Self-consistent bounds method for dissipative

PDEs
Daniel Wilczak and Piotr Zgliczyński1
arXiv:2502.09760v1 [math.AP] 13 Feb 2025

Jagiellonian University, Faculty of Mathematics and Computer Science,

Lojasiewicza 6, 30–348 Kraków, Poland
e-mail: {Daniel.Wilczak,Piotr.Zgliczynski}@uj.edu.pl

February 17, 2025

Abstract
We discuss the method of self-consistent bounds for dissipative PDEs
with periodic boundary conditions. We prove convergence theorems for a
class of dissipative PDEs, which constitute a theoretical basis of a general
framework for construction of an algorithm that computes bounds for the
solutions of the underlying PDE and its dependence on initial conditions.
We also show, that the classical examples of parabolic PDEs including
Kuramoto-Sivashinsky equation and the Navier-Stokes on the torus fit
into this framework.

1 Introduction
The main motivation for this research is to provide a theoretical framework
for construction of C 1 -like algorithm for a class of dissipative PDEs. By a
C 1 algorithm we understand computation that provides guaranteed bounds on
forward trajectories of underlying semi-flow and bound on the solutions to the
associated variational equations.
The method of self-consistent bounds introduced in [ZM] lead to design of
rigorous integrators for a class of dissipative PDEs [C, WZ, Z04, Z10], which
proved to be very efficient in studying dynamics of evolutionary dissipative
PDEs [BKZ, WZ, CW, CZ, Z, Z1, Z04, Z10].
The aim of this paper is threefold. First, we revise the method of self-
consistent bounds and reformulate it in an abstract setting. The statements
are proved under very general assumptions about norms used, sets on which
the existence of local semi-flow is proved and, the most important, assump-
tions on the vector field that guarantee uniform convergence of the solutions of
finite-dimensional approximations (Galerkin projections) to the solutions of the
original infinite-dimensional system. We will later refer to these assumptions by
the letters N for norms, S for sets and C for convergence conditions.
The second aim is to formulate and prove results about convergence of the
solutions to associated variational equations for Galerkin projections to the
solutions of (properly understood) variational equations of infinite-dimensional
1 Work of D.W. and P.Z. was supported by National Science Center (NCN) of Poland under

project No. UMO-2016/22/A/ST1/00077

1
systems. Here we will need additional convergence conditions for variational
equations, we refer to as VC.
Finally we will show, that convergence conditions C and VC are not very
restrictive, and they are naturally satisfied in a class of vector fields includ-
ing widely studied models such as Kuramoto-Sivashinsky [KT, S], Burgers [B],
Brusselator (reaction-diffusion system) and Navier-Stokes [T] equations. The
crucial fact which make the proposed framework applicable is so-called isola-
tion property of the vector field, which informally means that the vector field
is pointing inwards for all high modes leading to dissipative character of the
system.
To illustrate the isolation property we consider a class of PDEs of the fol-
lowing form
ut = Lu + N (u, Du, . . . , Dr u) , (1)
d
where u ∈ Rn , x ∈ Td = (R mod 2π) , L is a linear operator, N is a polynomial
and by Ds u we denote sth order derivative of u, i.e. the collection of all spatial
(i.e. with respect to variable x) partial derivatives of u of order s.
We require, that the operator L is diagonal in the Fourier basis {eikx }k∈Zd ,

Leikx = −λk eikx ,

with

L∗ |k|p ≤ λk ≤ L∗ |k|p , for all |k| > K and K, L∗ , L∗ ≥ 0,

p > r.

The assumption p > r makes the equation dissipative, we have a smoothing

effect.
If the solutions are sufficiently smooth, the problem (1) can be written as
an infinite ladder
P of ordinary differential equations for the Fourier coefficients
in u(t, x) = k∈Zd uk (t)eikx , as follows
duk

= Fk (u) = −λk uk + Nk {uj }j∈Zd , for all k ∈ Zd . (2)
dt
The crucial fact, which makes our approach possible is the isolation property,
which will be defined in Section 6. Here, informally we express it as follows.
Let  
C
W = {uk }k∈Zd : |uk | ≤
|k|s q |k|
where q ≥ 1, C > 0, s ≥ 0, satisfy some inequalities
Then under some assumptions there exists K > 0, such that for |k| > K there
holds
C
if u ∈ W, |uk | = , then uk · Fk (u) < 0.
|k|s q |k|
When q = 1 and s ≥ p + d + 1 this is established in Lemma 31 in Section 7.8
(see also Theorem 3.1 in [Z10]). In [WZ] we used sets of the type described
above with s = 0 and q > 1.

2
 Projection of the set W onto k th mode is a closed disc (or interval) centred
at zero and of radius rk = |k|sCq|k| . Geometrically the isolation property means
that if |uk | = rk for some k > K then the k-th component of the vector field is
pointing inwards the set. As a consequence, the only way a trajectory may leave
the set W forward in time is by increasing some of leading modes |uk |, k ≤ K
above the threshold rk . However, one has to be careful with what ”inwards”
means in the above statement because W has empty interior as a compact
subset in infinite dimensional space. On the other side it makes perfect sense
for Galerkin projections.
This property is used in our approach to obtain a priori bounds for uk (h)
for small h > 0 and |k| > K, while the leading modes uk for |k| ≤ K are
computed using tools for rigorous integration of ODEs [CAPD, Lo, NJP] based
on the interval arithmetics [Mo]. Moreover, from the point of view of topological
method, the isolation property is of crucial importance as it shows that on the
tail we have the entry behaviour, which enable us to apply the finite dimensional
tools from the dynamics like the Conley index [ZM], the fixed point index and
covering relations [Z10].
In the present work we focus on the issues related to the convergence of
Galerkin projections of equations (2) and its variational equations.
The paper is organized as follows. In Section 2 we introduce the notion
of good sequence space (GSS space) and establish some of their properties.
GSS will be later the spaces in which our Fourier series belong. In Section 3
we recall the notion of the logarithmic norm of a matrix and prove a useful
result regarding block decomposition for nonautonomous linear ODEs. This
is a crucial result for efficient implementation of an algorithm based on the
framework introduced in Section 4 and Section 5. There we state and prove the
main results about the convergence of solutions of finite-dimensional Galerkin
projections to solutions of underlying PDE for the main system (Theorem 11)
and for variational system (Theorem 14), respectively. In Section 6 we formally
define the isolation property and we prove that it implies the existence of a-priori
bounds, which appear in the assumptions the convergence theorems. Finally, in
Section 7 we show that the introduced framework applies to a class of dissipative
DPEs, that is the assumptions on sets S, convergence conditions C, VC and
the isolation property are satisfied for various choices of Banach spaces.

Notation
By Z+ we denote the set of all positive integers. Given two normed vector
spaces V, W by Lin(V, W ) we will denote the space of all bounded linear maps
from V to W .

2 Sequence spaces with good norms

The goal of this section is to define the sequence spaces on which we will work
later in this paper. We will be dealing with sequence spaces H ⊂ RZ+ equipped

3
with some norm k · k. Put ei = (δij )j∈Z+ . By e∗j ∈ H ∗ we denote the dual form,
that is e∗j (ei ) = δij . P
By πk we denote the projection onto k-th direction, i.e. πk ( wj ej ) = wk ek .
For a nonempty set J ⊂ Z+ by PJ we denote a projection defined by PJ (w) =
P
i∈J wi ei .
For n ∈ Z+ by Hn we denote a subspace spanned by {ej }j≤n . Put Pn :=
P{j≤n} and Qn = Id − Pn . By ιn : Hn → H we denote the embedding Hn into
H.
We define our standing assumptions N1–N5 on the space (H, k · k).

Definition 1 We will say that H ⊂ RZ+ with norm k · k is GSS (good sequence
space) if the following conditions are satisfied.
N1 H is a Banach space.
P
N2 ∀w ∈ H w = i wi ei .
P
N3 For all w ∈ H and for any α ∈ {−1, 1}Z+ there holds wα = i αi wi ei ∈ H
and kwk = kwα k.
N4 kPJ wk ≤ kwk, ∀w ∈ H, ∀J ⊂ Z+ .
N5 there exists constant G, such that for all w ∈ H and for all i |wi | ≤ Gkwk.

Examples: l2 , l1 , c0 (sequences converging to 0) with the norm k · k∞ .

However l∞ is not P
a GSS space, because N2 is not satisfied. The space defined
by convergence of i |wi |/2i does not satisfy N5.

Lemma 1 If (H, k · k) is GSS, then for any J ⊂ Z+ the projection PJ is con-

tinuous and for any w ∈ H there holds

kwk = sup kPn wk, (3)

n∈Z+

k(I − Pn )wk → 0, for n → ∞. (4)

Proof: Continuity of PJ follows immediately from N4. From N2 and N4 we

have
kwk = lim kPn wn k ≤ sup kPn wk ≤ kwk,
n→∞ n

which proves (3). From N4 the sequence kPn wk is non-decreasing, hence

limn→∞ kPn wk = supn kPn wk = kwk. In consequence,

k(I − Pn )wk ≤ kwk − kPn wk → 0.

Lemma 2 Assume that H is GSS. Then W ⊂ H is compact, iff it is closed,

bounded and W has uniform bounds on the tail, i.e. for every ǫ > 0 there exists
N , such that for all n ≥ N and for all w ∈ W there holds k(I − Pn )wk < ǫ.

4
Proof: Implication ⇒. Since W is compact, it is also bounded and closed. To
prove the existence of uniform bound for the tail let us fix ǫ > 0 and define an
open covering {Un }n∈Z+ of W by setting Un = {w ∈ H : k(I − Pn )wk < ǫ}. It
is indeed a covering due to (4) in Lemma 1. Observe that Un ⊂ Un+M for any
n, M ∈ Z+ (this a consequence of assumption N4). From compactness of W it
follows that there exists N , such that W ⊂ UN ⊂ UN +M for any M ∈ Z+ .
Implication ⇐. Fix a sequence ck ∈ W . We would like to prove that it
has a convergent subsequence. From the boundedness of W and the diagonal
argument it follows that we can find a subsequence of dm = ckm , such that for
each m ∈ Z+ the sequence (dm ¯
n )m∈Z+ is convergent to some dn . Therefore, for
m
each n the sequence (Pn d )m∈Z+ satisfies the Cauchy condition. We will show
that dm also satisfies the Cauchy condition.
Let us fix ǫ > 0. For j ∈ Z+ we have
kdm − dm+j k ≤ kPn dm − Pn dm+j k + k(I − Pn )dm k + k(I − Pn )dm+j k.
From the uniform bound on tail in W we find N such that k(I − PN )wk < ǫ
for w ∈ W , in particular for all dm . Given fixed N from the Cauchy condition
for the sequence (PN dm )m∈Z+ it follows that there exists M , such that for all
m ≥ M and all j ∈ Z+ there holds kPN dm − PN dm+j k < ǫ.
Hence we obtain for m ≥ M and all j ∈ Z+
kdm − dm+j k ≤ 3ǫ.
Therefore the sequence dm satisfies the Cauchy condition. Since H is complete
(assumption N1) there exists d ∈ H, such that dm → d. Since W is closed, we
conclude that d ∈ W .
From the proof of the above lemma it is easy to infer the following result.
Lemma 3 Assume that H is GSS and W ⊂ H is compact. Then, a sequence
ck in W is convergent to c ∈ W , iff for all j ∈ Z+ limk→∞ ckj = cj .

2.1 Linear forms on GSS

The following lemma gives a characterization of continuous linear forms on GSS.
Lemma 4 Assume that H is GSS and let f : H → R be bounded linear func-
tional. Put fj = f (ej ). Then, for every a ∈ H there holds
X
|fj | · |aj | ≤ kf k · kak and (5)
j∈Z+
X
f (a) = f j aj . (6)
j∈Z+

Proof: Put K = kf k. Let ā ∈ H be such that |aj | = |āj | and fj āj ≥ 0 for all j.
From N3 we have ā ∈ H and kak = kāk. Since each term fj āj is nonnegative,
using N4 we obtain
X X
Kkak = Kkāk ≥ KkPn āk ≥ f (Pn ā) = fj · āj = |fj | · |aj |
j≤n j≤n

5
hence we obtained (5).
Equation (6) is a consequence of the continuity of f and the assumption N2.

2.2 Bounded linear self-maps on GSS

Theorem 5 Let H be GSS space. Assume that V : H → H is bounded linear
map. Then there exists a collection of real numbers {Vij }i,j∈Z2+ , such that for
each a ∈ H there holds
X
|Vij | · |aj | ≤ kπi V k · kak, i ∈ Z+
j
 
X X
Va =  Vij aj  ei .
i j

Proof: We apply N2 and Lemma 4 to each πi V , which are bounded linear

forms.
In view of the above theorem we will often represent V ∈ Lin(H, H) as a
matrix V = {Vij } ∈ RZ+ ×Z+ and by V∗j we will denote its j-th column.

2.3 Galerkin projection of linear maps on GSS

We will be interested in the following question: given linear maps V n : Hn → Hn
(for example the Jacobian matrices for the flow defined by Galerkin projections),
does there exists a limit V : H → H, under some assumptions, which are
reasonable in the context of dissipative PDEs. To make sense of this question
we treat map V n : Hn → Hn as a linear map from H to H by setting V n (u) =
ιn (V n (Pn u)) ∈ Hn ⊂ H.
Observe that in view of Theorem 5 we can represent in basis {ej } each
continuous linear map V : H → H by its matrix {Vij }i,j∈Z+ . We also set
V∗j = V ej , which can be seen as j-th column of matrix Vij representing V .

Theorem 6 Let V n : H → H, n ∈ Z+ be linear maps such that kV n k ≤ K for

some K ≥ 0. Assume that
• for all i, j ∈ Z+ there exists Vij , such that

lim Vijn = Vij (7)

n→∞

• there exists a family of compact sets W j ⊂ H, j ∈ Z+ such that for all n

there holds
n
V∗j ∈ W j. (8)

Then there exists a bounded linear map V : H → H, such that

6
 1. for all a ∈ H there holds limn→∞ V n (a) = V a,
2. kV k ≤ K,
3. πi V (ej ) = Vij ei , ∀i, j ∈ Z+ and
P∞
4. for each i and all a ∈ H there holds j=1 |Vij | · |aj | < ∞.

Proof:
Let us fix a ∈ H. We will show that V n a is a Cauchy sequence. For any
n, k, m, M ∈ Z+ we have

kV n+k a − V n ak
≤ kV n+k (Pm a) − V n (Pm a)k + kV n+k (I − Pm )ak + kV n (I − Pm )ak
≤ kPM V n+k (Pm a) − PM V n (Pm a)k + k(I − PM )V n+k (Pm a)k
+ k(I − PM )V n (Pm a)k + kV n+k (I − Pm )ak + kV n (I − Pm )ak
M
X X
≤ kei k |Vijn+k − Vijn | · |aj | + k(I − PM )V n+k (Pm a)k
i=1 j≤m

+ k(I − PM )V n (Pm a)k + 2Kk(I − Pm )ak.

Let us fix ǫ > 0. Then by taking m big enough by Lemma 1 (condition (4)) we

obtain that 2Kk(I − Pm )ak ≤ ǫ. Let us fix such value of m.
Now we look at k(I − PM )V ℓ (Pm a)k for any ℓ. Observe that V ℓ (Pm a) is
linear combination of first m columns in V ℓ . Indeed,
X
V ℓ Pm a = V∗jℓ
aj .
j≤m

From this and assumption (8) it follows that for any ℓ there holds
 
X
(I − PM )V ℓ Pm a ⊂ (I − PM )  W j aj 
j≤m

P
The set j≤mW j aj is compact, hence from Lemma 2 it follows there exists M ,
P 
j
such that k(I − PM ) j≤m W aj k ≤ ǫ/2. We fix such value of M .
P P
From condition (7) it follows that M i=1 kei k
n+k
j≤m |Vij − Vijn | · |aj | ≤ ǫ for
n large enough and any k.
Summarizing, we obtained that kV n+k a − V n ak ≤ 3ǫ for n large enough and
therefore V n a converges to some V a, which defines a linear operator V : H → H
satisfying kV k ≤ K.
Since πi V n ej = ei Vijn , we see that Vij = limn→∞ Vijn .
P
Finally, from Lemma 4 applied to πi V it follows that ∞ j=1 |Vij | · |aj | < ∞.

7
2.4 Block decomposition
Fix M > 1 and let J1 , . . . , Jm be a partition of the set {1, . . . , M }. We set
Jm+1 = {M + 1, M + 2, . . . }. For ℓ = 1, . . . , m + 1 we set H<ℓ> = PJℓ H. This
defines a decomposition M
H= Hhℓi . (9)
ℓ≤m+1
P
For x ∈ H we have x = ℓ≤m+1 xhℓi , where xhℓi = PJℓ x.
For an operator A ∈ Lin(H, H) we define its blocks according to the above
decomposition by setting

Ahk1 ihk2 i x = PJk1 Ax, x ∈ Hhk2 i .

With this notation we have

m+1
X
(Ax)hki = Ahkihℓi xhℓi .
ℓ=1

Theorem 7 Let (H, k · k) be GSS and fix a block decomposition of H, as in (9).

Assume that
• V n : Hn → Hn , n > M is a family of linear maps, such that for all
i, j ∈ Z+ there exist a finite limit Vij := limn→∞ Vijn ,

• there exists a family W j ⊂ H, j ∈ Z+ of compact sets, such that for

n
n > M there holds V∗j ∈ W j,

• there exists B ∈ R(m+1)×(m+1) , such that

n
kVhkihℓi k ≤ Bkℓ , k, ℓ ≤ m + 1, n > M. (10)

Then there exists a bounded linear operator V : H → H such that

1. for all a ∈ H there holds limn→∞ V n (a) = V a,
P∞
2. for each i and all a ∈ H there holds j=1 |Vij | · |aj | < ∞,
P 
∞
3. πi V (a) = j=1 Vij a j ei and

4. we have the following bounds

kVhkihℓi k ≤ Bkℓ , k, ℓ ≤ m + 1. (11)

Proof: From (10) and condition N4 we have

 
m+1
X m+1
X
kV n ak ≤ n
kVhkihℓi k · kahℓi k ≤  Bkℓ  kak
k,ℓ=1 k,ℓ=1

8
for a ∈ H. This proves that kV n k are uniformly bounded. From Theorem 6 it
n
follows that for each a ∈ H the sequence
P∞V a converges to V a, V : H → H is
continuous and for each i there holds j=1 |Vij | · |aj | < ∞.
There remains to prove (11). From (10) we have
n
kVhkihℓi k = sup kVhkihℓi ahℓi k ≤ sup sup kVhkihℓi aℓ k ≤ Bkℓ ,
kaℓ k=1 n>M kaℓ k=1

which completes the proof.

3 Logaritmic norms and estimates for linear

equations
The goal of this section is to provide estimates for solutions to non-autonomous
linear ODEs, when only bounds on the time-dependent coefficients are known.
Later, in Section 5, we will use these results to provide uniform bounds on the
solutions to all Galerkin projections of variational equations.
The content of this subsection is an extension of the results from [KZ]. Let
k · k denote a norm on Rn and by the same symbol we will denote operator norm
induced on Rn×n .
Definition 2 [D58, L58] The logarithmic norm of an operator A is defined by

kId + hAk − 1
µ(A) = lim+ .
h→0 h
For the properties and usage of the logarithmic norm see [KZ] and the literature
given there.

Definition 3 Let f : U → Rn , where U ⊂ R is open. An absolutely continuous

function x : [a, b] → Rn is a weak solution of

x′ = f (x), x ∈ Rn

if for some t0 ∈ [a, b] and all t ∈ [a, b] there holds

Z t
x(t) = x0 + f (x(s))ds.
t0

Consider a linear equation

v ′ (t) = A(t) · v(t) + b(t), (12)

where v(t) ∈ Rk , A(t) ∈ Rk×k , b(t) ∈ Rk , A and b are bounded and measurable.

9
LnAssume that we have a decomposition of the phase space Rk of the form Rk =
ki k
i=1 R , ki ≥ 1 and accordingly for z ∈ R we will write z = (zh1i , . . . , zhni ),
where zhii ∈ Rki . Using this notion, equation (12) can be written as
n
X
′
zhii (t) = Ahiihji (t)zhji (t) + bhii (t), i = 1, . . . , n,
j=1

where zhii (t), bhii (t) ∈ Rki , i = 1, . . . , n and Ahiihji (t) ∈ L(Rki , Rkj ), i, j =
1, . . . , n are linear maps. To each block Ahiihji of the matrix A we will assign a
real number Jij and collect them in an n × n matrix J. The quantity Jij will
′
estimate the influence of zhji on zhii . The details will be given in the sequel.
The following lemma plays a crucial role in our considerations.
Ln
Lemma 8 [KZ, Lemma 4.1] Let z : [0, T ] → Rk = i=1 Rki be a weak solution
of the equation
z ′ (t) = A(t) · z(t) + b(t),
where A : [0, T ] → Rk×k and b : [0, T ] → Rk are bounded and measurable.
Assume that J : [0, T ] → Rn×n and c : [0, T ] → Rn are measurable and for
all t ∈ [0, T ] there holds
(
kAhiihji (t)k for i 6= j,
Jij (t) ≥
µ(Ahiihii (t)) for i = j,
ci (t) ≥ kbhii (t)k for i = 1, . . . , n.

Then, for i = 1, . . . , n and t ∈ [0, T ] there holds

kzhii k(t) ≤ yi (t),

where y : [0, T ] → Rn is a weak solution of

y ′ (t) = J(t)y(t) + c(t)

with an initial condition y(0) satisfying

kzhii (0)k ≤ yi (0), i = 1, . . . , n.

Next lemma gives bounds for fundamental matrix for a non-autonomous

linear equation.
Ln
Lemma 9 Fix a decomposition Rk = i=1 Rki . Let W : [0, T ] → Rk×k be a
weak solution of the equation

W ′ (t) = A(t) · W (t),

where A : [0, T ] → Rk×k is bounded and measurable.

10
 Assume J : [0, T ] → Rn×n is a measurable matrix function satisfying the
following inequalities for all t ∈ [0, T ]
(
kAhiihji (t)k for i 6= j,
Jij (t) ≥
µ(Ahiihii (t)) for i = j.

Then
kWhiihji (t)k ≤ Bij (t), t ∈ [0, T ], i, j = 1, . . . , n,
n×n
where B : [0, T ] → R is a weak solution of the equation
B ′ (t) = J(t)B(t),
with an initial condition B(0) satisfying
kWhiihji (0)k ≤ Bij (0), i, j = 1, . . . , n.

Proof: For any z 0 ∈ Rk the function z(t) given by

z(t) = W (t)z 0
is a weak solution of equation z ′ = A(t)z(t) with the initial condition z(0) =
W (0)z 0 . Let us fix j ∈ {1, . . . , n}. Then, for any z 0 such that kzhii
0
k = δij and
for all i = 1, . . . , n we have
n
X
kzhii (0)k = k(W (0)z 0 )hii k ≤ 0
kWhiihji (0)k · kzhji k ≤ Bij (0).
j=1

′
Take y(t) solving equation y = Jy with the initial condition y(0) = B∗j (0).
From Lemma 8 with b(t) ≡ 0 it follows that
0
kWhiihji (t)zhji k = k(W (t)z 0 )hii k = kzhii (t)k ≤ yi (t) = Bij (t).
0
Since kzhji k = 1 is arbitrary, we obtain kWhiihji (t)k ≤ Bij (t).
Consider the differential equation
x′ = f (x), f ∈ C 1. (13)
Let ϕ(t, x0 ) denote the solution of equation (13) with the initial condition x(0) =
x0 . By kxk we denote a fixed arbitrary norm in Rn .
From Lemma 8 one can easily derive the following result.
Lemma 10 Let y : [0, T ] → Rn be a piecewise C 1 function and ϕ(·, x0 ) be
defined for t ∈ [0, T ]. Suppose that Z is a convex set such that the following
estimates hold:
y([0, T ]), ϕ([0, T ], x0 ) ∈ Z
 
∂f
µ (η) ≤ l, for η ∈ Z
∂x
dy
(t) − f (y(t)) ≤ δ.
dt

11
Then for 0 ≤ t ≤ T there is
kϕ(t, x0 ) − y(t)k ≤ elt ky(0) − x0 k + δκl (t),
where (
elt −1
l if l 6= 0,
κl (t) = (14)
t if l = 0.

4 C 0-convergence
The aim of this section is to state and prove theorem about uniform conver-
gence of local flows induced by Galerkin projections to the local semi-flow for
the original infinite-dimensional system. The assumptions, called later the C 0 -
convergence conditions, will be about local properties of the vector field, only.
We will assume, that the vector field satisfies certain condition on some sets
(rough enclosures) satisfying the following geometric properties.
In the context of H being GSS, we define a special set of Galerkin projections
as
S follows. Let ∅ 6= J1 ( J2 ( J3 . . . be a family of finite sets, such that
n∈Z+ J n = Z+ . For n ∈ Z+ by Hn we denote a subspace spanned by {ej }j∈Jn .
Put Pn := PJn and Qn = Id − Pn . By ιn : Hn → H we denote the embedding
Hn into H and we define πk x = Pk x − Pk−1 x, i.e. πk is the projection onto
space spanned by {ej }j∈Jk \Jk−1 . Observe that we abuse a bit the notation here,
because in Section 2 Pn , Hn , πn and ιn had slightly different meaning, however
the results proven there apply also in the present context, it is enough to replace
n-th direction spanned by en by a finite dimensional space PJn \Jn−1 H which is
spanned by {ej }j∈Jn \Jn−1 . We will call family {Jn }n∈Z+ a Galerkin filtration of
H. The trivial Galerkin filtration of H is defined by Jn = {j, j ≤ n}.
Definition 4 Assume that (H, k·k) is GSS space. We say that W ⊂ H satisfies
condition S if
S1: W is convex and there exists M ≥ 1, such that Pn (W ) ⊂ W for n ≥ M ,
S2: W is compact.
For a vector field F : dom (F ) ⊂ H → H we define its n-th Galerkin projec-
tion F n : Hn → Hn by
u′ = F n (u) := Pn (F (in (u))). (15)
Definition 5 We say that F is admissible, if for all n ∈ Z+ the function F n
is defined on Hn and it is C 3 as a function on Hn .
Now we are in the position to present our C 0 convergence conditions for
infinite-dimensional vector fields.
Definition 6 Let (H, k · k) be GSS, W ⊂ H and F : dom(F ) ⊂ H → H. We
say that F satisfies condition C on W if F is admissible, W satisfies condition
S and

12
C1: W ⊂ dom (F ) and function F |W : W → H is continuous;
C2: there exists l ∈ R such that for all n ∈ Z+ there holds

sup µ (DF n (x)) ≤ l.

x∈Pn W

The main idea behind condition C2 is to ensure that the logarithmic norms (see
Section 3) for all Galerkin projections are uniformly bounded.
Observe that conditions S and C1 imply that F ◦ Pn converges uniformly to
F on W .
The following theorem is a generalization of Theorem 13 in [Z]. There it is
assumed that W is a trapping region and H = l2 was used, but the main idea
of the proof is the same.
Theorem 11 Let (H, k · k) be GSS and assume that F : dom(F ) ⊂ H → H
satisfies condition C on W ⊂ H. Let Z ⊂ W and T > 0 be such that for all
n > M1 there holds

ϕn (t, x) ∈ W for x ∈ Pn (Z), t ∈ [0, T ], (16)

where ϕn is a local dynamical system on Hn induced by the n-th Galerkin projec-

tion (15). Then there exists a continuous function ϕ : [0, T ] × Z → W satisfying
the following properties.
bn := ιn ◦ϕn ◦Pn converge uniformly
1. Uniform convergence: The functions ϕ
to ϕ on [0, T ] × Z.
2. Existence and uniqueness within W : For all x ∈ Z the function u(t) :=
ϕ(t, x) is a unique solution to the initial value problem u′ = F (u), u(0) = x
and satisfying u(t) ∈ W for t ∈ [0, T ].
3. Lipschitz constant: For x, y ∈ Z and t ∈ [0, T ] there holds

kϕ(t, x) − ϕ(t, y)k ≤ elt kx − yk.

4. Semidynamical system. The partial map ϕ : [0, T ] × W → W defines a

semidynamical system on W , that is
• ϕ is continuous;
• ϕ(0, u) = u;
• ϕ(t, ϕ(s, u)) = ϕ(t + s, u)
provided ϕ(t + s, u) exists.

Remark 12 Observe that the essential difficulty in application of the above

theorem is to find set W satisfying (16) and the convergence conditions. A
systematic way to construct it is based on the isolation property (see Section 6).
We will prove there that it is always possible to find a-priori bounds and make
one time step of a rigorous integration algorithm.

13
Proof of Theorem 11:
Let us fix k ≥ n and set xn = ϕn (·, u), xk = ϕk (·, v) for some u ∈ Pn Z and
v ∈ Pk Z. We start from estimation on the difference xn (t) − xk (t) for t ∈ [0, T ].
We have

(Pn xk (t))′ = Pn F (xk (t))

= Pn F n (Pn xk (t)) + Pn F (xk (t)) − Pn F (Pn xk (t)) .

Hence we can treat the function y(t) = Pn xk (t) as a solution to a perturbed

equation y ′ (t) = Pn F n (y(t)) + δ(t), where δ is uniformly bounded by

kδ(t)k = kPn F (xk (t)) − Pn F (Pn xk (t))k ≤ max kPn F (y) − Pn F (Pn y)k =: δn .
y∈W

Obviously δn → 0 for n → ∞, because F ◦ Pn converges uniformly to F on W

– this follows immediately from compactness of W (condition S2) and C1.
From C2 there is a uniform bound l on all logarithmic norms of DF n (W ),
n ∈ Z+ . Hence, by Lemma 10 we obtain

kxn (t) − Pn (xk (t))k ≤ elt kxn (0) − Pn xk (0)k + δn κl (t), t ∈ [0, T ] (17)

where κl is defined as in (14).

Convergence. For x ∈ Z, t ∈ [0, T ] and k ≥ n from (17) with u = Pn x and
v = Pk x we have

bn (t, x) − ϕ
kϕ bk (t, x)k ≤
kϕn (t, Pn x) − Pn (ϕk (t, Pk x))k + k(I − Pn )ϕk (t, Pk x)k ≤
δn κl (t) + k(I − Pn )xk (t)k ≤ δn κl (t) + k(I − Pn )W k.

From Lemma 2 we get limn→∞ k(I − Pn )W k = 0. The function κl is bounded

on [0, T ] and given that limn→∞ δn = 0 we see that ϕbn is a Cauchy sequence
in C([0, T ] × Z, H). Hence, it converges uniformly to a continuous function
ϕ : [0, T ] × Z → W .
Existence. The function ϕ bn satisfies the following integral equation for
x∈Z Z t
bn (t, x) = ϕ
ϕ bn (0, x) + bn (s, x))ds.
ιn Pn F (ϕ
0
From the uniform continuity of F on W and already established convergence of
bn (t, x) it follows that after passing to the limit n → ∞ there holds
ϕ
Z t
ϕ(t, x) = ϕ(0, x) + F (ϕ(s, x))ds,
0

which implies that for x ∈ Z the function u(t) = ϕ(t, x) is a solution of u′ = F (u)
with ϕ(0, x) = x.
Uniqueness. Let u : [0, T ] → W be a solution of u′ = F (u) with initial
condition u(0) = x. We will show that the sequence ϕ bn (·, x) converges uniformly

14
to u. Applying Lemma 10 to n-th Galerkin projection and the function Pn u(t)
we obtain an estimate

bn (t, x) − Pn (u(t))k ≤ δn κl (t).

kϕ

Since the tail k(I −Pn )u(t)k ≤ k(I −Pn )W k is by Lemma 2 uniformly converging
to zero as n → ∞, we see that ϕ bn (·, x) → u uniformly on [0, T ].
Lipschitz constant on W . From Lemma 10 applied to n-dimensional
Galerkin projection for different initial conditions we obtain

bn (t, x) − ϕ
kϕ bn (t, y)k = kϕn (t, Pn x) − ϕn (t, Pn y)k ≤ elt kPn x − Pn yk,

which after passing to the limit n → ∞ gives

kϕ(t, x) − ϕ(t, y)k ≤ elt kx − yk.

Assertion 4 follows easily from the previous ones.

5 C 1 -convergence of Galerkin projections

In Section 4 we gave (Theorem 11) sufficient and verifiable by means of rigor-
ous numerics conditions that guarantee uniform convergence of a sequence of
flows induced by Galerkin projections to a solution of the underlying infinite
dimensional system. In this section we would like to address similar question
for associated variational equations. Before we state the assumptions about the
vector field we need to introduce some notion.

5.1 Some remarks about C 1 functions on compact subsets

of GSS spaces
Let f : H ⊃ dom (f ) → R. In the applications we keep in mind, the function f
will be a component of a vector field, which might be not Frechét differentiable
on H. On the other hand, assuming that the restrictions f n := f ◦ in : Hn → R
are C 1 -smooth for all n, the derivative
 Df n (z) can be
 seen as a linear form on
∂f ∂f
Hn represented by a row-vector ∂x 1
(z), . . . , ∂x n
(z) . Formally Df n (z) ∈ Hn∗
for z ∈ Hn . We can embed it to H ∗ by setting Df n (z) ◦ Pn for z ∈ Hn .
 Here a natural question  arises: does the infinite row-vector
∂f ∂f
∂x1 (z), ∂x2 (z), . . . define a 1-form on H? If so, can we treat it as the deriva-
tive of f ? For this purpose we introduce the notion
g
Df (z) := lim Df n (Pn z) ◦ Pn . (18)
n→∞

and prove the following lemma.

Lemma 13 Let H be GSS, W ⊂ H and let f : H ⊃ dom (f ) → R. Assume

that

15
 • W satisfies condition S;
• f |W is continuous;
• f n is C 1 on Hn ;

• for all z ∈ W the limit (18) exists and the function g

Df : W → H ∗ is
continuous.
Then for every z, w ∈ W there holds
Z 1 
f (z) − f (w) = g(tz + (1 − t)w)dt (z − w)
Df
0
X Z 1
= g
D j f (tz + (1 − t)w)dt · (zj − wj ), (19)
j∈Z+ 0

where Dg g
j f (z) := Df (z)(ej ). Moreover, if for some z ∈ W and j there exists
∂f
δ > 0, such that z + [−δ, δ]ej ∈ dom (f ), then ∂x j
(z) exists and

g ∂f
D j f (z) = (z). (20)
∂xj
Proof: Observe that for any v, z ∈ H and n, k ∈ Z+ there holds

Df n+k (Pn+k (Pn z))(Pn+k (Pn v)) = Df n+k (Pn z)(Pn v)

= Df n (Pn z)(Pn v).

Hence
g(Pn z)(Pn v) = Df n (Pn z)(Pn v).
Df (21)
For any n ≥ M and z, w ∈ W using (21) we have
X Z 1 ∂f
f (Pn z) − f (Pn w) = (Pn (tz + (1 − t)w))dt · (zj − wj )
∂xj
j≤n 0
Z 1
= Df n (Pn (tz + (1 − t)w))dt · Pn (z − w)
0
Z 1
= g
Df (Pn (tz + (1 − t)w))dt · Pn (z − w).
0

Now we pass to the limit n → ∞ in the above equation. From continuity of f it

follows that f (Pn z)−f (Pn w) → f (z)−f (w) and from N2 we obtain Pn (z−w) →
(z − w). From continuity of g Df on W , which is compact, follows its uniform
continuity. Moreover, from Lemma 2 we get that Pn converges uniformly to the
R1
identity on W . Therefore the integral converges to 0 Dfg(tz + (1 − t)w)dt and
we obtain
Z 1 
f (z) − f (w) = g
Df (tz + (1 − t)w)dt (z − w).
0

16
 R1
g(tz + (1 − t)w)dt ∈ H ∗ , because it is bounded
Observe that 0 Df
by maxz∈W kgDf (z)k. Therefore from Lemma 4 we obtain assertion (19).
Assume that z ∈ W and z + hej ∈ W . Then from (19) we obtain
Z 1 
f (z + hej ) − f (z) = h g
Df (z + thej )dt (ej )
0
Z 1 Z 1  
=h g
D g
j f (z + thej )dt = hDj f (z) + h Dg g
j f (z + thej ) − Dj f (z) dt.
0 0

g
From the continuity of D j f it follows that the integral converges to 0 if h → 0.
Therefore we obtain (20).

5.2 Variational equations and Galerkin projections

Let ϕn be a local flow induced by (15) and consider the variational matrix for
ϕn given by
∂ϕni
Vijn (t, u) = (t, u).
∂uj
Since F is admissible, each column in Vijn evolves separately and satisfies the
system of variational equations
d n
x = F n (xn ), xn ∈ Hn , (22)
dt
d n
C = DF n (xn )C n , C n ∈ Hn . (23)
dt
The above system has Hn × Hn ⊂ H × H as the phase-space.
Following discussion at the beginning of this section, instead of DFi which
may not exists, we will use the notion DFgi and define a vector field on H × H
as
d gi (x)C}i∈Z+ ).
(x, C) = FV (x, C) := (F (x), {DF (24)
dt
To bring the variational system to the setting discussed in Section 4 we
proceeded as follows. On H × H we use the norm k(x, v)k = max(kxk, kvk).
This makes H ×H a GSS space and its basis is parameterized by Z+ ×Z+ . Given
{Jn } Galerkin filtration of H, we define {Jn′ } Galerkin filtration of H × H by
Jn′ = Jn × Jn . Then accoridingly Galerkin projection Pn on H × H is defined
by Pn (x, C) = (Pn x, Pn C).
For the remainder of this section we adopt notation used in Section 4 and
Theorem 11. In particular we have a Galerkin filtration of H and the induced
Galerkin filtration of H × H, so that Pn will always refer to these filtrations.

Definition 7 Let (H, k · k) be GSS, W × WV ⊂ H × H and F : dom(F ) ⊂

H → H. We say that F satisfies condition VC on W × WV if F is admissible,
W × WV satisfies condition S and

17
VC1: the function FV is continuous on W × WV ;
VC2: there exists a constant A, such that

sup µ(DPn FV |Hn ×Hn ) ≤ A.

(x,v)∈Pn (W ×WV )

The conditions VC1 and VC2 are nothing more than conditions C1 and
C2 for the vector field FV . Applying arguments from Section 4 to FV we
will obtain the existence of solution to variational equation for each column
(directional derivative), independently. Obtaining C 1 -like information about
full system needs some additional reasoning.
The following theorem is a generalization of Theorem 2 in [Z1].
Theorem 14 The same assumptions as in Theorem 11. Let {WVj }j∈Z+ ⊂ H
be a family of sets such that for each j ∈ Z+ F satisfies VC on W ×WVj and for
any n > j the solution of the variational problem (22–23) with initial conditions
xn (0) ∈ Z and C n (0) = ej satisfies

C n (t) ∈ WVj , t ∈ [0, T ]. (25)

Let l be a constant from condition C2 (uniform bound on logarithmic norm of

DF on W ).
Then there exists V : [0, T ] × Z → Lin(H, H), such that Vij : [0, T ] × Z → R
for i, j ∈ Z are continuous and the following properties are satisfied.
Convergence: For each j the function Vb∗j n n
(t, x) := ιn V∗j (t, Pn x) converges to
V∗j (t, x) uniformly on [0, T ] × Z, and

kV (t, x)k ≤ elt , (t, x) ∈ [0, T ] × Z,

X
V (t, x)ej = Vij (t, x)ei ,
i∈Z+

and for every a ∈ H the map [0, T ] × Z ∋ (t, x) 7→ V (t, x)a is continuous.
Smoothness: For any x, y ∈ Z and any t ∈ [0, T ] there holds
Z 1
ϕ(t, x) − ϕ(t, y) = V (t, y + s(x − y))ds · (x − y), (26)
0

∂ϕ
and for every j the partial derivative of the flow ∂xj (t, x) exists and

∂ϕ
(t, x) = V∗j (t, x). (27)
∂xj

Equation for V : V (t, u) satisfies the following variational equation

dV∗j X X ∂Fi
(t, u) = ei (ϕ(t, u))Vkj (t, u), (28)
dt i
∂uk
k

18
 with the initial condition V (0) = Id in the following sense: for each j the
dV
derivative dt∗j (t, u) exists, the series on r.h.s. of (28) converges uniformly
on [0, T ] × Z and equation (28) is satisfied.
Proof: We apply Theorem 11 to the variational system (24) for j-th column
(i.e. with initial condition C(0) = ej ) separately. From VC it follows that all its
assumptions are satisfied for system (24) with ZV = Z × {ej }, WV = W × WVj
and with l from condition VC2. Therefore, there exists a family of continuous
functions V∗j : [0, ∞)×Z → H for j ∈ Z+ , such that for each j the functions Vb∗j n

converge to V∗j uniformly on [0, T ] × Z and (x(t), V∗j (t, x)) satisfies variational
equation (24).
Now, we will prove (26) and (27). Let us fix t ∈ [0, T ] and x, y ∈ Z. For any
n we have
Z 1
ϕn (t, Pn x) − ϕn (t, Pn y) = V n (t, Pn y + s(Pn x − Pn y))ds · Pn (x − y)
0

We will pass to the limit n → ∞ in above equation. The limit of l.h.s. is

settled by Theorem 11.
Now we consider the r.h.s. We will use Theorem 6 to pass to the limit
n → ∞. We already have proved the convergence of Vijn . There remains to
show, that kV n k are uniformly bounded.
From condition VC2 and Lemmas 9 and 10 it follows that for all n ∈ Z+ ,
kV n (t, z)k ≤ elt , ∀n, (t, z) ∈ [0, T ] × Pn Z. (29)
n R1 n
Let V (t, [x, y]) = 0 V (t, y + s(x − y))ds. From (29) we have
n
kV (t, [Pn x, Pn y])k ≤ elt , t ∈ [0, T ].
From convexity of WVj it follows that for all j and for x, y ∈ Z there holds
n
Vb∗j
n
(t, x), ιn V (t, [Pn x, Pn y])∗j ∈ WVj .

Since Vbijn (t, ·) are continuous on Z for all i, j and converge uniformly to Vij (t, ·)
we obtain
Z 1 Z 1
Vbijn (t, y + s(x − y))ds → Vij (t, y + s(x − y))ds.
0 0
R1
Now we use Theorem 6 to conclude that V (t, x) ∈ Lin(H, H), 0 V (t, y + s(x −
y))ds ∈ Lin(H, H) and
Z 1 Z 1
n
lim V (t, y + s(x − y))ds · (x − y) = V (t, y + s(x − y))ds · (x − y).
n→∞ 0 0

Gathering this with

lim ϕn (t, Pn x) − ϕn (t, Pn y) = ϕ(t, x) − ϕ(t, y),
n→∞
lim Pn (x − y) = x−y
n→∞

19
we obtain (26).
Now we will show that from (26) we can conclude (27). Indeed, we have
Z 1
ϕ(t, x + hej ) − ϕ(t, x) = h V (t, x + shej )ds · ej
0
Z 1
= hV (t, x)ej + h (V (t, x + shej )ej − V (t, x)ej ) ds
0
Z 1
= hV∗j (t, x) + h (V∗j (t, x + shej ) − V∗j (t, x)) ds
0

and the result follows from continuity of V∗j (t, ·).

It remains to prove that for every a ∈ H the map [0, T ] × Z ∋ (t, z) 7→
V (t, z)a ∈ H is continuous. We know that for every j the map (t, x) 7→ V∗j (t, z)
is continuous. Let (t1 , z1 ), (t2 , z2 ) ∈ [0, T ] × Z. Then for any n holds

kV (t1 , z1 )a − V (t2 , z2 )ak ≤ kV (t1 , z1 )Pn a − V (t2 , z2 )Pn ak

+kV (t1 , z1 )(I − Pn )ak + kV (t2 , z2 )(I − Pn )ak ≤
X
kV (t1 , z1 )∗j − V (t2 , z2 )∗j k · |aj | + (et1 l + et2 l )k(I − Pn )ak
j∈Jn

For a given ǫ > 0 we take n so large that k(I − Pn )akemax(0,l)T < ǫ/2. We fix
such n. Then from the continuity of V∗j for j ∈ Jn we can make the sum less
that ǫ/2 when (t2 , z2 ) → (t1 , z1 ). This finishes the proof.

5.3 Block decomposition

Now, we would like to adapt Lemma 9 to the variational equation for dissipative
PDEs and its solution obtained as the limit of solutions of variational equation
of Galerkin projections. We will adopt the notation related to a decomposition
of H from Section 2.4.

Theorem 15 Same assumptions and notation as in Theorem 14. Consider a

decomposition of H given by (9).
Assume that matrix J ∈ R(m+1)×(m+1) satisfies
 n
supn>M supw∈W ∂Fhki (w) , for k 6= ℓ, l, ℓ ≤ m+1,
∂u
 ∂F n
hℓi 
Jkℓ ≥
sup hki
n>M supw∈W µ ∂uhki (w) , for k = ℓ ≤ m + 1

Then for any x ∈ Z and t ∈ [0, T ] holds

kVhkihℓi (t, x)k ≤ eJt kℓ , k, ℓ = 1, . . . , m + 1. (30)

Proof: Let us take any n > M . We consider the decomposition of Hn

M
Hn = Hhki ⊕ Yn , Yn = Pn Hhm+1i .
k≤m

20
 From Lemma 9 applied to n-th Galerkin projection with matrix J we obtain
that conditions (30) are satisfied by V n (t, Pn x) for x ∈ Z and t ∈ [0, T ]
Now we want to pass to the limit n → ∞. We know from Theorem 14 that

Vijn (t, Pn x) → Vij (t, x), ∀(i, j) ∈ Z2+ ∀t ∈ [0, T ], x ∈ Z.

The result now follows from Theorem 7.

6 Isolation property
In convergence theorems discussed in Section 4 and Section 5 the crucial as-
sumption was the existence of uniform a-priori bounds W and WV . It turns
out that such sets can be found in algorithmic way using the isolation property
of the vector field discussed in Section 7. The existence of such a-priori bounds
will be later used in construction of an algorithm for rigorous integration of the
flow and associated variational equations.
We adopt a notation used in Sections 4 and Section 5. In particular we have
some Galerkin filtration, which is used to define πk . Thus, for x ∈ H, πk x is
the k-th component of x and it is a vector of dimension #(Jk \ Jk−1 ).
In what follows for W ⊂ H we set Wk = πk (W ). We define the isolation
property in the following way.
Definition 8 Let W ⊂ H be a set satisfying condition S in a GSS space H.
We say that vector field F : H → H satisfies the isolation property on the set
W if F satisfies condition C on W and there exists K0 ∈ N such that

∀k≥K0 ∃Tk >0 ∀t∈(0,Tk ] ∀u∈W (uk ∈ bdWk =⇒ uk + tFk (u) ∈ intWk ) .

Geometrically the above condition means that the vector field F is pointing
inwards each component Wk on all far coordinates of W . In particular, each
component Wk , k ≥ K0 has nonempty interior.
Lemma 16 If F : H → H satisfies the isolation property on W ⊂ H then there
is N0 such that for any Galerkin projection n > N0 there holds

∀N0 ≤k≤n ∃Tk >0 ∀t∈(0,Tk ] ∀u∈Pn W (uk ∈ bdWk =⇒ uk + tFkn (u) ∈ intWk ) .

Proof: From S1 there is M ≥ 1 such that Pn W ⊂ W for n ≥ M . Let K0 be

the constant from the isolation property for F . Put N0 = max{K0 , M }. Now,
let us fix n > N0 , N0 ≤ k ≤ n and u ∈ Pn W such that uk ∈ bdWk . Let Tk be
the same constant as in the isolation property for F and index k.
Since k ≤ n and u ∈ Pn W ⊂ W we have Fkn (u) = Fk (u) and thus

uk + tFkn (u) = uk + tFk (u) ∈ intWk

for t ∈ (0, Tk ].

21
Definition 9 We say that a vector field F : H → H is isolating on the family
of sets W ⊂ 2H if F satisfies the isolation property on each set W ∈ W.

Notice, that the constant K0 for each set in the family W can be different.
The next theorem addresses the question about the existence of a-priori
bounds, as in Theorem 11.

Theorem 17 Assume the vector field F : H → H is isolating on the family of

sets W satisfying the following properties:
• there exists E ∈ W, such that 0 ∈ intEk for all k ∈ Z+ ;
• W is closed with respect to the addition

W1 , W2 ∈ W =⇒ W1 + W2 := {w1 + w2 : w1 ∈ W1 , w2 ∈ W2 } ∈ W.

Then for any Z ∈ W there exists T > 0 such that the assumptions of Theorem 11
are satisfied for the set W = Z + E.

Proof: We have to prove (16) with W = Z + E. From Lemma 16 there is N0

such that for n > N0 and N0 ≤ k ≤ n the vector field F n is pointing inwards
Wk on the boundary of Wk . Thus, for z ∈ Pn Z the only way a trajectory z n (t)
of the n-th Galerkin projection may escape the set Pn W is through one of the
leading coordinate zk , k < N0 .
Since W is compact, F is continuous on W and F n converge uniformly to
F on W we have
sup sup sup kFkn (u)k < ∞.
n>N0 k<N0 u∈W

Hence, there is T > 0 such that for all n > N0 and all k < N0 (finite number of
coefficients) there holds

[0, T ]Fkn (Pn W ) ⊂ intEk .

From this we get

Zk + [0, T ]Fkn (Pn W ) ⊂ intWk , k < N0 ,

which proves that for z ∈ Pn Z the solution of the n-th Galerkin projection z n (t)
exists for t ∈ [0, T ] and z n ([0, T ]) ⊂ Pn W .
In the applications we keep in mind, the family W may consists of sets with
polynomial, exponential or mixed decay of far coefficients, which are clearly
closed with respect to addition and the existence of E with 0 ∈ Ek is also
satisfied.
In a similar way we can address the question regarding the existence of
a-priori bounds for variational system needed in Theorem 14.
Theorem 18 Assume the vector field F : H → H is isolating on the family of
sets W satisfying the following properties:

22
 • there exists E ∈ W, such that 0 ∈ intEk for all k ∈ Z+ ;
• W is closed with respect to the addition

W1 , W2 ∈ W =⇒ W1 + W2 := {w1 + w2 : w1 ∈ W1 , w2 ∈ W2 } ∈ W;

• for all W ∈ W the vector field FV is isolating on W × E and satisfies

condition VC
Then for any Z ∈ W there exists T > 0 such that the assumptions of Theorem 14
are satisfied with W = Z + E and WVj = Cj E, for some Cj > 0, j ∈ Z+ .
Proof: From Theorem 17 there is T > 0 and N0 such that the assumptions
of Theorem 11 are satisfied, that is (16) holds true with W = Z + E.
From Lemma 16 and since FV is isolating on W × E it follows that there
is, possibly larger N1 such that the vector field FVn is pointing inwards W × E
for all n > N1 and N1 < k ≤ n. Take ZV := 21 E. Reasoning as in the proof
of Theorem 17 and shrinking T if necessary we obtain that for all n > N1 and
t ∈ [0, T ] there holds

xn (t) ∈ W for xn (0) ∈ Pn Z,

C n (t) ∈ E for C n (0) ∈ Pn ZV .

Let us fix j ∈ Z+ . Since each component of E is a convex set containing zero

we can find a constant Cj such that ej ∈ Cj ZV . Put WVj = Cj E. Now, due to
linearity of the variational equation, for n > N1 and j ≤ n we have

C n (t) ∈ WVj for C n (0) ∈ Pn (Cj ZV ).

In particular, if C n (0) = ej then C n (t) ∈ WVj for t ∈ [0, T ] and thus (25) is
satisfied.

7 Dissipative PDEs on the torus

The aim of this section is to prove that the framework introduced in Section 4
and Section 5 is applicable to a certain class of PDEs. We start with technical
estimates and then we show that conditions S, C and VC as well as the isolation
property are satisfied for this class on certain families of sets.
Consider
ut = Lu + N (u, Du, . . . , Dr u) , (31)
where u ∈ Rn , x ∈ Td = (R mod 2π)d , L is a linear operator, N is a polynomial
and by Ds u we denote sth order derivative of u, i.e. the collection of all spatial
(i.e. with respect to variable x) partial derivatives of u of order s.
We require, that the operator L is diagonal in the Fourier basis {eikx }k∈Zd ,

Leikx = −λk eikx ,

23
with

L∗ |k|p ≤ λk ≤ L∗ |k|p , for all |k| > K and K, L∗ , L∗ ≥ 0, (32)

p > r. (33)

If a(t, x) is a sufficiently
P regular solution of (31), then we can expand it in
Fourier series a(t, x) = k∈Zd ak (t)eik·x , ak ∈ Cn to obtain an infinite ladder of
ordinary differential equations for the coefficients ak
dak
= −λk ak + Nk (a), k ∈ Zd , (34)
dt
where Nk (a) is k-th Fourier coefficient of function N (a, Da, . . . , Dr a).
Observe that ak ’s might not be independent variables. For example, as-
sumption a(t, x) ∈ Rn forces the following reality condition

a−k = ak . (35)

In such situation we have to consider the subspace defined by condition (35).

This subspace is invariant for all Galerkin projections of (31) onto subspaces
containing both ak and a−k .
Other constraints like oddness or evenness of a(t, x) may cause the change
of set of basic functions to sin(kx) or cos(kx)
P or combinations thereof.
In any case we will have u(t, x) = k∈I ak (t)ek (x), where I is a countable
set and ak ∈ Cn or ak ∈ Rn , hence our sequence space H will be build of real
and imaginary parts of components of ak which after choosing a suitable norm
fits in the framework discussed in previous sections.

7.1 Some examples

7.1.1 Kuramoto-Sivashinsky equation
Let us consider the one-dimensional Kuramoto-Sivashinsky PDE [KT, S], which
is given by
ut = −νuxxxx − uxx + (u2 )x , ν > 0, (36)
where x ∈ R, u(t, x) ∈ R and we impose periodic boundary conditions
In the Fourier basis we obtain the following ladder of ordinary differential
equations for complex coefficients ak
X
ȧk = k 2 (1 − νk 2 )ak + ik ak ak−ℓ , (37)
ℓ∈Z

and the reality constraint (35). We can either treat (37) as equation acting on
sequence space indexed by k ∈ Z and consider invariant subspace defined by
reality condition (35) with Galerkin filtration Jn = {k ∈ Z, |k| ≤ n} or we can
eliminate variable ak for k < 0 and rewrite the convolution term in (37) in terms
of ak with nonnegative k’s.

24
 If for (36) we impose odd and periodic boundary conditions (38)

u(t, x) = −u(t, −x), u(t, x) = u(t, x + 2π), (38)

P
then we can represent u(t, x) = k∈Z+ −2ak (t) sin(kx), where ak ∈ R and
equation (36) becomes [CCP, ZM]
k−1
X X∞
dak
= k 2 (1 − νk 2 )ak − k an ak−n + 2k an an+k , k = 1, 2, 3 . . .
dt n=1 n=1

Observe that conditions (32) and (33) are satisfied for the KS equation.
Namely we have λk ∼ νk 4 , p = 4, r = 1.

7.1.2 Navier Stokes equations on the torus

The general d-dimensional Navier-Stokes system (NSS) is written for d unknown
functions u(t, x) = (u1 (t, x), . . . , ud (t, x)) of d variables x = (x1 , . . . , xd ) and
time t, and the pressure p(t, x).
d
∂uj X ∂uj ∂p
+ uk = ν△uj − + f (j) (39)
∂t ∂xk ∂xj
k=1
d
X ∂uj
div u = =0 (40)
j=1
∂xj

The functions f (j) are the components of the external forcing, ν > 0 is the
viscosity.
Therefore we have λk = ν|k|2 , hence we p = 2 and r = 1.
d
We consider (39),(40) on the torus Td = (R/2π) . This enables us to use
Fourier series. We write
X X
u(t, x) = uk (t)ei(k,x) , p(t, x) = pk (t)ei(k,x) (41)
k∈Zd k∈Zd

Observe that uk (t) ∈ Cd , i.e. they are d-dimensional vectors and pk (t) ∈ C. We
assume that f0 = 0 and u0 = 0.
Then (see [Z] for details) (40) is reduced to

(uk , k) = 0 k ∈ Zd , (42)

and on the space of functions satisfying (40) the pressure disappears and we
obtain the following infinite ladder of differential equations for uk
duk X
= −i (uk1 |k) ⊓k uk−k1 − νk 2 uk + ⊓k fk , (43)
dt
k1

where fk are components of the external forcing, ⊓k denotes the operator of

orthogonal projection onto the (d − 1)-dimensional plane orthogonal to k.

25
 Observe that the subspace defined by incompressibility condition (42) and
reality condition (35) is invariant under (43) and also is invariant for Galerkin
projections of (43), where for all k ∈ Zd all components of uk and u−k are
both included or excluded in the projection. This defines a Galerkin filtration.
Alternatively, using some more specific boundary conditions (see for example
[AK21, BB]), we define a set independent modes {uk }k∈I (maybe with respect to
some other function basis consisting of sin(kx) etc) and rewrite the convolution
term in (43) using only uk , k ∈ I. The formulas will be a bit more complicated,
but essentially still could be seen as some convolutions.

7.2 Preparatory remarks

Consider equation (31). We assume that N is a polynomial and by Ds u we
denote sth order derivative of u, i.e. the collection of all spatial (i.e. with re-
spect to variable x) partial derivatives of u of order s. The reason to consider
polynomial and not more general functions N is technical — we need to com-
pute the Fourier coefficients of N (u, Du, . . . , Dr u). This can be achieved by
taking suitable convolutions of Fourier expansions of u and its spatial partial
derivatives. For analytic N the results are a bit more involved, i.e. we will have
infinite series of convolutions. This still it is manageable but we omit it for the
sake of simplicity.
In what follows all considerations will be done assuming uk represent the
Fourier expansion of u,
X
u(t, x) = uk (t) exp(ik · x), (44)
k∈Zd

and uk are independent variables. The estimates and results obtained under
this assumptions can be easily translated to the case when some constraints
coming from the reality requirement (35), boundary conditions lead us to use
sin(kx), cos(kx) or combinations thereof as the basis for the expansion. In such
situation we have a natural imbedding into the previous case of the Fourier
expansion with respect to exp(ikx) and all estimates are the same up to some
constants.
In the sequel for k ∈ Zd we will use norm |k| which should satisfy
|(k1 , . . . , kd )| ≥ |kj | for j = 1, . . . , d. Moreover, uk can be vectors or complex
numbers. This can be easily reformulated to fit into the framework discussed
in previous section, we will use a countable set of indices (k, ℓ, a), where k ∈ I,
ℓ ∈ {1, . . . , d}, a ∈ {0, 1}, so that uk,ℓ,a is real (a = 0) or imaginary (a = 1) part
of ℓ-th component of vector uk ∈ Cd . The Galerkin filtration could be chosen
as follows
Jn = {(k, ℓ, a), |k| ≤ n, ℓ ∈ {1, . . . , d}, a ∈ {0, 1}}. (45)
After formally inserting the Fourier expansion for u, Du, . . . , Dr u in N () we
obtain a sum expressions of the following type for each monomial in N
X
vk1 · vk2 · · · · · vkl ,
k1 +···+kl =k

26
where each of the variables vkj , j = 1, . . . , l is some Fourier coefficient of one of
the components of u or its partial derivatives of the order less than or equal to
r. This is a formal expansion, the questions of convergence and differentiability
are treated in the following sections.

7.3 Derivatives with respect to Fourier coefficients

The goal of this section it to write compact formulas for derivatives with respect
of Fourier coefficients.
We will use the following notation to denote (partial) derivatives. For a
|α|
function f (x1 , . . . , xn ) and α ∈ Nn we set Dα f (x1 , . . . , xn ) = ∂xα1 ∂x∂α2 ...∂xαn ,
P 1 2 n
where |α| = j αj . For k ∈ Zn we define k α = k1α1 k2α2 · · · · · knαn .
When considering N (u, Du, . . . ) it is convenient to name the arguments of
N by u, Dα u, which we will use below. For example, for N (u, ux ) = uux we will
have ∂N ∂N
∂u (u, ux ) = ux and ∂ux (u, ux ) = u. We will use this convention below.
We have
X ∂N
N (u + h, . . . , Dr (u + h)) = N (u, . . . , Dr u) + (u, . . . , Dr u)Dα h
∂(Dα u)
0≤|α|≤r
X 1 ∂ N 2
+ (u, . . . , Dr u)Dα1 hDα2 h
α1 ,α2 2 ∂(D u)∂(Dα2 u)
α1

0≤|α1 |≤r,
0≤|α2 |≤r

+ O(khk3 + . . . + kDr hk3 )

Therefore we have
  !
∂Nk X ∂N r |α| α
= (u, . . . , D u) i j (46)
∂uj ∂(Dα u) k−j
|α|≤r

and
 
   
∂ 2 Nk  X ∂2N 
=
 α α
r
(u, . . . , D u) i |α1 |+|α2 | α1 α2 
j1 j2  . (47)
∂uj1 ∂uj2  α1 ,α2 ∂(D u)∂(D u)
1 2
k−j1 −j2 
0≤|α1 |≤r,
0≤|α2 |≤r

Formulas (46,47) after formally inserting Fourier expansions define us formal

∂ 2 Nk
power series, which we denote by symbols ∂N∂uj , ∂uj1 ∂uj2 without actually claim
k

that these functions represent partial derivatives. This fact is proven later for
arguments belong to some sets with good convergence properties.

7.4 Estimates for sets with polynomial decay

Throughout this section a polymial N (u, . . . , Dr u) is fixed and various estimates
given below will obviously depend on this polynomial, but this will not be clearly
indicated.

27
 In this subsection our goal is to prove the following bounds on the vector
field induced in Fourier domain and its derivatives on set with polynomial decay.
Theorem 19 Let s > s0 = d + r. If |ak | ≤ C/|k|s , |a0 | ≤ C, then the formal
∂ 2 Nk
series defining Nk , ∂N
∂aj , ∂aj1 ∂aj2 are absolutely convergent and there exists
k

D = D(C, s), D1 = D1 (C, s), D2 = D2 (C, s)

D
|Nk | ≤ , k 6= 0, and |N0 | ≤ D, (48)
|k|s−r
∂Nk D1 |j|r
≤ , k 6= j, and
∂aj |k − j|s−r
∂Nk
≤ D1 |j|r , k = j (49)
∂aj
∂ 2 Nk D2 |j1 |r |j2 |r
≤ , k 6= j1 + j2 , and
∂aj1 ∂aj2 |k − j1 − j2 |s−r
∂ 2 Nk
≤ D2 |j1 |r |j2 |r , k = j1 + j2 . (50)
∂aj1 ∂aj2
Assertion (48) has been proven as Lemma 3.1 in [Z10]. Before the proof of
Theorem 19 we need to establish several short lemmas. Some of them are taken
from Section 3.1 in [Z10], but we include their short proofs for the sake of
completeness.
Lemma 20 [Z10, Lemma 3.2] Let γ > 1. For any a, b ≥ 0 the following
inequality is satisfied
(a + b)γ ≤ 2γ−1 (aγ + bγ ).
Proof: This is an easy consequence of the convexity of function x 7→ xγ for
γ > 1. Namely
 γ  γ 
a+b a + bγ
(a + b)γ = 2γ ≤ 2γ = 2γ−1 (aγ + bγ ).
2 2

The following lemma was proved in [Sa]

Lemma 21 [Z10, Lemma 3.3] Assume that γ > d. Then there exists SQ (d, γ) ∈
R such that for any k ∈ Zd \ {0} holds
X 1 SQ (d, γ)
≤ .
|k1 |γ |k − k1 |γ |k|γ
k1 ∈Zd \{0,k}

Proof: From the triangle inequality and Lemma 20 we have

γ
|i|γ (|k − i| + |k|)
≤
|k − i|γ |k|γ |k − i|γ |k|γ
 
2γ−1 (|k − i|γ + |k|γ ) γ−1 1 1
≤ =2 + .
|k − i|γ |k|γ |k|γ |k − i|γ

28
Hence
X X  
1 2γ−1 1 1
≤ + <
|k|γ |i − k|γ |i|γ |k|γ |i − k|γ
k∈Zd \{0,i} k∈Zd \{0,i}
2γ X 1
.
|i|γ |k|γ
k∈Zd \{0}

Now we want to include also the vectors of zero length in the sum appearing
in Lemma 21. To make expression of some formulas less cumbersome in this
subsection for 0 = {0}d ∈ Zd we redefine its norm by setting |0| = 1.
Lemma 22 [Z10, Lemma 3.4] Assume that γ > d. Then there exists C2 (d, γ) ∈
R such that for any k ∈ Zd holds
X 1 C2 (d, γ)
γ |k |γ
≤ .
d
|k1 | 2 |k|γ
k1 ,k2 ∈Z
k1 +k2 =k

Proof: Consider two cases k = 0 and k 6= 0.

If k = 0, then there exists C(d, e γ) ∈ R such that
X 1 X 1
= 1+ e γ).
= C(d,
|k | γ |k | γ |k | 2γ
d 1 2 d 1
k1 ,k2 ∈Z k1 ∈Z \{0}
k1 +k2 =k

If k 6= 0, then from Lemma 21 it follows that

X 1 2 X 1 SQ (d, γ) + 2
γ |k |γ
= γ
+ γ |k |γ
≤ .
d
|k1 | 2 |k| d
|k1 | 2 |k|γ
k1 ,k2 ∈Z k1 ,k2 ∈Z \{0}
k1 +k2 =k k1 +k2 =k

e γ), SQ (d, γ) + 2).

Hence the assertion holds for C2 (d, γ) = max(C(d,

Lemma 23 [Z10, Lemma 3.5] Assume γ > d. For any n ∈ Z+ , n > 1 there
exists Cn (d, γ) ∈ R such that for any k ∈ Zd holds
X 1 Cn (d, γ)
γ |k |γ · · · · · |k |γ
≤ .
d
P n
|k1 | 2 n |k|γ
k1 ,k2 ,...,kn ∈Z , i=1 ki =k

Proof: By induction. Case n = 2 is contained in Lemma 22. Assume now that

the assertion holds for n. We have
X 1
γ |k |γ · · · · · |k γ
=
d
P n+1
|k1 | 2 n+1 |
k1 ,k2 ,...,kn+1 ∈Z , i=1 ki =k
 
X 1 X 1
 ≤
d
|kn+1 |γ d
Pn |k1 |γ |k2 |γ · · · · · |kn |γ
kn+1 ∈Z k1 ,k2 ,...,kn ∈Z , i=1 ki =k−kn+1
X 1 Cn (d, γ) C2 (d, γ)Cn (d, γ)
· ≤ .
|kn+1 |γ |k − kn+1 | γ |k|γ
kn+1 ∈Zd

29
Proof of Theorem 19: For the proof it is enough to assume that N is a
monomial. After formally inserting the Fourier expansion for u, Du, . . . , Dr u
we obtain the expression of the following type
X
Nk (u) = vk1 · vk2 · · · · · vkl , (51)
k1 +···+kl =k

where each of the variables vki , i = 1, . . . , l is some Fourier coefficient of one the
components of u or its partial derivatives of the order less than or equal to r.
Observe that for the Fourier coefficients of partial derivatives up to order r
we have the following estimates
∂ β1 +···+βl u C C
≤ ≤ . (52)
∂xβ1 1 . . . ∂xβd l |k|s−(β1 +···+βl ) |k|s−r
From conditions (51) and (52), and Lemma 23 we obtain
X Cn C n Cn (d, s − r)
|Nk (u)| ≤ s−r s−r
≤
|k1 | · · · · · |kn | |k|s−r
k1 +···+kn =k

This establishes (48).

To prove (49) we use (46). From previous reasoning applied to polynomials
∂N r
∂(Dα u) (u, . . . , D u) we immediately obtain
 
∂N r D1
(u, . . . , D u) ≤ ,
∂(Dα u) k−j |k − j|r−s
which together with bound |α| ≤ r gives (49).
The proof of (50) is analogous.
Later we will use the following lemma.
Lemma 24 Assume that s > 2r + d, then there exists C(d, s, r) such that
X |k|r
s−r
≤ C(d, s, r)(1 + |i|r ).
d
|i − k|
k∈Z ,k6=i

Proof:
X |k|r X (|i − k| + |i|)r
≤
|i − k|s−r |i − k|s−r
k∈Zd ,k6=i k∈Zd ,k6=i
X 2r−1 |i − k|r + 2r−1 |i|r
≤
|i − k|s−r
k∈Zd ,k6=i
 
X 1 X |i|r
= 2r−1  +  ≤ C(d, s, r)(1 + |i|r ).
|i − k|s−2r |i − k|s−r
k∈Zd ,k6=i d
k∈Z ,k6=i

Observe that the two infinite sums are convergent under assumption s > 2r + d.

30
7.5 Conditions S1 and S2
We will be interested in the following candidates for space H:
• c0 , kxk = supk∈Zd |xk |,
P

p 1/p
• lp , kxk = k∈Zd |xk | with p ≥ 1,
Pm P

p 1/p
• W m,p , kxk = i=0
ip
k∈Zd |k| |xk | with p ≥ 1 and m ≥ 0.
It is immediate that the space listed above are GSS spaces.
We consider sets of the form
 
C
WP (C, s) = |xk | ≤ , k 6= 0, |x0 | ≤ C , s > 0,
|k|s
 
S
Wexp (q, S) = |xk | ≤ |k| , q > 1.
q
Clearly they satisfy condition S1.
Remark 25 Observe, that if Wexp (q, S) ⊂ WP (C, s) ⊂ H then it is a closed
subset and thus it inherits S2, C1 and C2 from WP (C, s). Thus, in what follows
we will focus on sets with polynomial decay, only.
We will show that condition S1 on sets Wexp (q, S), WP (C, s) when consid-
ered in suitable spaces.
Theorem 26 The following statements hold true.
• WP (C, s) is a compact subset of c0 for any d and s > 1.
• WP (C, s) is a compact subset of lp and W m,p provided (s − m)p > d (with
m = 0 for space lp ).
• Wexp (q, S) is a compact subset of c0 , lp , W m,p for any d, m, p and q > 1.
Proof: We use Lemma 2 as a criterion for compactness. It is immediate that
the sets WP (C, s) and Wexp (q, S) are closed in all spaces considered (if contained
in them).
The case of c0 space is obvious. In other cases it is enough that to observe
that Z ∞
X
|k|a |xk |b ∼ rd−1 ra x(r)b dr,
|k|≥n n

where x(r) = rCs in case od WP and x(r) = Sq −r in the case od Wexp . From
this we obtain our assertion for Wexp for all d, p and m.
In the case of WP the largest exponent under integral must be less than −1
to obtain the convergence, which gives
d − 1 + mp − sp < −1.

31
7.6 Conditions C1,C2
7.6.1 Condition C1 for set with polynomial decay
Theorem 27 Consider (34). Assume that conditions (32) and (33) hold.
Let W = WP (C, s). Let H be one of space listed at the beginning of Sec-
tion 7.5.
Then
• if s > p + d, then W ⊂ c0 and F : W → c0 is continuous
• if (s − p − ℓ)q > d, then W ⊂ W ℓ,q and F : W → W ℓ,q is continuous (with
ℓ = 0 for space lq )

Proof:
The first question is whether W ⊂ dom F . Consider u ∈ W . From The-
orem 19 and condition (32) it follows that Fk (u) is defined and for |k| > K
holds
D2
|Fk (u)| ≤ L∗ C|k|p−s + D|k|r−s ≤ s−p . (53)
|k|
for some constants D and D2 . From Theorem 26 and our assumptions it follows
W is compact in various spaces listed in the assertion and F (W ) is also in the
listed space and is contained in some compact space (due to Lemma 2 and decay
estimates (53)). Moreover for all u ∈ W

lim Pn F (u) = F (u).

n→∞

uniformly on W .
Hence to prove the continuity of F : W → H it is enough to prove that
Fk : W → Hk is continuous.
Let us fix index k and assume un , u∗ ∈ W , for n ∈ N and un → u∗ for
n → ∞. We have (compare the proof of Theorem 19)
X
Fk (u) = λk uk + Nk (u) = λk uk + Nk,i (u),
i∈J

where J is some set of multindices and for each i ∈ J, Nk,i is monomial depend-
ing on the finite number of ul , i.e.

Nk,i = auk1 · uk2 · . . . ukl , for some a ∈ C and k1 + · · · + kl = k

The term λk uk is continuous, hence it is enough to consider Nk , only. Let

us fix ǫ > 0. From Theorem 19 it follows that there exists a finite set S ⊂ J,
such that X
|Nk,i (u)| < ǫ/3, for all u ∈ W . (54)
i∈J\S

32
There exists L, such that for all iP∈ S monomials Nk,i (u) depend in fact on
the variables ul for |l| ≤ L, hence i∈S Nk,i (u) is continuous on W . Therefore
there exists n0 , such that
X X
Nk,i (un ) − Nk,i (u∗ ) < ǫ/3. (55)
i∈S i∈S

From (55) and (54) we obtain for n > n0

X X
|Nk (un ) − Nk (u∗ )| ≤ Nk,i (un ) − Nk,i (u∗ ) +
i∈S i∈S
X X
n
|Nk,i (u )| + |Nk,i (u∗ )| < ǫ.
i∈J\S i∈J\S

n ∗
Hence limn→∞ Nk (u ) = Nk (u ).

7.6.2 Condition C2 for sets with polynomial decay

We will treat only the spaces l2 , c0 and l1 because for those spaces we have nice
and relatively compact formulas for logarithmic norms.
Namely, we have for A ∈ Rn×n
µ2 (A) = max(λ ∈ Sp(A + At )/2),
 
X
µ1 (A) = max aii + |aik | ,
i=1,...,n
k,k6=i
 
X
µ∞ (A) = max akk + |aik | .
k=1,...,n
i,k6=i

Lemma 28 The same assumptions and notation as in Lemma 27, but we re-
strict H to one of the following spaces c0 , l2 , l1 . Assume additionally that
s > 2r + d. Then F satisfies condition C2 on W .
Proof: First we deal with H = c0 . In this case the logarithmic norm is given
by µ∞ . From Theorem 19 and Lemma 24 we have on W
X ∂F n X D1 |k|r
i
S(row)k := ≤ ≤ C̃1 (1 + |i|r ).
∂ak |i − k|s−r
k,k6=i k∈Zd ,k6=i

Since by (32) akk ≤ −L∗ |k|p , hence since p > r (by (33)) we obtain
S(row)k + akk → −∞ for |k| → ∞. Therefore there exists l ∈ R, such that
µ∞ (Df n (Pn z)) < l for all n and z ∈ W .
Now we assume H = l1 and the logarithmic norm is µ1 . We have
X ∂F n X D1 |k|r
i
S(col)i := ≤ ≤ |k|r C̃2 .
∂ak |i − k|s−r
i,k6=i i∈Zd ,k6=i

33
We conclude exactly as in the case of µ∞ .
When H = l2 , then we use Gershogorin Theorem [G] to estimate the spec-
1 n n T
trum 2 DF + (DF ) . Observe the radius of k-th Gershgorin circle will be
bounded from above as follows
Ri ≤ (S(col)i + S(row)i )/2 ≤ C̃|i|r .
We conclude as in the previous cases.

7.7 Convergence conditions for variational system

Let us fix W = WP (C, s) with S1, S2 and such that the vector field F = L + N
satisfies C1 and C2 on W . On variational variables we also take set of the form
WV = WP (CV , s). We will show that F also satisfies VC on W × WV .
From Theorem 19 we have for z ∈ W
∂Nk D1 |j|r
(z) ≤ , k 6= j, and
∂aj |k − j|s−r
∂Nk
(z) ≤ D1 |j|r , k = j (56)
∂aj

7.7.1 Condition VC1 for sets with polynomial decay

Lemma 29 The same assumptions and notation as in Theorem 27. Assume
that s > r + d and if H = lq or H = W m,q , then (s − r − m)q > d (with m = 0
for space lq ).
Then F satisfies condition VC1 on W × WV .
Proof: From (56) and Lemma 21 it follows that for (z, C) ∈ W × WV holds
X D1 |j|r CV X 1
g i (u)C| ≤
|DF = D1 CV
|i − j| s−r |j| s |i − j| s−r |j|s−r
j j
1
= D1 CV SQ (d, s − r)
|i|s−r
From considerations as in the proof of Theorem 26 it follows that the image of
g (z)C will be contained in compact set WP (D1 CV SQ (d, s − r), s − r) in the
DF
following cases
• H = c0 if s − r > 1
• H = lq or H = W m,q provided (s − r − m)q > d (with m = 0 for space
lq ).
Therefore for continuity of map FV (u, C) it is enough that its Galerkin pro-
jections are continuous (which is obvious - see the proof of continuity of N in
Theorem 27), because then in view of the above estimates they converge to FV
on W × WV .

34
7.7.2 Condition VC2 for sets with polynomial decay
Lemma 30 The same assumptions and notation as in Theorem 27, but we
restrict H to one of the following spaces c0 , l2 , l1 . Assume additionally that
s > 2r + d. Then F satisfies condition VC2 on W × WV .

Proof: Observe that

 
DF n (x) 0
DPn FV |Hn ×Hn (x, C) =
D2 F n (x)C DF n (x)

To be more precise we have for |i|, |j| ≤ n

∂Fin
DPn FV |Hn ×Hn (x, C) x ,x = ,
i j ∂xj


DPn FV |Hn ×Hn (x, C) xi ,Cj = 0,

X ∂ 2 Fin
DPn FV |Hn ×Hn (x, C) Ci ,xj = Ck ,
∂xj ∂xk
|k|≤n

∂Fin
DPn FV |Hn ×Hn (x, C) Ci ,Cj = .
∂xj

Formulas for logarithmic norms in c0 , l1 , l2 has been given in section 7.6.2.

From these formulas it is clear that to be establish condition VC2 it is enough
to show that on W × WV the following holds for some constants E1 , E2
X

DPn FV |Hn ×Hn (x, C) Ci ,xj ≤ E1 |i|r , (57)
|j|≤n
X

DPn FV |Hn ×Hn (x, C) C ≤ E2 |j|r . (58)
i ,xj
|i|≤n

We start with (57). From Theorem 19 and Lemmas 21 and 24

X
X X ∂ 2 Fin
DPn FV |Hn ×Hn (x, C) C ≤ Ck
i ,xj ∂xj ∂xk
|j|≤n |j|≤n |k|≤n
X X 2
∂ Ni X X D2 |j|r |k|r CV
= Ck ≤
∂xj ∂xk |i − j − k|s−r |k|s
|j|≤n |k|≤n |j|≤n |k|≤n
XX |j|r
≤ D2 CV
j
|i − j − k|s−r |k|s−r
k
X r
|j|
≤ D2 CV SQ (d, s − r) ≤ D2 CV SQ (d, s − r)C(d, s, r)(1 + |i|r ).
j
|i − j|s−r

35
 Now we look establish (58). From Theorem 19 and Lemma 21 we obtain
X
X X ∂ 2 Fin
DPn FV |Hn ×Hn (x, C) C ≤ Ck
i ,xj ∂xj ∂xk
|i|≤n |i|≤n |k|≤n
X X 2
∂ Ni X X D2 |j|r |k|r CV
= · |Ck | ≤
∂xj ∂xk |i − j − k|s−r |k|s
|i|≤n |k|≤n |i|≤n |k|≤n
XX |j|r
≤ D2 CV
i
|i − j − k|s−r |k|s−r
k
X 1
≤ D2 CV SQ (d, s − r)|j|r
i
|i − j|s−r
 
X 1
= D2 CV SQ (d, s − r)  s−r
 |j|r
d
|i|
i∈Z

7.8 Isolation property for the main system

Lemma 31 The vector field (34) is isolating on the family of sets

W = {WP (C, s) : C > 0, s > d + r} .

Proof: Fix C > 0 and s > d + r and W = WP (C, s). From Theorem 19 it
follows that there exists D = D(C, s), such that

D
|Nk (u)| < , for all u ∈ W .
|k|s−r

Let L∗ > 0, p > r, K be constants as in (32). Then for u ∈ W and all |k| > K
C
such that |uk | = |k| s there holds

1 d
(uk |uk ) < −L∗ |k|p |uk |2 + |uk ||Nk (u)| ≤
2 dt


−L∗ C|k|p−s + D|k|r−s |uk |a = −L∗ C|k|p−r + D |k|r−s |uk |.
d
Since p > r we conclude that dt kuk k2 < 0 for |k| > K0 sufficiently large. The
same inequalities hold true for all Galerkin projections of (34) for n > K0 .

7.9 Isolation property for variational equation

We want to show the isolation property for the variational part of system (22)-
(23).

36
Lemma 32 The vector field FV induced by (34) is isolating on the family of
sets
W = {WP (C, s) × WP (CV , s) : C > 0, CV > 0, s > d + r} .
Moreover, the constant K0 does not depend on CV .

Proof: Fix W × WV ∈ W. In view of Lemma 31 we just have to prove

1 d CV
(Ck , Ck ) < 0, if |Ck | = |k|s . (59)
2 dt
Let L∗ > 0, p > r, K be constants as in (32). From (56) for any (z, V ) ∈
W × WV there holds (we use Lemma 21 and the assumption s − r > d)

X ∂Nk X |j|r
(z)Vj ≤ D1 CV
j
∂aj j
|k − j|s−r |j|s
X 1 1
= D1 CV = D1 CV SQ (d, s − r) s−r .
j
|k − j|s−r |j|s−r |k|

From this for |k| > K and z ∈ W and Ck ∈ bd(πk WV ) we get

1 d CV
(Ck , Ck ) < λk s + D1 CV SQ (d, s − r)|k|r−s
2 dt |k|
< −L∗ CV |k|p−s + D1 CV SQ (d, s − r)|k|r−s


= −L∗ |k|p−r + D1 SQ (d, s − r) CV |k|r−s .

Hence, for
  p−r
1
D1 SQ (d, s − r)
|k| > K0 :=
L∗
we obtain (59).

Remark 33 For our further investigations it is crucial that the constant K0 in

Lemma 32 does not depend on CV . This is a natural consequence of linearity
of equations with respect to variational variable.

References
[AK10] G. Arioli and H. Koch, Integration of dissipative PDEs: a case study,
SIAM J. of Appl. Dyn. Syst., 9:1119–1133, (2010).

[AK21] G. Arioli and H. Koch, A Hopf Bifurcation in the Planar

Navier–Stokes Equations Journal of Mathematical Fluid Mechanics
23(3) (2021)

37
[BB] JB van den Berg, M Breden, JP Lessard, L van Veen, Spontaneous
periodic orbits in the Navier–Stokes flow, Journal of Nonlinear Science
31 (2021), 1-64
[B] Burgers, J. M. A Mathematical Model Illustrating the Theory of Tur-
bulence. Advances in Applied Mechanics. 1 (1948): 171–199.
[BKZ] J. Banaśkiewicz, P. Kalita, P. Zgliczyński, Computer-assisted vali-
dation of the existence of periodic orbit in the Brusselator system
Advances in Differential Equations, 29 (11-12), 2024 , 815–862
[CAPD] T. Kapela, M. Mrozek, D. Wilczak, P. Zgliczyński, CAPD::DynSys:
a flexible C++ toolbox for rigorous numerical analysis of dynamical
systems, Communications in Nonlinear Science and Numerical Simu-
lation 101 (2021), 105578.
[CCP] F. Christiansen, P. Cvitanovic, V. Putkaradze, Spatiotemporal chaos
in terms of unstable recurrent patterns, Nonlinearity 10, (1997), 55–70
[C] J. Cyranka, Efficient and generic algorithm for rigorous integration
forward in time of dPDEs: part I, J. Sci. Comp. 59, 28–52 (2014).
[CW] J. Cyranka, T. Wanner, Computer-assisted proof of heteroclinic con-
nections in the one-dimensional Ohta-Kawasaki model, SIAM Journal
on Applied Dynamical. Systems (SIADS), 2018, 17(1), 694 -731
[CZ] J. Cyranka, P. Zgliczyński, Existence of globally attracting solutions
for one-dimensional viscous Burgers equation with nonautonomous
forcing - a computer assisted proof, SIAM Journal on Applied Dy-
namical Systems, 14 (2015), 787–821
[D58] G. Dahlquist, Stability and Error Bounds in the Numerical Intgration
of Ordinary Differential Equations, Almqvist & Wiksells, Uppsala,
1958; Transactions of the Royal Institute of Technology, Stockholm,
1959.
[G] S. Gerschgorin, Über die Abgrenzung der Eigenwerte einer Matrix,
Izv. Akad. Nauk. USSR Otd. Fiz.-Mat. Nauk, 6 (1931): 749–754.
[H] D. Henry, Geometric Theory of Semilinear Parabolic Equations, Lect.
Notes in Math. 840, Springer 1981.
[HNW] E. Hairer, S.P. Nørsett and G. Wanner, Solving Ordinary Differential
Equations I, Nonstiff Problems, Springer-Verlag, Berlin Heidelberg
1987.
[KZ] T. Kapela, P. Zgliczyński, A Lohner-type algorithm for control sys-
tems and ordinary differential inclusions, Discrete Cont. Dyn. Sys.
B, 11 (2009) 365–385.

38
[K] H.-O. Kreiss, Fourier expansions of the solutions of the Navier-Stokes
equations and their exponential decay rate, in Analyse mathématique
et applications, 245–262, Gauthier-Villars, Montrouge, 1988.
[KT] Y. Kuramoto, T. Tsuzuki, Persistent propagation of concentration
waves in dissipative media far from thermal equilibrium, Prog. Theor.
Phys., 55,(1976), 365
[Lo] R.J. Lohner, Computation of Guaranteed Enclosures for the Solutions
of Ordinary Initial and Boundary Value Problems, in: Computational
Ordinary Differential Equations, J.R. Cash, I. Gladwell Eds., Claren-
don Press, Oxford, 1992.
[L58] S. M. Lozinskii, Error esitimates for the numerical integration of or-
dinary differential equations, part I, Izv. Vyss. Uceb. Zaved. Matem-
atica,6 (1958), 52–90 (Russian)
[Mo] R.E. Moore, Methods and Applications of Interval Analysis,SIAM,
Philadelphia, 1979.
[NJP] Nedialkov, N. S., Jackson, K. R., Pryce, J. D., 2001. An effective
high-order interval method for validating existence and uniqueness of
the solution of an IVP for an ODE. Reliab. Comput. 7 (6), 449–465.
[Sa] Y. Sannikov, Mathematical Problems Related to Navier-Stokes Equa-
tions, Thesis, Princeton University, 2000
[S] G.I. Sivashinsky, Nonlinear analysis of hydrodynamical instability in
laminar flames – 1. Derivation of basic equations, Acta Astron. 4
(1977), no. 11-12, 1177–1206.
[T] Temam, R. Infinite-Dimensional Dynamical Systems in Mechanics
and Physics. In: Applied Mathematical Sciences, 2nd Edition, Vol-
ume 68 (1997), Springer-Verlag, New York.
[WZ] D. Wilczak, P. Zgliczyński, A geometric method for infinite-dimen-
sional chaos: Symbolic dynamics for the Kuramoto-Sivashinsky PDE
on the line, J. Diff. Eq., 269/10 (2020), 8509–8548.
[Z] P. Zgliczyński, Trapping regions and an ODE-type proof of the exis-
tence and uniqueness for Navier-Stokes equations with periodic bound-
ary conditions on the plane, Univ. Iag. Acta Math., 41 (2003) 89–113.
zz
[Z1] P. Zgliczyński, On smooth dependence on initial conditions for dis-
sipative PDEs, an ODE-type approach, J. Diff. Eq., 195/2 (2003),
271–283.
[ZM] P. Zgliczyński and K. Mischaikow, Rigorous Numerics for Partial
Differential Equations: the Kuramoto-Sivashinsky equation, Found.
Comp. Math., (2001) 1:255–288.

39
[Z04] P. Zgliczyński, Rigorous numerics for dissipative Partial Differential
Equations II. Periodic orbit for the Kuramoto-Sivashinsky PDE - a
computer assisted proof, Found. Comput. Math., 4 (2004), 157–185.
[Z10] P. Zgliczyński, Rigorous Numerics for Dissipative PDEs III. An ef-
fective algorithm for rigorous integration of dissipative PDEs, Topol.
Methods Nonlinear Anal., 36 (2010) 197-–262.

40