betaFeedback

WorksMathematics

Geometry property of stochastic parabolic equations and Prékopa-Leindler inequality on bounded domains

Abstract

We study the preservation of spatial concavity and log-concavity for stochastic parabolic equations posed on bounded convex domains. For the linear noise, by using Minkowski convolutions, we establish the convexity preservation of solution almost surely. For the nonlinear noise, by using Itô formula, the impact of noise is given. Moreover, we establish the relationship between the weighted (stochastic) Prékopa-Leindler inequality and properties of (stochastic) parabolic equation on bounded domain, which is new even for deterministic case.

Introduction

Geometric properties of functions and solutions to partial differential equations are often intimately linked to analytic regularity. For instance, Taylor’s formula implies that

f∈C2,f′′(x)≥0 ∀x∈R⇒f(λx+(1−λ)y)≤λf(x)+(1−λ)f(y),f \in C^{2}, \qquad f^{\prime\prime}(x) \ge0\ \forall x \in\mathbb{R} \Rightarrow f(\lambda x + (1-\lambda)y) \le\lambda f(x) + (1-\lambda)f(y),

for all x,y∈Rx,y \in\mathbb{R} and λ∈(0,1)\lambda\in(0,1). Here the second derivative is a local analytic quantity, whereas convexity is a global geometric property. The converse implication, however, generally fails: a convex function need not be twice differentiable. This basic observation motivates the study of geometric properties of solutions to evolution equations, where regularity may be limited or degenerate. In this paper, we investigate the preservation of concavity and log-concavity for solutions to stochastic parabolic equations posed on bounded domains.

Concavity and log-concavity play a central role in the qualitative analysis of partial differential equations. They are closely related to stability, uniqueness and monotonicity phenomena, and arise naturally in problems from potential theory, heat propagation and nonlinear diffusion. A fundamental question in this direction is whether geometric properties of the initial datum are preserved along the evolution of the equation.

In the deterministic setting, concavity properties of elliptic and parabolic equations have been extensively studied. The quasi-concavity of equilibrium potentials for convex bodies was established in [14, 15, 37], while probabilistic methods were employed by Borell [2] to analyze concavity for heat conduction problems. The systematic theory of log-concave functions was developed by Brascamp and Lieb [4]. A general framework for concavity-preserving properties of partial differential equations was initiated by Korevaar in [30, 31]; see also [16, 29, 33] and the references therein. Further related results for parabolic and degenerate diffusion equations can be found in [1, 5, 6, 7, 8, 17, 19, 23], as well as in the context of porous medium equations [3, 9, 12, 18, 32].

For parabolic equations on bounded domains, Ishige and Salani [20] studied space-time concavity for Dirichlet boundary value problems under suitable structural assumptions on the nonlinear terms. Their approach relies on viscosity solutions combined with a weak comparison principle. Subsequent extensions were obtained in [21], but the analysis remains restricted to power-type concavity indices p∈(0,1)p \in(0,1). A different approach was proposed in [22], where spatial concavity for parabolic systems was established via an auxiliary parabolic system and a refined comparison principle. While effective for spatial concavity, this method does not cover the borderline log-concave case p=0p = 0.

The presence of stochastic perturbations introduces substantial additional difficulties. Random forcing typically destroys pointwise order structures and prevents the direct application of classical maximum principles. In particular, for stochastic parabolic equations driven by Brownian noise, one cannot expect simultaneous preservation of concavity in both space and time. This raises the natural question of whether spatial concavity or log-concavity can still be preserved for stochastic parabolic equations on bounded domains, and under what structural conditions such properties remain valid.

The first objective of the present paper is to address this question. We consider stochastic parabolic equations of the form

{du(t,x)=[Δu(t,x)+f(t,x,u(t,x))]dt+σ(t,x,u(t,x))dB(t),x∈D, t>0,u(t,x)=0,x∈∂D, t>0,u(0,x)=u0(x),x∈D,(1)\left\{ \begin{aligned} du(t,x) &= [\Delta u(t,x) + f(t,x,u(t,x))]dt + \sigma(t,x,u(t,x))dB(t), && x \in D,\ t > 0,\\ u(t,x) &= 0, && x \in\partial D,\ t > 0,\\ u(0,x) &= u_0(x), && x \in D, \end{aligned} \right. \tag*{(1)}

where D⊂RnD \subset\mathbb{R}^{n} is a bounded convex domain and B(t)B(t) is a one-dimensional standard Brownian motion. Our aim is to identify sufficient conditions under which spatial concavity or log-concavity of the initial datum is preserved along the stochastic evolution, and to clarify the role played by the noise structure and boundary conditions.

Our analysis distinguishes between linear and nonlinear noise. In the linear case, we develop a viscosity-solution framework combined with a comparison principle to establish almost sure preservation of spatial concavity under appropriate structural assumptions on the reaction term. This approach allows us to treat the borderline case p=0p = 0, which appears to be new even for deterministic parabolic equations on bounded domains. For nonlinear noise, the F-concavity is studied. By using Itô formula, we obtain the impact of noise. More precisely, the relationship between the reaction term and diffusion is considered to make the solution have the same property as the initial data. An example is given to illustrate our main result.

The second objective of the present paper is to exploit the connection between the generalized Prékopa-Leindler inequality and properties of parabolic equation on bounded domain. Notice that Prékopa-Leindler inequality is a special case of Borell-Brascamp-Lieb inequality, and we will study it in Section 4. Recently, Ishige et al. [26] studied the relationship between Borell-Brascamp-Lieb inequality and properties of diffusion equations of porous medium type. [28] considered the Brascamp-Lieb-type inequalities on weighted Riemannian manifolds with boundary. There is only a small amount of work on bounded domain research about the Borell-Brascamp-Lieb inequality. In this paper, inspired by [26], the relationship between a weighted Prékopa-Leindler inequality and properties of parabolic equation on bounded domain will be given, which is new even for deterministic case.

Our contributions can be summarized as follows.

  1. We establish sufficient conditions for the preservation of concavity for the Dirichlet problem associated with (1).

  1. The impact of noise on concavity is given.

  1. We establish the relationship between a weighted (stochastic) Prékopa-Leindler inequality and properties of (stochastic) parabolic equation on bounded domain.

The rest of the paper is organized as follows. Section 2 is devoted to the linear noise. In Section 3, we investigate the impact of noise on concavity of solutions. In Section 4, we establish the relationship between the generalized Prékopa-Leindler inequality and properties of parabolic equation.

Linear Multiplicative noise

Inspired by [20, 22], in this section we investigate the Dirichlet boundary value problem for stochastic parabolic equations of the form

du(t,x)=[Δu(t,x)+f(t,x,u(t,x))]dt+σ(t,x,u(t,x))dB(t),x∈D, t>0,u(t,x)=0,x∈∂D, t>0,u(0,x)=u0(x),x∈D.(2)\begin{aligned} du(t,x) &= \left[\Delta u(t,x)+f(t,x,u(t,x))\right]dt+\sigma(t,x,u(t,x))dB(t), \quad x\in D,\ t>0,\\ u(t,x) &= 0, \quad x\in\partial D,\ t>0,\\ u(0,x) &= u_0(x), \quad x\in D. \tag*{(2)} \end{aligned}

where D⊂RnD\subset\mathbb{R}^{n} is a bounded convex domain and B(t)B(t) is a one-dimensional standard Brownian motion. If the noise depends on spatial vector, then it is impossible to consider the concavity of spatial vector in the sense of almost surely.

In particular, we focus on noise of the form σ(t,x,u)=σ(u)\sigma(t,x,u)=\sigma(u). It was shown in [27] that, if f∈Lr([0,T]×D)f\in L^{r}([0,T]\times D) and σ\sigma satisfies a Lipschitz condition, then the solution of (2) remains positive almost surely; see [27 Theorem 4.4]. Let (Ω,F,P)(\Omega,\mathcal{F},\mathbb{P}) be a complete probability space with filtration (Ft)t∈[0,T](\mathcal{F}_{t})_{t\in[0,T]} generated by B(t)B(t). The initial value u0u_0 is a F0\mathcal{F}_0-measurable variable in L2(D)L^{2}(D). A predictable L2(D)L^{2}(D)-valued stochastic process {u(t):t∈[0,T]}\{u(t):t\in[0,T]\} is called a weak solution of (2) if for any v∈H1(D)v\in H^{1}(D) and almost every t∈[0,T]t\in[0,T], the following equality holds

(u(t),v)L2(D)=(u0,v)+∫0t[(−u(s),Δv)+(f(s,⋅,u(s)),v)]ds+∫0t(σ(s,⋅,u(s)),v)dB(s).(u(t),v)_{L^{2}(D)}=(u_0,v)+\int_{0}^{t}\left[(-u(s),\Delta v)+(f(s,\cdot,u(s)),v)\right]ds+\int_{0}^{t}(\sigma(s,\cdot,u(s)),v)dB(s).

We say uu is a strong solution of (2) if u∈L∞(0,T;L2(D))∩L2(0,T;H1(D))u\in L^{\infty}(0,T;L^{2}(D))\cap L^{2}(0,T;H^{1}(D)) with H1(D):={u∈L2(D),∇u∈L2(D)}H^{1}(D):=\{u\in L^{2}(D),\nabla u\in L^{2}(D)\} and ∥u∥H1(D)2:=∥u∥L2(D)2+∥∇u∥L2(D)2\|u\|_{H^{1}(D)}^{2}:=\|u\|_{L^{2}(D)}^{2}+\|\nabla u\|_{L^{2}(D)}^{2}. In this subsection, we investigate the preservation of spatial log-concavity for solutions to (2). Due to the technical reason, we only consider the case σ(t,x,u)=σu\sigma(t,x,u)=\sigma u, where σ\sigma is a constant. We first introduce the following definitions.

Definition 2.1. Let DD be a convex domain in Rn\mathbb{R}^n. A nonnegative function ff on DD is said to be pp-concave in DD if

f((1−λ)x+λy)≥Mp(f(x),f(y);λ)f\left((1-\lambda)x+\lambda y\right)\geq M_p(f(x),f(y);\lambda)

for all x,y∈Dx,y\in D and λ∈(0,1)\lambda\in(0,1), where Mp(a,b;λ)M_p(a,b;\lambda) denotes the pp-mean of two nonnegative numbers aa and bb with weight λ∈(0,1)\lambda\in(0,1), defined by

Mp(a,b;λ):={max⁡{a,b},if p=∞,((1−λ)ap+λbp)1/p,if p∈R∖{0},a1−λbλ,if p=0,min⁡{a,b},if p=−∞.M_p(a,b;\lambda):= \begin{cases} \max\{a,b\}, & \text{if }p=\infty,\\ \left((1-\lambda)a^p+\lambda b^p\right)^{1/p}, & \text{if }p\in\mathbb{R}\setminus\{0\},\\ a^{1-\lambda}b^\lambda, & \text{if }p=0,\\ \min\{a,b\}, & \text{if }p=-\infty. \end{cases}

for a,b>0a,b>0, and Mp(a,b;λ):=0M_p(a,b;\lambda):=0 if ab=0ab=0. Clearly, if p=1p=1, this reduces to the usual notion of concavity. We often refer to the cases p=0p=0 and p=−∞p=-\infty as log-concavity and quasi-concavity, respectively.

Throughout the subsection, we denote

B(x,r):={z∈Rn:∣z−x∣<r},x∈Rn, r>0.B(x,r):=\{z\in\mathbb{R}^n:|z-x|<r\},\qquad x\in\mathbb{R}^n,\ r>0.

For x=(x1,…,xn)x=(x_1,\ldots,x_n) and y=(y1,…,yn)∈Rny=(y_1,\ldots,y_n)\in\mathbb{R}^n, we write x⊗yx\otimes y for the n×nn\times n matrix with entries (xiyj)(x_i y_j), i,j=1,…,ni,j=1,\ldots,n. For m∈{2,3,…}m\in\{2,3,\ldots\}, we set

Λm:={λ=(λ1,…,λm)∈(0,1)m:∑i=1mλi=1}.\Lambda_m:=\left\{\lambda=(\lambda_1,\ldots,\lambda_m)\in(0,1)^m:\sum_{i=1}^m\lambda_i=1\right\}.

For a=(a1,…,am)∈(0,∞)ma=(a_1,\ldots,a_m)\in(0,\infty)^m, λ∈Λm\lambda\in\Lambda_m, and p∈[−∞,∞]p\in[-\infty,\infty], we define

Mp(a;λ):={[λ1a1p+⋯+λmamp]1/p,if p≠−∞,0,∞,max⁡{a1,…,am},if p=∞,a1λ1⋯amλm,if p=0,min⁡{a1,…,am},if p=−∞,M_p(a;\lambda):= \begin{cases} \left[\lambda_1a_1^p+\cdots+\lambda_ma_m^p\right]^{1/p}, & \text{if }p\ne-\infty,0,\infty,\\ \max\{a_1,\ldots,a_m\}, & \text{if }p=\infty,\\ a_1^{\lambda_1}\cdots a_m^{\lambda_m}, & \text{if }p=0,\\ \min\{a_1,\ldots,a_m\}, & \text{if }p=-\infty, \end{cases}

which is the (λ\lambda-weighted) pp-mean of aa; see [20]. Obviously, Mp(a;λ)M_p(a;\lambda) generalizes the two-point mean Mp(a,b;λ)M_p(a,b;\lambda).

In order to study the spatial concavity, the definition of the pp-concave λ\lambda-envelope will be introduced, see [22].

Definition 2.2. Let DD be a bounded convex domain in Rn\mathbb{R}^n, and let uu be a nonnegative function on [0,∞)×D‾[0,\infty)\times\overline{D}. Fix λ=(λ1,…,λn+1)∈Λn+1\lambda=(\lambda_1,\ldots,\lambda_{n+1})\in\Lambda_{n+1} and p∈[−∞,∞]p\in[-\infty,\infty]. For any (t,x)∈[0,∞)×D‾(t,x)\in[0,\infty)\times\overline{D}, set

up,λ(t,x):=sup⁡{Mp(u(t,y1),…,u(t,yn+1);λ):{yi}i=1n+1⊂D‾, x=∑i=1n+1λiyi}.u_{p,\lambda}(t,x):=\sup\left\{M_p\left(u(t,y_1),\ldots,u(t,y_{n+1});\lambda\right):\{y_i\}_{i=1}^{n+1}\subset\overline{D},\ x=\sum_{i=1}^{n+1}\lambda_i y_i\right\}.

We call up,λu_{p,\lambda} the pp-concave λ\lambda-envelope of uu.

In particular, by taking yi≡xy_i \equiv x in Definition 2.2, we immediately obtain

up,λ(t,x)≥u(t,x),(t,x)∈[0,∞)×Dˉ.u_{p,\lambda}(t,x) \ge u(t,x), \qquad(t,x) \in[0,\infty) \times\bar{D}.

Consequently, letting uu be the solution of (2), in order to prove that up,λu_{p,\lambda} coincides with uu and hence is a viscosity solution of (2), it suffices to show that

up,λ(t,x)≤u(t,x),(t,x)∈[0,∞)×Dˉ.u_{p,\lambda}(t,x) \le u(t,x), \qquad(t,x) \in[0,\infty) \times\bar{D}.

We now recall the definition of viscosity solutions to (2).

Definition 2.3. (i) An upper semicontinuous function uu on [0,∞)×D[0,\infty) \times D is called a viscosity subsolution of (2) if, for any (t0,x0)∈[0,∞)×D(t_0,x_0) \in[0,\infty) \times D and any test function ϕ∈C1,2([0,∞)×D)\phi\in C^{1,2}([0,\infty) \times D) such that

ϕ(t0,x0)=u(t0,x0)andϕ(t,x)≥u(t,x) in a neighborhood of (t0,x0) a.s.,\phi(t_0,x_0) = u(t_0,x_0) \quad\text{and} \quad\phi(t,x) \ge u(t,x) \text{ in a neighborhood of } (t_0,x_0) \text{ a.s.},

the following inequality holds at (t0,x0)(t_0,x_0) almost surely:

ϕ(t0,x0)≤ϕ(0,x0)+∫0t0Δϕ(s,x0) ds+∫0t0f(s,x0,ϕ(s,x0)) ds+∫0t0σ(s,x0,ϕ(s,x0)) dB(s).\phi(t_0,x_0) \le\phi(0,x_0) + \int_0^{t_0} \Delta\phi(s,x_0)\,\mathrm{d}s + \int_0^{t_0} f(s,x_0,\phi(s,x_0))\,\mathrm{d}s + \int_0^{t_0} \sigma(s,x_0,\phi(s,x_0))\,\mathrm{d}B(s).

(ii) A lower semicontinuous function uu on [0,∞)×D[0,\infty) \times D is called a viscosity supersolution of (2) if, for any (t0,x0)∈[0,∞)×D(t_0,x_0) \in[0,\infty) \times D and any test function ϕ∈C1,2([0,∞)×D)\phi\in C^{1,2}([0,\infty) \times D) such that

ϕ(t0,x0)=u(t0,x0)andϕ(t,x)≤u(t,x) in a neighborhood of (t0,x0) a.s.,\phi(t_0,x_0) = u(t_0,x_0) \quad\text{and} \quad\phi(t,x) \le u(t,x) \text{ in a neighborhood of } (t_0,x_0) \text{ a.s.},

the following inequality holds at (t0,x0)(t_0,x_0) almost surely:

ϕ(t0,x0)≥ϕ(0,x0)+∫0t0Δϕ(s,x0) ds+∫0t0f(s,x0,ϕ(s,x0)) ds+∫0t0σ(s,x0,ϕ(s,x0)) dB(s).\phi(t_0,x_0) \ge\phi(0,x_0) + \int_0^{t_0} \Delta\phi(s,x_0)\,\mathrm{d}s + \int_0^{t_0} f(s,x_0,\phi(s,x_0))\,\mathrm{d}s + \int_0^{t_0} \sigma(s,x_0,\phi(s,x_0))\,\mathrm{d}B(s).

Define

up(t,x):=sup⁡λ∈Λn+1up,λ(t,x).u_p(t,x) := \sup_{\lambda\in\Lambda_{n+1}} u_{p,\lambda}(t,x).

Then, to prove that the strong solution uu of (2) is pp-concave with respect to the spatial variable, it suffices to show that

up(t,x)=u(t,x),(t,x)∈[0,∞)×Dˉ.u_p(t,x) = u(t,x), \qquad(t,x) \in[0,\infty) \times\bar{D}.

In this paper, we focus on the log-concavity of solutions to (2), corresponding to the case p=0p=0. More precisely, our goal is to prove that

u0,λ(t,x)=u(t,x),(t,x)∈[0,∞)×Dˉ.u_{0,\lambda}(t,x) = u(t,x), \qquad(t,x) \in[0,\infty) \times\bar{D}.

For stochastic partial differential equations of the form (2), it is well known that if the reaction term ff is monotone and the noise coefficient σ\sigma is Lipschitz continuous, then a suitable comparison principle holds. Inspired by [22], we obtain the following result.

Theorem 2.1. Let DD be a bounded convex smooth domain in Rn\mathbb{R}^n and uu be the strong solution of (2). Assume that f(t,x,u)=f(u)f(t,x,u)=f(u) satisfies a Lipschitz condition and that σ(t,x,u)=σu\sigma(t,x,u)=\sigma u with a positive constant σ\sigma. Suppose that the function v↦e−vf(ev)v\mapsto e^{-v}f(e^v) is concave. Then u0,λu_{0,\lambda} is a viscosity subsolution of (2). Moreover, if the comparison principle holds for (2), then

u(t,x)=u0,λ(t,x),(t,x)∈[0,T]×D‾,u(t,x)=u_{0,\lambda}(t,x), \qquad(t,x)\in[0,T]\times\overline{D},

for any T>0T>0, and uu is log-concave with respect to the spatial variable.

Proof. It follows from the classical theory [10, 11] that under the assumptions of Theorem 2.1, the (2) admits a unique strong solution. Fix (t∗,x∗)∈(0,T]×D(t_*,x_*)\in(0,T]\times D and λ=(λ1,…,λn+1)∈Λn+1\lambda=(\lambda_1,\ldots,\lambda_{n+1})\in\Lambda_{n+1}. By [27 Theorem 4.4], we have

u(t∗,x∗)>0almost surely.u(t_*,x_*)>0 \quad\text{almost surely}.

The set

{{yi}i=1n+1⊂D‾:x∗=∑i=1n+1λiyi}\left\{\{y_i\}_{i=1}^{n+1}\subset\overline{D}:x_*=\sum_{i=1}^{n+1}\lambda_i y_i\right\}

is closed and bounded in D‾\overline{D}, and the mapping

(y1,…,yn+1)⟼M0(u(t∗,y1),…,u(t∗,yn+1);λ)(y_1,\ldots,y_{n+1})\longmapsto\mathcal{M}_0\bigl(u(t_*,y_1),\ldots,u(t_*,y_{n+1});\lambda\bigr)

is continuous. Using the boundary condition u(t,⋅)∣∂D=0u(t,\cdot)|_{\partial D}=0, we can choose {xi}i=1n+1⊂D\{x_i\}_{i=1}^{n+1}\subset D such that

x∗=∑i=1n+1λixix_*=\sum_{i=1}^{n+1}\lambda_i x_i

and

u0,λ(t∗,x∗)=M0(u(t∗,x1),…,u(t∗,xn+1);λ)=∏i=1n+1u(t∗,xi)λi.u_{0,\lambda}(t_*,x_*)=\mathcal{M}_0\bigl(u(t_*,x_1),\ldots,u(t_*,x_{n+1});\lambda\bigr)=\prod_{i=1}^{n+1}u(t_*,x_i)^{\lambda_i}.

By the Lagrange multiplier method, we obtain

∇u(t∗,x1)u(t∗,x1)=⋯=∇u(t∗,xn+1)u(t∗,xn+1)=:Θ.\frac{\nabla u(t_*,x_1)}{u(t_*,x_1)}=\cdots=\frac{\nabla u(t_*,x_{n+1})}{u(t_*,x_{n+1})}=:\Theta.

Set

u∗:=u0,λ(t∗,x∗),ui:=u(t∗,xi),yi(x):=xi+(x−x∗).u_*:=u_{0,\lambda}(t_*,x_*),\qquad u_i:=u(t_*,x_i),\qquad y_i(x):=x_i+(x-x_*).

Then x=∑i=1n+1λiyi(x)x=\sum_{i=1}^{n+1}\lambda_i y_i(x). Define

ϕ(t,x):=M0(u(t,y1(x)),…,u(t,yn+1(x));λ)=∏i=1n+1u(t,yi(x))λi.\phi(t,x):=\mathcal{M}_0\bigl(u(t,y_1(x)),\ldots,u(t,y_{n+1}(x));\lambda\bigr)=\prod_{i=1}^{n+1}u(t,y_i(x))^{\lambda_i}.

The function ϕ\phi is C2C^2 with respect to the spatial variable almost surely and satisfies ϕ(t∗,x∗)=u∗\phi(t_*,x_*)=u_*. Moreover, by the definition of u0,λu_{0,\lambda},

u0,λ(t,x)≥ϕ(t,x)u_{0,\lambda}(t,x)\geq\phi(t,x)

in a neighborhood of (t∗,x∗)(t_*,x_*) almost surely. Since ff satisfies the Lipschitz condition, the classical comparison principle can be applied to (2), see [34]. It therefore suffices to show that

ϕ(t∗,x∗)≤ϕ(0,x∗)+∫0t∗Δϕ(s,x∗) ds+∫0t∗f(ϕ(s,x∗)) ds+∫0t∗σ(s,x∗,ϕ(s,x∗)) dB(s).(3)\phi(t_*,x_*) \leq\phi(0,x_*)+\int_{0}^{t_*}\Delta\phi(s,x_*)\,\mathrm{d}s+\int_{0}^{t_*}f(\phi(s,x_*))\,\mathrm{d}s+\int_{0}^{t_*}\sigma(s,x_*,\phi(s,x_*))\,\mathrm{d}B(s). \tag*{(3)}

A direct computation yields

dϕ(t∗,x∗)=ϕ(t∗,x∗)∑i=1n+1λidu(t∗,xi)ui,\mathrm{d}\phi(t_*,x_*)=\phi(t_*,x_*)\sum_{i=1}^{n+1}\lambda_i\frac{\mathrm{d}u(t_*,x_i)}{u_i},
∇ϕ(t∗,x∗)=ϕ(t∗,x∗)∑i=1n+1λi∇u(t∗,xi)ui=Θ,\nabla\phi(t_*,x_*)=\phi(t_*,x_*)\sum_{i=1}^{n+1}\lambda_i\frac{\nabla u(t_*,x_i)}{u_i}=\Theta,
Δϕ(t∗,x∗)=ϕ(t∗,x∗)∑i=1n+1λiΔu(t∗,xi)ui.\Delta\phi(t_*,x_*)=\phi(t_*,x_*)\sum_{i=1}^{n+1}\lambda_i\frac{\Delta u(t_*,x_i)}{u_i}.

where we used the identity

∑i=1n+1λi(λi−1)+∑1≤i,j≤n+1i≠jλiλj=0.\sum_{i=1}^{n+1}\lambda_i(\lambda_i-1)+\sum_{\substack{1\leq i,j\leq n+1\\i\ne j}}\lambda_i\lambda_j=0.

Consequently,

ϕ(t∗,x∗)−ϕ(0,x∗)−∫0t∗Δϕ(s,x∗) ds−∫0t∗f(ϕ(s,x∗)) ds−∫0t∗σ(s,x∗,ϕ(s,x∗)) dB(s)=∫0t∗[ϕ(s,x∗)∑i=1n+1λif(ui)ui−f(ϕ(s,x∗))]ds.(4)\begin{aligned} &\phi(t_*,x_*)-\phi(0,x_*)-\int_{0}^{t_*}\Delta\phi(s,x_*)\,\mathrm{d}s-\int_{0}^{t_*}f(\phi(s,x_*))\,\mathrm{d}s-\int_{0}^{t_*}\sigma(s,x_*,\phi(s,x_*))\,\mathrm{d}B(s)\\ &=\int_{0}^{t_*}\left[\phi(s,x_*)\sum_{i=1}^{n+1}\lambda_i\frac{f(u_i)}{u_i}-f(\phi(s,x_*))\right]\mathrm{d}s. \tag*{(4)} \end{aligned}

Let vi=log⁡uiv_i=\log u_i. By the assumed concavity of v↦e−vf(ev)v\mapsto e^{-v}f(e^v), we have

∑i=1n+1λif(ui)ui=∑i=1n+1λie−vif(evi)≤e−∑i=1n+1λivif(e∑i=1n+1λivi)=ϕ(s,x∗)−1f(ϕ(s,x∗)).\sum_{i=1}^{n+1}\lambda_i\frac{f(u_i)}{u_i}=\sum_{i=1}^{n+1}\lambda_i e^{-v_i}f(e^{v_i})\leq e^{-\sum_{i=1}^{n+1}\lambda_i v_i}f\left(e^{\sum_{i=1}^{n+1}\lambda_i v_i}\right)=\phi(s,x_*)^{-1}f(\phi(s,x_*)).

Substituting this inequality into (4) yields

ϕ(t∗,x∗)−ϕ(0,x∗)−∫0t∗Δϕ(s,x∗) ds−∫0t∗f(ϕ(s,x∗)) ds−∫0t∗σ(s,x∗,ϕ(s,x∗)) dB(s)≤0a.s.\phi(t_*,x_*)-\phi(0,x_*)-\int_{0}^{t_*}\Delta\phi(s,x_*)\,\mathrm{d}s-\int_{0}^{t_*}f(\phi(s,x_*))\,\mathrm{d}s-\int_{0}^{t_*}\sigma(s,x_*,\phi(s,x_*))\,\mathrm{d}B(s)\leq0\quad\text{a.s.}

This proves (3) and completes the proof. □

Remark 2.1. (1) The assumption σ(u)=σu\sigma(u)=\sigma u avoids difficulties caused by the lack of order preservation for stochastic integrals. More precisely, even if random variables X>YX>Y almost surely, one cannot in general conclude that

X dB(t)≥Y dB(t)a.s.X\,\mathrm{d}B(t)\geq Y\,\mathrm{d}B(t)\quad\text{a.s.}

In this case σ(u)=σu\sigma(u)=\sigma u, the noise term disappears after a logarithmic transformation, allowing us to deduce almost sure spatial log-concavity.

(2) By [27], solutions of (2) remain positive almost surely. Therefore, the heat energy

H(t)=∫DE[u(t,x)] dxH(t)=\int_D \mathbb{E}[u(t,x)]\,\mathrm{d}x

is well defined and is log-concave with respect to time.

(3) An example of a function ff satisfying the structural condition in Theorem 2.1 is

f(u)=−1+u2,u>0.f(u)=-\sqrt{1+u^2},\qquad u>0.

(4) If σ=0\sigma=0, Theorem 2.1 also applies to the deterministic equation (2), yielding preservation of spatial log-concavity in the deterministic setting.

Nonlinear Multiplicative Noise

In this section, we study the impact of noise of geometry property of parabolic equation on bounded domain. For simplicity, we only consider scalar case. Consider

{du(t,x)=[Δu(t,x)+f(u(t,x),∇u(t,x))]dt+σ(u(t,x),∇u(t,x))dB(t),x∈D, t>0,u(t,x)=0,x∈∂D, t>0,u(0,x)=u0(x),x∈D.(5)\begin{cases} \mathrm{d}u(t,x)=\left[\Delta u(t,x)+f(u(t,x),\nabla u(t,x))\right]\mathrm{d}t+\sigma(u(t,x),\nabla u(t,x))\mathrm{d}B(t), & x\in D,\ t>0,\\ u(t,x)=0, & x\in\partial D,\ t>0,\\ u(0,x)=u_0(x), & x\in D. \tag*{(5)} \end{cases}

where B(t)B(t) is one-dimensional Brownian motion similar to that in (2). It follows from section 2 that it is impossible to consider the geometry property of equation (5) almost surely. Now, we are ready to consider the moment of the solution of (5).

By using Itô formula, we have, for a function ψ\psi,

dψ(u)=ψ′(u)du+12ψ′′(u)du⋅du=[ψ′(u)Δu+ψ′(u)f(u,∇u)+12ψ′′(u)σ2(u)]dt+ψ′(u)σ(u,∇u)dB(t)=[Δψ(u)−ψ′′(u)∣∇u∣2+ψ′(u)f(u,∇u)+12ψ′′(u)σ2(u,∇u)]dt+ψ′(u)σ(u,∇u)dB(t).\begin{aligned} \mathrm{d}\psi(u)&=\psi'(u)\mathrm{d}u+\frac{1}{2}\psi''(u)\mathrm{d}u\cdot\mathrm{d}u\\ &=\left[\psi'(u)\Delta u+\psi'(u)f(u,\nabla u)+\frac{1}{2}\psi''(u)\sigma^2(u)\right]\mathrm{d}t+\psi'(u)\sigma(u,\nabla u)\mathrm{d}B(t)\\ &=\left[\Delta\psi(u)-\psi''(u)|\nabla u|^2+\psi'(u)f(u,\nabla u)+\frac{1}{2}\psi''(u)\sigma^2(u,\nabla u)\right]\mathrm{d}t+\psi'(u)\sigma(u,\nabla u)\mathrm{d}B(t). \end{aligned}

Obviously, if, for any (t,x)∈(0,∞)×D(t,x)\in(0,\infty)\times D, it holds that

−ψ′′(u)∣∇u∣2+ψ′(u)f(u,∇u)+12ψ′′(u)σ2(u,∇u)=0,-\psi''(u)|\nabla u|^2+\psi'(u)f(u,\nabla u)+\frac{1}{2}\psi''(u)\sigma^2(u,\nabla u)=0,

then letting v=ψ(u)v=\psi(u) and denote u=ψ−1(v)u=\psi^{-1}(v) (we assume the inverse function exists), we get

{dv(t,x)=Δv(t,x)dt+ψ′(ψ−1(v(t,x)))σ(ψ−1(v(t,x)),∇ψ−1(v(t,x)))dB(t),v(t,x)=ψ(0),v(0,x)=ψ(u0(x)).(6)\begin{cases} \mathrm{d}v(t,x)=\Delta v(t,x)\mathrm{d}t+\psi'(\psi^{-1}(v(t,x)))\sigma(\psi^{-1}(v(t,x)),\nabla\psi^{-1}(v(t,x)))\mathrm{d}B(t),\\ v(t,x)=\psi(0),\\ v(0,x)=\psi(u_0(x)). \tag*{(6)} \end{cases}

Moreover, if we take w=E[v]−ψ(0)w=\mathbb{E}[v]-\psi(0), then ww satisfies

{∂tw(t,x)=Δw(t,x),x∈D, t>0,w(t,x)=0,x∈∂D, t>0,w(0,x)=ψ(u0(x))−ψ(0),x∈D.(7)\begin{cases} \partial_t w(t,x)=\Delta w(t,x), & x\in D,\ t>0,\\ w(t,x)=0, & x\in\partial D,\ t>0,\\ w(0,x)=\psi(u_0(x))-\psi(0), & x\in D. \tag*{(7)} \end{cases}

Ishige et al. [24] considered the F-concavity of Dirichlet flow (7). In order to write it clearly, following [24, 25], we introduce the notion of FF-concavity.

Definition 3.1. Let a∈(0,∞]a \in(0,\infty], and set I=[0,a)I=[0,a) and int⁡I=(0,a)\operatorname{int} I=(0,a). A function F:I→[−∞,∞)F:I\to[-\infty,\infty) is said admissible on II if F∈C(int⁡I)F\in C(\operatorname{int} I), FF is strictly increasing on II, and F(0)=−∞F(0)=-\infty. Let A(I)\mathcal{A}(I) be the set of II-valued functions in Rn\mathbb{R}^{n}. Let FF be admissible on II. For any f∈A(I)f\in\mathcal{A}(I), we say that ff is FF-concave in DD if

F(f(λx+(1−λ)y))≥λF(f(x))+(1−λ)F(f(y))F(f(\lambda x+(1-\lambda)y))\geq\lambda F(f(x))+(1-\lambda)F(f(y))

for x,y∈Dx,y\in D and λ∈(0,1)\lambda\in(0,1). We denote by C[F]C[F] the set of FF-concave functions in Rn\mathbb{R}^{n}.

Let

h(z):=(4π)−12∫0∞e−∣z−w∣24 dwfor z∈R.h(z):=(4\pi)^{-\frac{1}{2}}\int_{0}^{\infty}e^{-\frac{|z-w|^{2}}{4}}\,dw\quad\text{for }z\in\mathbb{R}.

Then the function hh is smooth in R\mathbb{R}, lim⁡z→−∞h(z)=0\lim_{z\to-\infty}h(z)=0, lim⁡z→∞h(z)=1\lim_{z\to\infty}h(z)=1, and h′>0h'>0 in R\mathbb{R}. Thus the inverse function of hh exists, and denote by HH. Define an admissible function HaH_a on [0,a)[0,a) by

Ha(r):={H(ra)for r∈(0,a) if a>0,log⁡rfor r∈(0,a) if a=∞,−∞for r=0 and a∈(0,∞].H_a(r):= \begin{cases} H\left(\frac{r}{a}\right) & \text{for }r\in(0,a)\text{ if }a>0,\\ \log r & \text{for }r\in(0,a)\text{ if }a=\infty,\\ -\infty& \text{for }r=0\text{ and }a\in(0,\infty]. \end{cases}

”The F-concavity is preserved by Dirichlet heat flow (DHF)” means that the solution of (7) is F-concavity if the initial data is F-concavity. As for (7), we have the following result.

Theorem 3.1. Let I=[0,a)I=[0,a) with a∈(0,∞]a\in(0,\infty] and DD be a convex domain in Rn\mathbb{R}^{n} with n≥1n\geq1. Then the following properties hold:

(i) HaH_a-concavity is the strongest FF-concavity preserved by DHF in AD\mathcal{A}_D;

(ii) Log-concavity is the weakest FF-concavity preserved by DHF in AD\mathcal{A}_D for n≥2n\geq2; when n=1n=1, the same is true under C2C^{2}-regularity assumption on FF;

(iii) When n≥2n\geq2, if starting with an initial datum which is not log-concave, even quasi-concavity can be immediately broken by the corresponding solution of problem (7), so losing every possible reminiscence of concavity.

The proof of Theorem 3.1 is the exact same as that of Theorem 1.5 of [24].

Now, we want to show the impact of noise on F-concavity. It follows from (ii) of of Theorem 3.1 that if ψ(u0(x))−ψ(0)\psi(u_0(x))-\psi(0) is log-concavity, then solution ww is also log-concavity. Consider

{du(t,x)=[Δu(t,x)−u(t,x)−22∣∇u(t,x)∣] dt+[u(t,x)+2∣∇u(t,x)∣] dB(t),x∈D, t>0,u(t,x)=0,x∈∂D, t>0,u(0,x)=u0(x),x∈D,(8)\begin{cases} du(t,x)=[\Delta u(t,x)-u(t,x)-2\sqrt{2}|\nabla u(t,x)|]\,dt\\ \quad+[u(t,x)+\sqrt{2}|\nabla u(t,x)|]\,dB(t),\quad x\in D,\ t>0,\\ u(t,x)=0,\quad x\in\partial D,\ t>0,\\ u(0,x)=u_0(x),\quad x\in D, \tag*{(8)} \end{cases}

Let ψ(x)=x2\psi(x)=x^{2}. Direct calculations show that w=E[ψ(u)]−ψ(0)w=\mathbb{E}[\psi(u)]-\psi(0) satisfies

{∂tw(t,x)=Δw(t,x),x∈D, t>0,w(t,x)=0,x∈∂D, t>0,w(0,x)=u0(x)2,x∈D.\begin{cases} \partial_t w(t,x)=\Delta w(t,x),\quad x\in D,\ t>0,\\ w(t,x)=0,\quad x\in\partial D,\ t>0,\\ w(0,x)=u_0(x)^2,\quad x\in D. \end{cases}

Obviously, it follows from Theorem 3.1 that if u02u_0^2 satisfies Log-concavity, then the solution E[u2(t,x)]\mathbb{E}[u^2(t,x)] is also Log-concavity in spatial variable. But if σ=0\sigma=0, then w(t,x)=u2(t,x)w(t,x)=u^2(t,x) satisfies

{∂tw(t,x)=Δw(t,x)−∣∇w(t,x)∣22w(t,x)−2w(t,x)−2∣∇w(t,x)∣,x∈D, t>0,w(t,x)=0,x∈∂D, t>0,w(0,x)=u0(x)2,x∈D.(9)\begin{cases} \partial_t w(t,x)=\Delta w(t,x)-\dfrac{|\nabla w(t,x)|^2}{2w(t,x)}-2w(t,x)-\sqrt{2}|\nabla w(t,x)|, & x\in D,\ t>0,\\ w(t,x)=0, & x\in\partial D,\ t>0,\\ w(0,x)=u_0(x)^2, & x\in D. \tag*{(9)} \end{cases}

Set

F(w,∇w,Δw)=Δw−∣∇w∣22w−2w−2∣∇w∣.F(w,\nabla w,\Delta w)=\Delta w-\frac{|\nabla w|^2}{2w}-2w-\sqrt{2}|\nabla w|.

Then it follows from [22 Theorem 4.3] that the solution of (9) is Log-concavity under the condition that, for any fixed θ∈Rn\theta\in\mathbb{R}^n,

e−wF(ew,ewθ,ewA)e^{-w}F(e^w,e^w\theta,e^w A)

is convex with respect to (w,A)∈R×Sn(w,A)\in\mathbb{R}\times S^n, where SnS^n denotes the space of n×nn\times n real symmetric matrices. Unfortunately,

e−wF(ew,ewθ,ewA)=A−e−w∣θ∣−2−2∣θ∣,e^{-w}F(e^w,e^w\theta,e^w A)=A-e^{-w}|\theta|-2-\sqrt{2}|\theta|,

which is not convex because the term −e−w-e^{-w} is concave. Therefore, the noise can help the solution keep the same property as initial data.

Generalized Prékopa-Leindler inequality

In Sections 2, we established preservation results for spatial concavity and log-concavity of solutions to stochastic parabolic equations on bounded convex domains. It is well known that geometric properties of parabolic flows are closely connected with functional inequalities of Brunn–Minkowski type. In the whole space Rn\mathbb{R}^n, the Prékopa-Leindler inequality can be interpreted as an integral manifestation of the log-concavity preserving property of the heat semigroup.

Motivated by this perspective, we show in this section that the log-concavity preserving structure of the Dirichlet heat flow on bounded convex domains yields a generalized Prékopa-Leindler inequality involving the first Dirichlet eigenfunction. This provides a functional counterpart to the pointwise geometric preservation results obtained earlier.

We first recall the Borell-Brascamp-Lieb inequality. Let n∈Nn\in\mathbb{N}, λ∈(0,1)\lambda\in(0,1) and α∈[−1/n,∞]\alpha\in[-1/n,\infty]. Assume that nonnegative integrable functions ff, gg and hh in Rn\mathbb{R}^n satisfy

h((1−λ)y+λz)≥Mα(f(y),g(z);λ)h((1-\lambda)y+\lambda z)\ge M_\alpha(f(y),g(z);\lambda)

for almost all y,z∈Rny,z\in\mathbb{R}^n. Then

∫Rnh(x) dx≥Mα1+nα(∫Rnf(x) dx,∫Rng(x) dx;λ).(10)\int_{\mathbb{R}^n}h(x)\,dx\ge M_{\frac{\alpha}{1+n\alpha}}\left(\int_{\mathbb{R}^n}f(x)\,dx,\int_{\mathbb{R}^n}g(x)\,dx;\lambda\right). \tag*{(10)}

When α=0\alpha=0, this reduces to the classical Prékopa-Leindler inequality:

∫Rnh(x) dx≥(∫Rnf(x) dx)1−λ(∫Rng(x) dx)λ.(11)\int_{\mathbb{R}^n}h(x)\,dx\ge\left(\int_{\mathbb{R}^n}f(x)\,dx\right)^{1-\lambda}\left(\int_{\mathbb{R}^n}g(x)\,dx\right)^\lambda. \tag*{(11)}

Consider the Dirichlet heat equation

{∂tu(t,x)=Δu(t,x),x∈D, t>0,u(t,x)=0,x∈∂D, t>0,u(0,x)=u0(x),x∈D,(12)\begin{cases} \partial_t u(t,x)=\Delta u(t,x), & x\in D,\ t>0,\\ u(t,x)=0, & x\in\partial D,\ t>0,\\ u(0,x)=u_0(x), & x\in D, \tag*{(12)} \end{cases}

where D⊂RnD\subset\mathbb{R}^n is bounded convex smooth. The solution admits the spectral representation

u(t,x)=∑n=1∞cne−λntϕn(x),(13)u(t,x)=\sum_{n=1}^{\infty}c_n e^{-\lambda_n t}\phi_n(x), \tag*{(13)}

where cn=∫Du0(x)ϕn(x) dxc_n=\int_D u_0(x)\phi_n(x)\,dx and (λn,ϕn)(\lambda_n,\phi_n) are the Dirichlet eigenpairs of −Δ-\Delta:

−Δϕn=λnϕn,ϕn∣∂D=0,-\Delta\phi_n=\lambda_n\phi_n,\qquad\phi_n|_{\partial D}=0,

with

0<λ1<λ2≤λ3≤⋯ .0<\lambda_1<\lambda_2\leq\lambda_3\leq\cdots.

Theorem 4.1. Let DD be bounded convex smooth. Assume ψ0,ψ1,ψλ∈C(D)\psi_0,\psi_1,\psi_\lambda\in C(D) are nonnegative and satisfy

ψλ(λx+(1−λ)y)≥ψ0(x)λψ1(y)1−λ.(14)\psi_\lambda(\lambda x+(1-\lambda)y)\geq\psi_0(x)^\lambda\psi_1(y)^{1-\lambda}. \tag*{(14)}

Let uiu_i solve (12) with initial data ψi\psi_i. Then

uλ(t,λx+(1−λ)y)≥u0(t,x)λu1(t,y)1−λ(15)u_\lambda(t,\lambda x+(1-\lambda)y)\geq u_0(t,x)^\lambda u_1(t,y)^{1-\lambda} \tag*{(15)}

for all x,y∈Dx,y\in D and t>0t>0.

Proof. Define

u0,λ(t,x)=sup⁡{M0(u0(t,y),u1(t,z);λ):y,z∈D‾, x=λy+(1−λ)z}.u_{0,\lambda}(t,x)=\sup\left\{M_0(u_0(t,y),u_1(t,z);\lambda):y,z\in\overline{D},\ x=\lambda y+(1-\lambda)z\right\}.

By the same viscosity-envelope argument used in the proof of Theorem 2.1, replacing u(t,xi)u(t,x_i) by ui(t,xi)u_i(t,x_i), one verifies that u0,λu_{0,\lambda} is a viscosity solution of (12).

By uniqueness of solutions to the Dirichlet heat equation, u0,λu_{0,\lambda} coincides with the solution of (12) with initial data

ψ0,λ=sup⁡{M0(ψ0(y),ψ1(z);λ):y,z∈D‾, x=λy+(1−λ)z}.\psi_{0,\lambda}=\sup\left\{M_0(\psi_0(y),\psi_1(z);\lambda):y,z\in\overline{D},\ x=\lambda y+(1-\lambda)z\right\}.

Condition (14) implies ψλ≥ψ0,λ\psi_\lambda\geq\psi_{0,\lambda}. By the comparison principle,

uλ(t,λx+(1−λ)y)≥u0,λ(t,x).u_\lambda(t,\lambda x+(1-\lambda)y)\geq u_{0,\lambda}(t,x).

By definition of u0,λu_{0,\lambda},

u0,λ(t,x)≥u0(t,x)λu1(t,y)1−λ.u_{0,\lambda}(t,x)\geq u_0(t,x)^\lambda u_1(t,y)^{1-\lambda}.

The proof is complete. □\square

Using the asymptotic dominance of the first eigenfunction in (13), we now derive a weighted Prékopa-Leindler inequality on DD.

Theorem 4.2. Assume nonnegative integrable functions f,g,hf,g,h on convex domain DD satisfy

h((1−λ)y+λz)≥f(y)λg(z)1−λh((1-\lambda)y+\lambda z)\ge f(y)^\lambda g(z)^{1-\lambda}

for almost all y,z∈Dy,z\in D. Then

∫Dh(x)ϕ1(x) dx≥(∫Df(x)ϕ1(x) dx)λ(∫Dg(x)ϕ1(x) dx)1−λ.(16)\int_D h(x)\phi_1(x)\,dx\ge\left(\int_D f(x)\phi_1(x)\,dx\right)^\lambda\left(\int_D g(x)\phi_1(x)\,dx\right)^{1-\lambda}. \tag*{(16)}

Proof. First assume f,g,h∈C(D)f,g,h\in C(D). Let uλ,u0,u1u_\lambda,u_0,u_1 solve (12) with initial data h,f,gh,f,g. By Theorem 4.1,

∑n=1∞cn,λe−λntϕn(x)≥(∑n=1∞cn,0e−λntϕn(x))λ(∑n=1∞cn,1e−λntϕn(x))1−λ.\sum_{n=1}^{\infty}c_{n,\lambda}e^{-\lambda_n t}\phi_n(x)\ge\left(\sum_{n=1}^{\infty}c_{n,0}e^{-\lambda_n t}\phi_n(x)\right)^\lambda\left(\sum_{n=1}^{\infty}c_{n,1}e^{-\lambda_n t}\phi_n(x)\right)^{1-\lambda}.

Write

ui(t,x)=e−λ1t(c1,iϕ1(x)+∑n=2∞cn,ie−(λn−λ1)tϕn(x)).u_i(t,x)=e^{-\lambda_1t}\left(c_{1,i}\phi_1(x)+\sum_{n=2}^{\infty}c_{n,i}e^{-(\lambda_n-\lambda_1)t}\phi_n(x)\right).

Since λn>λ1\lambda_n>\lambda_1 for n≥2n\ge2, the higher modes decay exponentially. Dividing by e−λ1te^{-\lambda_1t} and letting t→∞t\to\infty, we obtain

c1,λϕ1(x)≥(c1,0ϕ1(x))λ(c1,1ϕ1(x))1−λ.c_{1,\lambda}\phi_1(x)\ge(c_{1,0}\phi_1(x))^\lambda(c_{1,1}\phi_1(x))^{1-\lambda}.

Recalling

c1,i=∫Dfi(x)ϕ1(x) dx,c_{1,i}=\int_D f_i(x)\phi_1(x)\,dx,

we obtain (16). The extension to L1L^1 data follows by approximation.

We now return to the stochastic setting considered in § 2. Consider

{du(t,x)=Δu(t,x) dt+σu(t,x) dB(t),u(t,x)∣∂D=0.(17)\begin{cases} du(t,x)=\Delta u(t,x)\,dt+\sigma u(t,x)\,dB(t),\\ u(t,x)|_{\partial D}=0. \tag*{(17)} \end{cases}

Define

v(t,x)=e−σB(t)u(t,x).v(t,x)=e^{-\sigma B(t)}u(t,x).

Then vv satisfies

∂tv(t,x)=Δv(t,x)−σ22v(t,x).\partial_t v(t,x)=\Delta v(t,x)-\frac{\sigma^2}{2}v(t,x).

Let w(t,x)=eσ22tv(t,x)w(t,x)=e^{\frac{\sigma^2}{2}t}v(t,x). Then ww solves the heat equation (12) almost surely. Applying Theorem 4.2 pathwise yields:

Theorem 4.3. Let f,g,hf,g,h be nonnegative integrable random functions satisfying

h((1−λ)y+λz)≥f(y)λg(z)1−λh((1-\lambda)y+\lambda z)\ge f(y)^\lambda g(z)^{1-\lambda}

almost surely. Then

∫Dh(x)ϕ1(x) dx≥(∫Df(x)ϕ1(x) dx)λ(∫Dg(x)ϕ1(x) dx)1−λ\int_D h(x)\phi_1(x)\,dx\ge\left(\int_D f(x)\phi_1(x)\,dx\right)^\lambda\left(\int_D g(x)\phi_1(x)\,dx\right)^{1-\lambda}

almost surely.

In the end of this section, we give another method to prove the inequality (16). First, we use the zero extension method to obtain the Prékopa-Leindler inequality on bounded domain. Assume the assumptions of Theorem 4.2 hold, then define

ws(x)={w(x),if x∈D,0,if x∈Rn∖D,w=f or g or h.w_s(x)= \begin{cases} w(x), & \text{if }x\in D,\\ 0, & \text{if }x\in\mathbb{R}^{n}\setminus D, \end{cases} \qquad w=f\text{ or }g\text{ or }h.

Then fs,gs,hsf_s,g_s,h_s satisfy the assumption of the Prékopa-Leindler inequality in whole space. That is to say, fs,gs,hsf_s,g_s,h_s are nonnegative integrable functions satisfying

h((1−λ)y+λz)≥f(y)λg(z)1−λh((1-\lambda)y+\lambda z)\ge f(y)^{\lambda}g(z)^{1-\lambda}

for almost all y,z∈Rny,z\in\mathbb{R}^{n}. Consequently, we have

∫Rnhs(x) dx≥(∫Rnfs(x) dx)λ(∫Rngs(x) dx)1−λ.\int_{\mathbb{R}^{n}}h_s(x)\,\mathrm{d}x\ge\left(\int_{\mathbb{R}^{n}}f_s(x)\,\mathrm{d}x\right)^{\lambda}\left(\int_{\mathbb{R}^{n}}g_s(x)\,\mathrm{d}x\right)^{1-\lambda}.

which implies

∫Dh(x) dx≥(∫Df(x) dx)λ(∫Dg(x) dx)1−λ.\int_D h(x)\,\mathrm{d}x\ge\left(\int_D f(x)\,\mathrm{d}x\right)^{\lambda}\left(\int_D g(x)\,\mathrm{d}x\right)^{1-\lambda}.

Note that the first eigenvalue function ϕ1\phi_1 is Log-concavity on convex domain [30, 31], i.e.,

ϕ1((1−λ)y+λz)≥ϕ1(y)λϕ1(z)1−λ\phi_1((1-\lambda)y+\lambda z)\ge\phi_1(y)^{\lambda}\phi_1(z)^{1-\lambda}

for almost all y,z∈Dy,z\in D. Combining the assumptions on f,g,hf,g,h, we get

(hϕ1)((1−λ)y+λz)≥(fϕ1)(y)λ(gϕ1)(z)1−λ(h\phi_1)((1-\lambda)y+\lambda z)\ge(f\phi_1)(y)^{\lambda}(g\phi_1)(z)^{1-\lambda}

for almost all y,z∈Dy,z\in D. Applying the Prékopa-Leindler inequality on bounded domain, we get (16). However, in this paper, we give a new methd to get the inequality (16).

Use of AI tools declaration

We did not use any AI tools to assist in the derivation and writing of this paper.

Acknowledgements

This work was supported by the NSF of China grants 12171247.

Data Availability Statements

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Conflicts of Interest

The authors declare that they have no competing interests.

References

  1. [1]U. Alan, Power concavity and boundary value problems, Indiana Univ. Math. J. 34(3) (1985) 687–704.
  2. [2]C. Borell, Brownian motion in a convex ring and quasiconcavity, Comm. Math. Phys. 86(1) (1982) 143–147.DOI
  3. [3]P. Bénilan, J. L. Vázquez, Concavity of solutions of the porous medium equation, Trans. Amer. Math. Soc. 299(1) (1987) 81–93.
  4. [4]H. Brascamp, E. Lieb, On extensions of the Brunn-Minkowski and Prékopa-Leindler theorems, including inequalities for log-concave functions, J. Funct. Anal. 22(4) (1976) 366–389.
  5. [5]A. Chau, B. Weinkove, Strong space-time convexity and the heat equation, Indiana Univ. Math. J. 70(4) (2021) 1189–1210.arxiv.org/abs/1903.03208
  6. [6]A. Chau, B. Weinkove, The Stefan problem and concavity, Calc. Var. Partial Differential Equations 60(2) (2021) 176.arxiv.org/abs/2004.04284
  7. [7]C. Chen, B. Hu, A microscopic convexity principle for space-time convex solutions of fully nonlinear parabolic equations, Acta Math. Sin. Engl. Ser. 29(4) (2013) 651–674.DOI
  8. [8]C. Chen, X. Ma, P. Salani, On space-time quasiconcave solutions of the heat equation, Mem. Amer. Math. Soc. 259 (2019) 1244.arxiv.org/abs/1405.6373
  9. [9]A. Chau, B. Weinkove, Non-preservation of power concavity for the porous medium equation, Adv. Math. 440 (2024) 109520.
  10. [10]P-L. Chow, Stochastic partial differential equations, Chapman Hall/CRC Applied Mathematics and Nonlinear Science Series. Chapman Hall/CRC, Boca Raton, FL, 2007. x+281 pp. ISBN: 978-1-58488-443-9.
  11. [11]G. Da Prato and J. Zabczyk, Stochastic equations in infinite dimensions, Cambridge University Press, Cambridge, 1992.
  12. [12]P. Daskalopoulos, R. Hamilton, K. Lee, All-time C∞-regularity of the interface in degenerate diffusion: a geometric approach, Duke Math. J. 108(2) (2001) 295–327.DOI
  13. [14]R. Gabriel, An extended principle of the maximum for harmonic functions in three dimensions, J. London Math. Soc. 30 (1955) 388–401.
  14. [15]R. Gabriel, A result concerning convex level surfaces of three-dimensional harmonic functions, J. London Math. Soc. 32 (1957) 286–294.
  15. [16]A. Greco, B. Kawohl, Log-concavity in some parabolic problems, Electron. J. Differential Equations 1999 (1999) 1–12.
  16. [17]K. Ishige, P. Salani, Is quasi-concavity preserved by heat flow? Arch. Math. (Basel) 90(5) (2008) 450–460.DOI
  17. [18]K. Ishige, P. Salani, Convexity breaking of the free boundary for porous medium equations, Interfaces Free Bound. 12(1) (2010) 75–84.DOI
  18. [19]K. Ishige, P. Salani, Parabolic quasi-concavity for solutions to parabolic problems in convex rings, Math. Nachr. 283(11) (2010) 1526–1548.
  19. [20]K. Ishige, P. Salani, Parabolic power concavity and parabolic boundary value problems, Math. Ann. 358(3–4) (2014) 1091–1117.
  20. [21]K. Ishige, K. Nakagawa, P. Salani, Power concavity in weakly coupled elliptic and parabolic systems, Nonlinear Anal. 131 (2016) 81–97.
  21. [22]K. Ishige, K. Nakagawa, P. Salani, Spatial concavity of solutions to parabolic systems, Ann. Sc. Norm. Super. Pisa Cl. Sci. 20(1) (2020) 291–313.
  22. [23]K. Ishige, P. Salani, A. Takatsu, New characterizations of log-concavity via Dirichlet heat flow, Ann. Mat. Pura Appl. 201(4) (2022) 1531–1552.
  23. [24]K. Ishige, P. Salani, A. Takatsu, Characterization of F-concavity preserved by the Dirichlet heat flow. Trans. Amer. Math. Soc. 377(8) (2024), 5705–5748.
  24. [25]K. Ishige, Eventual concavity properties of the heat flow. Math. Ann. 390 (2024), 5883–5922.
  25. [26]K. Ishige, Q. Liu, P. Salani, A parabolic PDE-based approach to Borel-Brascamp-Lieb inequality, Math. Ann. 392 (2025) 4891–4937.
  26. [27]N. Kavallaris, Y. Yan, Finite-time blow-up of a nonlocal stochastic parabolic problem, Stochastic Process. Appl. 130 (2020) 5605–5635.
  27. [28]A. Kolesnikov, E. Milman, Brascamp-Lieb-type inequalities on weighted Riemannian manifolds with boundary, J. Geom. Anal. 27(2) (2017) 1680–1702.
  28. [29]A. Kolesnikov, On diffusion semigroup preserving the log-concavity, J. Funct. Anal. 186 (2001) 196–205.
  29. [30]N. Korevaar, Capillary surface convexity above convex domains, Indiana Univ. Math. J. 32 (1983) 73–82.
  30. [31]N. Korevaar, Convex solutions to nonlinear elliptic and parabolic boundary value problems, Indiana Univ. Math. J. 32 (1983) 603–614.
  31. [32]K. Lee, J. Vazquez, Geometrical properties of solutions of the porous medium equation for large times, Indiana Univ. Math. J. 52(4) (2003) 991–1016.
  32. [33]K. Lee, Power concavity on nonlinear parabolic flows, Comm. Pure Appl. Math. 58(11) (2005) 1529–1543.DOI
  33. [34]G. Lv, J. Duan, Impacts of noise on a class of partial differential equations, J. Differential Equations 258(6) (2015) 2196–2220.
  34. [37]J. Lewis, Capacitary functions in convex rings, Arch. Rational Mech. Anal. 66 (1977) 201–224.DOI

Paper details

vlnc:W_b78fzc6c
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY-SA 4.0
Usage

views · 0

downloads · 2

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents