1 Correction to: J Theor Probab (2019) 32:1306–13 https://doi.org/10.1007/s10959-018-0814-4
2 Corrections
In this note, we correct claims made in [2]:
-
(i)
It is claimed that the generalized martingale problem introduced in [2] allows explosion in a continuous manner. However, because the cemetery \(\Delta \) is added to \(\mathbb {B}\) as an isolated point, explosion can only happen by a jump and is excluded by [2, Lemma 4.3]. In Sect. 2, we explain how the setup can be adjusted to include the possibility of explosion.
-
(ii)
In the proof of [2, Proposition 4.8], it is needed that the operator \(A\) has a non-empty resolvent set \(\rho (A)\), i.e., that
$$\begin{aligned} \rho (A) \triangleq \big \{ \lambda \in \mathbb {R}:(\lambda - A)^{-1} \text { exists in } L (\mathbb {B}, \mathbb {B})\big \} \not = \emptyset . \end{aligned}$$This assumption is missing in [2]. It is, e.g., satisfied in case \(A\) is the generator of a \(C_0\)-semigroup; see [4, Remark 1.1.3, Proposition 1.2.1].
3 A Setup Including Explosion
3.1 Modified Setup
In the following, we explain how \(\Omega , \tau _n\) and \(\tau _\Delta \) have to be redefined such that the setting includes the possibility of explosion.
For a function \(\omega :\mathbb {R}_+ \rightarrow \mathbb {B}_\Delta \), we define
where, as always, \(\inf (\emptyset ) \triangleq \infty \). Let \(\Omega \) to be the space of all right continuous functions \(\omega :\mathbb {R}_+ \rightarrow \mathbb {B}_\Delta \) which are càdlàg on \([0, \tau _\Delta (\omega ))\) and satisfy \(\omega (t) = \Delta \) for all \(t \ge \tau _\Delta (\omega )\). The difference in comparison with the setting in [2] is that \(\omega \in \{\tau _\Delta < \infty \}\) might not have a left limit at \(\tau _\Delta (\omega )\).
Denote by \(X\) the coordinate process, i.e., \(X_t(\omega ) = \omega (t)\) for all \(\omega \in \Omega \) and \(t \in \mathbb {R}_+\), and denote by \(\mathcal {F} \triangleq \sigma (X_t, t \in \mathbb {R}_+)\) the \(\sigma \)-field generated by \(X\). The proof of the following is given in Sect. 2.2.
Lemma 1
There exists a metric \(d_\Omega \) on \(\Omega \) such that \((\Omega , d_\Omega )\) is separable and complete and \(\mathcal {F}\) is the corresponding Borel \(\sigma \)-field.
Let \(\mathbf {F}= (\mathcal {F}_t)_{t \ge 0}\) be the filtration generated by \(X\), i.e. \(\mathcal {F}_t \triangleq \sigma (X_s, s \in [0, t])\) for \(t \in \mathbb {R}_+\). Note that \(\tau _\Delta \) is an \(\mathbf {F}\)-stopping time, because \(\{\tau _\Delta \le t\} = \{X_t = \Delta \}\in \mathcal {F}_t\). For \(\Gamma \subseteq \mathbb {B}\), we define
The proof of the following is given in Sect. 2.3.
Lemma 2
-
(i)
If \(\Gamma \subseteq \mathbb {B}\) is closed, then \(\tau (\Gamma )\) is an \(\mathbf {F}\)-stopping time.
-
(ii)
If \(\Gamma _1 \subseteq \Gamma _2 \subseteq \Gamma _3 \subseteq \cdots \) is an increasing sequence of open sets in \(\mathbb {B}\) such that \(\bigcup _{n \in \mathbb {N}} \Gamma _n = \mathbb {B}\), then \(\tau (\mathbb {B}\backslash \Gamma _n) \nearrow \tau _\Delta \) as \(n \rightarrow \infty \).
We define
By Lemma 2, \((\tau _n)_{n \in \mathbb {N}}\) is a sequence of \(\mathbf {F}\)-stopping times satisfying \(\tau _n \nearrow \tau _\Delta \) as \(n \rightarrow \infty \). In this modified setting, the GMP can be defined as in [2] and all results from [2] hold. In Sect. 3, we comment on necessary changes in the proofs.
3.2 Proof of Lemma 1
We adapt the proof of [1, Lemma A.7]. Define
where \(\omega _\Delta (t) = \Delta \) for all \(t \in \mathbb {R}_+\). For \(z \in [0, \infty ]\) and \(t \in \mathbb {R}_+\), we define
Moreover, we define \(\iota :\Omega \rightarrow \Omega ^\star \) by
Lemma 3
\(\iota \) is a bijection.
Proof
To check the injectivity, let \(\omega , \alpha \in \Omega \) be such that \(\iota (\omega ) = \iota (\alpha )\). In case \(\tau _\Delta (\omega ) = \tau _\Delta (\alpha ) \in \{0, \infty \}\), we clearly have \(\omega = \alpha \). In case \(0< \tau _\Delta (\omega ) = \tau _\Delta (\alpha ) < \infty \), we can deduce from the first coordinates of \(\iota (\omega )\) and \(\iota (\alpha )\) that \(\omega = \alpha \) on \([0, \tau _\Delta (\omega )) = [0, \tau _\Delta (\alpha ))\), which implies \(\omega = \alpha \).
To check the surjectivity, note that \( \iota (\omega _\Delta ) = (\omega _\Delta , 0) \) and that \( \iota ( \omega \circ \phi ^{-1}_{t} \mathbf {1}_{[0, t)} + \Delta \mathbf {1}_{[t, \infty )}) = (\omega , t) \) for all \((\omega , t) \in D(\mathbb {R}_+, \mathbb {B}) \times (0, \infty ]\). \(\square \)
Let \(d_D\) be the Skorokhod metric on \(D(\mathbb {R}_+, \mathbb {B}_\Delta )\) and let \(d_{[0, \infty ]}\) be the arctan metric on \([0, \infty ]\). We define
for \((\omega , t), (\alpha , s) \in D(\mathbb {R}_+, \mathbb {B}_\Delta ) \times [0, \infty ],\) and set
We note that \((\Omega ^\star , d_{\Omega ^\star })\) is separable and complete, because it is a \(G_\delta \) subspace of \((D(\mathbb {R}_+, \mathbb {B}_\Delta ) \times [0, \infty ], d_{D \times [0, \infty ]})\). Due to Lemma 3, we can equip \(\Omega \) with the metric
for \(\omega , \alpha \in \Omega .\) In this case, \(\iota \) is an isometry and \((\Omega , d_\Omega )\) is separable and complete. In the following, we equip \(\Omega \) with the topology induced by the metric \(d_{\Omega }\).
We now prove that \(\mathcal {F} = \mathcal {B}(\Omega )\). By the definition of the metric \(d_\Omega \), the maps
are continuous. For fixed \(t \in \mathbb {R}_+\), the map \([0, \infty ] \ni z \mapsto \phi ^{-1}_z(t) \in \mathbb {R}_+\) is Borel and, consequently, also
is Borel. Because right continuous adapted processes are progressively measurable, the map
is Borel. We conclude that for every \(t \in \mathbb {R}_+\) the map
is Borel. This implies that \(\mathcal {F} \subseteq \mathcal {B}(\Omega )\).
Note that \(\iota \) is \(\mathcal {F}/\mathcal {B}(\Omega ^\star )\) measurable. Let \(f :\Omega \rightarrow \mathbb {R}\) be a Borel function. Because \(\iota \) is an isometry, the inverse map \(\iota ^{-1} :\Omega ^\star \rightarrow \Omega \) is continuous and therefore Borel. We conclude that
is \(\mathcal {F}/\mathcal {B}(\mathbb {R})\) measurable as composition of the \(\mathcal {B}(\Omega ^\star )/\mathcal {B}(\mathbb {R})\) measurable map \(f \circ \iota ^{-1}\) and the \(\mathcal {F}/\mathcal {B}(\Omega ^\star )\) measurable map \(\iota \). This implies \(\mathcal {B}(\Omega ) \subseteq \mathcal {F}\) and the proof is complete. \(\square \)
3.3 Proof of Lemma 2
(i). We have to show that \(\{\tau (\Gamma ) \le t\} \in \mathcal {F}_t\) for all \(t \in \mathbb {R}_+\). For \(x \in \mathbb {B}\), we define \(d (x, \Gamma ) \triangleq \inf _{y \in \Gamma } \Vert x - y\Vert \) and set
Moreover, on \(\{t < \tau _\Delta \}\) we set
Because \(x \mapsto d(x, \Gamma )\) is Lipschitz continuous, the set \(\Gamma _n\) is open, and because \(\Gamma \) is closed, \(\Gamma = \{x \in \mathbb {B} :d(x, \Gamma ) = 0\}\). Define \(\tau \triangleq \sup _{n \in \mathbb {N}} \tau (\Gamma _n)\). Because \(\Gamma \subseteq \Gamma _n\), it is clear that \(\tau \le \tau (\Gamma )\). Next, we show that \(\tau \ge \tau (\Gamma )\). We claim that this inequality follows if we show that
We explain this: In case \(\tau \ge \tau _\Delta \), we have \(\tau = \tau (\Gamma ) = \tau _\Delta \). Take \(\omega \in \{\tau < \tau _\Delta \}\) and let \(\varepsilon > 0\) be such that \(\varepsilon < \tau _\Delta (\omega ) - \tau (\omega )\) in case \(\tau _\Delta (\omega ) < \infty \). For each \(n \in \mathbb {N}\), we find a \(t_n \in [\tau (\Gamma _n) (\omega ), \tau (\Gamma _n) (\omega ) + \varepsilon )\) such that \(F_{t_n} (\omega ) \cap \Gamma _n \not = \emptyset \). Note that \(t \triangleq \sup _{n \in \mathbb {N}} t_n \le \tau (\omega ) + \varepsilon < \tau _\Delta (\omega )\) and that \( F_t (\omega ) \cap \Gamma _n \not = \emptyset \text { for all } n \in \mathbb {N}. \) Consequently, in case (2.1) holds we have \(F_t(\omega ) \cap \Gamma \not = \emptyset \), which implies \(\tau (\Gamma ) (\omega ) \le t \le \tau (\omega ) + \varepsilon \). We conclude that \(\tau \ge \tau (\Gamma )\) as claimed. We proceed showing (2.1). Fix \(t \in \mathbb {R}_+\). Because on \(\{t < \tau _\Delta \}\)
it suffices to show that on \(\{t < \tau _\Delta \}\)
Take \(\omega \in \{t < \tau _\Delta \}\). Because \(\{\omega (\cdot \wedge t)\}\) is compact in \(D(\mathbb {R}_+, \mathbb {B})\), \(F_t(\omega )\) is compact in \(\mathbb {B}\) by [4, Problem 16, p. 152]. Consequently, due to its continuity, the function \(x \mapsto d(x, \Gamma )\) attains its infimum on \(F_t(\omega )\). Thus, because \(\Gamma = \{x \in \mathbb {B} :d(x, \Gamma ) = 0\}\), if \(\inf _{x \in F_t(\omega )} d(x, \Gamma ) = 0\), we have \(F_t(\omega ) \cap \Gamma \not = \emptyset \). We conclude that (2.1) holds and hence that \(\tau = \tau (\Gamma )\).
From the equality \(\tau = \tau (\Gamma )\), we deduce that for all \(t \in \mathbb {R}_+\)
Fix \(t \in \mathbb {R}_+\) and set \(\mathbb {Q}_+^t \triangleq ([0, t) \cap \mathbb {Q}_+) \cup \{t\}\). We note that
Because \(\Gamma _{n + 1}\) is open, we have
Thus, in case \(\tau (\Gamma _{n + 1}) \le t < \tau _\Delta \), the right continuity of \(X\) yields that \(X_{\tau (\Gamma _{n + 1})} \in cl _{\mathbb {B}}(\Gamma _{n + 1}) \subseteq \Gamma _n.\) We conclude that on \(\{t < \tau _\Delta \}\)
Now, (2.2), (2.3) and (2.4) imply that
Because
we conclude that \(\tau (\Gamma )\) is a stopping time. The proof of (i) is complete.
(ii). Because \(n \mapsto \tau (\mathbb {B} \backslash \Gamma _n)\) is increasing, \(\tau (\mathbb {B} \backslash \Gamma _n) \nearrow \tau \triangleq \sup _{n \in \mathbb {N}} \tau (\mathbb {B}\backslash \Gamma _n)\) as \(n \rightarrow \infty \). Because \(\tau \le \tau _\Delta \), it suffices to show that \(\tau \ge \tau _\Delta \). For contradiction, suppose that there exists an \(\omega \in \{\tau < \tau _\Delta \}\) and set \(\omega ' \triangleq \omega (\cdot \wedge \tau (\omega )) \in D(\mathbb {R}_+, \mathbb {B})\). Then,
Because \(\tau (\mathbb {B}\backslash \Gamma _n)\) is an \(\mathbf {F}\)-stopping time by (i), so is \(\tau \) and Galmarino’s test (see [6, Lemma III.2.43]) implies that \(\tau (\omega ) = \tau (\omega ') = \infty \). This is a contradiction and \(\tau = \tau _\Delta \) follows. The proof of (ii) is complete. \(\square \)
4 Modifications, Corrections and Comments on Proofs
4.1 [2, Lemma 4.3]
The last conclusion in [2, Lemma 4.3] is empty: In the setting of [2], it cannot happen that \(X_{\tau _\Delta -} = \Delta \).
4.2 [2, Lemmata 4.3, 4.5]
Due to the initial value and the possibility that \(X\) has no left limit at \(\tau _n\), some bounds in the proofs of [2, Lemmata 4.3, 4.5] are only valid on the open stochastic interval \(]]0, \tau _n[[\). Because singletons have Lebesgue measure zero, the arguments require no further changes.
The last conclusion in the proof of [2, Lemma 4.5] follows from the dominated convergence theorem.
4.3 [2, Proposition 4.8]
In the proof, it has been used that \(\rho (A^*) \not = \emptyset \), see [8, Lemma 4.1]. Because \(\mathbb {B}\) is separable and reflexive, its dual \(\mathbb {B}^*\) is separable and \(D\) in the proof of [2, Proposition 4.8] can be constructed more directly: The assumption \(\rho (A) \not = \emptyset \) implies that \(\rho (A^*) \not = \emptyset \), see [7, Theorem 5.30, p. 169]. Let \(D' \subset \mathbb {B}^*\) be a countable dense subset of \(\mathbb {B}^*\) and take \(\lambda \in \rho (A^*)\). Now, set \(R(\lambda , A^*) \triangleq (\lambda - A^*)^{-1}\) and define \(D \triangleq \{R(\lambda , A^*) x :x \in D' \} \subseteq D(A^*)\). We claim that for each \(x \in D(A^*)\) there exists a sequence \((x_n)_{n \in \mathbb {N}} \subset D\) such that \(x_n \rightarrow x\) and \(A^* x_n \rightarrow A^* x\) in the operator norm as \(n \rightarrow \infty \). To see this, take \(x \in D(A^*)\) and set \(y \triangleq \lambda x + A^* x\). There exists a sequence \((y_n)_{n \in \mathbb {N}}\subset D'\) such that \(y_n \rightarrow y\) as \(n \rightarrow \infty \). Finally, set \(x_n \triangleq R(\lambda , A^*) y_n \in D\). Because \(R(\lambda , A^*)\in L(\mathbb {B}^*, \mathbb {B}^*)\), we have \(x_n \rightarrow R(\lambda , A^*) y = x\) as \(n \rightarrow \infty \). Moreover, the triangle inequality yields that
The claim is shown.
4.4 [2, Lemma 4.10]
Due to Lemma 1, it is not necessary to pass to \(D(\mathbb {R}_+, \mathbb {B}_\Delta )\). Moreover, it can be seen more easily that \(\Phi \) is Borel. Indeed, \(\Phi \) is continuous.
4.5 [2, Lemma 4.11]
In the proof of \(P\)-a.s.
the variable \(n\) is used twice, which results in a conflict of notation. We correct the argument: Note that \(\tau _{n + k} \circ \theta _\xi + \xi \le \tau _{2 (n + k)}\) on \(\{\xi < \tau _{n + k}\}\) for all \(k \in \mathbb {N}\). Set \(\sigma _r \triangleq r \wedge \tau _n \circ \theta _\xi + \xi \). We obtain that \(P\)-a.s.
by the optional stopping theorem.
4.6 [2, Section 4.3.2]
Because \(X\) has no left limit at \(\tau _\Delta \), the random measure \(\mu ^X\) cannot be defined as in [2, Eq. 4.20]. We pass to a stopped version: Let \(\widehat{X}\) be defined as in Eq. 4.11 in [2] and set \(X^n \triangleq \widehat{X}_{\cdot \wedge \tau _n}\) and
We have the following version of [2, Lemmata 4.17, 4.18, 4.19]:
Lemma 4
For all \(n \in \mathbb {N}\) the random measure \(\mu ^n\) is \((\mathbf {F}^P, P)\)-optional with \(\mathscr {P}^P\)-\(\sigma \)-finite Doléans measure and \((\mathbf {F}^P, P)\)-predictable compensator \(\nu ^n\).
Because the proofs of [2, Lemmata 4.17, 4.18] contain typos and the proof of [2, Lemma 4.19] requires some minor modification, as the set \(\mathcal {Z}_1 \times \mathcal {Z}_2\) has not all claimed properties, we give a proof:
Proof
Due to [3, Theorem IV.88B, Remark below], the set \(\{\Delta X^n \not = 0\}\) is \(\mathbf {F}^P\)-thin. Hence, [6, II.1.15] yields that \(\mu ^n\) is \(\mathbf {F}^P\)-optional. It follows as in [9, Example 2, pp. 160] that \(M^P_{\mu ^n}\) is \(\mathscr {P}^P\)-\(\sigma \)-finite. Next, we show that \(\nu ^n\) is \(\mathbf {F}^P\)-predictable with \(\mathscr {P}^P\)-\(\sigma \)-finite Doléons measure \(M^P_{\nu ^n}\). For \(m \in \mathbb {N}\) we set \( G_m \triangleq \{x \in \mathbb {B} :\Vert x\Vert \ge \tfrac{1}{m}\} \cup \{0\}. \) Let \(W\) be a nonnegative \(\mathscr {P}^P \otimes \mathscr {B}(\mathbb {B})\)-measurable function which is bounded by a constant \(c > 0\). Because \(P\)-a.s.
we conclude that \(M^P_{\nu ^n}\) is \(\mathscr {P}^P\)-\(\sigma \)-finite. Furthermore, the process
is \(\mathbf {F}^P\)-predictable as the pointwise limit of an \(\mathbf {F}^P\)-predictable process. We conclude that \(\nu ^n\) is an \(\mathbf {F}^P\)-predictable random measure.
It remains to show that \(\nu ^n\) is the \((\mathbf {F}^P, P)\)-predictable compensator of \(\mu ^n\). Let \(\mathcal {Z}_1\) be the collection of sets \(A \times \{0\}\) for \(A \in \mathcal {F}^P_0\) and \([[0, \xi ]]\) for all \(\mathbf {F}^P\)-stopping times \(\xi \), and let \(\mathcal {Z}_2\) be the collection of all sets
for \(A \in \mathcal {B}(\mathbb {R}^d), y^*_1, \dots , y^*_d \in D(A^*)\) and \(d \in \mathbb {N}\). Note that \(M^P_{\mu ^n} (A \times \{0\} \times G) = M^P_{\nu ^n} (A \times \{0\} \times G) = 0\) for all \(A \in \mathcal {F}^P_0\) and \(G \in \mathcal {B}(\mathbb {B})\). Fix an \(\mathbf {F}^P\)-stopping time \(\xi \) and the cylindrical set \(G\) given by (3.1). Denote \(Y^n \triangleq (\langle X^n, y^*_1\rangle , \dots , \langle X^n, y^*_d\rangle ).\) By [2, Lemma 4.7], we obtain
which implies \(M^P_{\mu ^n} = M^P_{\nu ^n}\) on \(\mathcal {Z}_1 \times \mathcal {Z}_2\). Take a norming sequence \((x^*_m)_{m \in \mathbb {N}}\subset \mathbb {B}^*\) of unit vectors, see p. 522 in [5] for a definition, and note that
For \(m, k \in \mathbb {N}\) set
The dominated convergence theorem yields that
for all \(A \times B \in \mathcal {Z}_1 \times \mathcal {Z}_2\). Now, we conclude from the uniqueness theorem for measures that \(M^P_{\mu ^n} = M^P_{\nu ^n}\) on the trace \(\sigma \)-field \((\mathscr {P}^P \otimes \mathcal {B}(\mathbb {B}))\cap ([[0, \gamma (m, k)]]\times (B_m \cup \{0\}))\). Finally, taking \(k, m \rightarrow \infty \) and using the monotone convergence theorem show that \(M^P_{\mu ^n} = M^P_{\nu ^n}\) on \(\mathscr {P}^P \otimes \mathcal {B}(\mathbb {B})\). The proof is complete. \(\square \)
The candidate density process \(Z\) can be defined as in [2, Lemma 4.21] with \(\mu ^X\) and \(\nu ^X\) replaced by \(\mu ^n\) and \(\nu ^n\).
4.7 [2, Lemmata 4.21, 4.22]
In the proofs, the process \(X\) should be replaced by \(\widehat{X}\).
4.8 [2, Proposition 3.7]
The representation of the CMG densities and the function \(V^k\) in [2, Lemma 4.23] should be multiplied by \(\mathbf {1}_{\{\tau _n < \tau _\Delta \}}\). Moreover, in all Lebesgue integrals \(X_-\) should be replaced by \(X\).
4.9 [2, Lemma 3.16]
Instead of the Yamada–Watanabe argument, the uniqueness also follows from the observation that for a pseudo-contraction semigroup \((S_t)_{t \ge 0}\) and a square integrable Lévy process \(L\) the law of \(\int _0^\cdot S_{\cdot - s} dL_s\) is completely determined by \(L\). This can be seen with the approximation argument used in the proof of [11, Theorem 9.20].
5 Final Comment
Above [2, Proposition 3.9] it is noted that “in a non-conservative setting, one can try to conclude existence from an extension argument in a larger path space, [but] in this case one has to prove that the extension is supported on \((\Omega , \mathcal {F})\)” as defined in [2]. The larger path space, to which this comment refers, is the path space defined in this correction note. In our modified setting, it follows from Parthasarathy’s extension theorem (see [10]) that under the assumptions imposed in [2] the GMP \((A, b', a, K', \eta , \tau _\Delta -)\) has a solution whenever the GMP \((A, b, a, K, \eta , \tau _\Delta -)\) has a solution. This observation extends [2, Theorem 3.6].
References
Biagini, S., Bouchard, B., Kardaras, C., Nutz, M.: Robust fundamental theorem for continuous processes. Math. Finance 27(4), 963–987 (2017)
Criens, D.: Cylindrical martingale problems associated with Lévy generators. J. Theor. Probab. 32(3), 1306–1359 (2019)
Dellacherie, C., Meyer, P.: Probabilities and Potential. North-Holland, Amsterdam (1978)
Ethier, S., Kurtz, T.: Markov Processes: Characterization and Convergence. Wiley, Hoboken (2005)
Hytönen, T., van Neerven, J., Veraar, M., Weis, L.: Analysis in Banach Spaces: Volume I: Martingales and Littlewood–Paley Theory. Springer, Berlin (2016)
Jacod, J., Shiryaev, A.: Limit Theorems for Stochastic Processes, 2nd edn. Springer, Berlin (2003)
Kato, T.: Perturbation Theory for Linear Operators, 2nd edn. Springer, Berlin (1980)
Kunze, M.: On a class of martingale problems on Banach spaces. Electron. J. Probab. 18(104), 1–30 (2013)
Liptser, R., Shiryaev, A.N.: Theory of Martingales. Springer, Berlin (1989)
Parthasarathy, K.: Probability Measures on Metric Spaces. Academic Press, New York (1967)
Peszat, S., Zabczyk, J.: Stochastic Partial Differential Equations with Lévy Noise: An Evolution Equation Approach. Cambridge University Press, Cambridge (2007)
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.
Rights and permissions
About this article
Cite this article
Criens, D. Correction to: Cylindrical Martingale Problems Associated with Lévy Generators. J Theor Probab 33, 1791–1800 (2020). https://doi.org/10.1007/s10959-020-01012-1
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10959-020-01012-1