On the exponential convergence to equilibrium for ultrafast diffusion equations

Yi C. Huang, Xinhang Tong · 2025

Read On the exponential convergence to equilibrium for ultrafast diffusion equations by Yi C. Huang, Xinhang Tong (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-07382.txt.

Introduction

Some diffusion equations spread material so fast that standard math tools strain to describe them. Yi C. Huang and Xinhang Tong study ultrafast diffusion in a 2025 research paper. They prove that solutions settle toward equilibrium, the balanced end state, at an exponential rate. Their proof is deliberately simple.

The argument uses the Poincaré inequality directly, a classic tool linking a function's spread to its gradient. It avoids the optimal transport machinery used in an earlier one-dimensional result. That earlier work also had to exclude Gaussian weights. Huang and Tong remove both limits, extending exponential convergence to Gaussian measures in higher dimensions. A cleaner proof that covers more cases helps analysts apply the result to a wider family of equations.

Catalog record

CREATORYi C. Huang, Xinhang Tong
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE10 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-07382

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

On the exponential convergence to equilibrium for ultrafast diffusion equations Thanks: Research of YCH is partially supported by the National NSF grant of China (no. 11801274), the JSPS Invitational Fellowship for Research in Japan (no. S24040), and the Open Projects from Yunnan Normal University (no. YNNUMA2403) and Soochow University (no. SDGC2418). Both authors thank Professor Max Fathi for sharing generously his insights on this problem. YCH thanks in particular Jian-Yang Zhang (Shanghai) and Yuzhe Zhu (Chicago) for helpful communications, and Qi Guo (Beijing) for hospitality at Renmin University where this manuscript is finalised.
Yi C. Huang, Xinhang Tong

Abstract
. We propose a simple proof of the exponential convergence to equilibrium for ultrafast diffusion equations in ℝ n \mathbb{R}^{n} . Our approach, based on the direct use of Poincaré inequality, gets rid of the optimal transport arguments used in [ FI25 ] which are valid for Gaussian-excluded one-dimensional weights. This simplification allows us to extend their results to Gaussian measures in higher dimensions.



1.
Introduction

In
[
Iac19
]
, M. Iacobelli obtained the exponential convergence for the following so-called “ultrafast diffusion equation”
(see for example
[
BDMN17a
]
,
[
BDMN17b
]
and
[
IPS19
]
for the relevant background materials) with periodic boundary conditions:

\begin{cases}\partial_{t}f(t,x)=-r\,\partial_{x}\left(f(t,x)\,\partial_{x}\left(\frac{\rho(x)}{f^{r+1}(t,x)}\right)\right)&\quad\text{on }\quad(0,\infty)\times[0,1],\\[6.0pt]
f(t,0)=f(t,1)&\quad\text{on }\quad(0,\infty),\end{cases}

where
r>1
and both
\rho>0
and
f(t,\cdot)
are probability densities on
[0,1]
.
Recently in
[
FI25
]
, M. Fathi and M. Iacobelli extended the results of
[
Iac19
]
to the non-compact case (namely, the whole real line
\mathbb{R}
).
They considered the solutions
f(t,\cdot)
of

\partial_{t}f(t,x)=-r\partial_{x}\left(f(t,x)\partial_{x}\left(\frac{\rho(x)}{f^{r+1}(t,x)}\right)\right)\quad\text{on }\quad(0,\infty)\times\mathbb{R},

where
r>1
and
\rho>0
is a probability density on
\mathbb{R}
.
However, the approach in
[
FI25
]
relies crucially on the optimal transport arguments which are valid for restrictive (Gaussian-excluded) one-dimensional weights.
In this paper, with a rather simple argument, we extend the results of
[
FI25
]
to Gaussian measures in higher dimensions.

For the purpose as mentioned above, we shall consider the following (higher-dimensional) ultrafast diffusion equation with initial data
f_{0}(x):=f(0,x)
,

(1)

\partial_{t}f(t,x)=-r\,\mathrm{div}_{x}\left(f(t,x)\nabla_{x}\left(\frac{\rho(x)}{f^{r+1}(t,x)}\right)\right)\quad\text{on }\quad(0,\infty)\times\mathbb{R}^{n},

where
r>1
and
\rho>0
is a probability density on
\mathbb{R}^{n}
.
According to the framework laid out in
[
IPS19
]
and
[
FI25
]
, one is led to study the following functional difference

\mathcal{F}_{\rho}[f]-\mathcal{F}_{\rho}[m]:=\int_{\mathbb{R}^{n}}\frac{\rho(x)}{f^{r}(x)}dx-\int_{\mathbb{R}^{n}}\frac{\rho(x)}{m^{r}(x)}dx,

where, for the solution
f
we abuse the notation by dropping out the
t
variable, and

m:=\gamma\rho^{1/(r+1)}\quad\text{ with}\quad\gamma:=\frac{1}{\int_{\mathbb{R}^{n}}\rho^{1/(r+1)}dx}.

Here
m
is the equilibrium density of (
1
) and
\gamma
is the normalisation constant.

We have the following natural two-sided bounds.

Lemma 1
.

Let
\mathcal{P}(\mathbb{R}^{n})
denote the space of probability densities on
\mathbb{R}^{n}
.
Assume that

f_{0}\in\mathcal{P}_{c,C}:=\{g\in\mathcal{P}(\mathbb{R}^{n}):cm\leq g\leq Cm\},\quad C\geq 1\geq c>0.

Then there exist constants
k_{1}
and
k_{2}
, depending on
r
,
c
,
C
and
\gamma
, such that

\displaystyle k_{1}\int_{\mathbb{R}^{n}}\left|\frac{f}{m}-1\right|^{2}mdx\leq\mathcal{F}_{\rho}[f]-\mathcal{F}_{\rho}[m]\leq k_{2}\int_{\mathbb{R}^{n}}\left|\frac{f}{m}-1\right|^{2}mdx.

Remark 2
.

Under the assumption
f_{0}\in\mathcal{P}_{c,C}
,
the existence and uniqueness of a solution
f(t)
to (
1
) with
f(t,\cdot)\in\mathcal{P}_{c,C}
for all
t>0
will be justified in Appendix
A
.

As already well observed in
[
IPS19
]
and
[
FI25
]
,
the exponential convergence of solutions
f(t,\cdot)
to equilibrium
m
can be reduced to the following crucial estimate

(2)

\mathcal{F}_{\rho}[f]-\mathcal{F}_{\rho}[m]\leq K\,I_{\rho}[f],\quad\forall f\in\mathcal{P}_{c,C},

where
K
is a constant and, by letting
u=f/m
,

I_{\rho}[f]:=-\frac{d}{dt}\mathcal{F}_{\rho}[f]=r^{2}\int_{\mathbb{R}^{n}}u|\nabla(u^{-(r+1)})|^{2}mdx.

Given above lemma and in order to establish (
2
), it is natural to invoke

Definition 3

(Poincaré Inequality)
.

Let
m
be a probability measure on
\mathbb{R}^{n}
.
We say that the Poincaré inequality (associated to
m
) holds, with constant
C_{P}=C_{P}(m)
, if

\int_{\mathbb{R}^{n}}g^{2}mdx-\left(\int_{\mathbb{R}^{n}}gmdx\right)^{2}\leq C_{P}\int_{\mathbb{R}^{n}}|\nabla g|^{2}mdx,\quad\forall\,g.

Remark 4
.

Log-concave measures (in particular, Gaussian measures) satisfy the Poincaré inequality (see Bobkov
[
Bob99
]
).
Certain non-log-concave measures also satisfy the Poincaré inequality (see
[
BCE13
, Theorem 3]
and
[
BJ21
, Theorem 3.2]
).

Now we are able to state the main contribution of this paper.

Theorem 5
.

The estimate (
2
) holds with

K=\frac{C_{P}C^{2r+3}}{2r(r+1)\gamma^{r+1}c^{r+2}}.

Then, by Gronwall type arguments in conjunction with Lemma
1
, we obtain the exponential convergence to equilibrium for ultrafast diffusion equations in
\mathbb{R}^{n}
.

Theorem 6
.

Let
f(t)
be the solution to (
1
) with
f_{0}\in\mathcal{P}_{c,C}
.
Then for all
t>0
,

\int_{\mathbb{R}^{n}}\left|\frac{f(t)}{m}-1\right|^{2}mdx\leq\frac{1}{k_{1}}\left(\int_{\mathbb{R}^{n}}\frac{\rho(x)}{f_{0}^{r}(x)}dx-\frac{1}{\gamma^{r+1}}\right)\cdot e^{-\frac{1}{K}t},

where
k_{1}
is given in Lemma
1
and
K
is given in Theorem
5
.

Already for
m
being the Gaussians, our result answers the open problems in
[
FI25
]
.
Also, for compactly supported
f_{0}
and
m
, Theorem
6
covers
[
IPS19
, Theorem 1.4]
.

2.
Proofs

Proof of Lemma
1
.

Recall
u=\frac{f}{m}
. By a simple calculation, we obtain

(3)

\displaystyle\mathcal{F}_{\rho}[f]-\mathcal{F}_{\rho}[m]

\displaystyle=\frac{1}{\gamma^{r+1}}\int_{\mathbb{R}^{n}}\left(u^{-r}-1\right)mdx

\displaystyle=\frac{1}{\gamma^{r+1}}\int_{\mathbb{R}^{n}}\left(u^{-r}-1+r\left(u-1\right)\right)mdx.

Considering the Taylor expansion of
x^{-r}
at
x=1
, (
3
) can be rewritten as

\mathcal{F}_{\rho}[f]-\mathcal{F}_{\rho}[m]=\frac{r(r+1)}{2\gamma^{r+1}}\int_{\mathbb{R}^{n}}\theta^{-(r+2)}|u-1|^{2}mdx,

where
\theta\in(\min\{1,u\},\max\{1,u\})
. Since
0<c\leq u\leq C
(see Remark
2
), we obtain

\displaystyle\frac{r(r+1)}{2\gamma^{r+1}}C^{-(r+2)}\int_{\mathbb{R}^{n}}\left|u-1\right|^{2}mdx

\displaystyle\leq\mathcal{F}_{\rho}[f]-\mathcal{F}_{\rho}[m]

\displaystyle\leq\frac{r(r+1)}{2\gamma^{r+1}}c^{-(r+2)}\int_{\mathbb{R}^{n}}\left|u-1\right|^{2}mdx.

Here we also used implicitly
C\geq 1\geq c
. This proves the lemma.
∎

Proof of Theorem
5
.

Applying Poincaré inequality to
g=u-1=\frac{f}{m}-1
, and noting

\int_{\mathbb{R}^{n}}gm=\int_{\mathbb{R}^{n}}f-\int_{\mathbb{R}^{n}}m=1-1=0,

we obtain (upon tracing the expression of
k_{2}
in the proof of Lemma
1
)

\mathcal{F}_{\rho}[f]-\mathcal{F}_{\rho}[m]\leq C_{P}\frac{r(r+1)}{2\gamma^{r+1}}c^{-(r+2)}\int_{\mathbb{R}^{n}}\left|\nabla u\right|^{2}mdx.

Since
0<c\leq u\leq C
(see Remark
2
), we also have

\int_{\mathbb{R}^{n}}\left|\nabla u\right|^{2}mdx\leq\frac{C^{2r+3}}{(r+1)^{2}}\int_{\mathbb{R}^{n}}u|\nabla(u^{-(r+1)})|^{2}mdx=\frac{C^{2r+3}}{r^{2}(r+1)^{2}}I_{\rho}[f].

Collecting the above two inequalities then proves the theorem.
∎

Appendix A
On the existence and uniqueness of solutions

Here adapting
[
FI25
, Appendix A]
for the one-dimensional case,
we show the well-posedness of the ultrafast diffusion equation (
1
) in
\mathbb{R}^{n}
in the class
\mathcal{P}_{c,C}
.

Let
f_{0}\in\mathcal{P}_{c,C}
for some
c,C>0
and write
m=e^{-V}
.
Given
k\geq 1
, we consider

V_{k}:=\begin{cases}a_{k}V,&\text{in }B(0,k),\\[6.0pt]
+\infty,&\text{otherwise},\end{cases}

where
a_{k}>0
is chosen such that
m_{k}:=e^{-V_{k}}
is a probability density in
\mathbb{R}^{n}
,
or more precisely, in
B(0,k):=\{x\in\mathbb{R}^{n}:|x|\leq k\}
. Also, we consider the initial data

f_{0}^{k}:=\begin{cases}b_{k}f_{0},&\text{in }B(0,k),\\[6.0pt]
0,&\text{otherwise},\end{cases}

where
b_{k}>0
is chosen such that
f_{0}^{k}\in\mathcal{P}(B(0,k))
(that is,
f_{0}^{k}
is a probability density in
B(0,k)
).
Note that
a_{k}
,
b_{k}\to 1
as
k\to+\infty
, hence

f_{0}^{k}\in\mathcal{P}_{c_{k},C_{k}}^{k}:=\{g\in\mathcal{P}(B(0,k)):c_{k}m_{k}\leq g\leq C_{k}m_{k}\},

where
c_{k}\to c
and
C_{k}\to C
as
k\to+\infty
.

Now, since
m_{k}
is compactly supported, applying
[
IPS19
, Theorem 1.2]

we obtain the existence and uniqueness of the solution
f^{k}(t)\in\mathcal{P}^{k}_{c_{k},C_{k}}
to the ultrafast diffusion equation posed on
B(0,k)

with Neumann boundary conditions and initial data
f_{0}^{k}
.
In the process we have to invoke the maximum principle (see
[
IPS19
, Corollary 3.8]
) to guarantee that

f^{k}(t)
inherits faithfully the constants
c_{k}
and
C_{k}
from
f_{0}^{k}\in\mathcal{P}^{k}_{c_{k},C_{k}}
.
Moreover, for any
R<k
, we obtain from
[
IPS19
, Lemma 3.3]
the compactness of the family
\{f^{k}|_{B(0,R)}\}
.
Hence, recalling that
c_{k}\to c
and
C_{k}\to C
as
k\to+\infty
,
by a diagonal argument we see that
f^{k}
converges (up to a subsequence) to
f\in\mathcal{P}_{c,C}
, where
f(0)=f_{0}\in\mathcal{P}_{c,C}
and
f(t)
solves (
1
).
Also, upon localisation of
f_{0}
, the uniqueness of
f
follows from the
L^{1}
-contractivity (see
[
IPS19
, Proposition 3.6]
).

Compliance with ethical standards

Conflict of interest
The authors have no known competing financial interests or personal relationships that could have appeared to influence this reported work.

Availability of data and material

Read the complete text