Abstract
We construct a singular solution of a stationary nonlinear Schrödinger equation on \({\mathbb {R}}^2\) with square-exponential nonlinearity having linear behavior around zero. In view of Trudinger-Moser inequality, this type of nonlinearity has an energy-critical growth. We use this singular solution to prove non-uniqueness of mild solutions for the Cauchy problem of the corresponding semilinear heat equation. The proof relies on explicit computation showing a regularizing effect of the heat equation in an appropriate functional space.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In this paper, we consider the Cauchy problem for the following semilinear heat equation
where \(u(t, x) : (0, \infty ) \times {\mathbb {R}}^{d} \rightarrow {\mathbb {R}}\), \(d\ge 2\) is the unknown function, f is the nonlinearity and \(u_{0}\in L^q({\mathbb {R}}^d)\) with \(1\le q\le \infty \) is the given initial data.
It is well known that when f is \(C^1({\mathbb {R}},{\mathbb {R}})\), the Cauchy problem (1.1) possesses a unique classical solution if the initial data \(u_0\in L^\infty ({\mathbb {R}}^d)\). For unbounded initial data, this equation has been studied intensively since the pioneering work of Weissler [18]. For instance, in the pure power case i.e. \(f(u) = |u|^{p-1}u\; (p>1)\), equation (1.1) becomes scale invariant, that is, if u(t, x) satisfies (1.1), then so does
for any \(\lambda >0\). Moreover, if one defines the index
then the Lebesgue space \(L^{q_{c}}({\mathbb {R}}^d)\) becomes invariant and we have \(\Vert u_{\lambda }\Vert _{L^{q_{c}}} = \Vert u\Vert _{L^{q_{c}}}\) for all \(\lambda >0\). The critical exponent \(q_{c}\) then plays an important role for the well-posedness of the Cauchy problem. Indeed, first recall Weissler’s result [18] concerning the subcritical case \(q > q_{c}\) or the critical case \(q=q_c>1\). Weissler [18] proved that for any \(u_{0} \in L^{q}({\mathbb {R}}^{d})\), there exists a local time \(T = T(u_{0})>0\) and a solution \(u \in C([0, T); L^{q}({\mathbb {R}}^{d})) \cap L_{\text {loc}}^{\infty }((0, T); L^{\infty }({\mathbb {R}}^{d}))\) to (1.1). After that, Brezis and Cazenave [1] proved the unconditional uniqueness of Weissler’s solutions i.e. solution is unique in \( C([0, T); L^{q}({\mathbb {R}}^{d}))\) when the subcritical case \(q>q_c\), \(q\ge p\) or the critical case \(q=q_c>p\). In the supercritical case \(q < q_{c}\), Weissler [18], and Brezis and Cazenave [1] indicated that, for a specific initial data, there is no local solution in any reasonable weak sense. Moreover, Haraux and Weissler [7] proved non-uniqueness of the trivial solution in \(C([0, T); L^{q}({\mathbb {R}}^{d})) \cap L_{\text {loc}}^{\infty } ((0, T); L^{\infty }({\mathbb {R}}^{d}))\) when \(1 + 2/d< p < (d+2)/(d-2)\).
In the critical case \(q = q_{c}\) and \(d \ge 3\), when \(q_{c} > {p}\), Weissler [18] proved the existence of solutions to (1.1) and Brezis and Cazenave [1] obtained the unconditional uniqueness of the solutions. In the double critical case of \(q = q_{c} = p\; (= d/(d-2))\), Weissler [19] proved the conditional well-posedness (uniqueness in a subspace of \(C([0, T); L^{q}({\mathbb {R}}^{d}))\)). In the case where the underlying space is the ball of \({\mathbb {R}}^{d}\) with Dirichlet boundary condition, Ni and Sacks [15] showed that the unconditional uniqueness fails, while Terraneo [17] and Matos and Terraneo [12] extended this result to the entire space \({\mathbb {R}}^{d} \; (d \ge 3)\).
We note that the critical exponent \(q_{c}\) is also important for the blow-up problem (1.1). Let \(u_{0} \in L^{\infty }({\mathbb {R}}^{d})\) and \(T(u_{0})\) be the maximal existence of the time of the classical solution u. It is known that if \(T(u_{0}) < \infty \), the solution u satisfies \(\lim _{t \rightarrow T} \Vert u(t)\Vert _{L^{\infty }} = \infty \) and we say that the solution u blows up in finite time and \(T(u_{0})\) is called the blow-up time of u. In particular, the critical \(L^{q_{c}}\) norm blow-up problem has attracted attention for a long time. Namely, it is a question whether the solution satisfies
or not when \(T(u_{0}) < \infty \). Concerning this problem, Giga and Kohn [6] showed that if u is a positive radially decreasing blow-up solution to (1.1), Then, for any neighborhood N of 0 in \({\mathbb {R}}^{d}\), \(\lim _{t \rightarrow T} \Vert u(t)\Vert _{L^{q_{c}, \infty }(N)} = \infty \), where
and \(|\cdot |\) is the Lebesgue measure on \({\mathbb {R}}^{d}\). Since \(\Vert u\Vert _{L^{q_{c}, \infty }(N)} \le \Vert u\Vert _{L^{q_{c}}}\), it follows that \(\lim _{t \rightarrow T} \Vert u(t)\Vert _{L^{q_{c}}} = \infty \). Recently, Mizoguchi and Souplet [14] gave light on this problem and showed that if u is a type I blow-up solution of (1.1), that is, \(\limsup _{t \rightarrow T} (T - t)^{1/(p-1)} \Vert u(t)\Vert _{L^{\infty }} < \infty \), then we have \(\lim _{t \rightarrow T} \Vert u(t)\Vert _{L^{q_{c}}} = \infty \).
Now, let us pay our attention to the two dimensional case. When we consider the two space dimension, we see that any exponent \(1< p < \infty \) is subcritical, and thanks to the result of Weissler [18], we have the local well-posedness of the Cauchy problem in \(L^{q}({\mathbb {R}}^{2})\; (1< p \,{\le }\, q < \infty )\). However, for exponential type nonlinearities, like \(f(u) = \pm u(e^{u^{2}} -1)\), the result of Weissler [18] is not applicable for any Lebesgue space \(L^{q}({\mathbb {R}}^{2}) (1< q < \infty )\). On the other hand, we can show the local well-posedness for \(u_{0} \in L^{\infty }({\mathbb {R}}^{2})\), as we mentioned first. However, \(L^{\infty }({\mathbb {R}}^{2})\) is the subcritical space for the problem (1.1) with \(d=2\) and exponential type nonlinearities. Therefore, one can wonder if there is any notion of criticality in two space dimensions. In this regard, Ibrahim, Jrad, Majdoub and Saanouni [8] have considered the following problem in two space dimensions,
The nonlinearity \(f_{0}(u)\) has an energy-critical growth in view of Trudinger-Moser inequality. In [8], the authors proved the local existence and uniqueness in \(C([0,T],H^1({\mathbb {R}}^{2}))\) of the solution to (1.2) with the initial data \(u_{0} \in H^{1}({\mathbb {R}}^{2})\).
On the other hand, it is expected that the problem (1.2) for the heat equation can be solved in spaces which are defined by an integrability of functions such as the Orlicz space as an extension of the class of Lebesgue spaces. Ruf and Terraneo [16] showed the local existence of a solution to (1.2) for small initial data in the Orlicz space \(\text {exp}L^2\) defined by
endowed with Luxemburg norm
Subsequently, Ioku [9] has shown that these local solutions are indeed global-in time, and the behavior of \(f_0(u)\sim u^m\) near \(u\sim 0\) with \(m\ge 3\) was important in his result. Later on, Ioku, Ruf and Terraneo [10] proved that there is no local solution do not exist for the specific initial data
that belongs to the Orlicz space, while they showed the local well-posedness (local existence and uniqueness) for initial data in the subclass of the Orlicz space
The aim of this paper is to construct an explicit initial data, with neither small nor too large norm in Orlicz space, for which two corresponding distinct solutions exist. The idea is to use the concept of singular solutions as in Ni and Sacks [15]. Before stating our result, let us recall the strategy of the proof of [15]. Ni and Sacks first constructed a singular static solution \(\phi _{*}\) to (1.1) in the unit ball. Namely, \(\phi _{*}\) satisfies the following:
where \(B_{1}\) is the unit ball in \({\mathbb {R}}^{d}\; (d \ge 3)\) and \(f(\phi ) = |\phi |^{p-1}\phi \). Then, they showed that there exists a regular solution \(u_{\text {R}}\) to the Dirichlet problem corresponding to (1.1) with \(u(0, x) = \phi _{*}\). Setting \(u_{\text {S}} = \phi _{*}\), we see that both of \(u_{\text {S}}\) and \(u_{\text {R}}\) are solutions to (1.1), but \(u_{\text {S}} \ne u_{\text {R}}\) because for any \(t>0\), \(u_{\text {R}} (t,\cdot ) \in L^{\infty }(B_{1})\) while \(u_{\text {S}} \notin L^{\infty }(B_{1})\) (the subscripts S and R indicating ‘singular’ and ‘regular’ solutions). For the entire space \({\mathbb {R}}^{d}\), Terraneo [17] constructed a solution \(\phi \in C^{2}({\mathbb {R}}^{d} {\setminus } \{-x_{0}, x_{0}\})\) to the equation
such that \(\limsup _{x \rightarrow x_{0}} \phi (x) = \infty \) and \(\liminf _{x \rightarrow -x_{0}} \phi (x) = -\infty \), where \(d \ge 3, x_{0} = (1, \ldots , 1)\) and \(f(\phi ) = |\phi |^{p-1} \phi \). However, we cannot apply the method of [15] nor of [17] to two dimensional entire space \({\mathbb {R}}^{2}\) for \(f_{0}(u) = (e^{u^{2}} - 1)u\) directly. Actually, there is no positive solution to the equation (1.4) with \(f(u) = (e^{u^{2}} - 1)u\) (see the proof of Theorem 2.1 below). For this reason, we consider
where our nonlinearity \(f_{m}\), which depends on \(m >0\), satisfies
See (2.17) below for the precise form of \(f_{m}\). Here, we would like to stress that the essential characterization of the asymptotic form of our nonlinearity \(f_{m}\) at infinity and 0 is just given by (1.6). Let X be the Fréchet space defined as the intersection of the Lebesgue Banach spaces
endowed with the metric
Recall that the continuity of any map defined on X is equivalent to its continuity on all \(L^p\) spaces with \(1\le p<\infty \).
Definition 1
(Mild solution) We call u a mild solution to (1.5) if the corresponding integral equation
holds in C([0, T), X), where \(e^{t \Delta }\) denotes the heat semigroup on \({\mathbb {R}}^{2}\).
Then, our main result is the following:
Theorem 1.1
There exist a positive mass \(m_{*}>0\), a nonlinearity \(f_{m_*}\) satisfying (1.6), an initial condition \(\varphi _{*} \in X\) and a time \(T = T(\varphi _{*}) >0\) such that the Cauchy problem (1.5) with \(u_{0} = \varphi _{*}\) has two distinct mild solutions \(u_{S}, u_{R} \in C([0, T), X)\).
Remark 1
-
(i)
To prove Theorem 1.1, we construct a singular stationary solution \(\varphi _{*}\) to (1.5). Here, by a singular stationary solution, we mean a time independent function which satisfies the equation (1.5) in the distribution sense on the whole domain and diverges at some points. The result seems to be of independent interest.
-
(ii)
We essentially use the condition \(\lim _{u \rightarrow 0}f_{m}(u)/mu = -1\) only for the decay of a singular stationary soliton \(\varphi _{*}\) to (1.5), that is, \(\lim _{|x| \rightarrow \infty } \varphi _{*}(x) = 0\). The other argument in our proof works even without the condition. It is a challenging problem to study whether non-uniqueness still hold without the condition or not.
Remark 2
After completing this paper, it has been brought to our attention that Ioku, Ruf and Terraneo [11] obtained a similar non-uniqueness result for the following two dimensional heat equation on a ball with Dirichlet boundary condition:
for some \(\rho > 0\), and a nonlinearity f satisfying
with \(\alpha = e^{5/2} (5/2)^{-5/2}\) and \(\beta = (5/2)^{1/2}\). First, the authors constructed a stationary singular solution \(\varphi _{*}\) of (1.7). Then, they solved the Cauchy problem (1.7) with the initial data \(u_{0}(x) = \mu \varphi _{*}\) for \(\mu > 0\). They proved the following interesting result:
-
(i)
If \(\mu < 1\), the equation (1.7) has a unique regular local-in-time solution
-
(ii)
If \(\mu = 1\), the equation (1.7) has not only a singular solution \(u_{S} (= \varphi _{*})\) but also a regular solution \(u_{R}\) with the same initial data,
-
(iii)
If \(\mu > 1\), the equation has no non-negative solution on any positive time interval.
See [11, Theorem 2.1] for a more precise statement of their trichotomy result. From (ii), we see that non-uniqueness of a solution holds since there exists two distinct solutions \(u_{S}\) and \(u_{R}\) with the same initial data. Thus, the strategy of the proof for the non-uniqueness is same as that of our result. However, in addition to the fact that the settings are different, the ways of constructing the singular and regular solutions are not similar either. For example, their proof of the constructing of the singular stationary solution \(\varphi _{*}\) depended on the fact that the following equation
has an explicit solution \(U = \sqrt{-2 \log r}\). On the other hand, the singular solution we constructed is not explicit. We construct the singular solution in a ball by using the contraction mapping theorem and extend it to the entire space by the shooting method. Moreover, for the construction of the regular solution, they employed the sub-super-solution method, while we employ the contraction mapping theorem.
Remark 3
The problem of uniqueness of solutions for PDEs is a classical and old issue that can be tricky sometimes. It has caught a special attention in the last few years. Among many others, one can refer to the pioneer works of De Lellis and Székelyhidi [4] showing non-uniqueness of very weak solutions to the Euler problem. Their proof relies on convex integration techniques, which more recently, have been upgraded to show non-uniqueness of weak solutions of the Navier-Stokes system thanks to the work of Buckmaster and Vicol [2]. Davila, Del Pino and Wei [3] constructed non-unique weak solutions for the two-dimensional harmonic map, by attaching reverse bubbling. In [5], Germain, Ghoul and Miura investigated the genericity of the non-unique solutions of the supercritical heat flow map.
This paper is organized as follows. In Sect. 2, we construct a singular static soliton \(\varphi _{*}\) to (1.5). To this end, we first prove the existence of a singular soliton \(\phi _{*}\) to (1.3) with \(f(\phi ) = e^{\phi ^{2}}(\phi - 1)\) in the ball in \({\mathbb {R}}^{2}\), following Merle and Peletier [13]. Then, we seek the singular stationary soliton \(\varphi _{*}\) to (1.5) by the shooting method. In Sect. 3, we shall show the existence of a regular solution to (1.5) with \(u|_{t=0} = \varphi _{*}\) by the heat iteration and give a proof of Theorem 1.1. In Appendix A, we show a monotonicity property of solution to the linear heat equation.
1.1 Notation
Throughout the paper, C denotes a positive constant, that does not depend on the parameters, unless otherwise noted and may change from line to line.
2 Construction of singular soliton
2.1 Singular stationary solution on some disk
In this section, we construct a singular solution to the following elliptic equation on a disk
where \(R > 0\). More precisely, we shall show the following:
Theorem 2.1
There exist \(R_{\infty }>0\) and a singular solution \(u_{\infty } \in C^{\infty }(B_{R_{\infty }} {\setminus } \{0\})\) to (2.1) with \(R = R_{\infty }\) satisfying
To prove Theorem 2.1, we look for radially symmetric solutions to (2.1). We first pay our attention for \(0 < r = |x| \ll 1\) and employ the following Emden-Fowler transformation:
Then, we see that the equation in (2.1) is equivalent to the following:
We shall show the following:
Proposition 2.2
There exists \(\Lambda _{\infty } >0\) and a solution \(y_{\infty }(\rho )\) to (2.3) for \(\rho \in [\Lambda _{\infty }, \infty )\) satisfying
In order to prove Proposition 2.2, we write
and look for the ODE satisfied by \(\eta \). We have
Now, since
and
we obtain
In addition, since \( y^{2} - \phi ^{2} = 2\phi \eta + \eta ^{2} = 2 \rho ^{\frac{1}{2}} \eta + \eta ^{2} + 2(\phi - \rho ^{\frac{1}{2}}) \eta , \) we have
All this yields that
From (2.3), (2.5) and (2.6), we have
Namely, \(\eta \) satisfies the following:
where
To solve equation (2.7), we first consider the following linear equation: Footnote 1
Any solution \(\eta \) to (2.8) can be written explicitly as follows: we have
for some \(A, B \in {\mathbb {R}}\), where
Namely, \(\Phi \) and \(\Psi \) are form the fundamental system of solutions to (2.8). For a given function F, we seek a solution to the following problem:
By the variation of parameters, we see that (2.9) is equivalent to the following integral equation:
By integrating by parts, we can obtain the following:
Lemma 2.3
Let \(\sigma > 1\). Then, we have
Let \(g_{7}(\rho , \eta ) = - \frac{3}{16 \rho ^{2}} \eta \). In order to prove Theorem 2.1, we seek a solution to the following problem:
To this end, we prepare several estimates, which are needed later. First, note that
By (2.11) and elementary calculations, we can obtain the following:
Lemma 2.4
Let \(\phi \) be the function given by (2.4). Then, for sufficiently large \(\rho >0\), we have
In addition, we have the following estimates on the terms \(g_j\) of equation (2.10). More specifically, we have
Lemma 2.5
Suppose that \(\eta _{1}(\rho ), \eta _{2}(\rho ) = O(\rho ^{-\frac{3}{2}}\log \rho )\). Then, we have
Proof
From the definitions of \(g_{i}\) and (2.11), we have
Let \(y_{1}(\rho ) = \phi (\rho ) + \eta _{1}(\rho )\) and \(y_{2}(\rho ) = \phi (\rho ) + \eta _{2}(\rho )\). Then, we obtain
Since \(y_{i}^{2} - \phi ^{2} = 2\phi \eta _{i} + \eta _{i}^{2}\) for \(i = 1, 2\), we have
It follows that
-1pt From (2.1), we have
-1pt Using the mean value theorem and (2.1), we have
This yields that \(II \le C { \rho ^{-2} (\log \rho )} |\eta _{1} - \eta _{2}|\). Thus, we see that
as desired. \(\square \)
We are now in a position to prove Proposition 2.2.
Proof of Proposition 2.2
Note that (2.7) is equivalent to the following integral equation:
in which
where
Fix \(\Lambda >0\) sufficiently large and let X be a space of continuous functions on \([\Lambda , \infty )\) equipped with the following norm:
Lemmas 2.3 and 2.4 insure that there exists a constant \(C_{*}>0\) such that
For this constant \(C_{*}>0\), define the space
and we shall first show that \({\mathcal {T}}\) maps \(\Sigma \) to itself. From Lemma 2.5, we have
which yields that
Therefore, we can take \(\Lambda >0\) sufficiently large so that
for \(\rho \ge \Lambda \). We can easily find that for sufficiently large \(\Lambda >0\), we have
for \(\rho \ge \Lambda \). By (2.12), (2.13) and (2.14), we obtain
for \(\eta \in \Sigma \). This yields that \({\mathcal {T}}[\eta ] \in \Sigma \).
Next, we shall show that \({\mathcal {T}}\) is a contraction mapping. For \(\eta _{1}, \eta _{2} \in \Sigma \), we have
From Lemma 2.5 follows the estimate
This yields that
for sufficiently large \(\rho > 0\). Thus, we find that \({\mathcal {T}}\) is a contraction mapping. This completes the proof. \(\square \)
Proof of Theorem 2.1
Let \(u_{\infty }(r) = y_{\infty }(\rho )\), where \(y_{\infty }(\rho )\) is the solution to (2.3) obtained by Proposition 2.2. We note that \(u_{\infty }\) satisfies the following:
where \(R_{\infty } = e^{- \Lambda _{\infty }/2}\). Since \(u_{\infty }\) is a solution to the ordinary differential equation (2.15), we can extend \(u_{\infty }\) in the positive direction of r as long as \(u_{\infty }\) remains bounded. We claim that \(u_{\infty }\) has a zero at some point. Suppose the contrary that \(u_{\infty }(r) >0\) for all \(0< r < \infty \). Then, we see that \(u_{\infty }\) is monotone decreasing. Indeed, if not, there exists a local minimum point \(r_{*} \in (0, \infty )\). It follows that \(\partial ^{2}_{r} u_{\infty }(r_{*}) \ge 0\) and \(\partial _{r} u_{\infty }(r_{*}) = 0\). Then, from the equation (2.15), we obtain
which is a contradiction.
Since \(u_{\infty }\) is positive and monotone decreasing, there exists a constant \(C_{\infty } \ge 0\) such that \(u_{\infty }(r) \rightarrow C_{\infty }\) as \(r \rightarrow \infty \). To prove that \(C_{\infty } = 0\), suppose the contrary that \(C_{\infty } > 0\). This together with (2.15) yields that
which is absurd. Therefore, we see that \(C_{\infty } = 0\), that is, \(\lim _{r \rightarrow \infty }u_{\infty }(r) = 0\). Multiplying (2.15) by r and integrating the resulting equation from 0 to r yields that
This yields that for any \(R>0\), there exists a constant \(C_{1} >0\) such that \(- d u_{\infty }/dr(r) \ge C_{1}/r\) for all \(r > R\). It follows that
Letting \(r \rightarrow \infty \), we have
which is a contradiction. Therefore, there exists \(R_{\infty } > 0\) such that \(u_{\infty }(r)\) has a zero at \(r = R_{\infty }\).
Finally, we shall show that \(u_{\infty }\) is a solution (2.1) in a distribution sense. Let \(\eta (\rho )\) be the solution to (2.10), obtained in the proof of Proposition 2.2. Since \(\eta (\rho ) = O(\rho ^{-3/2} \log \rho )\), we see that \(u_{\infty }\) satisfies
This yields that there exists a constant \(C>0\) such that
This together with the monotonicity implies that \(u_{\infty } e^{u_{\infty }^{2}} \in L^{1}_{\text {loc}}(B_{R_{\infty }})\). Therefore, we see that \(u_{\infty }\) is a distributional solution to (2.1). This completes the proof. \(\square \)
2.2 Singular soliton by the shooting
Let \(R\in (0,R_\infty )\) be the unique point such that \(u_\infty (R)=2\). For each \(m\ge 0\), we put
where \(\chi \in C^\infty ({\mathbb {R}})\) is a cut-off function satisfying \(\chi (t)=1\) for \(|t|\le 1, \chi (t)=0\) for \(|t|\ge 2\), and \(t\chi '(t)\le 0\) for all \(t\in {\mathbb {R}}\). Consider a family of radial ODE’s with a parameter \(m\ge 0\):
where the prime mark denotes the differentiation with respect to r. Let \(\phi _m\) be the unique solution of the above. We shall show that there exists \(m_{*}>0\) such that \(\phi _{m_{*}}(r)\searrow 0\) as \(r\rightarrow \infty \). To this end, we show the following:
Proposition 2.6
Let \(m\ge 0\) and \(\phi _{m}\) be the solution to (2.18). There exists \(m_{S}>0\) and \(m_{L}>0\) such that if \(m \in [0, m_{S})\), \(\phi _{m}\) has a zero in \((R, \infty )\), and if \(m \in (m_{L}, \infty )\), \(\phi _{m}(r)\) is positive for all \(r \ge R\).
First, we define an energy density function \(E_{m}:[R,\infty )\rightarrow {\mathbb {R}}\) by
where \(F_m\) is the nonlinear potential energy defined by
which enjoys the standard superquadratic condition:
and, thanks to the monotonicity of \(\chi \),
It follows from (2.18) that
Thus, \(E_{m}(r)\) is a non-increasing function of r. Using (2.21), we shall show the following:
Lemma 2.7
Let \(m>0\) and \(\phi _{m}\) be a solution to (2.18). Let \(s\ge R\), \(E_{m}(s) <0\) and \(\phi _{m}(s)>0\). Then, we have \(\phi _{m}(r) >0\) for all \(r \in (s, \infty )\).
Proof
Suppose the contrary. Then there exists \(z \in (s, \infty )\) such that \(\phi _{m}(z) = 0\). This together with (2.21) yields that
which is a contradiction. \(\square \)
We are now in a position to prove Proposition 2.6.
Proof of Proposition 2.6
We note that \(m = 0\), \(\phi _{0}(r) = u_{\infty }\) has a zero at \(r = R_{\infty }\). Then, by the continuity of \(\phi _m\) with respect to m, we see that \(\phi _{m}(r)\) still has a zero if \(m>0\) is sufficiently small.
On the other hand, we have
for large \(m>0\), then Lemma 2.7 implies that \(\phi _m(r)>0\) for all \(r>R\). \(\square \)
We put
We see from Proposition 2.6 that \(0< m_{*} < \infty \). We extend \(\phi _{m_{*}}\) by defining \(\phi _{m_{*}}(r) = u_{\infty }(r)\) for \(r \in (0, R)\) (still denoted by the same symbol). Then, we have the following:
Theorem 2.8
Let \(m_{*} >0\) be the number defined by (2.22). Then, \(\phi _{m_{*}}\) is a singular positive radial solution to the following elliptic equation
where \(f_{m}(s)\) is defined by (2.17). Moreover, \(\phi _{m_{*}}\) is strictly decreasing in the radial direction. Moreover, for any \(m \in (0, m_{*})\) there exists \(C_{m} \in (0, \infty )\) such that
Proof
To prove Theorem 2.8, it suffices to show that \(\phi _{m_{*}}(r) > 0, \phi _{m_{*}}^{\prime }(r) < 0\) for all \(r > 0\), \(\lim _{r \rightarrow \infty } \phi _{m_{*}}(r) = 0\) and (2.24).
First, we shall show that \(\phi _{m_{*}}(r) >0\) on \(r \in (R, \infty )\). By definition of \(m_*\), there exists a sequence \(m_{n} \searrow m_{*}\) such that \(\phi _{m_{n}}(r) >0\) for all \(r> R\) and \(n \in {\mathbb {N}}\). Then \(\phi _{m_*}(r)=\lim _{n\rightarrow \infty }\phi _{m_n}(r) \ge 0\) for all \(r> R\). If \(\phi _{m_*}(r)=0\) at some \(r>R\), then \(\phi _{m_*}'(r)=0\) and so \(\phi _{m_*}\equiv 0\) by the ODE, a contradiction. Hence \(\phi _{m_*}(r)>0\) for all \(r>R\).
Next, we claim that
Suppose the contrary that there exists \(R_{*}>0\) such that \(E_{m_{*}}(R_{*}) < 0\). Then the continuity for m yields that \(\phi _m(r)>0\) on \(R\le r\le R_*\) and \(E_{m}(R_{*}) < 0\) when \(m\in (0,m_*)\) is close enough to \(m_*\). Then Lemma 2.7 implies \(\phi _m(r)>0\) for \(r\ge R_*\), hence for all \(r>R\), contradicting the definition of \(m_*\). Hence we have (2.25).
Next, we shall show that \(\phi _{m_{*}}^{\prime }(r) < 0\) for all \(r > R\). Suppose the contrary and let \(s>R\) be the first zero of \(\phi _{m_*}'\). Then we have \(0=\phi _{m_*}'(s)\le \phi _{m_*}''(s)=-f_{m_*}(\phi _{m_*}(s))\), \(0\le E_{m_*}(s)=F_{m_*}(\phi _{m_*}(s))\), and so \(G_{m_*}(\phi _{m_*}(s))\le 0\), contradicting (2.19).
Therefore \(\phi _{m_*}'(r)<0<\phi _{m_*}(r)\) for all \(r>R\), so \(\phi _{m_*}(r)\searrow \exists C_*\in [0,2)\) and \(\phi _{m_*}'(r)\rightarrow 0\) as \(r\rightarrow \infty \). Then we have \(0\le \lim _{r\rightarrow \infty }E_{m_*}(r)=F_{m_*}(C_*)\) and
which has to be 0 because \(\phi _{m_*}'(r)\rightarrow 0\). Hence \(G_{m_*}(C_*)\le 0\) and so \(C_*=0\) by (2.19).
Next, let \(m\in (0,m_*)\). Then \(\phi _{m_*}(r)\searrow 0\) together with the definition of \(f_m\) implies
for sufficiently large \(r>R\). Hence \(e^{\sqrt{m}r}(\sqrt{m}-\partial _r)\phi _{m_*}(r)\) is decreasing for large r, which implies the desired exponential decay.
Finally, it follows from (2.16) and (2.24) that \(f_{m_{*}} \in L^{1}({\mathbb {R}}^{2})\). Thus, we see that \(\phi _{m_{*}}\) is a distributional solution to (2.23). This completes the proof. \(\square \)
Remark 4
Let \(u_{\infty }\) be a solution to (2.1), which is obtained in Theorem 2.1. Since \(\phi _{m_{*}} (r) = u_{\infty }(r)\) for \(r \in (0, R]\), we see from (2.2) that \(\phi _{m_{*}}\) satisfies
Moreover, we have, by (2.10), Lemmas 2.3 and 2.4, that \(|d^{2} \eta /d \rho ^{2} (\rho )| \le C \rho ^{-5/2} \log \rho \). Thus, by integrating, we see that \(|d \eta /d \rho (\rho )| \le C \rho ^{-3/2} \log \rho \). Thus, \(\phi _{m_{*}}^{\prime }\) satisfies
3 Regular solution by the heat iteration
In what follows, we denote \(\phi _{m_{*}}\), which is obtained in Theorem 2.8, by \(\varphi _{*}\). Now we are ready to construct a regular solution to (1.5) with initial data \(\varphi _*\).
Theorem 3.1
Let \(u_0:=e^{t\Delta }\varphi _{*}\). Then for any \(t>0\), \(u_0(t)\) is bounded on \({\mathbb {R}}^2\). Moreover, there exists a small \(T>0\), and a solution \(u_{R}\) to (1.5) with \(u_{R}|_{t=0} = \varphi _{*}\) satisfying
Note that \(\varphi _{*}\in C^\infty ({\mathbb {R}}^2{\setminus }\{0\})\) is a positive radial function satisfying the asymptotic form,
where, as before, \(\rho :=|\log r^2|=2|\log r|\gg 1\). The above two equivalent expressions of remainder will be frequently switched in the following computations.
We need a precise estimate or asymptotic behavior around \(t,r\rightarrow 0\) of the iteration sequence. Consider the first (or zeroth) iteration
We shall show the following:
Lemma 3.2
There exists \(\varepsilon >0\) such that if \(\max \{t, |x|^{2}\} < \varepsilon ^{2}\), we have
Proof
We take \(\varepsilon >0\) sufficiently small so that for any \(r < \varepsilon \), \(\varphi _{*}(r)\) satisfies (3.1), and consider the region \((t, x) \in (0, \infty ) \times {\mathbb {R}}^{2}\) satisfying \(\max \{t, |x|^{2}\} < \varepsilon ^{2}\). Put
We shall estimate \(u_{0}\) by dividing the space-time region \(\max \{t, |x|^{2}\} < \varepsilon ^{2}\) into the three subregions: inside the parabolic cylinder \(\nu \le 1\), \(\nu > 8 \log \ell \gg 1\) and \(1\le \nu \le 8 \log \ell \).
First, we consider the region \(\nu \le 1\). It follows from Lemma A.1 that
Thus, by (3.1), we obtain
For \(0< s < t\), we have \(ts>s^{2}\). This yields via an integration by parts
For \(t< s < \frac{\varepsilon ^2}{t}\), we have \(|\log s| <\ell \), so the integrand is expanded
where we set \({\langle \log s \rangle } = \sqrt{1 + |\log s|^{2}}\). Since \(\varphi _{*}(r)\) is monotone decreasing in \(r > 0\), we have, by (3.1) and (3.5), that
for \(s \ge \frac{\varepsilon ^{2}}{t}\). Integration against \(e^{-s/4}\) yields
Estimates (3.4) and (3.6) together with (3.3) imply
Namely, we have
This is enough in the region \(\nu \le 1\).
Before focusing on the two remaining regions, let \(0<\delta \le \varepsilon \) be a small parameter and set
Decomposing \(u_{0}\) as
and writing
First, we consider \(z \in B^{C}\) satisfying \(|x - \sqrt{t}z| \le |x|\). One can apply the mean value theorem, and use (2.27) to write
for \(z \in B^{C}\) satisfying \(|x - \sqrt{t}z| \le |x|\). Here, \(C_{\delta }\) is a positive constant which depends on \(\delta \). Next, we consider \(z\in B^C\) satisfying \(|x - \sqrt{t}z| \ge |x|\). Since \(\varphi _{*}\) is monotone decreasing in \(r = |x|\), we see that
for \(z\in B^C\) satisfying \(|x - \sqrt{t}z| \ge |x|\). From (3.9)–(3.11), we conclude that
where we set
Now, for \(z\in B\), it follows that
Thus, we have \(|z|^2 \ge \nu (1- \delta )^{2}\), and so
Second, we estimate \(u_{0}^I\) in the region \(\nu > 8 \log \ell \gg 1\). For \(0<a\ll 1\) we have
by partial integration on \(e^{-t}\). Note that \(t \delta ^{2} \nu = \delta |x|^{2} \le \delta \varepsilon ^{2} \ll 1\). Plugging the formula (3.15) into (3.14) yields
In the region \(\nu >8\log \ell \gg 1\), we have \(\ell = |\log t| < e^{\nu /8}\), yielding
If we take \(\delta > 0\) sufficiently small so that \(- (1-\delta )^{2}/4 +1/16 < -1/6\), we have, by (3.16) and (3.17), that
Thus, we get the first intermediate estimate \(u_0^I \le \nu e^{- \nu /6}\). Moreover, using the fact that \( |\log t| < e^{\nu /8}\) again, we obtain
and therefore,
Finally, observe that in the same region \(\nu >8\log \ell \gg 1\), we have \(r^{2} \gg t\) and
This together with (3.8), (3.12) and (3.18) yields that for \(\nu >8\log \ell \gg 1\), we have obtained
Thus it remains to estimate \(u_0^I\) in the region
Let \(\zeta :=z-x/\sqrt{t}\). On \(z\in B\), we have \(|\zeta |<\delta \sqrt{\nu }\ll \ell \). We shall show that
for some constant \(C >0\), which is independent of \(\nu \) and \(\delta \). We first consider the region \(1 < |\zeta |\). Since \(1 < |\zeta | \ll \ell \), we have \(\ell ^{-1} \log |\zeta | < \ell ^{-1} \log \ell \ll 1\). Thus, by (3.1) and the Taylor expansion, we have
Next, we consider the region \(|\zeta | \le 1\). Since \(- \log |\zeta |^{2} \ge 0\), we have
Using this together with (3.1), we obtain
From (3.21) and (3.22), we see that (3.20) holds.
Moreover, it follows from (3.13) that \(|z|^{2} \ge \nu (1 - \delta )^{2}\). This yields that
for some constant \(C > 0\), which is independent of \(\nu \) and \(\delta \). It follows from (3.23) and (3.20) that
Hence, by (3.8), (3.12) and (3.24), we have
where the remainder on B is estimated by
Here, we have used (3.13) in the second inequality. This yields that
From (3.7), (3.19) and (3.25), we see that (3.2) holds. \(\square \)
From Lemma 3.2, we have obtained, denoting \(\rho :=|\log r^2|\),
One can use the radial monotonicity of \(u_0\) (cf. Lemma A.1) to extend, to all \(x\in {\mathbb {R}}^2\), the bounds of the functions \(u_0e^{u_0^2}\) and \(u_0^2e^{u_0^2}\) as follows:
We shall show the bound of the function \(u_{0} e^{u_{0}^{2}}\). From the monotonicity of \(u_{0}\), it is enough to consider the region \(0< r < \varepsilon ^{2}\) only.
Note that we have \(\ell ^{-3/2}, \rho ^{-3/2} < |\log (t + r^{2})|^{-3/2}\) for \(\max (t, r^{2}) < \varepsilon ^{2}\). Moreover, it follows that \(\min (\frac{1}{t}, \frac{1}{r^{2}}) < 2(t + r^{2})^{-1}\). These together with (3.26) yield that
We first consider the region \(t + r^{2} < \varepsilon ^{2}\). It follows from (3.28) that
Next, we consider the region \(\varepsilon ^{2} \le t + r^{2} < 2\varepsilon ^{2}\). Since the function \(s^{-1} |\log s|^{-3/2}\) is decreasing for sufficiently small \(s>0\), we have, by (3.28), that
From (3.29) and (3.30), we see that \(u_{0}e^{u_{0}^{2}} \le C F_{0}\) for \(t < \varepsilon ^{2}\). We can obtain the bound of the function \(u_{0}^{2} e^{u_{0}^{2}}\) similarly. Thus, (3.27) holds.
Hence by the mean value theorem, for any functions \(v_0\) and \(v_1\), we have, for \(t<{ \varepsilon ^{2}}\), that
where \(f_{0}(u) = u (e^{u^{2}} -1)\).
To estimate the second iteration, we prepare the following:
Lemma 3.3
Let \(\alpha > 0\) and \(0< \varepsilon < 1\). For any \((t, r) \in (0, \varepsilon ^{2}) \times (0, \infty )\), there exists a positive constant \(C_{*}\) such that
Proof
From Lemma A.1, we have
Then, the integral is estimated using the following formula
where the variables are changed by \(s\rightarrow t-s\), \(r^2\rightarrow \sigma \rightarrow s\eta \). Then, we split the integral as follows.
We first estimate I. Since \(t -s + s \eta \le \sqrt{t}\) for \(0< s < t\) and \(0< \eta < 1/\sqrt{t}\), we have
Next, we estimate II. We note that \(|\log \min \{t - s + s\eta , \varepsilon ^{2}\}|^{- \alpha } \le (- \log \varepsilon ^{2})^{-\alpha }\) and \(t -s + s\eta \ge t\) for \(0< s < t\) and \(\eta \ge 1/\sqrt{t}\). Therefore, we have
From (3.34) and (3.35), we obtain (3.32). We can obtain (3.33) by the similar argument above. This completes the proof. \(\square \)
Using Lemma 3.3, we shall show the following:
Lemma 3.4
Let \(\varepsilon >0\) be given by Lemma 3.3. For any space-time function v on \((0,\varepsilon ^{2})\times {\mathbb {R}}^2\), let
Then, there exists a positive constant \(C_{0}\) such that
Moreover, for any \(C_1\in (0,\infty )\), there exists \(C_2\in (0,\infty )\) such that for any \(v_0\) and \(v_1\) satisfying
and for any \(\alpha { >} -1\), we have
Proof
Similarly, (3.31) and Lemma 3.3 yield
for any \(\alpha { >} -1\). Noting that the constants are independent of \(v_0,v_1,\alpha ,t\), we arrive at the desired conclusion. \(\square \)
We are now in a position to prove Theorem 3.1
Proof of Theorem 3.1
We put
and \(I[v] := D[v] - E[v]\). Then, we are naturally led to consider the mapping \(v\mapsto I[v]\) for v in the following set
for some constant \(T\in (0, { \varepsilon ^{2}})\) to be determined, which is a closed ball of a Banach space with the norm \(X_T^{1/2}\). The estimates on \(D[\cdot ]\) in (3.36) and (3.37) with \(C_1=1\) imply
for any \(v_0,v_1\in B_T^{1/2}\). Moreover, we can easily obtain
Hence if \(T>0\) is small enough then \(v\mapsto I[v]\) is a contraction mapping on \(B_T^{1/2}\), so there is a unique fixed point \(v\in B_T^{1/2}\). Then \(u=u_0+v\) is a local (mild) solution on \(0<t<T\) of
This completes the proof. \(\square \)
We can prove Theorem 1.1 immediately from Theorems 2.8 and 3.1 .
Proof of Theorem 1.1
We also denote \(\phi _{m_{*}}\), which is the stationary singular soliton to (1.5) obtained in Theorem 2.8, by \(\varphi _{*}\). Let \(u_{S}(t) = \varphi _{*}\) and \(u_{R}(t)\) be the regular solution to (1.5), obtained in Theorem 3.1. We see that \(u_{S}(0) = u_{R}(0) = \varphi _{*}\).
We shall show that \(u_{S}\) belongs to C([0, T), X) and becomes a mild solution for each \(T > 0\). As we mentioned in the proof of Theorem 2.8, \(\varphi _{*}\) satisfies (2.23) in a distributional sense and \(f_{m_{*}}(\varphi _{*}) \in L^{1}({\mathbb {R}}^{2})\). It follows from the \(L^{p}-L^{q}\) estimate of the heat kernel (see e.g. [19, Proposition 1]) that for \(1 \le p < \infty \), we obtain
Thus \(u_{S}(t) = \varphi _{*} \in C([0, T), X)\) satisfies the Duhamel formula:
Next, we shall show that \(u_{R}\) belongs to C([0, T), X) for sufficiently small \(T > 0\) and becomes a mild solution. The fixed point \(v \in B_{T}^{1/2}\) of I can be obtained as the limit of \(v_{n} = I[v_{n-1}]\) starting from \(v_{0} = 0\). Then for \(T > 0\) small enough, there exists \(M >0\) such that \(\sup _{n \in {\mathbb {N}}} |\log t|^{1/2}\Vert v_{n}(t)\Vert _{L^{\infty }} \le M\) and \(\lim _{n \rightarrow \infty }v_{n}(t) = v(t)\) in \(L^{\infty }({\mathbb {R}}^{2})\) for \(0< t < T\). Let \(u_{n} = u_{0} + v_{n}\). From the estimate (3.26) on \(u_{0}\) together with Lemma A.1 and that \(\Vert v_{n}(t)\Vert _{L^{\infty }} \le M \ell ^{-1/2}\), we obtain
uniformly for \(0< t < T \ll 1\), \(x \in {\mathbb {R}}^{2}\) and \(n \in {\mathbb {N}}\). Hence, there exists a constant \(C_{m_{*}} > 0\)
where \(f_{m}\) is defined by (2.17). Suppose that \(\sup _{0< s< T} \Vert u_{n}(s)\Vert _{L^{p}} < \infty \) for any \(p \in [1, \infty )\). Using (3.39) in the Duhamel formula
we obtain
for \(0< t < T \ll 1\). Here, we have used the \(L^{p}\) contraction estimate of the heat kernel (see e.g. [19, Proposition 1]) in the first inequality. Thus, we obtain by the induction argument
for \(n \in {\mathbb {N}}\). Since, for \(0< t < T \ll 1\),
the above uniform bound (3.41) together with Fatou’s lemma implies that,
for \(0< t < T \ll 1\). In addition, the same estimate as in (3.40) implies that \(f_{m_{*}}(u_{R}) \in L^{1} ([0, T), L^{p}({\mathbb {R}}^{2}))\) and by the Duhamel formula, \(u_{R} \in C([0, T), L^{p}({\mathbb {R}}^{2}))\). Thus, we conclude that \(u_{R} \in C([0, T), L^{p}({\mathbb {R}}^{2}))\) for all \(p \in [1, \infty )\). That is \(u_R\in C([0,T), X)\). Then, we see that \(u_{R}\) also satisfies (3.38).
For all \(t \in (0, T)\), we have \(u_{S}(t) = \varphi _{*} \notin L^{\infty } ({\mathbb {R}}^{2})\) while \(u_{R}(t) \in L^{\infty }({\mathbb {R}}^{2})\). This implies that \(u_{S}(t) \ne u_{R}(t)\) for all \(t \in (0, T)\). This completes the proof. \(\square \)
Notes
Equation (2.8) is obtained by removing the better decaying terms in right hand of (2.7) and by adding \(3/(16 \rho ^{2})\) in the linear potential. By adding the term, we can write the solution to (2.8) just by the trigonometric functions. Otherwise, we need to use the Bessel functions of order 1. Thus, adding the extra term makes the proof a bit simpler.
References
Brezis, H., Cazenave, T.: A nonlinear heat equation with singular initial data. J. Anal. Math 68, 277–304 (1996)
Buckmaster, T., Vicol, V.: Nonuniqueness of weak solutions to the Navier-Stokes equation. Ann. Math. 89, 101–144 (2019)
Davila, J., Del Pino, M., Wei, J.: Singularity formation for the two-dimensional harmonic map flow into \(S^{2}\). Invent. Math. 219, 345–466 (2020)
De Lellis, C., Székelyhidi, L.: The Euler equations as a differential inclusion. Ann. Math. 170, 1417–1436 (2009)
Germain, P., Ghoul, T.E., Miura, H.: On uniqueness for the harmonic map heat flow in supercritical dimensions. Comm. Pure Appl. Math. 70, 2247–2299 (2017)
Giga, Y., Kohn, R.V.: Characterizing blowup using similarity variables. Indiana Univ. Math. J. 36, 1–40 (1987)
Haraux, A., Weissler, F.B.: Nonuniqueness for a semilinear initial value problem. Indiana Univ. Math. J. 31, 167–189 (1982)
Ibrahim, S., Jrad, R., Majdoub, M., Saanouni, T.: Local well posedness of a 2D semilinear heat equation. Bull. Belg. Math. Soc. Simon Stevin 21, 535–551 (2014)
Ioku, N.: The Cauchy problem for heat equations with exponential nonlinearity. J. Differ. Equ. 251, 1172–1194 (2011)
Ioku, N., Ruf, B., Terraneo, E.: Existence, non-existence, and uniqueness for a heat equation with exponential nonlinearity in \({{\mathbb{R}}}^{2}\). Math. Phys. Anal. Geom. 18(1), 19 (2015)
Ioku, N., Ruf, B., Terraneo, E.: Non-uniqueness for a critical heat equation in two dimensions with singular data. Ann. Inst. H. Poincaré Anal. Non Linéaire 36(7), 2027–2051 (2019)
Matos, J., Terraneo, E.: Nonuniqueness for a critical nonlinear heat equation with any initial data. Nonlinear Anal. 55, 927–936 (2003)
Merle, F., Peletier, L.A.: Positive solutions of elliptic equations involving supercritical growth. Proc. Roy. Soc. Edinb. Sect. A 118, 49–62 (1991)
Mizoguchi, N., Souplet, P.: Optimal condition for blow-up of the critical \(L^{q}\) norm for the semilinear heat equation. Adv. Math. 355, 106763 (2019)
Ni, W.-M., Sacks, P.: Singular behavior in nonlinear parabolic equations. Trans. Amer. Math. Soc. 287, 657–671 (1985)
Ruf, B., Terraneo, E.: The Cauchy problem for a semilinear heat equation with singular initial data. Progr. Nonlinear Differ. Equ. Appl. 50, 295–309 (2002)
Terraneo, E.: Non-uniqueness for a critical non-linear heat equation. Comm. Partial Differ. Equ. 27, 185–218 (2002)
Weissler, F.B.: Local existence and nonexistence for semilinear parabolic equations in \(L^{p}\). Indiana Univ. Math. J. 29, 79–102 (1980)
Weissler, F.B.: Existence and nonexistence of global solutions for a semilinear heat equation. Isr. J. Math. 38, 29–40 (1981)
Acknowledgements
The authors would like to thank the anonymous referees for their comments. This work was done while H.K. was visiting at University of Victoria. H.K. thanks all members of the Department of Mathematics and Statistics for their warm hospitality. The work of S.I. was supported by NSERC grant (371637-2019). The work of H.K. was supported by JSPS KAKENHI Grant Number JP17K14223. K.N. was supported by JSPS KAKENHI Grant Number JP17H02854. The research of J.W. is partially supported by NSERC of Canada.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Y. Giga.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A: Maximum point of solutions to the linear Heat equation
Appendix A: Maximum point of solutions to the linear Heat equation
In this appendix, we shall give a proof of the fact, which is used in Sect. 3. More precisely, we show the following:
Lemma A.1
Let \(\phi \) be a radially decreasing function. Set \(u(t) = e^{t \Delta } \phi \). Then, u(t) is also radially decreasing and
Proof
Note that u is also radial. Setting \(v = \partial _{r}u\), we see that v satisfies the following:
We put \(v_{+} = \max \{v, 0\}\). Multiplying (A.1) by \(v_{+}\) and integrating the resulting equation over \({\mathbb {R}}^{2}\), we have
From the assumption, we infer that \(v_{+}(0) = 0\). This together with (A.2) yields that \(v_{+} \equiv 0\) in \((0, \infty ) \times {\mathbb {R}}^{2}\). This completes the proof. \(\square \)
Rights and permissions
About this article
Cite this article
Ibrahim, S., Kikuchi, H., Nakanishi, K. et al. Non-uniqueness for an energy-critical heat equation on \({\mathbb {R}}^2\). Math. Ann. 380, 317–348 (2021). https://doi.org/10.1007/s00208-020-01961-2
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00208-020-01961-2