1. Introduction
An attractive and noteworthy issue in differential geometry is the concept of statistical manifolds which was initially defined by Amari [
1]. Statistical manifolds have many application areas such as machine learning, general relativity, neural networks, physics, image analysis, control systems and many more [
2,
3,
4]. A statistical structure on a Riemannian manifold is inspired from statistical mold, where probability distributions correspond to the manifold points. In the recent years, statistical manifolds have been actively investigated by many mathematicians and interesting results have been obtained [
5,
6,
7,
8,
9,
10,
11,
12,
13].
Suppose that
is an open subset of
and
is a sample space with parameter
. A statistical model
S is the set of probability density functions defined by
The Fisher information matrix
on
S is given as ([
14])
where
is the expectation of
with respect to
,
and
. The space
S with together the information matrices is a statistical manifold [
15]. It is seen that
is a Riemannian manifold. An affine connection ∇ with respect to
is described by
Statistical submanifolds were described by Vos in 1989 [
16]. In 2015, Milijevic showed that a semi-parallel totally real statistical submanifold with some natural conditions is totally geodesic if it is of non-zero constant curvature [
17]. In 2015, Aydin et al. investigated curvature properties of statistical submanifolds [
18]. Moreover, in 2017, they generalized Wintgen inequality for statistical submanifolds in statistical manifolds of constant curvature [
19]. In 2018, Aytimur and özgür focused on submanifolds of statistical manifolds with quasi-constant curvature and obtained similar inequalities [
20]. Statistical submanifolds of Hessian manifolds with constant Hessian curvature are studied in [
21]. Then, Uohashi described conformally flat statistical submanifold of a flat statistical manifold using a Hessian domain [
22]. Lee et al. derived extremities for normalized
-Casorati curvature for statistical submanifolds in statistical manifold with constant curvature [
23]. Alkhaldi et al. showed that normalized scalar curvature is bounded above by Casorati curvatures for statistical submanifolds in Sasaki-like statistical manifolds of constant
-sectional curvature in [
24]. In 2020, Jain et al. studied lightlike submanifolds of indefinite statistical manifolds and presented some conditions for the induced statistical Ricci tensor on a lightlike submanifold of indefinite statistical manifolds to be symmetric [
25]. In [
26], Aquib proved some of the curvature properties of submanifolds and provided a couple of inequalities for totally real statistical submanifolds of quaternionic Kaehler-like statistical space forms. In 2019, Chen et al. obtained a Chen first inequality for statistical submanifolds in Hessian manifolds of constant Hessian curvature [
27]. Decu et al. studied inequalities for the Casorati curvature of statistical manifolds in holomorphic statistical manifolds of constant holomorphic curvature in [
28]. Lone et al. defined Golden-like statistical manifolds and obtained certain interesting inequalities [
29]. Recently, Meli et al. considered null submanifold in [
30], where they presented statistical structures on a null hypersurface in the Lorentz-Minkowski space using the null second fundamental form. For some of the recent works, we refer to [
31,
32,
33,
34,
35,
36,
37].
In [
38] Balcerzak considered the affine combination of two affine connections
and
on a pseudo-Riemannian manifold and introduced a new connection as
, where
is a smooth function on the pseudo-Riemannian manifold. Until now, this type of connections on (sub)statistical manifolds have not been considered. This paper fills the gap. This work is intended as an attempt to introduce a new family of connections which is statistical on statistical submanifolds. We refer to these connections as
-statistical connections.
In this paper, we describe and study a family of statistical connections which called
-statistical connections on statistical manifolds in
Section 2. The Gauss and Weingarten formulas for
-statistical connections are given in
Section 3. Also, the geometric structures such as the
-second fundamental forms,
-mean curvatures, and the relations among them on a submanifold of a statistical manifold are examined. Then we show that if a statistical submanifold
N of a statistical manifold
is
-doubly autoparallel, then the fundamental form and its dual are vice versa with respect to 0-statistical connections. Also, the
-fundamental form and its dual are vice versa if
N is geodesic. Moreover, if
N is minimal with respect to
-statistical connection and its dual, then 0-mean curvature vector field and its dual are vice versa. We describe the concept of divergence for
-mean curvature vector fields and
-fundamental forms and we find the properties of them. In
Section 4, the equations of Gauss and Codazzi of
-statistical connections are obtained. Such structures with condition conjugate symmetric are discussed. We present a inequality for statistical submanifolds in real space forms with respect to
-statistical connections. Also, we obtain a basic inequality involving statistical Ricci curvature and the squared
-mean curvature of a statistical submanifold of statistical manifolds.
2. Statistical Manifolds
For an n-dimensional manifold
, consider
,
as a local chart of the point
. Considering the coordinates
on
, we have the local field
as frames on
. Assume that
is a pseudo-Riemannian metric on
. An affine connection
is called
Codazzi connection if the Codazzi equations satisfy:
for all
, where
The triple
is called a
statistical manifold% We prefer italics. if
is a statistical connection, i.e., a torsion-free Codazzi connection. In particular, it is known that if the cubic tensor field is zero, a torsion-free Codazzi connection
reduces to the Levi-Civita connection
. Moreover, the affine connection
of
defined by
is called the (dual) conjugate connection of
with respect to
. Immediately, one can see
and
forms a statistical manifold.
For a statistical structure
on
, if we consider a
-tensor field
described by
it follows that
satisfies
For an affine connection
, the curvature tensor
is defined as
On a statistical manifold
, we denote
and
by
and
, respectively for brief. It is known that the following hold
where
. Moreover, if
, the statistical manifold
is called a flat.
A statistical manifold
is called conjugate symmetric if the curvature tensors of the connections
and
, are equal, i.e., the following holds
Let
be a smooth manifold and
. The affine combination of two affine connections
and
on
is the connection
given by
Immediately, we see that
where
,
and
are the torsion tensors of
,
and
, respectively [
38].
Definition 1. For a statistical manifold , the family of connections given by affine combination of the conjugate connections and , i.e.,is called -statistical connection of . Assuming , and in the above definition, we obtain the statistical connections , and the Levi-Civita connections on , respectively.
Corollary 1. Let be a statistical structure on . Then
- 1.
is also a statistical manifold.
- 2.
The connection is dual of , i.e., - 3.
and satisfy the following
Proof. From Definition 1, it follows that the
-statistical connection
is torsion-free, i.e.,
Moreover, it satisfies the following condition
In the same way, (2) and (3) follow. □
3. Statistical Submanifolds
Consider
as an
-dimensional smooth manifold with a statistical structure
and
N as an
m-dimensional submanifold of
with the induced metric
h on it. Each tangent space of
has the orthogonal decomposition
where
. The Gauss and Weingarten formulas for dual connections are described by ([
39])
for any
and
. It results that
and
are statistical submanifolds, and
is the dual of ∇ with respect to
h and the tensor fields
,
,
A and
satisfy
Furthermore, the Levi-Civita connections
and
on
and
N, respectively are related to the fundamental form
by
Lemma 1. Let . The Gauss and Weingarten formulas for -statistical connections satisfyfor any and . Proof. Applying (
15) in Definition 1, we see that
Similarly, it follows (
21). □
According to the above equations, we set
where
In particular, setting
,
and
in (
23) we have (
15), (
16) and (
19), respectively.
Proposition 1. For a statistical submanifold of a statistical manifold with , we have
- 1.
The triple is a statistical submanifold in induced by .
- 2.
The connection is dual of , i.e., - 3.
The tensor fields and are related by - 4.
For any and , we have
Proof. We prove only (1). The parts (2)–(4) are obtained in a similar way. Since
is a statistical structure on
and
is also a statistical structure, we can write
which gives
On the other hand, for any
we have
Thus . Hence forms a statistical submanifold of . □
For a statistical submanifold
of a statistical manifold
, for any tangent vector fields
, we consider the difference tensor
K on
N as
From (
7), (
15) and the above equation, it follows that
Similarly, for
and
we have
where
Moreover, (
8) implies
which gives us
Regarding
, (
24) and the definitions of
and
imply
Now, consider
and
as orthonormal tangent and normal frames, respectively, on
N. The
-mean curvature vector field
with respect to
is defined by
where
. Moreover, the following identity holds
For
and
, we use the notations
and
, respectively. The above equation immediately leads to
Moreover, setting
and making use of the definitions
and
, we deduce
and
In particular, for
, we have
and
Lemma 2. The following relations holdand Proof. Using
, we get
The above two equations imply
Using (
31) in the last equation, we obtain (
33). In a similar way, (
34) follows. □
Definition 2. Considering as a statistical submanifold of a statistical manifold with , M is called f-autoparallel with respect to if . In addition, if and , N is f-doubly autoparallel. The statistical submanifold N is called geodesic if it is -autoparallel. If (), N is called -minimal (respectively, -minimal).
Proposition 2. On a statistical submanifold of a statistical manifold and , we have
- 1.
N is -doubly autoparallel if it is 0-doubly autoparallel.
- 2.
If N is -doubly autoparallel, then .
- 3.
if N is geodesic.
- 4.
If N is and -minimal, then .
Proof. If
N is 0-doubly autoparallel, we have
which gives
. Hence (1) follows. To prove (2), using
and
, we get
. Considering the part (3) of Proposition 1 and
, we get (3). Similar to proof (3) and using (
28), (4) is obtained. □
Example 1. The normal distribution manifold is described as Thus can be considered as a 2-dimensional manifold with a coordinate system . According to (1) and (2), the Fisher metric and the connection are obtained byand From (5), it follows that It follows thatwhich give is a statistical manifold. For the submanifoldwe get the Fisher metric of N as Thus is a statistical submanifold because Suppose that is a function on N. We get and on and N, respectively asand The above equations implyhence is a statistical submanifold. It is evident from the above results thatwhich give N is 0-doubly autoparallel and hence -doubly autoparallel. Lemma 3. The following relations hold:andfor any . Proof. Since
and using (
23) in the above equation, we can write
Thus (
35) holds. To prove (
36), considering
instead of
in (
35), it follows that
□
Consider
as an affine connection of a pseudo-Riemannian manifold
, the divergence of
is introduced as the trace of the covariant derivative
, i.e.,
In general for a tensor field
A of type
on
,
is obtained by
Now, suppose that
is a statistical manifold and
. (
7), (
14) and the above equation provide the explicit formula for
of
:
Proposition 3. On an -dimensional statistical manifold , where is an m-dimensional statistical submanifold of and , we have Proof. Using the definition of
, we can write
Thus from the above equation, it follows that
for any
. Considering
and summing over
s in the last equation, where
is an orthonormal basis on
N, we obtain (
38). Using (
37), we get
Then the above equations imply
Thus (
39) holds. Applying the above equation in (
38), we have
Subtracting the above two equations, we have (
40). □
Lemma 4. Let N be an m-dimensional submanifold with a statistical structure of an -dimensional statistical manifold and . Then
- (i)
- (ii)
- (iii)
Proof. Assume that
is an orthonormal basis on
N. Setting
in (
36) and summing over
s, we have
which gives us
The above equation leads to
(
40) and the last equation imply
Setting
and
in (
41), it follows
Applying the last equations in (
42), we see that (i) holds. The part (i) gives
On the other hand,
(see [
15]), hence (ii) is obtained. Using (
27) and (
35), we can write
In particular, for
, it follows
On the other hand, using (
28) and (
30), we get
The last three equations give (iii). □
Proposition 4. Let be a statistical submanifold in a statistical manifold and . Then we have
- 1.
If N is - (or -) minimal, then - 2.
If N is - (or -) minimal, then
Proof. According to Definition 2 and the part (iii) of Lemma 4, we have (1) and (2). □
Applying (
30) and Proposition 4, we conclude:
Corollary 2. The following conditions are equivalent:
- 1.
;
- 2.
;
- 3.
;
- 4.
;
- 5.
;
- 6.
.
4. The Gauss, Codazzi and Ricci Equations with Respect to the
-Statistical Connection
In this section, suppose that is a statistical submanifold in a statistical manifold and . We consider the statistical structures and , respectively on N and . For simplicity, we denote by , and , the -curvature tensor fields of the connections , and , respectively. Moreover, if (), (N) is called -flat.
Proposition 5. On a statistical submanifold in a statistical manifold , the curvature tensor satisfies the followingwhere and , for any . Proof. Since
, for any
, we obtain
The above equation and (
24) imply
On the other hand, we get
From the above two equations, the first formula follows. Similarly, the second formula follows from (
25). □
In the sequel, we review the following proposition:
Proposition 6 ([
40]).
Let be a statistical submanifold of a statistical manifold . The equations of Gauss, Codazzi, and Ricci are given byfor any and , whereSimilarly, one can see equations of Gauss, Codazzi, and Ricci for the curvature tensors and with respect to the connections and , respectively.
Now we get the -curvature tensor with respect to the -statistical connection , which is the counterpart of the above proposition in -statistical submanifolds.
Theorem 1. For a statistical submanifold of a statistical manifold with , the equations of Gauss, Codazzi, and Ricci for the -curvature tensor with respect to satisfy the followingwherein the same setting we have and for any and . Proof. Applying (
20) and (
21) in the above equation, it follows
By interchanging
and
in the last equation, we have
Again, using (
20) we obtain
Setting the last three equations in the definition of the curvature tensor and using
we obtain the first formula claimed by the theorem. In the same way, the other parts are concluded. □
Remark 1. Considering and in the above theorem, one can see the equations of Gauss, Codazzi, and Ricci for the curvature tensors , and with respect to the connections , and , respectively in Proposition 6.
According to the above theorem, we deduce the following corollaries:
Corollary 3. Let be a statistical manifold and be a doubly autoparallel statistical submanifold in . If is -flat where , then M is -flat.
Corollary 4. The following formulas holdfor any and . Lemma 5. The -curvature tensors and satisfy the following equations:for any . Proof. According to Theorem 1, we can write
Applying Proposition 5 in the above equation, we get the first part. Similarly, the second part follows. □
Corollary 5. Let be a doubly autoparallel statistical submanifold in a statistical manifold and . Then if and only if .
Proof. As
, from (
45) we get the assertion. □
Lemma 6. If is a conjugate statistical submanifold in a statistical manifold and , thenfor any . Moreover, if is -flat, then the following holds Proof. As
N is a conjugate statistical submanifold, so
. Hence, Lemma 5 implies (
47). Moreover, if
, we get (
49). □
From Lemmas 5 and 6, we conclude the following proposition:
Proposition 7. On a doubly autoparallel statistical submanifold of a statistical manifold and , if is a conjugate statistical structure on N, thenfor any . Moreover, if is constant, then we have In a statistical submanifold
, the Ricci curvature tensor
of the
-connection
is defined by
Similarly, of can be described analogously.
Theorem 2. Let be an -dimensional statistical manifold and be an m-dimensional statistical submanifold of . Let the statistical connection on has the constant curvature c, where . If and are orthonormal tangent and normal frames, respectively on N, then the Ricci tensor of N satisfiesfor any . Proof. Applying (
18) in Gauss equation of Theorem 1, we can write
for any
. As
thus from the above two equations and using (
51), it follows that
On the other hand, we have
Setting the above equations in (
53), we find the assertion. □
Moreover, one see that
where
.
Proposition 8. According to the hypothesis of Theorem 2, for any we have
- 1.
If N is conjugate symmetric with respect to the statistical connection ∇
, thenwhere . - 2.
If the function on N is constant, then the following holdand
Proof. (
44) gives
which gives
where
and
are the Ricci tensors associated with the statistical connections ∇ and
. The above equation and (
56) imply
If
N is conjugate symmetric, we have
and
. Hence (
61) implies (1). Using (
44) and (
51), we get
Setting
in the above equation and (
52), respectively, it follows
and
The above two equations imply (
57). From (
62), we have
On the other hand, (
54) yields
Moreover, considering
in (
54) we get
The last three equations give
Putting the last equation in (
64), it follows
From (
57) and the above equation, we get (
58). □
Corollary 6. We havefor any . Proof. Applying (
57) and (
58) in (
65), we have
. From (
66),
follows. □
Theorem 3. Let be an m-dimensional statistical submanifold of an -dimensional statistical manifold and . Let the statistical connection on has the constant curvature c. If least one of the following hold
- 1.
the Ricci tensor is symmetric;
- 2.
the Ricci tensor is symmetric;
- 3.
,
for any . Moreover, if is constant, the Ricci tensor is symmetric.
Proof. According to (
52), it follows
On the other hand, (
62) implies
From the above equations, we have
Moreover, from (
68) for
, we see that
Hence, (
67) is obtained and implies
if
is constant. □
Theorem 4. Let be an -dimensional statistical manifold of constant curvature c and be a statistical manifold of an m-dimensional statistical submanifold in . Thenwhere is the scalar curvature of , i.e., . Moreover, the equality holds if and only if perpendicular to . Proof. Setting
in (
53) and by summing over
, we have
As
and similarly
, so it follows
As
, the last equation implies
This and (
43) give us the assertion. □
Proposition 9. For an N-dimensional statistical submanifold in an -dimensional statistical manifold of the constant curvature c, if
- (i)
N has the property that and are perpendicular;
- (ii)
N is ∇- and -minimal;
Moreover, if is constant, it followswhere . Proof. On the other hand, (
70) yields
Considering
in the last equation, one see that
. Hence, from the above two equations, (
71) follows. (
72) is obtained from (
71). □
Let
be an
m-dimensional statistical submanifold in an
-dimensional statistical manifold
and
. For the statistical submanifold
induced by
, we define tensor fields
and
of type
on
and
N, respectively by
The tensor fields
and
are called
the statistical curvature tensor fields of
and
, respectively. The statistical Ricci curvature tensor and the statistical scalar curvature of
are described by
respectively. Now, let
and be an orthonormal frame on
N. We describe
-statistical sectional curvature of
for
, as
Shortly, we denote , , , , and , respectively by L, , , , and . Similarly, , and are denoted by , and , respectively.
Proposition 10. If is a statistical submanifold in a statistical manifold and , then for the statistical submanifold we getfor any . Proof. From
, (
73) and using the equations of Gauss in Theorem 1, we obtain the assertion. □
Lemma 7. Let be an m-dimensional statistical submanifold in an -dimensional statistical manifold . If , the statistical scalar curvature of satisfies the followingwhere is the statistical sectional curvature of . Proof. Proposition 10 and (
18) lead to
for any
. Considering
and
as orthonormal tangent and normal frames, respectively on
N and setting
and
in the above equation and summing over
, provides
Setting (
31) and (
32) in (
75), it follows that
On the other hand, setting
in the last equation we get
Hence the assertion is obtained from the above two relations. □
Proof. Applying Lemma 4 in the above equation, the assertion follows. □
From Lemma 7 and the last equation, we have the following corollary:
Lemma 8. The tensor field satisfy in the following Proof. As
, one can see that
which gives
□
Proof. Considering
instead of
in Lemma 8, it follows
Applying (
29) in the last relation, we get the assertion. □
Theorem 5. Let be an m-dimensional statistical submanifold in an -dimensional statistical manifold . Considering , for , , we havewhere is an orthonormal basis of . Proof. The above equation and Lemma 9 imply
Setting (
33) in the last relation, it follows that
Using (
76) and Lemma 7, we get
which gives us
□
Proof. Applying (
34) in Theorem 5, the assertion follows. □