A sufficient condition for the Liouville property of solutions to the Euler-Lagrange equations of variational integrals with nonstandard growth in terms of the ellipticity modulus near the origin
We consider entire solutions u: Rn→R of the Euler-Lagrange equation div(∇f(∇u))=0 with convex density f: Rn→R of nonstandard growth and nonuniform ellipticity. It turns out that the Liouville property can be characterized through the behavior of D2f(p) for points p∈Rn close to the origin: if, for example, D2f(p) behaves as ∣p∣ρ−21, 1=unit matrix, with exponent ρ≥2 and for ∣p∣≤1 then the validity of n≤ρ+2 together with the boundedness of u implies its constancy.
Introduction
In this note we discuss the Liouville property for entire solutions u∈C2(Rn) of the nonlinear elliptic equation
div(∇f(∇u))=0(1)
arising as the Euler-Lagrange equation of the variational integral ∫f(∇u)dx with convex density f:Rn→R of class C2. Here we mean by Liouville property the fact that any bounded entire solution u:Rn→R of (1.1) must be a constant function.
Surprisingly, it turns out that a positive answer just depends on the behavior of D2f(p) for p close to the origin, for example, if D2f(p) is of order ∣p∣ρ−2 as p→0 with exponent ρ≥2, then we have the Liouville property provided that n≤ρ+2, and this is true under rather general hypotheses concerning the growth of f at infinity.
Before going into details we like to remark that the study of equations with nonstandard growth and nonuniform ellipticity is a very active field, which has attracted the research of many scientists over the last decades focussing on the questions of existence and regularity of solutions. For a detailed overview including historical comments the reader is referred to [1]. However, we could not find a discussion of the Liouville property in this general setting, but it should be mentioned that there are contributions concerning quasilinear or uniformly elliptic equations, which can be found for instance in [2], [3], [4], [5] and in the references quoted therein. We also advise the reader to consult the textbooks [6], [7].
Let us now state our assumptions concerning the density f∈C2(Rn) occurring in equation (1.1): we have
D2f(p)(q,q)>0,p,q∈Rn,q=0,∇f(0)=0,(2)
implying the strict convexity of f. We further require (c,c1,… denoting positive constants)
∣∇f(p)∣≤c∣p∣m−1for p∈Rn,∣p∣≥1(3)
with an arbitrary exponent m>1, and addition we assume
p⋅∇f(p)≥c1∣p∣−μ∣p∣∣∇f(p)∣,p∈Rn,∣p∣≥1(4)
for some number μ∈[0,1) for which we also have
∣D2f(p)∣ln2(1+∣p∣)≤c2∣p∣−μ∣∇f(p)∣,p∈Rn,∣p∣≥1.(5)
The conditions (1.3)–(1.5) are conditions at infinity, since the bound ∣p∣≥1 for p∈Rn can be replaced by ∣p∣≥M for some arbitrary number M. Inequality (1.3) excludes exponential growth, but the exponent m may be as large as we want. Condition (1.4) states that the radial part of ∇f(p), i.e. the quantity ∇f(p)⋅∣p∣p, is not allowed to degenerate at infinity in the sense that we have ∣p∣−μ∣∇f(p)∣ as a lower bound for some exponent μ∈[0,1). Finally, the growth of ∣D2f(p)∣ as p→∞ is limited in terms of ∣∇f(p)∣ equipped with the weight function ∣p∣−μ, which is made precise in (1.5). We refer to § 2 for a list of prominent examples.
Theorem 1.1. Let (1.2)–(1.6) hold and let u∈C2(Rn) denote an entire solution of equation (1.1) for which
x→∞lim∣x∣β∣u(x)∣<∞(7)
with
β:=β(ρ):=ρ1(ρ+2−n).(8)
Then u is an affine function, i.e. ∂i∂ju=0 on Rn for $i,j=1,\ldots,n.
If n=2, then β=1 and in conclusion:
Corollary 1.2. Under the assumptions of Theorem 1.1 consider the case n=2 and let u∈C2(R2) denote an entire solution of (1.1) such that
x→∞lim∣x∣∣u(x)∣<∞.
Then u is an affine function.
Note that the number β=β(ρ) from (8) is ≥0 if and only if n≤ρ+2. This implies the Liouville property for dimensions n≤n(ρ) with upper bound n(ρ):=2+ρ determined just through the behavior of D2f(p) for p close to the origin as expressed in (6).
Corollary 1.3 (Liouville property). Let (2)–(6) hold and assume that n≤ρ+2. Then the constant functions are the only bounded entire solutions of equation (1).
In fact, if u∈L∞(Rn) and n≤ρ+2, then (7) clearly holds, hence u is an affine function and thereby constant.
Remark 1.4. It is possible to replace (6) by the anisotropic condition
with exponents 2≤ρ≤t. Then Theorem 1.1 continues to hold, provided we define in (8)
β:=β(t,ρ):=T1(2+T−n),T:=t−ρ+1t
as the appropriate value of β, and if we require ρ+2≥t. Some comments will be given in § 4.
Remark 1.5. One might wonder about the occurrence of the quantity ln2(1+∣p∣) on the left-hand side of inequality (5). Of course we can replace (5) by the stronger condition (keeping (4)) that
∣D2f(p)∣≤c2∣p∣−(μ+ε)∣∇f(p)∣,p∈Rn,∣p∣≥1
holds with exponent μ from (4) and for some arbitrarily small number ε>0. However, we prefer to work with a less restrictive assumption, and at the same time it turns out that the expression ∣D2f(∇u)∣ln2(1+∣∇u∣) occurs in a limit version of a Caccioppoli type inequality, which will be presented in Lemma 3.1 being valid for solutions u of equation (1) under very general hypotheses concerning the density f.
Our paper is organized as follows: in § 2 we collect some examples to which Theorem 1.1 and the corollaries apply. § 3 is devoted to an inequality of Caccioppoli type, which is crucial for the proof of Theorem 1.1 presented in § 4. The Caccioppoli inequality Lemma 3.1 will explain the occurrence of the factor ln2(1+∣p∣) on the left-hand side of (5) and during the proof of Lemma 3.1 the comments from Remark 1.5 will become more transparent.
Examples
We present some examples of densities f:Rn→R to which Theorem 1.1 can be applied.
Example 2.1 (anisotropic power growth). Let f∈C2(Rn) satisfy (2) and (6). In addition we require: there exist constants c1, c2 and exponents 1<τ≤r such that
c1∣p∣τ−2∣q∣2≤D2f(p)(q,q)≤c2∣p∣r−2∣q∣2(9)
holds for p,q∈Rn, ∣p∣≥1. Then we claim: if
r<τ+21,(10)
then we have (3), (4), (5) for some μ∈[0,1) and with arbitrary m≥r.
The validity of (3) for m≥r is immediate by the second inequality in (9). For (4) and (5) we choose p∈Rn, ∣p∣≥1, and observe that by (9)
In order to find a common value μ for μ1 and μ2 we need to know that r−τ<τ+1−r, which is just a consequence of (10) completing the discussion of (4) and (5).
Example 2.2 (densities f with linear growth). Again we consider f∈C2(Rn) with (2) and (6). We further assume that
a(∣p∣−1)≤f(p)≤A(∣p∣+1)p∈Rn(12)
holds with constants a,A>0. Moreover, we require
D2f(p)(q,q)≤Θ(∣p∣)∣q∣2p,q∈Rn,∣p∣≥1,(13)
for a function Θ:[0,∞)→[0,∞) such that
t→∞limsupΘ(t)ln2(t)<∞.(14)
Then we are in the situation of Theorem 1.1, i.e. the conditions (3)–(6) hold: for (3) we just observe that from the convexity of f (see (2)) together with (12) we immediately get the boundedness of ∇f.
Let
ϑ(r)=∣p∣=r∣q∣=1infD2f(p)(q,q),ϑ:[0,∞)→[0,∞).
Again from (2) we deduce that ϑ(r)>0 for r>0. Observing this, it follows for ∣p∣≥1:
∣p∣∣∇f(p)∣≥p⋅∇f(p)=∫01D2f(tp)(p,p)dt
≥∣p∣2∫01ϑ(t∣p∣)dt=∣p∣∫0∣p∣ϑ(s)ds
≥∣p∣∫01ϑ(s)ds=:α∣p∣
for some positive constant α, and the boundedness of ∇f clearly implies (4) with the choice μ=0. Condition (5) (with μ=0) is a consequence of (13) and (14).
Example 2.3 (densities with radial structure). We let f(p):=g(∣p∣), p∈Rn for a function g∈C2([0,∞)) such that
hence (1.5) holds for the choice μ=0. From p⋅∇f(p)=∣p∣∣∇f(p)∣ we obtain (1.4).
A Caccioppoli type inequality
A preliminary version of the following lemma can be found in Proposition 6.1 of [8]. Here we extend Proposition 6.1 in some sense to the limit case α=−1/2.
Lemma 3.1. Let f∈C2(Rn) denote a convex function, which means that D2f(p) is just positive semidefinite for any p∈Rn.
discussing the terms (I) and (II) separately: it holds
∣integrand of (II)∣=2∣∂iu∣∣η∣Φ(Γi)∣D2f(∇u)(∇∂iu,∇η)∣=2(Γi−1)1/2Γi−1/2ln(Γi)∣η∣∣D2f(∇u)(∇∂iu,∇η)∣=2D2f(∇u)(∣η∣Γi−1/4∇∂iu,Γi−1/4(Γi−1)1/2ln(Γi)∇η)≤2(D2f(∇u)(∇∂iu,∇∂iu)Γi−1/2η2)1/2
×(D2f(∇u)(∇η,∇η)Γi−1/2(Γi−1)ln2(Γi))1/2
where in the last step the Cauchy-Schwarz inequality has been used for the nonnegative symmetric bilinear form D2f(∇u). Thus, after using Hölder’s inequality, we find (recall (3.1))
∣(II)∣≤(∫spt(∇η)∩[Γi≥2]D2f(∇u)(∇∂iu,∇∂iu)Γi−1/2η2dx)1/2×(∫spt(∇η)∩[Γi≥2]D2f(∇u)(∇η,∇η)Γi−1/2(Γi−1)ln2(Γi)dx)1/2≤(r.h.s. of (3.2))(23)
In a similar way we proceed with (I): we have
∣integrand of (I)∣=2∣η∂iuΦ(Γi)∣∣D2f(∇u)(∇∂iu,∇η)∣≤c∣η(Γi−1)1/2ln(Γi)∣∣D2f(∇u)(∇∂iu,∇η)∣≤c(D2f(∇u)(∇∂iu,∇∂iu)η2(Γi−1))1/2×(D2f(∇u)(∇η,∇η)ln2(Γi))1/2
and using Hölder’s inequality again (recall (3.1))
∣(I)∣≤c(∫spt(∇η)∩[Γi<2]D2f(∇u)(∇∂iu,∇∂iu)η2(Γi−1)dx)1/2×(∫spt(∇η)∩[Γi<2]D2f(∇u)(∇η,∇η)ln2(Γi)dx)1/2≤(r.h.s. of (3.2))(24)
We return to (3.5) noting that, on account of (3.6) it holds (recall (3.1))
(l.h.s. of (3.5))≥c∫ΩD2f(∇u)(∇∂iu,∇∂iu)η2λ(Γi)dx.(25)
Thus, combining (3.5) with (3.7)–(3.9), inequality (3.2) follows, which completes the proof of Lemma 3.1.
Proof of Theorem 1.1
We want to prove Theorem 1.1 in the setting of Remark 1.4, which means that we assume the conditions (1.2)–(1.5) for the density f but replace (1.6) by (p,q∈Rn,∣p∣≤1)
c∣p∣t−2∣q∣2≤D2f(p)(q,q)≤C∣p∣ρ−2∣q∣2(26)
for exponents 2≤ρ≤t, t≤ρ+2 and constants c,C>0.
Our arguments proceed as follows: let i∈{1,…,n} and consider η∈C01(Rn) such that spt(η)⊂B2R, 0≤η≤1, η=1 on BR and ∣∇η∣≤c/R, where Br:={x∈Rn:∣x∣<r} for radii r>0. We will show that we can find a constant c∗∈(0,∞) for which
which goes to zero as R→∞ on account of (27) and (28).
Thus we have limR→∞Θi(R)=0 and in conclusion
D2f(∇u)(∇∂iu,∇∂iu)λ(Γi)=0on Rn(29)
for i∈{1,…,n}. The definition of λ (see (17)) shows that λ(Γi(x))=0 if and only if ∂iu(x)=0, and by Lemma 7.7, p. 152, in [6] it holds ∇∂iu=0 almost everywhere on [∂iu=0]. However, on the set [∂iu=0], (29) implies ∇∂iu=0, hence ∇∂iu=0 almost everywhere and by continuity everywhere on Rn, being true for i=1,…,n, completing the proof of Theorem 1.1.
Let us discuss our claim (27). On the set [∣∇u∣≥1] it holds by the definition of Λ given in (17) and on account of (5)
For proving (34) we select η∈C01(B25R−B2R), 0≤η≤1 such that η=1 on B2R−BR, ∣∇η∣≤c/R and let l∈N. Abbreviating TR:=B25R−B2R we quote (33) and use equation (1) to get
∣integrand of (I)∣≤∣∇f(∇u)∣η2l−1∣∇η∣∣u∣=∣∇f(∇u)∣1−ϵ∣∇f(∇u)∣ϵη2l−1∣∇η∣∣u∣≤c∣∇f(∇u)∣1−ϵ∣∇u∣ϵ(m−1)η2l−1∣∇η∣∣u∣≤δ∣∇f(∇u)∣∣∇u∣1−ϵ(m−1)ϵη1−ϵ2l−1+c(δ)∣∇η∣1/ϵ∣u∣1/ϵ,
and if we fix ϵ>0 such that 1−ϵϵ(m−1)≤1−μ (recall 0≤μ<1), we obtain after increasing l (recall (32)):
∣integrand of (I)∣≤δ∣∇f(∇u)∣∣∇u∣1−μη2l+c(δ)∣∇η∣1/ϵ∣u∣1/ϵ≤δM(∇u)η2l+c(δ)∣∇η∣1/ϵ∣u∣1/ϵ.
Returning to (35) and choosing δ sufficiently small, it is shown
∫TRη2lM(∇u)dx≤c{∣(II)∣+∫TR∣∇η∣1/ϵ∣u∣1/ϵdx}.
The remaining term (II) is handled as follows (see (26) and (32)):
∣integrand of (II)∣≤∣∇f(∇u)∣η2l−1∣u∣∣∇η∣≤c∣∇u∣ρ−1η2l−1∣∇η∣∣u∣≤δ∣∇u∣tη(2l−1)ρ−1t+c(δ)∣u∣T∣∇η∣T≤δM(∇u)η(2l−1)ρ−1t+c(δ)∣u∣T∣∇η∣T.(36)
with exponent T:=t/(t−ρ+1) as in Remark 1.4.
Finally, for l sufficiently large, we can arrange that (2l−1)t/(ρ−1)≥2l, and after appropriate choice of δ we deduce from (36)
We recall that ϵ>0 just has to satisfy 1−ϵϵ(m−1)≤1−μ. Now we require in addition that ϵ≤1/T holds and use (1.7) on the right-hand side of (37) with the choice of β=β(t,ρ) as defined in Remark 1.4. This gives (recall the choice of η) for R≫1:
r.h.s.of(4.12)≤cRn[RβT−T+Rϵβ−ϵ1]≤cRn+βT−T=cR2,
and (4.9) is a consequence of (37). This completes the proof of Theorem 1.1 and of Remark 1.4, respectively.
Revisiting Example 2.1
We finish by remarking that the arguments presented during the proof of Theorem 1.1 are rather general, which means that they can easily be adjusted to particular cases, for instance it is possible to weaken condition (2.2) required in Example 2.1. We can state the following extension:
Theorem 4.1. Let the density f∈C2(Rn) satisfy (1.2), (1.6) and (2.1), where in (2.1) we choose exponents 1<τ≤r such that
r<τ+1(38)
holds.
Then, if u∈C2(Rn) is an entire solution of (1.1) with the property (1.7), β defined according to (1.8), it follows that u is an affine function. Moreover, Corollary 1.2 and Corollary 1.3 extend to this situation.
Proof. We present the necessary adjustments. As before, it suffices to prove inequality (27). On the set [∣∇u∣≥1] we have by (9) (recall also (17))
From (41) in turn it follows that our claim will be a consequence of (compare (34))
R−2∫B2R−BRM(∇u)dx≤c∗(42)
again with a finite constant c∗ independent of R. For proving (42) we pass to inequality (35) discussing the quantity (II) exactly as before (recalling that now, for notational simplicity, we just assume that ρ=t). In addition we have for the integrand in (I) (compare (9) and (38))
Assume now that we have (1.7) with β(≤1) from (1.8). Observe that in (2.1) we can increase r such that r is very close to τ+1. Assuming this from the beginning, it follows that aβ−a≥bβ−b and (4.18) gives
∫TRη2lM(∇u)dx≤cRn+ρ(β−1)=cR2,
hence we have (4.17) and the proof is complete.
Michael Bildhauer Saarland University Department of Mathematics P.O. Box 15 11 50 66041 Saarbrücken, Germany [email protected]
Martin Fuchs Saarland University Department of Mathematics P.O. Box 15 11 50 66041 Saarbrücken, Germany [email protected]
References
[1]G. Mingione, V. Rădulescu: Recent developments in problems with nonstandard growth and nonuniform ellipticity. J. Math. Anal. Appl. 501 (2021), no. 1, Paper No. 125197, 41 pp.DOI
[3]A. Farina: Liouville-type theorems for elliptic problems. Handbook of differential equations: stationary partial differential equations. Vol. IV, 61–116. Elsevier/North-Holland, Amsterdam, 2007.DOI
[4]T. Adamowicz, P. Górka: The Liouville theorems for elliptic equations with nonstandard growth. Commun. Pure Appl. Anal. 14 (2015), no. 6, 2377–2392.arxiv.org/abs/1408.6359
[5]S. Dudek: The Liouville-type theorem for problems with nonstandard growth derived by Caccioppoli-type estimate. Monatsh. Math. 192 (2020), no. 1, 75–91.DOI
[6]D. Gilbarg, N. S. Trudinger: Elliptic partial differential equations of second order. Reprint of the 1998 edition. Classics Math. Springer-Verlag, Berlin, 2001. xiv+517 pp.
[7]O. A. Ladyzhenskaya, N. N. Ural’tseva: Linear and quasilinear elliptic equations. Academic Press, New York-London, 1968. xviii+495 pp.
[8]M. Bildhauer, M. Fuchs: On the global regularity for minimizers of variational integrals: splitting-type problems in 2D and extensions to the general anisotropic setting. J. Elliptic Parabol. Equ. 8 (2022), no. 2, 853–884.