J. Math. Anal. Appl.

360 (2009) 537–547

Contents lists available at ScienceDirect

Journal of Mathematical Analysis and

Applications
www.elsevier.com/locate/jmaa

On the structure of global solutions of the nonlinear heat equation

in a ball
Thierry Cazenave a,b,∗ , Flávio Dickstein c,1 , Fred B. Weissler d
a
UPMC Univ Paris 06, UMR 7598, Laboratoire Jacques-Louis Lions, F-75005, Paris, France
b
CNRS, UMR 7598, Laboratoire Jacques-Louis Lions, F-75005, Paris, France
c
Instituto de Matemática, Universidade Federal do Rio de Janeiro, Caixa Postal 68530, 21944–970 Rio de Janeiro, R.J., Brazil
d
Université Paris 13, LAGA, UMR CNRS 7539, 99 Avenue J.-B. Clément, 93430 Villetaneuse, France

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we consider the nonlinear heat equation

Received 27 May 2008
Available online 4 July 2009 ut − u = |u |α u , (NLH)
Submitted by C.V. Pao
in the unit ball Ω of R N with Dirichlet boundary conditions, in the subcritical case. More
Keywords: precisely, we study the set G of initial values in C 0 (Ω) for which the resulting solution
Nonlinear heat equation of (NLH) is global. We obtain very precise information about a specific two-dimensional
Global solutions slice of G , which (necessarily) contains sign-changing initial values. As a consequence of
Sign-changing stationary solutions our study, we show that G is not convex. This contrasts with the case of nonnegative
initial values, where the analogous set G + is known to be convex.
© 2009 Elsevier Inc. All rights reserved.

1. Introduction

This paper is concerned with regular solutions of the equation


ut − u = |u |α u ,
(1.1)
u |∂Ω = 0,
where

Ω = B1, (1.2)
is the (open) unit ball of R , and N

0 < α < α , (1.3)

with

∞ if N = 1, 2,
α = 4 (1.4)
N −2
if N  3.
As is well known (see e.g. Theorem 5.2.1 in [3]) the initial value problem associated with (1.1) is locally well posed in
C 0 (Ω), the space of continuous functions on Ω that vanish on the boundary.

* Corresponding author at: UPMC, LJLL, B.C. 187, 4 place Jussieu, 75252 Paris Cedex 05, France.
E-mail addresses: thierry.cazenave@upmc.fr (T. Cazenave), flavio@labma.ufrj.br (F. Dickstein), weissler@math.univ-paris13.fr (F.B. Weissler).
1
Partially supported by CNPq (Brazil).

0022-247X/$ – see front matter © 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.jmaa.2009.06.076
538 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547

We denote by G the set of initial values in C 0 (Ω) for which the resulting solution of (1.1) is global in time. It is well
known that all global solutions are bounded (in time) in C 0 (Ω), where the bound depends only on the sup norm of the
initial value (see [11,4,8,12]). It follows easily that G is a closed subset of C 0 (Ω). Moreover, every global solution has an
ω-limit set made up of stationary solutions of (1.1), see [6]. In particular, we denote by G0 the set of initial values in C 0 (Ω)
for which the resulting solution of (1.1) is global in time and converges uniformly to 0 as t → ∞. It is well known that G0
is open and thus contains a neighborhood of 0. Furthermore, G0 is the interior of G (see Theorem 8 in [7]). On the other
hand, as far as we are aware, it is not known if G = G0 , or even if all stationary solutions belong to G0 . (See Remark 1.8.)
If we consider only nonnegative solutions of (1.1), then the corresponding set G + = G ∩ {u 0  0} is well understood. It
is known that G + is convex. Moreover, if ψ ∈ C 0 (Ω), ψ > 0, there exists λ > 0 such that if 0 < λ < λ then λψ ∈ G0 ; if
λ > λ then λψ ∈ / G ; and λ ψ ∈ G \ G0 . It follows that G + = G0 ∩ G + . In particular, the positive stationary solution belongs
to G0 ∩ G + . (See [10] and Section 19.2 in [13].)
For general solutions, not just positive solutions, the situation is different and more complicated. Indeed, in a recent
paper [2] the authors have shown that for N  3 and α sufficiently close to α , G is not star-shaped around 0 (and in
particular not convex). We do not know if this is true in general. At this stage we are unable to give a description of G
and G0 which is even remotely close to the above description of G + and G0 ∩ G + . We can, however, give a precise description
of G ∩ V and of G0 ∩ V , where V is a specific two-dimensional subspace of C 0 (Ω). Note that (since Ω is connected) any
two-dimensional subspace of C 0 (Ω) contains sign-changing functions. We prove in particular that G ∩ V is not convex, and
therefore that G is not convex. Similarly, G0 is not convex. Moreover, for certain values of α , we construct initial values ψ
such that the open set {λ > 0; λψ ∈ G0 } is not connected. It follows that G0 is not star-shaped around 0.
In order to state our results, we first introduce some notation.

Definition 1.1. We denote by ϕ the unique positive radially symmetric, stationary solution of (1.1) and by Ψ the unique
radially symmetric, stationary solution of (1.1) such that Ψ (0) > 0 and Ψ (r ) has exactly one zero in (0, 1).

We next define the two-dimensional subspace V of C 0 (Ω). Let

 
V = λΨ + − μΨ − ; (λ, μ) ∈ R2 , (1.5)

so that V is naturally identified in this fashion with R2 . The sets G ∩ V and G0 ∩ V are then identified with the sets
 
F = (λ, μ) ∈ R2 ; λΨ + − μΨ − ∈ G , (1.6)
and
 
O = (λ, μ) ∈ R2 ; λΨ + − μΨ − ∈ G0 , (1.7)
respectively. Our main result of this paper is the following.

Theorem 1.2. Under the hypotheses (1.3)–(1.4) and with the notation (1.6)–(1.7), there exist two continuous, increasing functions
μ, μ : [−1, 1] → [−1, 1] such that the sets F and O are given by
 
F = (λ, μ) ∈ [−1, 1]2 ; μ(λ)  μ  μ(λ) , (1.8)

and
 
O = (λ, μ) ∈ (−1, 1)2 ; μ(λ) < μ < μ(λ) . (1.9)

Moreover, the functions μ and μ have the following properties.

(i) μ(x) < μ(x) for all −1 < x < 1. In particular, μ(0) < 0 < μ(0).
(ii) μ(−1) = μ(−1) = −1 and μ(1) = μ(1) = 1.
(iii) μ(x) = −μ(−x) for all −1  x  1.
(iv) The functions μ, μ are right-differentiable at −1 and left-differentiable at 1 and satisfy μ (−1+ ) = μ (−1+ ) = μ (1− ) =
μ (1− ).

See Fig. 1 for a (schematic, not numerically generated) visualization of the functions μ(λ) and μ(λ) and the set O . In
Fig. 1, we have shown μ (1− ) > 1, which we know to be true for N  3 and α close to α . (See formula (4.1)). On the other
hand, if N = 1, then μ (1− ) = 1. (See Proposition 4.1).
Theorem 1.2 has a number of consequences. In what follows, we say that a solution u of (1.1) which blows up at the
finite time T blows up positively if u − (t ) remains bounded as t ↑ T and blows up negatively if u + (t ) remains bounded as
t ↑ T.
 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547 539

Fig. 1. The functions μ(λ) and μ(λ).

Corollary 1.3. Let λ, μ ∈ R and let u be the solution of (1.1) with the initial condition u (0) = λΨ + − μΨ − . With the notation of
Theorem 1.2, the following properties hold.

(i) If −1 < λ < 1 and μ(λ) < μ < μ(λ), then u is global and u (t ) → 0 uniformly as t → ∞.
(ii) If −1 < λ < 1 and μ = μ(λ), then u is global and u (t ) converges uniformly to the negative stationary solution of (1.1) as t → ∞.
(iii) If −1 < λ < 1 and μ = μ(λ), then u is global and u (t ) converges uniformly to the positive stationary solution of (1.1) as t → ∞.
(iv) If −1  λ  1 and μ > μ(λ), then u blows up negatively in finite time.
(v) If −1  λ  1 and μ < μ(λ), then u blows up positively in finite time.

Corollary 1.4. With the notation (1.5)–(1.7), the following properties hold.

(i) The sets G ∩ V and G0 ∩ V are connected and simply connected, and G0 ∩ V = G ∩ V .
(ii) The sets G0 ∩ V and G ∩ V are not convex.

We emphasize that it is unknown if G and G0 themselves are connected.

Corollary 1.5. The sets G0 and G are not convex.

Corollary 1.6. If u 0 = λΨ with |λ| > 1, then the corresponding solution of (1.1) blows up in finite time.

Lastly, Theorem 1.2 combined with the results of [2] gives the following corollary.

Corollary 1.7. If N  3 and α sufficiently close to α , then there exist ψ ∈ C 0 (Ω) and 0 < ρ1 < ρ2 < ρ3  ρ4 such that the following
properties hold.

(i) If 0 < λ < ρ1 then λψ ∈ G0 .

(ii) If λ = ρ1 , then λψ ∈ G and the corresponding solution of (1.1) converges to the negative stationary solution as t → ∞.
(iii) There exist ρ1  λ1 < λ2  ρ2 such that λψ ∈ / G for all λ1 < λ < λ2 . More precisely, the corresponding solution of (1.1) blows
up negatively in finite time.
(iv) If λ = ρ2 , then λψ ∈ G and the corresponding solution of (1.1) converges to the negative stationary solution as t → ∞.
540 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547

(v) If ρ2 < λ < ρ3 then λψ ∈ G0 .

(vi) If λ = ρ3 or λ = ρ4 , then λψ ∈ G and the corresponding solution of (1.1) converges to the positive stationary solution as t → ∞.
/ G . Moreover, if ρ4 < λ  1, then the corresponding solution of (1.1) blows up positively in finite time.
(vii) If λ > ρ4 , then λψ ∈

Remark 1.8. Since (1, 1) ∈ O (see Theorem 1.2), we see that Ψ ∈ G0 ∩ V ⊂ G0 . In fact, the property Ψ ∈ G0 holds when Ψ is
any spherically symmetric stationary solution of (1.1), i.e. with possibly more than one node. See the end of Section 1 in [1].
Whether or not any general stationary solution Ψ belongs to G0 seems to be an open question.

Remark 1.9. Corollary 1.4 tells us that the sets G0 ∩ V and G ∩ V (and therefore G0 and G ) are not convex. If N  3 and α
is sufficiently close to α , Corollary 1.7 gives us the stronger result that G0 ∩ V and G ∩ V (and therefore G and G0 ) are not
even star-shaped around 0. In cases not covered by Corollary 1.7 (including N = 1 and N = 2), we do not know if these sets
are star-shaped around 0.

Remark 1.10. Corollary 1.7 provides an example of a curious phenomenon where by the orientation of the instability
produced by multiplication of the initial value by λ is reversed. For clarity, let  ϕ = −ϕ be the unique negative radially
symmetric, stationary solution of (1.1). As is well known, the solution of (1.1) with initial value λ ϕ blows up negatively in
finite time if λ > 1, and is global and converges to 0 as t → ∞ if 0 < λ < 1. The same phenomenon is true for initial values
which are on the stable manifold of  ϕ and sufficiently close to  ϕ . More precisely, let u (t ) be a global solution of (1.1) which
converges to  ϕ as t → ∞. If t 0 is sufficiently large, then u (t 0 ) < 0 on Ω , and it follows that the solution of (1.1) with initial
value λu (t 0 ) blows up negatively in finite time if λ > 1, and is global and converges to 0 as t → ∞ if 0 < λ < 1. However,
this phenomenon need not be true for all t 0  0. To see this, consider the solution u (t ) of (1.1) with the initial condition
u (0) = ρ2 ψ as given by Corollary 1.7. It follows from Corollary 1.7(iv) that u (t ) lies on the stable manifold of  ϕ . On the
other hand, by Corollary 1.7(v) there exists ε > 0 such that if 1 < λ < 1 + ε then the solution of (1.1) with the initial value
λu (0) is global and converges to 0. Furthermore, by Corollary 1.7(iii) there exists 0 < λ < 1 such that the solution of (1.1)
with the initial value λu (0) blows up negatively in finite time. Thus a change in orientation with respect to multiplication
by λ occurs somewhere along the trajectory u (t ).

One might ask whether a version of Theorem 1.2 is true for sign-changing, radially symmetric stationary solutions with
more than one node. Indeed, many of the arguments in the proof of Theorem 1.2 can be adapted to this more general
situation. However, the argument used to prove that given by (2.15) below equals 1 does not seem to extend to the
general case. Thus for the moment we are unable to prove a version of Theorem 1.2 for other stationary solutions.
The rest of the paper is organized as follows. In Section 2, we prove all of Theorem 1.2 with the exception of the
last statement (iv), using comparison and energy arguments. In Section 3, we prove statement (iv) of Theorem 1.2 by
a linearization technique. Finally in Section 4 we prove the corollaries and some additional results.

2. Proof of Theorem 1.2(i), (ii) and (iii)

Although the proof of properties (i) and (ii) requires a certain amount of technicalities, the basic idea is relatively simple.
We therefore begin with a brief outline of the argument.
In order to construct the functions μ and μ, we define the open set
 
Λ = λ ∈ R; ∃μ ∈ R, (λ, μ) ∈ O . (2.1)
By energy considerations O , and therefore Λ, are bounded sets. Thus given λ ∈ Λ, the set
 
Iλ = μ ∈ R; (λ, μ) ∈ O , (2.2)
is a nonempty, open, bounded subset of R. We now define μ(λ) and μ(λ) by
−∞ < μ(λ) = inf I λ < sup I λ = μ(λ) < ∞. (2.3)
As we will see, F is a closed set, so that
   
λ, μ(λ) ∈ F \ O, λ, μ(λ) ∈ F \ O, (2.4)
for all λ ∈ Λ. We will show, using comparison arguments, that the solution of (1.1) whose initial value corresponds by (1.5)
to (λ, μ(λ)) converges to the negative stationary solution as t → ∞ and that the solution of (1.1) likewise corresponding to
(λ, μ(λ)) converges to the positive stationary solution. Let Λ be any nonempty, connected component of Λ, and set

= sup Λ .
Comparison arguments will be used to show that

lim μ(λ) = lim μ(λ) := ν .

λ↑ λ↑
 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547 541

The key point will then be to determine the ω -limit set of the solution of (1.1) corresponding to the point ( , ν ) ∈ F \ O .
By the above considerations, continuous dependence and nonincrease of the zero number, this must be a one-node sign-
changing stationary solution of (1.1), i.e. ±Ψ . Since the energy functional restricted to V achieves its global maximum
precisely at the points ±Ψ and ±|Ψ |, and since ±|Ψ | are not stationary solutions, it must be that ( , ν ) = ±(1, 1). In
particular, = ±1. The same analysis works for the left endpoint of Λ , which implies in fact Λ = (−1, 1). Since Λ an
arbitrary nonempty, connected component of Λ, necessarily Λ = (−1, 1). With this information, properties (i) and (ii) will
easily follow.
We now begin work and we first introduce some notation.

Definition 2.1. Given μ, λ ∈ R, set

= λΨ + − μΨ − ,
λ,μ
u0 (2.5)
λ,μ
where Ψ is given by Definition 1.1, and let u λ,μ be the solution of (1.1) with the initial condition u λ,μ (0) = u 0 . We denote
by T λ, μ ∈ (0, ∞] the maximal existence time of u .
λ, μ

−λ,−μ λ,μ
Remark 2.2. Note that u 0 = −u 0 , so that u −λ,−μ (t ) ≡ −u λ,μ (t ). This shows that O and F are symmetric with respect
to (0, 0).

Lemma 2.3. F is a closed, bounded nonempty subset of R2 and O is an open, bounded, nonempty subset of R2 .

Proof. It is clear that (0, 0) ∈ O ⊂ F , so that F and O are nonempty. Next, given u ∈ H 01 (Ω), set
 
1 1
E (u ) = |∇ u |2 − | u |α + 2 . (2.6)
2 α+2
Ω Ω

If u (t ) is a nonstationary solution of (1.1), then E (u (t )) is a decreasing function of t. It is well known (see [9]) that if
λ,μ
E (u 0 ) < 0, then the solution of (1.1) such that u (0) = u 0 blows up in finite time. Since E (u 0 ) → −∞ as |λ| + |μ| → ∞, we
deduce that F (hence O ) is bounded. Furthermore, it is well known that 0 is an asymptotically stable stationary solution
of (1.1), so that O is open. Finally, to see that F is closed, recall (see [12]) that if (λ, μ) ∈ F , then

sup u λ,μ (t ) L∞
 C < ∞,
t 0

λ,μ λ,μ
where the constant C depends on u 0 L∞ . Since F is bounded, we see that u 0 L∞ is bounded, so that

sup sup u λ,μ (t ) L∞

< ∞. (2.7)
(λ,μ)∈F t 0

The closedness of F easily follows by standard arguments. 2

Lemma 2.4. The following properties hold, where ϕ is given by Definition 1.1.

(i) If (λ, μ) ∈ F \ O and (λ, μ) = ±(1, 1), then u λ,μ (t ) → ±ϕ in C 1 (Ω) as t → ∞.

(ii) Let (λn )n1 , (μn )n1 ⊂ R satisfy (λn , μn ) ∈ F for all n  1. Suppose u λn ,μn (t ) → ϕ (respectively, u λn ,μn (t ) → −ϕ ) as t → ∞
for all n  1 and (λn , μn ) → (λ, μ). If (λ, μ) = ±(1, 1), then u λ,μ (t ) → ϕ (respectively, u λ,μ (t ) → −ϕ ) as t → ∞.

Proof. We first prove property (i). Let E be defined by (2.6) and let ρ ∈ (0, 1) be the zero of Ψ . We claim that
 λ,μ 
E u0 < E (Ψ ) (2.8)

for all (λ, μ) = (±1, ±1). Indeed, note that v = [Ψ ]|{|x|<ρ } ∈ H 01 ({|x| < ρ }) satisfies − v = | v |α v in {|x| < ρ }. Similarly,
w = [Ψ ]|{ρ <|x|<1} ∈ H 01 ({ρ < |x| < 1}) satisfies − w = | w |α w in {ρ < |x| < 1}. It easily follows that

  λ2 |λ|α +2
E λΨ + = − |∇Ψ |2 ,
2 α+2
{|x|<ρ }

  μ 2
|μ|α +2
E μΨ − = − |∇Ψ |2 .
2 α+2
{ρ <|x|<1}
542 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547

Consequently,
 
 λ,μ  λ2 |λ|α +2 μ2 |μ|α +2
E u0 = − |∇Ψ |2 + − |∇Ψ |2 .
2 α+2 2 α+2
{|x|<ρ } {ρ <|x|<1}

On the one hand,


1 1
E (Ψ ) = − |∇Ψ |2 ,
2 α+2
Ω
α +2
and on the other hand, the function s → s2
2
− |sα|+2 achieves a strict maximum at s = ±1. The claim follows.
λ,μ
Next, note that if (λ, μ) = ±(1, −1), then u 0 is not a stationary solution of (1.1). Thus
   λ,μ 
E u λ,μ (t ) < E u 0 = E (Ψ ), (2.9)

for all 0 < t < T λ,μ .

Suppose (λ, μ) ∈ F \ O and (λ, μ) = ±(1, 1). It follows from (2.7) that u λ,μ (t ) is bounded uniformly in t  0. Therefore,
u λ, μ has (for the uniform topology) a nonempty, compact, connected ω -limit set ωλ,μ , which is made up of stationary
λ,μ
solutions of (1.1). (See [6].) Moreover, it follows from the asymptotic stability of 0 that 0 ∈ / ωλ,μ . Note that u 0 is radially
λ,μ
symmetric. Considered as a function of r ∈ (0, 1), u 0 has at most one zero. Therefore, u λ,μ (t ), considered as a function
of r ∈ (0, 1) has at most one zero for all 0 < t < T λ,μ (see [5]). It follows that any element of ωλ,μ is a nontrivial, radially
symmetric, stationary solution of (1.1) with at most one zero. Such solutions are ±ϕ and ±Ψ . Since (λ, μ) = ±(1, 1),
formulas (2.8) or (2.9) and the fact that E (u λ,μ (t )) is nonincreasing with respect to t imply that ±Ψ ∈ / ωλ,μ . Thus ωλ,μ ⊂
{ϕ , −ϕ }. Since ω λ, μ is connected, we deduce that ω λ, μ = ±{ϕ }. Property (i) easily follows by the C smoothing effect.
1

We finally prove property (ii). Assume u λn ,μn (t ) → ϕ , the other case being similar. Note that by property (i), either
u λ,μ (t ) → 0 or u λ,μ (t ) → ±ϕ . In the first case, (λ, μ) ∈ O . Since O is open, we would have (λn , μn ) ∈ O for all large n,
which is absurd. Suppose next that u λ,μ (t ) → −ϕ and let τ > 0 be such that u λ,μ (τ )  −ϕ /2. Since (λn , μn ) → (λ, μ), it
follows by continuous dependence that u λn ,μn (τ ) < 0 for all sufficiently large n. Therefore, u λn ,μn (t ) < 0 for all sufficiently
large n and all t  τ . This is again absurd, since u λn ,μn (t ) → ϕ as t → ∞. Thus the only remaining possibility is that
u λ,μ (t ) → ϕ , which is the desired conclusion. 2

We next give some comparison properties for solutions of (1.1).

Proposition 2.5. Let u 0 , v 0 ∈ G and let u , v be the corresponding solutions of (1.1). Suppose that u (t ) → ψ as t → ∞ where ψ is a
stationary solution of (1.1). If v 0  u 0 , v 0 ≡ u 0 and if ψ + ≡ 0, then v blows up positively in finite time. Similarly, if v 0  u 0 , v 0 ≡ u 0
and if ψ − ≡ 0, then v blows up negatively in finite time.

Proof. We only prove the first statement. The fact that v blows up in finite time follows immediately from Lemma 4 in [7].
Since v (t )  u (t ), we see that v − (t ) L ∞  u − (t ) L ∞ . Thus v − (t ) L ∞ remains bounded and v must blow up positively. 2

Corollary 2.6. Let u 0 , v 0 ∈ G and let u , v be the corresponding solutions of (1.1). Suppose further that u 0  v 0 and u 0 ≡ v 0 . Assume
that u (t ) → ψ and v (t ) → ψ  as t → ∞ where ψ, ψ  are stationary solutions of (1.1). Then the following properties hold.

(i) Either ψ = ψ = 0, or ψ = 0 and ψ  > 0, or ψ < 0 and ψ  = 0 or ψ < 0 and ψ  > 0.

(ii) If w 0 ∈ G satisfies u 0  w 0  v 0 , w 0 ≡ u 0 and w 0 ≡ v 0 , then w 0 ∈ G0 .

Proof. It follows from Proposition 2.5 that ψ + = 0 and, similarly, ψ − = 0. This proves property (i). We now prove prop-
erty (ii) and we denote by w the solution of (1.1) with the initial value w 0 . Note that u (t ) < w (t ) < v (t ), so that w is a
global solution. Therefore, it has an ω -limit set ω( w ) made of classical solutions of (1.1) (see [6]). Assume that there exists
ψ  ∈ ω( w ), ψ  ≡ 0, and suppose for example that ψ + = 0. It then follows from Proposition 2.5 that v blows up in finite
time, which is absurd. This completes the proof. 2

Proof of Theorem 1.2(i), (ii) and (iii). We use the notation given at the beginning of the section. Based on Lemma 2.3, the
entire discussion through formula (2.4) is completely justified. In addition, it follows from Corollary 2.6(ii) that I λ is an
interval, so that
 
O= {λ} × μ(λ), μ(λ) . (2.10)
λ∈Λ
 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547 543

We now consider the global solutions u λ,μ(λ) (t ) and u λ,μ(λ) (t ) of (1.1). By formula (2.4) and Lemma 2.4(i), these solutions
λ,μ(λ) λ,μ(λ) λ,μ(λ) λ,μ(λ)
each converge to a nontrivial stationary solution. Since u 0  u0 and u 0 ≡ u 0 , it follows from Corollary 2.6(i)
that

u λ,μ(λ) (t ) → ϕ and u λ,μ(λ) (t ) → −ϕ , (2.11)

uniformly as t → ∞.
We now prove that

Λ = (−1, 1), (2.12)

and

lim μ(λ) = lim μ(λ) = −1, lim μ(λ) = lim μ(λ) = 1. (2.13)
λ↓−1 λ↓−1 λ↑1 λ↑1

Note first that

1∈
/ Λ. (2.14)

Indeed, it follows from (2.10) and (2.11) that if λ ∈ Λ then for all μ ±ϕ
such that u λ,μ is global, u λ,μ (t ) converges to 0 or
as t → ∞. Since u 1,1 (t ) ≡ Ψ we see that (2.14) holds. Let now Λ be a nonempty, connected component of Λ, and set

= sup Λ . (2.15)

Since Λ is open, we see that / Λ . Λ being a connected component of Λ, we see that in fact
∈

∈
/ Λ. (2.16)

Let (λn )n1 ⊂ Λ satisfy λn ↑ as n → ∞. Set μn = μ(λn ) and μn = μ(λn ). By possibly extracting a subsequence, we may
assume that there exist

ν  ν, (2.17)

such that

μn −→ ν , μn −→ ν . (2.18)
n→∞ n→∞

Since F is closed, we see that ( , ν ), ( , ν ) ∈ F . We claim that

ν = ν := ν . (2.19)

Indeed, otherwise it follows from (2.11), Lemma 2.4(ii) and Corollary 2.6(ii) that ( , μ) ∈ O for all ν < μ < ν . In particular,
∈ Λ, which contradicts (2.16) and proves the claim (2.19). We next claim that

= ν = ±1. (2.20)

Indeed, suppose that ( , ν ) = ±(1, 1). It follows from (2.11), (2.18), (2.19) and Lemma 2.4(ii) that u ,ν (t )
→ ϕ and that
u ,ν (t ) → −ϕ as t → ∞. This is absurd and proves (2.20). Since the sequence (λn )n1 is arbitrary, we see that in fact,

= lim μ(λ) = lim μ(λ) = ±1. (2.21)

λ↑ λ↑

Setting  = inf Λ , one shows similarly that


= lim μ(λ) = lim μ(λ) = ±1. (2.22)
λ↓ λ↓

Since  < , we deduce in particular from (2.21)–(2.22) that  = −1 and = 1. Finally, since Λ is an arbitrary connected
component of Λ, we see that Λ = Λ = (−1, 1). The property (2.13) is precisely (2.21)–(2.22). We extend the functions μ
and μ to the closed interval [−1, 1] by setting μ(−1) = μ(−1) = −1 and μ(1) = μ(1) = 1.
We now prove that the functions μ and μ are increasing (−1, 1) → R. To see this, let λ ∈ (−1, 1), so that

u λ,μ(λ) (t ) → −ϕ as t → ∞ by formula (2.11). Given any −1 < λ < λ and μ  μ(λ), we see that u λ0 ,μ  u λ,
0
μ(λ)
and
λ ,μ λ,μ(λ) 
u 0 ≡ u 0 , so that the corresponding solution of (1.1) blows up in finite time (see Proposition 2.5). Thus μ(λ ) < μ(λ),
which shows that μ is increasing. One shows similarly that μ is increasing.
544 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547

We next show that the functions μ and μ are continuous [−1, 1] → R. Continuity at ±1 follows from (2.13). To show
continuity on (−1, 1), suppose by contradiction that there exists −1 < λ < 1 such that

μ0 =: lim μ(λ) < lim μ(λ) =: μ1 . (2.23)

λ↑λ λ↓λ
  λ ,μ0 λ ,μ1
It follows from formula (2.11) and Lemma 2.4(ii) that u λ ,μ0 (t ), u λ ,μ1 (t ) → −ϕ as t → ∞. Since u 0  u0 and
λ ,μ0 λ ,μ1
u0 ≡ u 0, this is impossible (see Proposition 2.5). This shows that μ is continuous, and one shows similarly that
μ is continuous.
At this stage, we have constructed the continuous, increasing functions μ, μ : [−1, 1] → R and we have established
Properties (i) and (ii). Property (iii) immediately follows from Remark 2.2. Also, formula (1.9) is a consequence of (2.10)
and (2.12).
It remains to prove (1.8). In view of (1.9), this amounts to showing that F = O . Since clearly F ⊃ O , we need only show
that F ⊂ O . To see this, note first that ±(1, 1) ∈ O by (2.10) and (2.13). Now let (λ, μ) ∈ F \ O , (λ, μ) = ±(1, 1). It follows
from Lemma 2.4(i) that u λ,μ (t ) → ±ϕ in C 1 (Ω) as t → ∞. Suppose for example that u λ,μ (t ) → ϕ . Since convergence holds
in C 1 (Ω), we see that there exists τ > 0 such that
1
u λ,μ (τ )  ϕ. (2.24)
2

By continuous dependence, we deduce that there exists ε > 0 such that if |λ − λ|  ε , then T λ ,μ > τ and

u λ ,μ (τ ) > 0. (2.25)
λ ,μ λ,μ λ ,μ λ,μ
If we now consider λ − ε  λ < λ, we see in particular that u 0  u0 and u 0 ≡ u 0 , so that
λ ,μ
u (t ) < u λ,μ (t ), (2.26)

for all t > 0. We easily deduce from (2.25)–(2.26) that the solution u λ ,μ is global and that

0 < u λ ,μ (t ) < u λ,μ (t ), (2.27)
for all t  τ . Thus in particular (λ , μ) ∈ F . It is clear by (2.27) that (λ , μ) = ±(1, 1). Thus we deduce from Lemma 2.4(i)
 
that either u λ ,μ (t ) → 0 or else u λ ,μ (t ) → ϕ as t → ∞. The last case is ruled out by Proposition 2.5, so that in fact
 ,μ
u λ (t ) → 0, i.e. (λ , μ) ∈ O . Letting λ ↑ λ, we conclude that (λ, μ) ∈ O . This completes the proof. 2


3. Proof of Theorem 1.2(iv)

We consider the self-adjoint operator − − (α + 1)|Ψ |α on L 2 (Ω) with domain H 2 (Ω) ∩ H 01 (Ω), and denote by
 
λ1 = λ1 − − (α + 1)|Ψ |α , (3.1)
its first eigenvalue and by ϕ1 the corresponding eigenvector, i.e.
−ϕ1 − (α + 1)|Ψ |α ϕ1 = λ1 ϕ1 , (3.2)
normalized by the conditions

ϕ1 > 0, ϕ1 L 2 (Ω) = 1. (3.3)

Given 0  θ  π /2, we set

 
f (θ) = ϕ1 Ψ + cos θ − Ψ − sin θ . (3.4)
Ω
Since ϕ1 > 0, we see that f is a decreasing function of θ , with f (0) > 0 and f (π /2) < 0. Therefore, there exists a unique
0 < θ0 < π /2 such that

f (θ0 ) = 0. (3.5)
Given 0 < θ < π /2 and 0 < ε < 1, we set

λε = 1 − ε cos θ ∈ (0, 1), με = 1 − ε sin θ ∈ (0, 1). (3.6)

Next, we let

ε zε (t ) = Ψ − u λε ,με (t ), (3.7)
 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547 545

for 0  t < T λε ,με , so that

⎧
⎨ ε 1  α  
zt =  zε + |Ψ |α Ψ − Ψ − ε zε  Ψ − ε zε ,
ε (3.8)
⎩ ε
z (0) = Ψ + cos θ − Ψ − sin θ.
Let z be the solution of

zt −  z − (α + 1)|Ψ |α z = 0,
(3.9)
z(0) = Ψ + cos θ − Ψ − sin θ.
It follows by continuous dependence that

T λε ,με −→ ∞, (3.10)
ε ↓0

and that for all 0 < δ < T < ∞,

zε → z, (3.11)

in C ([δ, T ], C (Ω)). We now fix 0 < θ < θ0 , so that

1


a1 := z(0)ϕ1 > 0, (3.12)

by (3.5). The solution z of (3.9) can be expressed as an expansion in terms of the eigenfunctions of the operator
− − (α + 1)|Ψ |α . Formula (3.12) says that the leading coefficient of this expansion for z(0) is positive, so that the leading
term in the expansion of the solution is a1 e −λ1 t ϕ1 . Therefore, there exists t 0 > 0 such that

z(t 0 )  cdΩ , (3.13)

where c > 0 and dΩ is the distance to ∂Ω . (See e.g. Lemma A.1 in [2].) By applying (3.11), (3.13) and (3.7), we conclude
that there exist 0 < ε0 < 1 such that

u λε ,με (t 0 ) < Ψ, (3.14)

for all 0 < ε < ε0 . We deduce from (3.14) and Proposition 2.5 that u λε ,με blows up in finite time. Moreover, it follows
from (3.14) that u λε ,με blows up while staying (for t close to the blow-up time) below Ψ . By comparison, we also have
T λε ,μ < ∞ for μ  με . This implies that μ(λε ) < με for all 0 < ε < ε0 . Since με = 1 − (1 − λε ) tan θ by (3.6), we see that
μ(λε ) < 1 − (1 − λε ) tan θ . Therefore, for all 0 < θ < θ0 , there exists 0 < λθ < 1 such that

μ(λ) < 1 − (1 − λ) tan θ, (3.15)

for all λθ < λ < 1. One shows similarly that for all θ0 < θ < π /2, there exists 0 < λθ < 1 such that

μ(λ) > 1 − (1 − λ) tan θ, (3.16)

for all λθ < λ < 1. It easily follows from (3.15)–(3.16) that the functions μ and μ are left-differentiable at λ = 1 and that

 −
 
 −

μ 1 =μ 1 = tan θ0 . (3.17)

Similarly,
   
μ −1+ = μ −1+ = tan θ0 . (3.18)

This completes the proof.

4. Further results and proofs of the corollaries

Proposition 4.1. If N = 1, then the sets F and O are symmetric with respect to the axis {λ = μ}. In particular, μ (−1+ ) = μ (−1+ ) =
μ (1− ) = μ (1− ) = 1.

μ,λ λ,μ
Proof. The function Ψ is symmetric about x = 1/2, i.e. Ψ (x) = −Ψ (1 − x); and so u 0 (x) = −u 0 (1 − x). Therefore,
(λ, μ) ∈ F if and only if (μ, λ) ∈ F and (λ, μ) ∈ O if and only if (μ, λ) ∈ O . 2
546 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547

Remark 4.2. Since μ(0) < 0 and μ(1) = 1, there exists a unique 0 < λ0 < 1 such that μ(λ0 ) = 0. It easily follows from
the convexity of the mapping u → |u |α u on [0, ∞) that μ is convex (and negative) on (0, λ0 ). Similarly, the function μ is
concave (and positive) on (−λ0 , 0).

Proof of Corollary 1.3. Part (i) follows from formula (1.9). Parts (ii) and (iii) are consequences of Lemma 2.4(i) and parts (iv)
and (v) follow from Proposition 2.5. 2

Proof of Corollary 1.4. Part (i) follows from formulas (1.8) and (1.9). Part (ii) is an immediate consequence of statement (iv)
of Theorem 1.2. 2

Proof of Corollary 1.5. This is an immediate consequence of Corollary 1.4(ii). 2

Proof of Corollary 1.6. This is immediate, since F ⊂ [−1, 1]2 by formula (1.8). 2

Proof of Corollary 1.7. By Proposition 2.1 in [2], we know that, if α is sufficiently close to α , then f (π /4) > 0 where f is
the function defined by (3.4). It follows that π /4 < θ0 < π /2, where θ0 is given by (3.5). Thus, by formula (3.17),
   
μ 1− = μ 1− > 1. (4.1)
The result in then obvious by looking at Fig. 1. We make this precise as follows. We deduce from (4.1) that there exist
0 < δ, ε < 1 such that

μ(λ)  λ − ε(1 − λ), (4.2)

for all 1 − δ  λ  1. We now let

ψ = Ψ + − γ Ψ −, (4.3)
where

γ = 1 − εδ ∈ (0, 1). (4.4)

Note that by (4.2), μ(1 − δ)  1 − δ − εδ < 1 − δ − εδ + εδ = γ (1 − δ), i.e.
2

μ(1 − δ) < γ (1 − δ). (4.5)

It follows that there exists 0 < λ1 < λ2 < 1 such that

μ(λ) < γ λ, (4.6)

μ(0) > 0 and μ(1) = 1 > γ , we may define
for all λ1 < λ < λ2 . Since
 
ρ1 = inf 0  λ  1; μ(λ) = γ λ ∈ (0, λ1 ], (4.7)
 
ρ2 = sup 0  λ  1; μ(λ) = γ λ ∈ [λ2 , 1). (4.8)
Finally, since μ(1) = 1 > γ and μ(ρ2 ) < μ(ρ2 ) = γρ2 , we define
 
ρ3 = inf ρ2  λ  1; μ(λ) = γ λ ∈ (ρ2 , 1), (4.9)
 
ρ4 = sup ρ2  λ  1; μ(λ) = γ λ ∈ [ρ3 , 1). (4.10)
It is straightforward to see that ρ1 , ρ2 , ρ3 , ρ4 , λ1 , λ2 satisfy the properties in Corollary 1.7. 2

References

[1] T. Bartsch, P. Poláčik, P. Quittner, Liouville-type theorems and asymptotic behavior of nodal radial solutions of semilinear heat equations, J. Eur. Math.
Soc. (JEMS), in press.
[2] T. Cazenave, F. Dickstein, F.B. Weissler, Sign-changing stationary solutions and blow up for the nonlinear heat equation in a ball, Math. Ann. 344 (2)
(2009) 431–449.
[3] T. Cazenave, A. Haraux, An Introduction to Semilinear Evolution Equations, Oxford Lecture Ser. Math. Appl., vol. 13, Oxford University Press, Oxford,
1998.
[4] T. Cazenave, P.-L. Lions, Solutions globales d’équations de la chaleur semi linéaires, Comm. Partial Differential Equations 9 (1984) 955–978.
[5] X.-Y. Chen, P. Poláčik, Asymptotic periodicity of positive solutions of reaction diffusion equations on a ball, J. Reine Angew. Math. 472 (1996) 17–51.
[6] C.M. Dafermos, Asymptotic behavior of solutions of evolution equations, in: M.G. Crandall (Ed.), Nonlinear Evolution Equations, Academic Press, New
York, 1978, pp. 103–123.
[7] F. Gazzola, T. Weth, Finite-time blow-up and global solutions for semilinear parabolic equations with initial data at high energy levels, Differential
Integral Equations 18 (9) (2005) 961–990.
 T. Cazenave et al. / J. Math. Anal. Appl. 360 (2009) 537–547 547

[8] Y. Giga, A bound for global solutions of semilinear heat equations, Comm. Math. Phys. 103 (1986) 415–421.
[9] H.A. Levine, Some nonexistence and instability theorems for formally parabolic equations of the form P ut = − Au + f (u ), Arch. Ration. Mech. Anal. 51
(1973) 371–386.
[10] P.-L. Lions, Asymptotic behavior of some nonlinear heat equations, nonlinear phenomena, Phys. D 5 (1982) 293–306.
[11] W.M. Ni, P.E. Sacks, J. Tavantzis, On the asymptotic behavior of solutions of certain quasilinear parabolic equations, J. Differential Equations 54 (1984)
97–120.
[12] P. Quittner, A priori bounds for global solutions of a semilinear parabolic problem, Acta Math. Univ. Comenian. (N.S.) 68 (2) (1999) 195–203.
[13] P. Quittner, P. Souplet, Superlinear Parabolic Problems. Blow-up, Global Existence and Steady States, Birkhäuser Adv. Texts, Birkhäuser Verlag, Basel,
2007.