Psychophysical Neuroeconomics of Decision Making: Nonlinear Time Perception Commonly Explains Anomalies in Temporal and Probability Discounting ()
1. Introduction
Canonical representations on Hermitian symmetric spaces
were introduced by Vershik-Gelfand-Graev [1] (for the Lobachevsky plane) and Berezin [2]. They are unitary with respect to some invariant non-local inner product (the Berezin form). Molchanov’s idea is that it is natural to consider canonical representations in a wider sense: to give up the condition of unitarity and let these representations act on sufficiently extensive spaces, in particular, on distributions. Moreover, the notion of canonical representation (in this wide sense) can be extended to other classes of semisimple symmetric spaces
, in particular, to para-Hermitian symmetric spaces, see [3]. Moreover, sometimes it is natural to consider several spaces
together, possibly with different
, embedded as open
-orbits into a compact manifold
, where
acts, so that
is the closure of these orbits.
Canonical representations can be constructed as follows. Let
be a group containing
(an overgroup),
a series of representations of
induced by characters of some parabolic subgroup
associated with
and acting on functions on
. The canonical representations
of
are restrictions of
to
.
In this talk we carry out this program for para-Hermitian symmetric spaces of rank one. These spaces are exhausted up to the covering by spaces
with
,
. For these spaces
, an overgroup is the direct product
and canonical representations turn out to be tensor products of representations of maximal degenerate series and contragredient representations. These tensor products are studied in [4], see also [5]. So we lean essentially on these papers [4,5]. We decompose canonical representations into irreducible constituents and decompose boundary representations. Notice that in our case the inverse of the Berezin transform
can be easily written: precisely it is the Berezin transform
.
Canonical and boundary representations for
in the case
(then
is the hyperboloid of one sheet in
) were studied in [6]. For the two-sheeted hyperboloid in
, it was done in [7].
In this paper we present only the main results. The detailed theory of canonical and boundary representations, for example, on a sphere with an action of the generalized Lorentz group, can be seen in [8].
Let us introduce some notation and agreements.
By
we denote
. The sign
denotes the congruence modulo 2.
For a character of the group
we shall use the following notation

where
,
,
.
For a manifold
, let
denote the Schwartz space of compactly supported infinitely differentiable
-valued functions on
, with a usual topology, and
the space of distributions on
—of anti-linear continuous functionals on
.
2. The Space
and the Manifold 
We consider the symmetric space
where
,
,
.
The group
acts on the space
by

Let us write matrices in
in block form according to the partition
of
. Let us take the matrix

The subgroup
is just the stabilizer of this point
, this subgroup consists of block diagonal matrices:

Thus, our space
is the
-orbit of
, it consists of matrices of rank one and trace one.
Equip
with the standard inner product
, let
. Let
be the sphere
. Let
be the Euclidean measure on
. The group
acts on
by
.
Let
be a cone in
consisting of matrices
of rank one. Therefore, the space
is the section of
by the hyperplane
.
Introduce a norm
in
by

where the prime denotes matrix transposition.
Let
be the section of
by
.
Define a map
by

It is a two-fold covering. The measure
defines a measure
on
by

The action of the group
on
gives the following action of
on
:

In particular, the subgroup
, a maximal compact subgroup, acts on
by translations:

Let us consider on
the function
(1)
The action on
has three orbits: namely, two open orbits (of dimension
):
and
and one orbit of dimension
:
. The orbit
is a Stiefel manifold, it is the boundary of
. Denote
. Each of orbits
can be identified with the space
. The map is constructed by means of generating lines of the cone
.
3. Maximal Degenerate Series Representations
Recall [4] maximal degenerate series representations
,
,
, of the group
. Let
be the subspace of
consisting of functions
of parity
:
. The representations
act on
by


4. Representations of
Associated with 
Recall [5] a series of representations
of the group
associated with the space
.
Denote by
the space of functions
in
of parity
:

The representation
acts on
by
(2)
Let
denote the following sesqui-linear form
(3)
Define an operator
on
by

It intertwines
and
. The operator
is a meromorphic function of
. Let us normalize this operator (multiplying it by a function of
) such that the normalized operator
is an entire non-vanishing function of
.
There are three series of unitarizable irreducible representations. The continuous series consists of
with
,
, the inner product is (3). The complementary series consists of
with
, the inner product is 
with a factor. The discrete series consists of the representations
where
,
,
, which are factor representations of
on the quotient spaces
. The representations
with the same
and different
are equivalent. It is convenient to take
where
for odd
and
for even
. The inner product is induced by the form
.
5. Canonical Representations
We define canonical representations
,
,
, of the group
as tensor products:

They can be realized on
: let
denote the subspace of
consisting of functions
of parity
:
, then the representation
acts on
by a formula similar to (2):

The inner product
(4)
is invariant with respect to the pair
, i.e.
(5)
Consider an operator
on
defined by

It turns out that the composition
is equal to the identity operator
up to a factor. We can take
such that

namely,

With the form (4) the operator
interacts as follows:
(6)
This operator
intertwines the representations
and
, i.e.

Let us call it the Berezin transform.
Let
be the space of distributions on
of parity
. We extend
and
to
by (5) and (6) respectively and retain their names and the notation.
Let us introduce the following Hermitian form
on
:

Let us call this form the Berezin form.
6. Boundary Representations
The canonical representation
gives rise to two representations
and
associated with the boundary
of the manifolds
(boundary representations). The first one acts on distributions concentrated at
, the second one acts on jets orthogonal to
.
We can introduce “polar coordinates” on
corresponding to the foliation of
into
-orbits. The
- orbits are level surfaces of the function
, see (1). For
the
-orbits are diffeomorphic to
. In these coordinates the measure
on
is

where
is the measure on
.
Let
be a function in
. Consider it as a function of polar coordinates. Consider its Taylor series
in powers of
. Here
are functions in
. Denote by
,
,
, the space of distributions in
, having the form

where
,
is the Dirac delta function on the real line,
its derivatives. Let
.
Denote by
Taylor coefficients of the function
. The distribution
acts on a function
as follows:
(7)
Denote by
the restriction of
to
. This representation is written as a upper triangular matrix with the diagonal
,
.
Distributions in
can be extended in a natural way to a space wider than
. Namely, let 
be the space of functions
of class
on
and
of parity
and having the Taylor decomposition of order
:

where
. Then (7) keeps for
with
.
Let
denote the column of Taylor coefficients
. The representation
acts on these columns:

It is written as a lower triangular matrix with the diagonal
,
.
The boundary representations
and
are in a duality.
7. Poisson and Fourier Transforms
Let us write operators
and
intertwining representations
and
. We call them Poisson and Fourier transforms associated with canonical representations.
The Poisson transform
is a map
given by

It intertwines
with
. Here we consider
as the restriction to
of the representation
acting on distributions in
.
For a
-finite function
and
the Poisson transform has the following decomposition in powers of
:

where
has polar coordinates
. Here
and
are certain operators acting on
. The factors
and
give poles of the Poisson transform in
depending on
:
(8)
where
and
,
. If a pole belongs only to one of series (8), then the pole is simple, and if a pole belongs to both series (8), then
and the pole is of the second or first order.
Let the pole
,
, be simple. The residue
of
at this pole is an operator
. Denote the image of this operator by
.
The Fourier transform
is a map
given by

It intertwines
with
.
The Fourier and Poisson transforms are conjugate to each other:

Poles in
of the Fourier transform are situated at points
(9)
where
and
,
. If a pole belongs only to one of the series (9), then the pole is simple, and if a pole belongs to both series (9), then
and the pole is of the second or first order.
Let the pole
,
, be simple.
The residue
of
at this pole is a “boundary” operator
,
. The operator
is defined in terms of Taylor coefficients
: it is a linear combination of functions
. Therefore, we may consider the following operator
acting on columns
of functions
: this operator to any column
assigns the column
of functions in the same space
—by the same formulas without
. This operator
is given by a lower triangular matrix.
8. Decomposition of Boundary Representations
The meromorphic structure of the Poisson and Fourier transforms is a basis for decompositions of boundary representations
and
.
Let the pole
of the Poisson transform is simple, in particular, it happens when
. Then the boundary representation
is diagonalizable which means that
decomposes into the direct sum of
, and the restriction of 
to
is equivalent to
(by means of
).
If a pole is of the second order, then the decomposition of
contains a finite number of Jordan blocks, this number depends on
.
Let the pole
of the Fourier transform is simple, in particular, when
. Then the matrix
is diagonalizable which means that
is a diagonal matrix. Its diagonal is
,
.
If a pole is of the second order, then the decomposition of
contains a finite number of Jordan blocks, this number depends on
.
9. Decomposition of Canonical Representations
Let us write decomposition of canonical representations. We restrict ourselves to a generic case:
lies in strips

Case (A):
.
Theorem 1 Let
. Then the canonical representation
decomposes—as the quasiregular representation [5]—into irreducible unitary representations of continuous and discrete series with multiplicity one. Namely, let us assign to a function
the family of its Fourier components
,
,
,
, and
,
. This correspondence is
equivariant. There is an inversion formula:
(10)
and a “Plancherel formula” for the Berezin form:
(11)
Here
and
stand for the Plancherel measure for
, see [5], the factor
is given by following formula:

Case (B):
.
Here we continue decomposition (10) analytically in
from
to
,
. Some poles in
of the integrand intersect the integrating line—the line
. They are poles
and
of the Poisson transform with
. They give additional summands to the right hand side. So after the continuation we obtain:
(12)
where the integral and the series mean the same as in (10) and

are some numbers.
Similarly, the continuation of (11) gives
(13)
where the integral and the series mean the same as in (11) and
are some numbers.
The operators
,
, can be extended from
to the space
and therefore to the sum

Then these operators
turn out to be projection operators onto
. Moreover, there are some “orthogonality relations” for them. Decomposition (13) can also be extended to the space
. This decomposition is a “Pythagorean theorem” for decomposition (12).
Theorem 2 Let
,
. Then the space
has to be completed to
. On this space the representation
splits into the sum of two terms: the first one decomposes as
does in Case (A), the second one decomposes into the sum of
irreducible representations
,
. Namely, let us assign to any
the family

where
,
,
. This correspondence is
-equivariant. There is an inverse formula, see (12), and a “Plancherel formula”, see (13).
Case (C):
.
Now we continue decomposition (10) analytically in
from
to
. Here poles
and
,
,
, of the integrand (they are poles of the Fourier transform) give additional terms. We obtain
(14)
where the integral and the series mean the same as in (10) and

some numbers. The operators
can be extended to the space
,
. Denote by
the image of
. It turns out that the operators
are projection operators onto
and for them there are some “orthogonality relations”.
Now we continue decomposition (11) from
to
. Poles of the integrand which intersect the integrating line
and give additional terms (they are poles of both Fourier transforms) turn out fortunately to be of the first order, since at these points the function
as a function of
has zero of the first order. After the continuation we obtain:
(15)
where the integral and the series mean the same as in (11),
some numbers. It is a “Pythagorean theorem” for decomposition (14).
Theorem 3 Let
,
. Then the representation
considered on the space
splits into the sum of two terms. The first one acts on the subspace of functions
such that their Taylor coefficients
are equal to 0 for
and decomposes as
does in Case (A), the second one decomposes into the direct sum of
irreducible representations
,
acting on the sum of the spaces
. There is an inversion formula, see (14), and a “Plancherel formula” for the Berezin form, see (15).