Linear Liouville Argument

• Linear Liouville argument is a rigidity technique showing that bounded solutions or subsolutions for certain linear operators must be constant or vanish.
• It connects key properties like the Khas’minskii condition, weak maximum principles at infinity, and stochastic completeness in diverse operator settings.
• Variants of the argument employ comparison principles, heat-kernel smoothing, global Harnack inequalities, and probabilistic methods to establish rigidity across elliptic, parabolic, and nonlocal frameworks.

Searching arXiv for recent and foundational papers on linear Liouville arguments and related equivalence frameworks. arXiv search query: "Linear Liouville theorem Khasminskii weak maximum principle stochastic completeness" Across the papers considered here, the expression linear Liouville argument can be understood as a family of rigidity arguments showing that bounded solutions, subsolutions, or invariant functions for linear operators are forced to be constant, and in some settings forced to vanish. In the Riemannian framework of operators modeled after the $p$-Laplacian with potential, the linear specialization $p=2$ yields operators of the form

$L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$

including the Laplace–Beltrami operator and Schrödinger-type operators, and the central structural fact is the equivalence between Liouville, the Khas’minskii property, and, in the Laplace–Beltrami case, stochastic completeness and the weak maximum principle at infinity (Mari et al., 2011). Other linear Liouville arguments retain the same rigidity goal but realize it through different mechanisms: heat-kernel smoothing for $-\Delta+c\cdot\nabla$ (Wang et al., 2018), global Harnack inequalities for Ornstein–Uhlenbeck and Kolmogorov operators (Kogoj et al., 2020), martingale stopping for affine rescaling equations (Bogachev et al., 2014), entropy and diffusive scaling in random environments (Fehrman, 2014), and subgroup-periodicity characterizations for Lévy and Courrège operators (Alibaud et al., 2018, Alibaud et al., 2019).

1. Linear operator classes and the Liouville property

In the operator class studied in "On the equivalence of stochastic completeness, Liouville and Khas'minskii condition in linear and nonlinear setting" (Mari et al., 2011), the general quasilinear divergence-form operator is

$L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$

with $A$ a Carathéodory bundle map satisfying coercivity, growth, and strict monotonicity, and $B(x,t)$ a Carathéodory potential satisfying monotonicity and sign conditions. The classical $p$-Laplacian with potential is recovered by

$$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$

so that

$L_F u = \Delta_p u - \lambda |u|^{p-2}u.$

In the linear case $p=2$0, one has $p=2$1 and

$p=2$2

and more generally

$p=2$3

with $p=2$4 symmetric uniformly elliptic. The sign convention in that paper subtracts the potential term.

The Liouville property is formulated in several equivalent linear forms. On a Riemannian manifold $p=2$5, the $p=2$6–Liouville property for $p=2$7, $p=2$8, states that the only bounded, nonnegative, continuous weak solution of

$p=2$9

is $L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$0. The parabolic version $L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$1 states that every bounded, nonnegative continuous weak solution of

$L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$2

is constant. In the Schrödinger setting

$L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$3

the Liouville property asserts that every bounded, nonnegative weak subsolution of $L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$4 is constant, and in the type 1 situation described in Section 5 of (Mari et al., 2011), zero.

Related linear formulations appear in other settings. For bounded continuous generalized harmonic functions on $L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$5, the equation

$L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$6

with constant drift $L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$7 has only constant bounded entire solutions (Wang et al., 2018). For translation-invariant operators satisfying the maximum principle in the sense of Courrège, bounded distributional solutions of $L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$8 are characterized by periodicity with respect to a subgroup determined by the operator coefficients, and Liouville becomes a statement about whether that subgroup fills $L u=\operatorname{div}(A(x)\nabla u)-V(x)u,$9 (Alibaud et al., 2019).

2. Equivalence schemes: Liouville, Khas’minskii, maximum principles, and stochastic completeness

A central feature of the linear Liouville argument is that the rigidity conclusion is usually not isolated. In the Riemannian and divergence-form settings, it sits inside an equivalence chain linking Liouville to barrier existence, maximum principles at infinity, and probabilistic non-explosion (Mari et al., 2011).

The Khas’minskii property for $-\Delta+c\cdot\nabla$0 is formulated by requiring, for every $-\Delta+c\cdot\nabla$1 with Lipschitz boundary and every $-\Delta+c\cdot\nabla$2, an exhaustion $-\Delta+c\cdot\nabla$3 such that

$-\Delta+c\cdot\nabla$4

For homogeneous operators such as $-\Delta+c\cdot\nabla$5, this simplifies to the existence of a nonnegative exhaustion $-\Delta+c\cdot\nabla$6 with $-\Delta+c\cdot\nabla$7 on $-\Delta+c\cdot\nabla$8 and $-\Delta+c\cdot\nabla$9 on $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$0.

The weak maximum principle at infinity is likewise operator-dependent. For the Laplace–Beltrami operator, it reads: for every $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$1 with $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$2, and every $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$3,

$L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$4

In the general framework with $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$5, $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$6, $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$7 a.e., and $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$8 satisfying the structure conditions, (Mari et al., 2011) defines the weak maximum principle $L_F u = \operatorname{div}(A(\nabla u)) - B(x,u),$9 for $A$0 and its parabolic version $A$1.

The principal equivalences may be summarized as follows.

Setting	Characterization	Source
$A$2	stochastic completeness $A$3 weak maximum principle at infinity $A$4 $A$5–Liouville for $A$6 $A$7 Khas’minskii’s condition	(Mari et al., 2011)
$A$8	$A$9	(Mari et al., 2011)
General $B(x,t)$0 under assumptions $B(x,t)$1	$B(x,t)$2 for $B(x,t)$3, $B(x,t)$4 for $B(x,t)$5, and $B(x,t)$6 are equivalent	(Mari et al., 2011)
Type 1/type 2 potentials $B(x,t)$7	$B(x,t)$8 or $B(x,t)$9 is equivalent to $p$0 and $p$1	(Mari et al., 2011)

In the pure Laplace–Beltrami case, stochastic completeness is the non-explosion of the associated diffusion, equivalently the conservativeness of the heat semigroup $p$2, and in heat-kernel terms

$p$3

Pigola–Rigoli–Setti are cited in (Mari et al., 2011) as identifying the weak maximum principle at infinity with stochastic completeness for $p$4.

This equivalence architecture is not universal, but it recurs in modified form. For general translation-invariant Courrège operators, bounded distributional solutions are exactly the a.e. $p$5-periodic functions, and Liouville holds if and only if

$p$6

so periodicity replaces Khas’minskii barriers as the decisive structural criterion (Alibaud et al., 2019).

3. The canonical comparison argument in the linear divergence-form case

The most explicit linear Liouville argument in (Mari et al., 2011) is the implication $p$7 for the linear operator

$p$8

where $p$9 is symmetric uniformly elliptic and $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$0. Assume the Khas’minskii property: there exists a compact $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$1 and an exhaustion $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$2 with

$$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$3

The goal is to show that every bounded nonnegative $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$4 satisfying $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$5 is constant, and in the type 1 case identically zero (Mari et al., 2011).

The proof proceeds by contradiction. First, the strong maximum principle, via Harnack’s inequality for supersolutions, gives $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$6 on $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$7 if $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$8 is nonconstant, where $$A(X)=|X|^{p-2}X,\qquad B(x,t)=\lambda|t|^{p-2}t,\quad \lambda\ge0,$$9. Then choose $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$0 close to $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$1 so that $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$2, with

$L_F u = \Delta_p u - \lambda |u|^{p-2}u.$3

Pick $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$4 with $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$5, choose an open $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$6, and set $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$7. The Khas’minskii property supplies a potential $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$8 such that $L_F u = \Delta_p u - \lambda |u|^{p-2}u.$9 on $p=2$00 and $p=2$01 on $p=2$02.

One then considers

$p=2$03

and lets $p=2$04 be the connected component of $p=2$05 containing $p=2$06. Because $p=2$07 is bounded and $p=2$08 is an exhaustion, $p=2$09, and on $p=2$10 one has $p=2$11. Inside $p=2$12,

$p=2$13

The weak comparison principle, based on the strict monotonicity of $p=2$14 and the monotonicity of $p=2$15, yields $p=2$16 on $p=2$17, contradicting the definition of $p=2$18. Hence a bounded nonnegative subsolution cannot be nonconstant. Proposition 5.1 in (Mari et al., 2011) then shows that for type 1 potentials the bounded nonnegative solution must be $p=2$19.

This argument depends on the technical toolkit assembled in Section 3 of (Mari et al., 2011): the weak comparison principle, obstacle problem solutions and their minimal supersolution characterization, Harnack inequality and strong maximum principle, and the pasting lemma for gluing supersolutions. A common misconception is that a “linear” Liouville argument in this sense must use only linear-algebraic manipulations; in this framework, the linear specialization still relies on obstacle problems, comparison, and maximum-principle technology.

4. Parabolic and semigroup realizations

A different linear Liouville argument replaces comparison with semigroup invariance and heat-kernel estimates. For bounded continuous generalized harmonic functions on $p=2$20 satisfying

$p=2$21

distributionally, (Wang et al., 2018) introduces the drifted heat kernel

$p=2$22

and the heat-flow average

$p=2$23

Because $p=2$24 solves

$p=2$25

and the distributional equation extends from $p=2$26 to Schwartz test functions, one obtains $p=2$27. Thus $p=2$28 for all $p=2$29, so $p=2$30 is smooth. Differentiating under the integral and using the explicit derivative of $p=2$31, the paper proves

$p=2$32

with

$p=2$33

Letting $p=2$34 forces $p=2$35, hence $p=2$36 and $p=2$37 is constant (Wang et al., 2018). In this setting, the Liouville conclusion is obtained from time-invariance of the heat flow plus quantitative decay of kernel derivatives.

For Ornstein–Uhlenbeck operators, (Kogoj et al., 2020) derives Liouville from a global Harnack inequality in space-time. The constant-coefficient operator is

$p=2$38

with the structural hypothesis

$p=2$39

The associated Kolmogorov operator is

$p=2$40

The paper proves a global Harnack inequality on left-translated backward paraboloids $p=2$41:

$p=2$42

for every nonnegative smooth solution of $p=2$43. A geometric lemma shows that for every $p=2$44, sufficiently negative times place $p=2$45 inside $p=2$46. Consequently, if $p=2$47 is bounded below and caloric, then

$p=2$48

Applying this to the time-independent extension $p=2$49 yields the one-sided Liouville theorem: every smooth solution of

$p=2$50

with $p=2$51 is constant (Kogoj et al., 2020).

These semigroup-based proofs show that the linear Liouville argument is not tied to elliptic comparison on static domains. It can be recast as rigidity of stationary states for associated diffusions, or as a statement about the long-time smoothing of a parabolic evolution.

5. Probabilistic, nonlocal, and algebraic variants

In several linear settings, Liouville rigidity is proved by embedding the equation into a Markovian or group-theoretic structure. For the archetypal rescaling equation

$p=2$52

(Bogachev et al., 2014) interprets bounded solutions as harmonic functions of the Markov chain

$p=2$53

Then $p=2$54 is a martingale, and for any almost surely finite stopping time $p=2$55,

$p=2$56

In the critical regime $p=2$57, stopping at

$p=2$58

shrinks the rescaling factor $p=2$59. Uniform continuity of $p=2$60 then implies $p=2$61, hence constancy. In the discrete-scaling case $p=2$62 with $p=2$63, the paper replaces uniform continuity by a lattice return-time argument and the Choquet–Deny theorem (Bogachev et al., 2014).

For isotropic diffusions in random environment, (Fehrman, 2014) proves that on a full-probability subset of environments, the constants are the only strictly sub-linear invariant maps, and also the only bounded ancient invariant maps. The strictly sub-linear argument combines an entropy–Cauchy–Schwarz estimate, control of the large-time second moment

$p=2$64

and averaged bounds on the physical entropy

$p=2$65

The resulting estimate forces

$p=2$66

and continuity then implies $p=2$67 is constant (Fehrman, 2014).

In the nonlocal setting, (Alibaud et al., 2018) studies symmetric pure-jump Lévy generators

$p=2$68

with $p=2$69 symmetric. The main theorem states that for $p=2$70,

$p=2$71

Hence Liouville holds if and only if the additive subgroup generated by $p=2$72 is dense in $p=2$73. In one dimension, this becomes an irrationality criterion on jump sizes. The more general Courrège-class classification in (Alibaud et al., 2019) replaces $p=2$74 by the combined period group $p=2$75, again identifying bounded solutions with periodic functions and Liouville with the filling of the whole space.

A plausible implication is that, across these variants, the linear Liouville argument can often be read as the triviality of an invariant $p=2$76-algebra, an annihilator subgroup, or a reciprocal lattice. The concrete realization changes—from martingale stopping to entropy to subgroup propagation—but the rigidity mechanism remains the elimination of nontrivial bounded invariants.

6. Geometric consequences, classical prototypes, and limits of the method

The classical prototype remains the Liouville theorem for harmonic functions. The survey "Liouville properties" (Colding et al., 2019) records several linear proofs on $p=2$77: the mean-value argument, Harnack-based arguments, and the gradient estimate

$p=2$78

Letting $p=2$79 gives $p=2$80 for bounded harmonic $p=2$81. On complete manifolds with $p=2$82, Yau’s generalization asserts that every bounded harmonic function is constant, with the Cheng–Yau gradient estimate and the Bochner formula as the principal tools. The same survey emphasizes that volume doubling and a scale-invariant Poincaré inequality suffice for finite dimensionality of spaces of polynomial-growth harmonic functions (Colding et al., 2019). This suggests that the linear Liouville argument is robust under rough geometric hypotheses, not only in smooth Euclidean settings.

A geometric application of the linear PDE strategy appears in (Warren, 2015). For a semiconvex function $p=2$83, the gradient graph

$p=2$84

can be rotated by Yuan’s orthogonal transformation so that $p=2$85 becomes a global graph with uniformly bounded slope over new coordinates $p=2$86. In those coordinates, the induced metric satisfies

$p=2$87

and the Laplace–Beltrami equation $p=2$88 becomes a linear uniformly elliptic divergence-form equation on $p=2$89. De Giorgi–Nash–Moser theory then implies that bounded harmonic functions on $p=2$90 are constant. The paper uses this Liouville property to deduce a Bernstein-type result for Hamiltonian stationary equations: if the Lagrangian phase angle satisfies

$p=2$91

or $p=2$92, then $p=2$93 is constant, the graph is special Lagrangian, and Yuan’s rigidity yields that $p=2$94 is quadratic (Warren, 2015).

Several limitations recur across the literature. Liouville is not a purely local boundedness statement. In the Riemannian setting, failure of the weak maximum principle at infinity or stochastic completeness leads to failure of the $p=2$95–Liouville property (Mari et al., 2011). In the nonlocal setting, if the closed subgroup generated by the jump support is proper, then bounded nonconstant periodic solutions exist (Alibaud et al., 2018). In the random-environment and rescaling settings, additional structure such as strict sub-linearity, uniform continuity, or full-probability environmental conditions is essential (Fehrman, 2014, Bogachev et al., 2014).

A common point of confusion is to identify “Liouville” solely with bounded harmonic functions on $p=2$96. The papers considered here show a broader taxonomy: Schrödinger operators, Kolmogorov and Ornstein–Uhlenbeck operators, translation-invariant Lévy generators, affine rescaling equations, and harmonic functions on gradient graphs all admit linear Liouville arguments, but the decisive invariant may be a barrier, a heat kernel, a Harnack inequality, a stopping-time identity, or a subgroup of translations. The shared content is rigidity of bounded invariants under a linear evolution or linear operator, together with a structural criterion explaining exactly when that rigidity holds.