See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/226769301

Blowup of solutions to generalized Keller–Segel model

Article in Journal of Evolution Equations · January 2009

DOI: 10.1007/s00028-009-0048-0 · Source: arXiv

CITATIONS READS
100 205

2 authors:

Piotr Biler Grzegorz Karch

University of Wroclaw University of Wroclaw
145 PUBLICATIONS 4,093 CITATIONS 102 PUBLICATIONS 2,429 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Chemotaxis, Keller-Segel model View project

Pattern formation, reaction-diffusion equations, mathematical biology View project

All content following this page was uploaded by Piotr Biler on 22 May 2014.

The user has requested enhancement of the downloaded file.

 Blow up of solutions
to generalized Keller–Segel model

Piotr Biler Grzegorz Karch

Instytut Matematyczny, Uniwersytet Wroclawski
arXiv:0812.4982v1 [math.AP] 29 Dec 2008

pl. Grunwaldzki 2/4, 50–384 Wroclaw, Poland

{Piotr.Biler,Grzegorz.Karch}@math.uni.wroc.pl

December 29, 2008

Abstract
The existence and nonexistence of global in time solutions is studied for
a class of equations generalizing the chemotaxis model of Keller and
Segel. These equations involve Lévy diffusion operators and general
potential type nonlinear terms.

Key words and phrases: nonlocal parabolic equations, blow up of solu-

tions, Lévy diffusion, chemotaxis, moment method

2000 Mathematics Subject Classification: 35Q, 35K55, 35B40.

1 Introduction

We consider in this paper the following nonlinear nonlocal evolution equation

generalizing the well known Keller–Segel model of chemotaxis

∂t u + (−∆)α/2 u + ∇ · (uB(u)) = 0, (1.1)

for (x, t) ∈ Rd ×R+ , where the anomalous diffusion is modeled by a fractional

power of the Laplacian, α ∈ (1, 2), and the linear (vector) operator B is
defined (formally) as
B(u) = ∇((−∆)−β/2 u). (1.2)

For β ∈ (1, d], and d ≥ 2, one may express the nonlocal nonlinearity in (1.1)
using convolution operators since
Z
x−y
B(u)(x) = sd,β u(y) dy, (1.3)
Rd |x − y|d−β+2

1
with some sd,β > 0, and the assumption β > 1 is needed for the convergence
of this integral.
Of course, the choice α = 2, β = 2 in (1.1)–(1.3) corresponds to the usual
Keller–Segel system studied mainly in space dimensions d = 1, 2, 3. It is
well known that if β = 2, the one-dimensional system (1.1)–(1.3) possesses
global in time solutions not only in the case of classical Brownian diffusion
α = 2 but also in the fractional diffusion case 1 < α < 2 as was shown
by C. Escudero in [17]. On the other hand, there are many results on the
nonexistence of global in time solutions with “large” initial data if d ≥ 2 and
α = 2, β = 2, see, e.g., [19, 2, 1, 12]. Even if d ≥ 2, α = 2 and 1 < β ≤ d,
there are results on the blow up of solutions with suitably chosen initial
data, see [10] (caution: the notation in [10, Prop. 4.2] differs from that in
the present paper). Let us finally recall that, in the limit case α = 2, β = d,
R
mass M = Rd u0 (x) dx is the critical parameter for the blow up, see [10,
Prop. 4.1] and [14].
The usual method of proving the nonexistence of global in time nonneg-
ative and nontrivial solutions, used in the abovementioned papers, consists
in the study the evolution of the second moment of a solution w2 (t) =
R 2
Rd |x| u(x, t) dx and to show (via suitable differential inequalities) that
w2 (t) vanishes for some t > 0. The second moment of a typical solution
to an evolution equation with fractional Laplacian cannot be finite, see e.g.
[13]. Hence, our goal in this paper is to generalize the classical virial method
and to show the blow up of solution system (1.1)–(1.3) by studying moments
of lower order γ ∈ (1, 2)
Z
wγ = |x|γ u(x) dx. (1.4)
Rd

After this paper was completed, we discovered a recent preprint [23]

where the authors show the blow up of solution to system (1.1)–(1.3) with
fractional diffusion in the particular case d = 2 and β = 2. Our argument
is different than that in [23], shorter, seems to be more direct, and applies
in more general situations. Moreover, we are able to formulate a simple
condition on the initial data which leads to the blow up in a finite time of

2
the corresponding solution.

Notation. The Lp -norm of a Lebesgue measurable, real-valued function

v defined on Rd is denoted by kvkp . The constants (always independent of
x, t) will be denoted by the same letter C, even if they may vary from line
to line. Sometimes, we write, e.g., C = C(∗) when we want to emphasize
the dependence of C on a parameter “∗”.

2 Main results

The crucial role in the approach in this paper is played by the following
scaling property of system (1.1)–(1.3)

uλ (x, t) = λα+β−2 u(λx, λα t) for all λ > 0, (2.1)

in the sense that if u is a solution to (1.1)–(1.3), then uλ is so. In par-

ticular, in our construction of solutions to (1.1)–(1.3) we use the fact that
the usual norm of the Lebesgue space Ld/(α+β−2) (Rd ) is invariant under the
transformation u0 (x) 7→ λα+β−2 u0 (λx) for every λ > 0.

Theorem 2.1. Assume that d ≥ 2, α ∈ (1, 2], and β ∈ (1, d]. Let
 
d 2d
max , < p ≤ d.
α+β−2 d+β−1
i) For every u0 ∈ Lp (Rd ) there exists T = T (ku0 kp ) and the unique local
in time mild solution u ∈ C([0, T ], Lp (Rd )) of system (1.1)–(1.3) with
u0 as the initial condition.

ii) There is ε > 0 such that for every u0 ∈ Ld/(α+β−2) (Rd ) satisfying

ku0 kd/(α+β−2) ≤ ε (2.2)

there exists a global in time mild solution u ∈ C([0, ∞), Lp (Rd )) of

system (1.1)–(1.3) with u0 as the initial condition.

Moreover, if u0 (x) ≥ 0, then the solution u in either i) or ii) above is non-

negative. Finally, if u0 ∈ L1 (Rd ), then the corresponding solution conserves
mass Z Z
u(x, t) dx = u0 (x) dx ≡ M. (2.3)
Rd Rd

3
 Recall that α > 1 is a usual assumption ([10, Th. 2.2], [8, 9]) which
permits us to control locally the nonlinearity in (1.1)–(1.3) by the linear
term.
The results stated above can be easily generalized for equations with
general Lévy diffusion operators considered in [8, 9] but we do not pursue
this question here. We refer the reader to the above mentioned papers, as
well as [10], for physical motivations to study such equations. On the other
hand, motivations stemming from probability theory (propagation of chaos
property for interacting particle systems) can be found in, e.g., [7].

Remark 2.2. For α ∈ (1, 2) and β > 1 satisfying α + β > d + 2, it can

be shown that local in time solutions can be continued to the global in
time ones, see [10, Th. 3.2]. However, in [10], another approach (via weak
solutions) is used to construct solutions of system (1.1)–(1.3).

The proof of Theorem 2.1 on local and global solutions to (1.1)–(1.3)

follows a more or less standard reasoning which we sketch in Section 3. Our
main goal, however, is to prove the finite time blow up of solutions to the
nonlocal system (1.1)–(1.3).

Theorem 2.3. Assume that d ≥ 2. The solution of (1.1)–(1.3) with a non-

negative and nonzero initial condition u(x, 0) = u0 (x) blows up in a finite
time in each of the following cases:

i) (large mass) for α = 2, β = d, u0 ∈ L1 (Rd , (1 + |x|2 ) dx), and if

Z
M= u0 (x) dx > 2d/sd,β ,
Rd
1
with the constant sd,β defined in (1.3); in particular: s2,2 = 2π so that
the threshold value of M is 8π if d = 2;

ii) (high concentration) for α ∈ (1, 2] and β ∈ (1, d] satisfying α + β <

d + 2, u0 ∈ L1 (Rd , (1 + |x|γ ) dx) for some γ ∈ (1, α), and if
R Z ! γ
|x|γ u (x) dx d+2−α−β
RRd 0
≤c u0 (x) dx (2.4)
Rd u0 (x) dx Rd

for certain (sufficiently small) constant c > 0 independent of u0 .

4
 The result stated in i) is essentially contained in [14]. The condition for
blow up in the form (2.4) appeared already in [2] and [25], of course, for
α = 2 only. Note that i) is a limit case of ii). Indeed, (2.4) written as
R γ  d+2−α−β
RRd |x| u0 (x) dx
γ
≤ cM,
Rd u0 (x) dx
becomes a condition on (sufficiently large) mass: 1 ≤ cM , when (α + β) ր
(d + 2).
We have to emphasize that Theorem 2.3.ii contains a result which is
new even for the classical parabolic-elliptic Keller–Segel model (i.e. equa-
tions (1.1)–(1.3) with α = β = 2). Indeed, for d ≥ 3, the conditions from
part ii) of Theorem 2.3 guarantee the blow up in a finite time if the moment
of order γ of the initial condition is finite for some γ ∈ (1, 2). All other
known proofs of the blow up required just γ = 2. An analogous result for
d = 2 is stated below.

Corollary 2.4. Assume that α = β = d = 2 in equations (1.1)–(1.3).

Suppose that there exists γ ∈ (1, 2) such that u0 ∈ L1 (R2 , (1 + |x|γ ) dx).
There exists Mγ > 0 such that if
Z
u0 (x) dx > Mγ ,
R2
then the solution of (1.1)–(1.3) with the initial condition u(x, 0) = u0 (x)
blows up in a finite time.

It is conjectured that M2 = Mγ = 8π for every γ ∈ (1, 2) and we

expect that it can be shown by a careful analysis of constants appearing in
inequalities (4.6)–(4.10) and in (4.18)–(4.20) in the proof of Theorem 2.3.
Here, one should recall that radially symmetric solutions to equations (1.1)–
(1.3) with α = β = d = 2 blow up in finite time under the assumption that
their mass is larger than 8π and no moment condition imposed on the initial
data is necessary, see [4, Prop. 2.2, Th. 3.1] for a detailed presentation.

Remark 2.5. Due to the translation invariance of problem (1.1)–(1.3), the

conditions on moments can be imposed on the quantity
Z Z
γ
inf |x − x0 | u0 (x) dx = |x − x̄|γ u0 (x) dx,
x0 ∈Rd Rd Rd

5
 R R
where x̄ = ( Rd xu0 (x) dx)/( Rd u0 (x) dx) is the center of mass of u0 .

Remark 2.6. Let us observe that the assumptions (2.2) and (2.4) are in
a sense complementary due to the following elementary inequality involving
R R
the Lp -norms, mass M = Rd u(x) dx, and the moment wγ = Rd |x|γ u(x) dx
of a nonnegative function u
 d
“ ”
1− p1
M γ
kukp ≥ CM . (2.5)
wγ
R
To prove (2.5) observe that wγ ≥ Rγ Rd \BR (0) u(x) dx, so that
Z Z
u(x) dx = M − u(x) dx ≥ M − R−γ wγ .
BR (0) Rd \BR (0)
“ ”
d 1− p1 1− p1
Multiplying both sides of this inequality by R we get with C = ωd
Z 1 Z !1− 1
p “ ”
p
p d p1 −1
Ckukp = u (x) dx dx R
Rd BR (0)
“ ”Z “ ”
d 1p −1 d p1 −1
≥ R u(x) dx ≥ R M − R−γ wγ .
BR (0)

Taking the optimal R, i.e. Rγ = Cwγ /M , we get (2.5).

Now, it is clear that if condition (2.4) for blow up is satisfied for some
γ ∈ (1, 2] and a suitable constant c, then for p = d/(α
“ +”β − 2) appearing in
γ d
− d+2−α−β 1− p1
Theorem 2.1.ii, we have kukp ≥ CM M γ
= C, so condition
(2.2) for global existence is violated for sufficiently small ε > 0. Vice versa,
if (2.2) is satisfied, (2.4) cannot be true with small constants c.

Remark 2.7. Note that one can prove in a similar way the inequality
 d
“ ”
1− p1
M γ
kukM p ≥ CM , (2.6)
wγ
where the Morrey space norm kukM p (≤ Ckukp ) is defined as
“ ”Z
d p1 −1
sup R |u(x)| dx,
R>0, x0 ∈Rd BR (x0 )

see [2, (15)] in the particular case γ = 2. The scale of Morrey spaces
is relevant to study mean field type problems related to the Keller–Segel

6
model, since the Morrey norms are particularly well suited to measure the
(local) concentration of densities, see e.g. [3].

We prove Theorem 2.3 and Corollary 2.4 by showing the extinction

R
in a finite time of the function w(t) = Rd ϕγ (x)u(x, t) dx where ϕγ (x) is
smooth and behaves like |x|γ with some γ ∈ (1, α) for large |x|, see (4.1)
below. Note that if α < 2, we cannot expect the existence of higher or-
der moments wγ defined in (1.4) with γ ≥ α. Indeed, even for the linear
equation ∂t v + (−∆)α/2 v = 0, the fundamental solution pα (x, t) behaves like

−1
pα (x, t) ∼ td/α + |x|d+α /t , and therefore the moment (1.4) with γ ≥ α
cannot be finite, see [13] and references given there. Thus, we cannot apply
the usual reasoning which involves an analysis of the evolution of the second
moment w2 of the solution because the integral defining w2 may diverge.
We recall in Proposition 3.4 a result showing that the moment (1.4) is
finite for a large class of initial conditions in the case of γ < α.

3 Existence of solutions

We are going to construct solutions to system (1.1)–(1.3) via the following

integral formulation
Z t

u(t) = Sα (t)u0 − ∇ · Sα (t − τ ) u(τ )Bu(τ ) dτ , (3.1)
0

i.e., we consider mild solutions. Here Sα (t)u0 = pα (t) ∗ u0 is the solution to

the linear Cauchy problem

∂t v + (−∆)α/2 v = 0, v(x, 0) = u0 , (3.2)

and pα (x, t) is the fundamental solution of (3.2) which can be represented

α
via the Fourier transform pbα (ξ, t) = e−t|ξ| . In particular,

pα (x, t) = t−d/α Pα (xt−1/α ),

α
where Pα is the inverse Fourier transform of e−|ξ| , see [18, Ch. 3] and [13]
for more details. It is well known that for every α ∈ (0, 2) the function Pα

7
is smooth, nonnegative, and satisfies the (optimal) estimates

0 < Pα (x) ≤ C(1 + |x|)−(α+d) and |∇Pα (x)| ≤ C(1 + |x|)−(α+d+1) (3.3)

for a constant C and all x ∈ Rd . Hence, it follows immediately from the

Young inequality for the convolution and from the self-similar form of the
kernel pα (x, t) that for every 1 ≤ q ≤ p ≤ ∞ there exists C = C(p, q, α) > 0
such that “ ”
d 1
−α − p1
kSα (t)u0 kp ≤ Ct q
ku0 kq (3.4)

and “ ”
d 1
−α − p1 1
−α
k∇Sα (t)u0 kp ≤ Ct q
ku0 kq (3.5)

for every u0 ∈ Lq (Rd ) and all t > 0.

The construction of solution to (3.1) in Lp spaces is based on the follow-

ing abstract result, see e.g. [22], [24].

Lemma 3.1. Let (X , k · kX ) be a Banach space and H : X × X → X

a bounded bilinear form satisfying kH(x1 , x2 )kX ≤ ηkx1 kX kx2 kX for all
x1 , x2 ∈ X and a constant η > 0. Then, if 0 < ε < 1/(4η) and if y ∈ X is
such that kyk < ε, the equation u = y + H(u, u) has a solution in X such
that kukX ≤ 2ε. This solution is the only one in the ball B̄(0, 2ε).

We skip an easy proof of this lemma which is a direct consequence of the

Banach fixed point theorem.

Sketch of the proof of Theorem 2.1. The proof consists in constructing solu-
tions to the “quadratic” equation (3.1) using Lemma 3.1 with y = Sα (t)u0
and with the bilinear form
Z t

H(u, v) = − ∇ · Sα (t − τ ) u(τ )Bv(τ ) dτ . (3.6)
0

Local existence of solutions. It suffices to obtain estimates required by

Lemma 3.1 in the Banach space XTp = C([0, T ], Lp (Rd )) supplemented with
the usual norm kukX p = supt∈[0,T ] ku(t)kp .
T

By inequality (3.4), we immediately obtain kSα (·)u0 kX p ≤ ku0 kp .

T

8
 Combining the definition (1.3) of the operator B with the Hardy–Little-
wood–Sobolev inequality we obtain

kB(u)kq ≤ sd,β | · |−d+β−1 ∗ u q

≤ Ckukp (3.7)

1 β−1
for every 1 < p < q < ∞ satisfying p − d = 1q . Moreover, inequality (3.5)
and the Hölder inequality lead to
“ ”
1 d 1
−α −α − p1
k∇Sα (t − τ )(uB(v))kp ≤ C(t − τ ) r
kuB(v)kr
1 d
“
1
” (3.8)
−α −α − p1
≤ C(t − τ ) r
kukp kB(v)kq .
1 1
where r = p + 1q . Hence, by inequalities (3.7) and (3.8), there exists a con-
stant C such that for all u, v ∈ XTp , the bilinear form (3.6) satisfies
Z t 1
“
d 1
”
−α −α − p1
kH(u, v)kX p ≤ C sup (t − τ ) r
ku(τ )kp kv(τ )kp dτ
T
t∈[0,T ] 0 (3.9)
“ ”
1 d 1
1− α −α − p1
≤ CT r
kukX p kvkX p .
T T

1 1 1 2 β−1
In estimates (3.9), we have used the relations r = p + q = p − d , and we
assume that

• p > d/(α + β − 2) in order to have 1/α + (d/α)(1/r − 1/p) < 1;

• p > 2d/(d + β − 1) to guarantee that r > 1;

• p ≤ d/(β − 1) to be sure that r ≤ p.

Choosing a sufficiently small T > 0 in (3.9) we complete the proof of

Theorem 2.1.i by an application of Lemma 3.1.

Global in time solutions. Here, the reasoning is completely analogous:

using inequalities (3.4), (3.5), (3.7)–(3.9) we estimate the bilinear form (3.6)
in the Banach space

X p =C([0, ∞), Ld/(α+β−2) (Rd ))

“ ”
p d
d 1
− α+β−2
∩ {u ∈ C((0, ∞), L (R )) : sup t α p d
ku(t)kp < ∞}
t>0

supplemented with the norm

“ ”
d 1
− α+β−2
kukX p = sup ku(t)kd/(α+β−2) + sup t α p d
ku(t)kp .
t>0 t>0

9
We skip further details of this standard reasoning.

Nonnegativity property. In order to prove that u0 ≥ 0 implies u(t) ≥ 0, it

suffices to study the function u− (x, t) = max{−u(x, t), 0} and to follow the
arguments either from [9, Prop. 3.1] or from [16, Prop. 2] in order to show
that u− (x, t) ≡ 0. Here, we do not give a detailed presentation because the
proof from [23, Lemma 2.7] can be rewritten in this more general case.

Conservation of the integral. If u0 ∈ L1 (Rd ), one should repeat the fixed

point argument from i) (or from ii)) in the space XTp ∩ C([0, T ], L1 (Rd )) (in
X p ∩C([0, ∞), L1 (Rd )), resp.) in order to have u(t) ∈ L1 (Rd ) for all t ∈ [0, T ]
(t ∈ (0, ∞), resp.). Next, it suffices to integrate over Rd the both sides of
equation (3.1). Using the following consequences of the Fubini theorem
Z Z
pα (t) ∗ u0 (x) dx = u0 (x) dx
Rd Rd

and Z
∇pα (t) ∗ v(x) dx = 0 for every v ∈ L1 (Rd ),
Rd
R R
we conclude that Rd u(x, t) dx = Rd u0 (x) dx for every t ∈ [0, T ].

Remark 3.2. The reasoning from the proof of Theorem 2.1 follows the lines
of the usual proof of the local in time existence (as well as the global in time
existence for small initial conditions) of solutions to system (1.1)–(1.3) with
α = 2. Since that argument is based on an integral representation analogous
to that in (3.1) and on counterparts of decay estimates from (3.4)–(3.5), it
can be easily adopted to the more general case of α ∈ (1, 2]. Examples
of such a reasoning applied to various semilinear models and realized in
miscellaneous Banach spaces can be found in [3, 5, 6, 11, 20, 21, 22, 24].

Next, we recall weighted estimates of solutions to the linear Cauchy

problem (3.2) which have been proved in, e.g., [13]. In the following, we use
the weighted L∞ spaces

L∞ d ∞ d
ϑ (R ) = {v ∈ L (R ) : kvkL∞
ϑ
≡ ess supx∈Rd (1 + |x|)ϑ |v(x)| < ∞}

for a fixed ϑ ≥ 0.

10
Lemma 3.3. ([13, Lemma 3.1]) Assume that v0 ∈ L∞ d
α+d (R ). There exists
C > 0 independent of v0 and t such that

kSα (t)v0 kL∞

α+d
≤ C(1 + t)kv0 kL∞
α+d
,
k∇Sα (t)v0 kL∞
α+d
≤ Ct−1/α kv0 kL∞
α+d
+ Ct1−1/α kv0 k1 ,

With these estimates, we are in a position to construct solutions to

system (1.1)–(1.3) also in the weighted space L∞ d 1 d ∞ d
α+d (R ) ⊂ L (R ) ∩ L (R ).

Proposition 3.4. Let α ∈ (1, 2] and b ∈ (1, d]. Assume that u is the
solution of system (1.1)–(1.3), constructed in Theorem 2.1, supplemented
with the initial condition u0 ∈ L∞ d ∞ d
α+d (R ). Then u ∈ C([0, T ], Lα+d (R )). In
R
particular, for each γ < α we have Rd |x|γ u(x, t) dx < ∞.

Sketch of proof. Here, it suffices to modify slightly the argument from [13,
Proof of Prop. 3.3.i] written the case of a convection equation with the frac-
tional Laplacian. In that reasoning (as well as in the above proof of Theorem
2.1), we construct solutions to equation (3.1) by applying Lemma 3.1 with
the Banach space XT = C([0, T ], L∞ d
α+d (R )). Obviously, y = Sα (·)u0 ∈ XT
by Lemma 3.3. Next, we show the following estimate of the bilinear form
from (3.6)
kH(u, v)kXT ≤ CT 1−1/α ku|XT kvkXT

for all u, v ∈ XT and a constant C independent of u, v. Here, in order to

estimate the singular integral operator (1.3) in the spaces L∞ d
ϑ (R )), one
should follow the reasoning from [5, Sec. 2]. Let us omit other details of
this classical argument.

4 Blow up of solutions

The main role in our proof of the blow up of solutions to (1.1)–(1.3) is played
by the following smooth nonnegative weight function on Rd

ϕ(x) = ϕγ (x) ≡ (1 + |x|2 )γ/2 − 1 (4.1)

11
with γ ∈ (1, 2]. Since (1 + |x|2 )γ ≤ (1 + |x|γ )2 , we have for each ε > 0,
suitably chosen C(ε) > 0, and for every x ∈ Rd

ϕ(x) ≤ |x|γ ≤ ε + C(ε)ϕ(x). (4.2)

Next, let us state two auxiliary results concerning the weight function ϕ
which will be used in the proof of Theorem 2.3. Here, for a given ϕ ∈ C 2 (Rd ),
we denote by D 2 ϕ its Hessian matrix. Moreover, the scalar product of
vectors x, y ∈ Rd is denoted by x · y. If A is either a vector or a matrix, the
expression |A| means its Euclidean norm.

Lemma 4.1. Let α ∈ (1, 2), γ ∈ (1, α), and ϕ be defined by (4.1). Then

(−∆)α/2 ϕ ∈ L∞ (Rd ). (4.3)

Proof. First note that by a direct computation we have

γ
∇ϕ(x) = γ(1 + |x|2 ) 2 −1 x (4.4)

and

γ
∂xj ∂xi ϕ(x) = γ(1 + |x|2 )δi j − γ(2 − γ)xi xj (1 + |x|2 ) 2 −2 . (4.5)

In particular, for every R > 0 there exists C(R, γ) > 0 such that for all
|x| ≥ R we have

|∇ϕ(x)| ≤ C(R, γ)|x|γ−1 and |D2 ϕ(x)| ≤ C(R, γ)|x|γ−2 . (4.6)

Now, we apply the following Lévy–Khintchine integral representation of

the fractional Laplacian
Z
α/2 ϕ(x + y) − ϕ(x) − ∇ϕ(x) · y
(−∆) ϕ(x) = C(d, α) dy (4.7)
Rd |y|d+α
(with a suitable constant C(d, α)) which is valid for every α ∈ (1, 2), see,
e.g. [16, Th. 1] for a detailed proof of that version of Lévy–Khintchine for-
mula. Using the Taylor expansion and estimates (4.6) one can immediately
show that (−∆)α/2 ϕ(x) is well defined for every x ∈ Rd and, moreover,
sup|x|≤R |(−∆)α/2 ϕ(x)| < ∞ for each R > 0.

12
 In order to obtain an estimate uniform in x ∈ Rd , we assume that |x| ≥ 1
and we shall estimate the integral on the right hand side of (4.7) for |y| ≤
|x|/2 and |y| > |x|/2, separately.
If |y| ≤ |x|/2, by the Taylor formula and the second inequality in (4.6),
we obtain
Z 1
1 2
|ϕ(x + y) − ϕ(x) − ∇ϕ(x) · y| ≤ |y| |D 2 ϕ(x + sy)| ds
2 0
Z 1
≤ C|y|2 |x + sy|γ−2 ds.
0

Since |y| ≤ |x|/2 and s ∈ [0, 1] we can estimate

1
|x + sy| ≥ |x| − s|y| ≥ |x| − |y| ≥ |x|.
2

Consequently, for γ − 2 < 0, we obtain

Z
ϕ(x + y) − ϕ(x) − ∇ϕ(x) · y
dy
|y|≤|x|/2 |y|d+α
Z (4.8)
γ−2 dy γ−α
≤ C|x| = C|x|
|y|≤|x|/2 |y|d+α−2

for all |x| ≥ 1 and a constant C > 0 independent of x.

If |y| ≥ |x|/2 and |x| ≥ 1, we combine first inequality from (4.6) (remem-
ber that γ − 1 > 0) with the Taylor expansion to show
Z 1

|ϕ(x + y) − ϕ(x)| ≤ |y| |∇ϕ(x + sy)| ds ≤ C|y| |x|γ−1 + |y|γ−1 .
0

Hence,
Z
ϕ(x + y) − ϕ(x) − ∇ϕ(x) · y
dy
|y|>|x|/2 |y|d+α
Z Z ! (4.9)
dy dy
≤ C |x|γ−1 d+α−1
+ d+α−γ
= C|x|γ−α
|y|>|x|/2 |y| |y|>|x|/2 |y|

for all |x| ≥ 1 and a constant C > 0 independent of x.

Finally, inequalities (4.8) and (4.9) complete the proof because γ < α.

13
Remark 4.2. Note that above we have, in fact, proved that


sup 1 + |x|α−γ (−∆)α/2 ϕ(x) < ∞ for every γ ∈ (1, α).
x∈Rd

Lemma 4.3. For every γ ∈ (1, 2], the function ϕ defined in (4.1) is locally
uniformly convex on Rd . Moreover, there exists K = K(γ) such that the
following inequality
K|x − y|2
(∇ϕ(x) − ∇ϕ(y)) · (x − y) ≥ (4.10)
1 + |x|2−γ + |y|2−γ

holds true for all x, y ∈ Rd .

Proof. Using the explicit expression for the Hessian matrix of ϕ in (4.5) we
obtain
P P 
γ(1 + |x|2 )|y|2 − γ(2 − γ) 2 2
i xi yi + i6=j xi xj y i y j
D 2 ϕ(x)y·y = (4.11)
(1 + |x|2 )2−γ/2

for every x, y ∈ Rd . Now, by the elementary inequality xi xj yi yj ≤ 12 (x2i yj2 +

x2i yj2 ) we immediately obtain
X X
xi xj yi yj ≤ x2i yj2 .
i6=j i6=j

Consequently,
X X X
x2i yi2 + xi xj yi yj ≤ x2i yj2 = |x|2 |y|2 . (4.12)
i i6=j i,j

Since γ ∈ (1, 2], applying estimate (4.12) to (4.11) we get the inequality

γ(1 + (γ − 1)|x|2 )|y|2

D 2 ϕ(x)y · y ≥
(1 + |x|2 )2−γ/2
which leads directly to the estimate from below
(γ − 1)|y|2
D 2 ϕ(x)y · y ≥ . (4.13)
(1 + |x|2 )1−γ/2
Finally, it follows from the integration of the second derivative of ϕ that
  Z 1


∇ϕ(x) − ∇ϕ(y) · (x − y) = D2 ϕ x + s(y − x) (x − y) · (x − y) ds.
0

14
Hence, using inequality (4.13) and the estimate

(1 + |x + s(y − x)|2 )1−γ/2 ≤ C(1 + |x|2−γ + |y|2−γ ),

valid for all x, y ∈ Rd , s ∈ [0, 1] and a constant C > 0 independent of x, y, s,

one can easily complete the proof of Lemma 4.3.

Proof of Theorem 2.3. We consider the function

Z
w = w(t) ≡ ϕ(x) u(x, t) dx,
Rd
where ϕ is defined in (4.1) and 1 < γ < α. Note that, in view of inequalities
(4.2), the quantity w is essentially equivalent to the moment wγ of order γ
of the solution u. Moreover, it satisfies the relation
Z Z
d α/2
w =− (−∆) u(x, t) ϕ(x) dx − u(x, t)Bu(x, t) · ∇ϕ(x) dx
dt d Rd
ZR
=− (−∆)α/2 ϕ(x) u(x, t) dx
Rd
Z Z  
sd,β u(x, t)u(y, t)
− ∇ϕ(x) − ∇ϕ(y) · (x − y) dx dy
2 Rd Rd |x − y|d−β+2
(4.14)
after using the definition of the form B in (1.3) and the symmetrization of
the double integral. This computation resembles the usual proof of blow up
involving the second moments, cf. [2, 10, 15, 14].

i) For α = 2 = γ (hence for ϕ(x) = |x|2 ) and for β = d, the equality

(4.14) can be rewritten as follows
d
w(t) = 2dM − sd,β M 2 .
dt
Evidently, for M > 2d/sd,β , this implies the equality w(T ) = 0 for some
T > 0, a contradiction with the global existence of nonnegative solutions.
Thus, we recover the result in [10, Prop. 4.1] refined in [15, 14].

ii) For 1 < β ≤ d and fixed M > 0, we are going to use the following
simple identity
Z Z
2
M = u(x, t)u(y, t) dx dy
Rd Rd
Z Z
δ
|x − y|ν 1 + |x|2−γ + |y|2−γ
= u(x, t)u(y, t) dx dy
Rd Rd (1 + |x|2−γ + |y|2−γ )δ |x − y|ν

15
with some ν > 0 and δ > 0. We apply now the Hölder inequality with the
p
powers p > 1 and p′ = p−1 chosen so that

νp = d − β, δp = 1, νp′ + (2 − γ)δp′ = γ. (4.15)

Of course, such a choice of ν, δ, p is possible whenever β < d and γ < 2

because we only need d − β + 2 − γ = γ(p − 1). If β = d, it suffices to take
ν = 0 and p = 2/γ > 1. As a consequence, we get

M 2 ≤ J(t)1/p
Z Z 1/p′
′
δp′
× u(x, t)u(y, t)|x − y|νp 1 + |x|2−γ + |y|2−γ dx dy ,
Rd Rd
(4.16)

where the integral J(t) satisfies

Z Z
u(x, t)u(y, t) dx dy
J(t) = d−β 2−γ
Rd Rd |x − y| 1 + |x| + |y|2−γ
Z Z   (4.17)
1 u(x, t)u(y, t)
≤ ∇ϕ(x) − ∇ϕ(y) · (x − y) dx dy
K Rd Rd |x − y|d−β+2
by Lemma 4.3.
It follows from relations (4.15) and inequalities (4.2) that there exists
a constant C1 > 0 such that
′
δp′  
|x − y|νp 1 + |x|2−γ + |y|2−γ ≤ C1 1 + ϕ(x) + ϕ(y) . (4.18)

Hence, (4.16) implies

 1/p′
1/p′
M 2 ≤ C1 J(t)1/p M 2 + 2M w(t)) . (4.19)

Going back to identity (4.14) we obtain from Lemma 4.1 and from inequal-
ities (4.17)–(4.19) that

d M 2p
w(t) ≤ C2 M − C3 (4.20)
dt (M 2 + 2M w(t))p/p′

with C2 = k(−∆)α/2 ϕk∞ and a suitable constant C3 > 0.

Now, we fix for a while M = M0 in (4.20) so large in order to have

M02p
C2 M0 − C3 ′ < 0. (4.21)
(M02 )p/p

16
Hence, there exists C4 = C4 (M0 ) > 0 such that for 0 < w(0) ≤ C4 we still
have
M02p
C2 M0 − C3 < 0.
(M02 + 2M0 w(0))p/p′
It is clear that if initially 0 < w(0) ≤ C4 then, by inequality (4.20) with
M = M0 , the function w(t) is decreasing in time. Moreover,

d M02p
w(t) ≤ C2 M0 − C3 <0
dt (M02 + 2M0 w(0))p/p′

and, consequently, w(T ) = 0 for some 0 < T < ∞. This contradicts the
global in time existence of regular nonnegative solutions of (1.1)–(1.3). Fi-
nally, note that due to the first inequality in (4.2), it suffices to assume
Z Z
γ
w(0) ≤ |x| u0 (x) dx ≤ C4 and u0 (x) dx = M0 , (4.22)
Rd Rd

in order to obtain the blow up in a finite time of the corresponding solution.

R R
Now, assume that Rd u(x, t) dx = Rd u0 (x) dx = M 6= M0 . Recall
that system (1.1)–(1.3) is invariant under the rescaling (2.1). Choosing
R R
λα+β−2−d = M0 /M we obtain Rd uλ (x, t) dx = Rd uλ0 (x) dx = M0 and, by
(4.22), the blow up of the solution takes place under the assumption
Z
|x|γ uλ0 (x) dx ≤ C4 .
Rd

Changing the variables and using the explicit form of λ we obtain the blow
up of solutions to (1.1)–(1.3) under the following assumption on the initial
condition
Z Z !1+ γ
γ d+2−α−β
−1+
|x|γ u0 (x) dx ≤ C4 M0 α+β−2−d u0 (x) dx .
Rd Rd

Proof of Corollary 2.4. We follow the proof of Theorem 2.3. In particular,

we choose Mγ so large that the inequality (4.21) holds true for all M0 > Mγ .
This leads to the blow up of the corresponding solution under the assumption
(4.22) imposed on the initial data.

17
 To complete the proof, we use the scaling argument again. By (2.1),
uλ (x, t) = λ2 u(λx, λ2 t) is a solution for every λ > 0. Note now that
Z Z Z Z
uλ0 (x) dx = u0 (x) dx and |x|γ uλ0 (x) dx = λ−γ |x|γ u0 (x) dx.
R2 R2 R2 R2
R
Hence, each initial data u0 ∈ L1 (R2 , (1 + |x|γ ) dx) satisfying R2 u0 (x) dx =
M0 > M γ leads to the blow up in a finite time of the corresponding solution
because the moment condition in (4.22) can be satisfied replacing u by uλ
and choosing λ large enough.

Acknowledgements. The preparation of this paper was partially sup-

ported by the Polish Ministry of Science grant N201 022 32/0902, the POLO-
NIUM project ÉGIDE no. 13886SG, and by the European Commission
Marie Curie Host Fellowship for the Transfer of Knowledge “Harmonic Anal-
ysis, Nonlinear Analysis and Probability” MTKD-CT-2004-013389. The
authors are greatly indebted to Tomasz Cieślak for pointing them out the
preprint [23].

Bibliography

1. P. Biler, Local and global solvability of parabolic systems modelling

chemotaxis, Adv. Math. Sci. Appl. 8 (1998), 715–743.

2. P. Biler, Existence and nonexistence of solutions for a model of gravi-

tational interaction of particles. III, Colloq. Math. 68 (1995), 229–239.

3. P. Biler, The Cauchy problem and self-similar solutions for a nonlinear

parabolic equation, Studia Math. 114 (1995), 181–205.

4. P. Biler, Radially symmetric solutions of a chemotaxis model in the

plane – the supercritical case, 31–42, in: Parabolic and Navier-Stokes
Equations, Banach Center Publications 81, Polish Acad. Sci., Warsaw,
2008.

5. P. Biler, L. Brandolese, On the parabolic-elliptic limit of the dou-

bly parabolic Keller–Segel system modelling chemotaxis, (2008), 1–21,
arXiv:0804.1000v1 [math.AP].

18
 6. P. Biler, M. Cannone, I. Guerra, G. Karch, Global regular and
singular solutions for a model of gravitating particles, Mathematische
Annalen 330 (2004), 693–708.

7. P. Biler, T. Funaki, W. A. Woyczyński, Interacting particle ap-

proximation for nonlocal quadratic evolution problems, Probab. Math.
Stat. 19 (1999), 267–286.

8. P. Biler, G. Karch, W. A. Woyczyński, Critical nonlinearity ex-

ponent and self-similar asymptotics for Lévy conservation laws, Ann.
Inst. H. Poincaré – Analyse non Linéaire 18 (2001), 613–637.

9. P. Biler, G. Karch, W. A. Woyczyński, Asymptotics for conserva-

tion laws involving Lévy diffusion generators, Studia Math. 148 (2001),
171–192.

10. P. Biler, W. A. Woyczyński, Global and exploding solutions for

nonlocal quadratic evolution problems, SIAM J. Appl. Math. 59 (1998),
845–869.

11. P. Biler, G. Wu, Two-dimensional chemotaxis models with fractional

diffusion, Math. Methods Appl. Sciences 32 (2009), 112–126;
DOI: 10.1002/mma.1036, 2008.

12. A. Blanchet, J. Dolbeault, B. Perthame, Two dimensional

Keller–Segel model: Optimal critical mass and qualitative properties
of the solutions, Electron. J. Diff. Eqns. 2006, 44, 1–33.

13. L. Brandolese, G. Karch, Far field asymptotics of solutions to con-

vection equation with anomalous diffusion, J. Evolution Equations 8
(2008), 307–326.

14. V. Calvez, B. Perthame, M. Sharifi tabar, Modified Keller–Segel

system and critical mass for the log interaction kernel, Contemporary
Math. 429, 45–62 (2007), Stochastic analysis and pde; Chen, Gui-Qiang
(ed.) et al., AMS.

19
 15. L. Corrias, B. Perthame, H. Zaag, Global solutions of some chemo-
taxis and angiogenesis systems in high space dimensions, Milan J. Math.
72 (2004), 1–29.

16. J. Droniou, C. Imbert, Fractal first order partial differential equa-

tions, Arch. Rat. Mech. Anal. 182 (2006), 299–331.

17. C. Escudero, The fractional Keller–Segel model, Nonlinearity 19

(2006), 2909–2918.

18. N. Jacob, Pseudo-differential Operators and Markov Processes, vol. 1:

Fourier analysis and semigroups, Imperial College Press, London, 2001.

19. W. Jäger, S. Luckhaus, On explosions of solutions to a system of

partial differential equations modelling chemotaxis, Trans. Amer. Math.
Soc. 329 (1992), 819–824.

20. G. Karch, Scaling in nolinear parabolic equations, J. Math. Anal.

Appl., 234 (1999), 534–558.

21. H. Kozono, Y. Sugiyama, Local existence and finite time blow-up of

solutions in the 2-D Keller-Segel system, J. Evol. Equ. 8 (2008), 353–
378.

22. P.-G. Lemarié-Rieusset, Recent Development in the Navier–Stokes

Problem, Chapman & Hall/CRC Press, Boca Raton, 2002.

23. D. Li, J. Rodrigo, X. Zhang, Exploding solutions for a nonlocal

quadratic evolution problem, 1–40, preprint on the webpage
http://www.warwick.ac.uk/staff/J.Rodrigo/research.html

24. Y. Meyer, Wavelets, paraproducts and Navier–Stokes equations, Cur-

rent developments in mathematics, 1996, Internat. Press, Cambridge,
MA 02238-2872 (1999).

25. T. Nagai, Behavior of solutions to a parabolic-elliptic system modelling

chemotaxis, J. Korean Math. Soc. 37 (2000), 721–732.

20

View publication stats