De Giorgi's Elliptic Regularisation in PDEs

• De Giorgi’s elliptic regularisation is a method that replaces non-elliptic or rough problems with a family of convex, elliptic approximants indexed by a parameter ε.
• In the semilinear wave-equation setting, the approach reformulates the problem into a fourth-order elliptic-in-time framework, leading to unique minimisers and rigorous finite-time convergence.
• The strategy extends to free-boundary and geometric problems, leveraging elliptic coercivity to upgrade weak solutions and ensure stability in space-time finite element implementations.

Searching arXiv for the cited papers to ground the article in current arXiv records. De Giorgi’s elliptic regularisation is a variational and analytic paradigm in which a non-elliptic, evolution, geometric, or rough elliptic problem is replaced by a family of elliptic, or elliptic-in-time, problems indexed by a parameter $\varepsilon>0$. In the semilinear wave-equation setting, it refers to De Giorgi’s 1996 proposal that solutions of nonlinear wave equations should arise as limits of minimisers of convex space-time functionals with exponential time weights; this finite-time conjecture was proved by Stefanelli, and the same viewpoint has been turned into a conforming space-time finite element method for semilinear waves (Stefanelli, 2010, Banjai et al., 30 Jul 2025). In related literatures, the same expression also denotes elliptic approximations of parabolic free-boundary problems, geometric regularisations of non-elliptic systems such as isometric immersion, and the broader regularising mechanism by which elliptic coercivity and convexity upgrade weak solutions to higher regularity (Audrito et al., 2023, Anderson, 2017, Qian, 2 Dec 2025).

1. Historical origin and conceptual scope

De Giorgi’s original conjecture proposed that the nonlinear wave equation

$u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$

should be recovered as the limit, as $\varepsilon\to 0$, of minimisers of a convex functional defined on entire space-time trajectories. In the formulation reproduced in the wave-equation literature, the regularised functional has the schematic form

$$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$

subject to initial conditions. For each fixed $\varepsilon>0$, the functional is convex, its minimiser is unique, and the Euler–Lagrange equation is fourth order in time and elliptic in the time variable; the hyperbolic equation is then expected, and in finite time proved, to emerge in the limit $\varepsilon\to 0$ (Stefanelli, 2010).

The later literature makes explicit that this is part of a broader weighted global variational principle. In the semilinear wave setting it is placed in the WIDE framework, while in parabolic free-boundary problems the closely related terminology WED or WIED, standing for weighted energy dissipation, is used for functionals of the same exponential-in-time type (Banjai et al., 30 Jul 2025, Audrito et al., 2023).

The expression does not designate a single formula valid in every subject. In the literature considered here, it denotes a family of constructions sharing the same structural features: convex or coercive space-time energies, elliptic approximants for each fixed $\varepsilon$, and a limiting procedure that recovers a non-elliptic target problem.

2. Variational formulation for semilinear wave equations

On a bounded domain $\Omega\subset\mathbb R^d$, $d=1,2,3$, over a finite interval $I=(0,T]$, the semilinear wave equation considered in the recent space-time formulation is

$u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$0

The regularised problem is posed on

$u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$1

with the affine constraint

$u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$2

For $u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$3, the functional is

$u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$4

Its strict convexity and coercivity on $u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$5 yield a unique minimiser $u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$6 (Banjai et al., 30 Jul 2025).

The Euler–Lagrange equation is written weakly as

$u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$7

where

$u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$8

and

$u_{tt}-\Delta u+\frac{p}{2}|u|^{p-2}u=0$9

Under sufficient regularity, the strong form becomes

$\varepsilon\to 0$0

The fourth-order time operator is the regularising mechanism: for fixed $\varepsilon\to 0$1 the problem is elliptic in time, while formally sending $\varepsilon\to 0$2 removes the higher derivatives and recovers the second-order wave equation (Banjai et al., 30 Jul 2025).

3. Analytical structure, coercivity, and the limit $\varepsilon\to 0$3

The regularised bilinear part induces the weighted energy norm

$\varepsilon\to 0$4

This coercive structure is central. It yields well-posedness by minimisation, and, together with monotonicity of the defocusing nonlinearity, produces stability estimates independent of discretisation parameters in the later finite element realisation (Banjai et al., 30 Jul 2025).

Stefanelli’s finite-time proof of De Giorgi’s conjecture established that, for each finite $\varepsilon\to 0$5, minimisers of the weighted elliptic functional converge, up to subsequences, to a variational solution of the semilinear wave equation. The convergence is weak in the natural energy spaces and strong in local $\varepsilon\to 0$6-type topologies, sufficient to pass to the nonlinear term and identify the limit equation in the distributional sense (Stefanelli, 2010).

The finite-time proof also clarifies a recurrent feature of De Giorgi’s elliptic regularisation: the approximating problems are better posed than the target hyperbolic problem because they arise from convex minimisation. The limiting passage is the delicate part. In Stefanelli’s formulation, the identification of $\varepsilon\to 0$7 is rigorous, whereas the identification of $\varepsilon\to 0$8 is described as more delicate in that framework (Stefanelli, 2010).

A common misunderstanding is to treat this construction as a standard viscous perturbation of the wave equation. In the wave literature considered here, it is not a step-by-step time discretisation and not an ad hoc damping device. It is a global space-time minimisation problem with an exponential weight, whose Euler–Lagrange equation is elliptic in time and whose hyperbolic limit is recovered only after letting $\varepsilon\to 0$9.

4. Space-time finite element realisation

The recent numerical realisation of De Giorgi’s viewpoint uses a conforming space-time Galerkin discretisation of the elliptic-regularised semilinear wave problem. The space-time cylinder $$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$0 is decomposed into time slabs and a spatial mesh $$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$1. The spatial finite element space is

$$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$2

while the temporal space uses globally $$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$3 splines,

$$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$4

together with the homogeneous initial conditions $$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$5. The resulting space-time finite element space is the tensor product

$$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$6

The discrete problem is

$$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$7

and is exactly the Galerkin discretisation of the elliptic-regularised variational problem (Banjai et al., 30 Jul 2025).

Because the discrete problem is itself the minimisation of the strictly convex functional $$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$8 on a finite-dimensional conforming subspace, it is well posed for any conforming choice of discretisation spaces. The same structure yields unconditional stability: there is no CFL-type restriction linking the time step to the spatial mesh size. The paper proves quasi-optimal error bounds and, in regimes where the energy norm controls the nonlinear term, derives the rate

$$I_\varepsilon(u)=\int_0^T\!\!\int_{\mathbb R^d} e^{-t/\varepsilon} \Big(|u_{tt}|^2+\varepsilon^{-2}|\nabla u|^2+\varepsilon^{-2}|u|^p\Big)\,dx\,dt,$$9

matching the interpolation orders in time and space (Banjai et al., 30 Jul 2025).

The implementation is not numerically trivial, because the exponential weights $\varepsilon>0$0 can underflow when $\varepsilon>0$1 is small. The paper studies this through a scalar fourth-order ODE model, introduces scaled matrices to factor out the dominant exponential weight, and reports that conditioning deteriorates when $\varepsilon>0$2, with empirical evidence that $\varepsilon>0$3 keeps the local weight in a moderate range. In one space dimension, the linear discrete system has a Kronecker-product structure

$\varepsilon>0$4

which can be exploited for efficient solvers and preconditioning (Banjai et al., 30 Jul 2025).

5. Extensions beyond semilinear wave equations

The same basic principle has been transferred to parabolic free-boundary problems. For

$\varepsilon>0$5

Audrito and Sanz-Perela consider the weighted energy-dissipation functional

$\varepsilon>0$6

Its minimisers satisfy the elliptic-in-space-time equation

$\varepsilon>0$7

and the paper proves existence of strong solutions by uniform energy estimates, non-degeneracy estimates, and passage to the weak limit. The authors state that this is, to their knowledge, the first use of elliptic regularisation in the context of parabolic obstacle problems (Audrito et al., 2023).

A distinct geometric instance is Anderson’s elliptic regularisation of the isometric immersion problem. Instead of prescribing the full induced metric $\varepsilon>0$8, the regularised data are

$\varepsilon>0$9

where $\varepsilon\to 0$0 is the conformal class, $\varepsilon\to 0$1 the conformal factor relative to a fixed background metric, and $\varepsilon\to 0$2 the mean curvature. For $\varepsilon\to 0$3 this reduces to the isometric immersion data, while for every $\varepsilon\to 0$4 the resulting system is elliptic and Fredholm. The corresponding variational interpolation uses

$\varepsilon\to 0$5

and yields boundary data essentially equivalent to

$\varepsilon\to 0$6

The paper sets up this De Giorgi-style framework, but explicitly does not carry out a full $\varepsilon\to 0$7-convergence or limiting analysis as $\varepsilon\to 0$8 (Anderson, 2017).

These extensions show that the phrase is not confined to hyperbolic equations. It also covers elliptic approximants for free-boundary evolution and geometric PDE where the original formulation is non-elliptic.

6. Relation to De Giorgi regularity theory and interpretive issues

In the regularity-theoretic literature, “De Giorgi’s elliptic regularisation” can also denote the intrinsic regularising effect of elliptic equations themselves. For uniformly elliptic divergence-form equations with rough coefficients, De Giorgi’s method uses truncations, Caccioppoli inequalities, Sobolev embeddings, iteration on level sets, and oscillation decay to upgrade weak $\varepsilon\to 0$9 solutions to locally Hölder functions. Recent lecture notes describe this as the mechanism by which weak energy solutions are “shown to become locally Hölder continuous,” and the same line is presented as central to the solution of Hilbert’s nineteenth problem (Brigati et al., 13 Oct 2025, Qian, 2 Dec 2025).

This broader use is already visible in later expositions of De Giorgi–Moser arguments, where the “regularisation” is not a family $\varepsilon$0 of parameter-dependent approximants, but the fact that the elliptic equation regularises its own weak solutions through truncation and measure decay. In that sense, the phrase encompasses both explicit $\varepsilon$1-regularisation and implicit regularisation by elliptic coercivity (Siljander, 2010).

Modern nonlinear regularity theory pushes this interpretation further. In recent work on strongly singular or degenerate elliptic and parabolic equations with ellipticity only outside a ball in gradient space, the De Giorgi lineage is expressed through nonlinear transforms of the gradient such as

$\varepsilon$2

for which one obtains Besov or $\varepsilon$3 regularity even though the original operator is flat or singular near $\varepsilon$4. There the regularising variable is not $\varepsilon$5 itself but a nonlinear function of $\varepsilon$6, and the ellipticity is described as being available only “at infinity” in the gradient variable (Ambrosio, 7 Nov 2025).

Two interpretive cautions follow. First, De Giorgi’s elliptic regularisation is not a single universally fixed construction; it is a family of elliptic or effectively elliptic reformulations adapted to different classes of problems. Second, not every paper using the phrase proves a full limit theorem. Some works, such as Stefanelli’s finite-time wave result, do prove convergence to the target equation, whereas others, such as Anderson’s geometric regularisation, primarily establish ellipticity, Fredholm properties, and variational structure without completing the singular limit (Stefanelli, 2010, Anderson, 2017).

Taken together, these uses identify a stable core idea: difficult dynamics, rough coefficients, or non-elliptic constraints are recast into a coercive elliptic framework, and the resulting minimisation, compactness, or energy estimates provide existence, stability, approximation, or regularity that is otherwise inaccessible.