Hyperfunctions

A brief introduction

University of Helsinki

Domast student seminar

November 3, 2023

Sami Vihko (UH) Hyperfunctions November 3, 2023 1 / 49

Presentation Overview

1 Formal introduction and motivation

2 A lightning look at distribution theory

3 What is a hyperfunction: Definitions and some properties

Sami Vihko (UH) Hyperfunctions November 3, 2023 2 / 49

Motivation for generalized functions

Why do we need generalized functions?

There are mathematical objects that we would like to treat as
functions, but they don’t quite fit into the classical definition of
a function.
The theory of generalized functions enlarges the arena, where
we can do for example differential calculus and Fourier
Analysis.

Sami Vihko (UH) Hyperfunctions November 3, 2023 3 / 49

Distributions vs. hyperfunctions I
The first class of generalized functions one usually encounters
are the Schwartz distributions. These can be defined on
C ∞ -manifolds.
Distributions are by definition linear functionals.

The hyperfunctions are another class of generalized functions

that can be defined on C ω (real analytic)-manifolds.
Hyperfunctions are boundary values of complex analytic
functions is some sense.

Distributions can also be represented as boundary values of

complex analytic functions. However, then the analytic functions
will need to satisfy certain growth condition near the boundary.
Remove the growth conditions ⇒ Hyperfunctions.

Sami Vihko (UH) Hyperfunctions November 3, 2023 4 / 49

Distributions vs. hyperfunctions II

Why hyperfunctions?
For example the ODE x2 f 0 (x) = f (x) on a domain containing 0
has non-smooth solutions, which are not distributions, but are
hyperfunctions.
The theory of hyperfunctions encompasses the distributions
as part of it.

Sami Vihko (UH) Hyperfunctions November 3, 2023 5 / 49

Distributions
Where do we encounter distributions
ODE/PDE-theory
Solutions that are not sufficiently differentiable in classical
sense.
The Green’s functions make sense only as distributions
Fourier analysis: Start from the test function space S(Rd )
instead of L1 (Rd ).
Integral operator theory: ”Allows more singular kernels.”
Stochastics: There exist interesting stochastic processes, which
do not have realizations on any nice function space, but rather
in some space of distributions.
Mathematical treatment of the so called Dirac calculus used by
physicists in NR Quantum theory.
Mathematical treatment of Quantum field theory: Even the
free (”trivial”) field theory requires distributions to make sense.
Sami Vihko (UH) Hyperfunctions November 3, 2023 6 / 49
Formal introduction to distributions: An example I

What is a distribution?
A prototypical example is the Dirac delta distribution δx
concentrated at a point x ∈ Rd .

Physicists like to call this the Dirac delta function, but

unfortunately there does not exist a classical function with the
desired properties for this object.
Most notably the two properties below are in contradiction: For
a measurable function δ : Rd → Rd : δx (·) ≡ δ(x − ·) = δ(· − x),
Z
f (y)δx (y)dy = f (x), ∀f ∈ C(Rd ) and δx (y) = 0, ∀y 6= x.
Rd

Sami Vihko (UH) Hyperfunctions November 3, 2023 7 / 49

Formal introduction to distributions: An example II
There are multiple ways to make sense of δx more intuitively
Set x = 0 and d = 1 for simplicity.
As the limit of sequence of C ∞ functions: ”fn → δ0 ” for eg.
(
1
0, if |x| ≥ 2n
fn (x) := 1
n, if |x| < 2n
n 2 2
fn (x) := √ e−n x
π
n 1 sin(nx)
fn (x) := sinc(nx) = .
π π x
As the derivative of the step function θ(x) := 1{x≥0} :

”θ0 (x) = δ0 (x)”

Sami Vihko (UH) Hyperfunctions November 3, 2023 8 / 49

Formal introduction to distributions: An example III

As a ”density” of probability measure, which has all its mass

concentrated at a single point: ”dP(x) = δ0 (x)dx”
As a boundary values of analytic functions: Breit-Wigner formula

i 1 1
 
f (x) = − ⇒ ” lim f = δ0 ”
2π x + i x − i &0

Note that z 7→ 1z , z = x + iy is analytic in both the open lower and the

upper half planes.

Three of these generalize to arbitrary distributions so these are the ways to

characterize distributions

Most important to us is the last one since it is the contact point to the theory
of hyperfunctions

Sami Vihko (UH) Hyperfunctions November 3, 2023 9 / 49

Rigorous definition of distributions

Definition: The space of distributions

Let Ω ∈ Rd be open set and M(Ω) a suitable test function space.
Then the space of corresponding distributions is the topological
dual space M0 (Ω), which consists of continuous linear functionals
F : M(Ω) → R (C).

ordinary test functions:

M(Ω) = D(Rd ) ≡ Cc∞ (Rd ) ≡ C0∞ (Rd ), which are smooth and have
compact support. (Could also take arbitrary Ω ⊂ Rd .)
This yields the space of ordinary distributions D0 (Rd ) (D0 (Ω)).
Thus, rigorously we define δx [ϕ] = ϕ(x), ∀ϕ ∈ D(Rd ) (or ∀ϕ ∈ D(Ω)
with δx [ϕ] = 0 if x ∈
/ Ω.)

Sami Vihko (UH) Hyperfunctions November 3, 2023 10 / 49

The continuity of a linear map F : D(Ω) → R
One possible way to characterize the continuity of a linear map
F : D(Ω) → R(C) is the following:

Thm: Continuity of linear functionals on D(Ω)

Denote kϕkK,j = sup|α|≤j supx∈K |(∂ α ϕ)(x)| for compact K ⊂ Ω and
ϕ ∈ D(Ω). Then F : D(Ω) → R is continuous iff ∀K ∈ Ω compact,
∃j ∈ N0 and CK > 0 such that

|hF, ϕi| ≤ CK kϕkK,j .

Sometimes the following concept is necessary.

Definition:The order of a distribution

The order of F ∈ D0 (Ω) is the smallest integer for, which the above
holds for all compact sets simultaneously. If this does not exist
then the order is infinite.
Sami Vihko (UH) Hyperfunctions November 3, 2023 11 / 49
Support of a distribution

Let F ∈ D0 (Ω) and U ⊂ Ω an open set. Then F is said to be zero on U

if F[ϕ] = 0 for all ϕ ∈ D(U).

Definition: the support of a distribution

The support of F is the complement S of cthe largest open set on
which F is zero, that is, supp(F) = ( U) , where the union is over all
open U ⊂ Ω such that F is zero on U.

Remark: To see that this definition makes sense we would the

localization/decomposition theorems of distributions.

Sami Vihko (UH) Hyperfunctions November 3, 2023 12 / 49

Examples of distribution spaces I
Other possible test function spaces
The space of smooth rapidly decreasing functions S(Rd ) (also
called the space of Schwartz functions). NOTE: Needs Ω = Rd !
These and all their derivatives decrease more rapidly than any
|x|→∞
inverse monomial: ϕ ∈ S(Rd ), |x|k (∂ α ϕ)(x) −→ 0, ∀k ∈ N and
for all multi-indeces α.
The space of smooth functions E(Rd ) ≡ C ∞ (Rd ) (could take
open Ω ⊂ Rd )
The duals
S 0 (Rd )=tempered distributions
E 0 (Rd )=distributions with compact support

We have the following embeddings

D(Rd ) ,→ S(Rd ) ,→ E(Rd ) ,→ E 0 (Rd ) ,→ S 0 (Rd ) ,→ D0 (Rd ).

Sami Vihko (UH) Hyperfunctions November 3, 2023 13 / 49

Examples of distribution spaces II
More possible test function spaces and further embeddings
Continuous functions with compact support: M(Ω) = Cc (Ω)
M0 (Ω) = MR (Ω) the space of all Radon Measures with support
on Ω.
Remark: One could take this as the definition of Radon measures!

Sobolev spaces: M(Ω) = W k,2 (Ω) ≡ Hk (Ω)

M0 (Ω) = H−k (Ω)
For our prototype distribution we have
δx ∈ D0 (Rd ), S 0 (Rd ), E 0 (Rd ), MR (Rd ) and δx ∈ H−k (Rd ), ∀k > d2 .

We also have the embedding L1loc (Ω) ,→ D0 (Ω) given by

Z
f [ϕ] := f (x)ϕ(x)dx for f ∈ L1loc (Ω), ∀ϕ ∈ D(Ω),
Ω

where
Z
L1loc (Ω) := {f measurable| |f (x)|dx < ∞, for all compact Λ ⊂ Ω}
Λ

Sami Vihko (UH) Hyperfunctions November 3, 2023 14 / 49

Topologies and notion of convergence I

The test function spaces S(Rd ) and E(Ω) are countably seminormed
spaces. There are different equivalent families of seminorms.
They are metrizable and even Frechet spaces (Complete with
respect to the induced Frechet metric).
They are NOT Banach spaces!

The test function spaces D(Ω) and Cc (Ω) are not metrizable and
their topology cannot be induced by a countable family of
seminorms.
Their topology is defined via so called strict inductive limit of
Frechet spaces of similar structure on fixed compact sets
K ∈ Ω.
They are also called LF-spaces

Sami Vihko (UH) Hyperfunctions November 3, 2023 15 / 49

Topologies and notion of convergence II
Definition: Seminorm
On a Vector space X over K = R, C, a mapping p : X → R satisfying
1 p(x + y) ≤ p(x) + p(y)
2 p(λx) = |λ|p(x)
3 p(0) = 0 and p(x) ≥ 0
for all x, y ∈ X and λ ∈ K is called a seminorm on X. NOTE: p(x) = 0 ; x = 0!

Definition:Induced Frechet metric

The mapping d : X × X → R.

∞
X 1 pn (x − y)
d(x, y) :=
2n 1 + pn (x − y)
n=1

is called the Frechet metric induced by the countable family P := {pn }n∈N of seminorms pn .

The family P induces the same topology as the metric d.

Sami Vihko (UH) Hyperfunctions November 3, 2023 16 / 49

Topologies and notion of convergence III
Without explicitly defining the topology we can characterize the
convergence in the space of ordinary test functions:

Convergence of sequences in D(Rd ) (D(Ω)):

A sequence (ϕn )n∈N converges in D(Rd ) (D(Ω)) if and only if ∃K ∈ Rd (Ω)
compact and such that
supp(ϕn ) ∈ K for all n ∈ N
limn→∞ supx∈K |(∂ α ϕn )(x) − (∂ α ϕ)(x)| = 0 for all multi-indeces α ∈ Nd0 ,
where by ∂ α f with α = (0, . . . , 0) we mean just the function f itself.

For the spaces S(Rd ) and E(Ω), the convergence with respect to the induced
Frechet metric is equivalent to convergence in the topology induced by the
seminorms.
Remark: Due to continuity ϕn → ϕ implies F(ϕn ) → F(ϕ) for ϕ in any of
the above spaces and F in its dual.

Sami Vihko (UH) Hyperfunctions November 3, 2023 17 / 49

Topologies and notion of convergence IV

On the distribution spaces D0 (Rd ), E 0 (Rd ) (or with Ω) and S 0 (Rd ) (or more
generally on the dual of TVS) one can define multiple topologies (see eg.
Treves sec. 19).

The weak-∗ topology is probably the most commonly used.

Definition: Convergence of sequences M0 (Rd ) (M0 (Ω))

A sequence (Fn )n∈N converges to F in the weak-∗ topology of M0 (Rd ) iff
Fn (ϕ) → F(ϕ) in R for all ϕ ∈ M(Rd ) (or similarly with Ω).

Sami Vihko (UH) Hyperfunctions November 3, 2023 18 / 49

Operations on distributions I
Let us introduce the dual pairing of ϕ ∈ M(Ω) and F ∈ M0 (Ω):
hF, ϕi := F[ϕ].
Then we can define the following operations of distributions
1 Multiplication by a smooth function: For f ∈ E(Ω) and F ∈ D0 (Ω), fF
is an ordinary distribution satisfying hfF, ϕi = hF, f ϕi, ∀ϕ ∈ D(Ω).
2 Differentiation: For F ∈ M0 (Ω) and all multi-indeces
α := (a1 , a2 . . . , ad ) ∈ Nd0 , ∂ α F := ∂dad . . . ∂2a2 ∂1a1 F is the distribution
satisfying h∂ α F, ϕi = (−1)|α| hF, ∂ α ϕi for all ϕ ∈ M(Ω).
Remark:
For f ∈ C ∞ (Ω) the definition coincides with the conventional
differentiation by simple integration by parts.
Every distribution posesses derivatives of all orders and the usual
rules of calculus apply
The differential operators Dα are linear and continuous.

Sami Vihko (UH) Hyperfunctions November 3, 2023 19 / 49

Operations on distributions II

3 Fourier transform: For F ∈ S 0 (Rd ), FF ∈ S 0 (Rd ) is the tempered

distribution satisfying hFF, ϕi = hF, Fϕi for all ϕ ∈ S(Rd ).
Remark: On S(Rd ) the Fourier transform is a bijective integral
operator.
4 Convolution with a test function: Denote ϕx (y) = ϕ(x − y) and note
that ϕ ∈ D(Rd ) implies that the function y 7→ ϕ(x − y) is also in D(Rd ).
Then the convolution of F ∈ D0 (Rd ) with ϕ ∈ D(Rd ) is the smooth
function

(F ∗ ϕ)(x) := hF, ϕx i = hF, ϕ(x − ·)i.

We can also define the convolution and tensor product of two distributions,
but we CANNOT define an ”ordinary” multiplication on M0 (Ω) or take
non-linear functions of distributions.

Sami Vihko (UH) Hyperfunctions November 3, 2023 20 / 49

Now we can state the three general characterizations of
distributions

The last one of these is the point of contact with

Hyperfunctions.

Sami Vihko (UH) Hyperfunctions November 3, 2023 21 / 49

distributions as derivatives of continuous functions
Let Fu denote the the distribution associated to the function u ∈ L1loc (Ω) defined by
Fu [ϕ] := Ω u(x)ϕ(x)dx for all ϕ ∈ D0 (Ω).
R

THM:Distributions as weak derivatives of continuous functions (Prop. 5.2

in [Blanchard and bruning])
Let F ∈ D0 (Ω). Then for each multi-index α, there exists a function
uα ∈ C(Ω) such that
1 Each compact set K ∈ Ω intersects with the supports of only finitely
many uα
2 For all ϕ ∈ D(Ω),
X
hF, ϕi = h∂ α Fuα , ϕi.
α

If the order of F is finite, then only finitely many uα are needed for such a
representation of F.

Sami Vihko (UH) Hyperfunctions November 3, 2023 22 / 49

Distributions as limits of sequences of smooth
functions
Definition: Regularizing sequence (Def 7.1 in [Blanchard and bruning])
A sequence (ϕj )j∈N with ϕj ∈ D(Rd ), is called regularizing sequence iff
1 ∃ϕ ∈ D(Rd ) s.t. ϕj (x) = jd ϕ(jx), ∀x ∈ Rd and j ∈ N.
2 0 ≤ ϕj (x) for all x ∈ Rd and j ∈ N.
ϕ (x)dx = 1 for all j ∈ N.
R
3
Rd j

Then we have the following theorem

THM:Approximation of distributions (Thm 7.2 in [Blanchard and bruning])
Let T ∈ D0 (Rd ) and (ϕj )j∈N be arbitrary distribution and regularizing sequence
respectively. Then (Tj )j∈N with T ∗ ϕj ∈ C ∞ (Rd ) converges to T in the weak topology
of D0 (Rd ).

Thus, every distribution is a limit of sequence of smooth functions (Or more

properly since there is no uniqueness an equivalence class of Cauchy
sequences of smooth functions)
Sami Vihko (UH) Hyperfunctions November 3, 2023 23 / 49
Distributions as boundary values of analytic
functions I
For simplicity we work in d = 1 from now on!
Let u be a holomorphic function on the upper half plane
C+ := {z = x + iy ∈ C|y > 0} and let Fu,y , y > 0 denote the family of
distributions defined by
Z
hFu,y , ϕi := u(x + iy)ϕ(x)dx
Ω

for all ϕ ∈ D(Ω).

Definition: Boundary value of a holomorphic function

Let u be a holomorphic function on C+ . Then u is said to have a boundary
value u+ ∈ D0 (Ω) iff limy&0 hFu,y , ϕi := hu+ , ϕi exists in C for every ϕ ∈ D(Ω).

We have analogous definition for the lower half plane

C− = {z = x + iy ∈ C|y < 0}!

Sami Vihko (UH) Hyperfunctions November 3, 2023 24 / 49

Distributions as boundary values of analytic
functions II
Theorem:Distributions as boundary values of analytic functions(Thm 11.3
in [Blanchard and Bruning] (a = 1), or thm 3.19 in [Berenstein and Gay])
For every F ∈ D0 (Ω) there exists an analytic function u on [supp(F)]c
satisfying the following
1 For every compact K ⊂ Ω there exists constants a, C > 0 and m ∈ N

C
|u(x + iy)| ≤ , ∀x ∈ K and |y| ∈ (0, a]
|y|m

We call u satisfying this condition a function of slow growth.

2

Z
hF, ϕi = lim [u(x + i) − u(x − i)]ϕ(x)dx
&0 Ω

for all ϕ ∈ D(Ω).

Then we write F = u(· + i0) − u(· − i0) ≡ u+ − u− .
Sami Vihko (UH) Hyperfunctions November 3, 2023 25 / 49
Distributions as boundary values III

Outline of the proof:

1 Begin with F ∈ E 0 (Ω)
The Cauchy transform b F of F ∈ E 0 (Ω) defined by
1 1
F(z) := 2πi
b hF, ·−z i for z ∈ [supp(F)]c is an analytic function on its
domain of definition and satisfies the growth conditions.
2 Show that
Z
hF, ϕi = lim F(x + i) − b
[b F(x − i)]ϕ(x)dx.
&0 Ω

for all ϕ ∈ D(Ω)

3 Approximate a general distribution F ∈ D0 (Ω) with distributions
of compact support Fn ∈ E 0 (Ω). This is done by a standard
argument with partition of unity on compact exhaustion of Ω.

Sami Vihko (UH) Hyperfunctions November 3, 2023 26 / 49

Notation
For the rest of the presentation we will fix the following notations.

Let W ⊂ C be a non-empty open set, then H (W) denotes the

space of holomorphic functions.
It is endowed with the topology of uniform convergence on
compact sets (This is a Frechet space).
Denote C∗ := C \ R = C+ ∪ C− and let Ω ⊂ R be a non-empty
open set.
Then set Ω̃ := C∗ ∪ Ω. We also denote

U(Ω) := {V ∈ C open |V ∩ R = Ω}

and call this the set of complex neighbourhoods of Ω.

Note that Ω is closed in any V ∈ U(Ω).
For W ⊂ C open we denote W σ := W ∩ Cσ
Lastly for any V1 , V2 ∈ U(Ω) with V1 ⊂ V2 we denote by ρVV12 the
restriction map H (V2 ) → H (V1 ).
Sami Vihko (UH) Hyperfunctions November 3, 2023 27 / 49
Definition and representations of hyperfunctions I

Definition: The space of hyperfunctions

We call the complex vector space
 ∗
B(Ω) := H (C∗ ) ρC
Ω̃
(H (Ω̃))

the space of hyperfunctions on Ω and a hyperfunction on Ω is by

definition an element in this space.

Remark 1: Since this space is defined as a quotient space an

element in this space is an equivalence class.
Remark 2: Heuristically by this definition a hyperfunction can be
thought as being a holomorphic function on C∗ modulo a
holomorphic function on a larger domain Ω̃ restricted to the
smaller domain.

Sami Vihko (UH) Hyperfunctions November 3, 2023 28 / 49

Definition and representations of hyperfunctions II
Theorem
For every V ∈ U(Ω) there exists a natural map
iV  V\Ω
B(Ω) → H (V \ Ω) ρV (H (V)) := BV (Ω)

that assigns F := [f ] ∈ B(Ω) with f ∈ H (VC∗ ) to F 0 := [f |V∩C∗ ] ∈ BV (Ω) and which is

an isomorphism, that is,

B(Ω) ∼
= BV (Ω).

∗
The map is induced by the restriction mapping ρCV\Ω .
The fact that iV is well defined, linear and injective are proved by elementary
considerations from the definitions.
Proof of surjectivity uses Mittag-Lefler theorem.
Remark: For any V ∈ U(Ω), iV−1 ([g]) with g ∈ H (V \ Ω) is hyperfunction on Ω and
this representation is unique. g is called the defining function of the
hyperfunction. Thus, we identify F ≡ F 0 from above.

Sami Vihko (UH) Hyperfunctions November 3, 2023 29 / 49

Definition and representations of hyperfunctions III

Theorem
Let V1 , V2 ∈ U(Ω) and gi ∈ H (Vi \ Ω) be the defining function of Fi ∈ B(Ω), i = 1, 2.
Then F1 = F2 iff g1 |V1 ∩V2 ∩C∗ − g2 |V1 ∩V2 ∩C∗ can be analytically continued across Ω.
In particular, If B(Ω) 3 F := [f ] with f ∈ H (C∗ ) and V ∈ U(Ω), then g ∈ H (V \ Ω)
represents F iff there exist h ∈ H (V) such that f |V∩C∗ = g|V∩C∗ + h|V∩C∗ .

Remark
We can decompose H (V \ Ω) 3 ϕ = ϕ+ + ϕ− with ϕσ ∈ H ([V \ Ω]σ ).
Then ϕ represents 0 ∈ B(Ω) iff ϕ+ can be analytically continued to ϕ− and vice
versa. Thus, F is the zero hyperfunction iff the defining function g can be
analytically continued across Ω.

Sami Vihko (UH) Hyperfunctions November 3, 2023 30 / 49

Operations on hyperfunctions I

1 Multiplication by an analytic function:

Definition
Let V ∈ U(Ω), g ∈ H (V \ Ω) is the defining function of F ∈ B(Ω) and
h ∈ H (V). Then Fh = hF is the hyperfunction on Ω with
hg ∈ H (V \ Ω) as the defining function.

For h ∈ H (V), 1V σ h ∈ H (V \ Ω) and the hyperfunction

associated to it is 1σΩ h, where 1σΩ is the hyperfunction on Ω
associated to 1Cσ .
−
Then we also have 1+
Ω h = −1Ω h since obviously 1C+ h + 1C− h is
an example of the previous remark.

Sami Vihko (UH) Hyperfunctions November 3, 2023 31 / 49

Operations on hyperfunctions II

Decomposition of hyperfunctions:
For every F ∈ B(Ω) with defining function g ∈ H (V \ Ω),
V ∈ U(Ω) we can write F = F + + F − , where F σ has the defining
function 1V σ g.
This multiplication is associative:
For h1 , h2 ∈ H (V) and F ∈ B(Ω), (h1 + h2 )F = h1 F + h2 F
2 Conjugation:

Definition:Conjugate of a hyperfunction
Let V ∈ U(Ω) and F ∈ B(Ω) have defining function g ∈ H (V \ Ω). Then the
hyperfunction denoted by F̄ and with the defining function
z 7→ −g(z̄) ∈ H (V̄ \ Ω), where V̄ := {z ∈ C|z̄ ∈ V} is called the conjugate of
F.

F̄¯ = F for all F ∈ B(Ω)

Sami Vihko (UH) Hyperfunctions November 3, 2023 32 / 49

Operations on hyperfunctions III

For λ1 , λ2 ∈ C and F1 , F2 ∈ B(Ω) we have

λ1 F1 + λ2 F2 = λ̄1 F̄1 + λ̄2 F2 .
The real and imaginary parts of F ∈ B(Ω) are respectively
Re (F) := 12 (F + F̄) and Im (F) := 1/(2i)(F − F̄).
F ∈ B(Ω) is called real hyperfunction if F̄ = F.
A hyperfunction F is real precisely, when the defining function g
satisfies

1)g ∈ H (C∗ ) 2)g(z) = −g(z̄)

for all z ∈ C∗ .

Sami Vihko (UH) Hyperfunctions November 3, 2023 33 / 49

Real analytic functions and hyperfunctions
Theorem([Berenstein and Gay 2] prop. 1.2.5,1.2.7, def. 1.2.6)
The algebra A(Ω) of complex valued real analytic functions can be
injectively mapped into B(Ω). B(Ω) can be considered as an
A(Ω)-module.
The image of A(Ω) under this injection is called the set of
holomorphic hyperfunctions and they can all be written in one the
following equivalent forms

1+
Ωf or 1−
Ωg or 1+ −
Ω f1 + 1Ω f2

for some V, V1 , V2 ∈ U(Ω) and f , g ∈ H (V) or fi ∈ H (Vi ) with i = 1, 2.

The image of ϕ ∈ A(Ω) under this injection has 1C∗ ∩V Φ as its defining
function.
If ϕ ∈ A(Ω) and g ∈ H (V \ Ω) is the defining function of F ∈ B(Ω) with
V ∈ U(Ω), then the hyperfunction ϕF has Φg as its defining function.
Above Φ ∈ H (V) is the extension of φ to V ∈ U(Ω).
Sami Vihko (UH) Hyperfunctions November 3, 2023 34 / 49
Differentiation of hyperfunctions I

Definition of the derivatives of a hyperfunction

Let V ∈ U(Ω) and g ∈ H (V \ Ω) be the defining function of the
hyperfunction F ∈ B(Ω). Then we define the nth derivative of the
hyperfunction F to be the hyperfunction Dn T ≡ T (n) that has
n
g (n) := ddzng ∈ H (V \ Ω) as its defining function.

Properties of the operators Dn : B(Ω) → B(Ω)

They are linear
A(Ω) is an invariant subspace for all n ∈ N
For ϕ ∈ A(Ω), Dn ϕ coincides with the usual derivative

Sami Vihko (UH) Hyperfunctions November 3, 2023 35 / 49

Differentiation of hyperfunctions II

For ϕ ∈ A(Ω) and F ∈ B(Ω) the Leibniz formula

n  
n n
(Dj ϕ)Dp−j F
X
D (ϕF) =
j
j=0

holds.
If Ω = (a, b) for some a < b ∈ R, then F ∈ B(Ω) is a
holomorphic hyperfunction with a polynomial of degree ≤ n as
a defining function iff F (n+1) = 0.
The nth order primitive of F ∈ B(Ω) is the hyperfunction G
satisfying G(n) = F unique up to a polynomial of order ≤ n − 1.
On Ω = (a, b) every hyperfunction admits primitives of all
orders.

Sami Vihko (UH) Hyperfunctions November 3, 2023 36 / 49

Examples

1 1
Delta function δ0 ∈ B({0}) is the hyperfunction with z 7→ 2πi z
as its defining function.
The step function θ ∈ B([0, ∞))) is the hyperfunction with the
−1
defining function z 7→ 2πi log(−z), where log is the principal
branch of the complex logarithm.
d −1 1 1
Remark 1: One can easily see dz ( 2πi log(−z)) = 2πi z , where this
0
makes sense. Thus θ = δ0 as hyperfunctions.
Remark 2: Both of the defining functions satisfy the slow growth
conditions so we expect that θ, δ0 ∈ D0 (Ω)
Let F ∈ B(Ω) with 0 ∈ Ω be a hyperfunction with defining
1
function z 7→ e± z . Then F ∈
/ D0 (Ω).

Sami Vihko (UH) Hyperfunctions November 3, 2023 37 / 49

Locality of hyperfunctions I
The hyperfunctions admit a concept of support!.
First we need the concept of restriction of a hyperfunction.

Ω1 : B(Ω2 ) → B(Ω1 )
Definition: Restriction mapping RΩ2

Let Ω1 ⊂ Ω2 ⊂ R be non-empty open sets and g ∈ H (V2 \ Ω2 ) with

V2 ∈ U(Ω2 ) be the defining function of F ∈ B(Ω2 ). Then the
restriction of F to Ω1 denoted by RΩ
Ω1 (F) or F|Ω1 is the hyper function
2

in B(Ω1 ) with defining function g|(V1 ∩V2 )\Ω1 for any V1 ∈ U(Ω2 ).

Remark: F|Ω1 is independent of the sets V1 , V2 and the representative defining

function g. The restriction mapping RΩ
Ω1 is linear.
2

Definition
We say that F ∈ B(Ω) is zero or a holomorphic hyperfunction on an open
U ⊂ Ω if F|U is the zero or a holomorphic hyperfunction respectively.

Sami Vihko (UH) Hyperfunctions November 3, 2023 38 / 49

Locality of hyperfunctions II
Definition:Support of a hyperfunction
The support of a hyperfunction F ∈ B(Ω) denoted by suppΩ (F) (or if clear
from context just supp(F)) is the complement of the largest open set
U ∈ Ω such that F = 0 on U.
The support of supp(F) has the following two properties
Pn Sn
supp( i=1 λi Fi ) ⊂ i=1 supp(Fi ) for any Fi ∈ B(Ω), λj ∈ C,
i = 1, 2, . . . , n.
suppΩ1 (F|Ω1 ) = suppΩ2 (F) ∩ Ω1 for F ∈ B(Ω), Ω1 ⊂ Ω2 .

Sami Vihko (UH) Hyperfunctions November 3, 2023 39 / 49

Locality of hyperfunctions III

The hyperfunctions have analogues of the locality and decomposition

theorems of distributions.

Theorem: Localization of hyperfunctions

Let (Ω
Sj )j∈J be an open covering of Ω, that is, Ωj is open for all j ∈ J and
Ω ⊂ j∈J Ωj . Let Tj ∈ B(Ωj ), j ∈ J be a collection of hyperfunctions such that
for all i, j ∈ J with Ωi ∩ Ωj 6= ∅ we have

Ti |Ωi ∩Ωj = Tj |Ωi ∩Ωj .

Then there exists unique T ∈ B(Ω) such that T|Ωj = Tj for all j ∈ J.

Sami Vihko (UH) Hyperfunctions November 3, 2023 40 / 49

Locality of hyperfunctions IV
Theorem: Decomposition of hyperfunctions
Let (Fj )j∈N be a locally finiteSsequence of of non-empty relatively closed
subset sets of Ω such that j∈N Fj = Ω. Let T ∈ B(Ω). Then there exists a
sequence (Tj )j∈N with Tj ∈ B(Ω) for all j ∈ N such that supp(Tj ) ⊂ Fj for all
j ∈ N and such that for every open Ω0 Ω, there exist n(Ω0 ) ≡ n so that
n
X
T|Ω0 = Tj |Ω0 .
j=1

Sami Vihko (UH) Hyperfunctions November 3, 2023 41 / 49

Locality of hyperfunctions V
However, the following theorem has no analogue in the theory of
distributions

Theorem
The restriction map RRΩ : B(R) → B(Ω) is surjective, that is, for any F ∈ there
exists F0 ∈ R such that F = F0 |Ω and supp(F0 ) ⊂ Ω̄.

Outline of proof:
Pick V = C \ ∂Ω, then V ∈ U(Ω) and the defining function
g ∈ H (V \ Ω) = H (C \ Ω̄) of F ∈ H (Ω). Let F0 ∈ B(R) have g|C\R as its
defining function. Then F0 |Ω = F and supp(F0 ) ⊂ Ω̄.

Sami Vihko (UH) Hyperfunctions November 3, 2023 42 / 49

Locality of hyperfunctions VI

Ω2
We can generalize this to situation where we have RΩ 1
with Ω1 ⊂ Ω R
and both open. We also have the corollaries

Corollary
1 Let Ωi , i = 1, 2 be as above, then

B(Ω1 ) ∼
= {F ∈ B(Ω2 )|supp(F) ⊂ Ω̄1 }.

Corollary
2 Let S ⊂ Ω be closed subset of the open set Ω. Then

B(S) ∼
= {F ∈ B(Ω)|supp(F) ⊂ S}.

For latter take Ω2 = Ω \ S and Ω1 = Ω in the former.

Sami Vihko (UH) Hyperfunctions November 3, 2023 43 / 49

Distributions and hyperfunctions
Theorem
There exists a canonical injection i : D0 (Ω) → B(Ω).
This injection preserves the support, that is,

supp(F) = supp(i(F)) for F ∈ D(Ω).

The image i(D0 (Ω)) is a subspace of B(Ω). F ∈ i(D0 (Ω)) have

defining functions that satisfy the slow growth condition in Ω.

Corollary
We have the following isomorphism
∗
D0 (Ω) ∼
= HSG,Ω (C∗ )/ρC
Ω̃
(H (Ω̃)),

where HSG,Ω (C∗ ) is the space of holomorphic functions in C∗ that

satisfy the slow growth condition in Ω.
Sami Vihko (UH) Hyperfunctions November 3, 2023 44 / 49
Analytic functionals and hyperfunctions I

Definition: Analytic functional

Let W ∈ C be open set. Then an analytic functional on W ∈ C is an
element of the dual space H 0 (W)

Definition: Locally analytic functional

Let K ∈ C be a compact set. Then a locally analytic functional is an
element in the dual space H 0 (K).

If we have open or compact subsets of the real line embedded into

the complex plane and real analytic functions on these sets we use
similar terminology.

Sami Vihko (UH) Hyperfunctions November 3, 2023 45 / 49

Analytic functionals and hyperfunctions II

Theorem
Let K ⊂ R ⊂ C be a compact set and Ω ⊂ R ⊂ C be an open set. Then
we have the following isomorphisms

A0 (K) ∼
=B(K)
A (Ω) ∼
0
=Bc (Ω),

where Bc (Ω) is the space of hyperfunctions on Ω with compact

support.

Sami Vihko (UH) Hyperfunctions November 3, 2023 46 / 49

References I

François Tréves
Topological vector spaces, Distributions and Kernels
Elsevier, 1967
Walter Rudin
Functional analysis
McGraw-Hill, 1974
Philippe Blanchard and Erwin Brüning
Mathematical methods in physics: Distributions, Hilbert spcae
operators, Variational methods, and applications in quantum physics,
Second edition
Birkhäuser, 2015
Yvonne Choquet-Bruhat, Cécile DeWitt-Morette (and Margaret
Dillard-Bleick)
Analysis, Manifolds and physics, Part I:Basics
Elsevier /North Holland, 1982

Sami Vihko (UH) Hyperfunctions November 3, 2023 47 / 49

References II

Carlos A. Berenstein and Roger Gay

Complex variables: An introduction
Springer-Verlag, 1991
Carlos A. Berenstein and Roger Gay
Complex analysis and special topics in harmonic analysis
Springer-Verlag, 1995
Mitsuo Morimoto
An introduction to Sato’s Hyperfunctions
American mathematical society, 1993
Translated from original Japanese version from 1976
Lars Hörmander
The analysis of linear partial differential operators I, second edition
Springer-Verlag, 1989

Sami Vihko (UH) Hyperfunctions November 3, 2023 48 / 49

 The End
Questions? Comments?