ ∂jZD(Cx).
A simple computation shows that the matrix
R =
1√C4
0 0
0√
L3 0
0 0√
L2
is convenient.Hence assumption (H) is satisfied.