São Paulo Journal of Mathematical Sciences 7, 2 (2013), 201–239

Nonlinear Generalized Functions: their origin,

some developments and recent advances

Jean F. Colombeau
Institut Fourier, Université de Grenoble (retired)

Abstract. We expose some simple facts at the interplay between math-

ematics and the real world, putting in evidence mathematical objects
” nonlinear generalized functions” that are needed to model the real
world, which appear to have been generally neglected up to now by
mathematicians. Then we describe how a ”nonlinear theory of gen-
eralized functions” was obtained inside the Leopoldo Nachbin group
of infinite dimensional holomorphy between 1980 and 1985 **. This
new theory permits to multiply arbitrary distributions and contains
the above mathematical objects, which shows that the features of this
theory are natural and unavoidable for a mathematical description of
the real world. Finally we present direct applications of the theory
such as existence-uniqueness for systems of PDEs without classical so-
lutions and calculations of shock waves for systems in non-divergence
form done between 1985 and 1995 ***, for which we give three examples
of different nature (elasticity, cosmology, multifluid flows).

* work done under support of FAPESP, processo 2011/12532-1, and

thanks to the hospitality of the IME-USP.
** various supports from Fapesp, Finep and Cnpq between 1978 and 1984.
*** support of ”Direction des Recherches, Études et Techniques”, Ministère
de la Défense, France, between 1985 and 1995.
jf.colombeau@wanadoo.fr

1. Mathematics and the real world. Let H be the Heaviside func-

tion. It is a function equal to 0 if x < 0, to 1 if x > 0 and not defined if
x = 0. For p, q ∈ N let us compute the integral
Z +∞
I= (H p (x) − H q (x))H 0 (x)dx. (1)
−∞

201
202 J.F. Colombeau

Physicists solve the problem with formal calculations:

H p+1 +∞ H q+1 +∞ 1 1
I=[ ]−∞ − [ ]−∞ = − 6= 0.
p+1 q+1 p+1 q+1
The justification is that the Heaviside function H is an idealization of a
smooth function whose values jump from 0 to 1 on a very small interval:
before idealization the calculations on the smooth function are well defined
and the result should of course be retained after idealization. Idealization
is needed for simplification in the manipulation of the mathematical models
of physics. Therefore the concept of Heaviside function is needed but one
is forced to consider that H p 6= H q if p 6= q.

In Lebesgue integration theory, therefore in Schwartz distribution the-

ory, one has H p = H q even if p 6= q since they differ only at the point
x = 0 which has Lebesgue measure 0. But one cannot multiply by H 0 and
integrate as requested in (1).

To be more convinced is it possible to visualize different Heaviside func-

tions from direct observation of the real world? The answer is yes: it suf-
fices to have in mind compound shock waves which you can produce with
a solid metallic ruler and a hammer if you knock strongly enough (figure 1).

The shock wave that propagates in the ruler is made of an elastic region
followed by a plastic region. In the elastic region the ”stress” S of the
electrons increases but they remain able to maintain the cristalline structure
of the metal. The plastic region begins when the electrons become unable
to maintain this structure: then in absence of their action the solid behaves
like a fluid with definitive deformation and possible breakings. Therefore if
the physical variables density and velocity vary throughout the shock wave
(along both regions) the stress varies only in the elastic region (in which
it increases in absolute value) and then remains fixed at its maximum (in
absolute value) in the plastic region. A 1-D model is provided by the set
of equations
ρt + (ρu)x = 0, (2)

(ρu)t + (ρu2 )x + (p − S)x = 0, (3)

St + uSx − k 2 (S)ux = 0, k 2 (S) = 1 if |S| < S0 , k 2 (S) = 0 if |S| ≥ S0 , (4)

p = const.ρ, (5)

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 203

where ρ=density, u=velocity, p=pressure, S=stress deviation tensor. |S|

increases in the elastic region; it stops when it reaches the value S0 : then

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

204 J.F. Colombeau

the material is in the plastic region. (2) and (3) are respectively the con-
tinuity and the Euler equations; (4) is a differential form of Hooke’s law
in which passage from the elastic region to the plastic region is modeled
in the Lamé coefficient k 2 , (5) is some simplified state law linking pressure
and density.
R
The integral u(x, t)Sx (x, t)dx (used to try to compute a jump condi-
tion for equation 4) has very different values depending on the Heaviside
functions representing u and S. For very strong shocks the elastic region is
small relatively to the plastic region and therefore the Dirac delta distribu-
tion Sx is located on the ” foot” of theR Heaviside function that represents
the jump of u, therefore the integral u(x, t)Sx (x, t)dx is close to 0 if u=0
on this foot. On the other hand for weak shocks the plastic region can be
void: the shock is purely elastic and the variables ρ, u, S vary throughout
the shock wave: admitting roughly
R they are represented by the same Heav-
iside function H the integral u(x, t)Sx (x, t)dx has a far from zero value.
See [30, 18] for details.

Here is an example of another idealization that does not make sense

in distribution theory. It appears in form of a square root of the Dirac
delta distribution. The Dirac delta distribution can be asymptotically rep-
resented by a function equal to 1 on [− 2 , 2 ] , 0 elsewhere; what about
the object asymptotically
√ represented by the square root of this object? It
should be denoted by δ. In the viewpoint of Schwartz distribution theory
such an object should √be 0 because the integral limited √ by its graph and
the x-axis is equal to  → 0. Does such an object δ exist in the real
world?

In general Relativity it was discovered from formal calculations that the

Reissner-Nordstrom field of a charged ultra-relastivistic (very fast rotat-
ing) black hole is null while the energy of this (null) field is not null! R.
Steinbauer [87] solved this paradox√ by discovering that the field itself is
a mathematical object such as δ above, which looks null in distribution
theory but whose square (the energy of the field) is similar to a Dirac delta
distribution i.e. not at all null.

The Keyfitz-Kranzer system of singular shocks

ut + (u2 − v)x = 0, (6)

u3
vt + ( − u)x = 0, (7)
3

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 205

has been studied in [58, 59, 80] and the observed solutions from numeri-
cal tests are given in [59, 80]. Here is a depiction of a singular shock wave:

30

20

10

-10

-20

-30

1.305 1.31 1.315 1.32 1.325 1.33 1.335 1.34

4
x 10

1800

1600

1400

1200

1000

800

600

400

200

0
1.31 1.315 1.32 1.325 1.33
4
x 10

Figure 2. A numerical solution of the Keyfitz-Kranzer system. On the top panel

one observes the function u with the two small peaks "square root of Dirac
distribution on both sides of the discontinuity. On the bottom panel one
observes the function v which carries a Dirac distribution on the discontinuity.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

206 J.F. Colombeau

In [58] p.426, explicit solutions are found in the form: ξ = xt , s = shock

speed,
ξ−s a ξ−s
u(x, t, ) = u0 + (u1 − u0 )h( p ) + √ ρ( ),
  

ξ−s a2 ξ−s
v(x, t, ) = v0 + (v1 − v0 )h( p ) + ρ2 ( ),
  
where ρ ∈ Cc∞ (] − 1, +1[), ρ2 = 1, ρ3 = 0, xρ(x) ≤ 0, h a C ∞ monotone
R R
approximation of the Heaviside function, h0 ∈ Cc∞ (] − 1, +1[), p > 0 a real
number. The function x 7−→ t1 ρ2 ( ξ−s  ) is a Dirac delta measure centered
R 1 2 xt −s
at x = st (indeed  ρ (  )dx = t ρ2 (y)dy = t). Therefore the term
R
√√ √
√a

ρ( ξ−s
 ) has the form a t δ. Contrarily to the standard notation δ this
R 3
term has a negative and a positive small peak due to the fact ρ (x)dx = 0
which forces the function ρ to have positive and negative values that can
be observed in figure 2.

As
√ indicated above the function u in figure 2 shows small peaks in form
of ” δ”. At the limit the space step tends to 0 they tend to 0 in the sense
of distributions. If we neglect them and state the solution as
u(x, t) = ul + [u]H(x − ct), (8)

v(x, t) = vl + [v]H(x − ct) + a(t)δ(x − ct), (9)

with a(t) a nonzero function of t, then in (6) ut shows a Dirac delta dis-
tribution from (8), u2 − v shows a Dirac delta distribution since a(t) 6= 0
therefore (u2 − v)x shows a term a(t)δ 0 . Therefore ut + (u2 − v)x shows
terms a(t)δ 0 + bδ with a(t) 6= 0 which is impossible from (6) since δ and δ 0
are linearly independent: equation (6) cannot be satisfied. Similarly con-
cerning (7): vt shows a nonzero δ 0 term (the observed value c is different
3
from 0) while from (8) ( u3 − u)x cannot show a δ 0 term: (7) cannot be
satisfied.
√
This proves that the δ small peaks in u cannot be omitted to explain
that the observed u, v are solutions of the equations (6, 7). In their presence
the above contradictions disappear at once: for instance in (6) u2 presents
now a δ that can cancel the δ from v and thus (6) can be satisfied.

Conclusion. New mathematical objects such as different Heavi-

side functions and square roots of Dirac delta distributions should

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 207

be introduced in mathematics in order to model real world situ-

ations.

Soon after 1950 when L. Schwartz ”proved” that ”multiplication of dis-

tributions is impossible in any mathematical context, possibly different of
distribution theory” mathematical physicists believed that the problems
with formal mathematical objects introduced by physicists that were not
distributions should be solved by modifying the formulation of physics so
as to model physics completely within the setting of distributions. Today
this viewpoint has completely failed. The unique option now is to introduce
these objects imposed by the real world into mathematics. Such introduc-
tion was done about 1980 inside the research group of Professor Leopoldo
Nachbin in Brazil.

2. Introduction and study of ”new generalized functions” in-

side the Leopoldo Nachbin group (J. Aragona, H. A. Biagioni, J. F.
Colombeau, João Toledo, R. Soraggi). About 1980 the group of Profes-
sor Leopoldo Nachbin was studying holomorphic functions defined on lo-
cally convex spaces (and also approximation theory not considered here)
[14, 15, 63, 66, 95].

Let Ω be an open set in Rn , let D = D(Ω) = Cc∞ (Ω) denote the space of all
C∞ functions on Ω with compact support and E 0 = E 0 (Ω) denote the space
of all distributions on Ω with compact support. These spaces are equipped
with their classical locally convex topologies defined in Schwartz distribu-
tion theory (the reader who does not know topological vector spaces can
follow the sequel). One can use as well holomorphic functions over E 0 (Ω)
and D(Ω). I chose C ∞ functions [26] because this is more general and will
give more numerous new mathematical objects. I did two very simple re-
marks.

Remark 1. C ∞ (E 0 ) ⊂ C ∞ (D).

Proof. This inclusion is obtained by the restriction map of a function

defined on E 0 to a function defined on D since it is well known in Schwartz
distribution theory that D is contained and dense in E 0 .

Remark 2. If we define an equivalence relation on C ∞ (E 0 ) by

Φ1 RΦ2 if f Φ1 (δx ) = Φ2 (δx ) ∀x ∈ Ω

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

208 J.F. Colombeau

where δx denotes the Dirac measure at the point x, then the set of all
equivalence classes is algebraically isomorphic to the classical algebra of all
C ∞ functions on Ω.

Proof. It is easy to prove that the map x 7−→ δx is C ∞ from Ω into

E0 therefore the map x 7−→ Φ(δx ) is C ∞ from Ω into C. Now the sur-
jection follows from the fact that if f ∈ C ∞ (Ω), f can be considered as
the class of the map Φ defined by Φ(T ) =< T, f > if <, > is the duality
brackets between E 0 and C ∞ (Ω). The injection follows from the fact that if
f1 6= f2 ∃x ∈ Ω / f1 (x) 6= f2 (x).

Therefore from these two remarks one has the diagram

quotient
C ∞ (E 0 (Ω)) 7−→ C ∞ (Ω)

↓ restriction

C ∞ (D(Ω)).

The natural idea is to close the diagram.

How to close the diagram? If Φ ∈ C ∞ (D(Ω)), Φ(δx ) does not make

Rsense. One naturally considers a sequence ρ,x → δx , ρ,x ∈ D(Ω),
1 λ−x 0
Rn ρ(x)dx = 1, ρ,x (λ) = n ρ(  ). Then ρ,x → δx in E (Ω). A nat-
ural idea to extend the relation R is to replace Φ(δx ), Φ ∈ C ∞ (E 0 (Ω)), by
Φ(ρ,x ), Φ ∈ C ∞ (D(Ω)). The statement ”Φ(ρ,x ) → 0” would not lead to
an algebra, so a more subtle definition had to be chosen. This was done as
follows in 1980-82 inside the group of Professor Leopoldo Nachin. If q ∈ N
set
Z Z
n
Aq := {ρ ∈ D(R ) / ρ(x)dx = 1, xi ρ(x)dx = 0 ∀i ∈ Nn , 1 ≤ |i| ≤ q}.
Rn Rn
(10)
Then if Φ ∈ C ∞ (E 0 (Ω)), Φ(δx ) = 0 is equivalent to

∀ ρ ∈ Aq ∀K ⊂⊂ Ω ∃C > 0, η > 0 / Φ(ρ,x ) = O(q+1 ) . (11)

A proof can be found in [24] p. 99-100, [28] p. 58-60. This character-

ization is basic since it puts in evidence an ideal in the subalgebra of the

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 209

algebra C ∞ (D(Ω)) of all C ∞ functions on D(Ω) made of those Φ such that

1
∀ K ⊂⊂ Ω ∃N ∈ N / supx∈K |Φ(ρ,x )| = O( N ) ∀ ρ ∈ A1 . (12)

This subalgebra was called the subalgebra of the ”moderate” elements of
C ∞ (D(Ω)), from the bounds in O( 1N ). Further one has to insert all partial
x-derivatives in these definitions. Finally it was possible to close the dia-
gram:

quotient
C ∞ (E 0 (Ω)) 7−→ C ∞ (Ω)

↓ ↓

quotient
suitable subalgebra of C ∞ (D(Ω)) 7−→ G(Ω).

This new space G(Ω) is a differential algebra i.e. presence of partial

derivatives and of a multiplication with all the usual algebraic properties.
We have the sequence of algebraic inclusions:

C ∞ (Ω) ⊂ C 0 (Ω) ⊂ D0 (Ω) ⊂ G(Ω). (13)

The partial derivatives in G(Ω) induce the partial derivatives in D0 (Ω) which
is a vector subspace of G(Ω). C ∞ (Ω) is a faithful subalgebra of G(Ω).

Two delicate points, that will be basic for the understanding of nonlin-
ear generalized functions appear at once: what is the real situation of the
algebra C 0 (Ω) in (13) and where are the different Heaviside functions we
need to represent the real world? Up to now one can check that C 0 (Ω) is
not a subalgebra of G(Ω) and that powers in G(Ω) of the Heaviside dis-
tribution ∈ D0 (Ω) ⊂ G(Ω) are not in D0 (Ω). These two points are due to
the Schwartz impossibility result we will consider later and will be (quite
satisfactorily) solved below.

The detailed definitions that lead to the definition of G(Ω) are as follows.
|k|
First we recall the notation Dk = k1∂ kn .
∂x1 ...∂xn

The subalgebra of the ”moderate” elements of C ∞ (D(Ω)) is defined as:

{Φ ∈ C ∞ (D) / ∀K ⊂⊂ Ω ∀k ∈ Nn ∃N ∈ N / ∀ρ ∈ A1 ∃C, η > 0 /
C
|Dk Φ(ρ,x )| ≤ N ∀ ∈]0, η[, ∀x ∈ K}.


São Paulo J.Math.Sci. 7, 2 (2013), 201–239

210 J.F. Colombeau

The kernel defining the equivalence relation is defined as:

{Φ ∈ C ∞ (D) / ∀K ⊂⊂ Ω ∀k ∈ Nn ∃N ∈ N /∀q > N, ∀ρ ∈ Aq ∃C, η > 0 /
|Dk Φ(ρ,x )| ≤ Cq−N ∀ ∈]0, η[, ∀x ∈ K}.
Then:
subalgebra of moderate elements
G(Ω) = .
above kernel
G(Ω) is often called the ”full algebra” of nonlinear generalized functions.
M. Grosser (Vienna) discovered later that the definition of the kernel could
be somewhat simplified but anyway these definitions are complicated with
too many quantifiers. A subalgebra of G(Ω), denoted by Gs (Ω) and called
”simplified” or ”special”, was constructed for this purpose. The canoni-
cal character of the inclusion of D0 (Ω) into G(Ω) is lost in the simplified
case. Also, the simplified, and even the full case, can be formulated (in a
very closely related form, not exactly equivalent) without the use of C ∞
or holomorphic functions defined over locally convex spaces, which makes
a departure from the main theme of Prof. Leopoldo Nachbin group, but,
later, one came back to C ∞ or holomorphic functions defined over locally
convex spaces for nonlinear generalized functions on manifolds. Similarly
the classical theory of distributions can be presented with locally convex
spaces or also even without infinite dimensional spaces. The simplified or
special algebra Gs (Ω) has been more used than the full algebra. It is defined
by a quotient like the full algebra.

Reservoir of representatives =
{f : Ω×]0, 1] 7−→ C, C ∞ in x f or any f ixed ,
C
/ ∀K ⊂⊂ Ω ∀k ∈ Nn ∃N, C > 0 / |Dk f (x, )| ≤ ∀ ∈]0, 1[, ∀x ∈ K}.
N
Kernel =
{f ∈ Reservoir / ∀K ⊂⊂ Ω ∀k ∈ Nn ∀q large enough, ∃Cq > 0 /
|Dk f (x, )| ≤ Cq q ∀ ∈]0, 1[, ∀x ∈ K}.
Then:
Reservoir
Gs (Ω) = .
Kernel
The full algebra as defined above is based on infinite dimensional analy-
sis (holomorphic or C ∞ functions defined over locally convex spaces) in the
same way as Schwartz presentation of distributions uses also infinite di-
mensional analysis (continuous linear functions defined over locally convex
spaces). The full algebra appears as a faithful extension of Schwartz pre-
sentation: a large functional space C ∞ (D(Ω)) of nonlinear functions and a

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 211

quotient that reduces on the small subspace L(D(Ω), C) of linear functions

(= the distributions) to absence of quotient, and thus gives back exactly
the definition of distributions as a particular case. The simplified algebra
does not use infinite dimensional analysis and is a direct extension of the
sequential definition of distributions that we recall now. To define the dis-
tributions on an open set Ω ⊂ Rn following the sequential approach one
first considers the set of all families {f }∈]0,1] of C ∞ functions on Ω such
that
Z
∀φ ∈ Cc∞ (Ω) f (x)φ(x)dx → a limit in R or C when  → 0.

This set forms obviously a vector space. On this vector space one defines
the equivalence relation
Z
{f }R{g } ⇔ ∀φ ∈ Cc∞ (Ω) (f − g )(x)φ(x)dx → 0 when  → 0.

The distributions are the equivalence classes. These distributions are the
same objects as the distributions defined by Schwartz using the theory of
locally convex spaces. Of these two presentations one could ask: which is
the best? The question has already no answer in distribution theory. The
success of Schwartz has occulted the sequential presentation (Mikusinski
and co-authors) but now one tends to drop the theory of locally convex
spaces, thus getting closer and closer to the sequential approach.

Prof. Leopoldo Nachbin made me publish 3 books in his collection ”Notas

de Matematica” published by North Holland Pub. Company, now Elsevier:
”Differential Calculus and Holomorphy ”, 1982, to prepare the background
of C ∞ functions over locally convex spaces,
”New Generalized Functions and Multiplication of Distributions”, 1984, to
introduce the full algebra G(Ω),
”Elementary Introduction to New Generalized Functions”, 1985, to intro-
duce the simplified algebra Gs (Ω).
Prof. Leopoldo Nachbin presented me to Prof. L. Schwartz who supported
this theory [25], and 4 members of his group, Jorge Aragona, Hebe Biagioni,
João Toledo and Roberto Soraggi started working on the field of nonlinear
generalized functions thus created [4, 18].

In his famous book ”Théorie des distributions”, L. Schwartz claims: ”

multiplication of distributions is impossible in any mathematical context,
possibly different from distribution theory”. Since D0 (Ω) ⊂ G(Ω) canoni-
cally, or D0 (Ω) ⊂ Gs (Ω) non canonically, the algebraic structure of G(Ω) or
Gs (Ω) points out a multiplication of distributions. Where is the mistake?
The first obvious answer is that the above claim is not a theorem, but the

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

212 J.F. Colombeau

interpretation of a theorem. So one has to go back to the Schwartz theorem

that we recall now.

Theorem (Schwartz 1954 [89]). Let A be an associative algebra contain-

ing the algebra C(R) as a subalgebra. Let us assume

• the map (x 7−→ 1) ∈ C(R) is unit element in A,

• there exists a map D : A 7−→ A called derivation such that
D(uv) = Du.v + u.Dv, and D|C 1 (R) is the usual derivative.

Then D2 (x 7−→ |x|) = 0 in A instead of 2δ as it should be.

One checks at once that all assumed properties hold in the context of
G(R), or Gs (R), except that the classical algebra C(R) is not a subalgebra
of G(R) or Gs (R), although C ∞ (R) is. Let f, g ∈ C(Ω). Then the new
product f • g ∈ G(Ω) is the class, modulo the kernel defining G, of the map
D(Ω) 7−→ C whose restriction to the set of the ρ,x is
Z Z
ρ,x 7−→ f (x + µ)ρ(µ)dµ. g(x + µ)ρ(µ)dµ.

On the other hand the classical product f.g, when considered in G(Ω)
through the inclusion map C(Ω) ⊂ G(Ω), is the class of
Z
ρ,x 7−→ f (x + µ)g(x + µ)ρ(µ)dµ.

The difference (f • g − f.g)(ρ,x ) → 0 when  → 0. From the two definitions

of the kernel above (in the full or simplified case) this is not a sufficient
smallness to ensure that the difference (f • g − f.g) is in any of the kernels
since a fast tendance to 0 (of the form o(q ) ∀q) is requested. This why the
two products are different.

Why f • g = f.g when f, g ∈ C ∞ (Ω)? This follows from Taylor’s formula

Rand the definition of the sets Aq . Indeed
0
[f (x)+µf (x)+· · ·+q µq f (q) (x)+q+1 µq+1 r(x, )]ρ(µ)dµ = f (x) ρ(µ)dµ+
R

f 0 (x) µρ(µ)dµ + · · · + q f (q) (x) µq ρ(µ)dµ + q+1 r(x, )µq+1 ρ(µ)dµ =

R R R

f (x) + q+1 r(x, )µq+1 ρ(µ)dµ = f (x) + O(q+1 ) if ρ ∈ Aq .

R

Therefore if f, g ∈ C ∞ and ρ ∈ Aq

(f • g − f.g)(ρ,x ) = O(q+1 )

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 213

which shows that the map (rather, the map defined on D(Ω) whose restric-
tion to the set of all ρ,x is):
ρ,x 7−→ (f • g − f.g)(ρ,x )
is in the kernel defining G(Ω) if f, g are C ∞ functions.

If f, g ∈ C n (Ω) one has only

(f • g − f.g)(ρ,x ) = O(n+1 )
if ρ ∈ Aq , whatever q > n, therefore f • g − f.g 6∈ kernel.

This defines various concepts of smallness in G(Ω), Gs (Ω). A convenient

concept which generalizes the concept defining the distributions is called
association and is defined as follows:

Definition. We say that G1 , G2 ∈ G(Ω) are associated iff they have re-
spective representatives Gi (ρ,x ), i = 1, 2 (then it works for arbitrary repre-
sentatives) such that
Z
∀ψ ∈ D(Ω), ∀ρ ∈ AN , N large enough, (G1 (ρ,x )−G2 (ρ,x ))ψ(x)dx → 0
Ω
(14)
when  → 0.

We note
G1 ≈ G2
to express that G1 and G2 are associated. We have
√
H n ≈ H ∀n, δ ≈ 0, (15)
if H is the Heaviside function and δ the Dirac delta distribution. From the
definition of association
∂ ∂
G1 ≈ G2 implies G1 ≈ G2 . (16)
∂x ∂x
But
G1 ≈ G2 does not imply G.G1 = G.G2 : (17)
√
it suffices to consider G1 = G = δ, G2 = 0.

Of course for every distribution T there are many generalized functions

G ∈ G(Ω) associated to T . In the sense of the definition of association they
have the same ”macroscopic
√ aspect” as T , but can differ completely from
T by deeper properties: δ and 0 for instance.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

214 J.F. Colombeau

Conclusion. In G(Ω) we have two very different concepts that

play the role of the classical equality:

• of course the ”true” equality in G(Ω) which is coherent with

all operations: multiplication, differentiation,

• the association, which is not an equality since different gen-

eralized functions can be associated, but extends faithfully the
concept of equality of distributions to this more subtle setting.
The association is coherent with differentiation but incoherent
with multiplication, as a reproduction of the properties of the
equality of distributions.

Note a property that could look as a paradox: we have an inclusion of

the space of distributions into the differential algebras of nonlinear gener-
alized functions (canonical in the full case, noncanonical in the simplified
case); then two generalized functions which are in the image of the space
of distributions are associated iff they are equal in G(Ω): this follows from
the fact that the other associated generalized functions are excluded since
they are not in this image.

One can interpret the classical products of distributions through the as-
sociation:

• the Schwartz product f.T, f ∈ C ∞ (Ω), T ∈ D0 (Ω) and the new product
f • T ∈ G(Ω) are associated [28, 24];

• the same holds for the Mikusinski product [28], and more particularly
for the Hormander product [28, 93] due to a result of [93].

• More generally the following coherence between the classical calcula-

tions and the calculations in G(Ω) has always been observed: whenever a
sequence of calculations makes fully sense in classical mathematics or in
distribution theory, the analog calculations in the G-context give a result
associated to the classical result.

Now one can understand the Schwartz impossibility result: the assump-
tion that the classical algebra C(R) is a subalgebra of the algebra A is
stronger than needed to ensure coherence of the new calculus in A and
the classical calculus. Indeed coherence concerns only the calculations that
fully make sense, from their beginning to their end, in classical mathematics

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 215

and distribution theory. In the case A = G(R) this coherence is obtained:

with the new product of continuous functions one obtains in G(R) a result
which is simply more precise than the classical result and it suffices to drop
this refinement (using the association) to have exactly the classical result.
The Schwartz assumption that C(R) would be a subalgebra of the algebra
A means that one could use the classical product even outside the domain
of classical calculations. For instance, in a sequence of calculations such
that one at least does not make sense classically or in distribution theory,
consider a previous calculation of the sequence that makes sense classically
such as a product of two continuous functions: if the Schwartz assumption
C(R) subalgebra of A would be satisfied one could use the classical product:
this is not the case because of the nonclassical calculations that follow this
classical calculation. One should use in this case the new product of contin-
uous functions: an infinitesimal difference such as H 2 −H can be multiplied
by an infinite quantity such as H 0 and give a significant result. Since this
difficulty concerns a sequence of calculations in which one at least does not
make sense within classical mathematics or distribution theory the use of
classical calculations in a part of them is more than needed for coherence
between classical mathematics and the calculations in the new setting. In
short in the context of G(Ω) one has complete coherence but only in form of
the minimum acceptable to have coherence, and the Schwartz assumption
that C(R) would be a subalgebra of A would give more than this minimum.
It is precisely this subtle fact that makes the difference between possibility
and impossibility.

The association also serves to interpret weak solutions of PDEs:

• Let us recall the definition of a weak asymptotic method as stated

by Danilov-Omel’yanov and Shelkovich [41] as an extension of Maslov as-
ymptotic methods: for simplicity consider the equation ut + f (u)x ” =
”0 where ” = ” means that the sense of this equality is not yet fixed.
Then a weak asymptotic method is a sequence (u )→0 such that ∀ψ ∈
D(R) ∀t ((u )t ψ − f (u )ψx )dx → 0 as  → 0. One recognizes the similar-

R
ity with the definition of the association in Gs (Ω) in variable x for fixed t:
∀ψ ∈ D(R) ∀t ((u )t ψ + (f (u ))x ψ)dx → 0 as  → 0. The difference lies
R
in that in the weak asymptotic method f (u ) needs not be C ∞ . Indeed,
provided the family u is made of C ∞ functions and has bounds in const N
on compact sets of Ω and for all partial derivatives, the two concepts are
same: starting from any weak asymptotic method these two requirements
(i.e. C ∞ and bounds) can be obtained by convolution and by reindexation
in  respectively. Therefore

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

216 J.F. Colombeau

There exists a weak asymptotic method iff there exists a solu-

tion in the sense of association.

The concept of weak asymptotic method as defined in [41] has been at

the origin of many important investigations on solutions of systems of con-
servation laws in recent years [1, 2, 36, 41, 37, 38, 39, 40, 42, 43, 44, 65, 73,
74, 84, 85, 86] . . . .

• In presence of Lp bounds, 1 ≤ p ≤ ∞,, anyway needed to produce a

measure valued solution, it has been proved that

The existence of a solution in the association sense is equivalent

to existence of a measure valued solution, [3].

The need of solutions in the association sense is clearly shown by the

following well known remark: the equation ut + uux ” = ”0 has not same
jump conditions as the equations obtained by multiplying it by u, for in-
stance uut + u2 ux ” = ”0. The interpretation is that the equation stated
in G in the strong form ut + uux = 0 (i.e. with the ”true” equality in G)
has no shock wave solution since multiplication by u would give same solu-
tion; the equation stated in G in its weak form ut + uux ≈ 0 (i.e. with the
association) has shock wave solutions but multiplication by u is forbidden,
which is put in evidence by the fact it changes the solutions.

Conclusion. In order to multiply distributions and √ to create

new needed mathematical objects: H n 6= H p if n 6= p, δ 6= 0, a
context of nonlinear generalized functions has been introduced
inside the Leopoldo Nachbin group (1980-1986): Jorge Aragona
and João Toledo (USP), Hebe Azevedo Biagioni (UNICAMP),
Jean François Colombeau (visitor at UNICAMP, USP, UFRJ),
Roberto Soraggi (UFRJ). One has defined and studied two dif-
ferential algebras (and variants)

Gs (Ω) ⊂ G(Ω)

based on a more refined concept of equality (asymptotic behavior

in o(q )∀q) instead of mere tendance to 0. Dropping the refinement
of this asymptotic behavior in favor of a tendance to 0 gives the
association (plus use of test functions in the standard definition
of association). The concept of association had already been de-
fined earlier in various circumstances by numerous authors such

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 217

as Mikusinski (products of distributions), Lojasiewicz (point val-

ues of distributions), Maslov, Danilov, Omel’yanov, Shelkovich
(asymptotic methods).

3. Direct subsequent applications after 1986. After 1986 the the-

ory evolved outside the Leopoldo Nachbin group; in particular let us quote:
• in France (J.F. Colombeau, A.Y. Le Roux and coauthors) under supervi-
sion of the technical branch of the Ministry of Defense (DRET): simulation
of strong collisions for armor, wave diffusion for furtivity.
• in Austria (M. Oberguggenberger and coauthors M. Kunzinger, C. Garetto,
G. Hormann,. . . ) and Serbia (S. Pilipovic and coauthors): solutions of lin-
ear and nonlinear Cauchy problems with = in G, propagation of singularities
for linear PDEs with irregular coefficients (numerous articles).
• in Austria (M. Grosser, M. Kunzinger, R. Steinbauer) and England (J.
Vickers): generalized functions on manifolds and General Relativity, see
the research-expository paper [69] in this volume.
• in Brazil (J. Aragona, S.O. Juriaans and coauthors): algebraic proper-
ties and use of the Scarpalezos sharp topology [7, 8, 9, 10, 11] and also
[51, 52, 53, 72, 81, 82].
• Stochastic equations (F. Russo, M. Oberguggenberger, D. Rajter, S.
Pilipovic, D. Selesi, P. Catuogno, C. Olivera,. . . ) [71, 78, 64, 20, 21, 76].
• Connections with nonstandard analysis have been developped by various
authors [56, 91, 72, 70], . . . , see [91] in this volume.

The theory of nonlinear generalized functions is now presented in many

books or research expository papers [4, 18, 22, 28, 29, 30, 31, 45, 54, 55,
67, 70, 77]. Now we sketch some applications that permit a better under-
standing of the theory and its use.

Symmetric linear hyperbolic systems (M. Oberguggenberger).

One considers linear symmetric hyperbolic systems

n
∂U X ∂U
= Ai (x1 , . . . , xn , t) (x1 , . . . , xn , t)+
∂t ∂xi
i=1

+B(x1 , . . . , xn , t)U (x1 , . . . , xn , t) + C(x1 , . . . , xn , t), (18)

where the Ai are real symmetric matrices with coefficients in G(Rn+1 ) which
further have a specific growth (in form of a bound log( 1 )) when  → 0, see
[60] for details.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

218 J.F. Colombeau

The authors obtain existence and uniqueness of solutions of the Cauchy

problem with = in the G context. Further the solution is associated to the
classical solution when the later exists.

This result has permitted to consider the case of discontinuous coeffi-

cients, very important in the applications since it models wave propagation
in an heterogeneous medium and it has been at the origin of numerous
works. The proof consists in applying to representatives a classical proof of
the C ∞ case that is refined more in detail so as to prove that the regularized
solutions have the bounds in 1N and to prove uniqueness from bounds in
o(q )∀q. This technique of working on representatives by refining a classical
C ∞ proof and using the definition of nonlinear generalized functions as a
quotient has been applied in many cases and it has permitted to extend
many C ∞ results to the case of nonlinear generalized functions.

A nonlinear very dissipative parabolic equation. Another exam-

ple of the above technique consisting in refining classical C ∞ proofs has
been used in [33, 61, 12] for the parabolic equation

ut − ∆u + u3 = 0, (19)
0
u(., 0) = f ∈ E (] − 1, 1[), (20)
u(x, t) = 0 on ∂Ω × [0, +∞[. (21)
When the initial condition is a distribution with compact support in ] −
1, +1[ one obtains existence and uniqueness of a solution with equality in the
algebra Gs ([−1, +1] × [0, +∞[) which is defined like the case of generalized
functions over an open set (representatives of class C ∞ in the interior, con-
tinuous as well as all derivatives up to the boundary, and uniform bounds
on compact subsets of [−1, +1] × [0, +∞[). This solution is associated to
the classical solution when the later exists.

An interesting point is as follows: it had been noticed [19] that if a Dirac

delta distribution is approximated as usual by a sequence (δ  ) then the
classical C ∞ solutions u corresponding to the initial conditions δ  tend to
0 uniformly on compact subsets of [−1, +1]×]0, +∞[ when  → 0. If as
usual the solution u is defined as a limit of the u then u = 0, and therefore
the initial condition defined as limit of u(., t) when t → 0 should be the
null function, while we started with the Dirac delta distribution as initial
condition. This paradox suggests to consider that problem (19-21) with
Dirac distribution as initial condition has no solution. In the G context the
explanation of the paradox is clear: the solution u ∈ Gs ([−1, +1] × [0, +∞[)
is associated to 0 in Ω×]0, +∞[, but it is not equal to 0, and its initial
condition which in the G context is defined as its restriction to t = 0 is by

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 219

construction the given distribution on ] − 1, +1[.

Weak solutions in the context of nonlinear generalized func-

tions. The solutions considered in the two above sections were strong so-
lutions in the sense that they were solutions with the equality in the algebra
of nonlinear generalized functions. Unfortunately many basic equations do
not have such solutions and one is forced to consider also solutions in the
association sense. This has already been noticed for the shock waves so-
lutions of the equation ut + uux ” = ”0 - where we recall that the symbol
” = ” means that the sense to be given to this equality is not yet fixed-
that we are forced to interpret as
ut + uux ≈ 0.
This is equivalent to a weak asymptotic method, see above, with weak
derivative both in x and t, i.e. a family (u ) of measurable functions in
the variables x, t such that, in the case of the equation ut + (f (u))x ≈ 0,
Z
∞ 2
∀ψ ∈ Cc (R ) (u ψt + f (u )ψx )dxdt → 0 when  → 0. (22)

A shock wave in the usual form of two constant values ul , ur separated

by a discontinuity moving with constant velocity c is represented by the
formula
u(x, t) = ul + [u]H(x − ct), (23)
where [u] := ur − ul is the jump of u and H is any Heaviside function in
G(R) or Gs (R). What is precisely a Heaviside generalized function?

A Heaviside generalized function H is defined as an element of G(R)

which has a representative (x, ) 7−→ H(x, ) uniformly bounded in  and
x such that H(x, ) tends to 0 if x < 0 and 1 if x > 0 in a sense that
can depend on authors, but that
R +∞ implies that ∀n H n ≈ H. All Heaviside
0
functions are associated and −∞ H (x, )dx = 1.

It will be very important for the sequel to remark that H should not
be the privilieged Heaviside function in D0 (R) from the (canonical or not)
injections of D0 (R) into the two algebras G(R) or Gs (R) of nonlinear gener-
alized functions. Indeed we shall deal with several physical variables such as
density ρ, velocity u,. . . , that in general cannot be represented by the same
Heaviside function (recall the example of the elastoplastic shock waves) so
that we will need different Heaviside functions and there is no reason to
priviliege one physical variable. We will consider equations in divergence
form also called conservative equations because they express conservation

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

220 J.F. Colombeau

of some quantity, and equations not in divergence form also called noncon-
servative equations.

Inserting (23) into the equation ut + (f (u))x ≈ 0 one obtains at once the
classical jump condition
f (ur ) − f (ul ) [f (u)]
c= = . (24)
ur − ul [u]
The detailed calculation is easy: inserting (23) into (24) one obtains −c[u]H 0 +
(f (ul + [u]H))0 ≈ 0. One can integrate an association as this is classically
the case for the equality in distribution theory; one obtains: −c[u]H +
f (ul + [u]H) + const ≈ 0. Using H(−∞) = 0 one obtains const = −f (ul );
then H(+∞) = 1 gives the jump condition (24).

One obtains easily: u in (23) is solution in the association sense iff the
jump condition (24) holds. A solution in the association sense does not
bring any information on the Heaviside function H: all Heaviside functions
fit as well: although there is uniqueness of jump condition, and so unique-
ness, modulo association, of solutions of form (23) for the Cauchy problem,
there is no uniqueness in the sense of equality in the G context since the
function H is not fixed.

Now we know that a nonlinear scalar equation has no shock wave in the
sense of equality in G (same as above for ut + uux = 0), so let us turn to
a system of two equations. We will distinguish systems in divergence (or
conservative) form and systems in nondivergence (or nonconservative) form
since the shock wave theory is quite different in each case.

Systems in divergence form are systems

ut + (f (u, v))x ” = ”0, (25)
vt + (g(u, v))x ” = ”0, (26)
i.e. they have the very particular feature that they are sums of derivatives.
Nonconservative systems are systems which are not in divergence form such
as the system
ut + (u2 )x ” = ”0, (27)
σt + u.σx ” = ”0, (28)
because of the term u.σx . The best known systems are in divergence form,
such as the famous system of ideal gases that we state here in one dimension:

ρt + (ρu)x ” = ”0, (29)

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 221

(ρu)t + (ρu2 )x + px ” = ”0, (30)

(ρe)t + (ρeu)x + (pu)x ” = ”0, (31)
u2
p” = ”(γ − 1)(ρe − ρ ), (32)
2
where ρ, u, p, e denote respectively the density, the velocity, the pressure
and the total energy per unit mass; γ is a constant =1.4 for the air in usual
conditions.

For systems in divergence form stated with the association

(ui )t + (fi (u1 , . . . , un ))x ≈ 0
we have noticed (23, 24) that shock waves ui (x, t) = (ui )l + [ui ]Hi (x −
ct), (ui )l , [ui ] ∈ R, Hi Heaviside generalized functions, satisfy the jump con-
[f ]
ditions c = [u i]
. For systems in nondivergence form the situation is com-
pletely different: consider for instance the equation ut + uvx − ux ≈ 0 and
set
u(x, t) = ul + [u]H(x − ct), v(x, t) = vl + [v]K(x − ct) (33)
where H, K are two Heaviside functions. Insertion of (33) into the equa-
tion gives −c[u]H 0 (ξ) + ul [v]K 0 (ξ) + [u][v]H(ξ)K 0 (ξ) − [u]H 0 (ξ) ≈ 0 as a
generalized function of the variable ξ = xR − ct. Next one would like to
integrate in ξ: this is impossible because H(ξ)K 0 (ξ)dξ is unknown and
can take a continuum of values according to H and K. Therefore we do
not obtain well defined jump conditions. This problem is important and
will be treated in detail in the sequel.

Examples of nonconservative systems. We give three classical ex-

amples of nonconservative systems in which the origins of the nonconserva-
tive equations have different natures: the state law for the elasticity system
of large deformation [27], the relativity equations [79] for the system of pri-
mordial cosmology [23], the statement of the basic laws of physics for the
multifluid system [90, 68]. For convenience the systems are given in 2-D
since the 3-D systems are obvious extensions of the 2-D systems.

Elasticity for large deformations. The following nonconservative equa-

tions (38-40) stem from a differential form of Hooke’s law. In elasticity
the nonconservative terms in (38) are v.(s11 )y and (uy − vx ).s12 . In elasto-
plasticity µ is null in the plastic region, [27], which introduces other ”non-
conservative terms” in the equations (38-40). This system is used for the
numerical simulations of strong collisions, in particular for the design of
armor [27, 30, 18].

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

222 J.F. Colombeau

The model in 2-D is (stating = in place of ”=”):

ρt + (ρu)x + (ρv)y = 0, (34)
(ρu)t + (ρu2 )x + (ρuv)y + (p − s11 )x − (s12 )y = 0, (35)
2
(ρv)t + (ρuv)x + (ρv )y + (p − s22 )y − (s12 )x = 0, (36)
(ρe)t + (ρeu)x + (ρev)y + ((p − s11 )u)x + ((p − s22 )v)y − (s12 v)x − (s12 u)y = 0,
(37)
4 2
(s11 )t + u.(s11 )x + v.(s11 )y − µux − µ.vy − (uy − vx ).s12 = 0, (38)
3 3
2 4
(s22 )t + u.(s22 )x + v.(s22 )y + µvx − µ.uy + (uy − vx ).s12 = 0, (39)
3 3
1
(s12 )t + u.(s12 )x + v.(s12 )y − µvx + µ.uy − (uy − vx ).(s11 − s22 ) = 0, (40)
2
u2 + v 2
p = Φ(ρ, e − ), (41)
2
where ρ = density; (u, v)=velocity vector, p =pressure, e=density of total
energy; (sij )=stress tensor (sij = sji ), µ= shear modulus which can be
considered as constant in elasticity, Φ = a state law.

Primordial cosmology. According to the standard big bang theory, in

its infancy the universe was ”small” and the photons were very energetic
(small wavelength). Therefore radiation dominated baryonic matter from
well known quantum effects. Later the expansion of the universe increased
the wavelength of the photons, and baryonic matter became free as it is to-
day. The equations in the radiation dominated universe are very important
since the seeds of galaxies were formed from black matter at this epoch,
which provided gravitational wells for baryonic matter when radiation dom-
ination ceased. The equations below are those in use by cosmologists [23].
They are issued from the equations of relativity, for which the lack of di-
vergence form is a well known problem [79]: the Euler equations (43, 44)
are not in divergence form. The equations are:

p p
ρt + ((ρ + 2 )u)x + ((ρ + 2 )v)y = 0, (42)
c c
p p
(ρ + 2 )(ut + uux + uvy ) + px + (ρ + 2 )Φx = 0, (43)
c c
p p
(ρ + 2 )(vt + vux + vvy ) + py + (ρ + 2 )Φy = 0, (44)
c c
3p
∆Φ = 4πG(ρ + 2 ) (45)
c

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 223

1
p = ρc2 , (46)
3
where ρ, (u, v), p, denote respectively the density of energy, the velocity vec-
tor, the radiation pressure; c is the velocity of light.

Multifluid flows. The system below is called the ”standard one pressure
model”. It models a mixture of two immiscible fluids which are strongly
mixed, for instance numerous bubbles of one fluid into the other, for which
there is only one pressure due to the mixture. The extremely complicated
geometry of the surface separating the two fluids is replaced by an averaging
from the respective volumic proportions α1 and α2 of the fluids, which
introduces nonconservative terms. The model is derived and discussed in
[90, 68]. It is used in offshore petroleum extraction to study the ”gas kick”,
i.e. a destructive possible explosion when a mixture of oil and natural
gas reaches the surface, see the extremely rich literature on the subject
in [68, 46, 47, 48, 49, 50, 92]. . . . This system is also used for the cooling
of nuclear power stations (mixture of boiling water and vapor of water).
Equations (49-52) are nonconservative. The 1-D model of a mixture in a
tube (index 1 for fluid 1 and index 2 for fluid 2) is [90, 68, 92]:
(ρ1 α1 )t + (ρ1 α1 u1 )x = 0, (47)
(ρ2 α2 )t + (ρ2 α2 u2 )x = 0, (48)
2
(ρ1 α1 u1 )t + (ρ1 α1 (u1 ) )x + α1 px + τ = ρ1 α1 g, (49)
2
(ρ2 α2 u2 )t + (ρ2 α2 (u2 ) )x + α2 px − τ = ρ2 α2 g, (50)
(ρ1 α1 e1 )t + (ρ1 α1 u1 e1 )x + (α1 pu1 )x + (pα1 u1 )x + p.(α1 )t = ρ1 α1 u1 g, (51)
(ρ2 α2 e2 )t + (ρ2 α2 u2 e2 )x + (α2 pu2 )x + (pα2 u2 )x + p.(α2 )t = ρ2 α2 u2 g, (52)
where, if i = 1, 2, αi is the volumic proportion of phase i in the mixture,
therefore,
α1 + α2 = 1, (53)
ρi , ui , ei are the density, velocity and total energy of phase i per unit mass, p
is the pressure in the mixture, τ denotes the interface-momentum exchange
term and g is the component of gravitational attraction in the direction of
the tube (which is not always vertical). The system is completed by the
two state laws of the liquid and the gas:
ui )2
p = Fi (ρi , ei − ), i = 1, 2. (54)
2
Jump conditions for nonconservative systems: the use of equal-
ity and association in the calculations of jump conditions for non-
conservative systems. Consider a typical nonconservative system of two

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

224 J.F. Colombeau

equations [30].
u2
ut + (
)x ” = ”σx , (55)
2
σt + uσx ” = ”ux , (56)
which is nonconservative because of the term uσx in (56), where ” = ”
means that the meaning of this equality is not really clear (it can be = in
G or ≈ or a similar concept). As usual we search solutions u, σ in form of
two constant values (ul , σl ) on the left and (ur , σr ) on the right separated
by a discontinuity that moves at constant speed c, i.e. of the form
u(x, t) = ul + [u]Hu (x − ct), (57)
σ(x, t) = σl + [σ]Hσ (x − ct), (58)
where [u] = ur − ul , [σ] = σr − σl are respectively the jumps of u, σ on the
discontinuity located at x = ct. Note the basic fact that we seek a solution
in (G(R))2 because the term uσx does not make sense in distribution the-
ory, and that one knows that one should a priori represent different physical
variables by different Heaviside functions, Hu , Hσ .

• No shock wave solutions with equality in G. Using the equality

in G(R) for both equations (55, 56) we are going to discover they have no
shock wave solution of the form (57, 58).

Inserting (57, 58) into (55) stated with = in G(R) one obtains
(−c + ul )[u]Hu0 + [u]2 Hu Hu0 = [σ]Hσ0 . (59)
Integration gives
1 [u]2
Hσ = ((−c + ul )[u]Hu + (Hu )2 ) + constant (60)
[σ] 2
in G(R). Since Hu (−∞) = Hσ (−∞) = 0 it follows that constant = 0. Now
Hu (+∞) = Hσ (+∞) = 1 gives a formula which is exactly the classical jump
formula of the conservative equation (55) (recall that the jump condition
of the equation ut +(f (u, v))x = 0 is c = [f ]
[u] where [f ] = f (ur , vr )−f (ul , vl )).

Inserting (57, 58) into (56) stated with = in G(R) gives the relation
(−c + ul )[σ]Hσ0 + [u][σ]Hu Hσ0 = [u]Hu0 (61)
between Hu and Hσ , very different from (59) since it cannot be integrated
like (59) because of the term Hu Hσ0 . One observes that relations (59) and
(61), which follow from equations (55), (56) respectively, stated with = in
G, are incompatible. Indeed these two relations give Hσ as two different

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 225

functions of Hu . These functions are respectively (top and bottom of p. 63

in [30]):
[u]2 [u]2 1 2Hu
Hσ = (1 − )Hu + (Hu )2 , Hσ = Log|1 + [σ]
|. (62)
2[σ] 2[σ] σ −1 + 2 [u] 2

Conclusion: system (55, 56) has no solution of the form (57, 58)
when both equations in system (55, 56) are stated with equality
in G(R). This is of course completely unsatisfactory since solu-
tions of the form (57, 58) do exist physically and are even of the
greatest importance in physics and engineering for nonconserva-
tive systems such as those given above.

• Undefinite jump conditions with association. Consider system

(55, 56) stated in the form of association for both equations and insert (57,
58). From equation (55) one obtains (59) stated in the association sense:
it can be integrated and it gives the classical jump formula for (55). From
equation (56) one obtains (61) with association in the place of equality, i.e.
(−c + ul )[σ](Hσ )0 + [u][σ]Hu (Hσ )0 ≈ [u](Hu )0 . (63)
Since this equation is stated in the sense of association it cannot be manip-
ulated by nonlinear calculations (17): nonlinear calculations would request
the equality in G(R). Further,
R as it is, relation (63) cannot be integrated
because of the term Hu Hσ0 : Hu Hσ0 can take a continuum of values depend-
ing on the relation between Hu and Hσ : for instance if Hu = (Hσ )n one
obtains Hu Hσ0 = (Hσ )n Hσ0 = n+11
(Hσn+1 )0 ≈ n+1
1
Hσ0 ≈ n+1
1
δ since H n ≈ H
if H is an Heaviside function (equal to 0 to 0 for x < 0 and to 1 for x > 0),
and since from (16) one can differentiate the association.

An algebraic relation between Hu and Hσ , such as for instance

one of the relations (62), would be needed to resolve the ambigu-
ity in the product Hu Hσ0 and to obtain a jump condition from (63).

We have observed that any such algebraic relation can be obtained by

stating (55) or (56) in the sense of equality in G(R), see (62).

Equation (63) shows that Hu (Hσ )0 is associated to a multiple of the Dirac

delta function (δ ≈ (Hu )0 ≈ (Hσ )0 ). Setting
Hu (Hσ )0 ≈ Aδ, A ∈ R, (64)
(63) can be integrated and gives
(−c + ul )[σ] + [u][σ]A = [u]. (65)

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

226 J.F. Colombeau

Now this jump condition depends on the arbitrary parameter A which

ca take a continuum of values and contains the ambiguity in the product
Hu (Hσ )0 .

From natural numerical schemes for equations (55, 56) formulas (64, 65)
have been checked with precision: the value A can be computed from (64)
when one has the set of values taken by Hu and Hσ in their jumps, which
makes sense since the jumps take place on several cells, see [34] and [30]
p. 44-48. One has observed that formula (65) was verified: [34] and [30] p.
44-48.

• Definite jump conditions with a mixed strong-weak statement

of the equations. One has observed on standard systems of physics [30]
that when one has a family of n independent physical variables satisfying a
system of n equations with no independent equation, the statement of n−1
equations with equality in G(R) gives at least implicitely all the Heaviside
functions representing the jump of n − 1 variables (by formulas such as
one of the formulas (62)) as a function of the Heaviside function of the nth
variable, thus resolving the ambiguities in nonconservative terms. The last
Heaviside function remains undetermined among all Heaviside functions in
G(R): this does not matter to obtain the jump conditions: the knowledge of
one Heaviside function Hu or Hσ in (64) as a function of the other permits
the calculation of the number A.

The jump conditions for the two formulations

ut + uux = σx , σt + uσx ≈ k 2 ux , (66)
and
ut + uux ≈ σx , σt + uσx = k 2 ux , (67)
k > 0, have been computed in [30] p. 62-64. Jump conditions for systems
of n equations, when n − 1 are stated with = and the last one is stated
with ≈, have been computed in [30] p. 70-77.

Numerical investigations to resolve systems stated with = for some equa-

tions and ≈ for the other equations are reported in [30] p. 64-68, [94] using
a large space step for the equations stated with ≈ and a small space step
for the equations stated with =. One has observed (figure 3) the expected
result that the jump conditions are those calculated by stating with = in G
the equation discretized with small steps and by stating with ≈ the equa-
tion discretized with large steps. Other numerical investigations consist in
the use of a large viscosity for the equations stated with ≈ and a small
viscosity for the equations stated with =, (figure 4) [75]. One obtains again

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 227

the expected result similar to the above for the double scale method.

•Choice of the statement of the equations on physical ground.

To apply the above method for a system of equations modelling a given
physical situation one has to decide on physical ground which equations
should be stated with = and with ≈. This can be delicate since most
equations involve approximations. In general one distinguishes between
the basic laws (conservation of mass, momentum, total energy and electric
charge) and the state laws (also called constitutive equations) which are
extracted from particular experiments and are usually far less accurate.

In case of a system of n equations with only one state law, it is natural

from physics to state the state law with ≈, [30] p. 69-77.

In the system (42-46) of primordial cosmology there are two state laws:
an extension of Newton’s law of gravitation (71), and the state law (72) for
the energy. It is natural to choose to state (72) with association since it is
an approximation.
p p
ρt + ((ρ + 2 )u)x + ((ρ + 2 )v)y = 0, (68)
c c
p p
(ρ + 2 )(ut + uux + uvy ) + px + (ρ + 2 )Φx = 0, (69)
c c
p p
(ρ + 2 )(vt + vux + vvy ) + py + (ρ + 2 )Φy = 0, (70)
c c
3p
∆Φ = 4πG(ρ + 2 ) (71)
c
1
p ≈ ρc2 . (72)
3
This statement would justify the fact cosmologists have already calculated
on equations (68-70) to obtain them from the equations of relativity.

For the multifluid system (47-54) in the case of a mixture of a liquid

and a gas one naturally states the state law of the liquid with = and the
state law of the gas with ≈ since the state law of the liquid (which is nearly
uncompressible) is far less affected than the state law of the gas inside the
shock waves, [30, 16, 6].

For the elasticity system (34-41) the situation is more difficult due to the
presence of the 3 scalar state laws (39-41), in fact a 2-D formulation of the
Hooke’law which is a tensorial law; an idea is, if the solid is isotropic, to
state that all components of the stress are modeled by the same Heaviside

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

228 J.F. Colombeau

function, thus giving three relations between Heaviside functions. Then

the statement of the state law (41) with association permits to propose
well defined jump conditions.

The paths in [35, 62] are exactly the microscopic profile of Heaviside
functions in [34, 18, 30, 27], i.e. the passage between the value 0 for x < 0
and the value 1 for x > 0. In this way quantities such as A in (64) are
expressed via the paths. The equivalence of the two formulations in the
setting of bounded variation functions is proved in [32]. The path formu-
lation as it is done in [35, 62] does not include the concepts of = and ≈
which has been used above to solve the ambiguity in values such as A in
(65) due to the products of a Heaviside function and a Dirac distribution,
and permits a choice of paths in a given physical situation.

• The continuity equations and the Euler equations in presence

of shock waves. The classical constitutive equation (29) and the clas-
sical Euler equation (30), when both stated with equality in G(R), imply
a remarkable connection between the Heaviside functions H 1 , Hu , Hp that
ρ
depict ρ1 , u, p respectively, according to
ω(x, t) = ωl + [ω]Hω (x − ct), (73)
ω = ρ1 , u, p.

Theorem. Equations (29, 30) stated with = in G(R) imply equality in

G(R) of the three a priori different Heaviside functions H 1 , Hu and Hp .
ρ

The proof can be found in [30] p. 69-72. Now consider the system of
isothermal fluids stated as
ρt + (ρu)x = 0, (74)
(ρu)t + (ρu2 )x + px = 0, (75)
p ≈ C.ρ, C a constant. (76)
From above one has H 1 = Hu = Hp . We deduce that there are no shock
ρ
waves that satisfy the state law (76) with equality in G(R): indeed p = Cρ
would imply Hp = Hρ , therefore H 1 = Hρ . But this last equality is
ρ
impossible: denoting this joint Heaviside function by H it would imply
(ρl + [ρ]H)( ρ1l + [ ρ1 ]H) = 1; using [ρ] = ρr − ρl and [ ρ1 ] = ρ1r − ρ1l one obtains
at once H 2 = H in G(R), which is proved to be impossible at the beginning
of the paper.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 229

• The form of the state laws inside the shock waves. This implies
that the state law p = Cρ of isothermal gases cannot be satisfied inside the
width of the shock waves, although the weak statement (76) implies that
it is satisfied on both sides from the formulas pl = Cρl , pr = Cρr . Is it
possible to modify the statement of (76) so as to state the modified state
law with equality in G(R) inside a shock wave?. Then

p 1 1
C= = (pl + [p]H)( + [ ]H) (77)
ρ ρl ρ

is no longer a constant as a generalized function ∈ G(R): it varies inside

1
u(x,t)−ul p(x,t)−pl ρ
(x,t)− ρ1
the shock wave. Since H(x) = ur −ul = pr −pl = 1
−1
l
the function
ρr ρl
C = C(x) imposed by equality in the state law can be expressed in func-
tion of the physical variables u, p, ρ and their left-right values on the shock
wave. Then one has at the same time equations (74), (75) and (76) stated
with the (strong) equality in G and presence of shock wave solutions. This
strong statement is suitable to have uniqueness since strong solutions are
unique in general.

• A physical interpretation of = and ≈ in case of shock waves.

The physical intuitive idea behind the formulation of shock waves with =
and ≈ is that shock waves would have an infinitesimal width, which is
the width of the Heaviside functions used to model the various physical
variables (this width has been observed experimentally). This infinitesimal
width could be thought to be of the order of magnitude of a few hundred
mean free paths in fluids (a mean free path is the average distance be-
tween two collisions) or cristal sizes in solids. This size of the widths of
shock waves is large enough so as to imagine small bags inside these widths,
and, by thought, one could state the conservation laws for the volume in
these bags (between times t and t + dt), which reproduces the classical
way physicists obtain the equations of continuum mechanics. This justifies
that these equations should be stated with the (strong) equality as done
above. On the other hand the state laws stated with ≈ are only valid on
both sides of the shock waves, in the same way as H and H 2 are equal on
both sides of the discontinuity, and are different inside the discontinuity.
By definition of a state law we know that such a law has been checked
(up to now) by experimental physicists in the case of (quasi) absence of
variation of the physical variables (absence of experimental apparatus to
observe inside widths of shock waves), that is, on both sides of the shock
waves but not inside the width of the shock waves. An interesting fact is
that the genuine state law that holds inside the width of the shock waves
can be calculated as a consequence of the equations when all are stated with

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

230 J.F. Colombeau

the strong equality (77), knowing the state laws outside of the shock waves.

• Numerical investigations. Numerical tests on the statement of

equations with = and ≈ are reported below (figures 3 and 4). Various nu-
merical methods have been devised to treat systems in which some equa-
tions are stated with = and the other ones with ≈. These methods consist
in a discretization of the equations stated with ≈ which should be much
coarser than the discretization of the equations stated with =. In [94] one
uses two discretization steps: a large step for the equations stated with
≈ and a small one for the equations stated with =. One obtains the ex-
pected results: one can observe that the numerical solutions so obtained
coincide with the explicit jump conditions corresponding to the various
statements (figure 3). In [75] one uses viscous second members with viscos-
ity coefficients larger for the equations stated with ≈. Again one obtains
the expected results (figure 4).

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

Nonlinear Generalized Functions: their origin, some developments and recent advances 231

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

232 J.F. Colombeau

In figures 3 and 4 we present the exact solutions for the system (55, 56)
stated successively with (≈, =), (≈, ≈) with same Heaviside functions Hu

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 233

and Hσ , and (=, ≈), and numerical results from various techniques of dis-
cretization of the equations.

In figure 3 one uses a two scale method for the discretization of deriva-
tives: for equation ut + uux ” = ”σx one sets f 0 = f (x,t)−fh1(x−h1 ,t) , f = u, σ
and for equation σt + uσx ” = ”ux one sets f 0 = f (x,t)−fh2(x−h2 ,t) , f = u, σ.
One tests various values of k := hh12 . For k = 16 the first equation is dis-
cretized in a far coarser way than the second and we observe the numerical
result coincides with the statement of the first equation with ≈ and the sec-
ond with = , as it could be expected. For k = 1 one observes the numerical
1
result corresponds to the same Heaviside functions for u and σ. For k = 16
we observe the numerical result corresponds to the statement of the first
equation with = and the second with ≈ as expected.

In figure 4 the system is stated with a small viscosity i.e.

ut + uux = σx + 1 uxx ,
σt + uσx = ux + 2 σxx ,

and one tests the quotient k := 21 . As expected, for k = 60

1
the fact that
2 << 1 implies that one observes the solution (≈, =). For k = 1 one
observes same Heaviside functions and for k = 60 one observes the solution
(=, ≈). This shows that the various statements of equations with = and ≈
are a fact of the real world (here numerical calculations) and that it can be
mastered numerically.
In [57] one proves that the above viscous Cauchy problem admits solutions
in the association sense when k = 1 for rather arbitrary initial data in
G(R). This is done by solving the equations for given  > 0. The result is
unknown when k 6= 1.

4. A topological continuation of the nonlinear generalized func-

tions: São Paulo 2012. In the theory of Banach spaces compactness
imposes the use of weak topologies but these topologies are uncompati-
ble with nonlinearity in that un → u, vn → v do not imply in general
un vn → uv. Correspondingly the locally convex spaces of distributions are
not topological algebras. The aim of this section is to present a context
containing irregular objects such as all distributions with compact support
in which one has compactness and nonlinearity at the same time. We con-
struct subalgebras of G(Ω) that enjoy Hausdorff locally convex topologies
suitable for the development of a functional analysis: they have very good

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

234 J.F. Colombeau

topological properties in particular concerning compactness (strong duals

of Fréchet-Schwartz spaces) at the same time as they are compatible both
with partial derivatives and nonlinearity.

In [5] we construct a family of differential algebras generically denoted

Ga (Ω) in the situation Ga (Ω) ⊂ Gs (Ω) such that

• The multiplication Ga (Ω) × Ga (Ω) 7−→ Ga (Ω), (F, G) 7−→ F G, and the
|α| F
differentiation Ga (Ω) 7−→ Ga (Ω), F 7−→ ∂xα1∂,...,∂x αn , are continuous,
1 n

• Ga (Ω) is a Hausdorff locally convex topological algebra which is a strong

dual of a Frechet-Schwartz space (in particular for any bounded set b there
is a larger bounded set B which is convex and balanced such that the vector
space ∪n nB normed with the gauge of B is a Banach space and such that
the set b is relatively compact in this Banach space),

• spaces of distributions such as E 0 (Ω) are contained in Ga (Ω) with con-

tinuous inclusions so that any bounded set in these spaces of distributions
is bounded in Ga (Ω), therefore relatively compact as explained above.

The construction in [5] is based on the properties of holomorphic func-

tions: uniqueness of analytic continuation, classification of point singular-
ities, Laurent expansion around a pole. This permits to have only one
representative in each class of the quotient reservoir
kernel defining Gs (Ω). The
absence of a quotient so obtained permits to have nice locally convex topolo-
gies.

This could be an interesting continuation of the theory of nonlinear gen-

eralized functions started inside the Leopoldo Nachbin group 30 years ago,
in which a topological structure with such properties remained lacking up
to now. A topology introduced early by H.A. Biagioni [18] (but somewhat
hidden in a minor point p. 44-45) and rediscovered independently by D.
Scarpalezos [81, 82] is not a locally convex vector space topology and has
no known compactness property. Anyway it has interested applications:
[7, 8, 9, 10, 11, 12, 18, 51, 53, 72, 81, 82] .

References
[1] S. Albeverio, O.S. Rozanova, V.M. Shelkovich. Transport and concentration
processes in the multidimensional zero-pressure gas dynamics model with the
energy conservation law. ArXiv: 1101.581v1, 30 Jan 2011.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 235

[2] S. Albeverio, V.M. Shelkovich. On delta shock front problem. In ”‘Analytical

approach to multibalance laws”’, chap. 2. Editor O. Rozanova, Nova Science
Pub. Inc. 2005, p. 45-88.
[3] J. Aleksik, J.F. Colombeau, M. Oberguggenberger, S. Pilipovic. Approximate
generalized solutions and measure valued solutions to conservation laws. Inte-
gral Transform and Special Functions. 20, 3-4, 2009, p. 163-170.
[4] J. Aragona, H.A. Biagioni. An Intrinsic Definition of the Colombeau Algebra
of Generalized Functions. Analysis Mathematica 17,1991, p. 75-132.
[5] J. Aragona, J.F. Colombeau, S.O. Juriaans. Locally convex topological algebras
of generalized functions: compactness and nuclearity in a nonlinear context.
Transactions AMS, in press.
[6] J. Aragona, J.F. Colombeau, S.O. Juriaans. Nonlinear generalized functions
and jump conditions for a standard one pressure liquid-gas model. preprint.
[7] J. Aragona, R. Fernandez, S.O. Juriaans. Natural topologies on Colombeau
algebras. Topological methods in nonlinear analysis, 34, 2009, p. 161-180.
[8] J. Aragona, R. Fernandez, S.O. Juriaans. A discontinuous Colombeau differen-
tial calculus. Monatsh. Math. 144, 2005, p. 13-29.
[9] J. Aragona, R. Fernandez, S.O. Juriaans. Integrating generalized functions over
membranes. preprint, Universidade de São Paulo, 2006.
[10] J. Aragona, R. Fernandez, S.O. Juriaans. M. Oberguggenberger. Differential
calculus and integration of generalized functions over membranes. Monatshefte
Math. 166, 2012, p. 1-18.
[11] J. Aragona, R. Fernandez, S.O. Juriaans. The sharp topology on the full
Colombeau algebra of generalized functions. Integral Transform and Special
Functions 17, 2006, p. 165-170.
[12] J.Aragona, A.R.G. Garcia, S.O. Juriaans. Generalized solutions of a nonlinear
parabolic equation with generalized functions as initial data. Nonlinear analysis
71, 2009, p. 5187-5207.
[13] J.Aragona, F. Villarreal. Colombeau’s theory and shock waves in a problem of
hydrodynamics. Journal d’Analyse Mathematique, 61, 1993, p. 113-144.
[14] J. A. Barroso (editor) Functional Analysis, Holomorphy and Approximation
Theory. (Rio de Janeiro 1980) North Holland Math. Studies 71, North-Holland,
1982.
[15] J. A. Barroso (editor) Aspects of Mathematics and its applications. North.
Holland Math; Library 34, North. Holland, 1986.
[16] F. Berger, J.F.Colombeau. Numerical solutions of one-pressure models in mul-
tifluid flows. SIAM J. Numer.Anal. 32, 4, 1995, p. 1139-1154.
[17] F. Berger, J.F.Colombeau, M. Moussaoui. Solutions mesures de dirac de
systmes de lois de conservation et applications numriqueq. Comptes rendus
acad. Sci. Paris serie 1, 316, 1993, p. 989-994.
[18] H.A. Biagioni. A Nonlinear Theory of Generalized Functions. Lecture Notes in
Math. 1421. Springer, 1990.
[19] H. Brezis, A. Friedman. Nonlinear parabolic equations involving measures as
initial conditions. J. math. Pures Appl. 62, 1983, p. 73-97.
[20] P. Catuogno, Ch. Olivera. On stochastic generalized functions. Inf. Dim. Anal;
quant. prob; Relat. topics. 14, 2011, 2, p. 237-260.
[21] P. Catuogno, Ch. Olivera. Tempered Generalized Functions and Hermite ex-
pansions. Nonlinear Analysis, T. M. A.. 2012, 74, p. 479-493.
[22] V.V. Christyakov. The Colombeau generalized nonlinear analysis and the
Schwartz linear distibution theory. J. Math. Sci. 93, 1999, p. 3-40.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

236 J.F. Colombeau

[23] P.Coles, F. Lucchin. Cosmology. the origin and evolution of cosmic structure?
Wiley, Chichester, England, 2002.
[24] J.F. Colombeau. A multiplication of distributions. J. Math.Ana. Appl. 94, 1,
1983, p. 96-115.
[25] J.F. Colombeau. Une multiplication generale des distributions. Comptes rendus
Acad. Sci. paris Serie 1, Math., 296, 1983, p. 357-360.
[26] J.F. Colombeau. Differential Calculus and Holomorphy. Noth-Holland-Elsevier,
1982.
[27] J.F. Colombeau. The elastoplastic shock problem as an example of the reso-
lution of ambiguities in the multiplication of distributions. J. Math. Phys. 30,
1989, p. 2273-2280.
[28] J.F. Colombeau. New Generalized Functions and Multiplication of Distribu-
tions. North-Holland-Elsevier, 1984.
[29] J.F. Colombeau. Elementary Introduction to New Generalized Functions.
North-Holland-Elsevier, 1985.
[30] J.F. Colombeau. Multiplication of distributions. Lecture Notes in Math. 1532,
SpringerVerlag, Berlin-Heidelberg-New York, 1992.
[31] J.F. Colombeau. Multiplication of distributions. Bull. Amer. Math. Soc., 23, 2,
1990, p. 251-268.
[32] J.F. Colombeau, A. Heibig. Nonconservative products in bounded variation
functions. SIAM J. Math. Anal. 23, 1992, p. 941-949.
[33] J.F. Colombeau, M. Langlais. Generalized silutions of nonlinear parabolic equa-
tions with distributions as initial conditions. J. Math. anal; appl. 145, 1, 1990,
p. 186-196.
[34] J.F. Colombeau, A.Y. Le Roux, A. Noussair, B. Perrot. Microscopic profiles of
shock waves and ambiguities in multiplication of distributions. SIAM J. Numer.
anal. 26,4, 1989, p. 871-883.
[35] G. Dal Maso, Ph. Le Floch, F. Murat. Definition and weak stability of a non-
conservative product. J. Math. Pures Appl. 9, 74, 1995, p. 253-273.
[36] V.G. Danilov. On the relation between the Maslov-Witham method and the
weak asymptotic methods. in Linear and Nonlinear Theory of Generalized Func-
tions and its Applications. p. 55-65. Banach Center Publications 88. Polish
Acad. of Sci. Warsaw, Poland.
[37] V.G. Danilov, D. Mitrovic. Delta shock wave formation in the case of triangular
hyperbolic system of conservation laws. J. Differential Equations 245, 2008, p.
3704-3734.
[38] V.G. Danilov, D. Mitrovic. Shock wave formation process for a multidimensional
scalar conservation law. Quarterly of Applied Mathematics vol 69, p. 613-634,
2011.
[39] V. G. Danilov, G.A. Omel’yanov. Weak asymptotic method and the interaction
of infinitely narrow δ-solitons. Nonlinear Analysis 54, 2003, p.773-799.
[40] V. G. Danilov, G.A. Omel’yanov. Weak asymptotic method for the study of
infinitely narrow δ-solitons. Electronic J. Diff. Equations 90, 2003, 27 p.
[41] V.G. Danilov, G.A. Omel’yanov, and V.M. Shelkovich. Weak Asymptotic
Method and Interaction of Nonlinear Waves, AMS Translations vol 208, 2003,
p. 33-164.
[42] V.G. Danilov, V.M. Shelkovich. On infinitely narrow solitons. In Nonlinear
Algebraic Analysis and Applications, A. Delcrois, M. Hasler, J.A. Marti,
V.Valmorin editors. Cambridge Scientific Pub. 2004, p. 57-72.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 237

[43] V.G. Danilov, V.M. Shelkovich. Dynamics of propagation and interaction of δ

shock waves in conservation law systems. J. Differential Equations 211, 2005,
p. 333-381.
[44] V.G. Danilov, V.M. Shelkovich. Delta-shock wave type solution of hyperbolic
systems of conservation laws. Quart. Appl. Math. 63, 2005, p. 401-427.
[45] Y. Egorov A theory of generalized functions. Russian Math. Surveys 45, 5, 1990,
p. 1-49.
[46] S. Evje. Weak solutions for a gas-liquid model relevant for describing gas-kick
in oil wells. SIAM J. Math.Anal. 43, 4, 2011, p. 1887-1922.
[47] S. Evje, K.K. Fjelde. Hybrid Flux-splitting Schemes for a two phase flow model.
J. Comput. Physics 175, 2002, p. 674-701.
[48] S. Evje, T. Flatten. Hybrid Flux-splitting Schemes for a common two fluid
model. J. Comput. Physics 192, 2003, p. 175-210.
[49] S.Evje, H.A. Friis. Global weak solutions for a gas-liquid model with external
forces and general pressure law. SIAM J. Appl. Math. 71, 2, 2011, p. 409-442.
[50] S. Evje, K.H. Karlsen. Global weak solutions for a viscous liquid-gas model
with singular pressure law. Comm. in Pure and Applied Analysis. 8, 6, 2009, p.
1867-1894.
[51] C. Garetto. Topological structure in Colombeau algebras: topological C mod-
ules and duality theory. Acta Appli. Math. 88, 2005, p. 81-123.
[52] C. Garetto. Topological structure in Colombeau algebras. Investigation of the
duals of Gc (Ω), G(Ω) and Gs (Rn ). Monatsh. math. 146, 3, 2005, p. 203-226.
[53] C. Garetto, H. Vernaeve. Hilbert C̃ modules: structural properties and ap-
plications to variational problems. Trans. Amer. Math. Soc. 363, 4, 2011, p.
2047-2090.
[54] M. Grosser, E. Farkas, M. Kunzinger, R. Steinbauer. On the Foundations of
Nonlinear Generalized Functions I and II. Memoirs of the AMS 153, 769, 2001.
[55] M. Grosser, M. Kunzinger, M. Oberguggenberger, R. Steinbauer. Geomet-
ric Theory of Generalized Functions with Applications to General Relativity.
Kluwer 2001.
[56] R.F. Hoskins, J. Sousa Pinto. Distributions, Ultradistributions and other Gen-
eralized Functions. Ellis-Horwood, New York-London, 1994.
[57] K; T. Joseph. Generalized solutions to a Cauchy problem for a nonconservative
hyperbolic system. J. Math. Anal. Appl. 207, 2, 1997, p. 361-387.
[58] B.L. Keyfitz, H.C. Kranzer. Spaces of weighted measures for conservation laws
with singular shock solutions. J. Diff. equations 118, 1995, p. 420-451.
[59] B.L. Keyfitz, H.C. Kranzer. A strictly hyperbolic system of conservation laws
admitting singular shocks. In Nonlinear Evolution Equations that change type,
ed? B.L. Keyfitz, M. Shearer, Springer, 1990, p. 107-125.
[60] F. Lafon, M. Oberguggenberger. Generalized solutions to symmetric hyperbolic
systems with discontinuous coefficients: the multidimensional case. J. Math.
Anal. Appl. 160, 1, 1991, p. 93-106.
[61] M. Langlais. Generalized functions solutions of monotone and semilinear para-
bolic equations with distributions as initial conditions. monatshefte Math; 110,
2, 1990, p. 117-136.
[62] Ph. Le Floch. Shock waves for nonlinear hyperbolic systems in nonconservative
form. Institute for mathematics and its Applications. University of Minnesotta,
preprint series 593, 1989.
[63] S. Machado(editor). Functional Analysis, Holomorphy and Approximation The-
ory. (Rio de Janeiro 1978) Lecture notes in Pure and Applied Math. 843,
Springer, 1979.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

238 J.F. Colombeau

[64] R. Mirkov, S. Pilipovic, D. Selesi. Generalized Stochastic Processes in algebras

of generalized functions J. math. anal. appl. 353, 1, 2009, p. 260-270.
[65] D. Mitrovic, V. Bojkovic, V.G. Danilov. Linearization of the Riemann problem
for a triangular system of conservation laws and delta shock wave formation
process. Mathematical Methods in the Applied Sciences 2010, 33, p. 904-921.
[66] J. Mujica. Complex analysis, Functional Analysis and Approximation Theory.
(Campinas 1984) North-Holland Math Studies 125, 1986.
[67] M. Nedeljkov, S. Pilipovic, D. Scarpalezos. The Linear Theory of Colombeau
Generalized Functions. Pitman Research Notes in Math. 1998.
[68] Svend Tollak Monkejord, Steinar Evje, Tore Flatten. A Musta scheme for a
nonconservative two-fluid model. SIAM J. Sci. Comput. 31, 4, 2009, p. 2587-
2622.
[69] E. Nigsch, C. Samann. Global algebras of nonlinear generalized functions with
applications in general relativity. In this volume and Arxiv. org. 2013.
[70] M. Oberguggenberger. Multiplication of Distributions and Partial Differential
Equations. Pitman Research Notes in Math. 259, Longman, Harlow, 1992.
[71] M. Oberguggenberger, F. Russo. Nonlinear SPDE, colombeau solutions and
pathwise limits. Stochastic analysis and related topics VI. editors L. Decreuse-
fond, J Geerde, B. Oksendal and A.S Ustunel. Progr.Proba. 42, Birkhauser,
Boston, Ma 1998. p. 319-332.
[72] M. Oberguggenberger, H. Vernaeve. Internal sets and internal functions in
Colombeau theory. J. Math. Anal. Appl. 341, 2008, p. 649-659.
[73] G.A. Omel’yanov, I. Segundo-Caballero. Asymptotic and numerical description
of the kink/antikink interaction. Electronic J. of Differential Equations, 2010,
150, p. 1-19.
[74] E.Yu. Panov, V.M. Shelkovich. δ’-shock waves as a new type of solutions to
systems of conservation laws. J. Differential Equations 228, 2006, p. 49-86.
[75] A. Prignet. Rapport de travail de fin d’ études. ENS Lyon, 1993.
[76] D. Rajter-Ciric, D. Selesi. Nonlinear stochastic diff. eq. containing generalized
delta processes. Monatshefte Math.168, 2012,1, p. 75-112.
[77] E.E. Rosinger. Part III of ”General solutions of nonlinear PDEs”. North-
Holland Elsevier, 1987.
[78] F. Russo. colombeau generalized functions and stochastic analysis. In stochastic
Analysis and Appl. in Physics. Editors A. Cardoso, M. de Faria, J. potthoff, R.
seneor, L. Streit. Kluwer 1994. p. 329-349.
[79] R.K. Sachs, H.H. Wu. General Relativity for Mathematicians. Graduate Texts
in math. 48, Springer, 1977.
[80] R. Sanders, M. Sever. The numerical study of singular shocks regularized by
small viscosity. J. Sci. comput. 19, 2003, p. 385-404.
[81] D. Scarpalezos. Generalized functions, topological structures, microlocal prop-
erties. A simplified point of view I. Bull. Cl. Sci. Math. Nat. Sci. Math. 25,
2000, p. 89-114. Same title II Publ. Institut Math. Beograd 76, 2004, p. 11-125.
[82] D. Scarpalezos. Some remarks on functoriality of Colombeau’s construction;
topological and microlocal aspects and applications. Integral Transforms and
Special Functions 6, 1998, 1-4, p. 295-307.
[83] M. Sever. Distribution solutions of nonlinear systems of conservation laws.
Memoirs of the AMS 889, 2007, p. 1-163.
[84] V.M. Shelkovich. δ− and δ 0 −shock wave types of singular solutions of systems
of conservation laws and transport and concentration processes. Russian Math.
surveys 63,3, 2008, p. 405-601.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239

 Nonlinear Generalized Functions: their origin, some developments and recent advances 239

[85] V.M. Shelkovich. The Riemann problem admitting δ−, δ’-shocks and vacuum
states; the vanishing viscosity approach. J. Diff. Eq. 231, 2006, p. 459-500.
[86] V.M. Shelkovich. Transport of mass, momentum and energy in zero-pressure
gas dynamics. In Proceedings of Symposia in Applied Mathematics 2009; vol.67.
Hyperbolic Problems: Theory, Numerics and Applications. Edited by E. Tad-
mor, Jian-Guo Liu, A.E. Tzavaras. AMS, 2009, p. 929-938.
[87] R. Steinbauer. The ultrarelativistic Reissner-Nordstrom field in the Colombeau
algebra. J. Math. Phys. 38, 1997, p. 1614-1622.
[88] R. Steinbauer, J. Vickers. The use of Generalized Functions and Distributions
in General Relativity. Class. Quant. Grav. 23, pR 91-114, 2006.
[89] L. Schwartz. Sur l ’impossibilité de la multiplication des distributions. comptes
Rendus Acad.Sci. Paris 239, 1954, p. 847-848.
[90] H.B. Stewart, B. Wendroff. Two-phase flows: Models and Methods. J. Comput.
Phys. 56, 1984, p. 363-409.
[91] T.D. Todorov. Lecture Notes: Nonstandard Approach to J.F. Colombeau’s
Theory of Generalized Functions. Arxiv.org 1010.3482v, 2010.
[92] I. Toumi, A. Kumbaro. An approximate linearized Riemann solver for a two
fluid model. J. Comput. Physics 66, 1986, p. 62-82.
[93] J. Tysk. Comparison of two methods of multiplying distributions. Proc. Amer.
math; Soc. 93, 1985, p. 35-39.
[94] I. Zalzali, H. Abbas. Double scale method for solving nonlinear systems involv-
ing multiplication of distributions. J. Comput. Anal. Appl. 8, 2006, p. 39-51.
[95] G. Zapata (editor) Functional Analysis, Holomorphy and Approximation The-
ory. (Rio de Janeiro 1979) Lecture notes in Pure and Applied Math. 83, Marcel
Dekker Inc. New York 1981.

São Paulo J.Math.Sci. 7, 2 (2013), 201–239