Core Concepts
The L2-error rate (1 + s)/2 is essentially optimal for approximating SDEs with a drift coefficient of fractional Sobolev regularity.
Abstract
The article discusses the strong approximation of stochastic differential equations (SDEs) with bounded drift coefficients and Sobolev regularity. It proves that the equidistant Euler approximation achieves an L2-error rate of at least (1 + s)/2. The study focuses on cases where the drift coefficient has fractional Sobolev regularity between 1/2 and 1. The analysis employs techniques like coupling of noise to establish lower error bounds. The results suggest that no numerical method can improve the L2-error rate beyond (1 + s)/2 for bounded drift coefficients in W s,p spaces.
Stats
For every n ∈ N, there exists c ∈ (0, ∞) such that for all n ∈ N, E sup t∈[0,1] |Xt - XEn,t|pi1/p ≤ c n(1+s)/2-ε .
Recently shown that for SDEs with piecewise differentiable drift coefficients and discontinuities, the best possible Lp-error rate achievable is at most 3/4 for all p ∈ [1, ∞).
The Euler approximation XEn,1 achieves an Lp-error rate of at least 1/2 even if the drift coefficient µ is only bounded and measurable.
Quotes
"We employ the coupling of noise technique to bound the L2-error of an arbitrary approximation."
"The estimate naturally leads to questioning whether better Lp-error rates can be achieved."
"Results show that adaptive approximations may outperform non-adaptive methods under certain conditions."