Maximum Principle in PDEs & Control

• Maximum Principle is a collection of assertions that control extremal solution values via boundary conditions, source terms, and admissibility constraints.
• It applies in elliptic, parabolic, fractional, and nonlocal equations, providing vital comparison principles, eigenvalue estimates, and uniqueness results.
• It underpins optimal control and geometric analysis by establishing Pontryagin-type conditions, Omori–Yau maximum forms, and effective curvature estimates.

Searching arXiv for the cited papers to ground the article in recent and related literature. Maximum principle denotes a family of structural assertions that control the sign or extremal values of solutions by boundary data, source terms, or admissibility constraints. In elliptic and parabolic PDE, it asserts that subsolutions cannot create interior positive maxima except under tightly constrained circumstances; in geometric analysis it appears in weak and Omori–Yau forms for trace operators on complete manifolds; in fractional and nonlocal equations it is reformulated through extremum principles for nonlocal derivatives or exterior-value conditions; and in optimal control it denotes Pontryagin-type necessary conditions expressed through Hamiltonians and adjoint equations (Bisterzo, 2023, Kirane et al., 2020, Du et al., 2012).

1. Classical PDE paradigm and global variants

In the elliptic setting, a maximum principle is a statement saying that a subsolution cannot attain a positive maximum in the interior unless it is constant. On bounded Euclidean domains, the standard model is

$$-\Delta u + c(x)u \le 0 \quad \text{in }\Omega,\qquad u\le 0 \quad \text{on }\partial\Omega,$$

with $0\le c\in C^0(\Omega)$, which implies $u\le 0$ in $\Omega$ for bounded-above solutions (Bisterzo, 2023). For second-order uniformly elliptic operators, the weak maximum principle yields sign preservation and comparison, while the strong maximum principle asserts that if $Lu\ge 0$ and $u$ attains its maximum at an interior point, then $u$ is constant; equivalently, if $u$ is nonnegative and vanishes at some interior point, then $u\equiv 0$ (Cassani et al., 2020).

On complete manifolds, the pointwise boundary formulation is frequently replaced by an Omori–Yau sequence. For the Laplacian, if $u\in C^2(M)$ and $0\le c\in C^0(\Omega)$0, there exists $0\le c\in C^0(\Omega)$1 such that

$0\le c\in C^0(\Omega)$2

A weak version drops the gradient condition. This mechanism extends to semi-elliptic trace operators

$0\le c\in C^0(\Omega)$3

with $0\le c\in C^0(\Omega)$4 symmetric positive semi-definite and $0\le c\in C^0(\Omega)$5, under suitable growth control through a proper function $0\le c\in C^0(\Omega)$6 and an auxiliary function $0\le c\in C^0(\Omega)$7 (Bessa et al., 2012). In this form, the maximum principle becomes a global compactness substitute for noncompact manifolds and a standard input for curvature estimates.

2. Principal eigenvalues, degeneracy, and unbounded elliptic domains

In unbounded domains, the absence of an outer boundary at infinity makes bounded-above subsolutions the natural class. For

$0\le c\in C^0(\Omega)$8

with Dirichlet or indefinite Robin boundary operator $0\le c\in C^0(\Omega)$9, the generalized principal eigenvalue is defined by

$u\le 0$0

Under the bounded potential drift condition

$u\le 0$1

positivity of $u\le 0$2 is necessary and sufficient for the validity of the maximum principle in unbounded domains; if $u\le 0$3, the relevant replacement is a Critical Maximum Principle, obtained under the growth condition

$u\le 0$4

(Nordmann, 2021). This gives a unified treatment of Dirichlet, Neumann, and indefinite Robin problems.

For fully nonlinear degenerate elliptic operators in bounded domains, positivity of the more familiar generalized eigenvalues can fail to characterize the maximum principle. The appropriate quantity is

$u\le 0$5

where the supersolution is required on an outer domain $u\le 0$6. Under the standard viscosity hypotheses (degenerate ellipticity, positive homogeneity, and Crandall–Ishii–Lions continuity), one has

$u\le 0$7

The outer-domain definition is essential because, for degenerate operators, $u\le 0$8, $u\le 0$9, or $\Omega$0 may be positive while the viscosity maximum principle fails (Berestycki et al., 2013). This corrects a common misconception inherited from uniformly elliptic theory.

A complementary geometric route uses Dirichlet parabolicity. On warped products $\Omega$1, if an unbounded domain is contained in a strip $\Omega$2 and $\Omega$3 satisfies suitable curvature or growth conditions, then the domain is $\Omega$4-parabolic, and bounded-above subsolutions of

$\Omega$5

obey an unbounded maximum principle (Bisterzo, 2023). This replaces spectral positivity by geometric smallness.

3. Parabolic and non-uniformly parabolic formulations

For divergence-form parabolic equations

$\Omega$6

the maximum principle can be recast as a global boundedness estimate under non-uniform ellipticity. The ellipticity is measured by

$\Omega$7

with only spatial integrability of $\Omega$8 and $\Omega$9, rather than uniform parabolicity. Under

$Lu\ge 0$0

together with the drift decomposition $Lu\ge 0$1, $Lu\ge 0$2, and mixed-norm assumptions on $Lu\ge 0$3, any Lipschitz weak solution with zero past satisfies

$Lu\ge 0$4

This is a De Giorgi–Moser type maximum principle for non-uniformly parabolic and possibly degenerate operators with supercritical drift, rather than a boundary supremum statement in the classical sense (Zhang, 2020).

The proof architecture is De Giorgi iteration on cylinders and levels, based on local energy inequalities, time-dependent variational embeddings, and multi-term variational inequalities adapted to the non-uniform ellipticity parameters $Lu\ge 0$5 and $Lu\ge 0$6 (Zhang, 2020). In the uniformly elliptic case, the estimate reduces to standard energy-plus-De Giorgi arguments; in the degenerate case, it yields global $Lu\ge 0$7 control where classical perturbative methods fail.

For quasilinear SPDEs on bounded domains, maximum principles are obtained from Itô formulas for the positive part together with energy estimates. In the non-obstacle case, local solutions that are non-positive on the lateral boundary, encoded by $Lu\ge 0$8, satisfy comparison and maximum principles without regularity assumptions on the coefficients beyond bounded measurable ellipticity (Denis et al., 2012). With obstacle, the same mechanism extends to reflected quasilinear SPDEs: the reflection measure $Lu\ge 0$9 acts only where the solution meets the obstacle, and one derives $u$0-estimates for $u$1 against an Itô barrier $u$2 (Laurent et al., 2012). In both cases, the maximum principle is quantitative and probabilistic rather than pointwise.

A further consequence of the non-uniformly parabolic estimate is stochastic. Applied to the backward Kolmogorov equation of

$u$3

it yields Krylov-type estimates, weak existence for degenerate SDEs with singular coefficients, and, via Krylov’s Markov selection theorem, existence of a strong Markov family (Zhang, 2020). This gives the maximum principle an indirect but decisive role in stochastic well-posedness.

4. Fractional and nonlocal maximum principles

In time-fractional diffusion with Atangana–Baleanu derivative, the core difficulty is that the time operator is nonlocal and has a non-singular Mittag–Leffler kernel. An extremum principle for the AB derivative shows that at a maximum the fractional derivative is nonnegative and at a minimum it is nonpositive, which leads to weak maximum and minimum principles for one-dimensional sub-diffusion with Cauchy–Dirichlet data, and then to uniqueness and continuous dependence of classical solutions for linear and nonlinear problems (Borikhanov et al., 2018).

A broader fractional theory is obtained for sequential Caputo derivatives. For

$u$4

the sequential derivative is nonpositive at a maximum and nonnegative at a minimum. This yields comparison principles for linear fractional differential equations and maximum principles for time-space fractional diffusion, pseudo-parabolic equations, elliptic equations with sequential Caputo derivatives, and fractional Laplace equations on cylindrical domains. In particular, the corresponding extremum principle answers positively the open problem on maximum principles for space and time-space fractional PDEs posed by Luchko in 2011 (Kirane et al., 2020).

For symmetric stable operators, including the fractional Laplacian, the natural formulation is exterior rather than boundary pointwise control. With

$u$5

the weak maximum principle holds for

$u$6

provided $u$7 a.e. on $u$8, $u$9 is a distributional subsolution, and the boundary layer satisfies

$u$0

This replaces upper semicontinuity at the boundary by an integral condition at the natural $u$1 scale of nonlocal Dirichlet problems (Grube et al., 2022). A plausible implication is that, for nonlocal operators, the “boundary” controlling the interior is the whole complement $u$2, not merely $u$3.

5. Higher-order operators and geometric trace-operator forms

For higher-order elliptic operators, positivity preserving and strong maximum principles generally fail. A model failure is the biharmonic operator with clamped boundary conditions. A restoration mechanism is to add a sufficiently strong second-order uniformly elliptic part. De Giorgi-type level estimates for $u$4, $u$5, together with a Harnack-type inequality for functions not assumed to belong to De Giorgi classes, lead in dimensions $u$6 to a strong maximum principle for operators such as

$u$7

and more generally

$u$8

provided $u$9 is large enough and second-order derivatives are taken into account (Cassani et al., 2020). The structural point is that the second-order part recovers the sign mechanism absent from the pure $u$0-th order operator.

In differential geometry, maximum principles are built for semi-elliptic trace operators

$u$1

where $u$2 is symmetric positive semi-definite. Under growth control through a proper function $u$3 and a function $u$4, one obtains a weak/Omori–Yau maximum principle for $u$5 on complete manifolds (Bessa et al., 2012). This framework is tailored to Newton tensors $u$6 associated with higher-order mean curvatures $u$7, since the geometric operators

$u$8

enter directly in hypersurface theory.

These operator-valued maximum principles have concrete geometric consequences. For hypersurfaces in warped products $u$9, they imply higher-order mean curvature estimates such as

$u\equiv 0$0

and slice theorems stating that complete hypersurfaces with constant positive $u\equiv 0$1 or $u\equiv 0$2, under sign and growth assumptions on the angle function, radial curvature, and warping function, must be slices $u\equiv 0$3 (Bessa et al., 2012). Here the maximum principle functions as a global rigidity instrument rather than a local comparison device.

6. Pontryagin-type maximum principles in stochastic and discrete-time control

In optimal control, maximum principle refers to a necessary optimality system rather than to boundary sign propagation. For semilinear stochastic evolution equations in Hilbert spaces,

$u\equiv 0$4

with control entering both drift and diffusion, non-convex control set, and unbounded operators in both parts, the Hamiltonian is

$u\equiv 0$5

If $u\equiv 0$6 is optimal, the first adjoint $u\equiv 0$7 solves a backward stochastic evolution equation, a second-order adjoint is represented by an operator-valued process $u\equiv 0$8, and the optimality condition is

$u\equiv 0$9

for almost every $u\in C^2(M)$0 and every $u\in C^2(M)$1 (Du et al., 2012). This is the infinite-dimensional analogue of Peng’s stochastic maximum principle.

For controlled SPDEs driven by a continuous Hilbert-space-valued martingale, with non-convex control domain and control entering the martingale term, the Hamiltonian takes the form

$u\in C^2(M)$2

and an optimal control $u\in C^2(M)$3 satisfies

$u\in C^2(M)$4

for all $u\in C^2(M)$5, almost surely and almost everywhere in time (Al-Hussein, 2012). The corresponding adjoint equation is a backward stochastic evolution equation with an additional orthogonal martingale component.

For systems driven by fractional Brownian motion with Hurst parameter $u\in C^2(M)$6, the maximum principle involves Malliavin derivatives and a backward stochastic differential equation driven simultaneously by the fractional Brownian motion and the underlying standard Brownian motion. This reflects the non-semimartingale structure of the state noise and differentiates the fractional maximum principle sharply from the Brownian case (Han et al., 2012).

A discrete-time analogue also exists. For deterministic nonstationary control problems with

$u\in C^2(M)$7

Gâteaux differentials yield a discrete-time maximum principle consisting of the adjoint recursion

$u\in C^2(M)$8

the stationarity condition

$u\in C^2(M)$9

and a transversality condition at infinity; in finite horizon the latter reduces to a terminal condition. The same framework yields the discrete-time Euler equation and necessary and sufficient conditions for Nash equilibria in dynamic games (Corella et al., 16 Jan 2026). This suggests that “maximum principle” names a unified variational idea across deterministic, stochastic, continuous-time, and discrete-time control, even though its analytic content differs from the PDE usage.