Monotonicity
Monotonicity
Monotonicity
(1) f (u) = a.
The inverse function theorem is a “local” result for this problem, and we desire
something stronger. As an example of what we expect, consider the following
result when n = 1.
Theorem 1. Suppose f : R → R is continuous and satisfies
(3) f (u) = a
has a solution a ∈ R.
- Furthermore, if f is monotone increasing, then, for each a ∈ R, the set of
solutions forms a closed interval.
- If f is strictly monotone, the solution is unique.
The proof is an elementary application of the intermediate value theorem and the
definition of monotone and continuous functions.
• In order to get an analogue of this result for mappings from Rn to Rn , we
must generalize both the growth condition (2) and the definition of monotone
increasing (order preserving).
Definition. We say that a function f : Rn → Rn is coercive if
f (u) · u
(4) → ∞ as |u| → ∞.
|u|
(6) f (u) = a
• To get a result when f is only monotone, we first prove the following lemma.
Typeset by AMS-TEX
1
2
(7) (f (v) − a) · (v − u) ≥ 0 ∀v ∈ Rn .
(8) (f (u + tw) − a) · w ≥ 0 ∀w ∈ Rn .
Letting t → 0 gives us
(f (u) − a) · w ≥ 0 ∀w ∈ Rn .
Sv = {u ∈ Rn | (f (v) − a) · (v − u) ≥ 0}.
Note that each set Sv is a closed half-space (and is hence convex). By our lemma
3 we can write the solution set as
\
K= Sv
v∈Rn
Since the arbitrary intersection of closed sets is closed and the arbitrary intersection
of convex sets is convex, our theorem is proved.
f (u) = u.
Proof of Theorem 6. When n = 1 the proof is easy to see graphically. In this case
the hypotheses of the theorem states that we have a continuous function f which
maps an interval [a, b] into itself. The graph of the function must lie in the box
[a, b] × [a, b] in the (x, y)-plane. There are only three possibilities:
(a) f (a) = a,
(b) f (b) = b, or
(c) the graph of f starts to the left of the line x = y and ends on the right.
Since the function is continuous the graph must cross the line x = y
somewhere. A crossing gives a fixed point.
Proof of Theorem 5. Step 1. We begin by noting that it is sufficient to show that
f (u) = 0 has a solution.
To see this, let a ∈ Rn be given and define
Thus, if we can show that for every coercive function f (u) = 0 has a solution, it
immediately follows that f (u) = a has one for every a ∈ Rn .
Step 2. We now convert f (u) = 0 to a fixed point problem.
(a) As a first step, we define g : Rn → Rn by
f (u) = a.
hT (u), ui
(28) → ∞ as kuk → ∞.
kuk
Remark. Using the definitions above, the result of the previous section can easily
be extended to mappings T from an n-dimensional Banach space to another n-
dimensional Banach space Y .
– The only real trick here is in defining the appropriate bilinear form
X × Y 3 x, y 7→ hx, yi ∈ R
X × X ∗ 3 x, y 7→ hx, yi ∈ R
in the definition in this section and the dot product in the previous section.
• The proofs of Lemma 3 and Theorem 4 were based on monotonicity and were
in no way dependent on the dimension of the spaces involved. Thus, we can get
the following two analogue results using only minor changes of notation in the
previous proofs.
Lemma 8. Let g ∈ X ∗ be given and let T : X → X ∗ be continuous. Then if
u ∈ X is a solution of the variational inequality
(31) (T (v) − g, v − u) ≥ 0 ∀v ∈ X.
(32) T (u) = g.
(38) T (u) = g;
i.e. T (X) = x∗ .
Proof. Let {xi }∞
i=1 be a set whose span is dense in X. No let
(40) Tn (un ) = gn ,
(41) gn := g .
Xn
(44) Tn (un ) = gn .
un * u in X
and
T (un ) * ge in X ∗
We now claim: ge = g and T (u) = g.
7
g − g, xi i = lim hT (un ) − g, xi i
he
n→∞
Thus e
g = g.
- The key result here is that, once again using (43), we can get
(Since the left-hand side is the product of two weakly convergent sequences, the
result is nontrivial.) We now use this and the monotonicity of T to get that for
every v ∈ X
where
1 1
(48) A(u) = f with f ∈ W −1,q (Ω), where p ∈ (1, ∞), + = 1.
p q
Definition. We say that u ∈ W01,p (Ω) is a weak solution of the Dirichlet problem
for the quasilinear PDE (48) if
H-4. There exists p ∈ (1, ∞), and a constant c0 > 0, a function h ∈ L1 (Ω) such
that for every x ∈ Ω an every δ1 = (xiα )|α|≤1 ∈ Rn+1 we have
X
(54) Aα (x, δ1 )xiα ≥ c0 |δ1 |p − h(x).
|α|≤1
p
H-5. There exists a constant c1 > 0, a function g ∈ Lq (Ω), (q := p−1 ),
i n+1
such that for every x ∈ Ω an every δ1 = (xα )|α|≤1 ∈ R we have
• We break the proof of this theorem into a series of lemmas, the first of which is
the following.
Lemma 12. For each fixed u ∈ W01,p (Ω), the mapping
such that
B(u, v) = (T (u), v), ∀u, v ∈ W01,p (Ω).
Furthermore, the nonlinear mapping T is bounded.
9
X Z
≤ [c1 |δ1 (u(x))|p−1 + g(x)]|Dα v(x)| dx
|α|≤1 Ω
X Z 1/q Z 1/p
q(p−1) α p
≤ c1 |δ1 (u(x))| dx |D v| dx
|α|≤1 Ω Ω
X Z 1/q Z 1/p
q α p
+ |g| dx |D v| dx
|α|≤1 Ω Ω
≤C(kukp−1
1,p + 1)kvk1,p
This shows that for each u ∈ W01,p (Ω). the mapping v 7→ B(u, v) is bounded. This
implies the existence of the mapping T .
(iii) Furthermore, the same estimate shows that T is bounded since
kT (u)k−1,q = sup |hT (u), vi| ≤ C(kukp−1
1,p + 1).
v∈W01,p (Ω)
kvk1,p =1
Thus T maps bounded sets in W01,p (Ω) into bounded sets in W −1,q (Ω).
Lemma 13. The operator T defined in Lemma 12 is monotone and coercive.
Proof. (i) To prove monotonicity we use H-3 to get
hT (u) − T (v), u − vi =B(u, u − v) − B(v, u − v)
X Z
= [Aα (x, δ1 (u)) − Aα (x, δ1 (v))][Dα u − Dα v] dx ≥ 0.
|α|≤1 Ω
• Thus, to apply the Browder-Minty theorem to the mapping T and complete the
proof of Theorem 11 we need only the following.
10
Xn
∂ ∂u p−2 ∂u
(61) A(u) := − + |u|p−2 u, where p ∈ (1, ∞).
i=1
∂x i ∂x i ∂xi
(Note that the case p = 2 simply the Laplacian plus a lower-order term.) Here
X n
X n
X
Aα (x, δ1 )xiα = |xii |p−2 xii xii = |xii |p ≥ c0 |δi |p .
|α|≤1 i=1 i=1
Thus the following existence result follows immediately from Theorem 10.
Theorem 15. Let the nonlinear second order partial differential operator A be
defined by (61). Then for every f ∈ W −1,q (Ω), there exists a weak solution u ∈
W01,p (Ω) of the equation
A(u) = f.
4. Nemytskii Operators
In this section we state without proof some important result on the composition of
Lp (Ω) with nonlinear functions.
11
Ω × Rm 3 (x, u) 7→ f (x, u) ∈ R
and
x 7→ f (x, u) is measurable for every u ∈ Ω.
Ω × Rm 3 (x, u) 7→ f (x, u) ∈ R
1
satisfy the Carathéodory conditions. In addition, let p ∈ (1, ∞), p + 1q = 1 and
g ∈ Lq (Ω) be given, and let f satisfy
Then the Nemytskii operator F defined by (63) is a bounded and continuous map
from Lp (Ω) to Lq (Ω).
uj * u in X
and
lim suphT (uj ), uj − ui ≤ 0.
j→∞
It follows that
• The following can be proved using only a slight modification of the proof of the
Browder-Minty theorem.
Theorem 17. Let X be a reflexive Banach space and suppose T : X → X ∗ is
continuous, coercive and pseudo-monotone. Then for every g ∈ X ∗ there exists a
solution u ∈ X of the equation
T (u) = g.
• In practice, the following condition is easier to verify than pseudo-monotonicity.
Definition. Let X be a reflexive Banach space. An operator T : X → X ∗ is
said to be of the calculus of variations type if it is bounded, and it has the
representation
T (u) = Tb(u, u)
where the mapping
X × X 3 (u, v) 7→ Tb(u, v) ∈ X ∗
satisfies the following hypotheses.
CV-1 ∀u ∈ X, the mapping v 7→ Tb(u, v) is bounded, continuous from X to X ∗
and
then
Tb(uj , v) * Tb(u, v) in X ∗ , ∀v ∈ X.
CV-4 If
uj * u in X and Tb(uj , v) * ψ in X ∗ ,
then
hTb(uj , v), uj i → hψ, ui.
As we indicated above, we have the following.
13
We first claim:
and
(67) hT (uj ), uj − ui → 0.
Thus if we define
we have
lim sup xj = lim sup[hT (uj ), uj − ui − hTb(uj , u), uj i − hTb(uj , u), ui] ≤ 0;
j→∞ j→∞
here we have used (65), (68) and (69). Since CV-1 implies that xj → 0 we get
lim xj = 0.
j→∞
Tb(uj , v) * Tb(u, v) in X ∗ , ∀v ∈ X.
which is (66). We now use this and the fact that xj ≥ 0 to get
This yields
θhT (uj ), u−vi ≥ −(1−θ)hT (uj ), uj −ui+(1−θ)hTb(uj , w), uj −ui+θhTb(uj , w), u−vi.
=hTb(u, w), u − vi
=hTb(u, (1 − θ)u + θv), u − vi
Since this argument holds for any subsequence of the original sequence, the inequal-
ity (70) holds for the entire original sequence. This completes the proof.
The following is immediate from the preceeding results.
Corollary 19. Let X be a real reflexive Banach space and suppose T : X → X ∗
is continuous, coercive and of calculus of variations type. Then for every g ∈ X ∗
there exists a solution u ∈ X of the equation
T (u) = g.
6. Application to PDEs
• Let Ω ⊂ Rn be a bounded domain with smooth boundary.
Xn
∂
A(u) = − (x, u(x), ∇u(x)) + a0 (x, u(x), ∇u(x)).
i=1
∂x i
• Note that the hypotheses HP-1, HP-2 and HP-3, the bivariate form B(u, v) is
well defined for u, v ∈ W 1,p (Ω) where p ∈ (1, ∞) is given in HP-3.
• As above, we say that u ∈ W01,p (Ω) is a weak solution of the Dirichlet problem
for the quasilinear PDE (48) if
B(u, v) = hf, vi, ∀v ∈ W01,p (Ω).
• Hypotheses HP-1, HP-2 and HP-3 also imply that for each fixed u ∈ W01,p (Ω),
the mapping
W01,p (Ω) 3 w 7→ B(u, w) ∈ R
is a bounded linear functional. Thus, there exists a mapping
W01,p (Ω) 3 u 7→ T (u) ∈ W −1,q (Ω)
such that
B(u, v) = (T (u), v), ∀u, v ∈ W01,p (Ω).
Furthermore, the nonlinear mapping T is bounded.
• We now show the following.
16
where Z X
n
∂w
B1 (u, v, w) := ai (x, u(x), ∇v(x)) (x)
Ω i=1 ∂xi
B0 (u, w) := a0 (x, u(x), ∇u(x))w(x) dx
Using the augment as above, we see that there exists a mapping
such that
B(u, v, w) = hTb(u, v), wi, ∀w ∈ W01,p (Ω).
Furthermore, we have
T (u) = Tb(u, u).
We now note that our results on Nemytskii operators and hypotheses HP-1, HP-2
and HP-3 immediately imply the following lemma.
Lemma 21. For each i = 0, · · · , n, the mapping
This and hypothesis HP-4 show that the conditions CV-1 and CV-2 are satisfied.
Thus, to show that Tb is of the calculus of variations type and complete the proof
of Theorem 22, we need only verify that conditions CV-3 and CV-4 are satisfied.
(i) To check the condition CV-3, we assume that uj * u in W01,p (Ω) and that
(72)
hTb(uj , uj ) − Tb(uj , u), uj − ui
= B1 (uj , uj , uj − u) − B1 (uj , u, uj − u)
Z X n
∂
= ai (x, u(x), ∇uj ) − ai (x, u(x), ∇u)) (uj − u) dx → 0,
Ω i=1 ∂x i
(i.1) Now by compact imbedding uj → u strongly in Lp (Ω). Thus, even without using
hypothesis (72), we have
Z X
n
∂w(x)
B1 (uj , v, w) = ai (x, u(x), ∇v(x))
Ω i=1 ∂xi
→B1 (u, v, w);
The proof of this in the general case is an exercise in measure theory which we
shall skip.
– In our example below we use a lower-order term of the form
where b1 and α0 are bounded continuous functions. For such a term, condition
(73) can be verified directly without using (72).
(ii) To check condition CV-4 we assume that
i.e.
To do this, we write
• There is one more lemma to prove to get our main result on PDEs.
Lemma 23. The operator T is coercive.
• The culmination of the previous results is the following existence theorem for
quasilinear PDEs.
Theorem 24. Let ai satisfy (HP-1)-(HP-6), and let p, q be as defined in the
hypotheses (HP-3). Then, for every f ∈ W −1,q (Ω), there exists a weak solution
u ∈ W01,p (Ω) of the Dirichlet problem for the quasilinear PDE (71).
Xn
∂ ∂u p−2 ∂u
(75) A(u) := − + b1 (x)∇u(x) + α0 (x)|u(x)|p−2 u(x),
i=1
∂x i ∂x i ∂xi
where p ∈ (1, ∞), and where b1 and α0 are bounded continuous functions. Note
that this is the same as the operator defined in (61) except for the difference in the
lower-order terms.
e > 0 such that if
If there exists a constant C
e
α0 (x) > −C, x ∈ Ω,