You are on page 1of 15

Journal of Geometry and Physics 170 (2021) 104382

Contents lists available at ScienceDirect

Journal of Geometry and Physics


www.elsevier.com/locate/geomphys

Harnack inequalities for a class of heat flows with nonlinear


reaction terms
Abimbola Abolarinwa a,∗ , Julius Osato Ehigie a , Ali H. Alkhaldi b
a
Department of Mathematics, Faculty of Science, University of Lagos, Akoka, Lagos, Nigeria
b
Department of Mathematics, College of Science, King Khalid University, 9004, Abha 61413, Saudi Arabia

a r t i c l e i n f o a b s t r a c t

Article history: A class of semilinear heat flows with general nonlinear reaction terms is considered on
Received 6 August 2021 complete Riemannian manifolds with Ricci curvature bounded from below. Two types of
Received in revised form 8 September 2021 (space-time and space only) gradient estimates are established for positive solutions to
Accepted 16 September 2021
the flow, and the corresponding Harnack inequalities are obtained to allow for compari-
Available online 23 September 2021
son of solutions. Some specific examples of the reaction term such as logarithmic reaction,
MSC: Fisher-KPP and Allen-Cahn equations are discussed as applications of the estimates so de-
35B50 rived. Referring to logarithmic nonlinearities, some discussions are made on Liouville type
53C44 properties of bounded solutions.
58J60 © 2021 Elsevier B.V. All rights reserved.
58J35
60J60

Keywords:
Riemannian manifolds
Gradient estimates
Harnack inequalities
Ricci curvature
Reaction-diffusion equation

1. Introduction

The main aim of this paper is to establish Li-Yau type (space-time) and Hamilton-Souplet-Zhang type (space only) Har-
nack inequalities for positive solutions w (x, t ) to a class of heat flows with nonlinear reactions


w (x, t ) =  w (x, t ) + g ( w (x, t )) (1.1)
∂t
on an n-dimensional complete (compact or noncompact) Riemannian manifold M, where t ∈ [0, ∞) and g (·) is a sufficiently
smooth nonlinear function on (0, ∞) and 0 < w = w (x, t ) is assumed smooth, atleast C 2 in x-variable and C 1 in t-variable.
Here, ∂∂wt , which may sometimes be written as w t , is the time derivative of w and  is the Laplace-Beltrami operator, while
the term  w accounts for the diffusion of heat. Thus (1.1) can be viewed as a reaction-diffusion model with ubiquitous
applications in sciences and engineering. If w is a stationary solution, that is w t = 0, then (1.1) becomes a generalised

*
Corresponding author.
E-mail addresses: A.Abolarinwa1@gmail.com, aabolarinwa@unilag.edu.ng (A. Abolarinwa), jehigie@unilag.edu.ng (J.O. Ehigie), ahalkhaldi@kku.edu.sa
(A.H. Alkhaldi).

https://doi.org/10.1016/j.geomphys.2021.104382
0393-0440/© 2021 Elsevier B.V. All rights reserved.
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

Poisson equation. The nonlinear reaction term appearing in (1.1) includes several cases from biology, physics and geometry.
For examples

(1) The case g ( w ) = c w (1 − w ), c > 0, 0 < w ≤ 1, is the Fisher-KPP equation (see [24] in one dimension and [33] in
two dimension), where w (x, t ) is the population density of genes at time t and w (1 − w ) is known as logistic growth
function [7,34]. In this case, (1.1) is related to Fitzhugh-Nagumo equation [25] which is used to model impulse growth
in nerve axons. It is also applicable in the studies of flame propagation and nuclear reactors, combustion theory and
heat transfer, [36,47].
(2) The case g ( w ) = c w (1 − w 2 ), c > 0, 0 < w ≤ 1, is the Allen-Cahn equation used in the theory of phase separation
in iron alloys, including order-disorder transitions [8]. Allen-Cahn equation is well studied in literature (see [14,21]),
while its connection with the theory of minimal hypersurfaces in differential geometry has been greatly exploited by
several authors, see [18,23,32] and references therein for instance. It also gave rise to the popular De Giorgi conjecture
[20] which states that an entire solution to the steady state part with | w | ≤ 1 in Rn , n ≤ 8, which is monotone in one
direction is necessarily one dimensional (see [28] and [9] for the case n = 2 and n = 3, respectively).
(3) The case g ( w ) = aw (log w )α , a, α ∈ R, is the evolution equation with logarithmic nonlinearity (see [44] for instance).
When α = 1, the stationary part of the equation can be linked with gradient Ricci solitons and logarithmic Sobolev
inequalities [22,27]. Recall that the gradient Ricci solitons are self similar solutions to the Ricci flow, while the logarith-
mic Sobolev inequalities are infinite dimensional version of Sobolev inequalities widely applied in the theory of partial
differential equations and constructive quantum field theory. The logarithmic Schrödinger equation

iut = (ε  − V )u + u log |u |2 on Rn ,
where V is a certain potential function, has considerable application in quantum mechanics, quantum optics, theory of
superfluidity, effective quantum gravity (see [51] and the reference therein).
(4) The case g ( w ) = | w |b−1 w , b > 1, is also of physical and geometric interests, we refer to [16,17,42,43] and the ref-
erences therein. Precisely, the authors in [43] studied the entire solution (see [42] for the steady state) in the case
1 < b < p  , where p  = nn+ 2
−2 for n ≥ 3 and p  = ∞ for n = 1, 2, is the critical Sobolev exponent, and they obtained
n(n+2)
Liouville type theorem for radial solutions. The cases b = 2 and 1 < b < (n−1)2 were respectively studied in [16] and
[17] for ancient solutions on Riemanniqan manifolds.
(5) The nonlocal heat flow preserving L 2 -norm on a closed manifold is another important case for

g ( w ) = λ(t ) w + A (x, t )
 
with λ(t ) := − M ( w  w + w A )dv and M w 2 dv = 1 chosen to preserve the norm. This case has diverse interpretations
as well [12,38].
(6) Consider the case g ( w ) = aw p − bw q , a, b > 0, q > p ≥ 1. The model (1.1) (with g ( w ) = aw p − bw q ) covers Yamabe-type
equation [1,40], Lichnerowich-type equation [22] and some other parabolic equations of mathematical physics values.

In the seminal paper [37], Li and Yau considered Schrödinger type heat flow
∂u
= u + q(x, t )u
∂t
on a general manifold, where q(x, t ) is restricted to satisfying C 2 in x and C 1 in t. Precisely, they arrived at the following
gradient estimates for q(x, t ) ≡ 0

|∇ u |2 ut nβ 2 k nβ 2
−β ≤ + , β >1 (1.2)
u2 u 2(β − 1) 2t
on a Riemannian manifold whose Ricci curvature, Ric ≥ −k, k ≥ 0. They then applied (1.2) to deducing the following
Harnack inequality
 t  n2β
2
u (x1 , t 1 ) ≤ u (x2 , t 2 ) e , (1.3)
t1
β d2 (x ,x ) β nk
where := 4(t −1 t 2) + 2(β−1) (t 2 − t 1 ), d(x1 , x2 ) is the geodesic distance between points x1 and x2 in M, 0 < t 1 < t 2 and ∇
2 1
is the gradient operator. The above Harnack inequality (1.3) was in turn used to prove various bounds on the fundamen-
tal solution (heat kernel) to the linear heat equation. Li-Yau gradient estimates can also be used to derive Liouville type
theorems for parabolic equations, estimates on Green’s functions and various bounds on Laplacian eigenvalues.
The Harnack inequalities of type (1.3) can only be used to compare solutions at different time, since the gradient type
estimates of the form (1.2) are time dependent. In order to circumvent this shortcoming, elliptic type (space only) gradient
estimates were developed, first by Hamilton [30] on closed manifolds, and later by Souplet and Zhang [46] on complete
noncompact manifolds, while their associated Harnack inequalities read as

2
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

u (x1 , t ) ≤ u (x2 , t )A(d(x1 ,x2 ),t ) e 1−A(d(x1 ,x2 ),t ) , (1.4)


which can thus be used to compare solutions at the same time.
Since the appearance of the above results, gradient and Harnack estimates have turned out to become one of the most
powerful tools in geometric analysis and the theory of linear and nonlinear partial differential equations. Li-Yau Harnack
inequalities played vital role in Perelman’s proof of noncollapsing theorem for the Ricci flow [41], which eventually led
to the final resolution of Poincaré conjecture. Quite a large number of literature has considered this topic under various
settings. For instance, the authors [10,50] found appropriate space-time and space only gradient estimates on bounded
solution to the heat equation on domains evolving by the Ricci flow. In a similar vein, the first author [2] derived Harnack
inequalities via Perelman type entropy formula under the Ricci-harmonic flow and in [3,4] derived Li-Yau gradient estimates
on heat type equation with potential under general geometric flow. Ma and Cheng [38] extended Li-Yau gradient estimates
to L 2 -norm preserving nonlocal heat flows on closed manifolds. Wang [48], on the other hand, extended Ma and Cheng’s
result to the Ricci flow setting. The first author [5], Bǎileşteanu [11] and Hou [31] studied differential Harnack inequalities
and gradient estimates in relation to the bounded solutions of Allen-Cahn equation. Differential Harnack estimates and
gradient estimates were also considered for Fisher-KPP equation in [15] and [26]. Wu [49] considered space-only time
gradient estimates on weighted heat equation, the first author [1] considered nonlinear weighted parabolic equation of
Yamabe type and in [6] weighted heat equation with logarithmic nonlinearity, Chen and Zhao [19] and Ma and Zeng [39]
studied this concept with respect to certain nonlinear parabolic equations (see also the recent paper by Dung and Khanh
[22]). We refer to Li and Xu [35] for a comprehensive discussion on the development of Harnack inequalities for linear heat
equation on manifolds and their applications to Perelman type entropies, Liouville type results and heat kernel estimates.
In this paper, we are concerned with space-time and space only gradient estimates and their associated Harnack inequal-
ities for the heat flow with nonlinear reaction term (1.1). The motivation for this stems out of the geometric and physical
applications of such a class of equations as highlighted above. The results of this paper therefore provide a unified treatment
of a large class of reaction-diffusion equations with nonlinear reaction terms. The rest of the paper is structured as follows.
Section 2 is devoted to discussing space-time gradient estimates and several Li-Yau Harnack type inequalities are derived as
corollaries. Specific estimates are shown for some semilinear heat equations in this class. In Section 3 we present space only
gradient estimates and their corresponding Harnack inequalities for bounded positive solutions. Some examples are given to
complement the results.

2. Li-Yau Harnack inequalities

Consider the positive smooth solution w = w (x, t ) to the nonlinear heat flow equation

wt = w + g(w) (2.1)

on M × [0, ∞). Let h = log w and define the Harnack quantity

F (x, t ) = t [|∇ h|2 − α (ht − ĝ (h))]

for α > 1, where ĝ (h) = g ( w )/ w. A simple computation using (2.1) shows that h satisfies

ht = h + |∇ h|2 + ĝ (h). (2.2)

The main results concerning the parabolic Harnack inequalities are discussed in this section. The required Li-Yau type
gradient estimates are stated and proved first. Meanwhile, we introduce the following notations for h = log w:

g(w) g(w) g(w)


ĝ (h) := , 
gh := g ( w ) − 
ghh := w g ( w ) − g ( w ) + .
w w w
Also

σ1+ := sup {
gh } and σ2− := inf {
ghh }.
M ×[0, T ] M ×[0, T ]

We like to remark that very crucial in the applications of the space-time gradient estimates is the solution dependent
quantity

c α ,n (σ1+ − σ2− ) ≥ 0,
where c α ,n > 0 is a constant depending on n and α.

Theorem 2.1. Let M be an n-dimensional complete compact manifold without boundary having Ric ( M ) ≥ −k, k ≥ 0. Suppose w (x, t )
is a positive smooth solution of (2.1) on M × [0, T ], T < ∞. There exists a time dependent constant A = A (α , n, T , σ1+ , σ2− , k) such
that

3
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

sup F ≤ A (α , n, T , σ1+ , σ2− , k), (2.3)


M ×[0, T ]

where F = t [|∇ h|2 − α (ht − ĝ (h))], h = log w, α > 1, σ1+ := sup M ×[0, T ] {
gh } and σ2− := inf M ×[0, T ] {
ghh }.

More explicitly under the same hypotheses as that of the previous theorem we have

Theorem 2.2. With the same hypotheses as of Theorem 2.1

α 2n
sup {|∇ h|2 − α ht + α ĝ (h)} ≤ A 1 (α , n, σ1+ , σ2− , k) + , (2.4)
M ×[0, T ] 2t

and

α 2nk α 2n
sup {|∇ h|2 − α ht + α ĝ (h)} ≤ A 1 (α , n, σ1+ , σ2− ) + + , (2.5)
M ×[0, T ] 2(α − 1) 2t

where σ1+ and σ2− are as defined above.

Proof of Theorem 2.1. Starting with the Harnack quantity

F = t [|∇ h|2 − α (ht − ĝ (h))],

we notice that
F
= |∇ h|2 − α (ht − ĝ (h))
t
and by using (2.2) we have

F α−1
h = − − |∇ h|2 .
αt α
Using the time evolution of |∇ h|2 , i.e., (|∇ h|2 )t = 2∇ h∇ ht , and the Bochner-Weitzenböck formula

(|∇ h|2 ) = 2|∇ 2 h|2 + 2∇ h∇ h + 2Ric (∇ h, ∇ h),


we obtain

(∂t − )|∇ h|2 = 2∇ h∇(|∇ h|2 + ĝ (h)) − 2(|∇ 2 h|2 + Ric (∇ h, ∇ h))
 
F
= 2∇ h ∇ + (1 − α ) ĝ (h) + αht − 2(|∇ 2 h|2 + Ric (∇ h, ∇ h)).
t
Note also that

(ht ) = ht + 2∇ h∇ ht + ĝt (h),
∂t
which implies

(∂t − )ht = 2∇ h∇ ht + ĝt (h)


and we therefore obtain
F 
(∂t − )(|∇ h|2 − αht ) =2∇ h∇ + (1 − α ) ĝ (h) + αht − 2(|∇ 2 h|2 + Ric (∇ h, ∇ h))
t
− 2α ∇ h∇ ht − α ĝt (h).
Thus,
 
F
(∂t − ) = (∂t − )[|∇ h|2 − α (ht − ĝ (h))]
t
= (∂t − )(|∇ h|2 − αht ) + α (∂t − ) ĝ (h)
F 
= 2∇ h ∇ + (1 − α ) ĝ (h) − 2(|∇ 2 h|2 + Ric (∇ h, ∇ h)) − α  ĝ (h).
t
Clearly, putting the above expression together we have

4
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

 
(∂t − ) F = (∂t − ) t [|∇ h|2 − α (ht − ĝ (h))] (2.6)
F F 
= + 2t ∇ h∇ + (1 − α ) ĝ (h) − 2t (|∇ 2 h|2 + Ric (∇ h, ∇ h)) − αt  ĝ (h).
t t
Now, we suppose sup M ×(0, T ] F > 0 at the point (x0 , t 0 ) ((x0 , t 0 ) being the maximum point on M × [0, T ] for some T > 0).
By the maximum principle then ∇ F = 0 and (∂t − ) F ≥ 0 at (x0 , t 0 ). Therefore by (2.6) we obtain

F
− αt 0  ĝ (h) ≥ 2t 0 (|∇ 2 h|2 + Ric (∇ h, ∇ h)) + 2t 0 (α − 1)∇ h∇ ĝ (h). (2.7)
t0
Applying the Ricci tensor condition, Ric ( M ) ≥ −k, k ≥ 0, and the Cauchy-Schwarz inequality one can easily derive the
following inequality

1
|∇ 2 h|2 + Ric (∇ h, ∇ h) ≥ (h)2 − k|∇ h|2 ,
n
which when combined with (2.7) yields
1 
F − αt 02  ĝ (h) ≥ 2t 02 (h)2 − k|∇ h|2 + 2t 02 (α − 1)∇ h∇ ĝ (h). (2.8)
n
|∇ h|2
Setting := (x0 , t 0 ), and we then suppose ≥ 0, otherwise the proposition of the theorem becomes trivial.
F
By direct computation we have

g(w) ∇w g(w) ∇ w
∇ ĝ (h) = ∇( ) = g (w) − =: 
gh ∇ h ,
w w w w
where 
gh := g ( w ) − g ( w )/ w, and similarly,

gh =: 
∇ ghh ∇ h,

where 
ghh := w g ( w ) − g ( w ) + g ( w )/ w. Noting that |∇ h|2 = F at (x0 , t 0 ), then

∇ h∇ ĝ (h) = 
gh |∇ h|2 = 
gh F (2.9)

and

gh = 
∇ h∇ ghh |∇ h|2 = 
ghh F . (2.10)

Thus,

 ĝ (h) = div( gh ∇ h + 
gh ∇ h) = ∇ gh h
 F α−1 
= ∇gh ∇ h + 
gh − − |∇ h|2
αt 0 α
α−1 F
gh ∇ h − 
= ∇ gh − 
g F
αt 0 α h
and then
F α−1
ghh F + 
− ĝ (h) = − gh + 
g F. (2.11)
αt 0 α h
One can also compute that

1 1 F α−1 2
(h)2 − k|∇ h|2 = + |∇ h|2 − k|∇ h|2
n n αt 0 α
λα t 0
= F2 −k F, (2.12)
nα 2 t 02

where λαt0 := (1 + (α − 1)t 0 )2 .


Inserting (2.9)–(2.12) into (2.8) we get

2 λα t 0
F − αt 02
ghh F + t 0
gh F + t 02 (α − 1)
gh F≥ F 2 − 2kt 02 F + 2t 02 (α − 1)
gh F.
nα 2
Upon rearranging the last inequality we have

5
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

2F t 02  1
2
≤ gh − α
2k − (α − 1) ghh + (1 + t 0
gh ) (2.13)
nα λα t 0 λα t 0
at (x0 , t 0 ).
Noting that (1 + (α − 1)t 0 ) ≥ 1, then we have

1 1
= ≤1
λα t 0 (1 + (α − 1)t 0 )2
and
1 (α − 1) 1
= = ≤ .
λα t 0 (1 + (α − 1)t 0 )2 (α − 1) (1 + (α − 1)t 0 )2 2(α − 1)t 0
Hence, for F at (x0 , t 0 ) (2.13) implies

nα 2 t 0  nα 2
F≤ gh − α
2k − (α − 1) ghh + (1 + t 0
gh )
4(α − 1) 2
nα 2 t 0  nα 2
= gh − α
2k + (α − 1) ghh + (2.14)
4(α − 1) 2
nα 2 t 0  nα 2
≤ 2k + (α − 1)σ1+ − ασ2− + ,
4(α − 1) 2
where α > 1, σ1+ := sup M ×[0, T ] {
gh } and σ2− := infM ×[0,T ] {
ghh }.
Equivalently

nα 2
F (x0 , t 0 ) ≤ A 1 (n, α , t 0 , σ1+ , σ2− , k) + (2.15)
2
or

nα 2 t 0 k nα 2
F (x0 , t 0 ) ≤ A 2 (n, α , t 0 , σ1+ , σ2− ) + + . (2.16)
2(α − 1) 2
This therefore completes the proof of (2.3). 

Proof of Theorem 2.2. Clearly, (2.4) and (2.5) follow directly from (2.15) and (2.16), respectively. 

Next, we derive the Li-Yau parabolic type Harnack inequalities from the preceding gradient estimates.

Corollary 2.3. Let M be an n-dimensional complete compact manifold without boundary having Ric ( M ) ≥ −k, k ≥ 0. Suppose w =
w (x, t ) is a positive smooth solution of (2.1) on M × [0, T ], T < ∞ such that 0 < t 1 < t 2 ≤ T . Then the following inequalities hold for
any α > 1
 t  n2α
2
w (x1 , t 1 ) ≤ w (x2 , t 2 ) e 1
, (2.17)
t1
2 1
where 1 := α4d(t(2x−1 t,1x2) ) + (t 2 − t 1 ) 0 1 ds and 1 = A 1 (n, α , σ1+ , σ2− , k) − ĝ (h).
 t  n2α
2
w (x1 , t 1 ) ≤ w (x2 , t 2 ) e 2
, (2.18)
t1
2
 1 
where 2 := α4d(t(2x−1 t,1x2) ) + (t 2 − t 1 ) 2(nαα−k1) + 0 2 ds and 2 = A 1 (n, α , σ1+ , σ2− ) − ĝ (h). For Ric ( M ) ≥ 0
 t  n2
2
w (x1 , t 1 ) ≤ w (x2 , t 2 ) e 3
, (2.19)
t1
d2 (x ,x )
1
where 3 := 4(t 1−t 2) + (t 2 − t 1 ) 0 3 ds and 3 = A 3 (n, σ1+ , σ2− ) − ĝ (h).
2 1
The approach to the proof of the above corollary is traditional [37,45].

Proof. Let γ (s) be a shortest geodesic joining x1 and x2 with γ : [0, 1] → M such that γ (0) = x2 and γ (1) = x1 . Define
a smooth curve η in M × [0, T ] by η : [0, 1] → M × [0, T ] with η(s) = (γ (s), τ (s)) and τ (s) = (1 − s)t 2 + st 1 . Obviously
η(0) = (x2 , t 2 ) and η(1) = (x1 , t 1 ). Let h = log w, then integrating (d/ds)(h(η(s)) along η, we have

6
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

1
d
h(x1 , t 1 ) − h(x2 , t 2 ) = h(η(s))ds
ds
0
1
 γ̇ , ∇ h 
= (t 2 − t 1 ) − ht ds
t2 − t1
0
1
 |γ̇ ||∇ h| 
≤ (t 2 − t 1 ) − ht ds. (2.20)
t2 − t1
0

Note that |γ̇ | = d(x1 , x2 ). Using the Harnack estimate (2.3) we have

1 nα
−ht ≤ A 1 (n, α , σ1+ , σ2− , k) − ĝ (h) − |∇ h|2 + . (2.21)
α 2t
Using (2.21), the integrand in (2.20) is estimated as follows

|γ̇ ||∇ h| |γ̇ ||∇ h| 1 nα


− ht ≤ − |∇ h|2 + A 1 − ĝ (h) +
t2 − t1 t2 − t1 α 2t
2
αd (x1 , x2 ) nα
≤ + A 1 − ĝ (h) + ,
4(t 2 − t 1 )2 2t
where we have used the fact that the quadratic function −ay 2 + by (with y = |∇ h|) attains its maximum at b2 /4a for
a > 0, b > 0.
Letting 1 (x, t ) := A 1 − ĝ (h), then (2.20) implies

w (x1 , t 1 )
log = h(x1 , t 1 ) − h(x2 , t 2 )
w (x2 , t 2 )
1 1
αd2 (x1 , x2 ) nα ds
≤ + (t 2 − t 1 ) ds + (t 2 − t 1 )
4(t 2 − t 1 ) 2 (1 − s)t 2 + st 1
0 0
1
αd2 (x1 , x2 ) nα t2
≤ + (t 2 − t 1 ) ds + log( ).
4(t 2 − t 1 ) 2 t1
0

Clearly, estimate (2.17) follows by exponentiating both sides of the last inequality.
In a similar manner we can obtain (2.18) by using the gradient estimate (2.5) to get

1 nα k nα
−ht ≤ A 2 (n, α , σ1+ , σ2− ) − ĝ (h) − |∇ h|2 + +
α 2(α − 1) 2t
and then by following the procedure.
Finally, setting k ≡ 0 and α → 1 in the Harnack estimate (2.18) gives rise to (2.19) since Ric ( M ) ≥ 0. 

Remark 2.4. We can write (2.16) more explicitly by reverting to (2.14) as follows

α 2n α 2nk α 2n + α
sup {|∇ h|2 − α ht + α ĝ (h)} ≤ + + σ1 − σ− .
M ×[0, T ] 2t 2(α − 1) 4 α−1 2
Thus (2.5) reads

α 2n α 2nk
sup {|∇ h|2 − α ht + α ĝ (h)} ≤ + + c α ,n (σ1+ − σ2− ), (2.22)
M ×[0, T ] 2t 2(α − 1)

where σ1+ := supM ×[0,T ] {


gh } and σ2− := inf M ×[0, T ] {
ghh }.

Finally in this section we give the following two examples as an application of the above estimates.

Example 2.5. Consider the heat flow with logarithmic nonlinearity of the form g ( w ) = aw log w, a ∈ R. Observe that ĝ (h) =
ah,  ghh = 0. Also σ1+ = a and σ2− = 0.
gh = a and 

7
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

Then, global gradient estimates can be obtained for the positive smooth solutions to the nonlinear heat equation

w t =  w + aw log w (2.23)
on M × [0, T ].

Corollary 2.6. Let w = w (x, t ) be a positive smooth solution to (2.23) on M × [0, T ], where M is an n-dimensional complete compact
Riemannian manifold (without boundary) satisfying Ric ( M ) ≥ −k, k ≥ 0. Then

|∇ w |2 wt nα 2
−α + αa log w ≤ + c α ,n (k + a), (2.24)
w2 w 2t
for α > 1, where c α ,n > 0 depends on α and n only.
Furthermore, if Ric ( M ) ≥ 0, then

|∇ w |2 wt n
− + a log w ≤ + cn a, (2.25)
w2 w 2t
where cn is a positive constant depending only on n.

Proof. The proof of (2.24) follows from Theorem 2.2 by using σ1+ = a and σ2− = 0 in (2.22). To obtain (2.25) we can set
k ≡ 0 and α → 1 since Ric ( M ) ≥ 0. 

Estimate (2.25) reduces to Li-Yau gradient estimate for the linear heat equation (that is, a = 0 in (2.23)) on M with
Ric ( M ) ≥ 0, that is,

|∇ w |2 wt n
− ≤ ,
w2 w 2t
which is optimal in the case M = Rn . Thus, if one solves ut = u on Rn × (0, ∞) one will obtain

|∇ u |2 ut n
− =
u2 u 2t
by considering the fundamental solution (Gauss heat kernel) u (x, t ) = (4π t )−n/2 e −|x| /4t on Rn × (0, ∞).
2

Estimate (2.24) looks better when restricting (2.23) to a ≤ −k on M with Ric ( M ) ≥ −k, k ≥ 0 or a ≤ 0 on M with
Ric ( M ) ≥ 0. An implication of this observation leads to derivation of Liouville type properties of bounded solutions to
(2.23).

Theorem 2.7. Let M be an n-dimensional complete compact Riemannian manifold without boundary. Consider

 w + aw log w = 0. (2.26)

(1) Suppose Ric ( M ) ≥ −k, k ≥ 0 and a ≤ −k then solution 0 < w ≤ 1 to (2.26) is a constant. If, in addition, a < 0, then w ≡ 1
identically.
(2) Suppose Ric ( M ) ≥ 0 and a ≤ 0 then solution 0 < w ≤ 1 to (2.26) is a constant. If, in addition, a < 0, then w ≡ 1 identically.

Proof. In the case Ric ( M ) ≥ −k, k ≥ 0, setting a + k ≤ 0 and t → +∞ (with observation that c α ,n > 0 is time independent),
we have

|∇ w |2
≤ −αa log w ≤ αkh ≤ 0,
w2
since α > 1, k ≥ 0 and h = log w ≤ 0. This therefore implies that ∇ w ≡ 0, and consequently w is a constant and w = 1
identically. 

Example 2.8. Consider the Fisher-KPP equation

w t =  w + aw (1 − w ), a > 0, 0 < w ≤ 1 (2.27)

and the parabolic Allen-Cahn equation

w t =  w + aw (1 − w 2 ), a > 0, 0 < w ≤ 1 (2.28)


on M × [0, ∞).

8
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

In (2.27), g ( w ) = aw (1 − w ) and by direct computation


gh = −aw = 
ghh

implying that

σ1+ := sup {
gh } = a sup {− w } = −a inf { w } ≤ 0,
M ×[0, T ] M ×[0, T ] M ×[0, T ]

while

σ2− := inf {
ghh } = a inf {− w } ≥ −a sup { w } ≥ −a
M ×[0, T ] M ×[0, T ] M ×[0, T ]

It then holds (from (2.22)) that

|∇ w |2 wt nα 2
−α + αa(1 − w ) ≤ + c α ,n (k + a)
w2 w 2t
for α > 1 and cn,α > 0 on M × [0, T ] satisfying Ric ( M ) ≥ −k, k ≥ 0.
Similar result can be obtained for (2.28). Here 
gh = −2aw 2 , ghh = −4aw 2 , σ1+ ≤ 0 and σ2− ≥ −4a.

3. Souplet-Zhang Harnack type inequalities

In this section we first consider space only gradient estimates which will lead to elliptic Harnack type inequality.
Let B2R (o) be the geodesic ball of radius 2R, R > 0 around the point o ∈ M. Let k(2R ) ≥ 0, we assume Ricci curvature
restricted to B2R (o) is bounded below by −k(2R ), that is, Ric (B2R (o)) ≥ −k(2R ). We then define

Q R , T = B R (x0 ) × [0, T ] ⊂ M × [0, ∞).

Consider a bounded positive solution 0 < w ≤ D (for some constant D) to the nonlinear heat flow

wt = w + g(w) (3.1)

on B2R × [0, T ]. Let h = log w, then

ht = h + |∇ h|2 + ĝ (h), (3.2)

where ĝ (h) = g ( w )/ w = e −h g ( w ). Notice that w̃ = w / D is a positive solution to

1
w̃ t =  w̃ + g ( D w̃ )
D
and 0 < w̃ ≤ 1. Thus, one can assume that 0 < w ≤ 1 in our next discussion.
We also like here that a vital role will be played in the applications of space only gradient estimates by a nonnegative,
h and ĝ (h) dependent quantity,
 1 +
M+
w := ĝ (h) + 
gh ,
1−h
where 
gh := g ( w ) − g ( w )/ w as defined before.
Next is the statement of the theorem concerning local elliptic gradient estimates on bounded positive solutions to (3.1).

Theorem 3.1. Let M be an n-dimensional complete noncompact Riemannian manifold with Ric ( M ) ≥ −k, k ≥ 0 on B2R . Suppose w
is a positive solution (3.1) satisfying 0 < w ≤ D for some constant D in Q2R , T . Then

|∇ w |  w  1 1 
(x, t ) ≤ c (n) 1 − log + √ + k + sup {M+
w} , (3.3)
w D R t Q R ,T

for all (x, t ) ∈ Q R , T /2 , where t = 0, c (n) is a positive constant depending only on n, and
 1 +
M+
w := ĝ (h) + 
gh .
1−h
9
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

Proof. By scaling w → w̃ = w / D, we assume 0 < w̃ ≤ 1 in what follows. Clearly, h = log w̃ ≤ 0 and then define W =
|∇ log(1 − h)|2 . By direct computation using (3.2)

2∇ h ∇ h t 2|∇ h|2 ht
Wt = +
(1 − h)2 (1 − h)3
2∇ h∇ h 2∇ h∇(|∇ h|2 ) 2∇ h∇ ĝ (h) 2|∇ h|2 h 2|∇ h|4 2|∇ h|2 ĝ (h)
= + + + + + .
(1 − h)2 (1 − h)2 (1 − h)2 (1 − h)3 (1 − h)3 (1 − h)3
Similarly,

2|∇ 2 h|2 2 ∇ h ∇ h 8|∇ h|2 ∇ 2 h 2|∇ h|2 h 6|∇ h|4


W = + + + +
(1 − h)2 (1 − h)2 (1 − h)3 (1 − h)3
(1 − h)4
2|∇ 2 h|2 2∇ h∇ h + 2Ric (∇ h, ∇ h) 8|∇ h|2 ∇ 2 h 2|∇ h|2 h 6|∇ h|4
= + + + + ,
(1 − h) 2 (1 − h)2 (1 − h) 3 (1 − h) 3 (1 − h)4
where we have used the Ricci identity [45] ∇ h − ∇ h = Ric ∇ h.
Therefore
 
∂ 2Ric (∇ h, ∇ h) 2|∇ 2 h|2 4|∇ h|2 ∇ 2 h 2|∇ h|4
( − ) W = − − + +
∂t (1 − h)2 (1 − h)2 (1 − h)3 (1 − h)4
 
2|∇ h|4 4|∇ h|2 ∇ 2 h 4|∇ h|2 ∇ 2 h 4|∇ h|4
+ + − −
(1 − h)3 (1 − h)2 (1 − h)3 (1 − h)4
2∇ h∇ ĝ (h) 2|∇ h|2 ĝ (h)
+ + . (3.4)
(1 − h)2 (1 − h)3
It can easily be checked that

2h 4|∇ h|2 ∇ 2 h 4|∇ h|4 4|∇ h|2 ∇ 2 h 4|∇ h|4


− ∇ h∇ W = + − − . (3.5)
1−h (1 − h)2 (1 − h)3 (1 − h)3 (1 − h)4
Note also that
 2
2|∇ 2 h|2 4|∇ h|2 ∇ 2 h 2|∇ h|4 ∇ 2h |∇ h|2
+ + =2 + ≥0 (3.6)
(1 − h)2 (1 − h)3 (1 − h)4 1−h (1 − h)2
and

∇ ĝ (h) = ∇(e −h g ( w )) = − w −1 g ( w )∇ h + w −1 ∇ g ( w )
so that

2∇ h∇ ĝ (h) 2w −1 ∇ h∇ g 2|∇ h|2 ĝ (h)


= − . (3.7)
(1 − h)2 (1 − h)2 (1 − h)2
Inserting (3.5)–(3.7) into (3.4) and using the Ricci tensor condition Ric ≥ −k, k ≥ 0 leads to

∂ 2h 2w −1 ∇ h∇ g 2h ĝ (h) W
( − ) W ≤ 2kW − ∇ h∇ W − 2(1 − h) W 2 + + . (3.8)
∂t 1−h (1 − h)2 1−h
Next we choose a smooth cut-off function η = η(x, t ) having compact support in Q2R ,T with x0 fixed in M such that the
following properties hold [50]:

(1) 0 ≤ η(r , t ) ≤ 1 with η(r , t ) ≡ 1 in Q R ,T /2 , where r = d(x, x0 ).



(2) η(r , t ) is radially decreasing and dr = 0 in Q R , T .
(3) |ηr |/ηa ≤ C a / R and |ηrr |/ηa ≤ C a / R 2 for any constant C a > 0 and 0 < a < 1.
(4) |ηt |/η1/2 ≤ C / T for some constant C > 0 and for all r ∈ [0, ∞).

Now suppose at (x0 , t 0 ) the function η W attains its maximum in Q R , T and assume (η W )(x0 , t 0 ) is positive, otherwise
there is nothing to prove. Note that η(x, t ) is smooth at (x, t ) whenever x = o and x is not in the cut locus of o [37]. Then
at (x0 , t 0 ) we have

∇(η W ) = 0, (η W )t ≥ 0 and (η W ) ≤ 0

10
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

which imply

η∇ W + ∇ η W = 0 (3.9)
and

( − )(η W ) ≥ 0. (3.10)
∂t
Thus, by (3.9) and (3.10)

∂ ∂ 2|∇ η|2
0≤( − )(η W ) = η( − ) W + W + ηt W − η W . (3.11)
∂t ∂t η
Combining (3.8) and (3.11), and rearranging yields

2|∇ η|2 2h 2w −1 ∇ h∇ g
2η(1 − h) W 2 ≤2kη W + W+ ∇ h∇ η W + η
η 1−h (1 − h)2
2h ĝ (h) W η
+ + ηt W − η W (3.12)
1−h
at (x0 , t 0 ).
The next is to use (3.12) to derive the required estimate at (x0 , t 0 ) where d(x, x0 ) ≥ 1. Thus, there is a need to bound
each term on the right hand side of (3.12). This is done by using properties (1)–(4) of η as stated above, Cauchy-Schwarz
and Young’s inequalities. (The detail computation of some of them are omitted here since the procedure is standard, see e.g.
[1,6,50].) In what follows we shall freely use constant C for possibly different constant value.
Now we estimate
 
2|∇ η|2 1 1
2kη W + W ≤ ηW 2 + C + k2 , (3.13)
η 4 R4
2h 2|h| 1 C h4
∇ h∇ η W ≤ |∇ h||∇ η| W ≤ η(1 − h) W 2 + . (3.14)
1−h 1−h 2 R4 (1 − h)3

By the Laplacian comparison theorem [45], r ≤ n−
r
1
+ n − 1k, η = ηrr |∇ r |2 + ηr r and then
   
∂ 1 1 1
−  ηW ≤ ηW 2 + C + + k 2
. (3.15)
∂t 4 R4 T2
Recall that ĝ (h) := g ( w )/ w and w −1 ∇ h∇ g ( w ) = g ( w )|∇ h|2 . Then, a direct computation shows that

2h 2w −1 ∇ h∇ g
ĝ (h) W η + η = 2η M w W , (3.16)
1−h (1 − h)2
where M w := 1− h
h
1
ĝ (h) + g ( w ) = 1− h
ĝ (h) + 
gh .
Using the bounds (3.13)–(3.15) and (3.16) in (3.12) and rearranging together with the facts that h ≤ 0, 1 − h ≥ 1 and
h4 /(1 − h)4 < 1 gives
 1 1 
ηW 2 ≤ C + + k 2 + 2η M w W .
R4 T2
As before using Young’s inequality we have
1
2η M w W ≤ η W 2 + η C ( sup {M+w })2 ,
2 Q R ,T
 +
where M+
w :=
1
1−h
ĝ (h) + 
gh . Hence
 1 1 
ηW 2 ≤ C + + k2 + C η( sup {M+
w })
2
R4 T2 Q R ,T

at (x0 , t 0 ). Since η ≡ 1 in Q R ,T /2 we arrive at



|∇ h| 1 1
≤c + + k + sup {M+
w },
1−h R2 T Q R ,T

by using the definition W = |∇ log(1 − h)|2 . The desired estimate then follows at once since T is arbitrary. 

11
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

Under a suitable assumption we can obtain the global version of the last theorem by letting R → +∞.

Corollary 3.2. Let M, w and c (n) be as defined in the above theorem. Then
 
|∇ w | w  1 
(x, t ) ≤ C (n) 1 − log w
√ + k+M (3.17)
w D t
 +
 w := sup
on M × [0, T ], where t = 0, M + + 1
ĝ (h) + 
M ×[0, T ] {M w } and M w := 1−h
gh .

The above corollary leads to the elliptic Harnack type inequality by integrating in space only.

Corollary 3.3. Let M be as defined in the Theorem 3.1. Suppose 0 < w ≤ D solves (3.1) for all (x, t ) ∈ M × [0, ∞). Then for x1 , x2 ∈ M,
t ∈ [0, T ]

w (x2 , t ) ≤ w (x1 , t )ξ(d(x1 ,x2 ),t ) e 1−ξ(d(x1 ,x2 ),t ) , (3.18)


    
where ξ(d(x1 , x2 ), t ) = exp − c (n) √1 + k + M  w d(x1 , x2 ) , t = 0 and c (n) is as in Theorem 3.1.
t

Proof. Let us consider the minimal geodesic joining x2 and x1 such that γ (s) : [0, 1] → M so that γ (0) = x2 and γ (1) = x1 .
Let h = log w, then
1
1 − h(x1 , t ) d log(1 − h(γ (s), t ))
log = ds
1 − h(x2 , t ) ds
0
1
γ (s)∇ h
= − ds
1 − h(γ (s), t ))
0
1
 1  
|∇ h|  w d(x1 , x2 ).
≤ |γ (s)| ds ≤ c √ + k + M
1 − h(γ (s), t )) t
0

The last inequality implies the desired estimate (3.18) and the proof is complete. 

Lastly we give the following examples as an application of the estimates in this section.

 w in (3.17) can be computed as


Example 3.4. Fisher-KPP equation. Here g ( w ) = aw (1 − w ), 0 < w ≤ 1, a > 0. M

 w := [(1/(1 − h)) ĝ (h) + 


M gh ]+ = a
since in this case, 1 − h ≥ 1 and
1 1
ĝ (h) + 
gh = a(1 − w ) − aw ≤ a(1 − 2w ) ≤ a.
1−h 1−h

 w in (3.17) can be computed as


Example 3.5. Allen-Cahn equation. Here g ( w ) = aw (1 − w 2 ), 0 < w ≤ 1, a > 0. M

 w := [(1/(1 − h)) ĝ (h) + 


M gh ]+ = a
since in this case, 1 − h ≥ 1 and
1 1
ĝ (h) + 
gh = a(1 − w 2 ) − 2aw 2 ≤ a(1 − 3w 2 ) ≤ a.
1−h 1−h

We now state the lemma which be useful in the next example on logarithmic nonlinearities.

Lemma 3.6. Let g (s) be a sufficiently smooth function for s > 0 and h = log s.

(1) If g (s) = asα | log s|β , where α and β = 1 are constants. Then
 
hg (s) α (1 − α )h
+ g (s) = + asα −1 |h|β + aβ sα −1 |h|β−2 h. (3.19)
(1 − h)s 1−h 1−h

12
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

(2) If g (s) = asα (log s)β , where α and β are constants. Then
 
hg (s) α (1 − α )h
+ g (s) = + asα −1 hβ + aβ sα −1 hβ−1 . (3.20)
(1 − h)s 1−h 1−h

Proof. By direct computation

g (s) = aα sα −1 | log s|β + aβ sα −1 | log s|β−2 log s


and

h/(1 − h) · g (s)/s = a(h/(1 − h))sα −1 | log s|β .


Putting these together proves (3.19). Whilst the proof of (3.20) is similar. 

Example 3.7. Logarithmic nonlinearities of the forms


(i) g ( w ) = aw α | log w |β , a ∈ R, α ≥ 1, β > 1
and
(ii) g ( w ) = aw α (log w )β , a ∈ R, α ≥ 1, β ≥ 1(integer).

For Case (i): g ( w ) = aw α | log w |β , α ≥ 1, β > 1. We compute


 w := [a]+ Lα ,β, w ,
M (3.21)
where

Lα ,β, w := ([dα ,h h̄]+ + β[h̄β−1 ]+ ) sup {e (α −1)h },


M ×[0, T ]

where h̄ = |h| h = log w, and dα ,h = α /(1 − h) + (1 − α )h/(1 − h) > 0. Note that we applied Lemma 3.6 ((3.19)) to arrive at
(3.21) in the sense that
1 h g(w)
ĝ (h) + 
gh = + g ( w ).
1−h 1−h w
For Case (ii): g ( w ) = aw α (log w )β , α ≥ 1, β ≥ 1. We compute
 w := [a]+ L∗α ,β, w ,
M (3.22)
where

L∗α ,β, w := ([dα ,h h]+ + β[hβ−1 ]+ ) sup {e (α −1)h },


M ×[0, T ]

where h = log w, and dα ,h = α /(1 − h) + (1 − α )h/(1 − h) > 0.


Finally, we state some Liouville type properties for the steady state of (3.1) with logarithmic nonlinearities of the forms
in Example 3.7. That is

 w + aw α | log w |β = 0, α ≥ 1, β > 1 (3.23)


and

 w + aw α | log w |β = 0, α ≥ 1, β ≥ 1. (3.24)

Theorem 3.8. Let M be an n-dimensional complete Riemannian manifold with Ric ( M ) ≥ 0.

(1) Suppose 0 < w ≤ D solves (3.23) on M. If a ≤ 0 then w is a constant, and necessarily, w ≡ 1 identically, if a < 0 in addition.
(2) Suppose 1 ≤ w ≤ D solves (3.24) on M. If a ≤ 0 then w is a constant, and necessarily, w ≡ 1 identically, if a < 0 in addition.

Proof. (1). Note that [a]+ = 0 since a ≤ 0. By (3.21) we have M w = 0. Using Corollary 3.2 with k = 0 since Ric ( M ) ≥ 0 and
t → +∞ and observe that constant c (n) > 0 is time independent. It then follows that ∇ w ≡ 0 meaning that w is a constant.
Necessarily w ≡ 1.
(2). The proof is as in (1) above except we use (3.22) and observe that [dα ,h (log w )β ]+ ≥ 0 and [(log w )β−1 ]+ ≥ 0 only
when w ≥ 1.
More explicitly, for β ≥ 2 (even integer) we know that for w ≥ 1, [dα ,h (log w )β ]+ ≥ 0 with [(log w )β−1 ]+ ≥ 0 and for
β ≥ 1 (odd integer) we have [(log w )β−1 ]+ ≥ 0 with [(log w )β ]+ ≥ 0. 

13
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

Remark 3.9. The exclusion of β = 1 from g ( w ) = aw α | log w |β is prompted by the breakdown of differentiability of modulus
function at the origin. Meanwhile, we see that similar estimates are obtained for g ( w ) = aw α (log w )β , β ≥ 1. Note that the
behaviour of the flow with aw α | log w |β , β > 1, coincides with that of the flow with aw α (log w )β for β ≥ 2 (even), but with
interesting differences for β ≥ 1 (odd), due to sign changing nature of logarithm. Physical applications are not unforeseeable,
though may involve isolated singularities. Singularity issues have been discussed in [13,27,29] for the steady state solutions.

CRediT authorship contribution statement

Abimbola Abolarinwa designed the model. All authors did the analysis, read and approved the final manuscript.

Acknowledgements

The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University for funding this
work through a research program under grant number R.G.P1/50/42. They are also grateful to the Department of Mathemat-
ics, University of Lagos, Nigeria.

References

[1] A. Abolarinwa, Elliptic gradient estimates and Liouville theorems for a weighted nonlinear parabolic equation, J. Math. Anal. Appl. 473 (2019) 297–312.
[2] A. Abolarinwa, Differential Harnack and logarithmic Sobolev inequalities along Ricci-harmonic map flow, Pac. J. Math. 278 (2) (2015) 257–290.
[3] A. Abolarinwa, Gradient estimates for a nonlinear parabolic equation with potential under geometric flow, Electron. J. Differ. Equ. 2015 (12) (2015)
1–11.
[4] A. Abolarinwa, Harnack estimates for heat equations with potentials on evolving manifolds, Mediterr. J. Math. 13 (5) (2016) 3185–3204.
[5] A. Abolarinwa, Differential Hanarck estimates for a nonlinear evolution equation of Allen-Cahn type, Mediterr. J. Math. 18 (2021) 200.
[6] A. Abolarinwa, A. Taheri, Elliptic gradient estimates for nonlinear f -heat equation on weighted manifolds with time dependent metrics and potentials,
Chaos Solitons Fractals 142 (2021) 110329.
[7] L. Alzaleq, V. Manoranjan, Analysis of Fisher-KPP with a time dependent Allee effect, IOP SciNotes 1 (2020) 025003.
[8] S.M. Allen, J.W. Cahn, A microscopic theory for antiphase boundary motion and its application to antiphase domain coarsening, Acta Metall. 27 (1979)
1085–1095.
[9] L. Ambrosio, X. Cabré, Entire solutions of semilinear elliptic equations in R 3 and a conjecture of De Giorgi, J. Am. Math. Soc. 13 (4) (2000) 725–739.
[10] M. Bǎileşteanu, X. Cao, A. Pulemotov, Gradient estimates for the heat equation under the Ricci flow, J. Funct. Anal. 258 (2010) 3517–3542.
[11] M. Bǎileşteanu, A Harnack inequality for the parabolic Allen-Cahn equation, Ann. Glob. Anal. Geom. 51 (2017) 367–378.
[12] L. Caffarelli, F. Lin, Nonlocal heat flows preserving the L 2 energy, Discrete Contin. Dyn. Syst. 23 (1–2) (2009) 49–64.
[13] L. Caffarelli, B. Gidas, J. Spruck, Asymptotic symmetry and local behavior of semilinear elliptic equations with critical Sobolev growth, Commun. Pure
Appl. Math. 42 (1989) 271–297.
[14] L. Calatroni, P. Colli, Global solution to the Allen-Cahn equation with singular potentials and dynamics boundary conditions, Nonlinear Anal. 79 (2013)
12–27.
[15] X. Cao, B. Liu, I. Pendleton, A. Ward, Differential Harnack estimates for Fisher’s equation, Pac. J. Math. 290 (2) (2017) 273–300.
[16] D. Castorina, C. Mantegazza, Ancient solutions of semilinear heat equations on Riemannian manifolds, Atti Accad. Naz. Lincei, Rend. Lincei, Mat. Appl.
28 (1) (2017) 85–101.
[17] D. Castorina, C. Mantegazza, Ancient solutions of superlinear heat equations on Riemannian manifolds, Commun. Contemp. Math. (2020).
[18] X. Chen, Generation and propagation of interface for reaction-diffusion equations, J. Differ. Equ. 96 (1) (1992) 116–141.
[19] Q. Chen, G. Zhao, Li-Yau type and Souplet-Zhang type gradient estimates of a parabolic equation for the V -Laplacian, J. Math. Anal. Appl. (2018).
[20] E. De Giorgi, Convergence problems for functionals and operators, in: Proceedings of the International Meeting on Recent Methods in Nonlinear
Analysis, Rome, 1978, Pitagora, Bologna, 1979, pp. 131–188.
[21] M. Del Pino, M. Kowalczyk, F. Pacard, J. Wei, Multiple solutions to the Allen-Cahn equation in R2 , J. Funct. Anal. 258 (2010) 458–503.
[22] N.T. Dung, N.N. Khanh, Gradient estimates for a class of semilinear parabolic equations and their applications, Vietnam J. Math. (2021).
[23] L.C. Evans, H.M. Soner, P.E. Souganidis, Phase transitions and generalized motion by mean curvature, Commun. Pure Appl. Math. 45 (9) (1992)
1097–1123.
[24] R.A. Fisher, The wave of advance of advantageous genes, Annu. Eugen. 7 (1937) 355–369.
[25] R. Fitzhugh, Impulses and physiological states in theoretical models of nerve membrane, Biophys. J. 1445 (1961).
[26] X. Geng, S. Hou, Gradient estimates for the Fisher-KPP equation on Riemannian manifolds, Bound. Value Probl. 2018 (2018) 25.
[27] M. Ghergu, S. Kim, H. Shahgholian, Exact behaviour around isolated singularity for semilinear elliptic equations with log-type nonlinearity, arXiv:
1806.04287, 2019, Adv. Nonlinear Anal. 8 (2019) 995–1003, https://doi.org/10.1515/anona-2017-0261.
[28] N. Ghoussoub, C. Gui, On a conjecture of De Giorgi and some related problems, Math. Ann. 311 (3) (1998) 481–491.
[29] B. Gidas, J. Spruck, Global and local behavior of positive solutions of nonlinear elliptic equations, Commun. Pure Appl. Math. 34 (1981) 525–598.
[30] R. Hamilton, A matrix Harnack estimate for the heat equation, Commun. Anal. Geom. 1 (1993) 113–126.
[31] S. Hou, Gradient estimates for the Allen-Cahn equation on Riemannian manifolds, Proc. Am. Math. Soc. 147 (2019) 619–628.
[32] T. Ilmanen, Convergence of the Allen-Cahn equation to Brakke’s motion by mean curvature, J. Differ. Geom. 38 (2) (1993) 417–461.
[33] A.N. Kolmogorov, I.G. Petrovskii, N.S. Piskunov, A study of the diffusion equation with increase in the amount of substance, and its application to a
biological problem, Bull. Moscow Univ. Math. Mech. 1 (1937) 1–26.
[34] S.A. Levin, H.C. Muller-Landau, R. Nathan, J. Chave, The ecology and evolution of seed dispersal: a theoretical perspective, Annu. Rev. Ecol. Syst. 34
(2003) 575–604.
[35] J. Li, X. Xu, Differential Harnack inequalities on Riemannian manifolds I: linear heat equation, Adv. Math. 226 (5) (2011) 4456–4491.
[36] G.S. Ludford, Combustion and Chemical Reactions, Amer. Math. Soc., Providence, R.I., 1986.
[37] P. Li, S-T. Yau, On the parabolic kernel of the Schrödinger operator, Acta Math. 156 (1986) 153–201.
[38] L. Ma, L. Cheng, Non-local heat flows and gradient estimates on closed manifolds, J. Evol. Equ. 9 (2009) 85–101.
[39] B. Ma, F. Zeng, Hamilton-Souplet-Zhang’s gradient estimates and Liouville theorems for a nonlinear parabolic equation, C. R. Math. Acad. Sci. Paris, Ser.
I 356 (5) (2018) 550–557.
[40] P. Mastrolia, M. Rigoli, A.G. Setti, Yamabe-Type Equations on Complete, Noncompact Manifolds, Progress in Mathematics, vol. 302, Birkhäuser Verlag,
Basel, 2012.

14
A. Abolarinwa, J.O. Ehigie and A.H. Alkhaldi Journal of Geometry and Physics 170 (2021) 104382

[41] G. Perelman, The entropy formula for the Ricci flow and its geometric application, arXiv:math.DG/0211159v1, 2002.
[42] P. Poláčik, P. Quittner, P. Souplet, Singularity and decay estimates in superlinear problems via Liouville-type theorems. I. Elliptic equations and systems,
Duke Math. J. 139 (3) (2007) 555–579.
[43] P. Poláčik, P. Quittner, P. Souplet, Singularity and decay estimates in superlinear problems via Liouville-type theorems. II. Parabolic equations, Indiana
Univ. Math. J. 56 (2) (2007) 879–908.
[44] A.D. Polyanin, V.F. Zaitsev, Handbook of Nonlinear Partial Differential Equations, second edition, CRC Press, Boca Raton, FL, 2012.
[45] R. Schoen, S.-T. Yau, Lectures on Differential Geometry, International Press, Cambridge, MA, 1994.
[46] P. Souplet, Q.S. Zhang, Sharp gradient estimate and Yau’s Liouville theorem for the heat equation on noncompact manifolds, Bull. Lond. Math. Soc.
38 (6) (2006) 1045–1053.
[47] A. Tremel, A. Cai, N. Tirtaatmadja, B.D. Hughes, G.W. Stevens, K.A. Landman, A.J. O’Connor, Cell migration and proliferation during monolayer formation
and wound healing, Chem. Eng. Sci. 64 (2003) 247–253.
[48] L. Wang, L 2 -preserving Schrodinger heat flow under the Ricci flow, Balk. J. Geom. Appl. 15 (2) (2010) 121–133.
[49] J.Y. Wu, Elliptic gradient estimates for a weighted heat equation and applications, Math. Z. 280 (2015) 451–468.
[50] Q. Zhang, Some gradient estimates for the heat equation on domains and for an equation by Perelman, Int. Math. Res. Not. 2006 (1) (2006) 92314.
[51] K.G. Zloshchastiev, Logarithmic nonlinearity in the theories of quantum gravity: origin of time and observational consequences, Gravit. Cosmol. 16
(2010) 288–297.

15

You might also like