Abstract
In this paper, we obtain the law of the iterated logarithm under a general moment condition with respect to sub-linear expectation. We present a novel proof by combining the exponential inequality with the subsequence method. Moreover, we show that the general moment condition we indicated is the weakest one for the law of the iterated logarithm under sub-linear expectation.
Keywords:
sub-linear expectation; law of the iterated logarithm; general moment condition; exponential inequalities MSC:
60F15
1. Introduction
The classical law of the iterated logarithm (LIL for short) is one of the foundational core results in probability theory, rooted in the additive properties of probability and expectation. In recent years, domains such as statistics and risk measurement have been confronted with model uncertainties that can no longer be sufficiently characterized within the classical linear framework. Consequently, non-additive probabilities and non-linear expectations have become more popular, as they provide a much more reliable way to deal with these uncertainties (cf. Huber and Strassen (1973), Peng (1997), Chen and Epstein (2002), Maccheroni and Marinacci (2005) [1,2,3,4]).
The framework of sub-linear expectation was introduced by Peng (2010) [5]. Within this setting, several classical limit theorems have been successfully extended. Chen et al. (2013) and Chen (2016) [6,7] obtained the strong law of large numbers for independent random variables under the finite -th moment condition () with respect to sub-linear expectation. Hu (2016) [8] established the strong law of large numbers for sub-linear expectation under a general moment condition based on a function class , and further indicated that such a general moment condition is the weakest moment condition for the strong law of large numbers with respect to sub-linear expectation. Hu (2018) [9] investigated the central limit theorems under sub-linear expectation without the assumption of identical distribution. Xu and Zhang (2020) [10] obtained the law of the logarithm for arrays of row-wise extended negatively dependent random variables under finite fourth moment conditions with respect to sub-linear expectation.
Among various limit theorems under sub-linear expectation, the LIL has been extensively investigated. Chen and Hu (2014) [11] established the LIL for sequences of uniformly bounded random variables within the framework of capacities, generalizing the classical Kolmogorov and Hartman–Wintner laws to non-additive probability spaces. Zhang (2016) [12] generalized the LIL to negatively dependent and identically distributed random variables, establishing the results under the moment condition with respect to Choquet expectation , and further showed that this condition is necessary and sufficient for the validity of the LIL. Zhang (2021) [13] established Wittmann’s LIL for independent but not necessarily identically distributed random variables under sub-linear expectation. Zhang (2022) [14] further derived the LIL for extended negatively dependent random variables under the moment condition with respect to sub-linear expectation for some .
However, in contrast to the necessary and sufficient moment condition with respect to Choquet expectation established by Zhang (2016) [12], all moment conditions currently available for the LIL with respect to sub-linear expectation are only sufficient for its validity. Since the Choquet expectation generated by a sub-additive capacity is larger than the corresponding sub-linear expectation, moment conditions formulated with respect to Choquet expectation are typically stronger than those with respect to sub-linear expectation. This motivates us to investigate whether the LIL can be established under a milder sub-linear expectation moment condition that relaxes the prevailing finite -th moment assumption for independent, not necessarily identically distributed random variables.
To precisely characterize the moment conditions, we follow Hu (2016) [8] and introduce the following function classes. Let (or, respectively, ) denote the set of functions defined on such that satisfies:
- (1)
- is non-negative and non-decreasing on and positive on for some . The series converges (respectively, diverges);
- (2)
- For any fixed , there exists such that for any .
The value of is not assumed to be the same for different . The functions and for any are examples of the class of functions . The functions and belong to the class .
Inspired by Hu (2016) [8], we derive that the LIL still holds under a general moment condition with respect to sub-linear expectation based on the above complementary function classes. Specifically, we establish that the moment condition for some is sufficient to guarantee the validity of the LIL, which is weaker than the -th moment condition required by Zhang (2022) [14]. We present a novel proof strategy that combines the exponential inequality techniques developed by De Acosta (1983) [15] with the classical subsequence method. We also prove that this moment condition is indeed the weakest possible for the LIL with respect to sub-linear expectation. Conversely, if the moment condition fails for all , then it can at most hold for some , and we can always construct an explicit counterexample showing that the LIL fails in this case.
The rest of this paper is organized as follows. In Section 2, we recall some basic concepts and lemmas related to sub-linear expectation. In Section 3, we state and prove our main result, that is the LIL under a general moment condition for sub-linear expectation. In Section 4, we illustrate that this condition is the weakest.
Throughout this paper, we define , . We say that two functions and are asymptotically equivalent, denoted by as , if . In addition, the symbol C stands for a positive constant whose value may differ at different places.
2. Basic Concepts and Lemmas
We use the definitions and framework proposed by Peng (2010) [5]. Let be a given measurable space and let be a linear space of real functions defined on such that if then for each , where denotes the linear space of (local Lipschitz) functions satisfying
for some , depending on . is considered as a space of random variables.
As follows we present some definitions and basic properties of sub-linear expectation.
Definition 1.
A sub-linear expectation on is a functional satisfying the following properties: for all , we have
- (a)
- Monotonicity: If , then ;
- (b)
- Constant preserving: , ;
- (c)
- Positive homogeneity: , ;
- (d)
- Sub-additivity: whenever is not of the form or .
The triple is called a sub-linear expectation space. Given a sub-linear expectation , we denote the conjugate expectation of by
Definition 2.
Let . A function is called a capacity if
The capacity V is said to be sub-additive if for all such that . We denote a pair of capacities by
where is the complement set of A. Then
It is obvious that is sub-additive, but and are not.
Also, we define the Choquet integrals/expectations () by
with V being replaced by and , respectively.
Definition 3.
A sub-linear expectation is said to be continuous if it satisfies:
- (1)
- Lower-continuity: if , then , where .
- (2)
- Upper-continuity: if , then , where .
A capacity V: is called a continuous capacity if it satisfies:
- (1)
- Lower-continuity: if , then , where .
- (2)
- Upper-continuity: if , then , where .
Definition 4.
(Independence). Let and be two random variables on . is said to be independent of X, if for each test function we have whenever for all x and . is said to be a sequence of independent random variables, if is independent of for each .
Definition 5.
(Negative dependence). On a sub-linear expectation space , the random vector is said to be negatively dependent (ND) to another random vector if for each pair of test functions and , we have
whenever and are both coordinatewise non-decreasing (or both coordinatewise non-increasing) with , , and all relevant expectations are finite.
Accordingly, a sequence of random variables is said to be negatively dependent if is negatively dependent to for each .
Remark 1.
It follows from Definitions 4 and 5 that negative dependence is a general extension of independence.
The following lemma provides the Borel-Cantelli Lemma under sub-linear expectation whose proof can be found in Chen et al. (2013) [6].
Lemma 1 (Borel-Cantelli Lemma).
Let be a sequence of events in and be the capacity induced by lower-continuous sub-linear expectation . If , then
The next lemma gives the Chebyshev’s inequality under sub-linear expectation.
Lemma 2 (Chebyshev’s Inequality).
Let X be a random variable defined on and function be non-decreasing and positive, we have:
Proof.
By the monotonicity of and , we have
Since , belongs to . Then by (2) we have:
□
The next lemma refines a key convergence property of and its proof can be found in Cui and Hu (2026) [16].
Lemma 3.
Assume that , then
The following lemma establishes the exponential inequality for negatively dependent random variables under sub-linear expectation and the proof can be found in Zhang (2016) [12].
Lemma 4.
Let be a sequence of negatively dependent random variables on with . Denote . Then for any , there exists a constant such that for any and ,
The next lemma is an exponential maximal inequality under sub-linear expectation. This lemma is stated in Hu et al. (2024) [17], while this result occurs first in Zhang (2016) [18].
Lemma 5.
Suppose that is a sequence of random variables defined on . Then for any ,
3. Main Results
Starting from this section, we consider the following general moment condition:
which will be used in all subsequent lemmas and main results. We state our main results and their proofs in what follows. The subsequent two lemmas are essential to the proofs of our theorems.
Lemma 6.
For any and , under the moment condition (5), there holds
Proof.
Noticing that , as , we have
where the last series converges by Lemma 3. □
Lemma 7.
Let X be a random variable defined on . Assume that the moment condition (5) holds. Then
Proof.
Since the convergence of implies , we have
as .
□
The next theorem states the main result of this paper which is the LIL under sub-linear expectation.
Theorem 1.
Let be a sub-linear expectation space where is lower-continuous and is the induced capacity. Let be a sequence of ND random variables satisfying the moment condition (5). Assume that for each . Denote . Then
Remark 2.
To investigate the limit properties of truncated random variables, Zhang (2021) [13] introduced the notation for the limit of the truncated sub-linear expectation, which is defined as
It can be verified that if , then holds. So in Theorem 1, under the moment condition (5), by Lemma 7 there holds for each ,
Then we have
Thus the moment condition (5) ensures that the asymptotic analysis of variances under sub-linear expectation can be reduced to the original sub-linear expectation setting without any loss of information.
Proof.
If (6) holds, then replacing with in (6), we have (7). Therefore, it suffices to prove (6). Without loss of generality, we may assume that . To establish (6), we only need to show that
Denote where is a positive constant to be specified subsequently. Define
Thus we have
Moreover, for any , it suffices to prove the following four statements:
- Step 1. Convergence of the far-tail truncated term:
For (12), noting that as , by the moment condition (5) and Lemma 3, we have
Hence, by the Borel-Cantelli lemma. Thus we obtain (12).
- Step 2. Convergence of the intermediate truncated term: ,
For (10), by the definition of , we have . It is sufficient to show that
Let . Then for any , by (3), we have
Note that . By Lemma 7, we have for all sufficiently large n. Thus, applying Lemma 6, we obtain
Hence, by the Borel-Cantelli lemma, we have
On the other hand,
Combining (16) with (15), (14) is proved. So (10) holds. Since for all , (11) holds immediately.
- Step 3. Convergence of the near-truncated term:
Fix the arbitrarily small positive constant mentioned above. By Remark 2 we have
Choose sufficiently small such that . Then, there exists a positive integer N such that for all , we have
Thus, we have
It is obvious that
Therefore, it suffices to prove
Let , where satisfies . Then it can be verified that and , as . Consider k sufficiently large such that . Then, for , we have
Without loss of generality, based on the earlier assumption for each , we have
By (19) and (20), it follows
Hence, to obtain (18), by the Borel-Cantelli lemma, we only need to prove and .
For , by the Chebyshev’s inequality, we have
where depends on n and will be determined later. By the elementary exponential inequality
and noting that for any ,
it follows that
Taking expectations on both sides, for , we have
where the last inequality holds since for all .
Consequently,
where the first inequality follows from the negative dependence of and the monotonicity of the exponential function. Let
By (17), we have
Let be a fixed small positive constant satisfying . Choose sufficiently small such that
Thus we have
Finally, we consider . By , as , we have
where depends on n and will be determined later. By (4), we have
where the second-to-last equality follows from the earlier assumption for each . Similar to the proof of (22), we obtain
Hence, combing (24)–(26), we have
Choose
For , by (20) we have
On the other hand
as . Combining (28) with (29), we obtain as .
For , to simplify the notation in subsequent proof, denote . Then we have
where the last inequality follows , for k sufficiently large. Taking small enough to satisfy , let k sufficiently large such that
Consequently, by (17) we have
Thus by (27), (30) and as , we have
Hence, by virtue of (21), (23), (31) and the Borel-Cantelli lemma, (9) holds. Now (9)–(12) are all established. The proof is complete. □
4. The Weakest Condition for Sub-Linear Expectation
In this section, we show that the moment condition (5) is the weakest moment condition with respect to sub-linear expectation to ensure the validity of the LIL. To illustrate this, we recall the following condition introduced by Zhang (2022) [19]:
(CC): The sub-linear expectation on satisfies
where , and is a countable-dimensionally weakly compact family of probability measures on . This compactness implies that, for any and any sequence , there exist a subsequence and a probability measure such that
According to Zhang (2021) [13], if is a complete separable metric space, each element in is a continuous function on . The sub-linear expectation is defined as
where is a weakly compact family of probability measures on the metric space , then condition (CC) is satisfied.
Zhang (2022) [19] demonstrated that for an independent and identically distributed (i.i.d.) sequence on , if condition (CC) holds, the necessity of the LIL follows, i.e.,
It seems that we cannot derive the similar necessary moment condition with respect to sub-linear expectation. However, we may employ the above moment condition with respect to Choquet expectation to prove that condition (5) is the weakest moment condition ensuring the validity of the LIL under sub-linear expectation.
The following counterexample illustrates that for any given , there always exists a sequence of i.i.d. random variables such that , but , which implies that the LIL in Theorem 1 fails to hold.
Example 1.
But
Let be a sequence of countable sample spaces, and be the σ-algebra consisting of all subsets of . Define the family of probability measures , on as follows:
- (i)
- The Dirac measure :
- (ii)
- For each integer :
It is easy to check that weakly converges to as . So each , is weakly compact.
Define the product measurable space and product probability family:
The sub-linear expectation is induced by the upper expectation: for any bounded random variable ,
Since each is weakly compact, by the diagonal argument for probability sequences, the product family is countable-dimensionally weakly compact and thus satisfies condition (CC).
Next, we construct a special sequence of random variables to make the LIL fail to hold. To simplify the notation, denote , Define random variables on each :
Then Z satisfies that for any ,
Further define the sequence on the product space by
It is easy to check that is a sequence of i.i.d. random variables under sub-linear expectation .
We now show that for all . For , we have . For , , we have
Then for any , we have . Hence
We proceed to discuss the convergence of the moment condition with respect to sub-linear expectation and the moment condition with respect to Choquet expectation for .
- Note that for any ,
Therefore by (32), we have successfully constructed a sequence of i.i.d. random variables under sub-linear expectation such that for any while , confirming that the LIL fails to hold under this setting.
Author Contributions
Methodology, C.H.; Validation, X.H.; Formal analysis, X.H. and C.H.; Writing—review and editing, X.H.; Supervision, C.H.; Funding acquisition, C.H. All authors have read and agreed to the published version of the manuscript.
Funding
This research was funded by the Outstanding Youth Innovation Team Program of Shandong Province (Grant No. 2024KJG010), the National Natural Science Foundation of China (Grant Nos. 12371148, 42007141) and the Natural Science Foundation of Shandong Province (Grant No. ZR2019BA038).
Data Availability Statement
Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.
Acknowledgments
The authors would like to thank the editor and five referees for their valuable comments, which improved the manuscript.
Conflicts of Interest
The authors declare no conflicts of interest.
References
- Huber, P.; Strassen, V. Minimax tests and the Neyman-Pearson lemma for capacities. Ann. Stat. 1973, 1, 251–263. [Google Scholar] [CrossRef] [Scilit]
- Peng, S. Backward SDE and related g-expectation. In Backward Stochastic Differential Equations; El Karoui, N., Mazliak, L., Eds.; Pitman Research Notes in Math; CRC Press: Boca Raton, FL, USA, 1997; Volume 364, pp. 141–159. [Google Scholar]
- Chen, Z.; Epstein, L. Ambiguity, risk, and asset returns in continuous time. Econometrica 2002, 70, 1403–1443. [Google Scholar] [CrossRef] [Scilit]
- Maccheroni, F.; Marinacci, M. A strong law of large number for capacities. Ann. Probab. 2005, 33, 1171–1178. [Google Scholar] [CrossRef] [Scilit]
- Peng, S. Nonlinear expectations and stochastic calculus under uncertainty. arXiv 2010, arXiv:1002.4546. [Google Scholar]
- Chen, Z.; Wu, P.; Li, B. A strong law of large numbers for non-additive probabilities. Int. J. Approx. Reason. 2013, 54, 365–377. [Google Scholar] [CrossRef] [Scilit]
- Chen, Z. Strong laws of large numbers for sub-linear expectations. Sci. China Math. 2016, 59, 945–954. [Google Scholar]
- Hu, C. A strong law of large numbers for sub-linear expectation under a general moment condition. Stat. Probab. Lett. 2016, 119, 248–258. [Google Scholar] [CrossRef] [Scilit]
- Hu, C. Central limit theorems for sub-linear expectation under the Lindeberg condition. J. Inequal. Appl. 2018, 2018, 316. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Xu, J.; Zhang, L.X. The law of logarithm for arrays of random variables under sub-linear expectations. Acta Math. Appl. Sin. Engl. Ser. 2020, 36, 670–688. [Google Scholar] [CrossRef] [Scilit]
- Chen, Z.; Hu, F. A law of the iterated logarithm under sublinear expectations. J. Financ. Eng. 2014, 1, 1450015. [Google Scholar]
- Zhang, L.X. Exponential inequalities under the sub-linear expectations with applications to laws of the iterated logarithm. Sci. China Math. 2016, 59, 2503–2526. [Google Scholar] [CrossRef] [Scilit]
- Zhang, L.X. On the laws of the iterated logarithm under the sub-linear expectations without the assumption on the continuity of capacities. Probab. Uncertain. Quant. Risk 2021, 6, 409–460. [Google Scholar]
- Zhang, L.X. Strong limit theorems for extended independent random variables and extended negatively dependent random variables under sub-linear expectations. Acta Math. Sci. 2022, 42, 467–490. [Google Scholar] [CrossRef] [Scilit]
- De Acosta, A. A new proof of the Hartman-Wintner law of the iterated logarithm. Ann. Probab. 1983, 11, 270–276. [Google Scholar] [CrossRef] [Scilit]
- Cui, M.; Hu, C. A strong law of large numbers for m-dependent random variables under sub-linear expectation. Stochastics 2026, 1–14. [Google Scholar] [CrossRef] [Scilit]
- Hu, C.; Wang, L.; Zong, G. Limit theorems for delayed sums under sublinear expectation. J. Math. Anal. Appl. 2024, 535, 128084. [Google Scholar] [CrossRef] [Scilit]
- Zhang, L.X. Self-normalized moderate deviation and laws of the iterated logarithm under G-expectation. Commun. Math. Stat. 2016, 4, 229–263. [Google Scholar] [CrossRef] [Scilit]
- Zhang, L.X. A note on the cluster set of the law of the iterated logarithm under sub-linear expectations. Probab. Uncertain. Quant. Risk 2022, 7, 85–100. [Google Scholar] [CrossRef] [Scilit]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.