We study (unreflected and reflected) Lévy processes that are modified at independent Poisson review epochs by a random state-dependent reset map. The first model is unreflected: between resettings the level evolves as a Lévy process, and at a resetting epoch the new level is F of the level seen just before the resetting. The second model is the reflected Lévy process at 0, but with reset maps chosen from a wider transform-solvable class. In this work, we give contraction-type ergodicity criteria for the existence of the steady-state. We identify this steady-state of the system as the solution of certain stochastic fixed-point equation. This allows to find the Laplace-transform, moments recursions, and regularly-varying-tail formulae. For non-reflected case, the examples include affine deterministic resets F(x)=cx+b, random-coefficient resets F(x)=Cx+b with two-point C for the finite-activity Lévy process, spectrally positive Lévy flight, reflected Brownian process. For the reflected cased we analyse affine resetting mechanism F(x)=cx, the alternating resetting F(x)=(B−x)+, and uniform proportional resetting F(x)=Cx with C having uniform distribution.
Introduction
There has been a vast amount of papers related to the resetting mechanism. This mechanisms arise in search processes (see, e.g., [2, 9, 16]), physics (see, e.g., [6]), biology (see, e.g., [15]), queueing theory (see, e.g., [12]) and in many other research fields; see also the survey [11]. The usual resetting mechanism is based either by sending the process to 0 (or any other fixed state) or by multiplying by possibly random amplitude at independent Poisson epochs. Lately, there have been efforts to consider more general resetting mechanisms, see for example [1] and references therein. Based on [14], we also consider more general possibly random resetting mechanisms F(x), where x is state prior resetting, that is applied to Lévy processes X and its reflected version Y at 0.
For unreflected Lévy process X, we give a contraction ergodicity criterion for the resetting mechanism to get steady-state of our system. For the stationary law we give a general transform equation and moment recursions for arbitrary reset maps F whenever powers of F(Z) can be expressed through the moments of Z. For the resetting mechanism F(x)=cx+b (where c,b are constants), using a perpetuity representation, we obtain an infinite-product transform of the stationary law. We also identify the moment recursions. Finally, we also present the behaviour of the asymptotic tail of the law of this stationary distribution under regular variation of the positive jumps. We managed also to analyse another interesting case when F(x)=Cx+b with two-point and three-point distribution of C.
For the reflection at 0 case we consider Lévy process without negative jumps X. Such process is called spectrally positive. For spectrally positive Lévy process Y reflected at 0 we also give sufficient conditions for the existence of the steady-state, and wrire an appropriate recursion identifying the limiting law. We analyse in detail the reſtting maps F(x)=cx, F(x)=(B−x)+ with exponential B, and F(x)=Cx with C having uniform distribution U(0,1). We also include the tail of distribution of the steady-state in the case when jumps have regularly varying jumps.
The present paper has two aims. Firstly, for very general process between resettings and quite general resetting mechanisms, we give the ergodicity conditions and show that the limiting stationary distributions satisfies a one-step recursive equation. This equations codes some recursive procedures to find moments of this stationary distribution and allows to identify the tail behaviour when jumps of the process between resettings is regularly varying. The second aim is to analyse the process reflected at 0 and then reset it. We show that, to some extend, the same ideas as for the process without reflection will still work.
The paper is organised as follows. In the next section we analyse the case without reflection and then we focus on the reflected case. Throughout this paper random reset coefficients are denoted by capital letters, such as C; deterministic reset constants are denoted by lower-case letters, such as c and b.
The unreflected reset model
Notations
For a real-valued Lévy process X, write
ψ(s):=log⟨e−sX(1)⟩,
whenever the expectation is finite. Then, for an independent exponential random variable eq∼Exp(q) with intensity q, we denote
m(s):=⟨e−sX(eq)⟩=∫0∞qe−qtetψ(s)dt=q−ψ(s)q,(1)
for s in the domain
Dq:={s:Re(q−ψ(s))>0}.
In the reflected spectrally positive part of the paper, we use the standard fluctuation notation
X(t):=0≤u≤tsupX(u),X(t):=0≤u≤tinfX(u).
Let Φ(q) be the largest non-negative solution of
ψ(Φ(q))=q.(2)
For a spectrally positive Lévy process, the Wiener–Hopf factorisation gives that −X(eq) has the exponential law with intensity Φ(q) and
ℓq(s):=⟨e−sX(eq)⟩=Φ(q)(ψ(s)−q)q(s−Φ(q)).(3)
The expression has the usual removable singularity at s=Φ(q).
Embedded chain
The unreflected process with resetting evolves as follows. Immediately after a resetting Poissonian epoch Tn with intensity q, the level is reset to F of the level just before that epoch. During the following inter-review interval, it evolves by adding an independent Lévy increment. We denote this Lévy process with resetting by Z(t). Let
Zn:=Z(Tn−),
be the pre-resetting level. Then
Zn+1=Fn+1(Zn)+ξn+1,ξn+1=dX(eq),(4)
with ξn+1 independent of Fn+1 and Zn. It is so-called one step analysis based on observing the process at resetting times. Assume that stationary law exists and is unique. Then, at stationarity,
Z=dF(Z)+ξ,ξ=dX(eq),(5)
where Z is a steady-state and it is independent of F and ξ on the right-hand side. This identity is a direct consequence of so-called PASTA property.
Ergodicity conditions
We start from identifying the sufficient ergodicity conditions for the existence of the stationary distribution. that follows directly from the iterated-random-functions theorem.
Theorem 2.1 (Unreflected contraction criterion). Assume that there are i.i.d. nonnegative random variables An such that
∣Fn(x)−Fn(y)∣≤An∣x−y∣for all x,y,
and that, for some x0,
⟨logA1⟩<0,⟨log+∣F1(x0)+ξ1∣⟩<∞.(6)
Then the chain (4) has a unique stationary distribution. Moreover, for any initial state Z0, Zn converges in distribution to the unique solution of (5).
Proof. Let Gn(x)=Fn(x)+ξn. Then
∣Gn(x)−Gn(y)∣=∣Fn(x)−Fn(y)∣≤An∣x−y∣.
Thus Gn is a random Lipschitz map with negative mean logarithmic Lipschitz constant. The second condition in (6) is the usual integrability condition. The standard backward iteration
G1∘G2∘⋯∘Gn(x0)
converges almost surely to a finite limit that is independent of x0, and the forward chain converges weakly to the law of that limit. □
For affine reset maps, the condition is straightforward: the contraction depends only on the multiplicative coefficient, while the additive part only enters through the integrability condition.
Corollary 2.2 (Affine and random-affine resets). Let
Fn(x)=Cnx+Bn,
where (Cn,Bn) are i.i.d. and independent of ξn. If
⟨log∣C1∣⟩<0,⟨log+∣B1+ξ1∣⟩<∞,
then the unreflected reset chain has a unique stationary distribution. In particular, if F(x)=cx+b with ∣c∣<1 and Elog+∣b+X(eq)∣<∞, then the stationary distribution exists and is unique.
Laplace transform of steady-state and moments recursion
If J is nonnegative and regularly varying with index −α, α>0, i.e.
P(J>x)=x−αL(x),
where L is slowly varying, then ξ=X(eq) has the one-big-jump tail
P(ξ>x)∼qλP(J>x).(22)
Using the perpetuity representation (11) and standard subexponential closure,
P(Z>x)∼j=0∑∞P(cjξj>x)∼qλ1−cα1P(J>x),x→∞,(23)
for 0<c<1.
Unreflected spectrally-positive α-stable flight under affine resetting
Let X be a spectrally-positive stable Lévy flight with index α∈(1,2) and negative linear component, normalised so that
ψ(s)=μs+κsα,μ>0,κ>0.(24)
The positive jumps have tail
Π(x,∞)∼Kαx−α,(25)
for a constant $K_\alpha>0 determined by the normalisation of the stable flight.
For F(x)=cx+b, 0<c<1, the transform is
v(s)=exp{−1−cbs}j=0∏∞q−μcjs−κcαjsαq.(26)
The domain is 0≤s<Φ(q), where Φ(q) is the positive root of μs+κsα=q. Since the positive jumps are regularly varying, the exponential-time increment satisfies
P(X(eq)>x)∼q1Π(x,∞)∼qKαx−α.(27)
Therefore
P(Z>x)∼q(1−cα)Kαx−α.(28)
Unreflected two-point random coefficient
2.9. Unreflected two-point random coefficient. Let
Assume now that X is spectrally positive and not a subordinator. Given an initial level x≥0, define the reflected process at 0 by
Y(t)=x+X(t)−0≤u≤tinf(x+X(u))∧0.(36)
At resetting Poissonian epochs with intensity q>0, the resetting using functional F is performed. Let
Yn:=Y(Tn−).
Then
Yn+1=dmax{Fn+1(Yn)+X(eq),X(eq)}.(37)
and at stationarity (assuming that it exists and is unique),
Y=dmax{F(Y)+X(eq),X(eq)},(38)
where by Y we denote the steady-state. Equivalently, by Wiener–Hopf factorisation,
Y=dX(eq)+(F(Y)−EΦ)+,(39)
where EΦ has the exponential distribution with parameter Φ(q) and X(eq), EΦ, and F(Y) are independent on the right-hand side.
Ergodicity conditions
The following is the Palmowski-Vlasiou [14] comparison criterion, written in the notation used here.
Theorem 3.1 (Reflected comparison criterion). Assume that X is not a subordinator and that there exist nonnegative i.i.d. random variables (An,Bn) such that
Fn(x)≤Anx+Bn,x≥0.(40)
Then the reflected reset chain has a unique stationary limiting distribution if one of the following holds:
(a) An=0 a.s.;
(b) An=1 a.s. and ⟨B1⟩+⟨X(1)⟩/q<0;
(c) ⟨logA1⟩<0 and ⟨log+B1⟩<∞.
Here, if B1=0 in (c), one may dominate by Anx+1 without changing the conclusion.
Proof sketch. Under (40), the original chain is stochastically dominated by the affine reflected chain
Yn+1a=dmax{An+1Yna+Bn+1+X(eq),X(eq)}.
Cases (a)–(c) imply return to a compact set. Since X is not a subordinator, the reflected process can hit zero, and zero is a regeneration state in the Harris-chain sense. The minorisation on a compact set then gives positive Harris recurrence and uniqueness of the stationary distribution. □
Remark 3.2. For the partial restting F(x)=Cx with 0≤C≤1 and P(C<1)>0, condition (c) holds after domination by Cx+1 whenever ⟨logC⟩<0. Note that this holds for example when C has a uniform law U(0,1), because in this case ⟨logC⟩=−1.
Laplace transform, moments and tail of steady-state
Let
v(s):=⟨e−sY⟩,hF(s):=⟨e−sF(Y)⟩.
Using (39), independence, and the elementary identity
⟨e−s(a−EΦ)+⟩=Φ(q)−sΦ(q)e−sa−se−Φ(q)a,a≥0,(41)
we get
v(s)=ℓq(s)Φ(q)−sΦ(q)hF(s)−shF(Φ(q)),(42)
where ℓq(s) is defined in (3). Substituting (3) and simplifying yields the reset-isolated form
For C having a uniform distribution U(0,1), we have ⟨Cn⟩=(n+1)−1 and KC=Φ(q)−1∫0Φ(q)v(u)du.
Following Palmowski–Vlasiou [14], we record the regularly varying tail expression for the most relevant partial resetting case with random proportion.
Proposition 3.3 (Regularly varying reflected Lévy process under partial resetting). Assume that X is spectrally positive, not a subordinator, and that its Lévy measure satisfies
Π(x):=Π(x,∞)∈RV−α,α>0.
If F(x)=Cx, where 0≤C≤1, then
P(Y>x)∼q(1−⟨Cα⟩)1Π(x).(49)
Proof sketch. The Wiener–Hopf representation (39) gives
Y=dX(eq)+(CY−EΦ)+.
The exponential EΦ is light-tailed and does not change a regularly varying tail. The stochastic recurrence is therefore tail-equivalent to the random-coefficient affine recursion
Y≈X(eq)+CY.
For regularly varying additive term with index α and ⟨Cα⟩<1, the standard one-large-term argument gives
Equations (54)–(55) give a completely explicit transform in terms of the Lévy exponent ψ.
For completeness, again using Palmowski–Vlasiou [14], we record the exponential-type asymptotics of the tail of the law of Y. Suppose there exists θ∗>0 such that
ψ(θ∗)=q,(56)
and the corresponding tilted ladder-height mean mq is finite. Let
⟨eθ∗F(Y)⟩<∞.(57)
Then the reflected Lévy process under resetting the asymptotic tail is given by
P(Y>x)∼De−θ∗x,(58)
with
D=P(EΦ>F(Z))θ∗mqκ(q,0),EΦ∼Exp(Φ(q)),(59)
where κ(q,0) is LST of ladder time. For bounded resets and for F(x)=(B−x)+ with light-tailed B, the condition (57) is immediate. For F(x)=Cx, it becomes a condition on v(−θ∗C), and therefore requires finite positive exponential moments of Y.
Reflected Brownian motion under partial resettting
Let F(x)=(B−x)+, where B has an exponential law Exp(β) that is independent of Y. The memoryless property closes the reset transform. Conditional on Y=z,
(B−z)+=0on {B≤z},
and, on {B>z}, the excess B−z is exponential with rate β. Therefore
P((B−Y)+>y)=⟨e−β(Y+y)⟩=v(β)e−βy,y≥0.(64)
Thus (B−Y)+ has an atom 1−v(β) at zero and an exponential density v(β)βe−βy on (0,∞). Hence
Reflected spectrally positive stable flight under partial restting
Let X have exponent
ψ(s)=μs+κsα,α∈(1,2),
as in (24). For deterministic proportion of the partial resetting, hence for F(x)=cx, 0<c<1, the transform is given by (54) with
a(s)=q−μs−κsαq(78)
and Φ(q) solving μΦ(q)+κΦ(q)α=q. Note that the Lévy tail satisfies (25). Then Proposition 3.3 gives
P(Y>x)∼q(1−cα)Kαx−α.(79)
For uniform partial resetting, one should replace (1−cα)−1 by (α+1)/α.
Conclusion
The unreflected model with resetting gives a clean baseline: one step analysis reduce the description of stationary state Z to the fixed-point equation Z=dX(eq)+F(Z) for the steady-state Z. The reflected model keeps the same reset structure but inserts the Wiener–Hopf correction term, producing the stationary state identity (39). The presented method is quite robust. For example, we can analyse other processes under resetting: unreflected Lévy process under affine random resetting, that is, F(x)=Cx+B with C having finite support, beta distributions, or mixtures with an atom at zero. Similarly, we can handle polynomial resetting mechanism with F(x)=b0+b1x+⋯+bdxd. Then (8) expresses mn through moments up to order dn. Very interesting seems to be the case the resetting is based on finite-support random functions. In this case F takes maps f1,…,fr with probabilities p1,…,pr. Then
v(s)=m(s)i=1∑rpi⟨e−sfi(Z)⟩.
The same resetting mechanisms can be considered for the reflected Lévy process. Within this context quite intriguing is the case of Brownian plus compound Poisson with phase-type positive jumps which yields rational ψ. Then all transform formulae reduce to products or finite sums involving rational factors. This is the natural setting for numerical inversion.
Faculty of Pure and Applied Mathematics, Wrocław University of Science and Technology Wyb. Wyspiańskiego 27, 50-370 Wrocław, Poland
[1]J. Assefa, M. Keller-Ressel, Z. Palmowski. Moments of polynomial processes under affine resetting. arXiv:2610.01363, 2026.arxiv.org/abs/2610.01363
[2]W.J. Bell. Searching behaviour: The behavioural ecology of finding resources. Chapman and Hall, 1991.
[6]M. Dahlenburg, A. Checkin, R. Schumer, R. Metzler. Stochastic resetting by a random amplitude. Phys. Rev. E 103:052123, 2021.DOI
[9]M.R. Evans and S.N. Majumdar. Diffusion with stochastic resetting. Phys.Rev. Lett. 106:160601, 2011.DOI
[11]S. Gupta and A.M. Jayannavar. Stochastic resetting: A (very) brief review. Front. Phys. 10:789097, 2022.DOI
[12]F.F. Guillemin, Ph. Robert, B. Zwart. AIMD algorithms and exponential functionals. Ann. Appl. Probab., 14:90–117, 2004.DOI
[14]Z. Palmowski and M. Vlasio. A Lévy input model with additional state-dependent services. Stochastic Processes and their Applications, 121:1546–1564, 2011.DOI
[15]E. Roldan, A. Lisica, D. Sanchez-Taltavull, S.W. Grill. Stochastic resetting in backtrack recovery by RNA polymerases. Phys. Rev. E 93(6):062411, 2016.DOI
[16]J. Whitehouse, M.R. Evans, and S.N. Majumdar. Effect of partial absorption on diffusion with resetting. Phys. Rev. E 8:022118, 2013.DOI