Abstract
We continue the study of compound renewal processes (c.r.p.) under Cramér’s moment condition initiated in [2, 3, 6, 7, 8, 4, 5, 9, 10, 12, 16, 13, 14, 15]. We examine two types of arithmetic multidimensional c.r.p. \({\mathbf {Z}}(n)\) and \({\mathbf {Y}}(n) \), for which the random vector \({\xi }=({\tau },{\boldsymbol {\zeta }})\) controlling these processes ( \(\tau >0 \) defines the distance between jumps, \(\boldsymbol {\zeta }\) defines the value of jumps of the c.r.p.) has an arithmetic distribution and satisfies Cramér’s moment condition. For these processes, we find the exact asymptotics in the local limit theorems for the probabilities
in the Cramér zone of deviations for \({\boldsymbol {x}}\in {\mathbb Z}^d \) (in [9, 10, 13, 14, 15], the analogous problem was solved for nonlattice c.r.p., where the vector \( {\boldsymbol {\xi }}=(\tau ,{\boldsymbol {\zeta }}) \) has a nonlattice distribution).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1. INTRODUCTION. STATEMENT OF THE PROBLEM
We will denote the elements of the \(d\)-dimensional Euclidean space \( {\mathbb R}^d\) by boldface symbols, for example, \({\boldsymbol {x}}=(x_{(1)},\dots ,x_{(d)})\), \({\boldsymbol {0}}=(0,\dots ,0)\), and \({\boldsymbol {\mu }}=({\mu }_{(1)},\dots ,{\mu }_{(d)})\). The scalar product of elements \({\boldsymbol {x}},{\boldsymbol {y}}\in {\mathbb R}^d \) will be designated as
The norm in \( {\mathbb R}^d\) will be denoted by \(|{\boldsymbol {x}}|:=\sqrt {{\boldsymbol {x}}{\boldsymbol {x}}} \). Random vectors with values in \({\mathbb R}^d \) will also be denoted by bold symbols, for example, \({\boldsymbol {\zeta }}=\big ({\zeta }_{(1)},\dots ,{\zeta }_{(d)}\big ) \). The notation
will be used for a random vector in \({\mathbb R}^{d+1} \).
Consider a sequence
of independent random vectors in \( {\mathbb R}^{d+1}\) such that \({\tau }_0=0 \), \({\boldsymbol {\zeta }}_0={\boldsymbol {0}} \), \({\tau }_1\ge 0\), \({\tau }_k>0 \) for \(k\ge 2 \). Suppose that the vectors \({\boldsymbol {\xi }}_k=({\tau }_k,{\boldsymbol {\zeta }}_k)\) for \(k\ge 2 \) have the same distribution as \({\boldsymbol {\xi }}=({\tau },{\boldsymbol {\zeta }})\). Put
and so \(T_0=0 \), \({\mathbf {Z}}_0={\boldsymbol {0}} \), \({\bf S}_0=(0,{\boldsymbol {0}}) \). Let \({\nu }(0)=\eta (0):=0 \). For \(t>0 \), we put
and so \(\eta (t)={\nu }(t)+1\) for \(t>0 \).
The first and second compound renewal processes (c.r.p.) \({\mathbf {Z}}(t)\), \({\mathbf {Y}}(t)\) are defined by the equalities (see, for example, [2, 11])
respectively. The standard commonly accepted model of a c.r.p. presumes that the time \({\tau }_1\) of the appearance of the first jump and the value \({\boldsymbol {\zeta }}_1 \) of this jump have a joint distribution that is in general different from the joint distribution of \(({\tau },{\boldsymbol {\zeta }})\) (see, for example, [1]). This is so, for instance, for c.r.p. with stationary increments.
If either \(({\tau }_1,{\boldsymbol {\zeta }}_1)\overset {d}{=}({\tau },{\boldsymbol {\zeta }})\) or \(({\tau }_1,{\boldsymbol {\zeta }}_1)\overset {d}{=}(0,{\boldsymbol {0}})\) then we will refer to the processes \({\mathbf {Z}}(t)\) and \({\mathbf {Y}}(t)\) as a homogeneous c.r.p. and otherwise as an inhomogeneous c.r.p.
Thus, the distributions of two random vectors \({\boldsymbol {\xi }}_1=({\tau }_1,{\boldsymbol {\zeta }}_1)\) and \({\boldsymbol {\xi }}=({\tau },{\boldsymbol {\zeta }})\)completely determine the distributions of the first c.r.p. \({\mathbf {Z}}(t)\), \(t\ge 0 \), and the second c.r.p. \({\mathbf {Y}}(t) \), \(t\ge 0\). At the event
we have the equalities \({\nu }(t)=k \), \(\eta (t)=k+1\), \({\mathbf {Z}}(t)={\mathbf {Z}}_k\), \({\mathbf {Y}}(t)={\mathbf {Z}}_{k+1}\). Consequently, the step processes \({\nu }(t) \), \({\mathbf {Z}}(t)\), \(\eta (t) \), and \({\mathbf {Y}}(t) \) are left continuous for \(t>0 \).
It follows from [16] that limit theorems for the first c.r.p. \({\mathbf {Z}}(t)\) and the second c.r.p. \( {\mathbf {Y}}(t)\) may substantially differ.
As in [9, 11, 12], in the present article we assume that the random vectors \( {\boldsymbol {\xi }}_1=({\tau }_1,{\boldsymbol {\zeta }}_1) \) and \({\boldsymbol {\xi }}=({\tau },{\boldsymbol {\zeta }})\) defining the c.r.p. \({\mathbf {Z}}(t)\) and \({\mathbf {Y}}(t) \) satisfy Cramér’s moment condition in the following form:
- \([\mathbf {C}_0] \) :
-
\(\enspace \enspace {\mathbb {E}} e^{{\delta }|{\boldsymbol {\xi }}_1|}<\infty , \enspace \enspace {\mathbb {E}} e^{{\delta }|{\boldsymbol {\xi }}|}<\infty \enspace \enspace \textit {for some} \enspace \enspace {\delta }>0 \).
The papers [9, 10, 13, 14, 15] in particular contain integro-local theorems for the processes \({\mathbf {Z}}(t) \) and \({\mathbf {Y}}(t) \) as \(t\to \infty \) in the case where the random vector \({\boldsymbol {\xi }}=({\tau },{\boldsymbol {\zeta }})\) “defining” these processes is nonlattice. In the present article (and in the one-dimensional case for \({\mathbf {Z}}(t)\) in [12]), the analogous problem is solved for arithmetic c.r.p. \({\mathbf {Z}}(t)\) and \({\mathbf {Y}}(t)\) (and their right continuous versions), i.e., in the case where the vectors \({\boldsymbol {\xi }}_1 \), \(\boldsymbol {\xi }\) lie on the integer lattice \({\mathbb Z}^{d+1}\). More exactly, we will assume that \({\mathbb {P}}\left ({\boldsymbol {\xi }}_1\in {\mathbb Z}^{d+1}\right )=1\), and the vector \(\boldsymbol {\xi }\) satisfies the following stronger arithmeticity condition, which is formulated in terms of the characteristic function of this vector (see [2, p. 69]):
where \({\boldsymbol {u}}{\boldsymbol {\xi }} \) is the scalar product of the vectors \(\boldsymbol {u} \) and \(\boldsymbol {\xi } \) in \({\mathbb R}^{d+1} \).
- \([\mathbf {Z}] \) :
-
(Arithmeticity condition) Every \( {\boldsymbol {u}}\in {\mathbb Z}^{d+1} \) satisfies the equality \( f(2\pi {\boldsymbol {u}})=1\), and for every \({\boldsymbol {u}}\in {\mathbb R}^{d+1}\setminus {\mathbb Z}^{d+1}\), the inequality \(\big \vert f(2\pi {\boldsymbol {u}})\big \vert <1 \) is valid.
We note here that if the random vector \({\boldsymbol {\xi }}=({\tau },{\boldsymbol {\zeta }})\) satisfies condition \([\mathbf {Z}] \) then this vector cannot be degenerate in \({\mathbb R}^{d+1}\), i.e., it cannot lie with probability 1 on the plane
defined by a unit normal \({\bf n}=(n_1,{\bf n}_2) \) and a constant \(c \). Indeed, if \({\mathbb {P}}\left ({\boldsymbol {\xi }}\in L_{{\bf n},c}\right )=1 \) then we can choose real \(r \) such that \({\boldsymbol {u}}=(u_1,{\boldsymbol {u}}_2):=(rn_1,r{\bf n}_2)\notin {\mathbb Z}^{d+1} \). Then for this \({\boldsymbol {u}}\notin {\mathbb Z}^{d+1}\) we have the equality
which is impossible in view of condition \([\mathbf {Z}] \).
Since the arithmetic c.r.p. \({\mathbf {Z}}(t)\) and \({\mathbf {Y}}(t)\) are changed only at the integer time moments \( t=1,2,\dots \) and are left continuous, it follows that, for every noninteger argument \(t>0\), we have the equalities
where, as usual, \([t] \) stands for the integer part of a nonnegative number \(t \). Therefore, we will consider only the integer values of argument \( t=n\in \left \{0,1,2,\dots \right \}\), i.e., we will study the random sequences
This somewhat simplifies the notation and does not diminish generality.
One can also consider right continuous versions of the processes \({\mathbf {Z}}(t) \) and \({\mathbf {Y}}(t) \) by setting
In other words, for defining the right continuous versions of the processes \( {\mathbf {Z}}(t)\) and \({\mathbf {Y}}(t) \), it suffices to consider the right continuous functionals
and for \(t>0 \), put
Thus, alongside the sequences in (1.4), one can study the sequences
However, studying the sequences (1.6) easily amounts to studying the sequences (1.4) since (see definitions (1.2) and (1.5))
and hence,
In the present article, we obtain local theorems for the multidimensional arithmetic c.r.p. \( {\mathbf {Z}}(n)\), \({\mathbf {Y}}(n) \), in which we find the exact asymptotics for the probabilities
for sequences \({\boldsymbol {x}}={\boldsymbol {x}}_n\in {\mathbb Z}^d\) such that the point \({\boldsymbol {\alpha }}:={\boldsymbol {x}}/n\) lies in some fixed compact set \(K\subset {\mathbb R}^d \) (see Theorems 2.1 and 2.2 below) or the point \(\boldsymbol {\alpha }\) converges to some fixed \({\boldsymbol {\alpha }}_0\) (see Theorems 2.1A and 2.2A below). By (1.7), we also find the exact asymptotics also for the probabilities (see Corollaries 2.1 and 2.2 below)
Earlier in [12], the analogous problem was solved for the one-dimensional (\(d=1\)) arithmetic c.r.p. \({\mathbf {Z}}(n)\). The proof of the local theorems for arithmetic c.r.p. is based on the local limit theorem for the renewal function
in which the exact asymptotics of \( H(t,{\boldsymbol {x}})\) is investigated for the sequences \( (t,{\boldsymbol {x}})=(t_n,{\boldsymbol {x}}_n)\in {\mathbb Z}^{d+1} \) such that the points \(({\theta },{\boldsymbol {\alpha }}):=\frac {1}{n}(t,{\boldsymbol {x}}) \) lie in some compact set \(K\subset {\mathbb R}^{d+1} \) (Theorem 2.3).
Historical reviews of related results can be found in [2, 9]. As regards the methods, the present article follows [9, 10, 13, 14, 15] where, under Cramér’s moment condition \([\mathbf {C}_0] \), integro-local theorems are obtained for the nonlattice c.r.p. \({\mathbf {Z}}(t) \), \({\mathbf {Y}}(t) \) and the corresponding renewal measure \(H(B) \) in the one- and multidimensional cases, respectively, and also follows [12] where analogous results are obtained in the one-dimensional case under conditions \([\mathbf {C}_0] \) and \([\mathbf {Z}] \). Therefore, Theorems 2.1–2.3 of the present article supplement the results of [9, 10, 13, 14, 15, 12].
The remaining part of the article consists of Sections 2 and 3. In Section 2, we formulate the main results of the article. The statements involve a number of “basic” functions whose sense and properties are desirable to know for understanding the nature of the established laws. Therefore, in Section 2, we give definitions and some properties of these functions. Their detailed description can be found in [9] (the one-dimensional case) and in [13] (the multidimensional case). The proof of main assertions of the article are exposed in Section 3.
2. STATEMENT OF THE MAIN ASSERTIONS
2.1. The definitions and properties of the necessary functions
In this section, we briefly give the definitions of the functions that will play a defining in local limit theorems.
For \((\lambda ,{\boldsymbol {\mu }})\in {\mathbb R}^{d+1} \), we put
Clearly, in accordance with condition \([\mathbf {C}_0]\), the interiors \((\mathcal {A})\) and \((\mathcal {A}_1)\) of the sets \(\mathcal {A} \) and \(\mathcal {A}_1 \) contain the point \(({\lambda },{\boldsymbol {\mu }})=(0,{\boldsymbol {0}})\) and are the domains of analyticity of the functions \(\psi (\lambda ,{\boldsymbol {\mu }}) \) and \(\psi _1(\lambda ,{\boldsymbol {\mu }})\), respectively.
A key role in the description of the asymptotics of the renewal function (see (1.8)) corresponding to the random walk \(\left \{\mathbf {S}_n\right \}_{n\geq 0}\) is played by the so called second deviation function of the arguments \( \theta \geq 0\) and \({\boldsymbol {\alpha }}\in \mathbb {R}^d\)
where
\(\partial \mathcal {A} \) is the boundary of \(\mathcal {A} \). The properties of the function \(D(\theta ,{\boldsymbol {\alpha }})\) are studied rather completely (see, for example, [2, Sec. 2.9; 16]). It is convex, lower semicontinuous, semiadditive, and linear along any ray starting from the point \( ({\theta }_0,{\boldsymbol {\alpha }}_0)=(0,{\boldsymbol {0}}) \).
Note that, by the linearity of the function \(D(\theta ,{\boldsymbol {\alpha }})\) along any ray starting from any point \( (0,{\boldsymbol {0}})\), for \({\theta }>0 \) we have the equality
and hence the function of two variables \(D({\theta },{\boldsymbol {\alpha }})\) is completely determined by the values of the function of one variable
Moreover, the main results of the present article will be formulated in terms of the function \(D({\boldsymbol {\alpha }}) \). By (2.1),
where
is the so-calledbasic function for the c.r.p. It was proved in [16] that the function \(A({\boldsymbol {\mu }}) \) is convex and lower semicontinuous, and hence we have the equality (see [16])
Let \(\big (\lambda ({\boldsymbol {\alpha }}),{\boldsymbol {\mu }}({\boldsymbol {\alpha }})\big )\) be any point at which the supremum
is attained if such a point exists. It was shown in [9] (in the one-dimensional case) and [13] (in the multidimensional case) that the functions \( D({\boldsymbol {\alpha }})\) and \(D(\theta ,{\boldsymbol {\alpha }})\) are analytic in the sets
respectively. Moreover, for \({\boldsymbol {\alpha }}\in \mathfrak {A}\), the point \(\big (\lambda ({\boldsymbol {\alpha }}),{\boldsymbol {\mu }}({\boldsymbol {\alpha }})\big ) \) is unique and
2.2. Local theorems for arithmetic compound renewal processes
First, consider the first c.r.p. The asymptotics of the probability
will be studied in the normalized deviation zone
and it is natural to call this zone theCramér zone by analogy with the domains of analyticity arising in the classical theorems for random walks. However, in contrast to these classical theorems, it is not always possible to obtain asymptotics (2.6) in the whole zone (2.7). It has to be restricted in a number of cases. If \({\lambda }_+>D({\boldsymbol {0}})\), where
then no restriction is required.
Consider the case of \({\lambda }_+\le D({\boldsymbol {0}}) \). In this case, the forbidden part of the zone \(\mathfrak {A} \) is the closed set
Thus, in Theorem 2.1, we will study the probability asymptotics (2.6) in the regular deviation zone
and in Theorem 2.1*, in zone (2.7).
Put
where the positive function \(C_H({\theta },{\boldsymbol {\alpha }})\) continuous in the cone \(\mathcal {D} \) is defined by formula (3.6) below. It is easy to note that if \({\boldsymbol {\alpha }}\in (\mathfrak {B}_Z)\) then
Let us now formulate the local theorem for the process \({\mathbf {Z}}(n)\).
Theorem 2.1 \(. \) Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled, a compact set
is fixed, and the admissible inhomogeneity conditions
hold. Then, for \( {\boldsymbol {\alpha }}:={{\boldsymbol {x}}}/{n}\in K \) , \( {\boldsymbol {x}}\in {\mathbb Z}^d\) , there is valid the representation
in which the remainder \(o(1)={\varepsilon }_n({\boldsymbol {x}}) \) satisfies the relation
Remark 2.1\(. \) Suppose that the compact set \(K \) in Theorem 2.1 has the form \(K=[{\boldsymbol {a}}]_{\varepsilon }\) for sufficiently small \({\varepsilon }>0\). Then the admissible inhomogeneity conditions (2.9) of Theorem 2.1 are fulfilled automatically.
In the one-dimensional case \(d=1\), assertion (2.10) in Theorem 2.1 was proved in [12]. Theorem 2.1 agrees with Theorem 1.1 in [9] and Theorem 3.1 in [13] and supplements these theorems in the arithmetic multidimensional case (as we already said, in [9] and [13], integro-local theorems were obtained for the nonlattice case in the one- and multidimensional cases, respectively). As was observed in [9] and [13], the form of the sum \(I_Z({\boldsymbol {\alpha }}) \) to some extent explains the substantiality of the presence of the forbidden set \(\mathfrak {B}_Z\): for \({\boldsymbol {\alpha }}\in (\mathfrak {B}_Z)\) (or, which is the same, \( {\lambda }({\boldsymbol {\alpha }})>{\lambda }_+ \)) the sum \(I_Z({\boldsymbol {\alpha }}) \) diverges (see (2.8)) and the asymptotics of the sequence \({\mathbb {P}}\big (\thinspace {\mathbf {Z}}(n)={\boldsymbol {x}}\thinspace \big ) \) is substantially different. The substantiality of condition (2.9) explains the presence of the factor \(\psi _1\big ({\lambda }({\boldsymbol {\alpha }}),{\boldsymbol {\mu }}({\boldsymbol {\alpha }})\big )\) on the right-hand side of (2.10).
Corollary 2.1\(. \) Under the conditions of 2.1, representation (2.10) is preserved if we replace \( {\mathbf {Z}}(n)\) by \( {\mathbf {Z}}_+(n)\) and \( I_Z({\boldsymbol {\alpha }})\) by \(I_{Z+}({\boldsymbol {\alpha }}){:=}e^{-{\lambda }({\boldsymbol {\alpha }})}I_Z({\boldsymbol {\alpha }}) \) therein.
Denote by
the defect up to the level \(n \) and overshoot through the level \(n\) of the random walk \(\left \{T_k\right \}\) respectively. Thus, the summand \({\tau }_{{\nu }(n)+1}\) “covering” the level \(n \) is representable as the sum \({\tau }_{{\nu }(n)+1}={\gamma }(n)+\chi (n)\).
A. A. Borovkov focused the authors’ attention on the following fact: If, in the local theorem for the c.r.p. \({\mathbf {Z}}(n)\), the defect \( {\gamma }(n)\) is fixed (or appropriately bounded) then the deviation zone can be substantially extended by replacing the condition \(K\subset \mathfrak {A}\setminus \mathfrak {B}_Z\) with the condition \(K\subset \mathfrak {A} \). In other words, we have
Theorem 2.1*\(. \) Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled, a compact set
is fixed, and the admissible inhomogeneity conditions (2.9) hold.
Then, for any fixed integers \( u\ge 1\), \( v\ge 0\), every sequence \( {\boldsymbol {x}}={\boldsymbol {x}}_n\in {\mathbb Z}^d \) with \( {\boldsymbol {\alpha }}:={{\boldsymbol {x}}}/{n}\in K \) admits the representations
where
and the remainders \(o_u(1)={\varepsilon }_n({\boldsymbol {x}})\) and \(o_{u,v}(1)={\varepsilon }_n({\boldsymbol {x}}) \) enjoy (2.11).
Theorems 2.1 and 2.1* will be proved in Subsection 3.3.
Let us now pass to formulating the local theorem for the second c.r.p. \({\mathbf {Y}}(n)\) and its right-continuous version \({\mathbf {Y}}_+(n)\). Put
By condition \( [\mathbf {C}_0]\), the interior of \(({\cal M}) \) contains the point \({\boldsymbol {\mu }}={\boldsymbol {0}}\). For the processes \({\mathbf {Y}}(n)\) and \({\mathbf {Y}}_+(n) \), the forbidden set is
Put
As follows from Lemma 3.6 (see below), if \({\boldsymbol {\alpha }}\in \mathfrak {A}\setminus \mathfrak {B}_Y \) then \( I_Y({\boldsymbol {\alpha }})<\infty \).
Conversely, if \({\boldsymbol {\alpha }}\in (\mathfrak {B}_Y) \) then
and, like in the case of process \(\mathbf {Z}(n)\), the probability asymptotics (2.15) in the domain \((\mathfrak {B}_Y)\) is substantially different.
Theorem 2.2\(. \) Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled and a compact set \(K\) such that
is fixed. Suppose that the initial random vector \({\boldsymbol {\xi }}_1=({\tau }_1,{\boldsymbol {\zeta }}_1)\) satisfies the admissible inhomogeneity condition
Then the following relation holds for \({\boldsymbol {\alpha }}:={{\boldsymbol {x}}}/{n}\in K\), \({\boldsymbol {x}}\in {\mathbb Z}^d \) as \( n\to \infty \):
Here the remainder \(o(1)\thinspace {=}\thinspace {\varepsilon }_n({\boldsymbol {x}})\) satisfies (2.11).
Remark 2.2\(. \) Suppose that the compact set in Theorem 2.2 has the form \( K=[{\boldsymbol {a}}]_{\varepsilon }\) for sufficiently small \({\varepsilon }>0\). Then the admissible inhomogeneity condition (2.14) of Theorem 2.2 is fulfilled automatically.
Corollary 2.2\(. \) Under the conditions of Theorem 2.2, representation (2.15) is preserved if we replace \( {\mathbf {Y}}(n)\) by \( {\mathbf {Y}}_+(n)\) and \( I_Y({\boldsymbol {\alpha }})\) by \(I_{Y+}({\boldsymbol {\alpha }}):=e^{-{\lambda }({\boldsymbol {\alpha }})}I_Y({\boldsymbol {\alpha }}) \) therein.
If we fix the overshoot \(\chi (n)\) in the local theorem for the c.r.p. \({\mathbf {Y}}(n)\) then we can substatially simplify the deviation zone by replacing the condition \(K\subset \mathfrak {A}\setminus \mathfrak {B}_Y\) (3.21) by the condition \(K\subset \mathfrak {A} \). In other words, we have
Theorem 2.2*\(. \) Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled, a compact set \(K\) is fixed such that
and the admissible inhomogeneity condition (2.14) is fulfilled.
Then, for any fixed integer \( v\ge 0\), any sequence \( {\boldsymbol {x}}={\boldsymbol {x}}_n\in {\mathbb Z}^d \) such that \( {\boldsymbol {\alpha }}:=\frac {{\boldsymbol {x}}}{n}\in K \) admits the representation
where
and the remainder \(o_w(1)={\varepsilon }_n({\boldsymbol {x}})\) satisfies (2.11).
Theorems 2.2 and 2.2* will be proved in Subsection 3.4.
Note in passing that the proofs of Theorems 2.1 and 2.2 differ substantially at the key points: In the first proof, we must fight against a “thick right tail” of the random variable \(\tau \), and in the second proof, with “thick tails” of the random vector \(\boldsymbol {\zeta } \). Therefore we cannot manage with Theorem 2.1 only for one version of the compound renewal process obtaining the desired assertion for the second version as an easy consequence.
Let us now turn to large and normal deviations for c.r.p.
For \({\boldsymbol {\alpha }}\in {\mathbb R}^d \), \({\varepsilon }>0 \) denote by
the closed ball of radius \(\varepsilon \) centered at \(\boldsymbol {\alpha } \). Put
Now, consider the local theorems for the processes \({\mathbf {Z}}(n) \) and \({\mathbf {Y}}(n) \) in the domains of moderate large and normal deviations. Put
where
is the covariance matrix of the random vector \({\boldsymbol {\zeta }}-{\boldsymbol {a}}{\tau }\) and \(\top \) is the transposition sign.
Lemma 2.1 \(. \) Suppose the fulfillment of condition \([\mathbf {C}_0] \) . Then
Lemma 2.1 will be proved in Subsection 3.5. With account taken of 2.1 and 2.2, Theorems 2.1 and 2.2 and Lemma 2.1 imply
Corollary 2.3 \(. \) Let conditions \( [\mathbf {C}_0]\) and \( [\mathbf {Z}]\) be fulfilled. Then, for every sequence \({\boldsymbol {x}}={\boldsymbol {x}}_n\in {\mathbb Z}^d \) such that \( {\boldsymbol {\alpha }}:=\frac {{\boldsymbol {x}}}{n}\rightarrow {\boldsymbol {a}} \) as \( n\rightarrow \infty \) , there are valid the relations
In particular, if \( {\boldsymbol {\alpha }}-{\boldsymbol {a}}=o\left (\frac 1{n^{1/3}}\right ) \) as \( n\to \infty \) then for \( {\boldsymbol {y}}:={\boldsymbol {x}}-n{\boldsymbol {a}} \) we have
which implies that
Thus, the local theorems for the c.r.p. \({\mathbf {Z}}(n) \) and \({\mathbf {Y}}(n) \) in the domains of large and normal deviations coincide.
2.3. The local theorem for the renewal function
In the arithmetic case, a substantial role for proving Theorems 2.1, and 2.2 is played by the local limit Theorem 2.3 for the renewal function (see (1.8)) in which the asymptotics of \(H(t,{\boldsymbol {x}}) \) (as \(n\to \infty \)) for a sequence \((t,{\boldsymbol {x}})=(t_n,{\boldsymbol {x}}_n)\in {\mathbb Z}^{d+1} \) in the case where the point \(({\theta },{\boldsymbol {\alpha }}):=\frac 1n(t,{\boldsymbol {x}})\) lies in some fixed compact set \(K\) included in the analyticity domain \(\mathcal {D}\) of the function \( D({\theta },{\boldsymbol {\alpha }})\).
Theorem 2.3\(. \) Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled and, for a fixed compact set \( K\subset \mathcal {D}, \) the admissible inhomogeneity condition
is fulfilled. Then for \( (t,{\boldsymbol {x}})\in {\mathbb Z}^{d+1} \), \( ({\theta },{\boldsymbol {\alpha }}):=\frac 1n(t,{\boldsymbol {x}}) \) we have the representation
in which the remainder \(o(1)={\varepsilon }_n(t,{\boldsymbol {x}})\) satisfies the relation
and the function \( C_{H}(\theta ,{\boldsymbol {\alpha }}) \) is defined by (3.6).
In the one-dimensional case \(d=1\), Theorem 2.3 was proved in [12]; it supplements Theorem 4.1 in [10] and Theorem 4.1 in [13] where integro-local theorems were established for the renewal function in the nonlattice one- and multidimensional cases, respectively.
3. PROOFS OF THE MAIN ASSERTIONS
3.1. The first deviation function. The local theorem for a one-dimensional random walk.
In the proof of the assertions of the present article (as in [9, 10, 13]), we use the known integro-local theorems for the sums \({\bf S}_n=(T_n,{\mathbf {Z}}_n)\) (see, for example, [2, Sec. 2.9] or Theorem 3.1 below). An important role here is played by the (first) deviation function
corresponding to the random vector \({\boldsymbol {\xi }}=({\tau },{\boldsymbol {\zeta }})\). This is the Legendre transform of the convex lower semicontinuous function \(A({\lambda },{\boldsymbol {\mu }})\); therefore the function \(\Lambda ({\theta },{\boldsymbol {\alpha }})\) is also convex and lower semicontinuous.
Agree on the following notations. Given a function \(F(u,{\boldsymbol {v}}) \) of two variables \(u\in {\mathbb R} \) and \({\boldsymbol {v}}\in {\mathbb R}^d\), in what follows, the subscripts \((1) \) and \((2) \) will designate the derivatives with respect to the first and second arguments respectively (the gradient); for example:
Denote by \(F^{\prime }=F^{\prime }(u,{\boldsymbol {v}})\) and \(F^{\prime {}\prime }=F^{\prime {}\prime }(u,{\boldsymbol {v}})\) the vector
of the first derivatives and the matrix
of order \(d+1 \) of the second derivatives respectively. Designate as \(|F^{\prime {}\prime }|\) the determinant of the matrix \(F^{\prime {}\prime }\).
Alongside the sets \(\mathcal {A}_1\) and \(\mathcal {A} \), we will need the analyticity domain \(\mathcal {L} \) of the function \(\Lambda ({\theta },{\boldsymbol {\alpha }})\). This domain contains the points \(({\theta },{\boldsymbol {\alpha }})\in {\mathbb R}^{d+1} \) for which the system of equations
for the coordinates of the point \(({\lambda },{\boldsymbol {\mu }}) \) (where the supremum in (3.1) is attained) has a solution
belonging to \((\mathcal {A}) \), and so
Since the function \(A({\lambda },{\boldsymbol {\mu }})\) is strictly convex in \((\mathcal {A}) \), such a solution is always unique. What has been said means that the conditions
are equivalent. Moreover, the mappings
are mutually inverse; they perform a one-to-one correspondence between the domains \(\mathcal {A} \) and \(\mathcal {L} \). Clearly, \(({\theta },{\boldsymbol {\alpha }})=(a_{\tau },{\boldsymbol {a}}_{\boldsymbol {\zeta }}) \), where \(a_{\tau }:={\mathbb {E}}{\tau }\), \({\boldsymbol {a}}_{\boldsymbol {\zeta }}:={\mathbb {E}}{\boldsymbol {\zeta }} \), always belongs to \(\mathcal {L} \); for it,
It is known (see, for example, [2, 9, 13]) that if \(({\theta },{\boldsymbol {\alpha }})\in \mathcal {L}\) then
where the pair of functions \({\lambda }({\theta },{\boldsymbol {\alpha }})\), \({\boldsymbol {\mu }}({\theta },{\boldsymbol {\alpha }})\) is the only solution to (3.2).
Theorem 2.3.2 in [2, p. 72] implies the following version of the local theorem for the sums \({\bf S}_n \) of random vectors in the arithmetic case:
Theorem 3.1 \(. \) For \( n\ge 1\) , let
be the sum of independent identically distributed random vectors. Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled, a compact set \(K\subset \mathcal {L} \) is fixed, and the point \( (t,{\boldsymbol {x}})\in {\mathbb Z}^{d+1} \) is such that \( ({\gamma },{\boldsymbol {\beta }}):=({t}/{n},{{\boldsymbol {x}}}/{n})\in K \) . Then, for
as \( n\to \infty \) , we have
where the remainder \( o(1)={\varepsilon }_n(t,{\boldsymbol {x}}) \) is uniform over \( ({\gamma },{\boldsymbol {\beta }})\in K \) ,
3.2. Proof of Theorem 2.3
First we expose the form of the local theorem for the renewal function equivalent to Theorem 2.3 (which is more convenient for proving).
Theorem 2.3A\(. \) Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled and, for a fixed point
the admissible inhomogeneity condition
holds. Then any sequence
such that
admits representation (2.19) in which the remainder \( o(1)={\varepsilon }_n\) satisfies the relation
In [12], a proof is given of the equivalence of Theorems 2.3 and 2.3A in the one-dimensional case \(d=1 \). This proof remains valid for \(d\ge 1 \).
For \(({\theta },{\boldsymbol {\alpha }})\in \mathcal {D} \), we put for brevity
and so the vectors
are functions on one variable \({{\boldsymbol {\alpha }}}/{{\theta }}\).
For proving Theorem 2.3A, we also need
Lemma 3.1 [9, 10] \(. \) Suppose that condition \([\mathbf {C}_0] \) is fulfilled and \( ({\theta },{\boldsymbol {\alpha }})\in \mathcal {D} \).
The the following hold:
-
I.
The minimum of the function \(L(r)=L_{{\theta },{\boldsymbol {\alpha }}}(r):=r\Lambda ({{\theta }}/{r},{{\boldsymbol {\alpha }}}/{r}) \) is attained at the only point
$$ r_{{\theta },{\boldsymbol {\alpha }}}= \frac {{\theta }}{\widehat {{\theta }}}\enspace , \enspace \enspace \mbox {and also}\enspace \enspace r_{{\theta },{\boldsymbol {\alpha }}}\widehat {{\boldsymbol {\alpha }}}={\boldsymbol {\alpha }}.$$The functions \( \widehat {\lambda }\) and \( \widehat {{\boldsymbol {\mu }}}\) admit the representations
$$ \widehat {\lambda }=\lambda \bigg (\frac {{\theta }}{r_{{\theta },{\boldsymbol {\alpha }}}}, \frac {{\boldsymbol {\alpha }}}{r_{{\theta },{\boldsymbol {\alpha }}}} \bigg ), \quad \widehat {{\boldsymbol {\mu }}}={\boldsymbol {\mu }} \bigg (\frac {{\theta }}{r_{{\theta },{\boldsymbol {\alpha }}}}, \frac {{\boldsymbol {\alpha }}}{r_{{\theta },{\boldsymbol {\alpha }}}} \bigg ), $$where the functions \( \lambda (\boldsymbol {\cdot }\thinspace ,\boldsymbol {\cdot }) \) and \( {\boldsymbol {\mu }}(\boldsymbol {\cdot }\thinspace ,\boldsymbol {\cdot }) \) are a solution to system (3.2). Moreover,
$$ L^{\prime }(r_{{\theta },{\boldsymbol {\alpha }}})=0, \quad L^{\prime {}\prime }(r_{{\theta },{\boldsymbol {\alpha }}})=\frac 1{r_{{\theta },{\boldsymbol {\alpha }}}} \big (\widehat {{\theta }}, \widehat {{\boldsymbol {\alpha }}} \big )\widehat {\Lambda }^{\prime {}\prime } \big (\widehat {{\theta }}, \widehat {{\boldsymbol {\alpha }}} \big )^\top >0, $$where the functions \( \widehat {{\theta }}=\widehat {{\theta }}({\theta },{\boldsymbol {\alpha }}) \), \( \widehat {{\boldsymbol {\alpha }}}=\widehat {{\boldsymbol {\alpha }}}({\theta },{\boldsymbol {\alpha }})\) are defined in (3.4), (3.5),
$$ \widehat {\Lambda }^{\prime {}\prime }:=\Lambda ^{\prime {}\prime }({\theta },{\boldsymbol {\alpha }}) \bigr \vert _{({\theta },{\boldsymbol {\alpha }})=(\widehat {{\theta }},\widehat {{\boldsymbol {\alpha }}} )}.$$ -
II.
Given \({\varepsilon }>0 \) , there exists \( {\varepsilon }_1>0\) such that
$$ \min _{|r-r_{{\theta },{\boldsymbol {\alpha }}}|\thinspace \ge {\varepsilon }}L(r)\ge L(r_{{\theta },{\boldsymbol {\alpha }}})+{\varepsilon }_1.$$
For \(({\theta },{\boldsymbol {\alpha }})\in \mathcal {D} \), we put
and so
Proof of Theorem 2.3A. Basing on Theorem 3.1 and Lemma 3.1, carry out the proof in two steps.
I. Denote by \(H_0(B)\) the renewal measure for the case where the vectors \({\boldsymbol {\xi }}_1=({\tau }_1,{\boldsymbol {\zeta }}_1)\) and \({\boldsymbol {\xi }}=({\tau },{\boldsymbol {\zeta }})\) (i.e., all the summands \( {\boldsymbol {\xi }}_i\) are equal to \(({\tau }_i,{\boldsymbol {\zeta }}_i)\) for \(i\ge 1 \)) are identically distributed. At Stage I, prove Theorem 2.3A for the homogeneous case. It suffices to prove that, for any fixed \(({\theta }_0,{\boldsymbol {\alpha }}_0)\in \mathcal {D}\) and any sequence \( (t,{\boldsymbol {x}})\in {\mathbb Z}^{d+1}\) such that \( ({\theta },{\boldsymbol {\alpha }}):=\frac {1}{n}(t,{\boldsymbol {x}})\to ({\theta }_0,{\boldsymbol {\alpha }}_0)\) for \(n\to \infty \), we have the relation
which coincides with (2.19) for \(\psi _1=\psi \). The proof of (3.7) on the basis of Theorem 3.1 and Lemma 3.1 is a repetition of the proof of the integro-local theorem for the renewal function in the homogeneous case in [13] (see also the proof of the local theorem in the one-dimensional arithmetic case in [12]); therefore we omit it.
II. At this stage, represent the renewal measure in the inhomogeneous case as
and use the following assertion (cf. Lemma 4.2 in [10], which was established in proving Theorem 3.1 in [10]).
Lemma 3.2 \(. \) Suppose the fulfillment of the hypotheses of Theorem 2.3A. Then, for some \(c>0 \) , \( C<\infty \) , \( n_0<\infty \) and for all \(n\ge n_0\) , we have
Since the proof of Lemma 3.2 repeats verbatim the proof of Lemma 4.2 in [10] (see also the proofs of Lemma 4.1 in [13] and Lemma 3.2 in [12]), we omit it.
By (3.8), for proving assertion (2.19) in the general case, it suffices to find the asymptotics of the mean
Using the result of Stage I and (3.8), we have
Since in \(|{\boldsymbol {\xi }}_1|\le \ln ^2n \), as \(n\to \infty \) we have
we conclude that the right-hand side of (3.9) coincides with the right-hand side of (2.19).
3.3. Proofs of Theorems 2.1, 2.1*, and Corollary 2.1
We now give another (equivalent) form of Theorem 2.1, which is more convenient for the proof (but less convenient for application).
Theorem 2.1A\(. \) Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled and, for a fixed point
he admissible inhomogeneity conditions
hold. Then, for any sequence \( {\boldsymbol {x}}={\boldsymbol {x}}_n\in {\mathbb Z}^d \) such that
we have representation (2.10) in which the remainder \(o(1)={\varepsilon }_n({\boldsymbol {x}})\) satisfies the relation
We need the following assertion.
Lemma 3.3 \(. \) Suppose the fulfillment of the hypotheses of Theorem 2.1A. Then, for some \(c>0 \) , \( C<\infty \) , \( n_0<\infty \) and, for all \(n\ge n_0\) , we have the inequality
The proof of (3.13) in the one-dimensional case \(d=1\) is given in [12] (see the proof of Lemma 3.3 in [12]). Since the proof of inequality (3.13) in the general case \(d\ge 1 \) repeats the proof of Lemma 3.3 in [12] with obvious changes, and we omit it.
Put
and so,
Lemma 3.4 \(. \) Under the conditions of Theorem 2.1A, there exists \({\delta }_0>0 \) such that
Proof. The exponential Chebyshev inequality implies that
where
is the deviation function for the random variable \(\tau \) satisfying the relation
Since, under the conditions of Theorem 2.1A, \({\lambda }({\boldsymbol {\alpha }}_0)<{\lambda }_+ \), for some \({\delta }_0>0 \) and some \(C<\infty \) for all \(m\ge 0 \) we have
Proof of Theorem 2.1A. We have
Since the inner sum on the right-hand side of the last formula “folds” into the renewal function \(H(n-m,{\boldsymbol {x}}) \), we obtain the representation
Now, applying Theorem 2.3, from (3.16) we obtain the representation
where
Obviously, for every \(m\in \left \{1,2,\dots ,\ln ^2n\right \}\), as \(n\to \infty \) (see (3.14)), we have the convergence
and the inequalities
hold for any \( {\delta }>0\) and all sufficiently large \(n \). Therefore, using Lemma 3.4, by the dominated convergence theorem, we obtain the relation
The relations (3.16)–(3.18) and Lemma 3.3 imply Theorem 2.1A.
Proof of Theorem 2.1*. Repeating the deduction of (3.15) with obvious changes, we obtain
Applying further the local theorem 2.3 for the renewal function, we obtain the assertion (2.12) of Theorem 2.1*. Since (2.13) is proved similarly, Theorem 2.1* is proved.
Proof of Corollary 2.1. First, prove that
For
we have
By (2.5),
Therefore, continuing (3.20) and using (2.5), we obtain
Equality (3.19) is proved. For proving Corollary 2.1, it remains to make use Theorem 2.1, equality (3.19), and the relation
3.4. Proofs of Theorems 2.2, 2.2*, and Corollary 2.3
We now give an equivalent version of Theorem 2.2, which is more convenient for proving.
Theorem 2.2A\(. \) Suppose that conditions \([\mathbf {C}_0] \) and \( [\mathbf {Z}]\) are fulfilled and a point
is fixed. Suppose that the admissible inhomogeneity condition
is fulfilled for the initial random vector \({\boldsymbol {\xi }}_1=({\tau }_1,{\boldsymbol {\zeta }}_1)\). Then, for any sequence \({\boldsymbol {x}}={\boldsymbol {x}}_n\in {\mathbb Z}^d\) such that
representation (2.15) is valid, in which the remainder \(o(1)={\varepsilon }_n \) satisfies the relation
We will need the following assertion.
Lemma 3.5 \(. \) Suppose the fulfillment of the hypotheses of Theorem 2.2A. Then there are constants \(c>0 \) , \( C<\infty \) , \( n_0<\infty \) such that the equality
holds for all \( n\ge n_0\) , where
Since the proof of Lemma 3.5 is an almost verbatim repetition of the proof of Lemma 2.3 in [15], we omit it.
Lemma 3.6 \(. \) Suppose the fulfillment of the conditions of Theorem 2.2A. Then there exists \( {\delta }>0\) such that
Proof. If \({\lambda }({\boldsymbol {\alpha }}_0)\ge 0\) then
where the distribution of the random vector \(\big ({}\thinspace \widehat {{\tau }},{}\thinspace \widehat {{{\boldsymbol {\zeta }}}}\thinspace \big ) \) is defined as follows:
Under our conditions, we have \(\big ({\lambda }({\boldsymbol {\alpha }}_0),{\boldsymbol {\mu }}({\boldsymbol {\alpha }}_0)\big )\in (\mathcal {A})\); therefore condition \([\mathbf {C}_0] \) is fulfilled for the new vector \(\big ({}\thinspace \widehat {{\tau }},{}\thinspace \widehat {{{\boldsymbol {\zeta }}}}\big ) \). Use the Cauchy inequality: for sufficiently small \({\delta }>0 \) and sufficiently large \(C<\infty \), we have
Since the random variable \({}\thinspace \widehat {{\tau }}\) satisfies condition \([\mathbf {C}_0] \), there exist constants \(c>0 \) and \(C_1<\infty \) such that
for all \(m\ge 0\). We have obtained the estimate
in which the right-hand side is finite for \({\delta }\in (0,c)\).
Let now \({\lambda }({\boldsymbol {\alpha }}_0)<0 \). Then
and, for estimating the right-hand side of the last inequality, we can use Hölder’s inequality: For any \(p,q\in (0,1) \) such that \(p+q=1 \), we have
Since, under our conditions, we have \({\boldsymbol {\mu }}({\boldsymbol {\alpha }}_0)\in (\mathcal {M}) \), there exist \({\delta }>0 \), \(c>0 \), \({1}/{p}>1 \), and \(C \), \(C_1<\infty \) such that
Therefore, from (3.24) we deduce
Since the right-hand side of the last inequality is finite for \({\delta }\in (0,c)\), inequality (3.23) is proved.
Lemma 3.7 \(. \) Suppose the fulfillment of the hypotheses of Theorem 2.2A. Then
The proof of Lemma 3.7 repeats the proof of Lemma 4.1(b) in [15]; therefore we omit it.
Proof of Theorem 2.2A. Alongside with the set \( L_n^{d+1}\) defined by (3.22), we will use the set
We infer
Since the inner sum on the right-hand side of the last expression is “compiled” into the renewal function, we obtain the representation
Applying now Theorem 2.3, we obtain
where
It is easy to see that, as \(n\to \infty \),
and for any \({\delta }>0 \) and all sufficiently large \(n \)
Now, applying Lemma 3.6, by the dominated convergence theorem, we obtain
Using Lemma 3.5, 3.7 and relations (3.25), (3.26), we infer
Assertion (2.15) of Theorem 2.2A and Theorem 2.2A are proved.
Proof of Theorem 2.2*. It is easy to see that, for all integers \(v\thinspace {\ge }\thinspace 0 \),
Therefore, calculating the right-hand side of (3.27) with the use of Theorem 2.1* and using the relation
which follows from (3.19), we obtain (2.16).
Proof of Corollary 2.2 is completely similar to that of Corollary 2.1.
3.5. Proof of Lemma 2.1.
Establish (2.18). Recall that, for points \(\boldsymbol {\alpha }\) in some neighborhood of the point \({\boldsymbol {\alpha }}_0={\boldsymbol {a}}\), we have the relation
where the function \(A({\boldsymbol {\mu }}) \) is a solution to the equation
and the function \({\boldsymbol {\mu }}({\boldsymbol {\alpha }})\) is inverse to the function \( A^{\prime }({\boldsymbol {\mu }})\), i.e., satisfies the equation
Verify that
The equalities in (3.29) are obtained from equation (3.28) by differentiating it sufficiently many times:
-
(0)
for \( {\boldsymbol {\mu }}={\boldsymbol {0}}\), by \(\psi (0,{\boldsymbol {0}})=1\) we have \(A({\boldsymbol {0}})=0\);
-
(1)
differentiating equality (3.28) once for \({\boldsymbol {\mu }}={\boldsymbol {0}}\), we obtain
$$ \psi ^{\prime }_{(1)}(0,{\boldsymbol {0}}) \big ({-}A^{\prime }({\boldsymbol {0}}) \big )+\psi ^{\prime }_{(2)}(0,{\boldsymbol {0}})={\boldsymbol {0}}; $$therefore,
$$ A^{\prime }({\boldsymbol {0}})=\frac {\psi ^{\prime }_{(2)}(0,{\boldsymbol {0}})} {\psi ^{\prime }_{(1)}(0,{\boldsymbol {0}})}={\boldsymbol {a}};$$ -
(2)
differentiating equality (3.28) twice for \({\boldsymbol {\mu }}={\boldsymbol {0}}\), we obtain
$$ \psi ^{\prime {}\prime }_{(1,1)}(0,{\boldsymbol {0}}) \big ({-}A^{\prime }({\boldsymbol {0}}) \big ) \big ({-}A^{\prime }({\boldsymbol {0}}) \big )^\top -2\psi ^{\prime {}\prime }_{(1,2)}(0,{\boldsymbol {0}}) \big (A^{\prime }({\boldsymbol {0}}) \big )^\top +\psi ^{\prime {}\prime }_{(2,2)}(0,{\boldsymbol {0}})-\psi ^{\prime }_{(1)}(0,{\boldsymbol {0}})A^{\prime {}\prime }({\boldsymbol {0}})={\boldsymbol {0}}{\boldsymbol {0}}^\top ;$$therefore,
$$ A^{\prime {}\prime }({\boldsymbol {0}})=\frac {1}{\psi ^{\prime }_{(1)}(0,{\boldsymbol {0}})} \big [{\mathbb {E}}{\zeta }{\zeta }^\top -2{\boldsymbol {a}}{\mathbb {E}}{\boldsymbol {\zeta }}{\tau }+{\boldsymbol {a}}{\boldsymbol {a}}^\top {\mathbb {E}}{\tau }^2 \big ]=B^2.$$
The equalities (3.29) are proved. Since the functions \(D^{\prime }({\boldsymbol {\alpha }}) \) and \(A^{\prime }({\boldsymbol {\mu }})\) are mutually inverse, relation (3.29) implies the equalities
The equalities (2.18) are proved.
Prove (2.17). Since
it suffices to verify the equality
The law of large numbers for the process \({\mathbf {Z}}(n)\) and Theorem 2.1 imply the existence of a sequence \( {\varepsilon }_n>0\) tending to \(0 \) slowly enough, for which
where
Since it is obvious that, as \({\varepsilon }_n\sqrt {n}\to \infty \),
REFERENCES
S. Asmussen and H. Albrecher, Ruin Probabilities. 2nd ed. (World Scientific, Hackensack, NJ, 2010).
A. A. Borovkov, Asymptotic Analysis of Random Walks. Rapidly Decreasing Distributions of Increments (Fizmatlit, Moscow, 2013) [in Russian].
A. A. Borovkov, “Large deviation principles in boundary problems for compound renewal processes,” Sibirsk. Mat. Zh. 57, 562 (2016) [Siberian Math. J. 57, 442 (2016)].
A. A. Borovkov and A. A. Mogul’skiĭ, “The second rate function and the asymptotic problems of renewal and hitting the boundary for multidimensional random walks,” Sibirsk. Mat. Zh. 37, 745 (1996) [Siberian Math. J. 37, 647 (1996)].
A. A. Borovkov and A. A. Mogul’skiĭ, “Chebyshev-type exponential inequalities for sums of random vectors and for trajectories of random walks,” Teor. Veroyatnost. i Primenen. 56, 3 (2011) [Theory Probab. Appl., 56, 21 (2012)].
A. A. Borovkov and A. A. Mogul’skiĭ, “Large deviation principles for the finite-dimensional distributions of compound renewal processes,” Sibirsk. Mat. Zh.56, 36 (2015) [Siberian Math. J. 56, 28 (2015)].
A. A. Borovkov and A. A. Mogul’skiĭ, “Large deviation principles for trajectories of compound renewal processes. I,” Teor. Veroyatnost. i Primenen. 60, 227 (2015) [Theory Probab. Appl., 60, 207 (2016)].
A. A. Borovkov and A. A. Mogul’skiĭ, “Large deviation principles for trajectories of compound renewal processes. II,” Teor. Veroyatnost. i Primenen. 60, 417 (2015) [Theory Probab. Appl., 60, 349 (2016)].
A. A. Borovkov and A. A. Mogul’skiĭ, “Integro-local limit theorems for compound renewal processes under Cramér’s condition. I,” Sibirsk. Mat. Zh.59, 491 (2018) [Siberian Math. J. 59, 383 (2018)].
A. A. Borovkov and A. A. Mogul’skiĭ, “Integro-local limit theorems for compound renewal processes under Cramér’s condition. II,” Sibirsk. Mat. Zh.59, 736 (2018) [Siberian Math. J. 59, 578 (2018)].
D. R. Cox and W. L. Smith, Renewal Theory (Sovetskoe Radio, Moscow, 1967) [in Russian].
A. A. Mogul’skiĭ, “Local theorems for arithmetic compound renewal processes when Cramer’s condition holds,” Sib. Èlektron. Mat. Izv. 16, 21 (2019).
A. A. Mogul’skiĭ and E. I. Prokopenko, “Integro-local theorems for multidimensional compound renewal processes, when Cramer’s condition holds. I,” Sib. Èlektron. Mat. Izv. 15, 475 (2018).
A. A. Mogul’skiĭ and E. I. Prokopenko, “Integro-local theorems for multidimensional compound renewal processes, when Cramer’s condition holds. II,” Sib. Èlektron. Mat. Izv. 15, 503 (2018).
A. A. Mogul’skiĭ and E. I. Prokopenko, “Integro-local theorems for multidimensional compound renewal processes, when Cramer’s condition holds. III,” Sib. Èlektron. Mat. Izv. 15, 528 (2018).
A. A. Mogul’skiĭ and E. I. Prokopenko, “The rate function and the fundamental function for multidimensional compound renewal process,” Sib. Èlektron. Mat. Izv. 16, 1449 (2019).
ACKNOWLEDGMENTS
The authors are grateful to A. A. Borovkov for the statement of Theorem 2.1* and to the anonymous referee for very benevolent and qualified remarks.
Funding
The work was supported by the Russian Science Foundation (project 18-11-00129).
Author information
Authors and Affiliations
Corresponding authors
About this article
Cite this article
Mogul’skiĭ, A.A., Prokopenko, E.I. Local Theorems for Arithmetic Multidimensional Compound Renewal Processes under Cramér’s Condition. Sib. Adv. Math. 30, 284–302 (2020). https://doi.org/10.1134/S1055134420040033
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1134/S1055134420040033