betaFeedback

WorksMathematics

On the domain of attraction of source–type solutions of the porous medium equations

Abstract

We study the rate of convergence to Barenblatt profiles of the solutions to porous medium equations in Rn,n≥1\R^n, n \ge 1, by means of a certain class of Energy distances, a class of discrepancies between probability measures which have been recently widely used in problems related to AI, in particular for tasks related to machine learning. If n≥3n \ge 3, it is shown both that the rate of decay in Energy distance is improved with respect to known results which are based on the decay of relative entropy, and that, with respect to this distance, any initial value of bounded mass belongs to the domain of attraction of the source–type solution. Similar results are shown to hold for the fast diffusion case.

Introduction

The study of the asymptotic behavior of the solutions f(x,t)f(x,t) of the nonlinear diffusion equation

∂f(x,t)∂t=Δfm(x,t),m>n−2n(1)\frac{\partial f(\mathbf{x},t)}{\partial t} = \Delta f^{m}(\mathbf{x},t), \qquad m > \frac{n-2}{n} \tag*{(1)}

where x∈Rn\mathbf{x} \in\mathbb{R}^{n}, n≥1n \ge1, is a classical problem which, starting with the porous medium case, characterized by m>1m > 1, has been originally studied in the 70’s of the last century in a pioneering paper by Kamin [21] (cf. [34, 36] for an exhaustive presentation of the results obtained on this topic up to the recent years). In a physical sense, the solution f(⋅,t)f(\cdot,t) to (1) represents the time-density of the diffusion process.

If we consider the Cauchy Problem for equation (1) posed in Rn×R+\mathbb{R}^n \times\mathbb{R}_+ with integrable and nonnegative initial data

f0(x)∈L1(Rn),f0(x)≥0(2)f_0(\mathbf{x}) \in L^1(\mathbb{R}^n), \qquad f_0(\mathbf{x}) \geq0 \tag*{(2)}

the source-type solutions of equation (1) provide the asymptotic representation of a large class of solutions starting from f0f_0 satisfying (2) [34].

The source-type or Barenblatt solutions of equation (1) are given by [5]

b(x,t,C)=t−nαFC(t−αx)(3)b(\mathbf{x},t,C)=t^{-n\alpha}F_C\left(t^{-\alpha}\mathbf{x}\right) \tag*{(3)}

where CC is a positive free constant, and FCF_C, the profile, has the form

FC(y)=(C−κ∣y∣2)+1/(m−1),α=1n(m−1)+2,κ=m−12mα.(4)F_C(\mathbf{y})=\left(C-\kappa|\mathbf{y}|^2\right)_+^{1/(m-1)}, \qquad\alpha=\frac{1}{n(m-1)+2}, \qquad\kappa=\frac{m-1}{2m}\alpha. \tag*{(4)}

In (4), the function (g)+(g)_+ denotes as usual the positive part of gg.

In this range of the parameter mm, for a given initial value f0f_0 of given mass MM, the constant C=C(f0)C=C(f_0) can be chosen to satisfy the rule of mass conservation

M=∫Rnf0(x) dx=∫Rnb(x,t,C) dx.(5)M=\int_{\mathbb{R}^n}f_0(\mathbf{x})\,\mathrm{d}\mathbf{x}=\int_{\mathbb{R}^n}b(\mathbf{x},t,C)\,\mathrm{d}\mathbf{x}. \tag*{(5)}

Further, in presence of additional conditions on the initial density, such as boundedness of the relative (to the Barenblatt solution FCF_C) Ralston-type entropy of order mm [10], or the boundedness of the relative Renyi entropy of order mm [9, 26], it is possible to reckon an explicit rate of convergence to the right source-type solution in L1(Rn)L^1(\mathbb{R}^n). It is interesting to note that the boundedness of energy allows to establish a direct connection between the asymptotic stabilization property of the porous medium flow and the stabilization to the Gaussian profile embodied in the classical central limit theorem [33]. Moreover, the boundedness of energy is essential to characterize in a precise way the best matching Barenblatt profile [14, 15]

Diffusion processes described by (1) arise in various fields, such as plasma physics, kinetic theory of gases, solid state physics, filtration models and transport in porous media. In some cases, including many metals and ceramic materials the diffusion coefficient can, over a wide range of temperatures, be approximated as ∣x∣ν|\mathbf{x}|^\nu for some exponent ν\nu such that 0<ν<20<\nu<2 [25], which leads to an exponent m<1m<1 in (1). Also, the special case ν=1\nu=1 emerges in the central limit approximation to Carleman’s model of the Boltzmann equation, according to [23, 24]. It is interesting to remark that, while less relevant from the point of view of applications, in mathematics the fast diffusion case m<1m<1 is better known than the porous medium case for its properties in connection with functional analysis.

In what follows, we approach the problem of identifying an explicit rate of convergence of the solution to equation (1) to the source-type solution in a new way, which requires only that the initial datum has a finite mass. This broadened range of attraction of the Barenblatt solution can be obtained only if the distance in which the explicit rate of convergence is proven is weaker than the classical L1L^1-distance; moreover, while as usual the rate of convergence is closely dependent on the space dimension nn, the proposed approach works in this generality only in dimension n≥3n \ge3.

To this end, we make use of arguments similar to the ones recently introduced in [4], which are based on applying Cramér distance [11] to one-dimensional Fokker–Planck equations with linear drift and nonlinear diffusion. As extensively discussed in [4], the Cramér distance—also known as the continuous ranked probability score—has gained renewed attention in AI-related problems, particularly in machine learning [7], where it is used for tasks such as distribution estimation, regression, and image generation. As first remarked in [7], the main motivation for adopting the Cramér distance, a metric for comparing one-dimensional probability distributions, lies in its advantageous properties, most notably its unbiased sample gradients, which make it especially amenable to optimization via stochastic gradient descent.

As we shall detail in this paper, the results proven in [4] for one-dimensional Fokker–Planck type equations follow by resorting to properties typical of Cramér’s distance, closely linked to some of the various forms in which this distance can be expressed. As a matter of fact, an interesting related distance, which, as shown in [1], shares most properties of Cramér’s distance, while designed to also compare multivariate probability distributions, is the Energy distance introduced by Székely [30, 31, 32] by resorting to one of the alternative expressions of Cramér’s distance.

The classical Energy distance, of wide use in Statistics, has been recently generalized in [1], where the relationships with other well-known metrics—such as the Wasserstein distance [37] and the Fourier-based metrics [9, 18]—were thoroughly analyzed, in reason of their applicability in the field of artificial intelligence [2, 3, 12, 27], as well as in the context of partial differential equations [13, 16, 17].

In this paper, we show that a certain class of the new Energy–type distances considered in [1] are very-well adapted to study the rate of convergence to the source–type solutions to non linear diffusion equations. These metrics are directly connected with the Cramér distance in the one-dimensional case. However, at variance with the one-dimensional Cramér distance, this class of nn-dimensional Energy distances do not work well on drift operators for n>1n > 1, and cannot be directly applied to study the large-time behavior of multi-dimensional Fokker–Planck equations.

In more details, in § 2 we briefly report the definitions and the main properties of both Cramér and Energy distances, as well as their connections. The equivalence between these two distances in the one-dimensional case relies on arguments first applied to obtain an alternative definition of Gini index [38]. Then, in § 3 we will show how a particular class of these distances can be fruitfully employed to study convergence to source–type solutions to the porous medium equation (1), departing from any initial value of bounded mass. § 4 will be devoted to proving that an explicit rate of convergence follows as soon as n≥3n \ge3. In § 5, we will show that a similar argument can be used to study the convergence rate to source–type solutions in the fast diffusion case m<1m < 1 if mm belongs to a certain range and if n≥3n \ge3. In Section 6, we will outline the conclusions of the paper.

From Cramér to Energy distances

Let ff be a probability density over Rn\mathbb{R}^n, n≥1n \ge1. We will associate ff with a random variable X\mathbf{X}, such that for a subset of the reals A⊆RnA \subseteq\mathbb{R}^n, we have

P(X∈A)=∫Af(x) dx.P(\mathbf{X} \in A) = \int_A f(\mathbf{x})\,\mathrm{d}x.

Given s>0s > 0, we denote by Ps(Rn,n≥1)P_s(\mathbb{R}^n, n \ge1) the class of all probability densities ff over Rn\mathbb{R}^n such that

ms(f)=∫Rn∣x∣sf(x) dx<+∞.m_s(f) = \int_{\mathbb{R}^n} |\mathbf{x}|^s f(\mathbf{x})\,\mathrm{d}x < +\infty.

If s=1s = 1, the quantity m1(f)m_1(f) defines the (absolute) mean value of X\mathbf{X}, say E(∣X∣)E(|\mathbf{X}|).

Given a probability density ff in Ps(Rn)P_s(\mathbb{R}^n), n≥1n \ge1, its Fourier transform is defined by

f^(ξ)=∫Re−2πiξ⋅xf(x) dx.\hat{f}(\boldsymbol{\xi}) = \int_{\mathbb{R}} e^{-2\pi i\boldsymbol{\xi}\cdot\mathbf{x}} f(\mathbf{x})\,\mathrm{d}x.

If n=1n = 1, the (cumulative) distribution function of the one-dimensional random variable XX is given by

F(x)=P(X≤x)=∫−∞xf(x) dx.(6)F(x) = P(X \le x) = \int_{-\infty}^{x} f(x)\,\mathrm{d}x. \tag*{(6)}

Given two one-dimensional random variables XX and YY and their cumulative distribution functions FF and GG, the Cramér distance between XX and YY is expressed by [11]

d(F,G)=∫R[F(x)−G(x)]2 dx.(7)d(F,G) = \int_{\mathbb{R}} [F(x)-G(x)]^2\,\mathrm{d}x. \tag*{(7)}

Note that, as written, Cramér distance is not a metric, while its square root is.

One of the interesting properties of Cramér’s distance is that it can be equivalently written in different ways. It is immediate to show that it can be written in terms of Fourier transforms. Indeed, owing to Parseval’s theorem, Cramér distance (7) can be equivalently written as

d(F,G)=∫R∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ.(8)d(F,G) = \int_{\mathbb{R}} \frac{|\hat{f}(\xi)-\hat{g}(\xi)|^2}{|\xi|^2}\,\mathrm{d}\xi. \tag*{(8)}

Also, assuming that the one-dimensional random variables XX and YY are mutually independent, and that X′X' and Y′Y' are independent copies of XX and YY, Cramér’s distance can be expressed as

d(F,G)=E(∣X−Y∣)−12E(∣X−X′∣)−12E(∣Y−Y′∣).(9)d(F,G) = E(|X-Y|) - \frac{1}{2}E(|X-X'|) - \frac{1}{2}E(|Y-Y'|). \tag*{(9)}

Note that (9) implies that the Cramér distance of FF and GG is bounded as soon as F,G∈P1(R)F,G\in P_1(\mathbb{R}). This third way of expressing Cramér’s distance has been shown in [32] (cf. [4] for a detailed proof based on the Gini index [19, 20]).

Expressing the Cramér distance as in (9) allows us to compute a weak metric between two probability distributions in closed form on every Rn\mathbb{R}^n, with n>1n>1. In a series of papers [30, 31, 32], Székely introduced the definition of Energy distance between two multidimensional random variables X\mathbf{X} and Y\mathbf{Y} with densities f(x)f(\mathbf{x}), and, respectively, g(x)g(\mathbf{x}), with x∈Rn\mathbf{x}\in\mathbb{R}^n, n≥1n\ge1. The Energy distance in Rn\mathbb{R}^n, n≥1n\ge1, of order α\alpha, with 0<α<20<\alpha<2 is defined by

En,α(f,g)=2∬Rn×Rn∣x−y∣αf(x)g(y) dxdy−∬Rn×Rn∣x−y∣αf(x)f(y) dxdy−∬Rn×Rn∣x−y∣αg(x)g(y) dxdy=−∬Rn×Rn∣x−y∣α[f(x)−g(x)][f(y)−g(y)] dxdy.(10)\begin{aligned} E_{n,\alpha}(f,g)={}&2\iint_{\mathbb{R}^n\times\mathbb{R}^n} \lvert\mathbf{x}-\mathbf{y}\rvert^\alpha f(\mathbf{x})g(\mathbf{y})\,\mathrm{d}\mathbf{x}\mathrm{d}\mathbf{y}\\ &-\iint_{\mathbb{R}^n\times\mathbb{R}^n} \lvert\mathbf{x}-\mathbf{y}\rvert^\alpha f(\mathbf{x})f(\mathbf{y})\,\mathrm{d}\mathbf{x}\mathrm{d}\mathbf{y} -\iint_{\mathbb{R}^n\times\mathbb{R}^n} \lvert\mathbf{x}-\mathbf{y}\rvert^\alpha g(\mathbf{x})g(\mathbf{y})\,\mathrm{d}\mathbf{x}\mathrm{d}\mathbf{y}\\ ={}&-\iint_{\mathbb{R}^n\times\mathbb{R}^n} \lvert\mathbf{x}-\mathbf{y}\rvert^\alpha[f(\mathbf{x})-g(\mathbf{x})][f(\mathbf{y})-g(\mathbf{y})]\,\mathrm{d}\mathbf{x}\mathrm{d}\mathbf{y}. \tag*{(10)} \end{aligned}

As for the Cramér distance, in dimension n>1n>1 the Energy distance of order α\alpha can be equivalently expressed in Fourier transform. In this case it holds [1, 32]

En,α(f,g)=cn,α∫Rn∣f^(ξ)−g^(ξ)∣2∣ξ∣n+α dξ,(11)E_{n,\alpha}(f,g)=c_{n,\alpha}\int_{\mathbb{R}^n}\frac{\lvert\widehat{f}(\boldsymbol{\xi})-\widehat{g}(\boldsymbol{\xi})\rvert^2}{\lvert\boldsymbol{\xi}\rvert^{n+\alpha}}\,\mathrm{d}\boldsymbol{\xi}, \tag*{(11)}

where the constant cn,αc_{n,\alpha}, determined as in [29] takes the value

cn,α=α2n+α−1πn/2Γ(n+α2)Γ(2−α2).(12)c_{n,\alpha}=\alpha2^{n+\alpha-1}\pi^{n/2}\frac{\Gamma\left(\frac{n+\alpha}{2}\right)}{\Gamma\left(\frac{2-\alpha}{2}\right)}. \tag*{(12)}

where Γ\Gamma is the Gamma function.

The passage to Fourier transform is the easiest way to verify that right-hand side in (10) is nonnegative, so that twice the mixed integral dominates the sum of the last two. Further, expression (18) allows one to relate in a precise way the energy distance of order α\alpha to homogeneous Sobolev spaces of fractional order.

Let us note that in one dimension, and for α=1\alpha=1, the Energy distance E1,1(f,g)E_{1,1}(f,g) is nothing but twice the Cramér distance, as defined by (9), so that

d(f,g)=12E1,1(f,g)=2∬R×R∣x−y∣2f(x)g(y) dxdy−∬R×R∣x−y∣2f(x)f(y) dxdy−∬R×R∣x−y∣2g(x)g(y) dxdy=−∬R×R∣x−y∣2[f(x)−g(x)][f(y)−g(y)] dxdy.(13)\begin{aligned} d(f,g)&=\frac{1}{2}E_{1,1}(f,g)=2\iint_{\mathbb{R}\times\mathbb{R}}\frac{\lvert x-y\rvert}{2}f(x)g(y)\,\mathrm{d}x\mathrm{d}y\\ &\quad-\iint_{\mathbb{R}\times\mathbb{R}}\frac{\lvert x-y\rvert}{2}f(x)f(y)\,\mathrm{d}x\mathrm{d}y-\iint_{\mathbb{R}\times\mathbb{R}}\frac{\lvert x-y\rvert}{2}g(x)g(y)\,\mathrm{d}x\mathrm{d}y\\ &=-\iint_{\mathbb{R}\times\mathbb{R}}\frac{\lvert x-y\rvert}{2}[f(x)-g(x)][f(y)-g(y)]\,\mathrm{d}x\mathrm{d}y. \tag*{(13)} \end{aligned}

Learning from Cramér distance

In what follow, we will show how Cramér distance works in connection with the porous medium equation (1). We assume that the initial value for the Cauchy problem for (1) satisfies conditions (2), and all solutions satisfy well-known regularity properties that are enough to justify all the computation that follows [35].

A key property of identity (13) is that for any given probability density h∈P1(R)h \in P_1(\mathbb{R}) it holds

d2dx2∫R∣x−y∣2h(y) dy=h(x),(14)\frac{d^2}{dx^2}\int_{\mathbb{R}}\frac{|x-y|}{2}h(y)\,dy=h(x), \tag*{(14)}

namely that the function

H(x)=d2dx2∣x∣2H(x)=\frac{d^2}{dx^2}\frac{|x|}{2}

coincides with the delta function δ0(x)\delta_0(x) centered in x=0x=0.

This implies that, if f(x,t)f(x,t) and g(x,t)g(x,t) are the two probability density functions solving the one-dimensional porous medium equation (1) departing from f0(x)f_0(x) (respectively g0(x)g_0(x)) satisfying (2) with mass equal to one, and such that d(f0,g0)d(f_0,g_0) is bounded, integrating twice by parts one has

ddtd(f(t),g(t))=12ddtE1,1(f(t),g(t))=−ddt∫R×R∣x−y∣2[f(x,t)−g(x,t)][f(y,t)−g(y,t)] dx dy=−2∫R×R∣x−y∣2ddt[f(x,t)−g(x,t)][f(y,t)−g(y,t)] dx dy=−2∫R×R∣x−y∣2[f(y,t)−g(y,t)]d2dx2[fm(x,t)−gm(x,t)] dx dy=−2∫R[fm(x,t)−gm(x,t)]d2dx2{∫R∣x−y∣2[f(y,t)−g(y,t)] dy} dx=−2∫R[fm(x,t)−gm(x,t)][f(x,t)−g(x,t)] dx≤0.(15)\begin{aligned} \frac{d}{dt}d(f(t),g(t))&=\frac{1}{2}\frac{d}{dt}E_{1,1}(f(t),g(t))\\ &=-\frac{d}{dt}\int_{\mathbb{R}\times\mathbb{R}}\frac{|x-y|}{2}[f(x,t)-g(x,t)][f(y,t)-g(y,t)]\,dx\,dy\\ &=-2\int_{\mathbb{R}\times\mathbb{R}}\frac{|x-y|}{2}\frac{d}{dt}[f(x,t)-g(x,t)][f(y,t)-g(y,t)]\,dx\,dy\\ &=-2\int_{\mathbb{R}\times\mathbb{R}}\frac{|x-y|}{2}[f(y,t)-g(y,t)]\frac{d^2}{dx^2}[f^m(x,t)-g^m(x,t)]\,dx\,dy\\ &=-2\int_{\mathbb{R}}[f^m(x,t)-g^m(x,t)]\frac{d^2}{dx^2}\left\{\int_{\mathbb{R}}\frac{|x-y|}{2}[f(y,t)-g(y,t)]\,dy\right\}\,dx\\ &=-2\int_{\mathbb{R}}[f^m(x,t)-g^m(x,t)][f(x,t)-g(x,t)]\,dx\leq0. \tag*{(15)} \end{aligned}

Hence, the Cramér distance between any two solutions to the porous medium equation is decreasing in time.

The key point in deriving this result is that the kernel ∣x∣/2|x|/2 in the definition of Cramér distance satisfies (14).

This result can be easily generalized to any dimension n>1n>1 by resorting to the concept of fundamental solution of the Laplacian Δ\Delta in Rn\mathbb{R}^n, namely a function Φ(x)\Phi(x) that satisfies in Rn\mathbb{R}^n the equation ΔΦ(x)=δ0(x)\Delta\Phi(x)=\delta_0(x) in the sense of distributions [28]. The fundamental solutions read

Φn(x)={1(2−n)∣Sn−1∣∣x∣2−n,n≥3,12πlog⁡∣x∣,n=2,x∈Rn∖{0},(16)\Phi_n(x)= \begin{cases} \frac{1}{(2-n)|S_{n-1}|}|x|^{2-n}, & n\geq3,\\ \frac{1}{2\pi}\log|x|, & n=2, \end{cases} \qquad x\in\mathbb{R}^n\setminus\{0\}, \tag*{(16)}

where ∣Sn−1∣|S_{n-1}| denotes the surface area of the unit sphere in Rn\mathbb{R}^n, given by

∣Sn−1∣=2πn/2Γ(n2).|S_{n-1}|=\frac{2\pi^{n/2}}{\Gamma\left(\frac{n}{2}\right)}.

Owing to (16) we introduce, for n≥3n \ge3, a (normalized) variant of the Energy distance between two nn-dimensional probability densities in the form

En(f,g)=2∫Rn×Rn∣x−y∣2−n(2−n)∣Sn−1∣f(x)g(y) dxdy−∫Rn×Rn∣x−y∣2−n(2−n)∣Sn−1∣f(x)f(y) dxdy−∫Rn×Rn∣x−y∣2−n(2−n)∣Sn−1∣g(x)g(y) dxdy=∫Rn×Rn∣x−y∣2−n(n−2)∣Sn−1∣[f(x)−g(x)][f(y)−g(y)] dxdy.(17)\mathcal{E}_n(f,g)=2\int_{\mathbb{R}^n\times\mathbb{R}^n}\frac{|\mathbf{x}-\mathbf{y}|^{2-n}}{(2-n)|S_{n-1}|}f(x)g(y)\,dxdy-\int_{\mathbb{R}^n\times\mathbb{R}^n}\frac{|\mathbf{x}-\mathbf{y}|^{2-n}}{(2-n)|S_{n-1}|}f(x)f(y)\,dxdy-\int_{\mathbb{R}^n\times\mathbb{R}^n}\frac{|\mathbf{x}-\mathbf{y}|^{2-n}}{(2-n)|S_{n-1}|}g(x)g(y)\,dxdy =\int_{\mathbb{R}^n\times\mathbb{R}^n}\frac{|\mathbf{x}-\mathbf{y}|^{2-n}}{(n-2)|S_{n-1}|}[f(x)-g(x)][f(y)-g(y)]\,dxdy. \tag*{(17)}

The Energy distance of negative order 2−n2-n, with n≥3n \ge3, defined in (17) is a particular case of the energy distances with negative exponent introduced and studied in [1]. As shown in in [1], the distance En\mathcal{E}_n is, up to a multiplicative constant cn,αc_{n,\alpha} given in (12), equal to a Fourier based metric, which in this case gives

En(f,g)=∫Rn∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ,(18)\mathcal{E}_n(f,g)=\int_{\mathbb{R}^n}\frac{|\widehat{f}(\xi)-\widehat{g}(\xi)|^2}{|\xi|^2}\,d\xi, \tag*{(18)}

so that En(f,g)\mathcal{E}_n(f,g) is the square of the distance in the Sobolev space H˙−1(Rn)\dot{H}_{-1}(\mathbb{R}^n).

The decay in Energy distance

Let us now consider two probability density functions f(x,t)f(x,t) and g(x,t)g(x,t), solutions of the nn-dimensional porous medium equation (1) departing from f0(x)f_0(x) (respectively g0(x)g_0(x)) and such that En(f0,g0)\mathcal{E}_n(f_0,g_0) is bounded. By means of Green’s third identity, we can repeat the computations in (13) for any dimension n>1n>1, obtaining the decay

ddtEn(f(t),g(t))=−4In(f(t),g(t)),(19)\frac{d}{dt}\mathcal{E}_n(f(t),g(t))=-4\mathcal{I}_n(f(t),g(t)), \tag*{(19)}

where

In(f,g)=∫Rn[fm(x)−gm(x)][f(x)−g(x)] dx≥0.(20)\mathcal{I}_n(f,g)=\int_{\mathbb{R}^n}[f^m(x)-g^m(x)][f(x)-g(x)]\,dx\ge0. \tag*{(20)}

Similar computations have been done in [4] in the (simpler) linear case m=1m=1 for the classical Energy distance corresponding to 0<α<20<\alpha<2.

In the next step, we obtain a lower bound on the functional In\mathcal{I}_n in terms of the energy distance En\mathcal{E}_n.

First of all, since m>1m>1, inequality

xm−am−(x−a)m≥0,x≥a≥0(21)x^m-a^m-(x-a)^m\ge0,\qquad x\ge a\ge0 \tag*{(21)}

implies

In(f,g)≥∫Rn∣f(x)−g(x)∣1+m dx.(22)\mathcal{I}_n(f,g)\ge\int_{\mathbb{R}^n}|f(x)-g(x)|^{1+m}\,dx. \tag*{(22)}

Further, since m>1m>1, thanks to the classical interpolation inequality

∫Rn∣h(x)∣2 dx≤[∫Rn∣h(x)∣ dx](m−1)/m[∫Rn∣h(x)∣1+m dx]1/m(23)\int_{\mathbb{R}^{n}} |h(x)|^{2}\,dx \le\left[\int_{\mathbb{R}^{n}} |h(x)|\,dx\right]^{(m-1)/m}\left[\int_{\mathbb{R}^{n}} |h(x)|^{1+m}\,dx\right]^{1/m} \tag*{(23)}

and to the fact that ff and gg are two probability densities on Rn\mathbb{R}^{n}, so that

∫Rn∣f(x)−g(x)∣ dx≤∫Rn[f(x)+g(x)] dx=2,\int_{\mathbb{R}^{n}} |f(x)-g(x)|\,dx \le\int_{\mathbb{R}^{n}} [f(x)+g(x)]\,dx=2,

we get the inequality

In(f,g)≥12m−1[∫Rn∣f(x)−g(x)∣2 dx]m(24)\mathcal{I}_{n}(f,g) \ge\frac{1}{2^{m-1}}\left[\int_{\mathbb{R}^{n}} |f(x)-g(x)|^{2}\,dx\right]^{m} \tag*{(24)}

which shows that, as m>1m>1, the functional In(f,g)\mathcal{I}_{n}(f,g) controls the mm-th power of the square of the L2L^{2}-distance between ff and gg.

Now, consider that, for n≥3n\ge3, we can bound from above the Energy distance En(f,g)\mathcal{E}_{n}(f,g) between two probability densities by their L2L^{2}-distance resorting to a Nash-type argument. For any given positive constant RR we have

∫Rn∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ=∫∣ξ∣≤R∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ+∫∣ξ∣>R∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ≤∫∣ξ∣≤R22∣ξ∣2 dξ+1R2∫Rn∣f^(ξ)−g^(ξ)∣2 dξ=4∣Sn−1∣∫ρ≤Rρn−1ρ2 dρ+1R2∫Rn∣f(x)−g(x)∣2 dx=CnRn−2+1R2∫Rn∣f(x)−g(x)∣2 dx,(25)\begin{aligned} \int_{\mathbb{R}^{n}} \frac{|\widehat{f}(\xi)-\widehat{g}(\xi)|^{2}}{|\xi|^{2}}\,d\xi &= \int_{|\xi|\le R}\frac{|\widehat{f}(\xi)-\widehat{g}(\xi)|^{2}}{|\xi|^{2}}\,d\xi+\int_{|\xi|>R}\frac{|\widehat{f}(\xi)-\widehat{g}(\xi)|^{2}}{|\xi|^{2}}\,d\xi\\ &\le\int_{|\xi|\le R}\frac{2^{2}}{|\xi|^{2}}\,d\xi+\frac{1}{R^{2}}\int_{\mathbb{R}^{n}}|\widehat{f}(\xi)-\widehat{g}(\xi)|^{2}\,d\xi\\ &=4|S_{n-1}|\int_{\rho\le R}\frac{\rho^{n-1}}{\rho^{2}}\,d\rho+\frac{1}{R^{2}}\int_{\mathbb{R}^{n}}|f(x)-g(x)|^{2}\,dx \\ &=C_{n}R^{n-2}+\frac{1}{R^{2}}\int_{\mathbb{R}^{n}}|f(x)-g(x)|^{2}\,dx, \tag*{(25)} \end{aligned}

where

Cn=8πn/2(n−2)Γ(n2)(26)C_{n}=\frac{8\pi^{n/2}}{(n-2)\Gamma\left(\frac{n}{2}\right)} \tag*{(26)}

Optimizing with respect to R∈R+R\in\mathbb{R}_{+} we obtain the bound

En(f,g)≤dn[∫Rn∣f(x)−g(x)∣2 dx](n−2)/n,(27)\mathcal{E}_{n}(f,g)\le d_{n}\left[\int_{\mathbb{R}^{n}}|f(x)-g(x)|^{2}\,dx\right]^{(n-2)/n}, \tag*{(27)}

where the constant dnd_{n} is explicit

dn=nn−224/nπΓ(n2)2/n.(28)d_{n}=\frac{n}{n-2}\frac{2^{4/n}\pi}{\Gamma\left(\frac{n}{2}\right)^{2/n}}. \tag*{(28)}

Owing to (24) we finally have

En(f,g)≤dn2(m−1)/mIn(f,g)(n−2)/nm.(29)\mathcal{E}_{n}(f,g)\le d_{n}2^{(m-1)/m}\mathcal{I}_{n}(f,g)^{(n-2)/nm}. \tag*{(29)}

Substituting into (19) we obtain that the time derivative of the Energy distance En\mathcal{E}_n satisfies the inequality

ddtEn(f(t),g(t))≤−D(m,n)En(f(t),g(t))mnn−2,(30)\frac{d}{dt}\mathcal{E}_n(f(t),g(t)) \leq-D(m,n)\mathcal{E}_n(f(t),g(t))^{\frac{mn}{n-2}}, \tag*{(30)}

where the explicit constant D(m,n)D(m,n) is given by

D(m,n)=4dn2(m−1)m.D(m,n)=\frac{4}{d_n2^{(m-1)m}}.

The differential inequality (30) can be integrated to give the time-decay

En(f(t),g(t))≤[(1En(f0,g0))γ(m,n)+D(m,n)γ(m,n)t]−1/γ(m,n).(31)\mathcal{E}_n(f(t),g(t))\leq\left[\left(\frac{1}{\mathcal{E}_n(f_0,g_0)}\right)^{\gamma(m,n)}+D(m,n)\gamma(m,n)t\right]^{-1/\gamma(m,n)}. \tag*{(31)}

In (31)

γ(m,n)=(m−1)n+2n−2.(32)\gamma(m,n)=\frac{(m-1)n+2}{n-2}. \tag*{(32)}

Note that, for any given m>1m>1 the value of γ(m,n)\gamma(m,n) is decreasing with n≥3n\geq3, and

m−1≤γ(m,n)≤3m−1.(33)m-1\leq\gamma(m,n)\leq3m-1. \tag*{(33)}

being the lower value m−1>0m-1>0 reached as n→+∞n\to+\infty.

Hence we proved

Theorem 1. Let n≥3n\geq3, and let f(x,t)f(x,t) and g(x,t)g(x,t) be the solutions of the nn-dimensional porous medium equation (1) departing from the initial values f0(x)f_0(x) (respectively g0(x)g_0(x)) of equal mass M=1M=1 such that the Energy distance En(f0,g0)\mathcal{E}_n(f_0,g_0) is bounded. Then, En(f(t),g(t))\mathcal{E}_n(f(t),g(t)) decays to zero as t→+∞t\to+\infty at the explicit rate (1/t)1/γ(m,n)(1/t)^{1/\gamma(m,n)}, where γ(m,n)\gamma(m,n) is defined in (32).

We now apply Theorem 1 to the case in which the initial configuration is the Dirac delta measure centered in zero. Since the solution to the porous medium equation departing from δ0\delta_0 is nothing but the Barenblatt solution b(t)b(t) of mass one defined in (3), we get both the convergence to zero and the rate of decay to the source-type solution in the Energy distance En(f(t),b(t))\mathcal{E}_n(f(t),b(t)), provided the Energy distance is bounded at the initial time t=0t=0. Indeed, when this condition is met, f0(x)f_0(x) belongs to the domain of attraction of the source-type solution b(x,t,C)b(x,t,C) with respect to the Energy distance En\mathcal{E}_n.

Applying definition (17) to the pair (f0,δ0)(f_0,\delta_0) yields

0≤En(f0,δ0)=2∫Rn∣x∣2−n(2−n)∣Sn−1∣f0(x) dx−∫Rn×Rn∣x−y∣2−n(2−n)∣Sn−1∣f0(x)f0(y) dx dy=∫Rn×Rn∣x−y∣2−n(n−2)∣Sn−1∣f0(x)f0(y) dx dy−2∫Rn∣x∣2−n(n−2)∣Sn−1∣f0(x) dx.(34)\begin{aligned} 0\leq\mathcal{E}_n(f_0,\delta_0) &=2\int_{\mathbb{R}^n}\frac{|x|^{2-n}}{(2-n)|S_{n-1}|}f_0(x)\,dx\\ &\quad-\int_{\mathbb{R}^n\times\mathbb{R}^n}\frac{|x-y|^{2-n}}{(2-n)|S_{n-1}|}f_0(x)f_0(y)\,dx\,dy\\ &=\int_{\mathbb{R}^n\times\mathbb{R}^n}\frac{|x-y|^{2-n}}{(n-2)|S_{n-1}|}f_0(x)f_0(y)\,dx\,dy\\ &\quad-2\int_{\mathbb{R}^n}\frac{|x|^{2-n}}{(n-2)|S_{n-1}|}f_0(x)\,dx. \tag*{(34)} \end{aligned}

Since the Energy distance is non-negative, and n≥3n \ge3, En(f0,δ0)\mathcal{E}_n(f_0,\delta_0) is bounded as soon as

∫Rn×Rn∣x−y∣2−nf0(x)f0(y) dxdy\int_{\mathbb{R}^n \times\mathbb{R}^n} \lvert x-y\rvert^{2-n}f_0(x)f_0(y)\,dxdy

is bounded.

Owing to the Hardy–Littlewood–Sobolev inequality [22], stating that

∣∫Rn×Rn∣x−y∣−λh(x)k(y) dxdy∣≤Np,λ,n∥h∥p∥k∥r\left\lvert\int_{\mathbb{R}^n \times\mathbb{R}^n}\lvert x-y\rvert^{-\lambda}h(x)k(y)\,dxdy\right\rvert\le N_{p,\lambda,n}\lVert h\rVert_p\lVert k\rVert_r

for all h∈Lp(Rn)h \in L^p(\mathbb{R}^n), k∈Lr(Rn)k \in L^r(\mathbb{R}^n) with 1<p,r<∞1 < p,r < \infty, 1/p+1/r+λ/n=21/p+1/r+\lambda/n=2 and 0<λ<n0 < \lambda< n, an explicit sufficient condition ensuring the boundedness of the Energy distance in (34) is to assume that f0∈Lp(Rn)f_0 \in L^p(\mathbb{R}^n) with p=2n/(n+2)p=2n/(n+2).

Let the initial density f0(x)∈L1(Rn)f_0(x) \in L^1(\mathbb{R}^n), and let f0^(ξ)\widehat{f_0}(\xi) its Fourier transform. Also, let ω(x)\omega(x) denote the Gaussian density

ω(x)=12πexp⁡{−∣x∣22}.(35)\omega(x)=\frac{1}{\sqrt{2\pi}}\exp\left\{-\frac{\lvert x\rvert^2}{2}\right\}. \tag*{(35)}

Owing to expression (18), and using the bound ∣f0^(ξ)∣≤1\lvert\widehat{f_0}(\xi)\rvert\le1, we obtain

En(f0,δ0)=∫Rn∣f0^(ξ)−1∣2∣ξ∣2 dξ=∫Rn∣f0^(ξ)−f0^(ξ)ω^(ξ)+f0^(ξ)ω^(ξ)−1∣2∣ξ∣2 dξ≤2∫Rn∣ω^(ξ)−1∣2∣f0^(ξ)∣2∣ξ∣2 dξ+2∫Rn∣f0^(ξ)ω^(ξ)−1∣2∣ξ∣2 dξ≤2∫Rn∣ω^(ξ)−1∣2∣ξ∣2 dξ+2∫Rn∣f0^(ξ)ω^(ξ)−1∣2∣ξ∣2 dξ≤C<+∞.(36)\begin{aligned} \mathcal{E}_n(f_0,\delta_0) &=\int_{\mathbb{R}^n}\frac{\left\lvert\widehat{f_0}(\xi)-1\right\rvert^2}{\lvert\xi\rvert^2}\,d\xi\\ &=\int_{\mathbb{R}^n}\frac{\left\lvert\widehat{f_0}(\xi)-\widehat{f_0}(\xi)\widehat{\omega}(\xi)+\widehat{f_0}(\xi)\widehat{\omega}(\xi)-1\right\rvert^2}{\lvert\xi\rvert^2}\,d\xi\\ &\le2\int_{\mathbb{R}^n}\frac{\left\lvert\widehat{\omega}(\xi)-1\right\rvert^2\lvert\widehat{f_0}(\xi)\rvert^2}{\lvert\xi\rvert^2}\,d\xi+2\int_{\mathbb{R}^n}\frac{\left\lvert\widehat{f_0}(\xi)\widehat{\omega}(\xi)-1\right\rvert^2}{\lvert\xi\rvert^2}\,d\xi\\ &\le2\int_{\mathbb{R}^n}\frac{\left\lvert\widehat{\omega}(\xi)-1\right\rvert^2}{\lvert\xi\rvert^2}\,d\xi+2\int_{\mathbb{R}^n}\frac{\left\lvert\widehat{f_0}(\xi)\widehat{\omega}(\xi)-1\right\rvert^2}{\lvert\xi\rvert^2}\,d\xi\le C<+\infty. \tag*{(36)} \end{aligned}

Indeed, since both ω\omega and f0⋆ωf_0\star\omega, the convolution of f0f_0 and ω\omega, belong to L1(Rn)∩L∞(Rn)L^1(\mathbb{R}^n)\cap L^\infty(\mathbb{R}^n), they also belong to Lp(Rn)L^p(\mathbb{R}^n) when p=2n/(n+2)p=2n/(n+2). In particular, their Energy distances from the Dirac delta located in zero, according to the Hardy–Littlewood–Sobolev inequality, are bounded.

Therefore we have

Theorem 2. Let n≥3n \ge3, and let f(x,t)f(x,t) be the solution of the nn-dimensional porous medium equation (1) departing from a probability density f0(x)f_0(x). Then, the Energy distance En(f(t),b(t,C))\mathcal{E}_n(f(t),b(t,C)), with b(t,C)b(t,C) defined as in (3), decays to zero as t→+∞t \to+\infty at the explicit rate (1/t)1/γ(m,n)(1/t)^{1/\gamma(m,n)}, where γ(m,n)\gamma(m,n) is defined in (32).

Theorem 2 shows that, if n≥3n \ge3, the domain of attraction of the source–type solution defined as in (3) with respect to the Energy distance En\mathcal{E}_n contains all initial values in L1(Rn)L^1(\mathbb{R}^n). Hence, the domain of attraction of the Barenblatt solution is larger than the one found in [8] by resorting to entropy methods, since there convergence in the stronger L1L^1-distance requires the initial density to have the second moment bounded. Moreover, we stress that in [8] the convergence of the solution to the porous medium equation towards the source-type solution in L1(Rn)L^1(\mathbb{R}^n) was shown with rate (1/t)1/μ(m,n)(1/t)^{1/\mu(m,n)}, where

μ(m,n)=(n+2)m−n2.\mu(m,n)=\frac{(n+2)m-n}{2}.

so that, for n=1,2n=1,2, we have γ(n,m)<μ(n,m)\gamma(n,m)<\mu(n,m). However, if n=3n=3, the two rates coincide, while if n>3n>3, the rate of convergence in Energy distance is better than the rate of convergence in L1L^1, and does not collapse to zero if n→∞n\to\infty.

The fast diffusion case

Let us now consider the fast diffusion case. To avoid confusion with the previous Sections, let us denote with p<1p<1 the characteristic exponent of the nonlinear diffusion (1) in the fast diffusion regime. Moreover, since we want to study the large time behavior of the solution to equation (1) in which the Barenblatt profile has bounded mass, we assume that p>(n−2)/np>(n-2)/n. It is immediate to realize that all the computations of Section 3 leading to equation (19), with In(f,g)\mathcal{I}_n(f,g) defined in (20) continue to hold. Hence, as in Section 4 we still obtain that the Energy distance En\mathcal{E}_n satisfies

ddtEn(f(t),g(t))=−4Jn(f(t),g(t)),(37)\frac{d}{dt}\mathcal{E}_n(f(t),g(t))=-4\mathcal{J}_n(f(t),g(t)), \tag*{(37)}

where now, for n−2n<p<1\frac{n-2}{n}<p<1,

Jn(f,g)=∫Rn[fp(x)−gp(x)][f(x)−g(x)] dx≥0.(38)\mathcal{J}_n(f,g)=\int_{\mathbb{R}^n}[f^p(\mathbf{x})-g^p(\mathbf{x})][f(\mathbf{x})-g(\mathbf{x})]\,d\mathbf{x}\geq0. \tag*{(38)}

In this case, however, we can not proceed as in Section 4, since inequality (21) is true if and only if m>1m>1. On the other hand, for any given m>1m>1, inequality (21) can be rewritten in the form

x−a−(x1/m−a1/m)m≥0,x≥a≥0.x-a-\left(x^{1/m}-a^{1/m}\right)^m\geq0,\qquad x\geq a\geq0.

namely, since 1/m=p<11/m=p<1

x−a−(xp−ap)1/p≥0,x≥a≥0(39)x-a-\left(x^p-a^p\right)^{1/p}\geq0,\qquad x\geq a\geq0 \tag*{(39)}

that implies

Jn(f,g)≥∫Rn∣fp(x)−gp(x)∣1+1/p dx.(40)\mathcal{J}_n(f,g)\geq\int_{\mathbb{R}^n}\left|f^p(\mathbf{x})-g^p(\mathbf{x})\right|^{1+1/p}\,d\mathbf{x}. \tag*{(40)}

We now provide a lower bound on the right-hand side of (40).

Let ff and gg two probability density functions and p∈[1/2,1)p \in[1/2,1). Thanks to the pointwise inequality

∣f(x)−g(x)∣≤∣fp(x)−gp(x)∣∣f1−p(x)+g1−p(x)∣,|f(\mathbf{x})-g(\mathbf{x})| \leq|f^p(\mathbf{x})-g^p(\mathbf{x})||f^{1-p}(\mathbf{x})+g^{1-p}(\mathbf{x})|,

for any given positive constant β\beta, the Holder inequality allows us to conclude that

∫Rn∣f(x)−g(x)∣β dx≤(∫Rn∣fp(x)−gp(x)∣βr dx)1r(∫Rn∣f1−p(x)+g1−p(x)∣βs dx)1s,(41)\int_{\mathbb{R}^n}|f(\mathbf{x})-g(\mathbf{x})|^\beta\,d\mathbf{x} \leq\left(\int_{\mathbb{R}^n}|f^p(\mathbf{x})-g^p(\mathbf{x})|^{\beta r}\,d\mathbf{x}\right)^{\frac{1}{r}}\left(\int_{\mathbb{R}^n}|f^{1-p}(\mathbf{x})+g^{1-p}(\mathbf{x})|^{\beta s}\,d\mathbf{x}\right)^{\frac{1}{s}}, \tag*{(41)}

with rr and ss such that 1/r+1/s=11/r+1/s=1. We then look for β\beta, rr, and ss such that

1r+1s=1;βr=1+pp;βs(1−p)=1.(42)\frac{1}{r}+\frac{1}{s}=1;\qquad\beta r=\frac{1+p}{p};\qquad\beta s(1-p)=1. \tag*{(42)}

Solving the system, we find the values

β=1+p1+p−p2;r=1+p−p2p;s=1+p−p21−p2.(43)\beta=\frac{1+p}{1+p-p^2};\qquad r=\frac{1+p-p^2}{p};\qquad s=\frac{1+p-p^2}{1-p^2}. \tag*{(43)}

Note that r,s>1r,s>1, while 1<β<21<\beta<2.

Since βs(1−p)=1\beta s(1-p)=1, and f,gf,g are probability densities, owing to the classical inequality

(aq+bq)1/q≤21+1/q(a+b),(a^q+b^q)^{1/q}\leq2^{1+1/q}(a+b),

valid for any given 0<q<10<q<1 and a,b>0a,b>0, we obtain

∫Rn∣f1−p(x)+g1−p(x)∣βs dx≤21+1/(1−p)∫Rn(f(x)+g(x)) dx=2(3−p)/(1−p).\int_{\mathbb{R}^n}|f^{1-p}(\mathbf{x})+g^{1-p}(\mathbf{x})|^{\beta s}\,d\mathbf{x}\leq2^{1+1/(1-p)}\int_{\mathbb{R}^n}(f(\mathbf{x})+g(\mathbf{x}))\,d\mathbf{x}=2^{(3-p)/(1-p)}.

Substituting into (41), and taking into account (40) we finally get, for any p∈(1/2,1)p\in(1/2,1)

Jn(f,g)≥cp(∫Rn∣f(x)−g(x)∣β dx)r,(44)\mathcal{J}_n(f,g)\geq c_p\left(\int_{\mathbb{R}^n}|f(\mathbf{x})-g(\mathbf{x})|^\beta\,d\mathbf{x}\right)^r, \tag*{(44)}

where β=β(p)\beta=\beta(p) and r=r(p)r=r(p) are given in (42), and

cp=2−(3+2p−p2)/p.(45)c_p=2^{-(3+2p-p^2)/p}. \tag*{(45)}

Hence J(f,g)\mathcal{J}(f,g) controls the LβL^\beta-distance of the probability densities ff and gg.

Since β\beta in (43) satisfies 1<β<21<\beta<2, the Beckner form of the Hausdorff–Young inequality [6]

(∫Rn∣f^(ξ)−g^(ξ)∣q∗ dξ)1/q∗≤(q1/qq∗1/q∗)n/2(∫Rn∣f(x)−g(x)∣q dx)1/q,\left(\int_{\mathbb{R}^n}|\widehat{f}(\boldsymbol{\xi})-\widehat{g}(\boldsymbol{\xi})|^{q^*}\,d\boldsymbol{\xi}\right)^{1/q^*}\leq\left(\frac{q^{1/q}}{q^{*1/q^*}}\right)^{n/2}\left(\int_{\mathbb{R}^n}|f(\mathbf{x})-g(\mathbf{x})|^q\,d\mathbf{x}\right)^{1/q},

which holds true for 1<q≤21<q\leq2 and 1/q+1/q∗=11/q+1/q^*=1, allows us to bound the Lβ∗L^{\beta^*}-distance of the Fourier transforms of the probability densities ff and gg.

Since βr=(1+p)/p\beta r = (1+p)/p, the Hausdorff–Young inequality implies

(∫Rn∣f(x)−g(x)∣β dx)r≥(β∗1/β∗β1/β)n(1+p)/(2p)(∫Rn∣f^(ξ)−g^(ξ)∣(1+p)/p2 dξ)p.(46)\left(\int_{\mathbb{R}^n} |f(x)-g(x)|^\beta\,\mathrm{d}x\right)^r \ge\left(\frac{\beta^{*1/\beta^*}}{\beta^{1/\beta}}\right)^{n(1+p)/(2p)} \left(\int_{\mathbb{R}^n} |\hat{f}(\xi)-\hat{g}(\xi)|^{(1+p)/p^2}\,\mathrm{d}\xi\right)^p . \tag*{(46)}

Therefore, inequality (44) allows us to conclude with the following

Jn(f,g)≥cp,n(∫Rn∣f^(ξ)−g^(ξ)∣(1+p)/p2 dξ)p,(47)J_n(f,g) \ge c_{p,n}\left(\int_{\mathbb{R}^n} |\hat{f}(\xi)-\hat{g}(\xi)|^{(1+p)/p^2}\,\mathrm{d}\xi\right)^p , \tag*{(47)}

where, clearly, the exponent (1+p)/p2>2(1+p)/p^2 > 2, while

cp,n=cp(β∗1/β∗β1/β)n(1+p)/(2p).(48)c_{p,n}=c_p\left(\frac{\beta^{*1/\beta^*}}{\beta^{1/\beta}}\right)^{n(1+p)/(2p)} . \tag*{(48)}

Thanks to (47), we can proceed as in § 4 to obtain

∫Rn∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ≤∫∣ξ∣≤R∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ+∫∣ξ∣≥R∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ(49)\int_{\mathbb{R}^n}\frac{|\hat{f}(\xi)-\hat{g}(\xi)|^2}{|\xi|^2}\,\mathrm{d}\xi\le\int_{|\xi|\le R}\frac{|\hat{f}(\xi)-\hat{g}(\xi)|^2}{|\xi|^2}\,\mathrm{d}\xi+\int_{|\xi|\ge R}\frac{|\hat{f}(\xi)-\hat{g}(\xi)|^2}{|\xi|^2}\,\mathrm{d}\xi \tag*{(49)}
≤CnRn−2+∫∣ξ∣≥R∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ,(50)\le C_nR^{n-2}+\int_{|\xi|\ge R}\frac{|\hat{f}(\xi)-\hat{g}(\xi)|^2}{|\xi|^2}\,\mathrm{d}\xi, \tag*{(50)}

where R>0R>0 is a given value and CnC_n is the constant in (26).

By the Holder inequality, we can estimate the right hand side of the latter inequality as it follows

∫∣ξ∣≥R∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ≤(∫∣ξ∣≥R1∣ξ∣2u dξ)1/u(∫Rn∣f^(ξ)−g^(ξ)∣2v dξ)1/v,(51)\int_{|\xi|\ge R}\frac{|\hat{f}(\xi)-\hat{g}(\xi)|^2}{|\xi|^2}\,\mathrm{d}\xi\le\left(\int_{|\xi|\ge R}\frac{1}{|\xi|^{2u}}\,\mathrm{d}\xi\right)^{1/u}\left(\int_{\mathbb{R}^n}|\hat{f}(\xi)-\hat{g}(\xi)|^{2v}\,\mathrm{d}\xi\right)^{1/v}, \tag*{(51)}

where uu and vv are such that 1/u+1/v=11/u+1/v=1. If v=1+p2p2>1v=\frac{1+p}{2p^2}>1, then u=1+p1+p−2p2u=\frac{1+p}{1+p-2p^2}, so that, making use of (47) we obtain

∫∣ξ∣≥R∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ≤(∫∣ξ∣≥R1∣ξ∣2(1+p)1+p−2p2 dξ)1+p−2p21+p(1cp,nJn(f,g))2p21+p.(52)\int_{|\xi|\ge R}\frac{|\hat{f}(\xi)-\hat{g}(\xi)|^2}{|\xi|^2}\,\mathrm{d}\xi\le\left(\int_{|\xi|\ge R}\frac{1}{|\xi|^{\frac{2(1+p)}{1+p-2p^2}}}\,\mathrm{d}\xi\right)^{\frac{1+p-2p^2}{1+p}}\left(\frac{1}{c_{p,n}}J_n(f,g)\right)^{\frac{2p^2}{1+p}} . \tag*{(52)}

Notice that ∣ξ∣−2(1+p)1+p−2p2|\xi|^{-\frac{2(1+p)}{1+p-2p^2}} is integrable on the set ∣ξ∣≥R|\xi|\ge R if and only if

2(1+p)1+p−2p2−n+1>1,\frac{2(1+p)}{1+p-2p^2}-n+1>1,

or, equivalently,

n(2p2−p−1)+2(1+p)>0.(53)n(2p^2-p-1)+2(1+p)>0. \tag*{(53)}

It is immediate to verify that, to solve (53) on R+\mathbb{R}_{+}, pp must be greater than

p(n)=n−24n+14n9n2−20n+4.p(n)=\frac{n-2}{4n}+\frac{1}{4n}\sqrt{9n^2-20n+4}.

Notice that if n=3n=3, then p(3)=1/2p(3)=1/2. Moreover, if n→∞n\to\infty, then p(n)→1p(n)\to1. For such values of pp, we have

(∫∣ξ∣≥R1∣ξ∣2(1+p)1+p−2p2 dξ)1+p−2p21+p≤Dp,nRn(1+p−2p2)1+p−2=Dp,nR−ν(p,n),(54)\left(\int_{|\xi|\geq R}\frac{1}{|\xi|^{\frac{2(1+p)}{1+p-2p^2}}}\,\mathrm{d}\xi\right)^{\frac{1+p-2p^2}{1+p}}\leq D_{p,n}R^{\frac{n(1+p-2p^2)}{1+p}-2}=D_{p,n}R^{-\nu(p,n)}, \tag*{(54)}

where we denoted

ν(p,n):=2−n(1+p−2p2)1+p>0,(55)\nu(p,n):=2-\frac{n(1+p-2p^2)}{1+p}>0, \tag*{(55)}

and

Dp,n=(∣Sn−1∣2(1+p)1+p−2p2−n)1+p−2p21+p.(56)D_{p,n}=\left(\frac{\lvert S_{n-1}\rvert}{\frac{2(1+p)}{1+p-2p^2}-n}\right)^{\frac{1+p-2p^2}{1+p}}. \tag*{(56)}

All in all, we get

∫Rn∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ≤CnRn−2+Dp,ncp,n2p1+pR−ν(p,n)Jn(f,g)2p1+p.(57)\int_{\mathbb{R}^n}\frac{\lvert\widehat{f}(\xi)-\widehat{g}(\xi)\rvert^2}{\lvert\xi\rvert^2}\,\mathrm{d}\xi\leq C_nR^{n-2}+D_{p,n}c_{p,n}^{\frac{2p}{1+p}}R^{-\nu(p,n)}\mathcal{J}_n(f,g)^{\frac{2p}{1+p}}. \tag*{(57)}

By optimizing the right hand side of (57) with respect to RR we get

∫Rn∣f^(ξ)−g^(ξ)∣2∣ξ∣2 dξ≤K(p,n)Jn(f,g)n−2np,(58)\int_{\mathbb{R}^n}\frac{\lvert\widehat{f}(\xi)-\widehat{g}(\xi)\rvert^2}{\lvert\xi\rvert^2}\,\mathrm{d}\xi\leq K(p,n)\mathcal{J}_n(f,g)^{\frac{n-2}{np}}, \tag*{(58)}

where the constant K(p,n)K(p,n), that can be explicitly computed in terms of CnC_n, cp,nc_{p,n}, ν(p,n)\nu(p,n) and Dp,nD_{p,n}, as given by (26), (48), (55) and (56), takes the value

K(p,n)=2np21+p((n−2)Cnν(p,n))ν(p,n)(1+p)2np2Dp,n(n−2)(1+p)2np2cp,n−n−2np.(59)K(p,n)=\frac{2np^2}{1+p}\left(\frac{(n-2)C_n}{\nu(p,n)}\right)^{\frac{\nu(p,n)(1+p)}{2np^2}}D_{p,n}^{\frac{(n-2)(1+p)}{2np^2}}c_{p,n}^{-\frac{n-2}{np}}. \tag*{(59)}

What is important to remark is that, apart from the different value of the constant K(p,n)K(p,n), the previous computations show that in the fast diffusion case the time decay of the energy distance En(f,g)\mathcal{E}_n(f,g) follows the law (30) relative to the porous medium equation.

If we now go back to the notation p=mp=m, where now m<1m<1, we conclude that the decay is given by (31), where now γ(m,n)\gamma(m,n) is well defined only when the fast diffusion coefficient m<1m<1 satisfies the bound

m>max⁡{12,n−2n}.(60)m>\max\left\{\frac{1}{2},\frac{n-2}{n}\right\}. \tag*{(60)}

Note that in this case, the measure of the interval ((n−2)/n,1)((n-2)/n,1) in which the coefficient γ(m,n)\gamma(m,n) is well defined converges to zero as n→∞n\to\infty.

All in all, we proved the following.

Theorem 3. Let n≥3n \ge3, and let f(x,t)f(\mathbf{x},t) and g(x,t)g(\mathbf{x},t) be the solutions of the nn-dimensional fast diffusion equation (1) departing from the initial values f0(x)f_0(\mathbf{x}) (respectively g0(x)g_0(\mathbf{x})) of equal mass M=1M=1 such that the Energy distance En(f0,g0)\mathcal{E}_n(f_0,g_0) is bounded. Then, if mm satisfies condition (60), En(f(t),g(t))\mathcal{E}_n(f(t),g(t)) decays to zero as t→+∞t\to+\infty at the explicit rate (1/t)1/γ(m,n)(1/t)^{1/\gamma(m,n)}, where γ(m,n)\gamma(m,n) is defined in (32).

Likewise, since the reasoning leading to Theorem 2, and in particular the estimate (36), do not depend on the exponent of the diffusion equation (1), but only on the properties of the initial value, we have the following.

Theorem 4. Let n≥3n \ge3, and let f(x,t)f(\mathbf{x},t) be the solution of the nn-dimensional fast diffusion equation (1) departing from a probability density f0(x)f_0(\mathbf{x}). Then, if mm satisfies condition (60), the Energy distance En(f(t),b(t,C))\mathcal{E}_n(f(t),b(t,C)) decays to zero as t→+∞t\to+\infty at the explicit rate (1/t)1/γ(m,n)(1/t)^{1/\gamma(m,n)}, where γ(m,n)\gamma(m,n) is defined in (32).

Conclusions

In this paper, we studied the rate of convergence of the solution to the porous medium equation to the Barenblatt source–type solution with the same mass by means of a particular class of Energy distances, which are equivalent to the square of the distance in the Sobolev space H˙−1(Rn)\dot{H}^{-1}(\mathbb{R}^n). It is shown that, as n≥3n \ge3, the rate of decay is improved with respect to known results which are based on the decay of relative entropy, and that the domain of attraction of the Barenblatt solution includes all initial value of finite mass.

With few modifications, it has been further shown that analogous results hold in the fast diffusion case, with some limitations in the diffusion exponent that are mainly due to guarantee that the source–type solution has a bounded mass.

The present results suggest that both Cramér and Energy distances could be of high interest in the study of both theoretical and numerical problems related to nonlinear diffusion equations.

Acknowledgement

This work has been written within the activities of GNFM (Gruppo Nazionale per la Fisica Matematica) of INdAM (Istituto Nazionale di Alta Matematica), Italy. G.T. acknowledges support of the Institute for Applied Mathematics and Information Technologies (IMATI), Pavia, Italy. G.T. wishes to thank the Editors for inviting him to contribute to this volume dedicated to Giuseppe Savaré, a friend and colleague at the Department of Mathematics at the University of Pavia for many years.

References

  1. [1]Auricchio, G., Brigati, G., Giudici, P., and Toscani, G., From kinetic theory to AI: a rediscovery of high-dimensional divergences and their properties. Mathematical Models and Methods in Applied Sciences, 36 (2026) 1–49.arxiv.org/abs/2507.11387
  2. [2]Auricchio, G., Codegoni, A., Gualandi, S., Toscani, G., and Veneroni, M., On the equivalence between Fourier-based and Wasserstein metrics, Rendiconti Lincei. Matematica e Applicazioni, 31 (2020) 627–649.arxiv.org/abs/2005.06530
  3. [3]Auricchio, G., Codegoni, A., Gualandi, S., and Zambon, L., The Fourier discrepancy function, Communications in Mathematical Sciences, 21 (2023) 627–639.
  4. [4]Auricchio, G., and Toscani, G., Energy distance and evolution problems: a promising tool for kinetic equations. Ricerche di Matematica 75 (2026) 849–863.arxiv.org/abs/2510.09123
  5. [5]Barenblatt, G.I., On some unsteady motions of a liquid and gas in a porous medium. Akad. Nauk SSSR Prikl. Mat. Meh. 16 (1952) 67–78.
  6. [6]Beckner, W., Inequalities in Fourier analysis. Annals of Mathematics, 102 (1) (1975) 159–182DOI
  7. [7]Bellemare, M. G., Danihelka, I., Dabney, W., Mohamed, S., Lakshminarayanan, B., and Hoyer, S., and Munos, R., The Cramér distance as a solution to biased Wasserstein gradients, arXiv preprint arXiv:1705.10743 (2017).arxiv.org/abs/1705.10743
  8. [8]Carrillo, J.A. and Toscani, G. Asymptotic L¹-decay of solutions of the porous medium equation to self-similarity. Indiana University Mathematics Journal, (2000) Apr 1:113–42.DOI
  9. [9]Carrillo, J.A. and Toscani, G., Contractive probability metrics and asymptotic behavior of dissipative kinetic equations, Rivista di Matematica dell’Università di Parma, 6 (2007) 75–198.
  10. [10]Carrillo, J.A. and Toscani, G., Renyi entropy and improved equilibration rates to self-similarity for nonlinear diffusion equations, Nonlinearity, 27 (2014) 3159–3177.DOI
  11. [11]Cramér, H., On the composition of elementary errors, Skandinavisk Aktuarietidskrift, 11, (1928) 141–180.
  12. [12]Cuturi, M. and Doucet, A., Fast computation of Wasserstein barycenters, International conference on machine learning, (2014) 685–693.arxiv.org/abs/1310.4375
  13. [13]Di Francesco, M., Fornasier, M., Hütter, J.C. and Matthes, D., Asymptotic behavior of gradient flows driven by nonlocal power repulsion and attraction potentials in one dimension. SIAM Journal on Mathematical Analysis, 46 (6), (2014) 3814–3837.
  14. [14]Dolbeault, J., and Toscani, G., Best matching Barenblatt profiles are delayed, J. Phys. A: Math. Theor., 48 (2015) 065206.DOI
  15. [15]Dolbeault, J., and Toscani, G., Nonlinear diffusions: extremal properties of Barenblatt profiles, best matching and delays. Nonlinear Anal. Series A, 138 (2016) 31–43.arxiv.org/abs/1501.03646
  16. [16]Fornasier, M., Haskovec, J. and Steidl, G., Consistency of variational continuous-domain quantization via kinetic theory. Applicable Analysis, 92 (6), (2013) 1283–1298.arxiv.org/abs/1112.1403
  17. [17]Fornasier, M. and Hütter, J.C., Consistency of probability measure quantization by means of power repulsion–attraction potentials. Journal of Fourier Analysis and Applications, 22 (3) (2016) 694–749.arxiv.org/abs/1310.1120
  18. [18]Gabetta, E., Toscani, G., and Wennberg, B., Metrics for probability measures and the trend to equilibrium for solutions of the Boltzmann equation, Journal of Statistical Physics, 81 (1995) 901–934.
  19. [19]Gini, C., Sulla misura della concentrazione e della variabilità dei caratteri, Atti del Reale Istituto Veneto di Scienze, Lettere ed Arti, 73 (1914) 1203–1248. English translation in Metron, 3–38, (2005).
  20. [20]Gini, C., Measurement of inequality of incomes, The Economic Journal, 31 (1921) 124–126.DOI
  21. [21]Kamenomostskaya (Kamin), S., The asymptotic behaviour of the solution of the filtration equation, Israel J. Math. 14 (1973) 76–87.DOI
  22. [22]Lieb, E.H., Sharp constants in the Hardy-Littlewood-Sobolev and related inequalities, Ann. Math. 118 (1983) 349–374.
  23. [23]Lions, P.L. and Toscani, G., Diffusive limit for finite velocity Boltzmann kinetic models. Revista Matematica Iberoamericana 13 (3) (1997) 473–513.DOI
  24. [24]McKean, H.P., The central limit theorem for Carleman’s equation. Israel Journal of Mathematics 21 (1) (1975) 54–92.DOI
  25. [25]Rosenau, P., Fast and superfast diffusion processes. Physical review letters, 74 (7)(1995) 1056–1059pubmed.ncbi.nlm.nih.gov/10058923
  26. [26]Savaré G., and Toscani, G., The concavity of Rényi entropy power. IEEE Transactions on Information theory, 60 (5) (2014) 2687–93.DOI
  27. [27]Scagliotti, A. and Farinelli, S., Normalizing flows as approximations of optimal transport maps via linear-control neural ODEs, Nonlinear Analysis, 257 (2025), 113811.arxiv.org/abs/2311.01404
  28. [28]Smoller, J., Shock Waves and Reaction–Diffusion Equations (2 ed.). Springer New York 1974.
  29. [29]Stein, E.M, and Weiss, G., Introduction to Fourier analysis on Euclidean spaces, Princeton University Press, Princeton, NJ (1970)
  30. [30]Székely, G.J., Potential and kinetic energy in statistics. Lecture Notes. Budapest Institute of Technology (Technical University) (1989).
  31. [31]Székely, G.J., E-statistics: Energy of statistical samples. Bowling Green State University, Department of Mathematics and Statistics Technical Report No. 03-05, 2003 (also technical reports by same title, from 2000–2003 and NSA Grant # MDA 904-02-1-0091 (2000–2002).)DOI
  32. [32]Székely, G.J. and Rizzo, M.I., Energy statistics: A class of statistics based on distances. Journal of Statistical Planning and Inference, 143 (2013) 1249–1272DOI
  33. [33]Toscani, G., A central limit theorem for solutions of the porous medium equation, J. Evol. Equ. 5 (2005) 185–203DOI
  34. [34]Vázquez, J.L, Asymptotic behaviour for the porous medium equation posed in the whole space. Nonlinear Evolution Equations and Related Topics: Dedicated to Philippe Bénilan. Basel: Birkhäuser Basel, 2003. 67–118.DOI
  35. [35]Vázquez, J.L., The porous medium equation: mathematical theory. Clarendon Press, 2006.
  36. [36]Vázquez, J.L., The mathematical theories of diffusion: nonlinear and fractional diffusion. In: Nonlocal and Nonlinear Diffusions and Interactions: New Methods and Directions: Cetraro, Italy 2016. Cham: Springer International Publishing, 2017. 205–278.arxiv.org/abs/1706.08241
  37. [37]Villani, C., Optimal transport: old and new, Vol. 338, (Springer-Verlag, Basel, 2009).
  38. [38]Xu, K., How has the literature on Gini’s index evolved in the past 80 years? Dalhousie University, Economics Working Paper (2003).DOI

Paper details

vlnc:W_v55ua9x6
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents