Propagation of singularities for generalized solutions to nonlinear wave equations

The paper is devoted to regularity theory of generalized solutions to semilinear wave equations with a small nonlinearity. The setting is the one of Colombeau algebras of generalized functions. It is shown that in one space dimension, an initial singularity at the origin propagates along the characteristic lines emanating from the origin, as in the linear case. The proof relies on a fixed point theorem in the ultra-metric topology on the algebras involved. The paper takes up the initiating research of the 1970s on anomalous singularities in classical solutions to semilinear hyperbolic equations and transplants the methods into the Colombeau setting.


Introduction
This paper addresses propagation of singularities for solutions to semilinear wave equations with a small nonlinearity. The equations are of the form where ε is small positive parameter and f is smooth, polynomially bounded, and f (0) = 0. The initial data u 0 and u 1 are generalized functions of compact support, with a singularity at the origin. Approximating the initial data by nets of smooth functions (u ε 0 , u ε 1 ) ε∈(0,1] , we establish the existence of a net of smooth solutions (u ε ) ε∈(0,1] , up to an asymptotic error term of O(ε ∞ ). Measuring regularity in terms of estimates as ε ↓ 0, we show that the initial singularity propagates only along the two characteristic lines emanating from the origin (the one-dimensional light cone), but the solution remains regular inside (and outside) the light cone.
The paper is formulated in the framework of Colombeau generalized functions. This will allow us to use the powerful tools from this theory to combine generalized function data with nonlinearities and to measure regularity. Our main tool will be the Banach fixed point theorem in the socalled sharp topology, a complete ultra-metric topology on the Colombeau algebras. To our knowledge, this is the first time in the literature that such fixed-point arguments have been used to establish existence and regularity of solutions to nonlinear wave equations in the Colombeau framework.
Let us first put our result in perspective with regard to the classical literature. The discovery that in semilinear hyperbolic equations and systems, propagation of singularities does not necessarily occur along bicharacteristics emanating from singularities of the initial data, goes back to the paper [30] of Michael Reed, followed by the paper [28] of Jeff Rauch. Michael Reed showed that for the nonlinear wave equation (1.1) (with ε = 1) in one space dimension, propagation of singularities is the same as in the linear case. That is, if classical initial data are smooth except at the origin, the solution will be smooth off the characteristic lines emanating from the origin. Jeff Rauch showed that this is not the case for space dimensions d ≥ 2; there will be a loss of regularity inside the light cone. These two papers sparked a whole new research direction-the investigation of anomalous singularities in semilinear hyperbolic equations and systems, with a lot of activity until the 1990s (for a summary of that period, see [3]).
In order to describe the results of the paper, a few informal words about Colombeau algebras are required. Let Ω be an open subset of R d . Denote by O M (R) the space of smooth functions such that each derivative grows at most polynomially at infinity, and by D ′ (Ω) the space of distributions on Ω. Colombeau algebras are algebras of families (u ε ) ε∈(0,1] of smooth functions modulo asymptotically vanishing families, i.e., families all whose derivatives are of O(ε ∞ ) on compact sets as ε ↓ 0.
A family (u ε ) ε∈(0,1] represents an element of the Colombeau algebra The inclusion D ′ (Ω) ⊂ G(Ω) holds, constructed by cut-off and convolution with a mollifier, and C ∞ (Ω) is a faithful subalgebra. G(Ω) is invariant by superposition with maps f ∈ O M (R). For the purpose of describing regularity properties of the elements of G(Ω), we shall single out two subalgebras. First, the subalgebra G ∞ (Ω) is characterized by the property that on every compact set, all derivatives ∂ α u ε are O(ε b ) with the same b, depending only on the compact set. It holds that G ∞ (Ω) ∩ D ′ (Ω) = C ∞ (Ω). However, G ∞ (Ω) is not invariant under nonlinear maps (except polynomials), hence does not deliver a general framework for regularity theory for nonlinear equations. Second, the subalgebra G 0 (Ω) is characterized by the property that all derivatives ∂ α u ε are O(1) on every compact set. It also holds that G 0 ∩ D ′ (Ω) = C ∞ (Ω). In addition, G 0 (Ω) is invariant under superposition with arbitrary functions f ∈ C ∞ (R).
Existence of solutions to semilinear wave equations in Colombeau algebras has been known for a long time. For a globally Lipschitz nonlinearity f ∈ O M (R), existence and uniqueness of a solution to (1.1) in G(R 2 ) follows, for example, from [21]. In the globally Lipschitz case, existence and uniqueness of solutions in G(R d+1 ) has been shown in space dimensions d = 1, 2, 3, for example, in [5,27]. For power nonlinearities, existence and uniqueness results in an L 2 -based Colombeau algebra have been obtained in space dimensions d ≤ 9 with suitable bounds on the polynomial growth of f in [19].
Intrinsic regularity theory in Colombeau algebras (i.e., without recourse to distributional limits), started with the introduction of G ∞ (Ω) in [22]. It turned out that in linear partial differential equations, most of the classical regularity theory could be replicated with G ∞ (Ω) in place of C ∞ (Ω): elliptic regularity, hypoellipticity, microlocal elliptic regularity, wave front sets, propagation along bicharacteristics, including the techniques of pseudodifferential operators and Fourier integral operators with Colombeau amplitudes and phase functions [11,12,13,14,15,18,20,26]. In addition, the G ∞singular support of solutions to linear wave equations with discontinuous coefficients could be precisely located in one space dimension and for radially symmetric solutions in higher space dimensions [6,7,8,17]. However, as G ∞ (Ω) is not invariant under non-polynomial smooth maps, it could not be used for nonlinear equations.
So far, only few and special results on propagation of singularities in Colombeau solutions to nonlinear hyperbolic equations and systems have been obtained [9,23,24,25]. The present paper suggests that progress can be made in nonlinear hyperbolic equations when G 0 (Ω) is used for measuring regularity. In fact, we follow the historical path and present here Michael Reed's result in the setting of Colombeau algebras.
The main result of the paper is that in the one-dimensional semilinear wave equation, propagation of G 0 -singularities is the same as in the linear case, that is, a G 0 -singularity at the origin does not affect the G 0 -regularity inside the one-dimensional light cone. At this stage, we can prove this result only for small nonlinearities. This is dictated by the method of proof, following Michael Reed's fixed point argument, but in the ultra-metric topology on Colombeau algebras. In this ultra-metric topology, a map is a contraction if and only if it lowers growth in ε ↓ 0, whence the factor ε in equation (1.1).
By the way, it is clear that anomalous G 0 -singularities occur for (3 × 3)hyperbolic systems in the Colombeau setting as well. One just needs to adapt the example of [29] or [33]. In the spirit of the result of Jeff Rauch [28], we also present an example where singularities spread into the interior of the light cone in three space dimensions.
The plan of the paper is as follows. In Section 2 we recall the required notions from Colombeau theory. Section 3 contains the main result on propagation of singularities for problem (1.1). It concludes with remarks on possible extension and the mentioned example of anomalous propagation of singularities.

Colombeau algebras
We will employ the special Colombeau algebra of generalized functions denoted by G s in [16] (called the simplified Colombeau algebra in [2]). However, here we will simply use the letter G instead. This section serves to recall the definitions and properties required for our purpose. For more details, see e. g. [4,5,16,20,22].
Given a non-empty open subset Ω of R n , the space of real valued, infinitely differentiable functions on Ω is denoted by C ∞ (Ω), while C ∞ (Ω) refers to the subspace of functions all whose derivatives have a continuous extension up to the closure of Ω.
Let C ∞ (Ω) (0,1] be the differential algebra of all maps from the interval The algebra G(Ω) of generalized functions is defined as the factor space Restrictions to open subsets. Let ω be an open subset of Ω and U ∈ G(Ω). Then the restriction U |ω, obtained by restriction of representatives, is a well defined element of G(ω). The support of a generalized function U ∈ G(Ω), denoted by supp U , is the complement of the largest open set ω ⊂ Ω such that U |ω = 0. An analogous definition applies to elements U ∈ G(Ω) with the sets ω, on which U vanishes, taken as open in Ω. Similarly, the restriction of elements of U ∈ G(Ω) or U ∈ G(Ω) to lower dimensional linear subspaces can be defined. In our case, this will give a meaning to the initial values of The ring of generalized numbers. We let E M be the space of nets (r ε ) ε∈(0.1] of real numbers such that |r ε | = O(ε b ) as ε ↓ 0 for some b ∈ R. Similarly, N comprises those sequences which are O(ε a ) as ε ↓ 0 for every a ≥ 0. The factor space R = E M /N is the Colombeau ring of generalized numbers.
Generalized functions of bounded type. Let Ω be an open subset of R n and L a subset of Ω. A generalized function U from G(Ω) is called of bounded type on L, if it has a representative (u ε ) ε∈(0,1] such that The subalgebra G 0 (Ω) comprises the generalized functions from G(Ω) all whose derivatives of any order are of bounded type on compact sets. The G 0singular support sing supp G 0 is the complement of the largest open set ω ⊂ Ω such that U |ω belongs to G 0 (ω). The bounded type property is defined similarly for generalized functions in G(Ω), as is the subalgebra G 0 (Ω).
Superposition by smooth maps of polynomial growth. If f ∈ O M (R) and is a representative of U , then (f (u ε )) ε∈(0.1] belongs to E M (Ω) and its class in G(Ω) does not depend on the choice of representative of U . In addition, G 0 (Ω) is invariant under superposition by arbitrary smooth maps.
Integration. Let K be a compact subset of Ω and U ∈ G(Ω). Then K U (x)dx, again defined on representatives, is a well defined element of R.
A proof of this fact can be found in [5]. Similarly, if Γ is a smooth, bounded curve in Ω, then Γ U dξ, where dξ denotes the line element, is a well defined element of R, see [1].
The sharp topology. The sharp topology, introduced in [2], can be defined through its family of neighborhoods We are going to recall that the sharp topology on G(Ω) can be defined in terms of an ultra-metric. This construction is due to [31,32] and has been further developed by [10]. Let (K n ) n∈N be an exhausting sequence of compact subsets of Ω. The seminorms µ n on C ∞ (Ω) are given by Obviously, (u ε ) ε∈(0,1] belongs to N (Ω) if and only if ν n (u ε ) ε∈(0,1] = ∞ for all n. Thus the valuations can be extended to the factor algebra G(Ω). The following properties hold: By means of these evaluations, a family of ultra-pseudo-seminorms on G(Ω) can be defined by p n (U ) = exp(−ν n (U )).
They have the properties (c) p n (λU ) = p n (U ) for all λ ∈ R.
Finally, an ultra-metric can be defined on G(Ω) by It is an easy exercise to show that the topology induced on G(Ω) by the ultrametric d is the same as the one given by the neighborhoods (2.1). Further, it is known [10,20,31] that G(Ω) with the uniform structure induced by d is complete, hence a complete ultra-metric space.
In particular, G 0 (Ω) is contained in the unit ball around zero in G(Ω).

Propagation of singularities
This section is devoted to presenting and proving our main result about propagation of singularities in the generalized solution to the Cauchy problems (1.1). We interpret problem (1.1) as in the Colombeau algebra of generalized functions. We make the following assumptions.
Before proving Theorem 3.1, we first explain some notation and definitions that we will use. Let D + be the directional derivative in the direction t +x, wheret andx are unit vectors in the t and x directions, and let D − be the directional derivative in the directiont −x. Put By assumption (C), there exists a > 0 such that

We remark that properties (I) and (II) imply that
We now introduce an ultra-metric d on G([0, T ] × R). Let (K ± n ) n∈N and (K n ) n∈N be exhausting sequences of compact subsets of [0, T ] × R \ Γ ∓ and [0, T ] × R, respectively. Take seminorms µ n on C ∞ ([0, T ] × R) which are defined by We next prove the following two lemmas.
Let dξ be the line element. Then for any A ∈ G([0, T ] × R), Proof. We will give the proof for B + . The proof for B − is similar. Let (a ε ) ε∈(0,1] be a representative of A. Then B + has a representative b ε a ε (ξ) dξ.

For any
This inequality and the assumption that A satisfies property (I) yield that B + satisfies property (I).
A simple calculation shows that Similarly, it holds that D α + b ε + = D α−1 + a ε for any α ∈ N. It follows from the assumption that A satisfies property (II) that B + satisfies property (II).
The D − -derivative of b ε + is calculated as follows: Similarly for any α ∈ N

Then for any
The first term on the right-hand side is uniformly bounded in ε, since A satisfies property (III).
This implies that for any Therefore, B + satisfies property (III). It is immediate to check that B + satisfies property (IV). Thus B + belongs to G([0, T ] × R).
Proof. By assumption, the integral (3.4) satisfies properties (IV). Since V , W satisfy property (I), so does the integral (3.4). Hence it remains to show that the integral (3.4) satisfies properties (II) and (III). By Lebesgue's dominated convergence theorem, we get On the other hand, noting that by assumption, We now turn to the proof of Theorem 3.1.
Proof of Theorem 1.1 . As mentioned in the introduction, we follow the ideas of Reed's article [30], in which the phenomenon of propagation of singularities in classical solutions to problem (1.1) with ε = 1 has been studied.
We fix T > 0 arbitrarily and consider the Cauchy problem (3.1). As stated above, by assumption (C), we have with some a > 0. Put V = ∂ t U − ∂ x U and W = ∂ t U + ∂ x U . By assumption (B), we have f (0) = 0. Hence, by finite propagation speed, we can expect that U vanishes on {(t, x) ∈ [0, T ] × R : |x| > t + a}). If that is the case, then U is expressed in the form and so problem (3.1) can be rewritten as the Cauchy problem for a first-order hyperbolic system where We may rewrite problem (3.9) as the pair of integral equations This can be seen as follows.
In fact, if F is a contraction on M([0, T ] × R), then by the Banach fixed point theorem, F has a fixed point (V, W ) ∈ M([0, T ] × R). We define Noting that (V, W ) satisfies problem (3.9), we find that U is a solution of problem (3.1). That U satisfies (3.2) can be seen as follows. Since V , W satisfy property (I), so does U . Hence for any K ⋐ [0, T ] × R, the function U is of bounded type on K. The fact that (V, W ) satisfies problem (3.9) gives the relationships and so for any α ∈ N It follows from the assumption that V , W satisfy properties (II) and (III) that for any K ⋐ [0, T ] × R \ (Γ − ∪ Γ + ) and α ∈ N, the functions D α ± U are of bounded type on K. Since the first equation in problem (3.1) can be rewritten as for any K ⋐ [0, T ] × R \ (Γ − ∪ Γ + ) and α, β ∈ N, the function D α − D β + U is of bounded type on K. Thus we obtain that i.e., assertion (3.2) follows.
By (3.14) and (3.15), the mapping F = (F 1 , F 2 ) satisfies This leads to Note that any countable subfamily of { p n } gives the same topology, since { p n } is increasing. Hence where the factor 2 comes from the shift j → j + 1. Since ν 0 (E) > log 2, it follows that F is a contraction.  Furthermore, U ′ 0 , U 1 are of bounded type on R, then for any T > 0, there exists a solution U ∈ G([0, T ] × R) to problem (3.1) such that We complete this section by an example of a solution to a semilinear wave equation in three space dimension, in which the initial singularity does not propagate along the light cone, but rather contaminates the interior. Example 3.6. Let Clearly, (u ε 0 ) ε∈(0,1] represents an element U 0 ∈ G(R 3 ) ∩ G 0 (R 3 \ {0}). Let U ∈ G([0, ∞) × R 3 ) be represented by the same family (3.16), i.e., U does not depend on t. A simple calculation shows that U solves the initial value problem and sing supp G 0 U = {(t, x) : x = 0, t ≥ 0}.