Weak Solutions of Stochastic Differential Equations over the Field of p-Adic Numbers
Hiroshi Kaneko, Anatoly N. Kochubei

TL;DR
This paper advances the theory of stochastic differential equations over p-adic numbers by refining stochastic integral definitions, establishing continuity of local times, and providing conditions for weak solutions driven by p-adic stable processes.
Contribution
It introduces an improved stochastic integral definition on p-adic fields and derives new existence conditions for weak solutions of p-adic SDEs with measurable coefficients.
Findings
Established joint continuity of local time for p-adic stable processes
Provided sufficient conditions for weak solutions of p-adic SDEs
Refined the stochastic integral framework for p-adic stochastic calculus
Abstract
Study of stochastic differential equations on the field of p-adic numbers was initiated by the second author and has been developed by the first author, who proved several results for the p-adic case, similar to the theory of ordinary stochastic integral with respect to Levy processes on the Euclidean spaces. In this article, we present an improved definition of a stochastic integral on the field and prove the joint (time and space) continuity of the local time for p-adic stable processes. Then we use the method of random time change to obtain sufficient conditions for the existence of a weak solution of a stochastic differential equation on the field, driven by the p-adic stable process, with a Borel measurable coefficient.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
Topicsadvanced mathematical theories · Mathematical Biology Tumor Growth
11footnotetext: Partially supported by the Ukrainian State Fund for Fundamental Research, Grant 10.01/004
Weak Solutions of Stochastic Differential Equations
over the Field of -Adic Numbers
Hiroshi Kaneko and Anatoly N. Kochubei*** 2000 Mathematical Subject Classification. Primary 60H10; Secondaly 11S80, 60G52.
Abstract
Study of stochastic differential equations on the field of -adic numbers was initiated by the second author and has been developed by the first author, who proved several results for the -adic case, similar to the theory of ordinary stochastic integral with respect to Lévy processes on the Euclidean spaces. In this article, we present an improved definition of a stochastic integral on the field and prove the joint (time and space) continuity of the local time for -adic stable processes. Then we use the method of random time change to obtain sufficient conditions for the existence of a weak solution of a stochastic differential equation on the field, driven by the -adic stable process, with a Borel measurable coefficient.
1 Introduction
Stochastic processes on the field \mbox{\boldmathQ}_{p} of -adic numbers have been studied in many papers, for example, by Albeverio-Karwowski [1, 2], Albeverio-Zhao [3, 4, 5], Evans [11], Figà-Talamanca et al. [6, 9], Ismagilov [15], Kaneko [18, 19], Karwowski–Vilela Mendes [17], Kochubei [21, 22, 23, 24], Varadarajan [29] and Yasuda [32, 33, 34] (here we do not mention papers on processes with -adic time and other related subjects). In particular, these authors constructed and studied wide classes of Markov processes on \mbox{\boldmathQ}_{p} (most of the results can be extended easily to more general local fields). Their infinitesimal generators are usually hypersingular integral operators; the first and simplest example is Vladimirov’s fractional differentiation operator [30, 31] corresponding to the -adic -stable process.
Alike for the case of the Euclidean space, for the field of -adic numbers, stochastic processes and analysis are tightly related. Indeed, we can find descriptions and facts in [30, 31, 23] which are transplanted for describing probablisitic concepts on . In [22], the second author initiated the theory of stochastic differential equations on , and in [18] the first author made several assertions on stochastic integrals on which are similar to ordinary Itô calculus on the Euclidean space. On the other hand, Yasuda [32] developed potential-theoretic notions related to -adic Lévy processes, in particular the stable process. An analytic potential theory over was developed by Haran [12].
In [18, 22] we considered only strong solutions of stochastic differential equations on driven by the -adic -stable process. Note that in the conventional theory of stochastic differential equations (over ), the study of weak solutions of such equations began only recently [35, 36, 10, 25].
In this paper, we initiate a theory of weak solutions for the -adic case. Since we need to discuss stochastic differential equations without continuity of the coeffficient as assumed in [18], we first provide a more general construction of stochastic integrals than the one in the first author’s paper. Our improved definition of stochastic integrals admits predictable integrand satisfying a finiteness condition on its moment. Then, we follow the approach by Zanzotto [35] based on the method of random time change. In order to use this technique in our situation, we prove the joint (with respect to the space and time variables) continuity of the local time for the -adic -stable process, as well as some facts on the occupation time on mesurable sets from with positive Haar measure. After that, we apply the theory of the Lévy system of a Markov process on a general state space [28]. In fact, this is one of the first applications of the general theory of Markov processes outside the usual Euclidean realm. In Section 4, we give a sufficient condition for the existence of a weak solution of the stochastic differential equation
[TABLE]
on , with respect to the -stable process , with a locally bounded coefficient . The conditions are much less restritive than those guaranteeing the existence of strong solutions of such equations [18, 22].
2 Stochastic integral
In this section, we will establish the notion of stochastic integrals of predictable stochastic process with respect to the Lévy process with determined by the sequence satisfying
- (1)
,
- (2)
and
- (3)
(see [2, 23, 32] for the basic notions regarding Lévy processes on ). Then, we can choose a filtration satisfying for any so that is independent of for every .
In this section, the smallest -field on based on which all left-continuous -adapted processes are measurable will be denoted by . The -valued stochastic process defined on is said to be predictable if it is measurable with respect to the -field . As standard notations in the theory of Markov process, the starting point of will be indicated in the notations for the probability measure and the expectation as and respectively.
Lemma 1**.**
The linear space of the -valued bounded stochastic processes defined on satisfying the following conditions * *(i) and (ii):
- (i)
contains all bounded left-continuous -adapted processes,
- (ii)
for any sequence , and imply ,
contains all bounded predictable processes.
Proof Since any bounded predictable process is described as the limit of bounded -measurable simple functions, it suffices to prove for any . For the family , we can verify that
(i) ,
(ii) and imply ,
(iii) and imply .
For any finite set of bounded left-continuous -adapted processes and finite set of balls in , we easily see . This is because is bounded -adapted left continuous process for any .
Since the family of the sets described as with finite set of bounded left-continuous -adapted processes and finite set of balls in is closed under finite intersection and , we can derive from Dynkin’s theorem (e.g., Lemma 5.1 in [14]) that .
For arbitrarily fixed and , we denote by the family of -valued predictable processes each of which is adapted to and satisfies . Here, we introduce a subfamily of each element of which admits the following expression:
[TABLE]
with some sequence of random variables satisfying and some .
Proposition 1 is a dense subfamily of with respect to the norm
[TABLE]
Proof For any , the sequence of stochastic processes defined by satisfies . Therefore, we may assume that vanishes outside .
We define to be the set of such that for any and there exists a sequence satisfying . Then, the space is a linear space and for any sequence , and imply .
If is a left-continuous -adapted process satisfying for any , then
[TABLE]
is in and satisfies . Accordingly, Lemma 1 shows that contains all -adapted predictable processes enjoying for any .
For any element given by with some sequence of random variables satisfying and some , the stochastic integral with respect to is defined by
[TABLE]
The stochastic processes can be regarded as a right continuous -adapted process.
As in [18], we can show the following
Lemma 2**.**
There exists a positive constant satisfying the following conditions* :*
- (i)
E_{0}\big{[}\|Z(t)\|_{\scriptstyle p}^{{}^{\scriptstyle\gamma}}\big{]}\leq C_{A,\gamma}t* for all .*
- (ii)
E_{0}\Bigg{[}\displaystyle\sup_{0\leq t\leq u}\Big{\|}\int_{0}^{t}\phi(s)dZ(s)\Big{\|}_{\scriptstyle p}^{{}^{\scriptstyle\gamma}}\Bigg{]}\leq C_{A,\gamma}\int_{0}^{u}E_{0}\big{[}\|\phi(s)\|_{\scriptstyle p}^{{}^{\scriptstyle\gamma}}\big{]}ds* for *
For any element , there exists a sequence , such that . Then one can derive from Proposition 1 that
[TABLE]
Therefore, the stochastic integral of with respect to can be defined as a unique -valued process satisfying
[TABLE]
Let us denote by \mathcal{D}([0,T]\to\mbox{\boldmathQ_{p}}) the space of all right continuous sample paths on the time interval to with left limit at every point. Since we already know that is a \mathcal{D}([0,T]\to\mbox{\boldmathQ_{p}})-valued stochastic process, we immediately see that is a \mathcal{D}([0,T]\to\mbox{\boldmathQ_{p}})-valued variable as well.
Again as in [18], we will obtain wider perspectives of -adic stochastic integral so that it covers the one with respect to the -stable processes. Indeed, we can consider a random walk corresponding to a sequence satisfying (1), (2) and
- (4)
In this case, defining sequences with properties (1), (2) and (3) by
[TABLE]
we can prove the following proposition.
Proposition 2 For any Lévy process corresponding to with properties (1),(2) and (4), we have the following :
(i) The -valued process defined by Z(M;t)=\int_{\mbox{\boldmathQ_{p}}}\int_{0}^{t}f_{M}(z)\Lambda(ds,dz) is a random walk corresponding to , where stands for the Poisson random measure of and denotes the -valued function defined by
[TABLE]
(ii) There exists a sequence of events satisfying and
[TABLE]
for any and a.s. on for any of .
(iii) For any sequence enjoying for any and a.s. on ,
[TABLE]
for any and a.s. on
Proof The proof can be done as in the proof of Proposition 2 in [18].
Accordingly, if a -valued predictable process adapted to admits predictable processes from satisfying on for any and , we can define the stochastic integral with respect to a stochastic process determined by satisfying (1), (2) and (4). Indeed, the stochastic integral is defined as a unique right continuous -adapted process satisfying
[TABLE]
3 Local Time and Related Properties of -Adic Stable Processes
Let be the symmetric -stable process on with . This stochastic process is characterized as the Hunt process satisfying with transition probability densities of the form
- (5)
\displaystyle\qquad\qquad\qquad P(t,x-y)=\int_{\mbox{\boldmathQ_{p}}}\chi(-(x-y)\xi)e^{-t||\xi||_{p}^{\alpha}}\mu(d\xi),
where is the canonical additive character on and is a Haar measure normalized by the requirement that the measure of a unit ball equals 1. The process is given as the Lévy process determined by the sequence of the form
[TABLE]
Below will denote the probability law induced by on the trajectory space \mathcal{D}([0,\infty)\to\mbox{\boldmathQ_{p}}) under , and will be the corresponding expectation. When the probability measure is provided, the -stable process starting from will be denoted again by .
Here, we can recall that Yasuda’s result ([32]) shows that implies that every point x\in\mbox{\boldmathQ_{p}} is regular for . Accordingly, we have , where (see [32]). Moreover, the Hunt process satisfies Hunt’s conditions (A) and (F) in [13], and for any x,y\in\mbox{\boldmathQ_{p}} (see [32, 23]).
Consider the random Borel measure
[TABLE]
on , called the occupation time measure, where stands for the indicator of the Borel set . If this random Borel measure admis a density with respect to the Haar measure , that is
[TABLE]
for any Borel set , then is called the local time of the process .
It is known [7] that in our situation admits the local time , for which the function is jointly measurable in and continuous and monotone non-decreasing in for every with probability one. We note that is satisfied a.s. for any with probability one. Let us prove almost sure joint continuity of in .
Let be arbitrary positive numbers. It is known (see (V.3.28) in [8]) that for any x,a,b\in\mbox{\boldmathQ_{p}}
- (6)
where , and .
For the function
[TABLE]
with positive parameter , we can derive from (5) that
[TABLE]
Since , the function is bounded and continuous in , and (see Lemma 3.2 in [32]).
Due to the spatial homogeneity, we have In our case, is satisfied ( depends actually on as the Fourier transform of a radial function) and so we have
- (7)
Next, we have the following lemma for the function
[TABLE]
Lemma 3**.**
For any N\in\mbox{\boldmathN}, there exists a positive constant such that
- (8)
**
Proof. We have
[TABLE]
with some positive constant , where
[TABLE]
For any x\in\mbox{\boldmathQ_{p}} with , from an integration formula in [31], we can derive that
[TABLE]
As , the expression in the braces tends to
[TABLE]
Therefore, we obtain the inequality (8) for any in .
Theorem 1**.**
The function from (0,\infty)\times\mbox{\boldmathQ_{p}} to is continuous a.s. Moreover, for any with , any and M\in\mbox{\boldmathZ_{+}}, there exists a random variable such that
- (9)
**
for all in .
Proof. Let us fix a positive integer and consider the set of elements in of the form
[TABLE]
Let us first prove that, for any satisfying , Inequality (9) holds almost surely with some constant independent of . If and , then it follows from (6), (7) and (8) that
[TABLE]
for any , where .
The number of pairs of elements of satisfying equals . This is because the quantity of all elements of must be multiplied by the number of elements whose distance from a given element equals , and then divided by two, so that each pair to be counted only once. Since we have , we have . By the Borel-Cantelli lemma, we see
[TABLE]
except a finite number of values of with probability one. In other words, almost surely, there exists a positive integer such that
- (10)
for any and in satisfying with some integer .
Now, suppose that are arbitrary elements in satisfying with some integer . Let us take canonical representations
[TABLE]
[TABLE]
for and with the coefficients from enjoying . For the sequences and given by
[TABLE]
and
[TABLE]
respectively, we have
[TABLE]
Since and as , we obtain that and in probability (see Corollary V.3.29 in [8]). Passing to the limit, we find that
[TABLE]
Hence, by (10), we get the inequality
[TABLE]
for satisfying . This is equivalent to (9). It is clear that the inequality (9) implies the required joint continuity.
Let us consider the behavior of the occupation time measure as .
Theorem 2**.**
For every Borel set B\subset\mbox{\boldmathQ_{p}} of a positive Haar measure,
- (11)
**
Proof. Consider the transition probability semigroup given by
[TABLE]
It can be verified by a direct calculation that
[TABLE]
for the indicator function of any ball in . Thus this identity holds for any locally constant function with a compact support, and therefore for any f\in L^{2}(\mbox{\boldmathQ_{p}};\mu). This shows that the Haar measure is an invariant measure for the -stable process .
It is proved in [32] (see also [23]) that the -stable process with is recurrent, so that it hits any open set in after any large time (for these notions in the context of processes on general locally compact Abelian groups see [26]). Together with the existence of an invariant measure, this implies the Harris recurrence of (Proposition X.3.9 in [27]), so that every Borel set of a positive Haar measure is recurrent. Due to Propositions X.3.11 and X.2.2 in [27], for the proof of (11), it suffices to show that the functional
[TABLE]
defined on bounded positive Borel measurable functions on , does not vanish. This follows from
[TABLE]
and .
The above properties of the local time make it possible to prove an analogue of the Engelbert-Schmidt zero-one law for our situation.
Theorem 3**.**
Let be a non-negative Borel function on . The following properties are equivalent :
- (i)
P_{0}\big{\{}\int\limits_{0}^{t}f(Z(s))\,ds<\infty,\mbox{ for all }t\geq 0\big{\}}>0.
- (ii)
P_{0}\big{\{}\int\limits_{0}^{t}f(Z(s))\,ds<\infty,\mbox{ for all }t\geq 0\big{\}}=1.
- (iii)
* is locally integrable on .*
4 Time change and moment estimate
In this section, we will discuss a weak solution of the stochastic differential equation
- (12)
\displaystyle\left\{\begin{array}[]{l}dX(t)=b(X(t-))dZ(t),\\ X(0)=x,\end{array}\right.
by taking some -stable process on . We will focus on the case that coefficient in the right-hand side is given by a -valued Borel measurable function defined on .
Definition 1**.**
An -adapted stochastic process defined on a probability measure space is called a solution of the stochastic differential equation (12) if there exists an -adapted -stable process with satisfying **
[TABLE]
Definition 2**.**
A solution of equation (12) is said to be trivial, if **
[TABLE]
Definition 3**.**
For x\in\mbox{\boldmathQ_{p}}, we say the coefficient in (12) satisfies Condition (H) with respect to if**
[TABLE]
for every integer .**
Here, we introduce an increasing process associated with the -stable process defined by
[TABLE]
Lemma 4**.**
If , then the increasing process satisfies the following :
- (i)
* for every .*
- (ii)
.
Proof For any integer , we define an -stopping time Then we have
[TABLE]
¿From the fact that , we can derive
[TABLE]
Condition (H) imposed on the coefficient implies the finiteness of the right-hand side. Since is equal to on and as , the first assertion has been proved.
Since for all with some , we can take B_{\varepsilon}=\{y\in\mbox{\boldmathQ_{p}}\mid 1/\|b(x+y)\|_{p}^{\alpha}\geq\varepsilon\} so that is satisfied. Therefore, the second assertion follows from Theorem 2 and the estimate \int_{0}^{t}1/\|b(x+Z(s))\|_{p}^{\alpha}ds=\int_{\mbox{\boldmathQ_{p}}}L_{t}^{y}/\|b(x+y)\|_{p}^{\alpha}\mu(dy)\geq\varepsilon\nu(t,B_{\varepsilon}).
Here, we define for every . Since Lemma 4 shows that except Lebesgue measure zero set in for every and x\in\mbox{\boldmathQ_{p}}, we have
[TABLE]
Let us now introduce an -adapted time changed process defined by , where stands for the filtration given by for every . The objective of this section is demonstrating that is a solution of the stochastic differential equation (12) by assuming is locally bounded.
Since the ball centered at the origin and with radius is given as a disjoint union of balls centered at , we can define a Markov process by
[TABLE]
By applying [28, VI-28,3], we can obtain the Lévy system of the Markov process . From this observation, we can derive that the Lévy system of the -stable process is given by and . This shows that, for any non-negative measurable function on (\mbox{\boldmathQ_{p}})_{\Delta} satisfying and any initial probability measure ,
[TABLE]
is the dual predictable projection of A_{f}(t)=\int_{[0,t]}\int_{(\mbox{\boldmathQ_{p}})_{\Delta}}f(y)\lambda(\omega,ds,dy)=\sum_{s\leq t}f(X(s)-X(s-)) under , where (\mbox{\boldmathQ_{p}})_{\Delta} stands for the one point compactification of with the topological Borel -field and stands for the jump measure of the -stable process with filtration .
Consider a time changed jump measure
[TABLE]
with respect to with . Then, another representation of is obtained. Indeed, it is not difficult to see that Y(t)=\int_{[0,t]}\int_{\mbox{\boldmathQ_{p}}}x\tilde{\lambda}(ds,dx).
Similarly to Lemma 2.20 in [35], we can show the following :
Lemma 5**.**
For almost all in , has Lebesgue measure zero.
On the other hand, by performing time change given by , it turns out that the stochastic process determined by this jump measure is characterized by the Lévy system given by and . Therefore, for any non-negative measurable function on (\mbox{\boldmathQ_{p}})_{\Delta} and any initial probabaility measure ,
[TABLE]
is the dual predictable projection of under (see [28, VI-28,1]). By introducing a random measure , we have \tilde{A}_{H}^{f}(t)=\int_{[0,t]}\int_{(\mbox{\boldmathQ_{p}})_{\Delta}}f(y)\tilde{\pi}(\omega,ds,dy). Accordingly, is the compensator of in the sense given in [14].
Let us denote the topological Borel -field of by and the topological Borel -field of by . Now, we define a map from ([0,\infty)\times\mbox{\boldmathQ_{p}},\mathcal{B}[0,\infty)\times\mathcal{B}) to ([0,\infty)\times(\mbox{\boldmathQ_{p}})_{\Delta},\mathcal{B}[0,\infty)\times\mathcal{B}_{\Delta}) given by
[TABLE]
for every .
Lemma 6**.**
The random measure defined by is characterized by the compensator (intensity) , where stands for the measure on (\mbox{\boldmathQ_{p}})_{\Delta} given as on and vanishing at
Proof Thanks to Theorem 2.1(a) in [16], it suffices to show that . Since is satisfied for almost all with probability one, we have
[TABLE]
with probability one. By performing changing of the variables in the integral, we obatin
[TABLE]
Now, we define an -adapted stochastic process given by
[TABLE]
Then it turns out that is an -stable process. This is because the random measure admits the compensator .
For and , we have . By modifying and outside as the procedure established in Section 2, we can define the stochastic integral of with respect to the -stable process . Similarly to Lemma 2.26 in [35], we have the following assertion on the stochastic integral :
Lemma 7**.**
.
Consequently, the following assertion on solutions of (12) is now concluded.
Theorem 4**.**
*If the coefficient is locally bounded and if *(H) * is satisfied with respect to , then the stochastic differential equation (12) * admits a non-trivial solution.
Here, we note that the -stable process on has an explicit description of the density of transition probability kernel. Indeed, we firstly recall that
[TABLE]
is obtained by the expression of transition probability found by Yasuda (1996, [3]). Then, by denoting the right-hand side by , we immediately see that is given as on .
We have the following assertion which gives a sufficient condition for Condition (H).
Proposition 3. For the -stable process with on , the integrability condition
[TABLE]
with some positive number satisfying implies
[TABLE]
Proof For any positive real number , one sees that
[TABLE]
with some positive constants and , where . Accordingly, it is not difficult to see that
[TABLE]
with some positive constant , where the right-hand side converges, when .
Since , a real number is introduced by so that is satisfied. Then, it turns out that
[TABLE]
with the positive number given by . This is because implies .
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] S. Albeverio and W. Karwowski, Diffusion on p 𝑝 p -adic numbers, Gaussian random fields (Nagoya, 1990), 86–99, Ser. Probab. Statist. 1, World Sci. Publ., River Edge, N. J., 1991.
- 2[2] S. Albeverio and W. Karwowski, A random walk on p 𝑝 p -adics, the generator and its spectrum, Stochastic. Process. Appl. 53 (1994), 1–22.
- 3[3] S. Albeverio and X. Zhao, On the relation between different constructions of random walks on p-adics, Markov Process. Related Fields 6 (2000), 239–256.
- 4[4] S. Albeverio and X. Zhao, Measure-valued branching processes associated with random walks on p-adics, Ann. Probab. 28 (2000), 1680–1710.
- 5[5] S. Albeverio and X. Zhao, A decomposition theorem of Lévy processes on local fields, J. Theoret. Probab. 14 (2001), 1–19.
- 6[6] P. Baldi, E. Casadio Tarabusi and A. Figà-Talamanca, Stable laws arising from hitting distributions of processes on homogeneous trees and the hyperbolic half-plane, Pacif. J. Math. 197 (2001), 257–273.
- 7[7] R. M. Blumenthal and R. K. Getoor, Local times for Markov processes, Z. Wahrschenlichkeitstheorie und Verw. Gebiete 3 (1964), 50–74.
- 8[8] R. M. Blumenthal and R. K. Getoor, Markov processes and potential theory, Pure and Applied Mathematics 29, Academic Press, New York, 1968.
