Abstract
Our aim is to construct high order approximation schemes for general semigroups of linear operators \(P_{t},t\ge 0\). In order to do it, we fix a time horizon T and the discretization steps \(h_{l}=\frac{T}{n^{l}},l\in {\mathbb {N}}\) and we suppose that we have at hand some short time approximation operators \(Q_{l}\) such that \(P_{h_{l}}=Q_{l}+O(h_{l}^{1+\alpha })\) for some \(\alpha >0\). Then, we consider random time grids \(\Pi (\omega )=\{t_0(\omega )=0<t_{1}(\omega )<\cdots <t_{m}(\omega )=T\}\) such that for all \(1\le k\le m\), \(t_{k}(\omega )-t_{k-1}(\omega )=h_{l_{k}}\) for some \(l_{k}\in {\mathbb {N}}\), and we associate the approximation discrete semigroup \(P_{T}^{\Pi (\omega )}=Q_{l_{n}}\ldots Q_{l_{1}}\). Our main result is the following: for any approximation order \(\nu \), we can construct random grids \(\Pi _{i}(\omega )\) and coefficients \(c_{i}\), with \(i=1,\ldots ,r\) such that
with the expectation concerning the random grids \(\Pi _{i}(\omega ).\) Besides, \(Card (\Pi _{i}(\omega ))=O(n)\) and the complexity of the algorithm is of order n, for any order of approximation \(\nu \). The standard example concerns diffusion processes, using the Euler approximation for \(Q_l\). In this particular case and under suitable conditions, we are able to gather the terms in order to produce an estimator of \(P_tf\) with finite variance. However, an important feature of our approach is its universality in the sense that it works for every general semigroup \(P_{t}\) and approximations. Besides, approximation schemes sharing the same \(\alpha \) lead to the same random grids \(\Pi _{i}\) and coefficients \(c_{i}\). Numerical illustrations are given for ordinary differential equations, piecewise deterministic Markov processes and diffusions.
Similar content being viewed by others
References
Agarwal, A., Gobet, E.: Finite variance unbiased estimation of stochastic differential equations (2018)
Alfonsi, A., Jourdain, B., Kohatsu-Higa, A.: Pathwise optimal transport bounds between a one-dimensional diffusion and its Euler scheme. Ann. Appl. Probab. 24(3), 1049–1080 (2014)
Alfonsi, A.: High order discretization schemes for the CIR process: application to affine term structure and Heston models. Math. Comput. 79(269), 209–237 (2010)
Alfonsi, A., Hayashi, M., Kohatsu-Higa, A.: Parametrix methods for one-dimensional reflected SDEs. In: Modern Problems of Stochastic Analysis and Statistics, volume 208 of Springer Proceedings in Mathematics and Statistics, pp. 43–66. Springer, Cham (2017)
Andersson, P., Kohatsu-Higa, A.: Unbiased simulation of stochastic differential equations using parametrix expansions. Bernoulli 23(3), 2028–2057 (2017)
Bally, V., Goreac, D., Rabiet, V.: Regularity and stability for the semigroup of jump diffusions with state-dependent intensity. Ann. Appl. Probab. 28(5), 3028–3074 (2018)
Bally, V., Kohatsu-Higa, A.: A probabilistic interpretation of the parametrix method. Ann. Appl. Probab. 25(6), 3095–3138 (2015)
Bally, V., Rey, C.: Approximation of Markov semigroups in total variation distance. Electron. J. Probab. 21, 44 (2016)
Beskos, A., Roberts, G.O.: Exact simulation of diffusions. Ann. Appl. Probab. 15(4), 2422–2444 (2005)
Constantine, G.M., Savits, T.H.: A multivariate Faà di Bruno formula with applications. Trans. Am. Math. Soc. 348(2), 503–520 (1996)
Frikha, N., Kohatsu-Higa, A., Li, L.: On the first hitting times of one dimensional elliptic diffusions. arXiv e-prints arXiv:1609.09327 (2016)
Giles, M.B.: Multilevel Monte Carlo path simulation. Oper. Res. 56(3), 607–617 (2008)
Henry-Labordère, P., Tan, X., Touzi, N.: Unbiased simulation of stochastic differential equations. Ann. Appl. Probab. 27(6), 3305–3341 (2017)
Ikeda, N., Watanabe, S.: Stochastic Differential Equations and Diffusion Processes, vol. 24 of North-Holland Mathematical Library, 2nd edn. North-Holland Publishing Co., Amsterdam (1989)
Jacobsen, M.: Point Process Theory and Applications. Probability and Its Applications. Birkhäuser, Boston (2006)
Kebaier, A.: Statistical Romberg extrapolation: a new variance reduction method and applications to option pricing. Ann. Appl. Probab. 15(4), 2681–2705 (2005)
Kunita, H.: Stochastic Differential Equations and Stochastic Flows of Diffeomorphisms. In: École d’été de probabilités de Saint-Flour, XII—1982, volume 1097 of Lecture Notes in Mathematics, pp. 143–303. Springer, Berlin (1984)
Lemaire, V., Pagès, G.: Multilevel Richardson–Romberg extrapolation. Bernoulli 23(4A), 2643–2692 (2017)
McLeish, D.: A general method for debiasing a Monte Carlo estimator. Monte Carlo Methods Appl. 17(4), 301–315 (2011)
Neveu, J.: Arbres et processus de Galton–Watson. Ann. Inst. H. Poincaré Probab. Stat. 22(2), 199–207 (1986)
Ninomiya, S., Victoir, N.: Weak approximation of stochastic differential equations and application to derivative pricing. Appl. Math. Finance 15(1–2), 107–121 (2008)
Oshima, K., Teichmann, J., Velušček, D.: A new extrapolation method for weak approximation schemes with applications. Ann. Appl. Probab. 22(3), 1008–1045 (2012)
Pagès, G.: Multi-step Richardson–Romberg extrapolation: remarks on variance control and complexity. Monte Carlo Methods Appl. 13(1), 37–70 (2007)
Rhee, C.-H., Glynn, P.W.: Unbiased estimation with square root convergence for SDE models. Oper. Res. 63(5), 1026–1043 (2015)
Vihola, M.: Unbiased estimators and multilevel Monte Carlo. Oper. Res. 66(2), 448–462 (2018)
Acknowledgements
Aurélien Alfonsi benefited from the support of the “Chaire Risques Financiers”, Fondation du Risque.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Technical results for the variance analysis
Technical results for the variance analysis
We introduce some notation. We consider smooth random fields, that is functions \(\varphi :\Omega \times {\mathbb {R}}^{d}\rightarrow {\mathbb {R}}^{d}\) which are measurable with respect to \((\omega ,x)\) and such that, for each \(\omega ,\) the function \(x\mapsto \varphi (\omega ,x)\) is of class \(C^{\infty }({\mathbb {R}}^{d})\). For such a random field we denote
Moreover, we will say that a sequence of random fields \(\varphi _{i},i=1,\ldots ,m\) are independent if there are some independent \(\sigma -\) algebras \({\mathcal {G}}_{i},i=1,\ldots ,m\) such that \(\varphi _{i}\) is \({\mathcal {G}}_{i}\otimes {\mathcal {B}}({\mathbb {R}}^{d})\) measurable. We will use this property as follows. Suppose that \(\Phi \) is \({\mathcal {G}}_{m}\otimes {\mathcal {B}}({\mathbb {R}}^{d})\) measurable and \(\Psi \) and \(\Theta \) are \(\vee _{i=1}^{m-1}{\mathcal {G}}_{i}\otimes {\mathcal {B}}({\mathbb {R}}^{d})\) measurable. Then, for every \(x\in {\mathbb {R}}^{d}\) and every \(p\ge 1\)
In the sequel we consider a sequence of smooth random fields \(\Phi _{i}:\Omega \times {\mathbb {R}}^{d}\rightarrow {\mathbb {R}}^{d}\) and \(\phi _{i}:\Omega \times {\mathbb {R}}^{d}\rightarrow {\mathbb {R}}^{d}\) , \(i\in {\mathbb {N}}\) and moreover, a vector field \(\varphi :\Omega \times {\mathbb {R}}^{d}\rightarrow {\mathbb {R}}^{d}\). We assume that \(\varphi \) and \((\Phi _{j},\phi _{j}),j\in {\mathbb {N}}\) are independent. We fix \(r\in {\mathbb {N}}\) and, for a set \( \Lambda \subset \{1,\ldots ,r\},\) we define
Moreover, given a multi-index \(\alpha ,\) we define
Proposition A.1
Suppose that for every \(p,q\in {\mathbb {N}}\), there exists \(C_{q,p}\) such that
Then, for every \(p\ge 1\) and every multi-index \(\alpha \) we have
for some C depending on r, \(\left| \alpha \right| \) and \(C_{\left| \alpha \right| +r,2|\alpha | p}.\)
Remark A.2
This proposition says the following: if at each step the error is of order \(\delta _{i}=\left\| \Phi _{i}-\phi _{i}\right\| _{q,p^{\prime },\infty }\), then after r steps we have an error of order \(\delta _{1}\times \cdots \times \delta _{r}.\) This may seem a little surprising, and one may have expected an error of order \(\delta _{1}+\cdots +\delta _{r}\), but this is due to the way how terms are summed with \(\sum _{\Lambda \subset \{1,\ldots ,r\}}(-1)^{Card ( \Lambda )}.\)
Proof
Step 1. We use the Faá di Bruno formula \(\partial ^{\alpha }[f\circ g]=\sum _{\left| \beta \right| \le \left| \alpha \right| }(\partial ^{\beta }f)(g)P_{\alpha ,\beta }(g)\) (see (8)) and the inequality between geometric and arithmetic means to upper bound the terms \(\vert \prod _{i=1}^{k}\partial ^{\gamma _{i}} g^{j_i} \vert \) defining \(P_{\alpha ,\beta }(g)\). We then obtain for random functions g
Besides, for two random fields \(g_{1}\) and \(g_{2}\), we write
Using the inequality between geometric and arithmetic means for the product on \(i\not =j\) and then the Cauchy–Schwarz inequality, we get
Step 2. We prove (84) for \(r=1.\) In this case \(\Lambda =\varnothing \) or \(\Lambda =\{1\}\) so that
with
and
Using (83) and the fact that \(\varphi \) is independent of \(\lambda \Phi _{1}+(1-\lambda )\phi _{1}\) we get (see (81))
so that \(\left\| A_{\beta }\right\| _{0,p,\infty }\le C\left\| \Phi _{1}-\phi _{1}\right\| _{0,p,\infty }.\) Moreover, using again (81 ) first and then (85), we get
so (84) is proved for \(r=1\).
Step 3. Suppose (84) is true for \(r-1,\) for every \(\alpha \) and every \(p\ge 1.\) We prove it for r. We do it first for \(\alpha =\varnothing \) (without derivatives) because it is simpler. We write
Note that we have made a slight abuse of notation here: the notation \(\theta _{(r-1)}^{\Lambda ^{\prime }}\) is used in fact for \(\theta ^{\Lambda '}_r \circ \dots \circ \theta ^{\Lambda '}_2\), not for \(\theta ^{\Lambda '}_{r-1} \circ \dots \circ \theta ^{\Lambda '}_1\). Since \((\lambda \Phi _{1}+(1-\lambda )\phi _{1})(x)\) is independent of \( \Gamma _{r-1}^{(i)}\varphi ,\) we have from (81)
Then, by using the induction hypothesis, we get
We prove now (84) for a general multi-index \(\alpha \) and make the same abuse of notation for \(\theta ^{\Lambda '}\). Using (8) and (9) for \(f=\varphi (\theta _{(r-1)}^{\Lambda ^{\prime }})\) and \(g_{1}=\Phi _{1},g_{2}=\phi _{1}\) we obtain
with
and
By assumption \(\theta _2,\dots ,\theta _r\) are independent of \((\Phi _{1},\phi _{1})\). Therefore, \(\Gamma _{r-1}^{\beta }\varphi (x)\) is independent of \((\Phi _{1},\phi _{1})\). We use (81) first and then the induction hypothesis and (85) to obtain
Moreover
Notice that \(\partial ^{i}\Gamma _{r-1}^{\beta }\varphi =\Gamma _{r-1}^{(\beta ,i)}\varphi \). Using again (81) and the recurrence hypothesis, we get
\(\square \)
Lemma A.3
Let \((X_t(x))_{t\ge 0} \) denote the flow of the SDE (1) and \({\hat{X}}_t(x)=x+b(x)t+\sigma (x)W_t\) the flow of the Euler scheme. We assume that b and \(\sigma \) are \(C^\infty \), bounded and with bounded derivatives. Then, we have
with \(a=2\) if \(\sigma =0\), \(a=3/2\) if \(\sigma (x)\) is a constant function and \(a=1\) in the general case.
Proof
We show this result by induction on q. We only focus on the general case, the cases \(\sigma =0\) or \(\sigma (x)\) constant can be then easily deduced. For \(q=0\), this result is stated for example in Proposition 1.2 [2]. For simplicity of notation, we do the proof in dimension \(d=1\) with \(b=0\). We note \(\sigma ^{(q)}\) the qth derivative of \(\sigma \). For \(q=1\), we have \({\hat{X}}_t^{(1)}(x)=1+\sigma ^{(1)}(x)W_t\) and
Since \(\sigma ^{(1)}\) is bounded, we have \(\forall t>0, \sup _{x}{\mathbb {E}}[\sup _{s\in [0,t]} |X_s^{(1)}(x)|^p]<\infty \). We write
Since \(\sigma ^{(1)}\) is bounded and Lipschitz, we get by using the Burkholder–Davis–Gundy inequality and then Jensen inequality
with a constant C that does not depend on x. We check then again with the BDG inequality that \({\mathbb {E}}[|X_s(x)-x|^p]\le Cs^{p/2}\) since \(\sigma \) is bounded and \({\mathbb {E}}[|X_s^{(1)}(x)-1|^p]\le Cs^{p/2}\) since \(\sigma ^{(1)}\) is bounded and (86). Thus, we have \({\mathbb {E}}[|{\hat{X}}_t^{(1)}(x)- X_t^{(1)}(x)|^p]\le C t^p\).
We suppose now the result true for \(q-1\in {\mathbb {N}}^*\) and that we have shown that
for each p, with a constant C that does not depend on x. We have \({\hat{X}}_t^{(q)}(x)=\sigma ^{(q)}(x)W_t,\) and by the Faá di Bruno formula
with \(A_t=\sum _{m_1+\dots + q m_q=q, m_1\not = q,m_q\not =0 }c_{m_1,\dots ,m_q} \prod _{k=1}^q( X^{(k)}_t(x))^{m_k} \sigma ^{(m_1+\dots +m_q)}(X_t(x))\). Note that in this sum is equal to 0 for \(q=2\) and otherwise there is at least one \(k\in \{2,\dots ,q-1\}\), such that \(m_k\ge 1\). This gives \({\mathbb {E}}[|A_t|^p]\le Ct^{p/2}\) by using the induction hypothesis (87) and Hölder type inequalities. Since \(X^{(q)}_0(x)=0\), \(\sigma ^{(1)}\) and \( \sigma ^{(q)}\) are bounded and \(\sup _{x}{\mathbb {E}}[\sup _{s\in [0,t]} |X_s^{(1)}(x)|^p]<\infty \) for any p, we get \({\mathbb {E}}[|X_t^{(q)}(x)|^p]\le Ct^{p/2}\) by using BDG and Gronwall inequalities. Therefore, \({\tilde{A}}_t=A_t+ X^{(q)}_t(x) \sigma ^{(1)}(X_t(x))\) also satisfies \({\mathbb {E}}[|{\tilde{A}}_t|^p]\le Ct^{p/2}\).
We now repeat the same arguments as for \(q=1\): from
we get \({\mathbb {E}}[|{\hat{X}}_t^{(q)}(x)- X_t^{(q)}(x)|^p]\le C t^p\).
\(\square \)
Rights and permissions
About this article
Cite this article
Alfonsi, A., Bally, V. A generic construction for high order approximation schemes of semigroups using random grids. Numer. Math. 148, 743–793 (2021). https://doi.org/10.1007/s00211-021-01219-2
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00211-021-01219-2