一个科研设想:对于离散的动态规划问题,可以用鲁棒优化类似处理;对于连续的动态规划问题,是否有响应的鲁棒优化理论?
Recall that Linear Optimization (LO) problem is of the form
(1)
min
x
{
c
T
x
+
d
:
A
x
≤
b
}
\min_x\{c^Tx+d:Ax\leq b\}\tag{1}
xmin{cTx+d:Ax≤b}(1) where
x
∈
R
n
x\in\mathbf{R}^n
x∈Rn is the vector of decision variables,
c
∈
R
n
c\in\mathbf{R}^n
c∈Rn and
d
∈
R
d\in\mathbf{R}
d∈R form the objective,
A
A
A is an
m
×
n
m\times n
m×n constraint matrix, and
b
∈
R
m
b\in\mathbf{R}^m
b∈Rm is the right hand side vector. The stucture of the problem can be given by the following data matrix
D
=
[
C
T
d
A
b
]
.
D=\begin{bmatrix} C^T & d \\ A &b\end{bmatrix}.
D=[CTAdb].
Definition 1. An uncertain Linear Optimization problem ( LO U ) (\text{LO}_\mathcal{U}) (LOU) is a collection (2) { min x { c T + d : A x ≤ b } } ( c , d , A , b ) ∈ U \{\min_x\{c^T+d:Ax\leq b\}\}_{(c,d,A,b)\in\mathcal{U}}\tag{2} {xmin{cT+d:Ax≤b}}(c,d,A,b)∈U(2) of LO problems (instances) min x { c T x + d : A x ≤ b } \min_x\{c^Tx+d:Ax\leq b\} minx{cTx+d:Ax≤b} of common structure (i.e., with common numbers m m m of constraints and n n n of variables) with the data varying in a given uncertainty set    U ⊂ R ( m + 1 ) × ( n + 1 ) \;\mathcal{U}\subset\mathbf{R}^{(m+1)\times(n+1)} U⊂R(m+1)×(n+1).
The uncertainty set can be parameterized, in an affine fashion, by pertubation vector
ζ
\zeta
ζ varying in a given pertubation set
Z
\mathcal{Z}
Z:
(3)
U
=
{
[
c
T
d
A
b
]
=
[
c
0
T
d
0
A
0
b
0
]
⎵
nominal data
D
0
+
∑
ℓ
=
1
L
ζ
ℓ
[
c
ℓ
T
d
ℓ
A
ℓ
b
ℓ
]
⎵
basic shifts
D
ℓ
:
ζ
∈
Z
⊂
R
L
}
\mathcal{U}=\left\{\begin{bmatrix}c^T&d\\ A &b\end{bmatrix} = \underbrace{\begin{bmatrix}c_0^T&d_0\\ A_0 &b_0\end{bmatrix}}_{\text{nominal data } D_0} + \sum_{\ell=1}^{L}\zeta_{\ell} \underbrace{\begin{bmatrix}c_\ell^T&d_\ell\\ A_\ell &b_\ell\end{bmatrix}}_{\text{basic shifts } D_\ell} : \zeta\in\mathcal{Z}\subset\mathbf{R}^L \right\}\tag{3}
U=⎩⎪⎪⎨⎪⎪⎧[cTAdb]=nominal data D0
[c0TA0d0b0]+ℓ=1∑Lζℓbasic shifts Dℓ
[cℓTAℓdℓbℓ]:ζ∈Z⊂RL⎭⎪⎪⎬⎪⎪⎫(3)
Definition 2. A vector x ∈ R n x\in\mathbf{R}^n x∈Rn is a robust feasible solution to ( LO U ) (\text{LO}_\mathcal{U}) (LOU), if it satisfies all realizations of the constraints from the uncertainty set, that is (4) A x ≤ b , ∀ ( c , d , A , b ) ∈ U Ax\leq b,\qquad \forall(c,d,A,b)\in\mathcal{U}\tag{4} Ax≤b,∀(c,d,A,b)∈U(4)
Definition 3. Given a candidate solution x x x, the robust value c ^ ( x ) \hat{c}(x) c^(x) of the objective in ( LO U ) (\text{LO}_\mathcal{U}) (LOU) at x x x is the largest value of the “true” objective c T + d c^T+d cT+d over all realizations of the data from the uncertainty set: (5) c ^ ( x ) = sup ( c , d , A , b ) ∈ U [ c T x + d ] \hat{c}(x)=\sup_{(c,d,A,b)\in\mathcal{U}}\left[c^Tx+d\right]\tag{5} c^(x)=(c,d,A,b)∈Usup[cTx+d](5)
Definition 4. The Robust Counterpart of the uncertain LO problem ( LO U ) (\text{LO}_\mathcal{U}) (LOU) is the optimization problem (6) min x { c ^ ( x ) = sup ( c , d , A , b ) ∈ U [ c T + d ] : A x ≤ b , ∀ ( c , d , A , b ) ∈ U } \min_x\left\{\hat{c}(x)=\sup_{(c,d,A,b)\in\mathcal{U}}\left[c^T+d\right]:Ax\leq b, \forall (c,d,A,b)\in\mathcal{U}\right\}\tag{6} xmin{c^(x)=(c,d,A,b)∈Usup[cT+d]:Ax≤b,∀(c,d,A,b)∈U}(6) of minimizing the robust value of the objective over all robust feasible solutions to the uncertian problem.
OBSERVATIONS.
A. The Robust Counterpart of
LO
U
\text{LO}_\mathcal{U}
LOU can be rewritten equivalently as the problem
(7)
min
x
{
t
  
:
  
c
T
−
t
≤
−
d
,
 
A
x
≤
b
,
  
∀
(
c
,
d
,
A
,
b
)
∈
U
}
\min_x\left\{t\;:\;c^T-t\leq -d,\,Ax\leq b,\;\forall (c,d,A,b)\in\mathcal{U}\right\}\tag{7}
xmin{t:cT−t≤−d,Ax≤b,∀(c,d,A,b)∈U}(7)
B. The Robust Counterpart of uncertain LO problem with certain objective
min
x
{
c
T
+
d
:
A
x
≤
b
,
∀
(
A
,
b
)
∈
U
}
\min_x\{c^T+d:Ax\leq b, \forall (A,b)\in\mathcal{U}\}
xmin{cT+d:Ax≤b,∀(A,b)∈U} is purely “constraint-wise” construction: to get RC, we act as follows:
(1) preserve the original certain objective as it is, and
(2) replace every one of the original constraints
(8)
(
A
x
)
i
≤
b
i
⇔
a
i
T
x
≤
b
i
(Ax)_i\leq b_i\Leftrightarrow a_i^Tx\leq b_i\tag{8}
(Ax)i≤bi⇔aiTx≤bi(8) (
a
i
T
a_i^T
aiT is
i
i
i-th row in
A
A
A) with its Robust Counterpart
(9)
a
i
T
x
≤
b
i
,
∀
[
a
i
,
b
i
]
∈
U
i
a_i^Tx\leq b_i, \forall[a_i,b_i]\in\mathcal{U}_i\tag{9}
aiTx≤bi,∀[ai,bi]∈Ui(9) where
U
i
\mathcal{U}_i
Ui is the projection of
U
\mathcal{U}
U on the space of data of
i
i
i-th constraint:
U
i
=
{
[
a
i
,
b
i
]
:
[
A
,
b
]
∈
U
}
.
\mathcal{U}_i=\{[a_i,b_i]:[A,b]\in\mathcal{U}\}.
Ui={[ai,bi]:[A,b]∈U}. The RC of uncertain LO problem with certain objective remains intact when the original uncertainty set
U
\mathcal{U}
U is extended to the direct product
 
U
^
=
U
1
×
⋯
×
U
m
\,\hat\mathcal{U}=\mathcal{U}_1\times\cdots\times\mathcal{U}_m
U^=U1×⋯×Um of its projections onto the spaces of data of respective constraints.
C. If x x x is a robust feasible solution of (8), then x x x remains robust feasible when we extend the uncertainty set   U i \,\mathcal{U}_i Ui to its convex hull Conv ( U i ) \text{Conv}(\mathcal{U}_i) Conv(Ui).
D. When modeling an uncertain LO problem one should avoid whenever possible converting inequality constraints into equality ones, unless all the data in the constraints in question are certain.
Definition 5. A set X + ⊂ R x n × R u k X^+\subset \mathbf{R}_x^n\times\mathbf{R}_u^k X+⊂Rxn×Ruk is said to represent a set X ⊂ R x n X\subset\mathbf{R}_x^n X⊂Rxn, if the projection of X + X^+ X+ onto the space of x x x-variables is exactly X X X.
Why the representation argument is important? A representation can posses desired properties which are absent in the original problem. Such strategy allows us to focus on a single uncertainty-affected linear inequality: a family { a T x ≤ b } [ a , b ] ∈ U \{a^Tx\leq b\}_{[a,b]\in\mathcal{U}} {aTx≤b}[a,b]∈U of linear inequalities with the data varying in the uncertainty set (10) U = { [ a ; b ] = [ a 0 ; b 0 ] + ∑ ℓ = 1 L ζ ℓ [ a ℓ , b ℓ ] : ζ ∈ Z } \mathcal{U}=\left\{[a;b]=[a^0;b^0]+\sum_{\ell=1}^L\zeta_{\ell}[a^{\ell},b^{\ell}]:\zeta\in\mathcal{Z}\right\}\tag{10} U={[a;b]=[a0;b0]+ℓ=1∑Lζℓ[aℓ,bℓ]:ζ∈Z}(10) naming tractable representation.