b b

International Journal of Mathematics and M

Computer Science, 15(2020), no. 1, 325–345 CS

On Existence and invariance of sphere, of

solutions of constrained evolution equation

Javed Hussain

Department of Mathematics
Sukkur IBA University
Sukkur Sindh, Pakistan

email: [email protected]

(Received August 10, 2019, Revised October 4, 2019,

Accepted November 4, 2019)

Abstract

Hilbert manifolds are infinite dimensional manifolds modeled on

Hilbert spaces. The aim of this study is twofold. First, we study
the existence of the solution of a constrained nonlinear initial value
problem involving a diagonal operator, with values on a particular
submanifold (unit sphere) in a Hilbert space. Secondly, we investigate
the invariance of the submanifold; i.e., we find a sufficient condition
for the solutions to stay on the submanifold.

1 Introduction
The motivation for this work comes from the projected parabolic gradient
flows studied by Rybka [24] and Cafferelli, Lin [16]. Rybka studied the heat
equation in L2 (O) projected on a manifold M, where
 Z 
2 k
M= u ∈ L (O) ∩ C (O) : u (x)dx = Ck , k = 1, 2, ..., N ,
Ω

Key words and phrases: Hilbert Space, Constrained evolution equation,

Invariance of submanifold.
AMS (MOS) Subject Classifications: 35K15, 35R01.
ISSN 1814-0432, 2020, http://ijmcs.future-in-tech.net
326 J. Hussain

and D be a bounded, connected region in R2 . Rybka proved the global exis-

tence and uniqueness of the solution of the following projected heat equation,
 du P
= ∆u − N k=1 λk u
k−1
in Ω ⊂ R2
dt
∂u (1.1)
∂n
= 0 on ∂Ω, u(0, x) = u0

where λk = λk (u) are such that ut is orthogonal to Span uk−1 for a more
regular initial data. He also showed that the solutions to (1.1) converge to a
steady state as time tends to ∞.
Secondly, in [16], Cafarelli and Lin proved global existence and uniqueness
of energy-conserving solution to the heat equation. Their main result was to
prove the strong convergence of the solutions of these perturbed systems to
some weak-solutions of the limiting constrained non-local heat flows of maps
into a singular space.
Recently, in [7], Zdzisaw Brzeźniak, Gaurav Dhariwal and Mauro Mariani
studied 2D Navier–Stokes equations with a constraint forcing the conserva-
tion of the energy of the solution. They proved the existence and uniqueness
of a global solution for the constrained Navier-Stokes equation on R2 and
T2 , by a fixed point argument. They also showed that the solution of the
constrained equation converges to the solution of the Euler equation as the
viscosity ν vanishes.
In this paper, we consider a problem which links the aforementioned works.
Some of the classical and modern references on partial differential equations
on manifolds are [3], [4], [8], [9], [10], [15], [17], [18]. An interesting paper is
the recent work by Irem Akbulut and Cemil Tunç [20].

Let us assume that (H, h·, ·i) is an abstract Hilbert Space and let M be
a unit sphere in it; i.e.,

M = {u ∈ H : |u|H = 1} .

Assume that f denotes a vector field on H such that there exists a unique
solution of following class of constrained initial value problems (comprising
of an abstract evolution equation)

dx
= f (x (t)) , t ∈ [0, ∞) (1.2)
dt
x(0) = x0,

for every initial data x ∈ H.

On Existence and invariance of sphere... 327

It is interesting to observe that the trajectories (x(t))t≥0 of the solution of

above initial value problem do not necessarily satisfy the constraint to stay
on the sphere M even if we take x0 ∈ M. One of the key reasons that the
vector field f involved in the above initial value problem is not necessarily a
tangent vector to M; i.e., it does not satisfy

f (x) ∈ Tx M, for x ∈ D(f ) ∩ M. (1.3)

To turn a vector field f into a tangent vector on M at point x, We propose

to replace f with a new F in such a way that such a modification becomes
tangent to M and satisfies the property (1.3). The modification that we
propose is to orthogonally project the vector field f on M at x. For this we
are going to use the orthogonal projection πx : H → Tx M described by:

πx (y) = y − hy, xi x.
By applying the orthogonal projection to vector field f we infer that

F (x) ≡ πx (f ) = f − hf, xi x (1.4)

One can easily see that for every x ∈ M ∩ D(f ), we have

hF (x), xi = hπx (f ), xi = 0

Hence F satisfies property (1.3). We will verify the last fact in detail in
section 4.
It is worth noting that for the existence of the solution ( local at least)
of the initial value problem, it is important that f must be Lipschitz. More-
over, whenever f is Lipschitz and D(f ) = H, its modification F (i.e. the
orthogonal projection of f ) will also be globally Lipschitz and D(F ) = H.
The satisfaction of Lipschitz property guarantees that the modified equation
dx
= F (x (t)) , t ≥ 0 (1.5)
dt
x(0) = x0

has at least a local solution for every x ∈ M. In addition, this solution

satisfies the invariance property; i.e., this solution stays on M whenever
x ∈ M.

The key question that we are going to address now is that what if the
vector field f is not globally defined; i.e., D(f ) ⊂ H? Are we going to have
328 J. Hussain

invariance of the manifold? This aspect is completely unexplored. We will

study this invariance problem in a very particular case where f (x) = Ax
(which is not globally defined; i.e. D(f ) ⊂ H), and H = ℓ2 (R), where A
is discrete diagonal operator. Thus we are interested in the existence and
invariance of sphere, for the solutions of following initial values constrained
problem

dx
(t) = Ax(t), t ≥ 0 (1.6)
dt
x(0) = x0 ,

where x0 ∈ M = {u ∈ H : |u|H = 1} .

2 Notation and preliminaries

Definition 2.1. [29] Let ℓ2 (R) denote the space of the all infinite real se-
P
∞
quences u = (ui )ni=1 such that u2i < ∞. The norm on ℓ2 (R) is defined in
i=1
the following manner, v
u∞
uX
kuk := t u2 . i
i=1

Definition 2.2. [29](Continuous Embedding) Let X and Y be normed spaces.

Then a mapping I : X → Y : x → x is a continuous embedding if the opera-
tor I is continuous; i.e., kxkY ≤ N kxkX , where N can be taken as kIkL(X,Y )
.

Definition 2.3. [29](Bochner Measurable) Let a Banach space Y and a bounded

interval I ⊂ R be given. Then a function x : I → Y is a Bochner Measur-
able if it is point-wise limit of a sequence {xi }i∈N of simple functions; i.e.,
xi (t) → x(t) for all t ∈ I.

Definition 2.4. [29](Bochner Integrable function) Let a Banach space Y

and a bounded interval I ⊂ R be given. Then a function x : I → Y is called
a Bochner Integrable functions if
Z T
lim kx(t) − xi (t)kY dt = 0
i→∞ 0
On Existence and invariance of sphere... 329

implies Z Z
T T
x(t)dt = lim xi (t),
0 i→∞ 0

where {xi }i∈N is a sequence of simple functions.

Definition 2.5. [29](Bochner Spaces ,L2 (I, Y ), C(I, Y )) Let a Banach space
Y and a bounded interval I ⊂ R be given. R T Thenp a linear space Bochner
Integrable functions x : I → Y satisfying 0 kx(t)kY dt < ∞, for 1 < p < ∞,
is a Bochner Space, denoted by Lp (I; Y ), with norm is defined as follows
Z T  p1
kxkLp (I;Y ) := kx(t)kpY dt .
0

Definition 2.6. [29] (Isometric Isomorphism) let X and Y be normed vec-

tor spaces. An operator T : X → Y which is both an isometry and an
isomorphism, is called anisometric isomorphism.

kf ∗ gkr ≤ kf kp kgkq .

Definition 2.7. Let A : D(A) → H with

( ∞
)
X
D (A) = x ∈ H : λ2k, xk , xk ∈ R ,
k=1

where 0 < λ1 < λ2 ..., defined by

Ax := (λ1 x1 , λ2 x2 , ...) ,

where x ∈ D(A), is called a diagonal operator.

Lemma 2.8. ([26], Lemma III 1.2) Let V, H be two Hilbert spaces, V ∗ and
H∗ be corresponding dual spaces. Assume that V ֒→ H = H∗ ֒→ V ∗ , where
the embedding i dense as well. If a function u belongs to L2 (0, T ; V) and its
weak derivative u′ belongs to L2 (0, T ; V), then u is a.e. equal to an absolutely
continuous function from [0, T ] into H, and the following equality holds in
sense of distributions on (0, T ):
d
ku(t)k2H = 2 hu′ , ui . (2.1)
dt
Lemma 2.9. [29] (Young’s inequality) Let 1 ≤ p, q ≤ ∞ and r satisfies
1
r
= p1 + 1q − 1. Suppose f ∈ Lp (R) and g ∈ Lq (R) then
330 J. Hussain

2.1 Important space and results

In this section, we assume that V, H are two Hilbert spaces and V ∗ , H∗ be
corresponding dual spaces respectively. Moreover, assume that embeddings
V ֒→ H = H∗ ֒→ V ∗ are dense and continuous. Recall the following well-
known sequence spaces:
( ∞
)
X
ℓ2 R) = (x)∞
i=1 ∈ R
∞
: |xi |2 < ∞ ,
i=1
( ∞
)
X
ℓ2λ (R) = (x)∞ 2
i=1 ∈ ℓ (R) : λi |xi |2 < ∞ ,
i=1
( ∞
)
X
ℓ2λ−1 (R) = (x)∞ 2
i=1 ∈ ℓ (R) : λ−1 2
i |xi | < ∞ .
i=1

The aim of the section is to show that we can identify the V, H, V ∗ as

ℓ2λ (R), ℓ2 (R) and ℓ2λ−1 (R) respectively, through isometric isomorphisms. We
will do our analysis in sequence spaces and demonstrate that these spaces
satisfy the assumptions of the Lemma 2.8.

Theorem 2.10. Let H be a Hilbert Space with the standard basis {ei }∞
=1 .
2 2
Then ℓ (R) is isometrically isomorphic to ℓ (R)

Proof.
P∞ Since {ek }∞
k=1 is a basis of H, each x ∈ H can be uniquely written as
x = i=1 xi ei . Define a map T : ℓ2 (R) → H by
∞
X
T (x) = T ((xi )∞
i=1 ) = xi ei ,
i=1

where x = (xi )∞ 2
i=1 ∈ ℓ (R) . Clearly T is linear and bijective. We will show
that T is an isometry. If x ∈ ℓ2 (R) , then
∞
X ∞
X
kT (x)kH = xi ei = |xi |2 = kxkℓ2 (R)
i=1 H i=1

Thus T is an isometric isomorphism.

Theorem 2.11. If V be a Hilbert space and densely and continuously em-

bedded in H, then V is isometrically Iiomorphic to ℓ2λ (R).
On Existence and invariance of sphere... 331

Proof. We begin by showing that H ≃Pℓ2λ (R). Since H has basis {ek }∞ k=1 each
x ∈ H can be uniquely written as x = ∞ x e
i=1 i i . Define a map T : ℓ 2
λ (R) →H
by
X ∞
∞
T (x) = T ((xi )i=1 ) = λi xi ei ,
i=1
Next we show that T is an isometry.
∞
X ∞
X
kT (x)kV = xi ei = λi |xi |2 = kxkℓ2 (R) .
λ
i=1 V i=1

Thus V is isometrically isomorphic to ℓ2λ (R) .

Theorem 2.12. Let V ′ be the Dual space of a Hilbert Space V. Then V ′ is

isometrically isomorphic to ℓ2λ−1 (R).
Proof. To prove that V ′ is isometrically isomorphic to ℓ2λ−1 (R) , it is sufficient
to show that the dual of ℓ2λ (R) is ℓ2λ−1 (R) . To do so, let x ∈ ℓ2λ−1 (R) . Define
Tx : ℓ2λ (R) → R by
∞
X
Tx (y) = xi yi , where y ∈ ℓ2λ (R) ,
i=1

Using Cauchy-Schwartz inequality,

∞ ∞
!1/2 ∞
!1/2
X X X |xi |2
|Tx (y)| = xi yi ≤ λi |yi |2
i=1 i=1 i=1
λi
= kykℓ2 (R) kxkℓ2 (R) <∞ (2.2)
λ λ−1

∗
which shows that Tx ∈ (ℓ2λ (R)) . Linearity of Tx is obvious. To show Tx is
bounded we can use inequality (2.2),
|Tx (y)| ≤ kykℓ2 (R) kxkℓ2 (R)
λ λ−1
!
|Tx (y)|
sup ≤ kxkℓ2 (R)
y6=0,y∈ℓ2λ (R) kykℓ2 (R) λ−1
λ

kTx k(ℓ2 (R))∗ ≤ kxkℓ2 (R) . (2.3)

λ λ−1

∗
Hence Tx ∈ (ℓ2λ (R)) . This observation allows us to define a map φ : ℓ2λ−1 (R) →
∗
(ℓ2λ (R)) by
φ(x) = Tx , x ∈ ℓ2λ−1 (R) .
332 J. Hussain

We show that φ is an isometric isomorphism. φ is well defined because for

any x, y ∈ ℓ2λ−1 (R) such that x = y we have Tx = Ty and so φ(x) = φ(y). In
addition, linearity is evident. To show isometry it is sufficient to show that
kTx (y)k(ℓ2 (R))∗ = kxkℓ2 (R) . In inequality (2) we have already shown that
λ λ−1

kTx k(ℓ2 (R))∗ ≤ kxkℓ2 (R) .

λ λ−1

It remains to show that kTx k(ℓ2 (R))∗ ≥ kxkℓ2 (R) . Let y ∈ ℓ2λ (R) . Then
P λ λ−1
y= ∞ y e
i=1 i i . Using linearity of T x , we get
∞
! ∞ ∞
X X X
Tx (y) = Tx yi ei = yi Tx (ei ) = yi γi ,
i=1 i=1 i=1

where γi = Tx (ei ) . Now define yn = (yin )∞ n

n=1 , where yi is given by

|γi |2
yin = , if i ≤ n, γi 6= 0
γ i λi
= 0, if i > n.

Consider the value Tx at yin ,

∞
X n
X n
X
|γi |2 |γi |2
Tx (yn ) = yin γi = γi = . (2.4)
i=1 i=1
γ i λi i=1
λi

Using continuity of T we get

∞
!1/2 ∞
!1/2
X X |γi |2
Tx (yn ) ≤ kTx k(ℓ2 (R))∗ λi |yin |2 = kTx k(ℓ2 (R))∗
λ
i=1
λ
i=1
λi

and now using (2.3) we get

n ∞
!1/2
X |γi |2 X |γi |2
≤ kTx k(ℓ2 (R))∗
i=1
λi λ
i=1
λi

Take n → ∞ and divide by the last factor on both sides we get

∞
!1/2
X |γi |2
≤ kTx k(ℓ2 (R))∗ i.e. kxkℓ2 (R) ≤ kTx k(ℓ2 (R))∗ .
i=1
λi λ λ−1 λ
On Existence and invariance of sphere... 333

Thus we conclude that kxkℓ2 (R) = kTx k(ℓ2 (R))∗ and hence the dual of ℓ2λ (R)
λ−1 λ
∗
is ℓ2λ−1 (R) . It remains to show that φ is surjective; i.e., for each η ∈ (ℓ2λ (R))
there exist x ∈ ℓ2λ−1 (R) such that φ(x) = η or Tx = η. For n ∈ N, define a
sequence yn = (yin ), where

|γi |2
yin = , ifi ≤ n, γi 6= 0ifTx (ei ) = γi
λi γ i
= 0 otherwise.
Pn
Observe that yn = i=1 yin ei . So
n
X n
X n
X |γi |2
Tx (yn ) = yin Tx (ei ) = yin γi = .
i=1 i=1 i=1
λi

Therefore
n n
!1/2 n
!1/2
X |γi |2 X X |γi |2
= Tx (yn ) ≤ kTx k(ℓ2 (R))∗ λi |yin |2 = kTx k(ℓ2 (R))∗
i=1
λi λ
i=1
λ
i=1
λi

which implies that

n
!1/2
X |γi |2
≤ kTx k(ℓ2 (R))∗ , forall n.
i=1
λi λ

Pn |γi |2
It follows that i=1 λi
< ∞ and so (γi ) ∈ ℓ2λ−1 (R) as required. This
completes the proof.

3 Abstract Cauchy problem with diagonal op-

erators
In this section we will show that the trajectories of the solution of the follow-
ing abstract Cauchy operator with Diagonal operator satisfies the conditions
of Lemma 2.8 . Consider
dx
(t) + Ax(t) = f (t), t ≥ 0 (3.1)
dt
x(0) = x0 ,
334 J. Hussain

where x0 ∈ V ≃ ℓ2λ (R) and f = (fk ) ∈ L2 (0, T ; V) . To do so, we show that

x = (xk )∞ 2 ′ 2 ′ 2
k=0 ∈ L ((0, T ), V) and x ∈ L ((0, T ), V ) , where V ≃ ℓλ (R) and
V ′ ≃ ℓ2λ−1 (R). The solution of the above problem can be given as,

Zt
x(t) = x0 + e−(t−s)A f (s)ds,
0

which is equivalent to
Zt
xk (t) = x0,k + e−(t−s)λk fk (s)ds = x0,k + (ϕk ∗ fk )(t).
0

Here

ϕk (t) = e−tλk , if t > 0

= 0, if t ≤ 0

and

fk (t) = fk (t), if t > 0

= 0, if t ≤ 0

and
Z t
(ϕk ∗ fk )(t) = ϕ(t − s)f (s)ds = 0, if t ≤ 0
−∞
Z t
= ϕ(t − s)f (s)ds, if t > 0.
−∞

Let us begin by showing that x = (xk )∞ 2

k=0 ∈ L ((0, T ), V) and compute
2
kxkL2 ((0,T ),V) .
Z T
kxk2L2 ((0,T ),V) = kx(t)k2V dt
0
Z ∞
!
T X
= λk |xk (t)|2 dt
0 k=1
∞
X Z T
= λk |xk (t)|2 dt
k=1 0
On Existence and invariance of sphere... 335

∞
X Z T
= λk |x0,k + (ϕk ∗ fk )(t)|2 dt
k=1 0

X∞ Z T

≤ 2 λk |x0,k |2 + |(ϕk ∗ fk )(t)|2 dt
k=1 0

X∞ ∞
X Z T
2
≤ 2 λk |x0,k | + λk |(ϕk ∗ fk )(t)|2 dt
k=1 k=1 0
∞
X
= 2 kx0 k2V + λk kϕk ∗ fk kL2 (0,T )
k=1

Moreover, by using Young’s inequality 2.9, for p = 1, q = 2 and r = 2, we

get
∞
X
kxk2L2 ((0,T ),V) ≤ 2 kx0 k2V +2 λk kϕk (t)kL1 (0,T ) kfk (t)k2L2 (0,T )
k=1
X∞ Z T 2
= 2 kx0 k2V +2 λk e −tλk
dt kfk (t)k2L2 (0,T )
k=1 0

X∞ 2 
1 − e−T λk
= 2 kx0 k2V
+2 λk kfk (t)k2L2 (0,T )
k=1
T λk
∞
2
2 2 X 1 − e−T λk
= 2 kx0 kV + kfk (t)k2L2 (0,T )
T λk
k=1

Using the elementary inequality, (1 − ex )2 ≤ x2 , for all x ≤ 0 and Tenolli’s

Theorem
∞
1 X T 2 λ2k
kxk2L2 ((0,T ),V) ≤ 2 kx0 k2V + kfk (t)k2L2 (0,T )
T k=1 λk
∞
X
= 2 kx0 k2V +T λk kfk (t)k2L2 (0,T )
k=1
∞ Z
X T
= 2 kx0 k2V + T λk |fk (t)|2 dt
k=1 0
Z ∞
!
T X
= 2 kx0 k2V +T λk |fk (t)| 2
dt
0 k=1
Z T
= 2 kx0 k2V +T kfk (t)k2V dt
0
336 J. Hussain

= 2 kx0 k2V + T kf k2L2 (0,T ;V)

Since x0 ∈ V and f ∈ L2 (0, T ; V) , kxkL2 ((0,T ),V) < ∞; i.e., x ∈ L2 (0, T ; V) .
Next we show x′ ∈ L2 (0, T ; V) . Let us compute kx′ kL2 (0,T ;V ′) .
Z T Z T X ∞
! Z T X ∞
!
′ 2 ′ 2
2 2 |xk (t)| |(ϕk ∗ fk )(t)|
kx′ kL2 (0,T ;V ′) = kx′ (t)kV dt = dt = dt
0 0 λk 0 λk
k=1 k=1
X∞ Z T X∞
1 2 1 2
= |(ϕ′k ∗ fk )(t)| dt = kϕ′k ∗ fk kL2 (0,T )
k=1
λ k 0
k=1
λ k

Using Young’s inequality for p = 1, q = 2 and r = 2,

X∞
′ 2 1
kx kL2 (0,T ;V ′) = kϕ′k kL1 kfk kL2 (0,T ) (3.2)
k=1
λ k

Since
d −tλk

ϕ′k (t) = e = −λk e−tλk ,
dt
Z T Z T Z T
1 − e−T λk
kϕ′k kL1 = |ϕ′k (t)| dt = −λk e−tλk
dt = λk e−tλk dt = .
0 0 0 T
 −T λk
2
2 1−e
kϕ′k kL1 ≤ ≤ T 2 λ2k .
T
Therefore, using last inequality in inequality (3.2), we infer,
X∞ X∞ Z T
′ 2 T 2 λ2k
kx kL2 (0,T ;V ′) ≤ kfk kL2 (0,T ) = T 2
λk |fk (t)|2
λk 0
k=1 k=1
Z TX ∞ Z T
= T2 λk |fk (t)|2 dt = T 2 kfk (t)k2V dt
0 k=1 0

= T kf kL2 (0,T ;V) < ∞.

Since f ∈ L2 (0, T ; V) , kx′ kL2 (0,T ;V ′) < ∞. Thus x′ ∈ L2 ((0, T ), V ′ ).

4 Invariance of Sphere
Let H be an abstract Hilbert-space. Consider the evolution equation
dx
= f (x), x ∈ H (4.1)
dt
x(0) = x0 ,
On Existence and invariance of sphere... 337

where f is a vector field on H and x0 ∈ V ∩ M. Assume that

M = {x ∈ H : ϕ(x) = 0} ,
and ϕ : H → R is a fixed smooth function and chosen in such a way that M
can be a Hilbert submanifold of H .
Our aim is to study the sufficient conditions on the vector field f that
assures the invariance of submanifold M; i.e., if x0 ∈ M and all x(t) of the
solution of constrained problem, belong of M, for all t ≥ 0
Remark 4.1. 1. A vector field f is said to satisfy the property (∗ ) on
manifold M if for each x ∈ M and f (x) ∈ Tx M.
2. For each x ∈ M, we have ∇x ϕ ∈ H defined as
dx ϕ(y) = h∇x ϕ, yi , where y ∈ H. (4.2)
where dx ϕ(y) is the Frećhet derivative of ϕ at x evaluated at y.
3. One can see easily that dx ϕ ∈ L(H, R) for all x ∈ M, and
Tx M = {y ∈ H : dx ϕ(y) = 0} .
Proposition 4.2. For all x ∈ M,
Tx M = {y ∈ H : hdx ϕ, yi = 0} .
Proof. Set A = {y ∈ H : hdx ϕ, yi = 0} . We show that Tx M = A. Let y ∈
Tx M. Then dx ϕ(y) = 0 which implies that hdx ϕ, yi = h0, yi = 0 so y ∈ B
and Tx M ⊂ A. For the converse, take an arbitrary y ∈ A.
hdx ϕ, yi = 0.
Since dx ϕ(y) = h∇x ϕ, yi
hh∇x ϕ, yi , yi = 0
h∇x ϕ, yi h1, yi = 0
Clearly h∇x ϕ, yi is scalar and y is arbitrary. So h1, yi is not necessarily zero.
It follows that
h∇x ϕ, yi = 0, forall y ∈ A.
Hence, by equation (4.2),
dx ϕ(y) = h∇x ϕ, yi = 0, forall y ∈ A.
Hence A ⊂ Tx M. Since the inclusion is two sided, Tx M = A.
338 J. Hussain

Consider now the case when the f involved in equation (4.1) does not
satisfy the property (∗). In this case, we can modify f so that F satisfies
the property (∗). We can achieve this by projecting vector field onto tangent
space at a point on M i.e. πx : H → Tx M

πx (f ) ≡ F (x) := f (x) − hf (x), xi x. (4.3)

We now state the key result of this section.

Theorem 4.3. Let M = {x ∈ H : ϕ(x) = 0} and f does not satisfy (∗). For
a fixed x ∈ M ,we define πx : H → Tx M as
 
∇x ϕ ∇x ϕ
πx (y) = y − y, .
k∇x ϕk k∇x ϕk

Then πx (y) ∈ Tx M, for all y ∈ M, and f (x) is modified as F (x) = πx (f (x))

such that it satisfies property (∗).

b x ϕ(y) =
Proof. Set ∇ ∇x ϕ(y)
. For y ∈ H, consider
k∇x ϕ(y)k

D D E E
hπx (y), ∇xϕ(y)i = y − y, ∇b xϕ(y) ∇b x ϕ(y), ∇xϕ(y)
DD E E
b b
= hy, ∇xϕ(y)i − y, ∇xϕ(y) ∇x ϕ(y), ∇xϕ(y)
D ED E
= hy, ∇xϕ(y)i − y, ∇b xϕ(y) ∇ b x ϕ(y), ∇xϕ(y)
hy, ∇xϕi h∇x ϕ(y), ∇x ϕ(y)i
= hy, ∇xϕ(y)i −
k∇x ϕ(y)k2
hy, ∇xϕ(y)i k∇x ϕ(y)k2
= hy, ∇xϕ(y)i −
k∇x ϕ(y)k2
= hy, ∇xϕ(y)i − hy, ∇xϕ(y)i = 0

so that πx (y) ∈ Tx M. In particular, if we choose y = f (x), then

F (x) := πx (f (x)) ∈ Tx M, forall x ∈ M.

On Existence and invariance of sphere... 339

4.1 Key result regarding invariance of Sphere in ℓ2λ (R)

In this section, we assume that H = ℓ2 (R), V = ℓ2λ (R) and V ′ = ℓ2λ−1 (R).
Keeping in view the notation of the last section and choosing ϕ(x) = kxk2H −1,
we infer that
M = {x ∈ H : ϕ(x) = 0}
= {x ∈ H : kxkH = 1} .
In the next lemma, we compute the Frećhet derivative of ϕ(x) = kxk2H −1.
Lemma 4.4. Let the map ϕ : H → R be defined by
ϕ(u) = kuk2H − 1.
Then ϕ has Frećhet derivative dx ϕ given by
du ϕ (v) ≡ h∇u ϕ(x), vi = hu, vi ,
for all v ∈ H.
Proof. Since ϕ is a real-valued polynomial, it is smooth. Let us first calculate
the first order Frećhet derivatives of ϕ. For any u and h ∈ H,
1 1
ϕ (u + th) − ϕ (u) = ku + thk2H − kuk2H ,
2 2
1 1 1
= kuk2H + kthk2H + hu, thi − kuk2H ,
2 2 2
t2
= khk2H + t hu, hi
2
ϕ (u + th) − ϕ (u) t
lim = lim khk2H + hu, hi
t→0 t t→0 2
= hu, hi ,
for allh ∈ H.
Define a functional du ϕ : H → R by
du ϕ (h) := hu, hi , (4.4)
for all h ∈ H.
Clearly, du ϕ is linear. Moreover, to prove continuity, it is sufficient to prove
boundedness. For h ∈ H, consider
|du ϕ (h)| = |hu, hi| ≤ kukH khkH = M khkH ,
340 J. Hussain

where M = |u|H < ∞. Thus

|ϕ (u + h) − ϕ (u) − du ϕ (h)| o (khkH )

= → 0 as khkH → 0.
khkH khkH

Hence, we conclude the first order continuous Frećhet derivative of ϕ, given

by
h∇u ϕ(u), vi ≡ du ϕ (h) = hu, hi , (4.5)
for all h ∈ H.

Remark 4.5. From the last lemma, it follows that ∇u ϕ(x) = u, for all
u ∈ H.

We are now interested in studying the evolution equation (4.1) for the
case f (x) := Ax, where A is diagonal operator on H , defined in Definition
2.7.
dx
= −Ax, (4.6)
dt
x(0) = x0 .

Note that D(A) is not globally defined in H; i.e., D(A) ⊂ H. We claim

that if we project f (x) := Ax orthogonally onto the tangent space of the
sphere M in H and modify f by F (i.e., F = πx (f )) in problem 4.6 in the
following manner,
dx
= πx (−Ax), (4.7)
dt
x(0) = x0 ,

where x0 ∈ V ∩ M. We mainly want to show that in the above projected

problem if we assume that initial data of the problem x0 ∈ M, then the all
trajectories of the solution of problem 4.7 are going to also belong to M; i.e.,
we have invariance of sphere. Before getting into the proof of the invariance
of sphere consider the following important observation

Proposition 4.6. If πx : H → Tx M be the orthogonal projection and f :

H → H be any vector field on H, then F (x) = πx (f (x)) satisfies the property
(∗) on M = {x ∈ M : kxkH = 1} .
On Existence and invariance of sphere... 341

Proof. Recall property (∗): f is said to satisfy the property (∗) on manifold
M if for each x ∈ M and f (x) ∈ Tx M. We will verify this for F (x) =
πx (f (x))
For x ∈ M so

hF (x), xiH = hπx (f (x)) , xiH

= hf (x) − hf (x), xi x, xiH
= hf (x), xiH − hhf (x), xi x, xiH
= hf (x), xiH − hf (x), xiH hx, xiH
= hf (x), xiH − hf (x), xiH kxkH
= hf (x), xiH − hf (x), xiH = 0.

Hence F (x)⊥x, for all x ∈ M; i.e., F (x) ∈ Tx M, for all x ∈ M.

Remark 4.7. Recall that we have chosen H = ℓ2 (R), V = ℓ2λ (R), for this sec-
tion. Let x = (xi )∞
i=1 ∈ D(A) H, where A is diagonal operator defined in Def-
inition 2.7. We want to explicitly compute the expression for the πx (−Ax).
Consider

πx (−Ax) = −Ax − h−Ax, xiH x

= −Ax + h(λi xi )∞ ∞
i=1 , (xi )i=1 iH x
∞
!
X
= Ax − λi x2i x
i=1

= −Ax + kxk2V x

Hence problem 4.7 can be reformulated as

dx
= πx (−Ax) = −Ax + kxk2V x, (4.8)
dt
x(0) = x0 .

We intend to prove our key result.

Theorem 4.8. If x0 ∈ V ∩ M, then all trajectories of solution of 4.8 belong

to M; i.e., x(t) ∈ M, for all t ≥ 0.

Proof. Assume that x0 ∈ V ∩M. We show that all trajectories of the solution
belong to M. We employ Lemma 2.1. Note that problem 4.8 becomes a
342 J. Hussain

special case of the initial value problem 3.1, when f (t) = kxk2V x(t). In the last
section, we have already shown that x ∈ L2 ((0, T ), V) (i.e., L2 ((0, T ), ℓ2λ(R)))
and x′ ∈ L2 ((0, T ), V ′ ) (i.e., L2 ((0, T ), ℓ2λ−1 (R))). Hence the conditions of the
Lemma are satisfied. so we begin by using the identity (2.1) i.e.
1d

kx(t)k2H − 1 = hx′ (t), x(t)iH
2 dt
From equation (4.8) we infer that, for t ≥ 0,
1d

kx(t)k2H − 1 = −Ax(t) + kx(t)k2V x(t), x(t) H
2 dt
= − hAx(t), x(t)i + kx(t)k2V x(t), x(t) H
= − hAx(t), x(t)i + kx(t)k2V hx(t), x(t)iH
= − hAx(t), x(t)i + kx(t)k2V kx(t)k2H

From Remark 4.7, we know that h−Ax(t), x(t)iH = kx(t)k2V , and using this
fact in the last equation it follows that
1d

kx(t)k2H − 1 = kx(t)k2V kx(t)k2H − kx(t)k2V
2 dt

= kx(t)k2V kx(t)k2H − 1

Set u(t) = kx(t)k2H − 1

du
(t) = kx(t)k2V u(t),
dt
Since kx(t)k2V is polynomial, it must be continuous in time. Hence the solu-
tion of the above evolution equation is given as
2
u(t) = u(0)ekx(t)kV

Since x(0) = x0 ∈ M so u(0) = kx(0)k2H − 1 = kx0 k2H − 1 = 0, it follows

from the last equation, that

u(t) = 0,
kx(t)k2H − 1 = 0,
kx(t)k2H = 1.

Thus we show that x(t) ∈ M, for all t ≥ 0. This completes the proof of the
invariance of the sphere M.
On Existence and invariance of sphere... 343

References
[1] H. Brezis, Functional Analysis, Sobolev Spaces and Partial Differential
Equations, Springer, 2010.

[2] Z. Brzeźniak, T. Zastawniak, Basic Stochastic Process, Springer, 1999.

[3] Z. Brzeźniak, K. D. Elworthy, Stochastic differential equations on Ba-

nach manifolds, Methods Funct.Anal. Topology, 6, no. 1, (2000), 43–84.

[4] Z. Brzeźniak, A. Carroll, Approximations of the Wong–Zakai type for

stochastic differential equations in M-type 2 Banach spaces with appli-
cations to loop spaces, Sémi. de Prob. XXXVII, Springer Lecture Notes
in Math., 1832, (2003), 251–289.

[5] Z. Brzeźniak, E. Hausenblas, P. Razafimandimby. Stochastic Non-

parabolic dissipative systems modeling the flow of Liquid Crystals:
Strong solution. Preprint arXiv:1310.8641. See also RIMS Kôkyûroku
“Proceeding of RIMS Symposium on Mathematical Analysis of Incom-
pressible Flow, February 2013”.

[6] Z. Brzeźniak, B. Maslowski, J.Seidler., Stochastic nonlinear beam equa-

tions., Prob. Theory & Rel. Fields, 132, no. 1, (2005), 119–149.

[7] Z. Brzeźniak, G. Dhariwal, M. Mariani, 2D constrained Navier-

Stokes equations, J. Differential Equations, (2017), 27 pages,
https://doi.org/10.1016/j.jde.2017.11.005

[8] Z. Brzeźniak, M. Ondrejat, Stochastic Geometric Wave Equations.,

Stoch. Anal. Lecture Series, Sringer-Verlag, 68,, (2010), 157–188.

[9] Z. Brzeźniak, M. Ondrejat, Weak solutions to stochastic wave equations

with values in Riemannian manifolds, Comm. Part. Diff. Eqs., 36, no.
9, ( 2011), 1624–1653.

[10] Z. Brzeźniak, B. Goldys, M. Ondrejat, Stochastic Geometric Partial

Differential Equations, New Trends in Stoch. Anal. and Related Topics,
A Volume in Honor of Professor K. D. Elworthy, 12, World Sci. Pub.
Co.,(2012), 1–32.

[11] Z. Brzeźniak, A. Millet, On the stochastic Strichartz estimates and

the stochastic nonlinear Schrödinger equation on a compact riemannian
manifold, Potential Analysis, 41, no. 2, (2014), 269–315.
344 J. Hussain

[12] Z. Brzeźniak, E. Motyl, Existence of a martingale solution of Stochastic

Naviour-Stokes equation in unbounded 2D and 3D domains, Preprint

[13] Z. Brzeźniak, E. Hausenblas, P. Razafimandimby, Stochastic Non-

parabolic dissipative systems modeling the flow of Liquid Crystals: Strong
solution, Kurenai, 1875, (2014), 41–72.

[14] Z. Brzeźniak, B. Goldys, T. Jegraj., Large deviations and tran-

sitions between equilibria for Stochastic Landau-Lifshitz equation,
http://www.maths.usyd.edu.au/u/pubs/publist/preprints/2015/goldys-
7.pdf

[15] Z. Brzeźniak, A. Carroll, The stochastic nonlinear heat equation, in

preparation.

[16] L. Caffarelli, F. Lin, Nonlocal heat flows preserving the L2 -energy, Dis-
crete and Continuous Dynamical Systems, 32, (2009), 49–64.

[17] A. Carroll, The stochastic nonlinear heat equation. PhD thesis, The Uni-
versity of Hull, 1999.

[18] T. Funak, A stochastic partial differential equation with values in man-

ifold, J. Fun. Anal., 109, no. 2, (1992), 257–258.

[19] D. Henry, Geometric theory of semilinear parabolic equations, Lecture

Notes in Math., Springer , 840, 1981.

[20] Irem Akbulut, Cemil Tunç, On the stability of solutions of neutral dif-
ferential equations of first order, Int. J. Math. Comput. Sci., 14, no. 4,
(2019), 849-866.

[21] M. Jandoubi, A Simple Unified Approach to Some Convergence Theo-

rems of L. Simon, Journal of Functional Analysis 153, (1998), 187–202.

[22] E. Pardoux, Stochastic differential equations and filtering of diffusion

process, Stochastics, 3, (1979), 127–167.

[23] A. Pazy, Semigroups of Linear operator and applications to partial dif-

ferential equations, Springer-Verlag, 1983.

[24] P. Rybka, Convergence Of Heat Flow On a Hilbert Manifold, Proceedings

of the Royal Society of Edinburgh, 136A, (2006), 851–862.
On Existence and invariance of sphere... 345

[25] Z. Songu, Non-Linear evolution equations, Champan and Hall/CRC,

2004.

[26] R. Temam, Infinite-Dimensional Dynamical Systems in Mechanics and

Physics, App. Math. Sciences, 68, (1988).

[27] R. Temam, Navier-Stokes Equations, AMS Chelsa Pub., 2000.

[28] I. I. Vrabie, C0 -semigroups and application., North-Holland Math. Stud-

ies, 191, (2003).

[29] N. Young, An introduction to Hilbert space, Cambridge Univ. Press,

1995.