![Page 1: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/1.jpg)
D.1 Convex Optimization
D.2 Linear Matrix Inequality(LMI)
D. Linear Matrix Inequality
D.3 Control Design and LMI Formulation
Robust and Optimal Control, Spring 2015Instructor: Prof. Masayuki Fujita (S5-303B)
![Page 2: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/2.jpg)
2
Convex Optimization Problems
minimizesubject to
(affine)( : convex)( : convex)
Note: A problem is quasiconvexif is quasiconvex and are convex.
The feasible set of a convex (or quasiconvex) optimization problem is convex.
Convex function Quasi-convex function
![Page 3: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/3.jpg)
3
Semidefinite Programming Problem (SDP)
minimizesubject to
where
(LMI)
: a set of symmetric matrix (size ): a symmetric matrix is a negative semidefinite if the following inequality holds.
Note: Multiple constraints are trivially combined into a single (larger) constraint,
iff
![Page 4: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/4.jpg)
4
Linear Matrix Inequality (LMI)
: Constant Symmetric Matrices
: Affine Functions: Symmetric Matrix,Ex.
Set of satisfying is convex, i.e.,
: Variables
General Formulation
for every satisfying
Convex Optimization Problem
Formulation
,
![Page 5: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/5.jpg)
5
LMI Numerical Optimization Problems
Convex Feasibility Problem(CFP)[Ex.]
Convex Optimization Problem(COP)[Ex.]
Quasi-convex Optimization Problem(QOP)[Ex.]
find s.t.
s.t.
s.t. , and
: Affine functions
Fact Set of satisfying LMI condition is a convex set.Suppose any satisfy . Then
![Page 6: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/6.jpg)
6
LMI Numerical Optimization Problems[Ex.] Scaled Norm Condition
For a internally stable system ,
where and , which is a structured sym. matrix set
[CFP] find such that
[COP] min subject to (*) over and
(*)
[Ex.]
[QOP] min subject to (*) over , and
Standard Solvers: SDP, Projection Algorithm(interior-point method)
![Page 7: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/7.jpg)
7
Matrix norm minimization (Maximum singular value)
minimizewhere is an LMI
The equivalent SDP
minimize
subject to : a positive semidefinite
Decision variables:
Note: The constraint equivalence follows from a Schur complement argument
[Ex.]Semidefinite Programming Problem (SDP)
![Page 8: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/8.jpg)
8
LMI Programming: CVX
http://cvxr.com/cvx/
MATLAB Software for Disciplined Convex Programming
Stephen P. Boyd
Michael C. Grant
![Page 9: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/9.jpg)
9
LMI Programming: CVXcvx_begin sdpvariable P(n,n) symmetricA’*P + P*A <= -eye(n)P >= eye(n)
cvx_end
CVX Command[Ex.] Proving the stability of a system:
cvx_statusNote: is a string returning the status of the optimization
(*)
(i)
(iii)(ii)
The following conditions are equivalent:(*) is stable, i.e.,
(iv)(v)
A candidate of Lyapunov function ,Stability Condition:
![Page 10: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/10.jpg)
10
LMI Programming: CVX[Ex.] Proving the stability of two systems,
and
The stability can be proven with a single Lyapunov function,
(*) is stable for
cvx_begin sdpvariable P(n,n) symmetricA1’*P + P*A1 <= -eye(n)A2’*P + P*A2 <= -eye(n)P >= eye(n)
cvx_end
CVX Command
(i)
(iii)
(ii)
The following conditions are equivalent:
![Page 11: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/11.jpg)
11
• Singular Matrix Type Control Problem
• Gain Scheduled Control Problem
Riccati-based Solution
LMI-based Solution
• Control Problem (See 5th doc.)
There is NO assumption about general plants
Beyond Riccati-based Control
Assumptions Full rank on the imaginary axis
Inequalities
• Multi-objective Control Problem
• Matrix Polytope Type Control Problem
Equalities
• Quadratic Stabilization Problem
• Norm Bounded Uncertainty Type Problem
![Page 12: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/12.jpg)
12
Robust Stability Condition
Algebraic Riccati Equation(ii)(i) is stable and
Riccati Inequality(iii)
Stability of -norm
LMI(iv)
Stable LTI system
Given , the following conditions are equivalent.
![Page 13: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/13.jpg)
13
LMI(v)
LMI(v)’
LMI(iv)
Robust Stability Condition (Cont’d)
Stable LTI system
Given , the following conditions are equivalent.
Schur Complement
![Page 14: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/14.jpg)
14
LMI formulation: Structured Singular Value
Upper Bound
Stability
[SP05, p. 478]
minimizesubject to
Then is an upper bound for
Quasiconvex optimization problem: Generalized eigenvalue problem
If varies monotomically, the feasible regions of are nested
![Page 15: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/15.jpg)
15
LMI formulation: via Main Loop TheoremState-space performance test
is stable
In this case
Consider (without loss of generality) finding such that
![Page 16: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/16.jpg)
16
LMI formulation: Bounded Real LemmaState-space performance test
(take )
Bounded Real Lemmais stable
s.t.
![Page 17: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/17.jpg)
17
LMI formulation: Bounded Real LemmaDiscrete-time
Continuous-time
is stable
s.t.
Discrete-time Lyapunov condition
is stable
![Page 18: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/18.jpg)
18
State feedback control
: stabilizable
and
Continuous-timeminimizesubject to
,
gives stable and
![Page 19: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/19.jpg)
19
State feedback control: CVX
P = ss(A, [Bw, Bu], [Cz; eye(n,n)], [Dzw, Dzu; zeros(n,nw+nu)]) ;cvx_begin sdpvariable Q(n,n) symmetric;variable F(nu,n);variable eta;minimize eta ;subject to:
Q > 0 ;[ Q*A’ + F’*Bu’ + A*Q + Bu*F, Bw, Q*Ce’ + F’*Dzu’ ;Bw’, -eye(nw,nw), Dzw’ ;Cz*Q + Dzu*F, Dzw, -eta*eye(nz,nz)] < 0 ;
cvx_endK = F*inv(Q) ;Aclp = A + Bu*K ;Disp( eig(Aclp) ) ; % always check that it really is a good controller.
CVX Command
minimizesubject to
![Page 20: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/20.jpg)
20
Output feedback control
: stabilizable
and
: detectable
Continuous-timeis stable
![Page 21: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/21.jpg)
21
Output feedback controlPartition as:
and
Define an inertia-preserving transform via:
![Page 22: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/22.jpg)
22
Output feedback controlminimize
subject to
gives stable and
![Page 23: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/23.jpg)
23
Output feedback control: CVX
P = ss(A, [Bw, Bu], [Cz; Cy], [Dzw, Dzu; Dyw, zeros(ny,nu)]) ;
cvx_begin sdpvariable X(n,n) symmetric;variable Y(n,n) symmetric;variable Ah(n,n);variable Bh(n,ny);variable Ch(nu,n);variable eta;
minimize eta ;subject to:
[ X, eye(n,n) ; eye(n,n), Y ] > 0 ;
[ A*X + Bu*Ch + X*A’ + Ch’*Bu’, A+Ah’, Bw, X*Ce’ + Ch’*Dzu’ ;A’+Ah, Y*A + A’*Y + Bh*Cy + Cy’*Bh’, Y*Bw + Bh*Dyw, Ce’ ; Bw’, Bw’+Y + Dyw’*Bh’, -eye(nw,nw), Dzw’ ;Cz*X + Dzu*Ch, Cz, Dzw, -eta*eye(nz,nz)] < 0 ;
cvx_end
CVX Command
Same as hinfsyn
[Khi,CLhi,ghi,hiinfo] = hinfsyn(P,ny,nu,’Method’,’lmi’);MATLAB Command
![Page 24: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/24.jpg)
24
control
Theoremis stable
s.t.
Discrete-timeContinuous-times.t. s.t.
![Page 25: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/25.jpg)
25
State feedback control
: stabilizable
and
Continuous-time is stable and iffand s.t.
![Page 26: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/26.jpg)
26
control: LQG Problem
: stabilizable
and
LQG Objective:
and
![Page 27: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/27.jpg)
27
Output feedback control
: stabilizable
and
: detectable
Continuous-time is stable and iff
s.t.
, and
![Page 28: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/28.jpg)
28
Output feedback controlminimize
subject to
gives stable and
![Page 29: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/29.jpg)
29
control: CVX
cvx_begin sdpvariable X(n,n) symmetric;variable Y(n,n) symmetric;variable W(nz,nz) symmetric;variable Ah(n,n);variable Bh(n,ny);variable Ch(nu,n);variable gamma;
minimize gamma ;subject to:
trace(W) < gamma ;[ W, Cz*X+Dzu*Ch, Cz ; X*Cz’+Ch’*Dzu’, X, eye(n,n) ;Cz’, eye(n,n), Y ] > 0 ;
[ A*X + Bu*Ch + X*A’ + Ch’*Bu’, A+Ah’, Bw ;A’+Ah, Y*A + A’*Y + Bh*Cy + Cy’*Bh’, Y*Bw + Bh*Dyw ; Bw’, Bw’+Y + Dyw’*Bh’, -eye(nw,nw) ] < 0 ;
cvx_end
CVX Command
Same as h2syn
[K2,CL2,g2,hiinfo] = hinfsyn(P,ny,nu);MATLAB Command
![Page 30: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/30.jpg)
30
Design ProblemBounding error amplitudes for bounded amplitude inputs
Use impulse response matrices and a Youla parametrization to set up the design problem:
Robust problems can also be set up and solved as (large) optimization problems
![Page 31: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/31.jpg)
31
Pole Region Constraints( -Stability)Definitions
,: a region of the complex plane
[Ex.] [Ex.] [Ex.]
Re
Im
Re
Im
Re
ImConic sector
![Page 32: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/32.jpg)
32
Pole Region Constraints: LMI conditionsDefinitions
,
Theorems.t.
[Ex.] All closed loop poles have real part less than
Re
Im
s.t.
![Page 33: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/33.jpg)
33
Multi-objective Analysis
Control Problem
Control Problem
Pole Region Constraintss.t.
s.t.
s.t.
![Page 34: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/34.jpg)
34
Multi-objective Design
Control Problem
Control Problem
Pole Region Constraintss.t.
s.t.
s.t.
For Synthesis
Conservative Design
![Page 35: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/35.jpg)
35
Schur Complement
(i)
(iii)(ii)
The following conditions are equivalent:
(i) ⇔ (iii)
(i) ⇔ (ii)
![Page 36: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/36.jpg)
36
LMI Formulation: Root mean square(RMS) Gain
and s.t.
RMS gain of the stable LTI system is the minimum value of the solution satisfying the following statement.
RMS gain is the value that the average size for the sustainable signal
Covariance matrix
Power spectrum densityIf satisfies ergodicity, i.e., a stationary stochastic signal,
![Page 37: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/37.jpg)
37
Relaxations for Structured Uncertainty
(Quadratic) StabilityStability
Fundamental Stability (NS)s.t. s.t.and
s.t.and s.t.and
![Page 38: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/38.jpg)
The matrix is Hurwitz (stable) and
38
Strictly Bounded Real Lemma
(i) Stability of -norm
LMI(ii)
Suppose .Then the following are equivalent conditions.
There exists a symmetric matrix such that
KYP Lemma(iii)and there exist symmetric
matrices and matrices such that
V.M. Popov
R.E.Kalman L. Yakubovich
A. Hurwitz
![Page 39: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/39.jpg)
39
Single Constraint Quadratic OptimizationOriginal SDP Problem
minimizesubject to
minimizesubject to
Not a convex optimization problem
A linear objective function, A linear inequality constraint and a nonlinear equality constraint
Relaxation/ Dual problem of the SDPminimizesubject to
![Page 40: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/40.jpg)
40
S-procedure (Yakubovich’s S-lemma)
Sufficient Condition is clear.
Definitions
where and
where and
if and only if
s.t.
If there exists satisfying (S), then given ,Note:
(S)
![Page 41: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/41.jpg)
where .
41
Generalized S-ProcedureWe often encounter problems with constraints of the form
satisfying(S1)(S2)
A potentially conservative but useful algebraic sufficient condition for (S1) and (S2) is the existence of positive-semidefinite functions
such that
for all (S3)
for all
The set-containment constraint
For the case in which are quadratic functions, the sufficient condition in (S3) is known as the S-procedure relaxation for (S1) and (S2).
![Page 42: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/42.jpg)
42
LMI Programming: YALMIP
http://users.isy.liu.se/johanl/yalmip/
YALMIP: Yet Another LMI Parser
Johan Löfberg
![Page 43: D. Linear Matrix Inequality D.1 Convex Optimization D.2 ... · 3 Semidefinite Programming Problem (SDP) minimize subject to where (LMI): a set of symmetric matrix (size ): a symmetric](https://reader034.vdocument.in/reader034/viewer/2022050513/5f9d5f5af4f6d35ae65e4f02/html5/thumbnails/43.jpg)
43
SDP Solvers in YALMIPLinear Programming
(free) CDD, CLP, GLPK, LPSOLVE, QSOPT, SCIP, (commercial) CPLEX, GUROBI, LINPROG, MOSEK, XPRESS
Mixed Integer Linear Programming(free) CBC, GLPK, LPSOLVE, SCIP, (commercial) CPLEX, GUROBI, MOSEK, XPRESS
Quadratic Programming(free) BPMPD, CLP, OOQP, QPC, qpOASES, quadprogBB, (commercial) CPLEX, GUROBI, MOSEK, NAG, QUADPROG, XPRESS
Mixed Integer Quadratic Programming(commercial) CPLEX, GUROBI, MOSEK, XPRESS
Second-order Cone Programming(free) ECOS, SDPT3, SeDuMi (commercial) CPLEX, GUROBI, MOSEK
Mixed Integer Second-order Cone Programming(commercial) CPLEX, GUROBI, MOSEK
Semidefinite Programming(free) CSDP, DSDP, LOGDETPPA, PENLAB, SDPA, SDPLR, SDPT3, SDPNAL, SeDuMi (commercial) LMILAB, MOSEK, PENBMI, PENSDP
General Nonlinear Programming and other solvers