1. Introduction
Diffusion processes are often used to describe the development of dynamic systems in a broad variety of scientific disciplines, including physics, biology, population dynamics, neurology, finance, and queueing. There is much interest in analyzing the “first-passage time” (FPT) issue in various situations. This entails determining the probability distribution of a random variable that describes the moment at which a process, beginning from a fixed initial state, reaches a defined boundary or threshold for the first time, which may also be time-varying. Unfortunately, closed-form solutions for the FPT densities are only accessible in a limited number of instances, leaving the more difficult job of determining the FPT densities through time-dependent boundaries.
Some general methods to solve FPT problems are based on:
- Analytical methods to determine the Laplace transform of FPT probability density function (pdf) and its moments for time-homogeneous diffusion process and constant boundaries (cf., for instance, Darling and Siegert [ 1- ], Blake and Lindsey [ 2- ], Giorno et al. [ 3- ]); 
- Symmetry properties of transition density to obtain closed-form results on the FPT densities through time-dependent boundaries and other related functions (cf., for instance, Di Crescenzo et al. [ 4- ]); 
- Construction of FPT pdf by making use of certain transformations among diffusion processes (cf., for instance, Gutiérrez et al. [ 5- ], Di Crescenzo et al. [ 6- ], Giorno and Nobile [ 7- ]); 
- Formulation of integral equations for the FPT density (cf., for instance, Buonocore et al. [ 8- ], Gutiérrez et al. [ 9- ], Di Nardo et al [ 10- ]); 
- Analysis of the asymptotic behavior of FPT pdf for large boundary or large times (cf., for instance, Nobile et al. [ 11- , 12- ]) 
- Efficient numerical algorithms and simulation procedures to estimate FPT pdf’s (cf., for instance, Herrmann and Zucca [ 13- ], Giraudo et al. [ 14- ], Taillefumier and Magnasco [ 15- ], Giorno and Nobile [ 16- ], Naouara and Trabelsi [ 17- ]). 
In the present paper, we focus on the FPT problem for the Feller-type diffusion process.
Let 
, 
, be a time-inhomogeneous Feller-type diffusion process, defined in the state space 
, which satisfies the following stochastic differential equation:
      where 
 is a standard Wiener process. Hence, the infinitesimal drift and infinitesimal variance of 
 are
      
      and we assume that 
, 
, 
 are continuous functions for all 
.
The Feller diffusion process plays a relevant role in different fields: in mathematical biology to model the growth of a population (cf. Feller [
18], Lavigne and Roques [
19], Masoliver [
20], Pugliese and Milner  [
21]), in queueing systems to describe the number of customers in a queue (cf. Di Crescenzo and Nobile [
22]), in neurobiology to analyze the input–output behavior of single neurons (see, for instance, Giorno et al. [
23], Buonocore et al. [
24], Ditlevsen and Lánský [
25], Lánský et al. [
26], Nobile and Pirozzi [
27], D’Onofrio et al. [
28]), in mathematical finance to model interest rates and stochastic volatility (see Cox et al. [
29], Tian and Zhang [
30], Maghsoodi [
31], Peng and Schellhorn [
32]). In population dynamics, the Feller-type diffusion process arises as a continuous approximation of a linear birth–death process with immigration (cf., for instance, Giorno and Nobile [
33]). The Feller process has the advantage of having a state space bounded from below, a property that in the neuronal models allows the inclusion of the effect of reversal hyperpolarization potential. In this context, the statistical estimation of parameters of the Feller process starting from observations of its first-passage times plays a relevant role (cf., for instance, Ditlevsen and Lánský [
25], Ditlevsen and Ditlevsen [
34]). The study of the Feller process is also interesting in chemical reaction dynamics (cf., for instance, [
35]).
For the Feller-type diffusion process 
, we assume that the total probability mass is conserved in 
 and we denote by 
 the transition pdf of 
 in the presence of a zero-flux condition in the zero state (cf., for instance, Giorno and Nobile [
33]). Moreover, for the process 
, we consider the random variable
      
      which denotes the FPT of 
 from 
 to the continuous boundary 
. The FPT pdf 
 satisfies the first-kind Volterra integral equation (cf., for instance, Fortet [
36]):
The renewal Equation (3) expresses that any sample path that reaches 
, after starting from 
 at time 
, must necessarily cross 
 for the first time at some intermediate instant 
. Research on the FPT problem for the Feller diffusion process has been carried out by Giorno et al. [
37], Linetsky [
38], Masoliver and Perelló [
39], Masoliver [
40], Chou and Lin [
41], Di Nardo and D’Onofrio [
42], Giorno and Nobile [
43]).
The paper is structured as follows. In 
Section 2, we consider the time-homogeneous Feller process with a zero-flux condition in the zero state. For this process, we analyze the FPT problem through a constant boundary 
S starting from the initial state 
 by determining the Laplace transform of the FPT density and the ultimate FPT probability in the following cases: 
(a)  and 
(b) . In particular, a closed-form expression for the FPT pdf through the zero state is given. Moreover, some connections between the FPT densities of the Feller process and the Wiener and Ornstein–Uhlenbeck processes are investigated. In 
Section 3, making use of the iterative Siegert formula, the first three FPT moments are obtained and analyzed. In 
Section 4, we study the asymptotic behavior of the FPT density when the time-varying boundary 
 moves away from the starting point 
 for large time by distinguishing two cases: 
 is an asymptotically constant boundary and 
 is an asymptotically periodic boundary.
Section 5 is dedicated to the time-inhomogeneous Feller process in the proportional case. Specifically, we assume that 
 is a real function, 
 and 
, with 
. For this case, we determine the closed-form expression of the FPT density through the zero state. Furthermore, for 
 and 
, we obtain the FPT density through a specific time-varying boundary and the related ultimate FPT probability. Finally, in 
Section 6, an asymptotic exponential approximation is derived for asymptotically constant boundaries.
 Various numerical computations are performed both for the time-homogeneous Feller process and for the time-inhomogeneous Feller-type process to analyze the role of the parameters.
  2. FPT Problem for a Time-Homogeneous Feller Process
We consider the time-homogeneous Feller process 
 with drift 
 and infinitesimal variance 
, defined in the state space 
. As proved by Feller [
44], the state 
 is an exit boundary for 
, a regular boundary for 
 and an entrance boundary for 
. The scale function and the speed density of 
 are (cf. Karlin and Taylor [
45]):
      respectively. In this section, we assume that 
 and suppose that a zero-flux condition is placed in the zero state.
  2.1. Transition Density
When 
, 
 and 
, imposing a zero-flux condition in the zero state, the transition pdf of 
 can be explicitly obtained (cf., for instance, Giorno et al. [
37], Sacerdote [
46]). Indeed, when 
, 
 and 
, the transition pdf is:
        whereas if 
, 
 and 
, one obtains:
        where
        
        denotes the modified Bessel function of the first kind and 
 is Eulero’s gamma function. Here and elsewhere, whenever the multiple-valued functions such as 
 appear, they are assumed to be taken as their principal branches. We note that the transition pdf 
 in (5) and (6) satisfies the following relation:
Moreover, when 
, 
 and 
, the time-homogeneous Feller process allows a steady-state density:
        which is a gamma density of parameters 
 and 
. In the sequel, we denote by
        
        the Laplace transform (LT) of the function 
.
  2.2. Laplace Transform of the Transition Density
By performing the LT to (5) and (6), for 
 one has (cf. Giorno et al. [
37], Chou and Lin [
41]):   
        where
        
        denotes the modified Bessel function of the second kind (cf. Gradshteyn and Ryzhik [
47], p. 928, no. 8.485) and
        
        are the Kummer’s functions of the first and second kinds, respectively (cf. Gradshteyn and Ryzhik [
47], p. 1023, no. 9.210.1 and no. 9.210.2). Kummer’s functions satisfy the following relations (cf. Tricomi [
48]):
        and
        
        where
        
        denotes the incomplete gamma function. By performing the Laplace transform to both sides of (8), the following result is obtained:
  2.3. Laplace Transform of the FPT Density
An analytic approach to analyze the FPT problem through a non-negative constant boundary 
 is based on the Laplace transform. Indeed, from (3), one has:
        so that the LT of the FPT pdf 
 can be evaluated by knowing the LT of the transition pdf 
.
To determine  via (17), we consider the following cases: (a)  and (b) .
        
- (a)
- FPT downwards for the time-homogeneous Feller process 
For 
, by virtue of (16) and (17), one has:
Then, making use of (10) in (18), for 
, one obtains:
From (19), one derives the ultimate FPT probability through 
S starting from 
, with 
:
        with the use of (11) and (14). Furthermore, if 
, taking the limit as 
 in (19), for 
, one has:
        where the relation
        
        has been used for 
, whereas the identity
        
        has been applied for 
. From (21), one obtains the ultimate FPT probability through zero state starting from 
, with 
:
        where
        
        denotes the incomplete gamma function.
For 
 and 
, the inverse LT of 
, given in (21), can be explicitly evaluated:
Indeed, since (cf. Erdelyi et al. [
49], p. 283, no. 35)
        
        the start of (23) follows from (21) for 
. Moreover, for 
 making use of the first of (14) in (21) and recalling that (cf. Tricomi [
48], p. 90)
        
        the second part of (23) is obtained.
In 
Figure 1, the FPT pdf 
, given in (23), is plotted as function of 
t for some choices of 
 and 
r, with 
.
       
- (b)
- FPT upwards for the time-homogeneous Feller process 
By virtue of (10), from (17), for 
, one has
        
        whereas for 
 and 
, it results that:
From (24) and (25), one derives that the first passage through 
S starting from 
 is a sure event, i.e.,
        
  2.4. Relations between the FPT Densities for the Feller and the Wiener Processes
The FPT pdf  for the time-homogeneous Feller process can be explicitly obtained for  and  or for  and , as proved in Proposition 1 and in Proposition 2, respectively. Moreover, in these cases, there is a relationship between the FPT pdf of Feller process and the FPT pdf of the standard Wiener process.
Proposition 1. Let  be a time-homogeneous Feller diffusion process, having  and , with a zero-flux condition in the zero state.
- If , one has: - and . 
- If , one obtains: - and . 
 Proof.  We assume that 
 and 
. In this case, the zero state is a regular reflecting boundary. Making use of the relations (cf. Abramowitz and Stegun [
50], p. 443, no. 10.2.14 and p. 444, no. 10.2.17)
          
          from (19), (21), (24) and (25) with 
 and 
, it follows that:
          
When 
, the right-hand side of (30) identifies with the LT 
 of the FPT pdf 
 through 
 for a standard Wiener process originated in 
. Hence, for 
 and 
, one has
          
          from which (27) follows. Instead, for 
, the right-hand side of (30) is the LT 
 of the FPT pdf 
 through 
 for a standard Wiener process, starting from 
, restricted to 
 with 0 reflecting boundary (cf., for instance, Giorno and Nobile [
3]). Then, for 
 and 
, one obtains:
          
          from which (28) follows. The alternative expression (29) is derived by performing the inverse LT to the second expression in (30) and by using formula 33.149, p. 190 in Spiegel et al.’s work [
51].    □
 We note that by setting  in (27) we obtain (23) with  and .
In 
Figure 2, the FPT pdf (28) is plotted as function of 
t for 
, 
 and various choices of parameters 
r and 
S.
Proposition 2. Let  be a time-homogeneous Feller diffusion process, having  and , with a zero-flux condition in the zero state.
 Proof.  We assume that 
 and 
. In this case, the zero state is an entrance boundary. Making use of the relations (cf. Abramowitz and Stegun [
50], p. 443, no. 10.2.13 and p. 444, no. 10.2.17)
          
          from (19), (24) and (25) with 
 and 
, it follows that:
          
We note that when 
, the right-hand side of (36) identifies with the LT 
 of the function 
, where 
 is the FPT pdf through 
 of a standard Wiener process originated in 
. Hence, for 
 and 
, one has
          
          that leads to (32). Instead, for 
 the right-hand side of (36) is the LT 
 of the function 
, where 
 is the first-exit time pdf through 
 for a standard Wiener process, starting from 
, defined in 
 with 0 absorbing boundary (cf., for instance, Giorno and Nobile [
3]). Then, for 
 and 
, one has
          
          from which (32) follows. The alternative expression (33) can be obtained by performing the inverse LT to the second expression in (36) and by using formula 33.148, p. 190 in Spiegel et al. [
51] (by changing the sign). Finally, (34) and (35) follow by taking the limit as 
 in (32) and (33), respectively.    □
 In 
Figure 3, the FPT pdf (32) is plotted as function of 
t for 
, 
 and various choices of parameters 
r and 
S. We note that, due to the different nature of the zero state, the peaks of FPT densities of 
Figure 3 are more pronounced with respect to those of 
Figure 2.
  2.5. Relations between the FPT Densities for the Feller and the Ornstein–Uhlenbeck Processes
For  and  or  and , the FPT pdf  of the Feller process can be related to the FPT pdf of the Ornstein–Uhlenbeck process.
Proposition 3. Let  be a time-homogeneous Feller diffusion process, having  and  (), with a zero-flux condition in the zero state.
 Proof.  Let 
 and 
. We assume that the state 
 is a regular reflecting boundary. Recalling that (cf. Tricomi [
48], p. 219, no. (1)):
          
          for 
 from (19) one obtains (37). Furthermore, for 
 and 
, from (21) with 
 and 
, making use of (40), we have
          
Equation (41) identifies with (37) for 
, being (cf. Tricomi [
48], p. 221, no. (9)):
          
Since (cf. Tricomi [
48], p. 234, no. 15 and p. 235, no. 18):
          
          by setting 
 in (37), one obtains (38).
Instead, for , from (24) and (25), with  and , one immediately obtains (39). Consequently, by setting  and making use of the second expression in (13), it follows that .    □
 We note that, for 
, the right-hand side of (37) identifies with the LT 
 of the FPT pdf 
 from 
 through 
 for the Ornstein–Uhlenbeck process with infinitesimal drift 
 and infinitesimal variance 
. Hence, for 
 and 
 from (37) one has:
Furthermore, for 
 the right-hand side of (39) is the LT 
 of the FPT pdf 
 from 
 to 
 for the Ornstein–Uhlenbeck process with infinitesimal drift 
 and infinitesimal variance 
, defined in 
, with 0 reflecting boundary. Therefore, for 
 and 
 from (39), one obtains:
For 
 and 
, relations (44) and (45) show that the FPT density of the Feller process can be also interpreted as the the FPT density of an Ornstein–Uhlenbeck process, that is known only when 
. Therefore, from (44), one has:
        which identifies with (23) for 
 and 
.
Proposition 4. Let  be a time-homogeneous Feller diffusion process, having  and  (), with a zero-flux condition in the zero state.
- If , one has: 
- If , one obtains: - and . 
 Proof.  Let 
 and 
, so that the state 
 is an entrance boundary. For 
, recalling that (cf. Tricomi [
48], p. 219, no. (2))
          
          from (19), with 
 and 
, one obtains (46). Moreover, making use of relation 
 and of (43), one has
          
          so that, by setting 
 in (46), one obtains (47).
Instead, for  from (24) and (25), with  and , Equation (48) immediately follows. Finally, by setting  in (48) and making use of the second expression in (13), one has .    □
 For 
, we note that the right-hand side of (46) identifies with the LT 
 of 
, where 
 is the FPT pdf from 
 through 
 for the Ornstein–Uhlenbeck process with infinitesimal drift 
 and infinitesimal variance 
. Hence, for 
 and 
 one has:
For  and , Equation (51) shows that a functional relationship between the FPT densities of the Feller and Ornstein–Uhlenbeck processes exists.
  4. Asymptotic Behavior of the FPT Density for the Time-Homogeneous Feller Process
In 
Section 2 and 
Section 3, we analyzed the FPT problem for a time-homogeneous Feller process and we assumed that the boundary 
S is constant. Nevertheless, the inclusion of a time-varying boundary 
 is often useful to model various aspects of the time varying behavior of dynamic systems.
Let 
, with 
, where 
 denotes the set of continuously differentiable functions on 
. For a time-homogeneous diffusion process, having drift 
 and infinitesimal variance 
, the FPT pdf 
 is the solution of the second-kind non-singular Volterra integral equation (cf. Buonocore [
8]):   
      with 
 if 
 and 
 if 
, and where
      
The knowledge of the transition pdf 
 of the considered diffusion process allows the formulation of effective numerical procedures to obtain 
 via (57) (cf., for instance, Buonocore et al. [
8], Di Nardo et al. [
10]).
For the Feller process, having 
 and 
, with a zero-flux condition in the zero state, recalling (5) and (6), for 
 from (58), one obtains:
      where the relation (cf. Gradshteyn and Ryzhik [
47], p. 928 no. 8.486.4)
      
      has been used.
Let . We focus our analysis on the asymptotic behavior of the FPT pdf for the Feller diffusion process, with ,  and , by considering separately two cases:  is an asymptotically constant boundary and  is an asymptotically periodic boundary.
  4.1. Asymptotically Constant Boundary
We consider the FPT problem for the Feller process through the asymptotically constant boundary
        
        with 
, where 
 is a bounded function that does not depend on 
S, such that
        
Since 
, the function 
 approaches a constant value as 
. Making use of (60), for 
, one has:
        where (9) has been used. From (57), for 
 and for large times the FPT density exhibits an exponential behavior (cf. Nobile et al. [
12]). Specifically, for 
 and 
, one has:
The goodness of the exponential approximation increases as the boundary progressively moves away from the starting point.
We now assume that the boundary 
 is constant, i.e., 
. By virtue of (53) for 
, with 
 and 
 defined in (4), and recalling (63), for 
 and 
 one has
        
        implying that for 
 the FPT mean can be approximated by 
 for large values of 
S. Furthermore, by virtue of (64), for 
 and 
, one obtains:
In 
Table 2, the FPT moments 
 and their exponential approximations 
, with 
, are listed for increasing values of the boundary 
, showing a good degree of precision in the approximations. We emphasize that the exponential approximation of the FPT density (64) provides the growth trend of the FPT moments (65) for large constant boundaries 
S. Moreover, the goodness of the approximation depends on the parameters of the process that determine the exact shape of the FPT pdf.
  4.2. Asymptotically Periodic Boundary
We consider the FPT problem for the Feller process through an asymptotically periodic boundary 
, with 
, where 
 is a bounded function, that does not depend on 
S, such that
        
        with 
 being a periodic function of period 
 satisfying the condition:
Since 
, the function 
 approaches a periodic function as 
. Indeed, making use of (60) and recalling (9), for 
, one obtains:
By virtue of (57), for 
 and for large times, the FPT density shows a non-homogeneous exponential behavior. Specifically, for 
 and 
, one has:
Hence, for , the FPT pdf of the Feller process through an asymptotically periodic boundary exhibits damped oscillations taking the form of a sequence of periodically spaced peaks whose amplitudes exponentially decrease.
  6. Asymptotic Behavior of the FPT Density for a  Time-Inhomogeneous Feller-Type Process
In the following proposition, we prove that the FPT density  of the process (69), with a zero-flux condition in the zero state, is a solution of a second-kind non-singular Volterra integral equation.
Proposition 8. Let , with . For the time-inhomogeneous Feller-type diffusion process (69), with ,  and , the FPT pdf  is a solution of the integral Equation (57) with  if  and  if , where  Proof.  The FPT pdf 
 of the process 
, with infinitesimal drift 
 and infinitesimal variance 
, with a zero-flux condition in the zero state, is a solution of the following integral equation
        
        where, due to (59) with 
 and 
, one has:
        
Multiplying both-sides of Equation (91) by 
, performing the transformation 
 in the integral and recalling (79), we obtain the integral Equation (57) with
        
Then, (90) follows from (93), making use of (74) and (92).    □
 Let 
. We focus on the asymptotic behavior of the FPT pdf of the Feller-type diffusion process (69), with a zero-flux condition in the zero state, through the asymptotically constant boundary (61), with 
, where 
 is a bounded function, that does not depend on 
S, such that (62) holds. We assume that
      
      so that the process allows a steady-state density. Under such assumptions, from (90), one has:
Finally, by virtue of (57), for 
 and for long periods, the FPT density through the asymptotically constant boundary (61) of the time-inhomogeneous Feller-type process (69) exhibits the following exponential behavior:
  7. Conclusions
In this paper, we have considered the first-passage time problem for a Feller-type diffusion process, having infinitesimal drift 
 and infinitesimal variance 
, defined in 
, with 
, 
, 
 continuous functions. In 
Section 2, for the time-homogeneous process, we have determined the Laplace transform of the downwards and upwards FPT densities. In Propositions 1 and 2, some connections between the FPT densities for the Feller and the Wiener processes (
) have been discussed, whereas in Propositions 3 and 4 we have analyzed some relations between the FPT densities for Feller and Ornstein–Uhlenbeck processes (
). Furthermore, in 
Section 3, the FPT moments have been investigated by using the Siegert formula. In 
Section 4, for 
, the asymptotic behavior of the FPT density through a time-dependent boundary has been discussed for an asymptotically constant boundary and for an asymptotically periodic boundary. Furthermore, the first three moments of FPT density through a constant boundary have been compared with the corresponding asymptotic approximations. 
Section 5 is dedicated to a time inhomogeneous Feller-type diffusion process with 
, for 
. In Propositions 6 and 7, the FPT density has been obtained for an exponential time-varying boundary. The FPT densities have been plotted for periodic noise, showing the presence of damped oscillations having the same periodicity as the noise intensity. In 
Section 6, a second-kind Volterra integral equation was derived for the FPT density of a time-inhomogeneous Feller-type process through a general time-dependent boundary. Finally, such an equation has been used to derive the asymptotic exponential trend of the FPT pdf through an asymptotically constant boundary.
Analytical, asymptotic and computational methods for the evaluation of FPT densities through time-varying boundaries for more general time-inhomogeneous diffusion processes will be the object of future research focused also on contexts of statistical inference.