Exact simulation of the first-passage time of SDEs to time-dependent thresholds

Devika Khurana, Sascha Desmettre, Evelyn Buckwar · 2024

Read Exact simulation of the first-passage time of SDEs to time-dependent thresholds by Devika Khurana, Sascha Desmettre, Evelyn Buckwar (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-13060.txt.

Introduction

Khurana, Desmettre, and Buckwar work on a math problem that asks how long a random process takes to first reach a set limit. In many real settings that limit moves over time rather than staying fixed. Their 2024 paper extends an exact computer method that worked for fixed limits so that it can handle moving ones.

Rather than tracing the whole random path step by step, the method draws on the known behavior of simple random motion. A correction based on a classic change-of-measure rule then decides whether each draw fits the harder process. The authors test speed and accuracy against older methods through worked examples. They also show a use in predicting when a nerve cell will fire.

Catalog record

CREATORDevika Khurana, Sascha Desmettre, Evelyn Buckwar
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-13060

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Exact simulation of the first-passage time of SDEs to time-dependent thresholds Devika Khurana, Sascha Desmettre, Evelyn Buckwar Abstract The first-passage time (FPT) is a fundamental concept in stochastic processes, representing the time it takes for a process to reach a specified threshold for the first time. Often, considering a time-dependent threshold is essential for accurately modeling stochastic processes, as it provides a more accurate and adaptable framework. In this paper, we extend an existing Exact simulation method developed for constant thresholds to handle time-dependent thresholds. Our proposed approach utilizes the FPT of Brownian motion and accepts it for the FPT of a given process with some probability, which is determined using Girsanov’s transformation. This method eliminates the need to simulate entire paths over specific time intervals, avoids time-discretization errors, and directly simulates the first-passage time. We present results demonstrating the method’s effectiveness, including the extension to time-dependent thresholds, an analysis of its time complexity, comparisons with existing methods through numerical examples, and its application to predicting spike times in a neuron. Keywords— First-passage time, Exact simulation, Stochastic differential equations, Neuroscience MSC 2010 Classification —- 37M05, 65C30, 60G05, 60H35, 68Q87, 92C20 1 Introduction Dynamical systems are frequently modeled using one-dimensional diffusion processes in various fields. Understanding their behavior involves examining the time it takes for the diffusion to reach a specified threshold. The time at which the diffusion reaches the threshold for the first time is known as the first-passage time. For instance, in financial markets, the pricing of barrier options, cf. [ 1 ] , depends on whether the system crosses the threshold before the option’s expiry time. In oncology, first-passage time analysis is critical for assessing treatment efficacy. A notable application is the study of tumor growth delay [ 43 ] , where first-passage time metrics are used to compare the time required for a tumor to reach a certain volume (threshold) with and without the treatment. Additionally, the theory of first-passage time extends to diverse domains, including physics [ 19 ] , reliability engineering [ 11 ] , and neuroscience. In our focus on the neuroscience model [ 32 ] , the diffusion process describes the evolution of the membrane voltage of a neuron. When the membrane voltage reaches a certain threshold, it generates a spike, and the information is transmitted to the next neuron. The time at which the membrane voltage hits the threshold is known as the spike time, which can be modeled as the first-passage time. The corresponding mathematical problem can be formulated as follows: Let \left(\Omega,\mathcal{F},\mathbb{P}\right) be a probability space with filtration \mathbb{F}=\{\mathcal{F}_{t},t\geq 0\} . We consider the stochastic process \{Y_{t},t\geq 0\} adapted to the filtration \mathbb{F} as a solution to a time-homogeneous Itô stochastic differential equation: dY_{t}=\mu\left(Y_{t}\right)dt+\sigma\left(Y_{t}\right)dB_{t},\hskip 28.45274ptY_{0}=y_{0}\in\mathbb{R}\,, (1) where \{B_{t},t\geq 0\} is the standard one-dimensional Brownian motion adapted to \mathbb{F} . Assumption 1.1 The drift \mu and diffusion \sigma of SDE described in ( 1 ) are defined in a manner that ensures the existence of a unique strong solution. We denote \tau_{\beta} as the first-passage time of the diffusion process Y to a time-dependent deterministic threshold \beta\left(t\right) , where \beta:\mathbb{R}^{+}_{0}\rightarrow\mathbb{R} . The first-passage time \tau_{\beta} , inherits randomness from Y and thus is a random variable defined as follows: \tau_{\beta}:=\inf\{t\geq 0:Y_{t}=\beta(t)\}. (2) Due to randomness inherited by \tau_{\beta} , path dependence in the process Y , and the dynamic nature of the threshold function over time, determining the first-passage time can be tedious. The existing methods for evaluating the first-passage time \tau_{\beta} , based on the dynamics in which this time is obtained, can be classified as follows: \left(i\right) Explicit expressions of probability density function for the first-passage time of a stochastic differential equation to a time-dependent threshold function, or even to a constant one, are exceptionally rare. When available, these expressions often correspond to basic mathematical models that often lack direct applicability [ 29 ] . \left(ii\right) Approximation of the probability density function of the first-passage time. One prominent approach is to use the Fokker-Planck equations [ 37 ] . The probability that the process Y reaches the threshold \beta\left(t\right) before T is given by \phi_{y}=F\left(y,T\right) , where the function F\left(y,T\right) satisfies a Fokker-Planck equation. Given the complexity of such equations, exact solutions are rarely feasible. A novel strategy as explored in [ 5 ] involved transforming the space-time domain to \left(0,1\right)\times\left(0,1\right) and using a fast-converging series expansion for the first-passage time distribution to simplify the Fokker-Planck equation. More work in this direction can be found in [ 6 , 40 ] . Another noteworthy approach focuses on finding the transitional density g(\beta(\tau_{\beta}),\tau_{\beta}|y_{0},t_{0}) of the FPT \tau_{\beta} . This density is the solution to a Volterra-type integral equation [ 15 , 48 ] . However, the kernel of this equation is weakly singular, making numerical approximation inherently unstable [ 15 ] . Additionally, one can only compute the solution of the equation for one initial value of the diffusion process at a time. Relevant studies on this topic include [ 17 , 47 , 42 ] . An alternative approach is the Method of Images [ 12 ] . This method involves the construction of a measure A using a linear combination of simpler measures. This construction facilitates an approximate threshold for which the first-passage time distribution can be computed [ 49 ] . Another approach is transforming the pair of SDE and threshold to another pair for which the pdf of first-passage time can be easily computed [ 10 , 8 , 9 ] . \left(iii\right) Approximation of the value of the first-passage time can be obtained by simulating the path of the process Y and then determining the time at which the simulated path crosses \beta\left(t\right) . The path of Y can be simulated using various time-discretisation schemes. The most classical one is the Euler-Maruyama method, which is a widely used approach due to its simplicity in implementation. However, it tends to overestimate the first-passage time if the step size is not sufficiently small [ 7 ] . Furthermore, reducing the step size to decrease error leads to increased variance. That is why there is always a challenge to balance the time-discretization and estimation errors. One improvement was introduced in [ 35 ] , where the probability that the process Y crosses the threshold within a time step was established, and later Gobet in [ 20 ] proved that the method has a o\left(1/2\right) time discretization error, compared to the o\left(1\right) error of the classical Euler-Maruyama method. Other notable improvements can be found in [ 28 , 21 , 18 ] . Another technique, the random walk on spheres introduced by Muller [ 36 ] , focuses on a multi-dimensional Brownian motion Z . It aims to determine the first exit time of Z from its domain D . This method begins by constructing the largest sphere centered at initial value Z_{0} contained within D . A point is then chosen uniformly from the boundary of this sphere to serve as the hitting position of the sphere, denoted z_{1} . Next, a new sphere is constructed, centered at z_{1} and contained within D . The exit time from this new sphere is determined by uniformly selecting a point

Read the complete text