Abstract
We consider a class of generalized time-fractional evolution equations containing a fairly general memory kernel k and an operator L being the generator of a strongly continuous semigroup. We show that a subordination principle holds for such evolution equations and obtain Feynman-Kac formulae for solutions of these equations with the use of different stochastic processes, such as subordinate Markov processes and randomly scaled Gaussian processes. In particular, we obtain some Feynman-Kac formulae with generalized grey Brownian motion and other related self-similar processes with stationary increments.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In this paper, we study a general class of evolution equations
where k is a fairly general memory kernel and L is the infinitesimal generator of a strongly continuous semigroup \((T_t)_{t\geqslant 0}\) acting on some Banach space X. In particular, the operator L may be the generator \(L_0\) of a Markov process \(\xi \) on some state space Q, or \(L:=L_0+b\nabla +V\) for a suitable potential V and drift b. Moreover, L may be the generator of a subordinate semigroup or a Schrödinger type group. This class of evolution equations includes in particular time- and space- fractional heat and Schrödinger type equations as well as equations with generalized time-fractional derivatives of Caputo type (cf. Remarks 1, 2 below). Such equations are widely discussed in the literature (see, e.g., [2, 10, 14, 18, 19, 24,25,26,27] and references therein), in particular, in connection with models of anomalous diffusion. We refer to [4] for additional background information and for detailed discussion of the considered memory kernels k.
In this paper, we show that the solution operator of equation (1.1) can be written in the form
for a family \((\mathcal {P}_{A(t)})_{t\geqslant 0}\) of probability measures on the positive real line, which depends on k only. We, thus, consider this representation as a subordination principle associated to the memory kernel k. We state the subordination principle in Section 2, and in particular discuss how to obtain stochastic representations of the solution, if the operator L is (a Bernstein function of) the infinitesimal generator of a Markov process (plus a potential). The most natural stochastic reprensentations of such an approach are given in terms of time-changed Markov processes. In Section 3, we explain, however, how to arrive at representations in terms of non-Markovian processes such as generalized grey Brownian motion or even in terms of solutions of stochastic differential equations driven by more general randomly scaled fractional Brownian motions. Such stochastic processes are attractive for modelling since they are self-similar and with stationary increments (cf. [26, 27]). Finally, the proofs are provided in Section 4. While the main results can be considered as generalizations of our previous results in [4] beyond the case of pseudo-differential operators L associated to Lévy processes, the proofs are completely different, relating (an approximate version of) the subordination principle to a family of Volterra equations via the Hille-Phillips functional calculus.
2 Main results
Assumption 1
Let X be a Banach space with a norm \(\Vert \cdot \Vert _X\). Let \((T_t)_{t\geqslant 0}\) be a strongly continuous semigroup on X with generator \((L,\mathrm{Dom}(L))\).
We consider the evolution equation (1.1) with operator L as in Assumption 1, with \(u_0\in \mathrm{Dom}(L)\), \(u\,:\,[0,\infty )\rightarrow X\) and k satisfying the following Assumptions 2–3.
Assumption 2
We consider a Borel-measurable kernel \(k\,:\,(0,\infty )\times (0,\infty )\rightarrow {{\mathbb {R}}}\) satisfying the condition: \(\exists \,\alpha ^*\in [0,1)\) and \(\exists \,\varepsilon >0\) such that for each \(T>0\)
In order to identify the family of probability measures \((\mathcal {P}_{A(t)})_{t\geqslant 0}\) for the subordination, we specify their Laplace transform in terms of the memory kernel k. To this end we define the function \(\varPhi :[0,\infty )\times \mathbb {C}\rightarrow \mathbb {C}\) via
It has been shown in [4] that, under Assumption 2, the function \(\varPhi \) is well-defined and, for fixed t, entire in \(\lambda \).
Assumption 3
Let the function \(\varPhi \) be constructed from the kernel k via formulas (2.1), (2.2). We assume that the restriction of the function \(\varPhi (t,-\cdot )\) on \((0,\infty )\) is completely monotone for all \(t\geqslant 0\), i.e., for each \(t\geqslant 0\), there exists a nonnegative random variable A(t) whose distribution \(\mathcal {P}_{A(t)}\) has the Laplace transform given by \(\varPhi (t,-\cdot )\):
Note that \(\mathcal {P}_{A(0)}=\delta _0\) and \(A(0)=0\) a.s. since \(\varPhi (0,-\lambda )\equiv 1\).
Typical examples of kernels k satisfying Assumptions 2–3 are kernels of convolution type and homogeneous kernels related to operators of generalized fractional calculus (cf. [4]). Recall that a kernel k is homogeneous of degree \(\theta -1\) for some \(\theta >0\) if \(k(t,ts)=t^{\theta -1}k(1,s)\), \(t\in (0,\infty )\), \(s\in (0,1)\).
Theorem 1
Let Assumption 1 hold. Let k satisfy Assumption 2 and assume that the corresponding function \(\varPhi \) satisfies Assumption 3. Then:
(i) For each \(t\geqslant 0\), the operator \(\varPhi (t,L)\) given by the Bochner integral
is well defined, and it is a bounded linear operator on X.
(ii) For each \(t>0\) and each \(u_0\in \mathrm{Dom}(L)\), the function
solves equation (1.1) and it holds \(\lim _{t\searrow 0} u(t)=u_0\).
(iii) Suppose additionally that k is homogeneous of order \(\theta -1\) for some \(\theta >0\). Then one can choose \(A(t):=At^\theta \) in (2.4), where A is a nonnegative random variable such that
We next wish to apply the semigroup \((T_t)_{t\geqslant 0}\) associated to a generator L in order to represent the solution of the evolution equation with memory kernel k and the (space-)fractional operator \(-(-L)^\gamma \). We use subordination in the sense of Bochner [5, 30] which is a random time change of a given process \((\xi _t)_{t\geqslant 0}\) by an independent subordinator, i.e. an 1-dimensional increasing Lévy process (with killing) \((\eta ^f_t)_{t\geqslant 0}\). Any subordinator can be characterized in terms of its Laplace exponent f: \(\mathbb {E}\left[ e^{-\lambda \eta ^f_t} \right] =e^{-tf(\lambda )}\); any such f is a Bernstein function and is determined uniquely by its Lévy-Khintchine representation \(f(\lambda )=a+b\lambda +\int _{(0,\infty )}\left( 1-e^{-\lambda s} \right) \nu (ds)\), where a, \(b\geqslant 0\) and \(\nu \) is a measure on \((0,\infty )\) satisfying \(\int _{(0,\infty )}\min (s,1)\nu (ds)<\infty \). Let \((T_t)_{t\geqslant 0}\) be as in Assumption 1 and additionally a contraction semigroup. The family of operators \((T^f_t)_{t\geqslant 0}\) defined by the Bochner integral
is said to be subordinate to \((T_t)_{t\geqslant 0}\) with respect to the convolution semigroup of measures \(\left( \mathcal {P}_{\eta ^f_t} \right) _{t\geqslant 0}\), where \(\mathcal {P}_{\eta ^f_t}\) is the distribution of \(\eta ^f_t\). The family \((T^f_t)_{t\geqslant 0}\) is again a strongly continuous contraction semigroup on the space X whose generator \((L^f,\mathrm{Dom}(L^f))\) is the closure of the operator \((-f(-L),\mathrm{Dom}(L))\), where
If \((T_t)_{t\geqslant 0}\) is the transition semigroup of a Feller process \((\xi _t)_{t\geqslant 0}\) and \((\eta ^f_t)_{t\geqslant 0}\) is an independent subordinator, then \((T^f_t)_{t\geqslant 0}\) is the transition semigroup of the (again Feller) process \((\xi _{\eta ^f_t})_{t\geqslant 0}\). Further information on subordination in the sense of Bochner and all related objects can be found e.g. in [31].
Consider now the function \(\varPhi ^f(t,-\cdot ):=\varPhi (t,-f(\cdot ))\). If the function \(\varPhi (t,-\cdot ) \) is completely monotone, so is the function \(\varPhi ^f(t,-\cdot )\), as a composition of a Bernstein function f and a completely monotone function \(\varPhi (t,-\cdot )\). Hence there exists a family of nonnegative random variables whose Laplace transform is given by \(\varPhi ^f(t,-\cdot )\), \(t\geqslant 0\). Using distributions of these random variables and a strongly continuous contraction semigroup \((T_t)_{t\geqslant 0}\) with generator \((L,\mathrm{Dom}(L))\), one can define the operator \(\varPhi ^f(t,L)\) analogously to (2.4).
Corollary 1
Let Assumption 1 hold and \((T_t)_{t\geqslant 0}\) be a contraction semigroup. Let k satisfy Assumption 2 and the corresponding function \(\varPhi \) satisfy Assumption 3. Let \((A(t))_{t\geqslant 0}\) be a family of nonnegative random variables satisfying (2.3). Let \((\eta ^f_t)_{t\geqslant 0}\) be a subordinator corresponding to a Bernstein function f which is independent from \((A(t))_{t\geqslant 0}\). Then
Moreover, for each \(t>0\) and each \(u_0\in \mathrm{Dom}(L^f)\), the function \(u(t):=\varPhi ^f(t,L)u_0\) solves the evolution equation
If the semigroup \((T_t)_{t\geqslant 0}\) has a stochastic representation, then the family \((\varPhi ^f(t,L))_{t\geqslant 0}\) as well has a stochastic representation due to (2.7).
Example 1
Let Q be a separable completely metrizable topological space endowed with a Borel \(\sigma \)-field \(\mathcal {B}(Q)\). Let \((\varOmega ,{\mathcal {F}},\mathbb {P}^x,(\xi _t)_{t\geqslant 0})_{x\in Q}\) be a (universal) Markov process with state space \((Q,\mathcal {B}(Q))\). Assume that the corresponding transition semigroup \((T^0_t)_{t\geqslant 0}\), \( T^0_t u_0(x):=\mathbb {E}^x\left[ u_0(\xi _t) \right] , \) is a strongly continuous semigroup on some Banach space \(X\subset B_b(Q)\) (where \(B_b(Q)\) is the space of all bounded Borel measurable functions on Q). Let \((L_0,\mathrm{Dom}(L_0))\) be the generator of \((T^0_t)_{t\geqslant 0}\). Let \(V\,:\,Q\rightarrow (-\infty ,0]\) be a Borel measurable function such that the (closure of the) operator \((L_0+V,\mathrm{Dom}(L_0+V))\) generates a strongly continuous semigroup \((T_t)_{t\geqslant 0}\) on X with stochastic representation
Note that (2.9) is the classical Feynman-Kac formula which holds under very mild assumptions on processes and potentials, cf., e.g., [6, 7, 21]. Let assumptions of Corollary 1 hold and \((\xi _t)_{t\geqslant 0}\) be independent from \((A(t))_{t\geqslant 0}\) and \((\eta ^f_t)_{t\geqslant 0}\). Then for \(u_0\in \mathrm{Dom}((L_0+V)^f)\) the function
solves the evolution equation
Suppose additionally that k is homogeneous of order \(\theta -1\) for some \(\theta >0\). Let A be a nonnegative random variable which satisfies (2.6) and is independent from \((\eta ^f_t)_{t\geqslant 0}\) and \((\xi _t)_{t\geqslant 0}\). Then we can take \(A(t):=At^\theta \) in (2.10).
Remark 1
Theorem 1 (and Corollary 1) can be applied also to generalized time-fractional Schrödinger type equations. Note, that different types of fractional analogues of the standard Schrödinger equation have been discussed in the literature, see, e.g., [2, 10, 14]. Such equations seem to be physically relevant; in particular, some of them arise from the standard quantum dynamics under special geometric constraints [19, 29]. So, let \(X:=L^2({{\mathbb {R}}}^d)\) be the Hilbert space of complex-valued square integrable functions; X plays the role of the state space of a quantum system. Let \((\mathcal {H},\mathrm{Dom}(\mathcal {H}))\) be a (bounded from below) self-adjoint operator in X playing the role of the Hamiltonian (energy operator) of this quantum system. Then \((L,\mathrm{Dom}(L)):=(-i\mathcal {H},\mathrm{Dom}(\mathcal {H}))\) does generate a strongly continuous contraction semigroup \((T^\mathcal {H}_t)_{t\geqslant 0}\) on X by the Stone theorem. Let k, \(\varPhi \), \((A(t))_{t\geqslant 0}\) be as in Theorem 1. Then, by Theorem 1,
solves the generalized time-fractional Schrödinger type equation
where the equality above is understood as the equality of two elements of the space X. For a few particular choices of the Hamiltonian, some stochastic representations of the corresponding semigroup \((T^\mathcal {H}_t)_{t\geqslant 0}\) are known in the literature (see, e.g., [8, 9, 18]). Inserting these stochastic representations into (2.12), one obtains Feynman-Kac formulae (which may be local in the space variables) for the corresponding generalized time-fractional Schrödinger type equation (2.13).
Remark 2
(i) Consider \(k(t,s):={\mathfrak {K}}(t-s)\), where \({\mathfrak {K}}\) is such that its Laplace transform \(\mathcal {L}[{\mathfrak {K}}]=1/h\) for some Bernstein function h. Then this k satisfies Assumptions 2, 3. And one may take an inverse subordinator corresponding to h as \((A(t))_{t\geqslant 0}\) in this case (cf. Sec. 2.3 of [4]). Moreover, evolution equation (1.1) is then equivalent (what can be shown by applying the Laplace transform w.r.t. time-variable to both equations) to the Cauchy problem
where \(\mathcal {D}^h_t\) is a generalized time-fractional derivative of Caputo type, which is defined (for sufficiently good functions \(v\,:\,(0,\infty )\rightarrow {{\mathbb {R}}}\) of time variable t) via the Laplace transform (cf. [1]) by
Therefore, the results of Theorem 1 and Corollary 1 provide solutions for evolution equations of the form (2.14) with generalized time-fractional derivatives of Caputo type \(\mathcal {D}^h_t\). In the case \(h(\sigma ):=\sigma ^\beta \), \(\beta \in (0,1)\), the generalized time-fractional derivative \(\mathcal {D}^h_t\) coincides with the Caputo derivative of order \(\beta \). The kernel \({\mathfrak {K}}_1\) below corresponds to a mixture of Caputo time-fractional derivatives of orders \(\beta \), \(\beta _1,\ldots ,\beta _m\). In the case of Bernstein function \(h(\sigma ):=\int _0^1 \sigma ^\beta \mu (d\beta )\) with a finite Borel measure \(\mu \) concentrated on the interval (0, 1), the corresponding derivative \(\mathcal {D}^h_t\) is known as distributed order fractional derivative.
(ii) Let us mention the following functions \({\mathfrak {K}}_1\) and \({\mathfrak {K}}_2\) providing kernels k as in part (i) (cf. [3]): for \(1\geqslant \beta>\beta _1>\ldots>\beta _m>0\), \(b_j>0\), \(j=1,\ldots ,m\)
with the corresponding Bernstein function \(h_1(\sigma ):=\left( \sigma ^{-\beta }+\sum _{j=1}^m b_j\sigma ^{-\beta _j} \right) ^{-1}\) and
with multinomial Mittag-Leffler function [13, 17] (for \(z_j\in {{\mathbb {C}}}\), \(\beta \in {{\mathbb {R}}}\), \(\alpha _j>0\), \(j=1,\ldots ,m\))
The kernel \({\mathfrak {K}}_2\) corresponds to the Bernstein function \(h_2(\sigma ):=\sigma ^{\beta }+\sum _{j=1}^mb_j\sigma ^{\beta _j}\). The corresponding functions \(\varPhi _1(t,-\lambda )\) and \(\varPhi _2(t,-\lambda )\) are found in [3] in terms of the multinomial Mittag-Leffler function:
3 Feynman-Kac formulae with randomly scaled Gaussian processes
Due to Corollary 1, the most natural stochastic representations for evolution equations of the form (1.1) with L being (a Bernstein function of) the generator of a Markov process (plus a potential term) are given in terms of time-changed Markov processes. In the special case when the memory kernel k is homogeneous, one may sometimes use randomly scaled Gaussian processes in the obtained stochastic representations. For this recall first a class of memory kernels k from [4] which satisfy Assumptions 2-3 and are homogeneous.
Example 2
Let \(b>0\), \(a\geqslant b\), \(\mu \geqslant \frac{b}{a}-1\), \(\nu >\max \left\{ a-b,-a\mu \right\} \). Consider the Marichev-Saigo-Maeda kernel (cf. Sec. 4 in [4])
where \(0<s<t\) and \(F_3\) is Appell’s third generalization of the Gauss hypergeometric function. The kernel k is homogeneous of degree \(b-1\) and satisfies Assumptions 2, 3. The corresponding function \( \varPhi \) has the following form: \( \varPhi (t,\lambda )= \varGamma (q_2) E_{q_1,q_2}^{q_3}(\lambda t^b)\), where \(q_1:=\frac{b}{a}\), \( q_2:=\frac{\nu }{a}+\mu \), \(q_3:= 1+\frac{\nu -a}{b}\), and \(E_{q_1,q_2}^{q_3}\) is the three parameter Mittag-Leffler function \( E_{q_1,q_2}^{q_3}(\lambda ):=\sum _{n=0}^\infty \frac{\left( q_3\right) _n}{\varGamma \left( q_1 n+q_2\right) n!}\,\lambda ^n. \) As corresponding random variables \((A(t))_{t\geqslant 0}\) one may take \(A(t):=A_{b,a,\mu ,\nu }t^b\), where \(A_{b,a,\mu ,\nu }\) is a non-negative random variable whose Laplace transform is given by \(\varGamma (q_2) E_{q_1,q_2}^{q_3}(-\lambda )\). In the special case \(\mu :=0\), \(b:=\alpha \), \(a:=\frac{\alpha }{\beta }\), \(\nu :=a\) for some \(\alpha \in (0,2)\), \(\beta \in (0,1]\), the Marichev-Saigo-Maeda kernel (3.1) reduces to the kernel which appears in the governing equation for generalized grey Brownian motion:
The corresponding function \(\varPhi \) reduces to the classical Mittag-Leffler function: \(\varPhi (t,\lambda )=E_\beta (\lambda t^\alpha )\). And, as the corresponding random variables \((A(t))_{t\geqslant 0}\), one may take \(A(t):=A_{\beta }t^\alpha \), where \(A_{\beta }\) is a non-negative random variable with Laplace transform \(E_\beta (-\cdot )\).
Let us now present some Feynman-Kac formulae for evolution equations of type (1.1) with homogeneous kernel k on the base of randomely scaled Gaussian processes.
Example 3
(i) Under the assumptions of Corollary 1 consider the Bernstein function \(f(\lambda ):=\lambda ^\gamma \), \(\gamma \in (0,1]\). Then, in the case \(\gamma \in (0,1)\), the operator \(L^f\) is the fractional power of the operator L, i.e. \(L^f=-(-L)^\gamma \) (cf. [31]), and \((\eta ^f_t)_{t\geqslant 0}\) is a \(\gamma \)-stable subordinator. In the case \(\gamma =1\), we take \(\eta ^f_t:=t\), \(t\geqslant 0\). Let k be homogeneous of degree \(\theta -1\) for some \(\theta >0\) and take \(A(t)=At^\theta \) according to Corollary 1 and Theorem 1 (iii). Then the random variable \(\eta ^f_{A(t)}\) has the same distribution as \(A^{1/\gamma }\eta ^f_1 t^{\theta /\gamma }\). We may replace the “subordinator” \(\left( \eta ^f_{A(t)}\right) _{t\geqslant 0}\) in (2.10) by a new “subordinator” \(\left( \mathcal {A}t^{\theta /\gamma }\right) _{t\geqslant 0}\) with
This allows to split randomness and time-dependence in the random time-change. Thus, we obtain the following Feynman-Kac formula
for the evolution equation
(ii) Let k, A, \((\eta ^f_t)_{t\geqslant 0}\) and \(\mathcal {A}\) be as in part (i) of this example. Let \(V:=c\) for some \(c\leqslant 0\), \(\xi _t:=x+B_t+wt\) under \(\mathbb {P}^x\), where \((B_t)_{t\geqslant 0}\) is a standard d-dimensional Brownian motion, which is independent from A and \((\eta ^f_t)_{t\geqslant 0}\), and \(w\in {{\mathbb {R}}}^d\) is some fixed vector. Let \(X^{\mathcal {A},\gamma ,\theta }_t:=B_{\text{\AA} t^{\theta /\gamma }}\) or \(X^{\mathcal {A},\gamma ,\theta }_t:=\sqrt{\mathcal {A}} B_{ t^{\theta /\gamma }}\), or, if \(H:=\frac{\theta }{2\gamma }\in (0,1)\), \(X^{\mathcal {A},\gamma ,\theta }_t:=\sqrt{\mathcal {A}} B^H_{ t}\), where \(\left( B^{H}_t \right) _{t\geqslant 0}\) is a d-dimensional fractional Brownian motion with Hurst parameter H which is independent from A and \((\eta ^f_t)_{t\geqslant 0}\). Note that all three options of the process \((X^{\mathcal {A},\gamma ,\theta }_t)_{t\geqslant 0}\) have the same one-dimensional marginal distributions. Then, due to Feynman-Kac formula (3.4),
solves the evolution equation
Therefore, we have obtained a Feynman-Kac formula (3.5) for the evolution equation (3.6) in terms of two different classes of randomly scaled Gaussian processes: randomly scaled slowed-down / speeded-up Brownian motion \(\left( \sqrt{\mathcal {A}} B_{ t^{\theta /\gamma }} \right) _{t\geqslant 0}\) and (if \(H:=\frac{\theta }{2\gamma }\in (0,1)\)) randomly scaled fractional Brownian motion \(\left( \sqrt{\mathcal {A}} B^H_{ t} \right) _{t\geqslant 0}\). If k is a Marichev-Saigo-Maeda kernel (3.1) then \(\theta =b\), \(A=A_{b,a,\mu ,\nu }\) in distribution. In the special case of the GGBM-kernel (3.2), we have \(\theta =\alpha \), \(A=A_\beta \) in distribution, and hence we may use generalized grey Brownian motion in formula (3.5) as the process \((X^{\mathcal {A},\gamma ,\theta }_t)_{t\geqslant 0}\).
The result of Example 3 (ii) can be generalized beyond the case of a constant diffusion coefficient, as detailed in the case of dimension \(d=1\) in space in the proposition below. As can be seen from the proof, this generalization requires to move from a Brownian motion to a stochastic differential driven by a Brownian motion in the Stratonovich sense in order to apply Corollary 1.
Proposition 1
Let \(\gamma \in (0,1]\) and suppose the kernel k is homogeneous of order \(\theta -1\) for some \(\theta >0\) and Assumption 2, Assumption 3 are satisfied. Let \(\mathcal {A}\) be a non-negative random variable constructed by (3.3) in Example 3 (i). Assume \(w\in {{\mathbb {R}}}\), \(c\leqslant 0\), and \(\sigma \in C^2({{\mathbb {R}}})\) is a bounded function with bounded first and second derivatives. Consider the linear operator \((L_{(\sigma ,w)},\mathrm{Dom}(L_{(\sigma ,w)}))\) in \(C_\infty ({{\mathbb {R}}})\) which is defined by
Let \(u_0\in \mathrm{Dom}(L_{(\sigma ,w)})\) and denote by \(g_\sigma \,:\,[0,\infty )\times {{\mathbb {R}}}\rightarrow {{\mathbb {R}}}\) the solution to the parametrized family of ODEs
Let \((B_t)_{t\geqslant 0}\) be a standard Brownian motion independent from \(\mathcal {A}\). Let \(X^{\mathcal {A},\gamma ,\theta }_t:=B_{\text{\AA} t^{\theta /\gamma }}\) or \(X^{\mathcal {A},\gamma ,\theta }_t:=\sqrt{\mathcal {A}} B_{ t^{\theta /\gamma }}\), or, if \(H:=\frac{\theta }{2\gamma }\in (0,1)\), \(X^{\mathcal {A},\gamma ,\theta }_t:=\sqrt{\mathcal {A}} B^H_{ t}\), where \(\left( B^{H}_t \right) _{t\geqslant 0}\) is a 1-dimensional fractional Brownian motion with Hurst parameter H which is independent from \(\mathcal {A}\). Then
solves the evolution equation
The proof of Proposition 1 will be given in Section 4.
Remark 3
Let \(H:=\frac{\theta }{2\gamma }\in (0,1)\) and \(X^{\mathcal {A},\gamma ,\theta }_t:=\sqrt{\mathcal {A}} B^H_{ t}\), where \(\left( B^{H}_t \right) _{t\geqslant 0}\) is a 1-dimensional fractional Brownian motion with Hurst parameter H as in Proposition 1. We now interpret the Feynman-Kac formula (3.9) from the point of view of stochastic differential equations with respect to \(\left( X^{\mathcal {A},\gamma ,\theta }_t \right) _{t\geqslant 0}\) in the rough path sense. Note that almost every path of \(\left( X^{\mathcal {A},\gamma ,\theta }_t \right) _{t\geqslant 0}\) is Hölder continuous with each index less than H. Assume \(H>1/3\). Consider now \(\mathbb {X}_{t,s}:=\frac{1}{2}\left( X^{\mathcal {A},\gamma ,\theta }_t-X^{\mathcal {A},\gamma ,\theta }_s \right) ^2\). Then \(\mathcal {X}:=(X^{\mathcal {A},\gamma ,\theta }, \mathbb {X})\) is a lift to a geometric rough path (see [12]). Consider \(Z^x_t:=g_\sigma \left( X^{\mathcal {A},\gamma ,\theta }_t,x \right) \). Then, by the Itô formula for geometric rough paths, see again [12],
since \(g_\sigma \in C^3({{\mathbb {R}}})\). Hence, the stochastic representation for the solution of (3.10) in (3.9) can be rewritten in the form
This form resembles the classical Feynman-Kac formula for parabolic Cauchy problems in terms of stochastic differential equations driven by a Brownian motion. However, the stochastic differential equation (3.11) is driven by a randomly scaled fractional Brownian motion, which is neither a semimartingale nor a Markov process (unless \(H=1/2\)), to account for the memory kernel and the space fractionality in (3.10), while maintaining the stationary increment property of the driving process.
4 Proofs
Proof of Theorem 1
(i) Let Assumptions 1, 2 and 3 hold. Since the function \(\varPhi (t,\cdot )\), \(t\geqslant 0\), is entire by Theorem 1 in [4], the function \(\varPhi (t,i(\cdot ))\) is also entire and is the characteristic function of the distribution \(\mathcal {P}_{A(t)}\), which is concentrated on \([0,\infty )\). Therefore, we have by the Raikov theorem (cf. Theorem 3.2.1 in [22])
Further, for any strongly continuous semigroup \((T_t)_{t\geqslant 0}\) there exist constants \(M\geqslant 1\), \(c\geqslant 0\) such that \(\Vert T_t\Vert \leqslant M e^{ct}\), \(\forall \,t\geqslant 0\), and the mapping \(t\mapsto T_t\varphi \) is continuous for any \(\varphi \in X\). Thus, we have \(\int _0^\infty \Vert T_a\varphi \Vert _X \mathcal {P}_{A(t)}(da)<\infty \) and the Bochner integral in the r.h.s. of (2.4) is well defined for any \(\varphi \in X\). Moreover, the operator \(\varPhi (t,L)\) defined by (2.4) is a bounded linear operator on X and \(\varPhi (0,L)=\mathop {\mathrm {Id}}\nolimits \).
(ii) Recall that the following statement was proved in [4] (cf. Corollary 1 of [4]):
Lemma 1
Let Assumption 2 hold. Then, for each \(\lambda \in {{\mathbb {C}}}\), there exists a unique solution \(\varPhi (\cdot ,-\lambda )\in B_b([0,T],{{\mathbb {C}}})\), \(\forall \,\, T>0\), of the following Volterra equation of the second kind
Moreover, \(\lim _{t\searrow 0}\varPhi (t,-\lambda )=1\) locally uniformly with respect to \(\lambda \in {{\mathbb {C}}}\), \(\varPhi (t,\cdot )\) is an entire function for all \(t\geqslant 0\) and equalities (2.1) and (2.2) hold.
Our aim is to lift the equality (4.2) to the level of operators \(\varPhi (t,L)\). To this aim we use the so-called Hille-Phillips functional calculus. Let us recall the main facts about this functional calculus (cf. [15, 16]).
Let \((T_t)_{t\geqslant 0}\) be as in Assumption 1. Consider first the case when \((T_t)_{t\geqslant 0}\) is uniformly bounded (i.e. \(\Vert T_t\Vert \leqslant M\) for some \(M\geqslant 1\) and all \(t\geqslant 0\)). Denote by \(LS({{\mathbb {C}}}_+)\) the space of functions that are Laplace transforms of complex measures on \(([0,\infty ),\mathcal {B}([0,\infty )))\). Let \(g\in LS({{\mathbb {C}}}_+)\) and \(m_g\) be the (unique) complex measure whose Laplace transform \(\mathcal {L}[m_g]\) is given by g. One defines the operator \(g(-L)\) as follows:
The right hand side of (4.3) is a well-defined Bochner integral and \(g(-L)\) is a bounded linear operator on X, i.e. \(g(-L)\in \mathcal {L}(X)\). The mapping \(\mathcal {C}_T\,:\,LS({{\mathbb {C}}}_+)\rightarrow \mathcal {L}(X)\), \(g\mapsto g(-L)\), is called the Hille-Phillips calculus for \(-L\). Note that \(\mathcal {C}_T\) is an algebra homomorphism and hence \(\mathcal {C}_T(g_1g_2)= g_1(-L)\circ g_2(-L)=g_2(-L)\circ g_1(-L)\) and \( \mathcal {C}_T( a g_1+b g_2)= a g_1(-L)+b g_2(-L)\) for any \(g_1\), \(g_2\in LS({{\mathbb {C}}}_+)\), \(a,b\in {{\mathbb {R}}}\).
Consider now the case when \((T_t)_{t\geqslant 0}\) is of type \(c\geqslant 0\) (i.e., \(\Vert T_t\Vert \leqslant M e^{ct}\) for some \(M\geqslant 1\), \(c\geqslant 0\) and all \(t\geqslant 0\)). Then the rescaled semigroup \((T^c_t)_{t\geqslant 0}\), \(T^c_t:=T_te^{-ct}\), is uniformly bounded, strongly continuous and has generator \((L-c,\mathrm{Dom}(L))\). Then one may use the Hille-Phillips calculus \(\mathcal {C}_{T^c}\) for \(-(L-c)\). Consider now the space \(LS({{\mathbb {C}}}_+-c):=\left\{ g\,:\,g(\cdot -c)\in LS({{\mathbb {C}}}_+) \right\} \). Let \(g\in LS({{\mathbb {C}}}_+-c)\) and \(m^c_g\) be the (unique) complex measure with \(\mathcal {L}[m^c_g]=g(\cdot -c)\). One defines the operator \(g(-L)\) as follows:
Let now m be a complex measure such that \(e^{ca}m(da)\) is again a complex measure. Let \(g^*:=\mathcal {L}[m]\). Then it holds \(\mathcal {L}[e^{ca}m(da)](\lambda )=\int _0^\infty e^{-\lambda a}e^{ca}m(da)=g^*(\lambda -c)\). Hence \(g^*\in LS({{\mathbb {C}}}_+-c)\) and \(m^c_{g^*}(da)=e^{ca}m(da)\). Therefore, \( g^*(-L)\varphi =\int _0^\infty T^c_a\varphi \,m^c_{g^*}(da)=\int _0^\infty T_a\varphi \, m(da)\), \(\varphi \in X. \) Thus, the operator \(\varPhi (t,L)\) defined in (2.4) can be interpreted as \(\mathcal {C}_{T^c}\big (\varPhi (t,-(\cdot -c)) \big )\) in terms of Hille-Phillips calculus due to (4.1).
Now we are ready to transfer equality (4.2) to the level of operators by means of Hille-Phillips calculus. Let \((T_t)_{t\geqslant 0}\) be of type \(c\geqslant 0\) and \(\rho (L)\) be the resolvent set of operator L, i.e. the resolvent operator \(R_\lambda (L):=(\lambda -L)^{-1}\) is a well defined bounded operator on X for each \(\lambda \in \rho (L)\). Let \(\gamma >c\). Hence \(\gamma \in \rho (L)\). And equality (4.2) implies the equality: \(\forall t>0\), \(\forall \lambda \in \mathbb {C}: \Re \lambda \geqslant -c\)
Let us present each component of (4.4) as the Laplace transform of some complex measure on \(([0,\infty ),\mathcal {B}([0,\infty )))\). As we have already discussed
Furthermore, we have with Dirac delta-measure \(\delta _0\) and with exponential distribution \(Exp(\gamma )\):
Note that \(R_\gamma (L)\) is a bounded operator since \(\gamma \in (c,\infty )\subset \rho (L)\) (cf. [11, 28]) and \(\Vert \gamma R_\gamma (L)\varphi -\varphi \Vert _X\rightarrow 0\) as \(\gamma \rightarrow \infty \) for any \(\varphi \in X\). Further,
Note that \(L_\gamma \) is the so-called Yosida-Approximation of L (cf. [11, 28]); \(L_\gamma \) is a bounded operator and \(\Vert L\varphi -L_\gamma \varphi \Vert _X\rightarrow 0\) as \(\gamma \rightarrow +\infty \) for each \(\varphi \in \mathrm{Dom}(L)\).
Without loss of generality we now assume \(k(t,s)\ge 0\) (else divide into negative and nonnegative part) and define a family of measures on \(([0,\infty ),\mathcal {B}([0,\infty )))\) via
The right hand side in the above formula is well-defined since the mapping \(s\mapsto \mathcal {P}_{A(s)}(B)\) is a bounded Borel-measurable function on \([0,\infty )\) for any \(B\in \mathcal {B}([0,\infty ))\). Indeed, the mapping \(s\mapsto \varPhi (s,-\lambda )\) is Borel measurable for any \(\lambda \in {{\mathbb {C}}}\) due to Assumption 2 and representation formulas (2.1), (2.2). And for any \(s,x\geqslant 0\) holds (cf. Lemma 1.1 and the proof of Prop. 1.2 in [31])
Further, it holds for measurable \(g:[0,\infty )\rightarrow [0,\infty )\)
which can be seen via approximation of g by step functions from below and the use of Beppo Levi’s Theorem. By choosing \(g(a):=e^{-\lambda a}\) we see that
Thereby \(\Psi (t,-\lambda )\) is the Laplace transform of an appropriate measure and we get the correspondence
where \({\nu }^c_t(da):=e^{c a}\nu _t(da)\). Note that \({\nu }^c_t\) is a bounded measure on the measurable space \(([0,\infty ),\mathcal {B}([0,\infty )))\) due to (4.1). Furthermore, similar to property (4.5), it holds for any Bochner-integrable function \(g\,:\,[0,\infty )\rightarrow X\)
and therefore, for any \(\varphi \in X\),
Thereby, all components of (4.4) have been transferred. Taking everything together and using that for \(u_0\in \mathrm{Dom}(L)\) holds (cf. [16])
we get
Taking the limit \(\gamma \rightarrow +\infty \) we obtain (with \(\varPhi (s,L) L u_0=L \varPhi (s,L) u_0\) for all \(u_0\in \mathrm{Dom}(L)\))
Therefore, the function \(u(t):=\varPhi (t,L)u_0\) solves evolution equation (1.1) for any \(u_0\in \mathrm{Dom}(L)\).
For continuity at zero we evaluate equality (2.3) at \(\lambda =-c-i\rho \), \(\rho \in {{\mathbb {R}}}\), resulting in \( \int _0^\infty e^{i\rho a} e^{c a} \mathcal {P}_{A(t)}(da) = \varPhi (t,i\rho +c),\) \( \forall (t,\rho )\in [0,\infty )\times {{\mathbb {R}}}\). According to Lemma 1
and by Lévy’s Continuity Theorem it follows that \(e^{c a}\mathcal {P}_{A(t)}(da) \xrightarrow {\text {weakly}} \delta _0(da)\), \( t\searrow 0\). We now write
where \(f:{{\mathbb {R}}}\rightarrow {{\mathbb {R}}}\), \(f(a):=\Vert T_au_0-u_0\Vert _X e^{-c a}\) for \(a\geqslant 0\) and \(f(a):=0\) for \(a<0\), is a bounded and continuous function. Now
by weak convergence and thus continuity at zero is shown.
(iii) Let k be homogeneous of order \(\theta -1\) for some \(\theta >0\). By the recursion formula (2.2) for all \(t>0\), \(n\in \mathbb {N}\)
And, thus, we have for all \( t\geqslant 0\), \(\lambda \in \mathbb {C}\) (cf. Theorem 2 in [4]):
Let \(A(t):=At^\theta \), where A is a nonnegative random variable satisfying (2.6). Then \(\mathcal {L}(\mathcal {P}_{A(t)})(\lambda ) = \mathbb {E}\left[ e^{-\lambda At^\theta }\right] = \mathcal {L}(\mathcal {P}_A)(\lambda t^\theta ) = \varPhi (1,-\lambda t^\theta ) = \varPhi (t,-\lambda )\). Therefore, \(A(t):=A t^\theta \) has the required distribution. And Theorem 1 is proved. \(\square \)
Proof of Corollary 1
(i) \((T_t^f)_{t\geqslant 0}\) is a strongly continuous contraction semigroup on the Banach space X. Therefore, \((T_t^f)_{t\geqslant 0},\ k\) and \(\varPhi \) fulfill all assumptions of Theorem 1 and thus
is well-defined. Let now \((A(t))_{t\geqslant 0}\) and \((\eta ^f_t)_{t\geqslant 0}\) be as in the statement of Corollary 1. Consider the family of random variables \(\left( \eta ^f_{A(t)}\right) _{t\geqslant 0}\). Then
Starting with the strongly continuous contraction semigroup \((T_t)_{t\geqslant 0}\) and the completely monotone function \(\varPhi ^f(t,-\cdot ):=\varPhi (t,-f(\cdot ))\), one may define
Due to Fubini’s theorem
Therefore, for any \(u_0\in \mathrm{Dom}(L^f)\), equation (2.8) is solved by \(\varPhi (t,L^f)u_0=\varPhi ^f(t,L)u_0\) according to Theorem 1 (ii).
(ii) Since \(V\le 0\), \((T_t)_{t\geqslant 0}\) is a strongly continuous contraction semigroup and so is \((T_t^f)_{t\geqslant 0}\). It follows from Theorem 1 (ii) that \(u(t,x):=\varPhi (t, (L+V)^{f})u_0\) solves evolution equation (2.11) and due to Fubini’s theorem
(iii) Follows immediately from Theorem 1 (iii). \(\square \)
Proof of Proposition 1
First, note that, under our assumptions on \(\sigma \), the operator \((L_{(\sigma ,w)},\mathrm{Dom}(L_{(\sigma ,w)}))\) does generate a strongly continuous semigroup on \(C_\infty ({{\mathbb {R}}})\) (cf. [23], Sec. 3.1.2). Second, consider the pair \(\left( (\xi _t)_{t\geqslant 0},(\mathbb {P}^x)_{x\in {{\mathbb {R}}}}\right) \) where \((\xi _t)_{t\geqslant 0}\) solves the Stratonovich SDE with respect to a standard 1-dimensional Brownian motion \((B_t)_{t\geqslant 0}\)
with \(\xi _0=x\) under \(\mathbb {P}^x\). By Remark 5.2.22 in [20], the pair \(\left( (\xi _t)_{t\geqslant 0},(\mathbb {P}^x)_{x\in {{\mathbb {R}}}}\right) \) is a Markov process with generator \(L_{(\sigma ,w)}\). We apply the Doss-Sussmann technique to find an explicit expression for \((\xi _t)_{t\geqslant 0}\). So, let \((B_t)_{t\geqslant 0}\) be a standard 1-dimensional Brownian motion with respect to some probability measure \(\mathbb {P}\). Let \(g_\sigma \) be as in the statement of Proposition 1. Then, by the Itô formula for the Stratonovich integral
Hence \(\text {Law}\big ( (g_\sigma (B_t+wt,x))_{t\geqslant 0},\mathbb {P}\big )=\text {Law}\big ( (\xi _t)_{t\geqslant 0},\mathbb {P}^x \big )\) for every \(x\in {{\mathbb {R}}}\). In view of Corollary 1 and Example 3, there is a nonnegative random variable \(\mathcal {A}\) (constructed from k and \(\gamma \) as in (3.3)) which is independent of \((B_t)_{t\geqslant 0}\) and such that
solves the evolution equation (3.10). Note that \(\left( B_{\text{\AA} t^{\theta /\gamma }}+w\text{\AA} t^{\theta /\gamma } \right) _{t\geqslant 0}\), conditionally on \(\mathcal {A}\), is a Gaussian process with mean \(w\mathcal {A}t^{\theta /\gamma } \) and variance \(\text{\AA} t^{\theta /\gamma }\). The process \(\left( \sqrt{\mathcal {A}} B_{t^{\theta /\gamma }}+w\mathcal {A}t^{\theta /\gamma } \right) _{t\geqslant 0}\) and, if \(H:=\frac{\theta }{2\gamma }\in (0,1)\), the process \(\left( \sqrt{\mathcal {A}} B^H_{t}+w\text{\AA} t^{\theta /\gamma } \right) _{t\geqslant 0}\) have the same conditional law, where \((B^H_t)_{t\geqslant 0}\) is a 1-dimensional fractional Brownian motion independent of \(\mathcal {A}\). Hence Feynman-Kac formula (3.8) is shown. Further, we have
Hence Feynman-Kac formula (3.9) is shown. \(\square \)
References
Ascione, G., Mishura, Yu., Pirozzi, E.: Time-changed fractional Ornstein-Uhlenbeck process. Fract. Calc. Appl. Anal. 23(2), 450–483 (2020). https://doi.org/10.1515/fca-2020-0022
Bayin, S.S.: Time fractional Schrödinger equation: Fox’s H-functions and the effective potential. J. Math. Phys. 54(1), 012103, 18 (2013)
Bazhlekova, E.: Completely monotone multinomial Mittag-Leffler type functions and diffusion equations with multiple time-derivatives. Fract. Calc. Appl. Anal. 24(1), 88–111 (2021). https://doi.org/10.1515/fca-2021-0005
Bender, C., Butko Ya.A.: Stochastic solutions of generalized time-fractional evolution equations. Fract. Calc. Appl. Anal. 25(2), 488-519 (2022). https://doi.org/10.1007/s13540-022-00025-3
Bochner, S.: Diffusion equation and stochastic processes. Proc. Nat. Acad. Sci. U.S.A. 35, 368–370 (1949)
Chung, K.L., Zhao, Z.X.: From Brownian Motion to Schrödinger’s Equation. Springer, Berlin (1995)
Demuth, M., van Casteren, J.A.: Stochastic Spectral Theory for Selfadjoint Feller Operators. A Functional Integration Approach. Probability and its Applications. Birkhäuser, Basel (2000)
Doss, H.: Sur une résolution stochastique de l’équation de Schrödinger à coefficients analytiques. Comm. Math. Phys. 73(3), 247–264 (1980)
Doss, H.: On a probabilistic approach to the Schrödinger equation with a time-dependent potential. J. Funct. Anal. 260(6), 1824–1835 (2011)
Dubbeldam, J.L.A., Tomovski, Z., Sandev, T.: Space-time fractional Schrödinger equation with composite time fractional derivative. Fract. Calc. Appl. Anal. 18(5), 1179–1200 (2015). https://doi.org/10.1515/fca-2015-0068
Engel, K.J., Nagel, R.: One-Parameter Semigroups for Linear Evolution Equations. Springer, New York (2000)
Friz, P.K., Hairer, M.: A Course on Rough Paths. With an Introduction to Regularity Structures. Universitext. Springer, Cham (2020)
Gorenflo, R., Luchko, Yu.: Operational method for solving generalized Abel integral equation of second kind. Integral Transform. Spec. Funct. 5(1–2), 47–58 (1997)
Górka, P.a., Prado, H., Trujillo, J.: The time fractional Schrödinger equation on Hilbert space. Integral Equations Operator Theory 87(1), 1–14 (2017)
Haase, M.: The Functional Calculus for Sectorial Operators. Birkhäuser, Basel (2006)
Haase, M.: Lectures on Functional Calculus - 21st International Internet Seminar. https://www.math.uni-kiel.de/isem21/en/course/phase1/isem21-lectures-on-functional-calculus (2018)
Hadid, S.B., Luchko, Yu.: An operational method for solving fractional differential equations of an arbitrary real order. PanAmer. Math. J. 6(1), 57–73 (1996)
Hu, Y., Kallianpur, G.: Schrödinger equations with fractional Laplacians. Appl. Math. Optim. 42(3), 281–290 (2000)
Iomin, A.: Fractional time quantum mechanics. In: Handbook of Fractional Calculus with Applications. Vol. 5, pages 299–315. De Gruyter, Berlin (2019)
Karatzas, I., Shreve, S.E.: Brownian Motion and Stochastic Calculus Springer, New York (2013)
Knopova, V.: On the Feynman-Kac semigroup for some Markov processes. Mod. Stoch. Theory Appl. 2(2), 107–129 (2015)
Linnik, J.V.: Razlozheniya Veroyatnostnykh Zakonov (In Russian). Izdat. Leningrad. Univ, Leningrad (1960)
Lunardi, A.: Analytic Semigroups and Optimal Regularity in Parabolic Problems. Modern Birkhäuser Classics. Birkhäuser/Springer Basel AG, Basel (1995)
Metzler, R., Jeon, J.H., Cherstvy, A.G., Barkai, E.: Anomalous diffusion models and their properties: non-stationarity, non-ergodicity, and ageing at the centenary of single particle tracking. Phys. Chem. Chem. Phys. 16, 24128–24164 (2014)
Metzler, R., Klafter, J.: The random walk’s guide to anomalous diffusion: a fractional dynamics approach. Phys. Rep. 339(1), 77 (2000)
Mura, A., Mainardi, F.: A class of self-similar stochastic processes with stationary increments to model anomalous diffusion in physics. Integral Transforms Spec. Funct. 20(3–4), 185–198 (2009)
Pagnini, G.: Fractional kinetics in random/complex media. In: Handbook of Fractional Calculus with Applications. Vol. 5, pages 183–205. De Gruyter, Berlin (2019)
Pazy, A.: Semigroups of Linear Operators and Applications to Partial Differential Equations. Springer, New York (1983)
Petreska, I., de Castro, A.S.M., Sandev, T., Lenzi, E.K.: The time-dependent Schrödinger equation in three dimensions under geometric constraints. J. Math. Phys. 60(3), 032101, 8 (2019)
Phillips, R.S.: On the generation of semigroups of linear operators. Pacific J. Math. 2, 343–369 (1952)
Schilling, R.L., Song, R., Vondraček, Z.: Bernstein Functions. Theory and Applications. De Gruyter, Berlin (2012)
Acknowledgements
Yana Kinderknecht (Butko) thanks René Schilling for a fruitful discussion and interesting references.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
This article is published under an open access license. Please check the 'Copyright Information' section either on this page or in the PDF for details of this license and what re-use is permitted. If your intended use exceeds what is permitted by the license or if you are unable to locate the licence and re-use information, please contact the Rights and Permissions team.
About this article
Cite this article
Bender, C., Bormann, M. & Butko, Y.A. Subordination principle and Feynman-Kac formulae for generalized time-fractional evolution equations. Fract Calc Appl Anal 25, 1818–1836 (2022). https://doi.org/10.1007/s13540-022-00082-8
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13540-022-00082-8
Keywords
- Time-fractional evolution equations
- Subordination principle
- Feynman-Kac formulae
- Randomly scaled Gaussian processes
- Generalized grey Brownian motion
- Time-changed Markov processes
- Hille-Phillips functional calculus