Global nullcontrollability and nonnegativecontrollability of slightly superlinear heat equations
Abstract.
We consider the semilinear heat equation posed on a smooth bounded domain of with Dirichlet or Neumann boundary conditions. The control input is a source term localized in some arbitrary nonempty open subset of . The goal of this paper is to prove the uniform large time global nullcontrollability for semilinearities where which is the case left open by Enrique FernandezCara and Enrique Zuazua in 2000. It is worth mentioning that the free solution (without control) can blowup. First, we establish the smalltime global nonnegativecontrollability (respectively nonpositivecontrollability) of the system, i.e., one can steer any initial data to a nonnegative (respectively nonpositive) state in arbitrary time. In particular, one can act locally thanks to the control term in order to prevent the blowup from happening. The proof relies on precise observability estimates for the linear heat equation with a bounded potential . More precisely, we show that observability holds with a sharp constant of the order for nonnegative initial data. This inequality comes from a new Carleman estimate. A KakutaniLeraySchauder’s fixed point argument enables to go back to the semilinear heat equation. Secondly, the uniform large time nullcontrollability result comes from three ingredients: the global nonnegativecontrollability, a comparison principle between the free solution and the solution to the underlying ordinary differential equation which provides the convergence of the free solution toward in norm, and the local nullcontrollability of the semilinear heat equation.
Contents
 1 Introduction
 2 Main results
 3 Parabolic equations: Wellposedness and regularity

4 Global nonnegativecontrollability of the linear heat equation with a bounded potential
 4.1 Statement of the result
 4.2 A precise  observability inequality for the linear heat equation with bounded potential and nonnegative initial data
 4.3 A new Carleman estimate
 4.4 Proof of the  observability inequality: creftypecap 4.4
 4.5 Proof of the linear global nonnegativecontrollability: creftypecap 4.1
 5 A fixedpoint argument to prove the smalltime nonlinear global nonnegative controllability
 6 Application of the global nonnegativecontrollability to the large time global nullcontrollability
 7 Dirichlet boundary conditions
 8 Comments
1. Introduction
Let , , be a bounded, connected, open subset of of class and be the outer unit normal vector to . We consider the semilinear heat equation with Neumann boundary conditions:
(1) 
where .
Remark 1.1.
All our results stay valid for Dirichlet boundary conditions (see Section 7).
In (1), is the state to be controlled and is the control input supported in , a nonempty open subset of .
We assume that satisfies
(2) 
In this case, solves (1) with and .
In the following, we will also assume that satisfies the restrictive growth condition
(3) 
Under the hypothesis (3), blowup may occur if in (1). Take for example with . The mathematical theory of blowup for
(4) 
was established in [24] and [25]. It was shown that blowup

occurs globally in the whole domain if ,

is of pointwise nature if ,

is ’regional’, i.e., it occurs in an open subset of if .
See [26, Section 2 and Section 5] for a survey on this problem.
The goal of this paper is to analyze the nullcontrollability properties of (1).
Let us define . We recall two classical definitions of nullcontrollability.
Definition 1.2.
We have the following wellknown local nullcontrollability result.
Theorem 1.3.
For every , (1) is locally nullcontrollable in time .
The proof of creftypecap 1.3 is a consequence of the (global) nullcontrollability of the linear heat equation with a bounded potential (due to Andrei Fursikov and Oleg Imanuvilov, see [23] or [21, Theorem 1.5]) and the small perturbations method (see [3, Lemma 6] and [1], [5], [30], [33], [40] for other results in this direction).
The following global nullcontrollability (positive) result has been proved independently by Enrique FernandezCara, Enrique Zuazua (see [22, Theorem 1.2]) and Viorel Barbu under a sign condition (see [4, Theorem 2] or [6, Theorem 3.6]) for Dirichlet boundary conditions. It has been extended to semilinearities which can depend on the gradient of the state and to Robin boundary conditions (then to Neumann boundary conditions) by Enrique FernandezCara, Manuel GonzalezBurgos, Sergio Guerrero and JeanPierre Puel in [19] (see also [13] for the Dirichlet case).
Theorem 1.4.
Remark 1.5.
The following global nullcontrollability (negative) result has been proved by Enrique FernandezCara, Enrique Zuazua (see [22]).
Theorem 1.6.
Remark 1.7.
Such a function does satisfy (3) for any because as . Then, creftypecap 1.6 shows that (1) can fail to be nullcontrollable for every under the hypothesis (3) with . creftypecap 1.6 comes from a localized estimate in that shows that the control cannot compensate the blowup phenomena occurring in (see [22, Section 2]).
When the nonlinear term is dissipative, i.e., for every , then blowup cannot occur. Furthermore, such a nonlinearity produces energy decay for the uncontrolled equation, therefore naively one may be led to believe that it can help in steering the solution to zero in arbitrary short time. The results of Sebastian Anita and Daniel Tataru show that this is false, more precisely that for ‘strongly’ superlinear one needs a sufficiently large time in order to bring the solution to zero. An intuitive explanation for this is that the nonlinearity is also damping the effect of the control as it expands from the controlled region into the uncontrolled region (see [3]).
Theorem 1.8.
Remark 1.9.
In particular, for such a as in creftypecap 1.8, (1) is not globally nullcontrollable in small time . creftypecap 1.8 is due to pointwise upper bounds on the solution of (1) which are independent of the control (see [3, Section 3]).
2. Main results
2.1. Smalltime global nonnegativecontrollability
We introduce a new concept of controllability.
Definition 2.1.
The first main result of this paper is a smalltime global nonnegativecontrollability result for (1).
Theorem 2.2.
Remark 2.3.
creftypecap 2.2 is almost sharp because it does not hold for according to creftypecap 1.8. The case where as is open.
Remark 2.4.
creftypecap 2.2 does not treat the case with because of the sign condition.
2.2. Large time global nullcontrollability
The second main result of this paper is the following one.
Theorem 2.5.
Remark 2.6.
creftypecap 2.5 proves that creftypecap 1.6 is almost sharp. Indeed, let us take with , then by creftypecap 2.5, there exists sufficiently large such that (1) is globally nullcontrollable in time . In particular, one can find a localized control which prevents the blowup from happening. The case is open.
Remark 2.7.
creftypecap 2.5 does not treat the case with because of the sign condition.
2.3. Proof strategy of the smalltime global nonnegativecontrollability
We will only prove the global nonnegativecontrollability result. The nonpositivecontrollability result is an easy adaptation.
The proof strategy of creftypecap 2.2 will follow Enrique FernandezCara and Enrique Zuazua’s proof of creftypecap 1.4 (see [22]).
The starting point is to get some precise observability estimates for the linear heat equation with a bounded potential for nonnegative initial data. More precisely, we show that observability holds with a sharp constant of the order for nonnegative initial data (see creftypecap 4.4 below). This is done thanks to a new Carleman estimate in (see creftypecap 4.9 below). This leads to a nonnegativecontrollability result in in the linear case with an estimate of the control cost of the order which is the key point of the proof (see creftypecap 4.1 below).
We end the proof of creftypecap 2.2 by a KakutaniLeraySchauder’s fixedpoint strategy. The idea of taking short control times to avoid blowup phenomena is the same as in [22] and references therein. More precisely, the construction of the control follows two steps. The first step consists in steering the solution of (1) to in time with an appropriate choice of the control. Then, the two conditions: and the dissipativity of in imply that the free solution of (1) (with ) defined in stays nonnegative and bounded by using a comparison principle (see Section 5).
2.4. Proof strategy of the large time global nullcontrollability
We will only treat the case where for . The other case, i.e., for is an easy adaptation.
The proof strategy of creftypecap 2.5 is divided into three steps.
First, for every initial data , one can steer the solution of (1) in time (for instance) to a nonnegative state by using creftypecap 2.2.
Secondly, we let evolve the system without control and we remark that
with independent of and when . This kind of argument has already been used by JeanMichel Coron in the context of the Burgers equation (see [10, Theorem 8]).
Finally, by using the second step, for sufficiently large, belongs to a small ball of centered at , where the local nullcontrollability holds (see creftypecap 1.3). Then, one can steer to with an appropriate choice of the control.
3. Parabolic equations: Wellposedness and regularity
The goal of this section is to state wellposedness results, dissipativity in time in norm, maximum principle and  estimates for linear parabolic equations. We also give the definition of a solution to the semilinear heat equation (1). The references of these results only treat the case of Dirichlet boundary conditions but the proofs can be easily adapted to Neumann boundary conditions.
3.1. Wellposedness
We introduce the functional space
(6) 
which satisfies the following embedding (see [15, Section 5.9.2, Theorem 3])
(7) 
Linear parabolic equations
Definition 3.1.
Let , and . A function is a solution to
(8) 
if for every ,
(9) 
and
(10) 
The following wellposedness result in holds for linear parabolic equations.
Proposition 3.2.
Let , and . The Cauchy problem (8) admits a unique weak solution .
Moreover, there
exists such that
(11) 
The proof of creftypecap 3.2 is based on Galerkin approximations, energy estimates and Gronwall’s argument (see [15, Section 7.1.2]).
We also have the following classical estimate for (8).
Proposition 3.3.
Let , and . Then the solution of (8) belongs to and there exists such that
(12) 
The proof of creftypecap 3.3 is based on Stampacchia’s method (see the proof of [28, Chapter 3, Paragraph 7, Theorem 7.1]).
Let us also mention the dissipativity in time of the norm of the heat equation with a bounded potential.
Proposition 3.4.
Let , and . Then, there exists such that the solution of (8) with , satisfies for every ,
(13) 
The proof of creftypecap 3.4 is based on the application of the variational formulation (9) with a cutoff of and a Gronwall’s argument.
Nonlinear parabolic equations
We give the definition of a solution of (1).
Definition 3.5.
The uniqueness of a solution to (1) is an easy consequence of the fact that is locally Lipschitz because .
3.2. Maximum principle
We state the maximum principle for the heat equation.
Proposition 3.6.
Let , and . Let and be the solutions to
(16) 
Then, we have the comparison principle
(17) 
The proof of creftypecap 3.6 is based on the comparison principle for smooth solutions of (16) (see [41, Theorem 8.1.6]) and a regularization argument.
We state a comparison principle for the semilinear heat equation (1) without control .
Proposition 3.7.
Let , . We assume that there exist a subsolution and a supersolution in of (1), i.e., (respectively ) satisfies (14), (15) replacing the equality by the inequality (respectively by the inequality ). Moreover, we suppose that and are ordered in the following sense
Then, there exists a (unique) solution of (1). Moreover, satisfies the comparison principle
(18) 
For the proof of creftypecap 3.7, see [41, Corollary 12.1.1].
3.3.  estimates
We have the wellknown regularizing effect of the heat semigroup.
4. Global nonnegativecontrollability of the linear heat equation with a bounded potential
4.1. Statement of the result
Let . We consider the heat equation with a bounded potential
(20) 
and the following adjoint equation
(21) 
The goal of this section is to prove the following theorem.
Theorem 4.1.
For every , (20) is globally nonnegativecontrollable in time . More precisely, for every , there exists , with
(22) 
such that for every , there exists such that
(23) 
and
(24) 
Remark 4.2.
Actually, by looking carefully at the proof of creftypecap 4.1 (see Section 4.5 below), we can see that the control in creftypecap 4.1 can be chosen constant in the time and the space variables.
Remark 4.3.
It is wellknown that (20) is globally nonnegativecontrollable in time because it is globally nullcontrollable in time (see [20, Theorem 2]) but the most interesting point is the cost of nonnegativecontrollability given in creftypecap 4.1. In particular, the exponent of the term will be the key point to prove creftypecap 2.2 (see Section 5).
4.2. A precise  observability inequality for the linear heat equation with bounded potential and nonnegative initial data
The proof of creftypecap 4.1 is a consequence of this kind of observability inequality.
Theorem 4.4.
An immediate corollary of creftypecap 4.4 is this observability inequality  that we state to discuss it below, but that will not be used in the present article.
Corollary 4.5.
It is wellknown that nullcontrollability in is equivalent to an observability inequality in for every (see [9, Theorem 2.44]). The main idea behind creftypecap 4.5 is the fact that nonnegativecontrollability in is a consequence of an observability inequality in for every (see Section 4.5).
Remark 4.6.
It is interesting to mention that (26) holds with of the form
(27) 
for every (see [20, Theorem 2]). The exponent of the term is the key point to prove creftypecap 1.4. Note that the optimality of the exponent has been proved by Thomas Duyckaerts, Xu Zhang and Enrique Zuazua in the context of parabolic systems in even space dimensions and with Dirichlet boundary conditions (see [14, Theorem 1.1] and also [44, Theorem 5.2] for the main arguments of the proof). creftypecap 4.5 shows that we can actually decrease the exponent to the exponent for nonnegative initial data. In some sense, we can make the connection between the recent preprint of Camille Laurent and Matthieu Léautaud who disprove the Miller’s conjecture about the shorttime observability constant of the heat equation in the general case and show that the conjecture holds true for nonnegative initial data by using LiYau estimates (see [29] and [32]).
Remark 4.7.
In the context of the wave equation in one space dimension, the (optimal) constant of observability inequality for the linear wave equation with a bounded potential is actually (see [42, Theorem 4]) which leads to the exact controllability of the semilinear wave equation in large time for semilinearities satisfying (3) with (see [42, Theorem 1] and also [7, Problem 5.5] for the presentation of the related open problem in the multidimensional case). Roughly speaking, as an ordinary differential argument would indicate, this constant of observability inequality is very natural because the wave operator is of order two in the time and the space variables. Then, by analogy and by taking into account that the heat operator is of order one in the time variable and of order two in the space variable, one could rather expect a constant of obervability inequality of the order or which seem to be more intuitive than the term .
4.3. A new Carleman estimate
The goal of this section is to establish a Carleman estimate for nonnegative initial data (see creftypecap 4.9 below). First, we introduce some classical weight functions for proving Carleman inequalities.
Lemma 4.8.
Let be a nonempty open subset. Then there exists such that in , in , and in .
A proof of this lemma can be found in [9, Lemma 2.68].
Let be a nonempty open set satisfying and let us set
(28) 
(29) 
for , where is the function provided by creftypecap 4.8 for this and is a parameter.
We have the following new Carleman estimate.
Theorem 4.9.
There exist two constants and , such that,
(30) 
for every , the nonnegative solution of (21) satisfies
(31) 
Proof.
Unless otherwise specified, we denote by various positive constants varying from line to line which may depend on , but independent of the parameters and .
We introduce other weights which are similar to and
(32) 
(33) 
The following estimates
(34) 
will be very useful for the proof.
Let . The general case comes from an easy density argument by using the fact that is dense in for the topology.
The solution of (21) is nonnegative by applying the maximum principle given in creftypecap 3.6 with and .
We define
The proof is divided into five steps:

Step 1: We integrate over an identity satisfied by .

Step 2: We get an estimate which looks like to (31) up to some boundary terms.

Step 3: We repeat the step 1 for .

Step 4: We repeat the step 2 for .

Step 5: We sum the estimates of the step 2 and the step 4 to get rid of the boundary terms.
Remark 4.10.
Step 1: An identity satisfied by . We readily obtain that
(35) 
where
(36)  
Remark 4.11.
The starting point, i.e., the identity (35) is the same as in the classical proof developed by Andrei Fursikov and Oleg Imanuvilov in [23] (see also [21, Proof of Lemma 1.3] or [31, Section 7]). But, from now, the proof strategy of the Carleman estimate is very different from the usual one of the Carleman estimate. Indeed, we will focus on the fourth right hand side term of (36)
It is nonnegative because is nonnegative and it is of order two in the parameter whereas the seventh right hand side term of (36)
is of order in the parameter . This comparison suggests to integrate the identity (35) in order to obtain (31) for and as defined in (30).
We integrate (35) over
(37) 
Note that all the terms in (37) are welldefined. Indeed, by using and the parabolic regularity in to (21) (see [12, Theorem 2.1]), we deduce that then .
Step 2: Estimates for . As a consequence of the properties of (see creftypecap 4.8), we have
(38) 
which yields
(39)  
By combining (37) and (39), we have
(40) 
We have the following integration by parts
(41) 
(42) 
(43) 
where .
From (40), (41), (42), (43), we have
(44) 
By using the first two lines of (34) and , we have
(45) 
By combining (44) and (45), we get
(46) 