Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Introduction to Differential Inclusions
September 26, 2007
Rick Barnard
Student Seminar on Control Theory and Optimization Fall 2007
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
1 Definitions
2 Selections
3 Differential Inclusions
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
References
Aubin, J.P. and Cellina, A. Differential Inclusions.
Springer-Verlag, Berlin, 1984.
Clarke, F.H. Optimization and Nonsmoorth Analysis.
Centre de Recherches Mathe´matiques, Montre´al, 1989.
Smirnov, G.S. Introduction to the Theory of Differential
Inclusions. AMS, Providence, 2002.
C. Cai, R. Goebel, R. Sanfelice, A. Teel. Lecture Notes
to Workshop on Robust Hybrid Systems: Theory and
Applications. CDc ’06.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Multifunctions
A multifunction F : Rm → Rn is a map from Rm to the
subsets of Rn, that is for every x ∈ Rm, we associate a
(potentially empty) set F (x).
Its graph, denoted Gr(F ) is defined by
Gr(F ) = {(x , y)|y ∈ F (x)}.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Measurability
A multifunction F : S → Rn is measurable if for every open
(closed) C ⊆ Rn,
{x ∈ S : F (x) ∩ C 6= ∅}
is Lebesgue measurable.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Continuity
A multifunction F is called upper semi-continuous at x0
if for any open M containing F (x0) there is a
neighborhood Ω of x0 so that F (Ω) ⊂ M.
A multifunction F is called lower semi-continuous at x0
if for any y0 ∈ F (x0) and any neighborhood M of y0
there is a negihborhood Ω of x0 so that
F (x) ∩M 6= ∅, ∀x ∈ Ω.
A multifunction is continuous at x0 if it is both upper and
lower semi-continuous at x0.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Continuity
A multifunction F is called upper semi-continuous at x0
if for any open M containing F (x0) there is a
neighborhood Ω of x0 so that F (Ω) ⊂ M.
A multifunction F is called lower semi-continuous at x0
if for any y0 ∈ F (x0) and any neighborhood M of y0
there is a negihborhood Ω of x0 so that
F (x) ∩M 6= ∅, ∀x ∈ Ω.
A multifunction is continuous at x0 if it is both upper and
lower semi-continuous at x0.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Lipschitz Continuity
A multifunction F is said to be Lipschitz continuous if there
is a k ≥ 0 so that for any x1, x2 ∈ Rm we have
F (x1) ⊂ F (x2) + k|x1 − x2|B.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
The selection problem
Given a multifunction F : Rm → Rn, a single-valued map
f : Rm → Rn is a selection if
f (x) ∈ F (x), ∀x ∈ Rm.
For what multifunctions are we assured of the existence of a
selection?
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
The selection problem
Given a multifunction F : Rm → Rn, a single-valued map
f : Rm → Rn is a selection if
f (x) ∈ F (x), ∀x ∈ Rm.
For what multifunctions are we assured of the existence of a
selection?
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Michael’s Selection Theorem and a Measurable
Selection Theorem
Theorem
Let F be a closed, convex, and lower semi-continuous
multifunction. Then there is a continuous selection from F .
Theorem
Let F be measurable, closed, and nonempty on S . Then
there is a measurable selection from F .
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Approximate Selections
Theorem
Let F be a convex, upper semi-continuous multifunction.
Then for ² > 0 there is a locally Lipschitz continuous
function f² whose range is in the convex hull of the range of
F and
Gr(f²) ⊂ Gr(F ) + ²B.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Differential Inclusions
We now turn our attention to the problem of solving
Differential Inclusions:
x˙(t) ∈ F (x(t)), t ∈ [0,T ], (1)
with x(0) = x0. We will assume that F is closed, convex,
and Lipschitz continuous with constant k > 0. We shall see
presently that these assumptions are not too restrictive,
especially when we concern ourselves with problems from
control theory.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
An Example from Control Theory
Consider the control system
x˙ = f (x , u)
where u ∈ U ⊂ Rk . We assume that f is Lipschitz in x , and
we allow any measurable functions u : [0, t]→ Rk so that
u(t) ∈ U a.e. and that for all x , f (x , u(t)) ∈ L1([0,T ]).
Consider the multifunction F defined by
F (x) = f (x ,U) =
⋃
u∈U
f (x , u).
We do the same for nonautonomous systems using the
right-hand side F (t, x(t)).
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
An Example from Control Theory
Consider the control system
x˙ = f (x , u)
where u ∈ U ⊂ Rk . We assume that f is Lipschitz in x , and
we allow any measurable functions u : [0, t]→ Rk so that
u(t) ∈ U a.e. and that for all x , f (x , u(t)) ∈ L1([0,T ]).
Consider the multifunction F defined by
F (x) = f (x ,U) =
⋃
u∈U
f (x , u).
We do the same for nonautonomous systems using the
right-hand side F (t, x(t)).
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Filippov’s Theorem
Theorem
Let f : Rm × Rk → Rn be continuous, and let v : Rm → Rn
be measurable. Assume U is compact so that v(x) ∈ f (x ,U)
a.e. Then there is a measurable u : Rm → U satisfying
v(x) = f (x , u(x)).
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Filippov’s Theorem
Sketch of Proof: We know that
U(x) = {u ∈ U| v(x) = f (x , u)}
has compact values. Let u(x) = (u1(x), . . . , uk(x)) ∈ U(x)
with u1(x) the smallest possible. We show that if ui (x) is
measurable on a compact set A for i < p then up(x) is as
well. Then we use the Lusin theorem to find a set A² where
the ui (x)’s and v(x) are continuous and m(A \ A²) ≤ ². We
then show that the sublevel sets of up(x) restricted to A² are
closed. So up(x) is measurable on A² and we then repeat
Lusin on this set to get an A2². Since ² is arbitrary we use
Lusin again and see it is measurable on A.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Filippov’s Theorem
Theorem
Let f : Rm × Rk → Rn be continuous, and let v : Rm → Rn
be measurable. Assume U is compact so that v(x) inf(x ,U)
a.e. Then there is a measurable u : Rm → U satisfying
v(x) = f (x , u(x)).
Obviously, trajectories of the control system are solutions to
the differential inclusion. This result means the converse
holds and so the system and the inclusion are equivalent.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Continuity and Relaxation of Differential
Inclusions
Proposition
Let U be compact and f : Rm × U → Rn be continuous
(Lipschitz) in x . Then F : x → f (x ,U) is continuous
(Lipschitz).
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Continuity and Relaxation of Differential
Inclusions
Proof.
We know that F has closed graph and is locally bounded.
Let N be a neighborhood of F (x) with sequences {xi}, {yi}
so that xi → x0, yi ∈ F (xi ), and yi /∈ N. Since
cl{F (x)| |x − x0| < δ} is compact, WLOG assume yi → y0.
But the graph is closed, so y0 ∈ F (x0). So F is upper
semi-continuous. The lower semi-continuity follows from the
continuity of x → f (x , u), ∀u ∈ U. The Lipschitz
statements are similarly straightforward.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Continuity and Relaxation of Differential
Inclusions
A trajectory is an absolutely continuous function
x : [a, b]→ Rn such that
x˙(t) ∈ F (t, x(t)) a.e.
We say F is integrably bounded if there is an integrable
function φ(·) such that |v | ≤ φ(t) for all v in F (t, x).
Proposition
A relaxed trajectory y is a trajectory for coF . That is,
y˙(t) ∈ coF (t, y(t)). If F is integrably bounded then any
relaxed trajectory y is within δ of a trajectory for F in the
sup norm.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Continuity and Relaxation of Differential
Inclusions
Proposition
Let U be compact and f : Rm × U → Rn be continuous
(Lipschitz) in x . Then F : x → f (x ,U) is continuous
(Lipschitz).
Proposition
A relaxed trajectory is a trajectory for coF . If F is integrably
bounded then any relaxed trajectory y is within δ of a
trajectory for F in the sup norm.
Introduction to Differential Inclusions
Introduction to
Differential
Inclusions
Definitions
Selections
Differential
Inclusions
Definitions
Selections
Differential Inclusions
Existence Theorem
Theorem
Assume that F (x) is closed, convex, and Lipschitzian. For
any x0 ∈ Rn there exist solutions to (1) with x(0) = x0.
Introduction to Differential Inclusions
Definitions
Selections
Differential Inclusions